diff options
Diffstat (limited to 'src')
155 files changed, 6383 insertions, 4225 deletions
diff --git a/src/egl/drivers/xdri/Makefile b/src/egl/drivers/xdri/Makefile index afd551dea5c..a721b997e69 100644 --- a/src/egl/drivers/xdri/Makefile +++ b/src/egl/drivers/xdri/Makefile @@ -48,7 +48,7 @@ $(TOP)/$(LIB_DIR)/$(DRIVER_NAME): $(OBJECTS) $(TOP)/bin/mklib -o $(DRIVER_NAME) \ -noprefix \ -major 1 -minor 0 \ - -L $(TOP)/$(LIB_DIR) \ + -L$(TOP)/$(LIB_DIR) \ -install $(TOP)/$(LIB_DIR) \ $(OBJECTS) $(DRM_LIB) $(MISC_LIBS) diff --git a/src/egl/drivers/xdri/egl_xdri.c b/src/egl/drivers/xdri/egl_xdri.c index 83d4b86d98f..3b3e312746e 100644 --- a/src/egl/drivers/xdri/egl_xdri.c +++ b/src/egl/drivers/xdri/egl_xdri.c @@ -654,7 +654,10 @@ xdri_eglInitialize(_EGLDriver *drv, EGLDisplay dpy, xdri_drv->Base.Initialized = EGL_TRUE; - snprintf(name, sizeof(name), "X/DRI:%s", xdri_drv->dri_driver_name); + if (xdri_drv->dri_driver_name) + snprintf(name, sizeof(name), "X/DRI:%s", xdri_drv->dri_driver_name); + else + snprintf(name, sizeof(name), "X/DRI"); xdri_drv->Base.Name = name; /* we're supporting EGL 1.4 */ diff --git a/src/gallium/auxiliary/draw/draw_context.c b/src/gallium/auxiliary/draw/draw_context.c index 78249054f2a..b439bc4059f 100644 --- a/src/gallium/auxiliary/draw/draw_context.c +++ b/src/gallium/auxiliary/draw/draw_context.c @@ -274,6 +274,14 @@ draw_enable_point_sprites(struct draw_context *draw, boolean enable) } +void +draw_set_force_passthrough( struct draw_context *draw, boolean enable ) +{ + draw_do_flush( draw, DRAW_FLUSH_STATE_CHANGE ); + draw->force_passthrough = enable; +} + + /** * Ask the draw module for the location/slot of the given vertex attribute in * a post-transformed vertex. diff --git a/src/gallium/auxiliary/draw/draw_context.h b/src/gallium/auxiliary/draw/draw_context.h index 0ab3681b647..3eeb4535311 100644 --- a/src/gallium/auxiliary/draw/draw_context.h +++ b/src/gallium/auxiliary/draw/draw_context.h @@ -160,6 +160,9 @@ void draw_set_render( struct draw_context *draw, void draw_set_driver_clipping( struct draw_context *draw, boolean bypass_clipping ); +void draw_set_force_passthrough( struct draw_context *draw, + boolean enable ); + /******************************************************************************* * Draw pipeline */ diff --git a/src/gallium/auxiliary/draw/draw_pipe_vbuf.c b/src/gallium/auxiliary/draw/draw_pipe_vbuf.c index c0cf4269dbb..9825e116c32 100644 --- a/src/gallium/auxiliary/draw/draw_pipe_vbuf.c +++ b/src/gallium/auxiliary/draw/draw_pipe_vbuf.c @@ -231,9 +231,9 @@ vbuf_set_prim( struct vbuf_stage *vbuf, uint prim ) unsigned emit_sz = 0; unsigned src_buffer = 0; unsigned output_format; - unsigned src_offset = (vbuf->vinfo->src_index[i] * 4 * sizeof(float) ); + unsigned src_offset = (vbuf->vinfo->attrib[i].src_index * 4 * sizeof(float) ); - switch (vbuf->vinfo->emit[i]) { + switch (vbuf->vinfo->attrib[i].emit) { case EMIT_4F: output_format = PIPE_FORMAT_R32G32B32A32_FLOAT; emit_sz = 4 * sizeof(float); diff --git a/src/gallium/auxiliary/draw/draw_private.h b/src/gallium/auxiliary/draw/draw_private.h index 626a2e3e304..5d531146c5f 100644 --- a/src/gallium/auxiliary/draw/draw_private.h +++ b/src/gallium/auxiliary/draw/draw_private.h @@ -163,12 +163,15 @@ struct draw_context struct { boolean bypass_clipping; + boolean bypass_vs; } driver; boolean flushing; /**< debugging/sanity */ boolean suspend_flushing; /**< internally set */ boolean bypass_clipping; /**< set if either api or driver bypass_clipping true */ + boolean force_passthrough; /**< never clip or shade */ + /* pipe state that we need: */ const struct pipe_rasterizer_state *rasterizer; struct pipe_viewport_state viewport; @@ -193,7 +196,7 @@ struct draw_context const float (*aligned_constants)[4]; - float (*aligned_constant_storage)[4]; + const float (*aligned_constant_storage)[4]; unsigned const_storage_size; diff --git a/src/gallium/auxiliary/draw/draw_pt.c b/src/gallium/auxiliary/draw/draw_pt.c index 669c11c993c..87ec6ae20c2 100644 --- a/src/gallium/auxiliary/draw/draw_pt.c +++ b/src/gallium/auxiliary/draw/draw_pt.c @@ -69,26 +69,26 @@ draw_pt_arrays(struct draw_context *draw, return TRUE; } - - if (!draw->render) { - opt |= PT_PIPELINE; - } - - if (draw_need_pipeline(draw, - draw->rasterizer, - prim)) { - opt |= PT_PIPELINE; - } - - if (!draw->bypass_clipping && !draw->pt.test_fse) { - opt |= PT_CLIPTEST; + if (!draw->force_passthrough) { + if (!draw->render) { + opt |= PT_PIPELINE; + } + + if (draw_need_pipeline(draw, + draw->rasterizer, + prim)) { + opt |= PT_PIPELINE; + } + + if (!draw->bypass_clipping && !draw->pt.test_fse) { + opt |= PT_CLIPTEST; + } + + if (!draw->rasterizer->bypass_vs) { + opt |= PT_SHADE; + } } - - if (!draw->rasterizer->bypass_vs) { - opt |= PT_SHADE; - } - - + if (opt == 0) middle = draw->pt.middle.fetch_emit; else if (opt == PT_SHADE && !draw->pt.no_fse) diff --git a/src/gallium/auxiliary/draw/draw_pt_emit.c b/src/gallium/auxiliary/draw/draw_pt_emit.c index d4eca80588b..d520b05869b 100644 --- a/src/gallium/auxiliary/draw/draw_pt_emit.c +++ b/src/gallium/auxiliary/draw/draw_pt_emit.c @@ -84,11 +84,11 @@ void draw_pt_emit_prepare( struct pt_emit *emit, unsigned emit_sz = 0; unsigned src_buffer = 0; unsigned output_format; - unsigned src_offset = (vinfo->src_index[i] * 4 * sizeof(float) ); + unsigned src_offset = (vinfo->attrib[i].src_index * 4 * sizeof(float) ); - switch (vinfo->emit[i]) { + switch (vinfo->attrib[i].emit) { case EMIT_4F: output_format = PIPE_FORMAT_R32G32B32A32_FLOAT; emit_sz = 4 * sizeof(float); diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch_emit.c b/src/gallium/auxiliary/draw/draw_pt_fetch_emit.c index 5a4db6cfe56..3966ad48ba7 100644 --- a/src/gallium/auxiliary/draw/draw_pt_fetch_emit.c +++ b/src/gallium/auxiliary/draw/draw_pt_fetch_emit.c @@ -121,7 +121,7 @@ static void fetch_emit_prepare( struct draw_pt_middle_end *middle, memset(&key, 0, sizeof(key)); for (i = 0; i < vinfo->num_attribs; i++) { - const struct pipe_vertex_element *src = &draw->pt.vertex_element[vinfo->src_index[i]]; + const struct pipe_vertex_element *src = &draw->pt.vertex_element[vinfo->attrib[i].src_index]; unsigned emit_sz = 0; unsigned input_format = src->src_format; @@ -129,7 +129,7 @@ static void fetch_emit_prepare( struct draw_pt_middle_end *middle, unsigned input_offset = src->src_offset; unsigned output_format; - switch (vinfo->emit[i]) { + switch (vinfo->attrib[i].emit) { case EMIT_4F: output_format = PIPE_FORMAT_R32G32B32A32_FLOAT; emit_sz = 4 * sizeof(float); diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c index 73fc70c1bc9..f7e6a1a8eeb 100644 --- a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c +++ b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c @@ -79,6 +79,7 @@ static void fse_prepare( struct draw_pt_middle_end *middle, unsigned num_vs_inputs = draw->vs.vertex_shader->info.num_inputs; const struct vertex_info *vinfo; unsigned i; + unsigned nr_vbs = 0; if (!draw->render->set_primitive( draw->render, @@ -102,7 +103,7 @@ static void fse_prepare( struct draw_pt_middle_end *middle, fse->key.viewport = !draw->identity_viewport; fse->key.clip = !draw->bypass_clipping; - fse->key.pad = 0; + fse->key.const_vbuffers = 0; memset(fse->key.element, 0, fse->key.nr_elements * sizeof(fse->key.element[0])); @@ -116,16 +117,23 @@ static void fse_prepare( struct draw_pt_middle_end *middle, */ fse->key.element[i].in.buffer = src->vertex_buffer_index; fse->key.element[i].in.offset = src->src_offset; + nr_vbs = MAX2(nr_vbs, src->vertex_buffer_index + 1); } + for (i = 0; i < 5 && i < nr_vbs; i++) { + if (draw->pt.vertex_buffer[i].pitch == 0) + fse->key.const_vbuffers |= (1<<i); + } + if (0) debug_printf("%s: lookup const_vbuffers: %x\n", __FUNCTION__, fse->key.const_vbuffers); + { unsigned dst_offset = 0; for (i = 0; i < vinfo->num_attribs; i++) { unsigned emit_sz = 0; - switch (vinfo->emit[i]) { + switch (vinfo->attrib[i].emit) { case EMIT_4F: emit_sz = 4 * sizeof(float); break; @@ -153,8 +161,8 @@ static void fse_prepare( struct draw_pt_middle_end *middle, * numbers, not to positions in the hw vertex description -- * that's handled by the output_offset field. */ - fse->key.element[i].out.format = vinfo->emit[i]; - fse->key.element[i].out.vs_output = vinfo->src_index[i]; + fse->key.element[i].out.format = vinfo->attrib[i].emit; + fse->key.element[i].out.vs_output = vinfo->attrib[i].src_index; fse->key.element[i].out.offset = dst_offset; dst_offset += emit_sz; @@ -162,13 +170,7 @@ static void fse_prepare( struct draw_pt_middle_end *middle, } } - - /* Would normally look up a vertex shader and peruse its list of - * varients somehow. We omitted that step and put all the - * hardcoded "shaders" into an array. We're just making the - * assumption that this happens to be a matching shader... ie - * you're running isosurf, aren't you? - */ + fse->active = draw_vs_lookup_varient( draw->vs.vertex_shader, &fse->key ); @@ -177,18 +179,17 @@ static void fse_prepare( struct draw_pt_middle_end *middle, return ; } + if (0) debug_printf("%s: found const_vbuffers: %x\n", __FUNCTION__, + fse->active->key.const_vbuffers); + /* Now set buffer pointers: */ - for (i = 0; i < num_vs_inputs; i++) { - unsigned buf = draw->pt.vertex_element[i].vertex_buffer_index; - - fse->active->set_input( fse->active, - i, - - ((const ubyte *) draw->pt.user.vbuffer[buf] + - draw->pt.vertex_buffer[buf].buffer_offset), - - draw->pt.vertex_buffer[buf].pitch ); + for (i = 0; i < draw->pt.nr_vertex_buffers; i++) { + fse->active->set_buffer( fse->active, + i, + ((const ubyte *) draw->pt.user.vbuffer[i] + + draw->pt.vertex_buffer[i].buffer_offset), + draw->pt.vertex_buffer[i].pitch ); } *max_vertices = (draw->render->max_vertex_buffer_bytes / diff --git a/src/gallium/auxiliary/draw/draw_vertex.c b/src/gallium/auxiliary/draw/draw_vertex.c index 1446f785c51..3214213e445 100644 --- a/src/gallium/auxiliary/draw/draw_vertex.c +++ b/src/gallium/auxiliary/draw/draw_vertex.c @@ -49,7 +49,7 @@ draw_compute_vertex_size(struct vertex_info *vinfo) vinfo->size = 0; for (i = 0; i < vinfo->num_attribs; i++) { - switch (vinfo->emit[i]) { + switch (vinfo->attrib[i].emit) { case EMIT_OMIT: break; case EMIT_4UB: @@ -81,8 +81,8 @@ draw_dump_emitted_vertex(const struct vertex_info *vinfo, const uint8_t *data) unsigned i, j; for (i = 0; i < vinfo->num_attribs; i++) { - j = vinfo->src_index[i]; - switch (vinfo->emit[i]) { + j = vinfo->attrib[i].src_index; + switch (vinfo->attrib[i].emit) { case EMIT_OMIT: debug_printf("EMIT_OMIT:"); break; diff --git a/src/gallium/auxiliary/draw/draw_vertex.h b/src/gallium/auxiliary/draw/draw_vertex.h index 16c65c43175..a943607d7ed 100644 --- a/src/gallium/auxiliary/draw/draw_vertex.h +++ b/src/gallium/auxiliary/draw/draw_vertex.h @@ -75,12 +75,41 @@ struct vertex_info { uint num_attribs; uint hwfmt[4]; /**< hardware format info for this format */ - enum interp_mode interp_mode[PIPE_MAX_SHADER_INPUTS]; - enum attrib_emit emit[PIPE_MAX_SHADER_INPUTS]; /**< EMIT_x */ - uint src_index[PIPE_MAX_SHADER_INPUTS]; /**< map to post-xform attribs */ uint size; /**< total vertex size in dwords */ + + /* Keep this small and at the end of the struct to allow quick + * memcmp() comparisons. + */ + struct { + ubyte interp_mode:4; /**< INTERP_x */ + ubyte emit:4; /**< EMIT_x */ + ubyte src_index; /**< map to post-xform attribs */ + } attrib[PIPE_MAX_SHADER_INPUTS]; }; +static INLINE int +draw_vinfo_size( const struct vertex_info *a ) +{ + return ((const char *)&a->attrib[a->num_attribs] - + (const char *)a); +} + +static INLINE int +draw_vinfo_compare( const struct vertex_info *a, + const struct vertex_info *b ) +{ + unsigned sizea = draw_vinfo_size( a ); + return memcmp( a, b, sizea ); +} + +static INLINE void +draw_vinfo_copy( struct vertex_info *dst, + const struct vertex_info *src ) +{ + unsigned size = draw_vinfo_size( src ); + memcpy( dst, src, size ); +} + /** @@ -91,14 +120,15 @@ struct vertex_info */ static INLINE uint draw_emit_vertex_attr(struct vertex_info *vinfo, - enum attrib_emit emit, enum interp_mode interp, + enum attrib_emit emit, + enum interp_mode interp, /* only used by softpipe??? */ uint src_index) { const uint n = vinfo->num_attribs; assert(n < PIPE_MAX_SHADER_INPUTS); - vinfo->emit[n] = emit; - vinfo->interp_mode[n] = interp; - vinfo->src_index[n] = src_index; + vinfo->attrib[n].emit = emit; + vinfo->attrib[n].interp_mode = interp; + vinfo->attrib[n].src_index = src_index; vinfo->num_attribs++; return n; } diff --git a/src/gallium/auxiliary/draw/draw_vs.h b/src/gallium/auxiliary/draw/draw_vs.h index 45992d19867..68c24abad3b 100644 --- a/src/gallium/auxiliary/draw/draw_vs.h +++ b/src/gallium/auxiliary/draw/draw_vs.h @@ -64,7 +64,7 @@ struct draw_vs_varient_key { unsigned nr_outputs:8; unsigned viewport:1; unsigned clip:1; - unsigned pad:5; + unsigned const_vbuffers:5; struct draw_varient_element element[PIPE_MAX_ATTRIBS]; }; @@ -76,7 +76,7 @@ struct draw_vs_varient { struct draw_vertex_shader *vs; - void (*set_input)( struct draw_vs_varient *, + void (*set_buffer)( struct draw_vs_varient *, unsigned i, const void *ptr, unsigned stride ); diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.c b/src/gallium/auxiliary/draw/draw_vs_aos.c index a556477a767..87232865e23 100644 --- a/src/gallium/auxiliary/draw/draw_vs_aos.c +++ b/src/gallium/auxiliary/draw/draw_vs_aos.c @@ -92,9 +92,9 @@ struct x86_reg aos_get_x86( struct aos_compilation *cp, assert(which_reg == 1); offset = Offset(struct aos_machine, constants); break; - case X86_ATTRIBS: + case X86_BUFFERS: assert(which_reg == 0); - offset = Offset(struct aos_machine, attrib); + offset = Offset(struct aos_machine, buffer); break; default: assert(0); @@ -196,6 +196,18 @@ static void spill( struct aos_compilation *cp, unsigned idx ) } +void aos_spill_all( struct aos_compilation *cp ) +{ + unsigned i; + + for (i = 0; i < 8; i++) { + if (cp->xmm[i].dirty) + spill(cp, i); + aos_release_xmm_reg(cp, i); + } +} + + static struct x86_reg get_xmm_writable( struct aos_compilation *cp, struct x86_reg reg ) { @@ -1939,6 +1951,11 @@ static boolean build_vertex_program( struct draw_vs_varient_aos_sse *varient, save_fpu_state( &cp ); set_fpu_round_nearest( &cp ); + aos_init_inputs( &cp, linear ); + + cp.x86_reg[0] = 0; + cp.x86_reg[1] = 0; + /* Note address for loop jump */ label = x86_get_label(cp.func); @@ -2018,13 +2035,7 @@ static boolean build_vertex_program( struct draw_vs_varient_aos_sse *varient, /* Incr index */ - if (linear) { - x86_inc(cp.func, cp.idx_EBX); - } - else { - x86_lea(cp.func, cp.idx_EBX, x86_make_disp(cp.idx_EBX, 4)); - } - + aos_incr_inputs( &cp, linear ); } /* decr count, loop if not zero */ @@ -2065,15 +2076,13 @@ static void vaos_set_buffer( struct draw_vs_varient *varient, unsigned stride ) { struct draw_vs_varient_aos_sse *vaos = (struct draw_vs_varient_aos_sse *)varient; - unsigned i; - for (i = 0; i < vaos->base.key.nr_inputs; i++) { - if (vaos->base.key.element[i].in.buffer == buf) { - vaos->attrib[i].input_ptr = ((char *)ptr + - vaos->base.key.element[i].in.offset); - vaos->attrib[i].input_stride = stride; - } + if (buf < vaos->nr_vb) { + vaos->buffer[buf].base_ptr = (char *)ptr; + vaos->buffer[buf].stride = stride; } + + if (0) debug_printf("%s %d/%d: %p %d\n", __FUNCTION__, buf, vaos->nr_vb, ptr, stride); } @@ -2086,10 +2095,12 @@ static void PIPE_CDECL vaos_run_elts( struct draw_vs_varient *varient, struct draw_vs_varient_aos_sse *vaos = (struct draw_vs_varient_aos_sse *)varient; struct aos_machine *machine = vaos->draw->vs.aos_machine; + if (0) debug_printf("%s %d\n", __FUNCTION__, count); + machine->internal[IMM_PSIZE][0] = vaos->draw->rasterizer->point_size; machine->constants = vaos->draw->vs.aligned_constants; machine->immediates = vaos->base.vs->immediates; - machine->attrib = vaos->attrib; + machine->buffer = vaos->buffer; vaos->gen_run_elts( machine, elts, @@ -2105,10 +2116,13 @@ static void PIPE_CDECL vaos_run_linear( struct draw_vs_varient *varient, struct draw_vs_varient_aos_sse *vaos = (struct draw_vs_varient_aos_sse *)varient; struct aos_machine *machine = vaos->draw->vs.aos_machine; + if (0) debug_printf("%s %d %d const: %x\n", __FUNCTION__, start, count, + vaos->base.key.const_vbuffers); + machine->internal[IMM_PSIZE][0] = vaos->draw->rasterizer->point_size; machine->constants = vaos->draw->vs.aligned_constants; machine->immediates = vaos->base.vs->immediates; - machine->attrib = vaos->attrib; + machine->buffer = vaos->buffer; vaos->gen_run_linear( machine, start, @@ -2127,7 +2141,7 @@ static void vaos_destroy( struct draw_vs_varient *varient ) { struct draw_vs_varient_aos_sse *vaos = (struct draw_vs_varient_aos_sse *)varient; - FREE( vaos->attrib ); + FREE( vaos->buffer ); x86_release_func( &vaos->func[0] ); x86_release_func( &vaos->func[1] ); @@ -2140,6 +2154,7 @@ static void vaos_destroy( struct draw_vs_varient *varient ) static struct draw_vs_varient *varient_aos_sse( struct draw_vertex_shader *vs, const struct draw_vs_varient_key *key ) { + unsigned i; struct draw_vs_varient_aos_sse *vaos = CALLOC_STRUCT(draw_vs_varient_aos_sse); if (!vaos) @@ -2147,17 +2162,22 @@ static struct draw_vs_varient *varient_aos_sse( struct draw_vertex_shader *vs, vaos->base.key = *key; vaos->base.vs = vs; - vaos->base.set_input = vaos_set_buffer; + vaos->base.set_buffer = vaos_set_buffer; vaos->base.destroy = vaos_destroy; vaos->base.run_linear = vaos_run_linear; vaos->base.run_elts = vaos_run_elts; vaos->draw = vs->draw; - vaos->attrib = MALLOC( key->nr_inputs * sizeof(vaos->attrib[0]) ); - if (!vaos->attrib) + for (i = 0; i < key->nr_inputs; i++) + vaos->nr_vb = MAX2( vaos->nr_vb, key->element[i].in.buffer + 1 ); + + vaos->buffer = MALLOC( vaos->nr_vb * sizeof(vaos->buffer[0]) ); + if (!vaos->buffer) goto fail; + debug_printf("nr_vb: %d const: %x\n", vaos->nr_vb, vaos->base.key.const_vbuffers); + #if 0 tgsi_dump(vs->state.tokens, 0); #endif @@ -2179,8 +2199,8 @@ static struct draw_vs_varient *varient_aos_sse( struct draw_vertex_shader *vs, return &vaos->base; fail: - if (vaos && vaos->attrib) - FREE(vaos->attrib); + if (vaos && vaos->buffer) + FREE(vaos->buffer); if (vaos) x86_release_func( &vaos->func[0] ); diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.h b/src/gallium/auxiliary/draw/draw_vs_aos.h index 7fe6f79db0d..264387517b0 100644 --- a/src/gallium/auxiliary/draw/draw_vs_aos.h +++ b/src/gallium/auxiliary/draw/draw_vs_aos.h @@ -87,9 +87,10 @@ struct lit_info { #define MAX_SHINE_TAB 4 #define MAX_LIT_INFO 16 -struct aos_attrib { - const void *input_ptr; - unsigned input_stride; +struct aos_buffer { + const void *base_ptr; + unsigned stride; + void *ptr; /* updated per vertex */ }; @@ -123,7 +124,7 @@ struct aos_machine { const float (*immediates)[4]; /* points to shader data */ const float (*constants)[4]; /* points to draw data */ - const struct aos_attrib *attrib; /* points to ? */ + const struct aos_buffer *buffer; /* points to ? */ }; @@ -175,12 +176,15 @@ void aos_adopt_xmm_reg( struct aos_compilation *cp, unsigned idx, unsigned dirty ); +void aos_spill_all( struct aos_compilation *cp ); + struct x86_reg aos_get_shader_reg( struct aos_compilation *cp, unsigned file, unsigned idx ); -boolean aos_fetch_inputs( struct aos_compilation *cp, - boolean linear ); +boolean aos_init_inputs( struct aos_compilation *cp, boolean linear ); +boolean aos_fetch_inputs( struct aos_compilation *cp, boolean linear ); +boolean aos_incr_inputs( struct aos_compilation *cp, boolean linear ); boolean aos_emit_outputs( struct aos_compilation *cp ); @@ -210,7 +214,7 @@ do { \ #define X86_NULL 0 #define X86_IMMEDIATES 1 #define X86_CONSTANTS 2 -#define X86_ATTRIBS 3 +#define X86_BUFFERS 3 struct x86_reg aos_get_x86( struct aos_compilation *cp, unsigned which_reg, @@ -232,7 +236,8 @@ struct draw_vs_varient_aos_sse { struct draw_vs_varient base; struct draw_context *draw; - struct aos_attrib *attrib; + struct aos_buffer *buffer; + unsigned nr_vb; vaos_run_linear_func gen_run_linear; vaos_run_elts_func gen_run_elts; diff --git a/src/gallium/auxiliary/draw/draw_vs_aos_io.c b/src/gallium/auxiliary/draw/draw_vs_aos_io.c index 26297c74f82..dd79bc799aa 100644 --- a/src/gallium/auxiliary/draw/draw_vs_aos_io.c +++ b/src/gallium/auxiliary/draw/draw_vs_aos_io.c @@ -54,6 +54,7 @@ static void emit_load_R32G32B32( struct aos_compilation *cp, struct x86_reg data, struct x86_reg src_ptr ) { +#if 1 sse_movss(cp->func, data, x86_make_disp(src_ptr, 8)); /* data = z ? ? ? */ sse_shufps(cp->func, data, aos_get_internal_xmm( cp, IMM_IDENTITY ), SHUF(X,Y,Z,W) ); @@ -62,6 +63,16 @@ static void emit_load_R32G32B32( struct aos_compilation *cp, /* data = ? 0 z 1 */ sse_movlps(cp->func, data, src_ptr); /* data = x y z 1 */ +#else + sse_movups(cp->func, data, src_ptr); + /* data = x y z ? */ + sse2_pshufd(cp->func, data, data, SHUF(W,X,Y,Z) ); + /* data = ? x y z */ + sse_movss(cp->func, data, aos_get_internal_xmm( cp, IMM_ONES ) ); + /* data = 1 x y z */ + sse2_pshufd(cp->func, data, data, SHUF(Y,Z,W,X) ); + /* data = x y z 1 */ +#endif } static void emit_load_R32G32( struct aos_compilation *cp, @@ -95,28 +106,6 @@ static void emit_load_R8G8B8A8_UNORM( struct aos_compilation *cp, -static void get_src_ptr( struct aos_compilation *cp, - struct x86_reg src, - struct x86_reg elt, - unsigned a ) -{ - struct x86_reg attrib = x86_make_disp(aos_get_x86( cp, 0, X86_ATTRIBS ), - a * sizeof(struct aos_attrib)); - - struct x86_reg input_ptr = x86_make_disp(attrib, - Offset(struct aos_attrib, input_ptr)); - - struct x86_reg input_stride = x86_make_disp(attrib, - Offset(struct aos_attrib, input_stride)); - - /* Calculate pointer to current attrib: - */ - x86_mov(cp->func, src, input_stride); - x86_imul(cp->func, src, elt); - x86_add(cp->func, src, input_ptr); -} - - /* Extended swizzles? Maybe later. */ static void emit_swizzle( struct aos_compilation *cp, @@ -128,22 +117,60 @@ static void emit_swizzle( struct aos_compilation *cp, } + +static boolean get_buffer_ptr( struct aos_compilation *cp, + boolean linear, + unsigned buf_idx, + struct x86_reg elt, + struct x86_reg ptr) +{ + struct x86_reg buf = x86_make_disp(aos_get_x86( cp, 0, X86_BUFFERS ), + buf_idx * sizeof(struct aos_buffer)); + + struct x86_reg buf_stride = x86_make_disp(buf, + Offset(struct aos_buffer, stride)); + if (linear) { + struct x86_reg buf_ptr = x86_make_disp(buf, + Offset(struct aos_buffer, ptr)); + + + /* Calculate pointer to current attrib: + */ + x86_mov(cp->func, ptr, buf_ptr); + x86_mov(cp->func, elt, buf_stride); + x86_add(cp->func, elt, ptr); + if (buf_idx == 0) sse_prefetchnta(cp->func, x86_make_disp(elt, 192)); + x86_mov(cp->func, buf_ptr, elt); + } + else { + struct x86_reg buf_base_ptr = x86_make_disp(buf, + Offset(struct aos_buffer, base_ptr)); + + + /* Calculate pointer to current attrib: + */ + x86_mov(cp->func, ptr, buf_stride); + x86_imul(cp->func, ptr, elt); + x86_add(cp->func, ptr, buf_base_ptr); + } + + cp->insn_counter++; + + return TRUE; +} + + static boolean load_input( struct aos_compilation *cp, unsigned idx, - boolean linear ) + struct x86_reg bufptr ) { unsigned format = cp->vaos->base.key.element[idx].in.format; - struct x86_reg src = cp->tmp_EAX; + unsigned offset = cp->vaos->base.key.element[idx].in.offset; struct x86_reg dataXMM = aos_get_xmm_reg(cp); /* Figure out source pointer address: */ - get_src_ptr(cp, - src, - linear ? cp->idx_EBX : x86_deref(cp->idx_EBX), - idx); - - src = x86_deref(src); + struct x86_reg src = x86_make_disp(bufptr, offset); aos_adopt_xmm_reg( cp, dataXMM, @@ -179,20 +206,128 @@ static boolean load_input( struct aos_compilation *cp, return TRUE; } - -boolean aos_fetch_inputs( struct aos_compilation *cp, boolean linear ) +static boolean load_inputs( struct aos_compilation *cp, + unsigned buffer, + struct x86_reg ptr ) { unsigned i; - + for (i = 0; i < cp->vaos->base.key.nr_inputs; i++) { - if (!load_input( cp, i, linear )) - return FALSE; - cp->insn_counter++; + if (cp->vaos->base.key.element[i].in.buffer == buffer) { + + if (!load_input( cp, i, ptr )) + return FALSE; + + cp->insn_counter++; + } + } + + return TRUE; +} + +boolean aos_init_inputs( struct aos_compilation *cp, boolean linear ) +{ + unsigned i; + for (i = 0; i < cp->vaos->nr_vb; i++) { + struct x86_reg buf = x86_make_disp(aos_get_x86( cp, 0, X86_BUFFERS ), + i * sizeof(struct aos_buffer)); + + struct x86_reg buf_base_ptr = x86_make_disp(buf, + Offset(struct aos_buffer, base_ptr)); + + if (cp->vaos->base.key.const_vbuffers & (1<<i)) { + struct x86_reg ptr = cp->tmp_EAX; + + x86_mov(cp->func, ptr, buf_base_ptr); + + /* Load all inputs for this constant vertex buffer + */ + load_inputs( cp, i, x86_deref(ptr) ); + + /* Then just force them out to aos_machine.input[] + */ + aos_spill_all( cp ); + + } + else if (linear) { + + struct x86_reg elt = cp->idx_EBX; + struct x86_reg ptr = cp->tmp_EAX; + + struct x86_reg buf_stride = x86_make_disp(buf, + Offset(struct aos_buffer, stride)); + + struct x86_reg buf_ptr = x86_make_disp(buf, + Offset(struct aos_buffer, ptr)); + + + /* Calculate pointer to current attrib: + */ + x86_mov(cp->func, ptr, buf_stride); + x86_imul(cp->func, ptr, elt); + x86_add(cp->func, ptr, buf_base_ptr); + + + /* In the linear case, keep the buffer pointer instead of the + * index number. + */ + if (cp->vaos->nr_vb == 1) + x86_mov( cp->func, elt, ptr ); + else + x86_mov( cp->func, buf_ptr, ptr ); + + cp->insn_counter++; + } + } + + return TRUE; +} + +boolean aos_fetch_inputs( struct aos_compilation *cp, boolean linear ) +{ + unsigned j; + + for (j = 0; j < cp->vaos->nr_vb; j++) { + if (cp->vaos->base.key.const_vbuffers & (1<<j)) { + /* just retreive pre-transformed input */ + } + else if (linear && cp->vaos->nr_vb == 1) { + load_inputs( cp, 0, cp->idx_EBX ); + } + else { + struct x86_reg elt = linear ? cp->idx_EBX : x86_deref(cp->idx_EBX); + struct x86_reg ptr = cp->tmp_EAX; + + if (!get_buffer_ptr( cp, linear, j, elt, ptr )) + return FALSE; + + if (!load_inputs( cp, j, ptr )) + return FALSE; + } } return TRUE; } +boolean aos_incr_inputs( struct aos_compilation *cp, boolean linear ) +{ + if (linear && cp->vaos->nr_vb == 1) { + struct x86_reg stride = x86_make_disp(aos_get_x86( cp, 0, X86_BUFFERS ), + (0 * sizeof(struct aos_buffer) + + Offset(struct aos_buffer, stride))); + + x86_add(cp->func, cp->idx_EBX, stride); + sse_prefetchnta(cp->func, x86_make_disp(cp->idx_EBX, 192)); + } + else if (linear) { + /* Nothing to do */ + } + else { + x86_lea(cp->func, cp->idx_EBX, x86_make_disp(cp->idx_EBX, 4)); + } + + return TRUE; +} @@ -203,7 +338,7 @@ static void emit_store_R32G32B32A32( struct aos_compilation *cp, struct x86_reg dst_ptr, struct x86_reg dataXMM ) { - sse_movups(cp->func, dst_ptr, dataXMM); + sse_movaps(cp->func, dst_ptr, dataXMM); } static void emit_store_R32G32B32( struct aos_compilation *cp, @@ -306,7 +441,7 @@ boolean aos_emit_outputs( struct aos_compilation *cp ) if (data.file != file_XMM) { struct x86_reg tmp = aos_get_xmm_reg( cp ); - sse_movups(cp->func, tmp, data); + sse_movaps(cp->func, tmp, data); data = tmp; } diff --git a/src/gallium/auxiliary/draw/draw_vs_llvm.c b/src/gallium/auxiliary/draw/draw_vs_llvm.c index 2ce30b9a02b..727977bc3af 100644 --- a/src/gallium/auxiliary/draw/draw_vs_llvm.c +++ b/src/gallium/auxiliary/draw/draw_vs_llvm.c @@ -32,6 +32,7 @@ * Brian Paul */ +#include "util/u_memory.h" #include "pipe/p_shader_tokens.h" #include "draw_private.h" #include "draw_context.h" diff --git a/src/gallium/auxiliary/draw/draw_vs_sse.c b/src/gallium/auxiliary/draw/draw_vs_sse.c index 0efabd9de8b..b11ae316627 100644 --- a/src/gallium/auxiliary/draw/draw_vs_sse.c +++ b/src/gallium/auxiliary/draw/draw_vs_sse.c @@ -37,7 +37,7 @@ #include "draw_vs.h" -#if defined(PIPE_ARCH_X86) +#if defined(PIPE_ARCH_X86) && defined(PIPE_ARCH_SSE) #include "pipe/p_shader_tokens.h" diff --git a/src/gallium/auxiliary/draw/draw_vs_varient.c b/src/gallium/auxiliary/draw/draw_vs_varient.c index 4daf05dae7c..7ee567d4789 100644 --- a/src/gallium/auxiliary/draw/draw_vs_varient.c +++ b/src/gallium/auxiliary/draw/draw_vs_varient.c @@ -64,10 +64,10 @@ struct draw_vs_varient_generic { -static void vsvg_set_input( struct draw_vs_varient *varient, - unsigned buffer, - const void *ptr, - unsigned stride ) +static void vsvg_set_buffer( struct draw_vs_varient *varient, + unsigned buffer, + const void *ptr, + unsigned stride ) { struct draw_vs_varient_generic *vsvg = (struct draw_vs_varient_generic *)varient; @@ -265,7 +265,7 @@ struct draw_vs_varient *draw_vs_varient_generic( struct draw_vertex_shader *vs, vsvg->base.key = *key; vsvg->base.vs = vs; - vsvg->base.set_input = vsvg_set_input; + vsvg->base.set_buffer = vsvg_set_buffer; vsvg->base.run_elts = vsvg_run_elts; vsvg->base.run_linear = vsvg_run_linear; vsvg->base.destroy = vsvg_destroy; diff --git a/src/gallium/auxiliary/gallivm/gallivm_builtins.cpp b/src/gallium/auxiliary/gallivm/gallivm_builtins.cpp index 0fc5c4ec5ca..fcc5c05794a 100644 --- a/src/gallium/auxiliary/gallivm/gallivm_builtins.cpp +++ b/src/gallium/auxiliary/gallivm/gallivm_builtins.cpp @@ -1,140 +1,140 @@ static const unsigned char llvm_builtins_data[] = { -0x42,0x43,0xc0,0xde,0x21,0x0c,0x00,0x00,0x29,0x02,0x00,0x00,0x01,0x10,0x00,0x00, +0x42,0x43,0xc0,0xde,0x21,0x0c,0x00,0x00,0x27,0x02,0x00,0x00,0x01,0x10,0x00,0x00, 0x10,0x00,0x00,0x00,0x07,0x81,0x23,0x91,0x41,0xc8,0x04,0x49,0x06,0x10,0x32,0x39, 0x92,0x01,0x84,0x0c,0x25,0x05,0x08,0x19,0x1e,0x04,0x8b,0x62,0x80,0x14,0x45,0x02, 0x42,0x92,0x0b,0x42,0xa4,0x10,0x32,0x14,0x38,0x08,0x18,0x49,0x0a,0x32,0x44,0x24, 0x48,0x0a,0x90,0x21,0x23,0x44,0x72,0x80,0x8c,0x14,0x21,0x86,0x0a,0x8a,0x0a,0x64, -0x0c,0x1f,0x00,0x00,0x49,0x18,0x00,0x00,0x02,0x00,0x00,0x00,0x0b,0x04,0x00,0x0c, -0x00,0x00,0x00,0x00,0x51,0x20,0x00,0x00,0x12,0x00,0x00,0x00,0x32,0x22,0x48,0x09, -0x20,0x65,0x82,0x84,0x00,0x26,0x45,0x48,0x05,0x09,0x26,0x45,0xc6,0x05,0x42,0x52, -0x26,0x08,0xae,0x19,0x80,0x61,0x04,0x02,0x98,0x23,0x00,0x83,0x29,0x80,0x21,0x00, -0xb2,0x73,0x04,0x01,0x51,0x8a,0xf4,0x08,0x92,0xa4,0x39,0x47,0x80,0x50,0x2b,0x03, -0x00,0xa0,0x08,0x21,0x5c,0x46,0x2b,0x44,0x08,0x21,0xd4,0x40,0x14,0x01,0x80,0x11, -0x80,0x22,0x88,0x00,0x13,0xa2,0x74,0xb0,0x03,0x3c,0xb0,0x83,0x36,0x80,0x87,0x71, -0x68,0x03,0x76,0x48,0x07,0x77,0xa8,0x07,0x7c,0x68,0x83,0x73,0x70,0x87,0x7a,0xd8, -0x70,0x0f,0xe5,0xd0,0x06,0xf0,0xa0,0x07,0x73,0x20,0x07,0x7a,0x30,0x07,0x72,0xa0, -0x07,0x73,0x20,0x07,0x6d,0x90,0x0e,0x71,0xa0,0x07,0x78,0xa0,0x07,0x78,0xd0,0x06, -0xe9,0x80,0x07,0x7a,0x80,0x07,0x7a,0x80,0x07,0x6d,0x90,0x0e,0x71,0x60,0x07,0x7a, -0x10,0x07,0x76,0xa0,0x07,0x71,0x60,0x07,0x6d,0x90,0x0e,0x73,0x20,0x07,0x7a,0x30, -0x07,0x72,0xa0,0x07,0x73,0x20,0x07,0x6d,0x90,0x0e,0x76,0x40,0x07,0x7a,0x30,0x07, -0x72,0xa0,0x07,0x76,0x40,0x07,0x6d,0x60,0x0e,0x73,0x20,0x07,0x7a,0x30,0x07,0x72, -0xa0,0x07,0x73,0x20,0x07,0x6d,0x60,0x0e,0x76,0x40,0x07,0x7a,0x30,0x07,0x72,0xa0, -0x07,0x76,0x40,0x07,0x6d,0x60,0x0f,0x76,0x40,0x07,0x7a,0x60,0x07,0x74,0xa0,0x07, -0x76,0x40,0x07,0x6d,0x60,0x0f,0x71,0x20,0x07,0x78,0xa0,0x07,0x71,0x20,0x07,0x78, -0xa0,0x07,0x71,0x20,0x07,0x78,0xd0,0x06,0xe1,0x00,0x07,0x7a,0x00,0x07,0x7a,0x60, -0x07,0x74,0xd0,0x06,0xe6,0x80,0x07,0x70,0xa0,0x07,0x71,0x20,0x07,0x78,0xa0,0x07, -0x71,0x20,0x07,0x78,0xa0,0xf3,0x40,0x88,0x04,0x32,0x32,0x02,0x04,0x20,0x76,0x46, -0xfc,0x6c,0x48,0x92,0x00,0x40,0x00,0x00,0x00,0x00,0x0c,0x49,0x12,0x20,0x00,0x00, -0x00,0x00,0x80,0x21,0x89,0x02,0x00,0x01,0x00,0x00,0x00,0x30,0x24,0x59,0x00,0x20, -0x08,0x00,0x00,0x00,0x86,0x24,0x0a,0x00,0x04,0x00,0x00,0x00,0xc0,0x90,0x84,0x01, -0x02,0x00,0x00,0x00,0x00,0x18,0x92,0x1c,0x40,0x00,0x00,0x00,0x00,0x00,0x43,0x12, -0x05,0x00,0x02,0x00,0x00,0x00,0x60,0x48,0x72,0x00,0x01,0x00,0x00,0x00,0x00,0x0c, -0x49,0x14,0x00,0x08,0x00,0x00,0x00,0x80,0x21,0x49,0x01,0x00,0x41,0x00,0x00,0x00, -0x90,0x05,0x02,0x00,0x10,0x00,0x00,0x00,0x32,0x1e,0x98,0x10,0x19,0x11,0x4c,0x90, +0x0c,0x1f,0x00,0x00,0x49,0x18,0x00,0x00,0x03,0x00,0x00,0x00,0x0b,0x84,0xff,0xff, +0xff,0xff,0x1f,0xc0,0x00,0x00,0x00,0x00,0x51,0x20,0x00,0x00,0x12,0x00,0x00,0x00, +0x32,0x22,0x48,0x09,0x20,0x65,0x82,0x84,0x00,0x26,0x45,0x48,0x05,0x09,0x26,0x45, +0xc6,0x05,0x42,0x52,0x26,0x08,0xae,0x19,0x80,0x61,0x04,0x02,0x98,0x23,0x00,0x83, +0x29,0x80,0x21,0x00,0xb2,0x73,0x04,0x01,0x51,0x8a,0xf4,0x08,0x92,0xa4,0x39,0x47, +0x80,0x50,0x2b,0x03,0x00,0xa0,0x08,0x21,0x5c,0x46,0x2b,0x44,0x08,0x21,0xd4,0x40, +0x14,0x01,0x80,0x11,0x80,0x22,0x88,0x00,0x13,0x30,0x7c,0xc0,0x03,0x3b,0xf8,0x05, +0x3b,0xa0,0x83,0x36,0xa8,0x07,0x77,0x58,0x07,0x77,0x78,0x87,0x7b,0x70,0x87,0x36, +0x60,0x87,0x74,0x70,0x87,0x7a,0xc0,0x87,0x36,0x38,0x07,0x77,0xa8,0x87,0x0d,0xf7, +0x50,0x0e,0x6d,0x00,0x0f,0x7a,0x60,0x07,0x74,0xa0,0x07,0x76,0x40,0x07,0x7a,0x60, +0x07,0x74,0xd0,0x06,0xe9,0x10,0x07,0x7a,0x80,0x07,0x7a,0x80,0x07,0x6d,0x90,0x0e, +0x78,0xa0,0x07,0x78,0xa0,0x07,0x78,0xd0,0x06,0xe9,0x10,0x07,0x76,0xa0,0x07,0x71, +0x60,0x07,0x7a,0x10,0x07,0x76,0xd0,0x06,0xe9,0x30,0x07,0x72,0xa0,0x07,0x73,0x20, +0x07,0x7a,0x30,0x07,0x72,0xd0,0x06,0xe9,0x60,0x07,0x74,0xa0,0x07,0x76,0x40,0x07, +0x7a,0x60,0x07,0x74,0xd0,0x06,0xe6,0x30,0x07,0x72,0xa0,0x07,0x73,0x20,0x07,0x7a, +0x30,0x07,0x72,0xd0,0x06,0xe6,0x60,0x07,0x74,0xa0,0x07,0x76,0x40,0x07,0x7a,0x60, +0x07,0x74,0xd0,0x06,0xf6,0x60,0x07,0x74,0xa0,0x07,0x76,0x40,0x07,0x7a,0x60,0x07, +0x74,0xd0,0x06,0xf6,0x10,0x07,0x72,0x80,0x07,0x7a,0x10,0x07,0x72,0x80,0x07,0x7a, +0x10,0x07,0x72,0x80,0x07,0x6d,0x10,0x0e,0x70,0xa0,0x07,0x70,0xa0,0x07,0x76,0x40, +0x07,0x6d,0x60,0x0e,0x78,0x00,0x07,0x7a,0x10,0x07,0x72,0x80,0x07,0x7a,0x10,0x07, +0x72,0x80,0x07,0x3a,0x0f,0x84,0x48,0x20,0x23,0x24,0x40,0x00,0x62,0x67,0x88,0x9f, +0x19,0x92,0x24,0x00,0x10,0x04,0x00,0x00,0x00,0x43,0x92,0x04,0x08,0x00,0x00,0x00, +0x00,0x60,0x48,0xa2,0x00,0x40,0x10,0x00,0x00,0x00,0x0c,0x49,0x16,0x00,0x08,0x02, +0x00,0x00,0x80,0x21,0x89,0x02,0x00,0x41,0x00,0x00,0x00,0x30,0x24,0x61,0x80,0x00, +0x00,0x00,0x00,0x00,0x86,0x24,0x07,0x10,0x00,0x00,0x00,0x00,0xc0,0x90,0x44,0x01, +0x80,0x20,0x00,0x00,0x00,0x18,0x92,0x1c,0x40,0x00,0x00,0x00,0x00,0x00,0x43,0x12, +0x05,0x00,0x82,0x00,0x00,0x00,0x60,0x48,0x52,0x00,0x40,0x10,0x00,0x00,0x00,0x64, +0x81,0x00,0x00,0x00,0x10,0x00,0x00,0x00,0x32,0x1e,0x98,0x10,0x19,0x11,0x4c,0x90, 0x8c,0x09,0x26,0x47,0xc6,0x04,0x43,0x8a,0x8a,0x59,0x8b,0x43,0x50,0xd2,0x09,0x02, 0x81,0xd2,0x73,0x50,0xc9,0x0c,0x2a,0x99,0x41,0x25,0x33,0xa8,0x64,0x56,0x28,0x66, 0x2d,0x0e,0x41,0xcf,0x2a,0x15,0x04,0x4a,0xcf,0x41,0x25,0x33,0xa8,0x64,0x06,0x95, 0xcc,0xa0,0x92,0x59,0x01,0x00,0x00,0x00,0x53,0x82,0x26,0x0c,0x04,0x00,0x00,0x00, 0x22,0x00,0x00,0x00,0x13,0x04,0x41,0x2c,0x10,0x00,0x00,0x00,0x05,0x00,0x00,0x00, 0x04,0xc6,0x08,0x40,0x10,0x04,0xe1,0x70,0x18,0x23,0x00,0x41,0x10,0x84,0xc3,0x60, -0x04,0x00,0x00,0x00,0x93,0x0c,0xce,0x43,0x4c,0x31,0x3c,0x8e,0x34,0xc9,0x30,0x41, -0xc2,0x14,0x03,0x34,0x51,0x93,0x0c,0x4d,0x44,0x4c,0x31,0x44,0x8d,0x35,0x56,0x01, -0x04,0xc3,0x55,0x21,0x16,0x0e,0x04,0x00,0x0f,0x00,0x00,0x00,0x46,0x41,0x08,0xcc, -0x73,0x9b,0x05,0x21,0x30,0xcf,0x6e,0x18,0x84,0x00,0x2c,0x8b,0x35,0x04,0x80,0x39, -0x04,0x81,0x5d,0x20,0x80,0x0f,0x0c,0x43,0xe4,0xd3,0x36,0x81,0x04,0x3e,0x30,0x0c, -0x91,0x4f,0x5b,0x05,0x12,0xf8,0xc0,0x30,0x44,0x7e,0x7d,0x00,0x05,0xd1,0x4c,0x11, -0x66,0x12,0x83,0xc0,0x3c,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x61,0x20,0x00,0x00, -0x2a,0x00,0x00,0x00,0x13,0x04,0x43,0x2c,0x10,0x00,0x00,0x00,0x08,0x00,0x00,0x00, +0x04,0x00,0x00,0x00,0xc3,0x0d,0xce,0x43,0x4c,0x37,0x3c,0x8e,0x34,0xdc,0x30,0x41, +0xc2,0x74,0x03,0x34,0x51,0xc3,0x0d,0x4d,0x44,0x4c,0x37,0x44,0x8d,0x35,0x56,0x01, +0x04,0xc3,0x55,0x21,0x16,0x0e,0x04,0x00,0x0f,0x00,0x00,0x00,0xd6,0x10,0x00,0xe6, +0x10,0x04,0x76,0x81,0x00,0x3e,0x30,0x0c,0x91,0x4f,0x1b,0x05,0x21,0x30,0x8f,0x6d, +0x13,0x48,0xe0,0x03,0xc3,0x10,0xf9,0xb4,0x55,0x20,0x81,0x0f,0x0c,0x43,0xe4,0xd7, +0x66,0x41,0x08,0xcc,0xa3,0x1f,0x40,0x41,0x34,0x53,0x84,0x99,0xc4,0x20,0x30,0x8f, +0x61,0x10,0x02,0xb0,0x2c,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x61,0x20,0x00,0x00, +0x27,0x00,0x00,0x00,0x13,0x04,0x43,0x2c,0x10,0x00,0x00,0x00,0x08,0x00,0x00,0x00, 0x24,0x8a,0xa0,0x0c,0x46,0x00,0x4a,0x80,0xc2,0x1c,0x84,0x55,0x55,0xd6,0x1c,0x84, 0x45,0x51,0x16,0x81,0x19,0x80,0x11,0x80,0x31,0x02,0x10,0x04,0x41,0xfc,0x03,0x00, -0x63,0x08,0x0d,0x34,0xc9,0x70,0x55,0xc2,0x2c,0x43,0x20,0x60,0x73,0x0c,0xd3,0x15, +0x63,0x08,0x0d,0x34,0xdc,0x70,0x55,0xc2,0x2c,0x43,0x20,0x60,0x73,0x0c,0xd3,0x15, 0x8d,0x21,0x34,0xd1,0x18,0x42,0xf3,0x8c,0x55,0x00,0x81,0xa0,0x6d,0x73,0x0c,0x19, -0xe7,0x60,0x87,0x52,0x38,0x10,0x00,0x00,0x13,0x00,0x00,0x00,0x17,0x60,0x20,0xc5, -0x74,0x10,0x8d,0x65,0x14,0x13,0xf3,0xd4,0xb4,0x6d,0x14,0x13,0xf3,0xd4,0xb8,0x69, -0x14,0x13,0xf3,0xd4,0xb6,0x75,0x14,0x13,0xf3,0xd4,0xba,0x35,0x0c,0x13,0xf3,0x9c, -0x80,0xe4,0x36,0x48,0x81,0x10,0xc3,0x4a,0x4c,0x54,0xd4,0x6c,0x8b,0x23,0x28,0x76, -0x41,0x4c,0xcc,0xa3,0x1b,0x07,0x21,0x00,0xcb,0x72,0x00,0x05,0xd1,0x4c,0x11,0x66, -0x18,0x83,0xc0,0x3c,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x61,0x20,0x00,0x00, -0x81,0x00,0x00,0x00,0x13,0x04,0x4d,0x2c,0x10,0x00,0x00,0x00,0x04,0x00,0x00,0x00, -0x24,0xca,0x60,0x04,0xa0,0x04,0x8a,0x80,0xc2,0x0c,0x00,0x91,0x11,0x00,0x00,0x00, -0x63,0x08,0x4d,0x64,0x16,0xc1,0x49,0x86,0xab,0x22,0x66,0x19,0x02,0x01,0x1b,0x43, -0x70,0xa2,0x59,0x82,0x61,0x0c,0xe1,0x89,0x66,0x09,0x86,0x81,0x0a,0x20,0x0b,0x34, -0x61,0x8e,0x81,0xda,0xa2,0x31,0x84,0x46,0xb2,0x8e,0xe0,0x24,0x83,0x57,0x11,0xb3, -0x0c,0x44,0xf1,0x8d,0x21,0x38,0xd2,0x2c,0x81,0x31,0x86,0xf0,0x48,0xb3,0x04,0xc6, -0x40,0x05,0x00,0x06,0x44,0x18,0x14,0x73,0x0c,0x9c,0x18,0x48,0x63,0x08,0xcd,0x64, -0x64,0x40,0x70,0x92,0xa1,0x0c,0x2a,0x62,0x96,0xe1,0x40,0xcc,0x60,0x0c,0xc1,0x99, -0x66,0x09,0x92,0x31,0x84,0x67,0x9a,0x25,0x48,0x06,0x2a,0x80,0x33,0x38,0xd0,0x00, -0x99,0x63,0x18,0x83,0x34,0x98,0xc6,0x10,0x1a,0xc8,0xd6,0x80,0xe0,0x24,0x03,0x1b, -0x54,0xc4,0x2c,0x83,0xb2,0xb4,0xc1,0x18,0x82,0x03,0xcd,0x12,0x30,0x63,0x08,0x0f, -0x34,0x4b,0xc0,0x0c,0x54,0x00,0x6e,0xa0,0xbc,0xc1,0x32,0xc7,0xa0,0x06,0x70,0x00, -0x61,0x1c,0x84,0x03,0x01,0x00,0x00,0x00,0x4e,0x00,0x00,0x00,0x76,0x52,0x4c,0xcc, -0x73,0xd3,0x24,0x05,0x64,0xec,0xcd,0x8d,0xcc,0xe5,0x87,0x46,0xc6,0x50,0x8a,0x89, -0x79,0xee,0xdb,0x54,0x8a,0x89,0x79,0xee,0xdd,0x1a,0x88,0x89,0x79,0x68,0x73,0x20, -0x26,0xe6,0xa9,0xed,0x81,0x98,0x98,0xc7,0x36,0x0b,0x62,0x62,0x9e,0xdb,0x32,0x88, -0x89,0x79,0x72,0xd3,0x20,0x26,0xe6,0xd9,0x8d,0x83,0x98,0x98,0xa7,0xb7,0x95,0x62, -0x62,0x9e,0xbb,0x27,0x2d,0x20,0x63,0x6f,0x6e,0x64,0x2e,0x3a,0x34,0x35,0x56,0x62, -0x08,0x4e,0x53,0xd9,0xba,0xb5,0x14,0x02,0xf3,0xe0,0xf5,0x25,0x2c,0x82,0xd3,0x0c, -0xbe,0xe0,0x34,0xd3,0x8d,0x9b,0x88,0x21,0x38,0xcd,0x60,0xd7,0x24,0x01,0x63,0xec, -0xcd,0x8d,0xcc,0x45,0x87,0x44,0x80,0x8c,0xbd,0xb9,0x91,0xb9,0xfc,0xc4,0xd0,0x90, -0x02,0x8c,0xb1,0x37,0x37,0x32,0x97,0x1f,0x73,0x29,0x26,0xe6,0xc1,0x71,0x7b,0x29, -0x26,0xe6,0xc1,0x77,0xfb,0x28,0x04,0xe6,0xa9,0x6f,0x52,0x01,0x32,0xf6,0xe6,0x46, -0xe6,0xa2,0x13,0x73,0x63,0x18,0x83,0xc0,0x3c,0xb6,0x41,0x08,0x4e,0x33,0x58,0x47, -0x31,0x31,0x4f,0x5d,0x1f,0xc3,0x22,0x38,0xcd,0xe0,0x0b,0x4e,0x33,0xe1,0xbc,0xa5, -0x18,0x82,0xd3,0x0c,0x77,0x6e,0x20,0xc5,0xc4,0x3c,0xb5,0x4e,0x3a,0x40,0xc6,0xde, -0xdc,0xc8,0x5c,0x7e,0x64,0x70,0x2c,0xa4,0x98,0x98,0xa7,0xee,0xed,0x82,0x10,0x9c, -0xa6,0xba,0x81,0x44,0x70,0x9a,0xc1,0x17,0x9c,0x66,0x32,0x93,0x42,0x60,0x1e,0x7b, -0xb7,0x98,0x62,0x62,0x9e,0xbc,0x36,0x16,0x43,0x70,0x9a,0x0a,0xa7,0x6d,0xa4,0x98, -0x98,0xc7,0xbe,0x8d,0xa4,0x98,0x98,0xc7,0xce,0x0d,0xc6,0x10,0x9c,0x66,0xc0,0x7b, -0x12,0x02,0x32,0xf6,0xe6,0x46,0xe6,0xa2,0x33,0x13,0x73,0x06,0x8b,0xe0,0x34,0x83, -0x2f,0x38,0xcd,0x64,0xd3,0xe6,0x61,0x08,0x4e,0x53,0xd5,0xf6,0x01,0x14,0x44,0x33, -0x45,0x18,0x00,0x00,0x00,0x00,0x00,0x00,0x61,0x20,0x00,0x00,0x4a,0x00,0x00,0x00, -0x13,0x04,0x41,0x2c,0x10,0x00,0x00,0x00,0x07,0x00,0x00,0x00,0x24,0xca,0x60,0x04, -0xa0,0x04,0x8a,0x80,0xc2,0x0c,0x00,0xb9,0x61,0x0c,0x04,0x10,0x1e,0xe1,0x19,0xc6, -0x40,0x02,0xe1,0x11,0x1e,0x00,0x00,0x00,0x63,0x08,0xcd,0x63,0x15,0xc1,0x31,0x84, -0x06,0xb2,0x8b,0xe0,0x18,0x42,0x13,0x59,0x46,0x70,0x0c,0xa1,0x71,0x6c,0x23,0x38, -0x16,0x02,0x04,0xc7,0x64,0x61,0x1a,0x37,0x16,0x01,0x04,0x48,0x35,0xc7,0x20,0x79, -0xcf,0x58,0x04,0x10,0x20,0xd5,0x1c,0xc3,0x07,0x06,0xd0,0x58,0x04,0x10,0x20,0xd5, -0x1c,0x43,0x18,0x88,0x41,0x34,0x16,0x01,0x04,0x48,0x35,0xc7,0x30,0x06,0x64,0xe0, -0x98,0x47,0xd0,0xc0,0x80,0xa0,0x89,0x01,0x41,0x23,0x03,0x82,0x63,0x21,0x40,0x70, -0x50,0x66,0x70,0x06,0x68,0x90,0x06,0x58,0x06,0xe1,0x40,0x00,0x25,0x00,0x00,0x00, -0x56,0x52,0x4c,0xcc,0x73,0xd3,0x56,0x41,0x4c,0xcc,0x53,0xdb,0x05,0x31,0x31,0xcf, -0x6d,0x19,0xc4,0xc4,0x3c,0xba,0x6d,0x10,0x13,0xf3,0xf4,0xd6,0x41,0x08,0xc0,0xb2, -0x18,0x46,0x21,0x38,0x4d,0x85,0x9b,0x46,0x21,0x38,0x4d,0xb5,0x9b,0x8a,0x21,0x00, -0xcb,0x82,0xdf,0x66,0x62,0x08,0x4e,0x53,0xdd,0xb7,0x9d,0x18,0x82,0xd3,0x54,0xb7, -0x6e,0x28,0x86,0xe0,0x34,0xd5,0xdd,0xdb,0x47,0x31,0x31,0x4f,0x9d,0x9b,0x87,0x21, -0x00,0xcb,0x52,0xdf,0x06,0x62,0x08,0xc0,0xb2,0xd4,0xbc,0x59,0x10,0x82,0xd3,0x54, -0x96,0x62,0x08,0x4e,0x53,0xe1,0xb6,0x85,0x14,0x13,0xf3,0xd8,0xb4,0x8d,0x14,0x13, -0xf3,0xd8,0xb9,0x89,0x18,0x02,0xb0,0x2c,0xf6,0x6d,0x24,0x86,0x00,0x2c,0x8b,0xcd, -0x1b,0x87,0x21,0x38,0x4d,0x55,0xd3,0xd6,0x30,0x54,0xc0,0x72,0x00,0x05,0xd1,0x4c, -0x11,0x06,0x00,0x00,0x00,0x00,0x00,0x00,0x61,0x20,0x00,0x00,0x19,0x00,0x00,0x00, -0x13,0x04,0x41,0x2c,0x10,0x00,0x00,0x00,0x03,0x00,0x00,0x00,0x24,0x4a,0x60,0x04, -0x80,0xc2,0x0c,0x00,0x00,0x00,0x00,0x00,0x63,0x08,0xcd,0x33,0x16,0x01,0x04,0x48, -0x34,0xc7,0x00,0x49,0xcf,0x58,0x04,0x10,0x28,0xd1,0x1c,0xc3,0x44,0x39,0x58,0x85, -0x03,0x01,0x00,0x00,0x0a,0x00,0x00,0x00,0x16,0x41,0x4c,0xcc,0x63,0xdb,0x04,0x31, -0x31,0x4f,0x6e,0x0d,0x43,0x05,0x2c,0x07,0x50,0x10,0xcd,0x14,0x61,0x56,0x41,0x4c, -0xcc,0xd3,0x1b,0x45,0x21,0x00,0xcb,0xb2,0x9b,0x04,0x21,0x00,0xcb,0x02,0x00,0x00, -0x00,0x00,0x00,0x00,0x61,0x20,0x00,0x00,0x1b,0x00,0x00,0x00,0x13,0x04,0x41,0x2c, -0x10,0x00,0x00,0x00,0x03,0x00,0x00,0x00,0x24,0xca,0x60,0x04,0xa0,0x04,0x8a,0x80, -0xc2,0x0c,0x00,0x00,0x63,0x08,0xcd,0x33,0x16,0x01,0x04,0xca,0x34,0xc7,0x20,0x51, -0xcf,0x1c,0x43,0x45,0x41,0x73,0x0c,0x16,0x15,0xcd,0x31,0x5c,0x94,0x83,0x58,0x38, -0x10,0x00,0x00,0x00,0x0b,0x00,0x00,0x00,0x86,0x51,0x4c,0xcc,0x53,0xe7,0x76,0x51, -0x4c,0xcc,0x53,0xdb,0x36,0x41,0x4c,0xcc,0x63,0x5b,0x05,0x31,0x31,0x8f,0x6e,0x0d, -0x43,0x05,0x2c,0x66,0x41,0x4c,0xcc,0xd3,0x1f,0x40,0x41,0x34,0x53,0x84,0x19,0x05, -0x21,0x00,0xcb,0x02,0x00,0x00,0x00,0x00,0x61,0x20,0x00,0x00,0x2f,0x00,0x00,0x00, -0x13,0x04,0x45,0x2c,0x10,0x00,0x00,0x00,0x03,0x00,0x00,0x00,0x24,0xca,0xa0,0x04, -0x46,0x00,0x8a,0x80,0xc0,0x08,0x00,0x00,0x63,0x08,0x0d,0x34,0xc9,0x30,0x49,0xc4, -0x2c,0x03,0x11,0x50,0x63,0x08,0xcd,0x33,0xc9,0x50,0x49,0xc4,0x2c,0x03,0x21,0x58, -0x63,0x08,0x4d,0x34,0xc9,0x70,0x49,0xc4,0x2c,0x03,0x31,0x60,0x63,0x08,0x8d,0x33, -0xc9,0x90,0x49,0x84,0x69,0x22,0x70,0xc3,0x27,0x1c,0x08,0x00,0x1a,0x00,0x00,0x00, -0x96,0x51,0x4c,0xcc,0x53,0xdf,0x66,0x41,0x08,0xcc,0x83,0xdb,0x04,0x31,0x31,0x4f, -0x6d,0x15,0xc4,0xc4,0x3c,0xb7,0x61,0x10,0x02,0xf3,0xf0,0x47,0x20,0xb9,0x0d,0x52, -0x20,0xc4,0xb0,0x12,0x13,0x15,0x35,0xdb,0xe2,0x08,0x8a,0x5d,0x10,0x13,0xf3,0xec, -0x37,0x90,0x2c,0x4e,0xf4,0x47,0x87,0x54,0xd7,0x17,0x70,0x2c,0x4e,0xf4,0x47,0x87, -0x74,0x02,0xc8,0xe2,0x44,0x7f,0x74,0x48,0xb9,0x69,0x14,0x02,0xf3,0xd4,0xb8,0x6d, -0x18,0x11,0x31,0x55,0xc0,0x62,0x0d,0x43,0x05,0x2c,0x07,0x50,0x10,0xcd,0x14,0x61, -0x46,0x31,0x08,0xcc,0x03,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x71,0x20,0x00,0x00, -0x12,0x00,0x00,0x00,0x66,0x40,0x54,0x82,0x23,0x59,0xc2,0x20,0x09,0x92,0x1d,0x18, -0x4f,0x84,0x34,0x53,0x61,0x03,0xc4,0xe3,0x58,0x85,0x05,0x14,0xbe,0x34,0x45,0xb5, -0x21,0x10,0x82,0x23,0x15,0x46,0x30,0x2c,0xc8,0x64,0x02,0x06,0xf0,0x3c,0x91,0x73, -0x19,0x00,0xe1,0x4b,0x53,0x64,0x0a,0x84,0x84,0x34,0x85,0x31,0x10,0x0a,0xb2,0x3c, -0x56,0x30,0x08,0xcc,0x63,0x0b,0x44,0x25,0x21,0x0d,0x00,0x00,0x00,0x00,0x00,0x00}; +0xe7,0x60,0x87,0x52,0x38,0x10,0x00,0x00,0x10,0x00,0x00,0x00,0x27,0x50,0x20,0x05, +0xd1,0x0c,0x17,0x60,0x20,0xc5,0x74,0x10,0x8d,0x65,0x14,0x13,0xf3,0xd4,0xb4,0x6d, +0x14,0x13,0xf3,0xd4,0xb8,0x69,0x14,0x13,0xf3,0xd4,0xb6,0x75,0x14,0x13,0xf3,0xd4, +0xba,0x35,0x0c,0x13,0xf3,0xd8,0x05,0x31,0x31,0x8f,0x6e,0x1c,0x84,0x00,0x2c,0xcb, +0x01,0x14,0x44,0x33,0x45,0x98,0x61,0x0c,0x02,0xf3,0x00,0x00,0x00,0x00,0x00,0x00, +0x61,0x20,0x00,0x00,0x81,0x00,0x00,0x00,0x13,0x04,0x4d,0x2c,0x10,0x00,0x00,0x00, +0x04,0x00,0x00,0x00,0x24,0xca,0x60,0x04,0xa0,0x04,0x8a,0x80,0xc2,0x0c,0x00,0x91, +0x11,0x00,0x00,0x00,0x63,0x08,0x4d,0x64,0x16,0xc1,0xe1,0x86,0xab,0x22,0x66,0x19, +0x02,0x01,0x1b,0x43,0x70,0xa2,0x59,0x82,0x61,0x0c,0xe1,0x89,0x66,0x09,0x86,0x81, +0x0a,0x20,0x0b,0x34,0x61,0x8e,0x81,0xda,0xa2,0x31,0x84,0x46,0xb2,0x8e,0xe0,0x70, +0x83,0x57,0x11,0xb3,0x0c,0x44,0xf1,0x8d,0x21,0x38,0xd2,0x2c,0x81,0x31,0x86,0xf0, +0x48,0xb3,0x04,0xc6,0x40,0x05,0x00,0x06,0x44,0x18,0x14,0x73,0x0c,0x9c,0x18,0x48, +0x63,0x08,0xcd,0x64,0x64,0x40,0x70,0xb8,0xa1,0x0c,0x2a,0x62,0x96,0xe1,0x40,0xcc, +0x60,0x0c,0xc1,0x99,0x66,0x09,0x92,0x31,0x84,0x67,0x9a,0x25,0x48,0x06,0x2a,0x80, +0x33,0x38,0xd0,0x00,0x99,0x63,0x18,0x83,0x34,0x98,0xc6,0x10,0x1a,0xc8,0xd6,0x80, +0xe0,0x70,0x03,0x1b,0x54,0xc4,0x2c,0x83,0xb2,0xb4,0xc1,0x18,0x82,0x03,0xcd,0x12, +0x30,0x63,0x08,0x0f,0x34,0x4b,0xc0,0x0c,0x54,0x00,0x6e,0xa0,0xbc,0xc1,0x32,0xc7, +0xa0,0x06,0x70,0x00,0x61,0x1c,0x84,0x03,0x01,0x00,0x00,0x00,0x4e,0x00,0x00,0x00, +0x76,0x52,0x4c,0xcc,0x73,0xd3,0x24,0x05,0x64,0xec,0xcd,0x8d,0xcc,0xe5,0x87,0x46, +0xc6,0x50,0x8a,0x89,0x79,0xee,0xdb,0x54,0x8a,0x89,0x79,0xee,0xdd,0x1a,0x88,0x89, +0x79,0x68,0x73,0x20,0x26,0xe6,0xa9,0xed,0x81,0x98,0x98,0xc7,0x36,0x0b,0x62,0x62, +0x9e,0xdb,0x32,0x88,0x89,0x79,0x72,0xd3,0x20,0x26,0xe6,0xd9,0x8d,0x83,0x98,0x98, +0xa7,0xb7,0x95,0x62,0x62,0x9e,0xbb,0x27,0x2d,0x20,0x63,0x6f,0x6e,0x64,0x2e,0x3a, +0x34,0x35,0x56,0x62,0x08,0x4e,0x53,0xd9,0xba,0xb5,0x14,0x02,0xf3,0xe0,0xf5,0x25, +0x2c,0x82,0xd3,0x0c,0xbe,0xe0,0x34,0xd3,0x8d,0x9b,0x88,0x21,0x38,0xcd,0x60,0xd7, +0x24,0x01,0x63,0xec,0xcd,0x8d,0xcc,0x45,0x87,0x44,0x80,0x8c,0xbd,0xb9,0x91,0xb9, +0xfc,0xc4,0xd0,0x90,0x02,0x8c,0xb1,0x37,0x37,0x32,0x97,0x1f,0x73,0x29,0x26,0xe6, +0xc1,0x71,0x7b,0x29,0x26,0xe6,0xc1,0x77,0xfb,0x28,0x04,0xe6,0xa9,0x6f,0x52,0x01, +0x32,0xf6,0xe6,0x46,0xe6,0xa2,0x13,0x73,0x63,0x18,0x83,0xc0,0x3c,0xb6,0x41,0x08, +0x4e,0x33,0x58,0x47,0x31,0x31,0x4f,0x5d,0x1f,0xc3,0x22,0x38,0xcd,0xe0,0x0b,0x4e, +0x33,0xe1,0xbc,0xa5,0x18,0x82,0xd3,0x0c,0x77,0x6e,0x20,0xc5,0xc4,0x3c,0xb5,0x4e, +0x3a,0x40,0xc6,0xde,0xdc,0xc8,0x5c,0x7e,0x64,0x70,0x2c,0xa4,0x98,0x98,0xa7,0xee, +0x6f,0x20,0x11,0x9c,0x66,0xf0,0x05,0xa7,0x99,0xec,0x82,0x10,0x9c,0xa6,0x32,0x93, +0x42,0x60,0x1e,0x7b,0xb7,0x98,0x62,0x62,0x9e,0xbc,0x36,0x16,0x43,0x70,0x9a,0x0a, +0xa7,0x6d,0xa4,0x98,0x98,0xc7,0xbe,0x8d,0xa4,0x98,0x98,0xc7,0xce,0x0d,0xc6,0x10, +0x9c,0x66,0xc0,0x7b,0x12,0x02,0x32,0xf6,0xe6,0x46,0xe6,0xa2,0x33,0x13,0x73,0x06, +0x8b,0xe0,0x34,0x83,0x2f,0x38,0xcd,0x64,0xd3,0x07,0x50,0x10,0xcd,0x14,0x61,0xe6, +0x61,0x08,0x4e,0x53,0xd5,0x36,0x00,0x00,0x00,0x00,0x00,0x00,0x61,0x20,0x00,0x00, +0x4a,0x00,0x00,0x00,0x13,0x04,0x41,0x2c,0x10,0x00,0x00,0x00,0x07,0x00,0x00,0x00, +0x24,0xca,0x60,0x04,0xa0,0x04,0x8a,0x80,0xc2,0x0c,0x00,0xb9,0x61,0x0c,0x04,0x10, +0x1e,0xe1,0x19,0xc6,0x40,0x02,0xe1,0x11,0x1e,0x00,0x00,0x00,0x63,0x08,0xcd,0x63, +0x15,0xc1,0x31,0x84,0x06,0xb2,0x8b,0xe0,0x18,0x42,0x13,0x59,0x46,0x70,0x0c,0xa1, +0x71,0x6c,0x23,0x38,0x16,0x02,0x04,0xc7,0x64,0x61,0x1a,0x37,0x16,0x01,0x04,0x48, +0x35,0xc7,0x20,0x79,0xcf,0x58,0x04,0x10,0x20,0xd5,0x1c,0xc3,0x07,0x06,0xd0,0x58, +0x04,0x10,0x20,0xd5,0x1c,0x43,0x18,0x88,0x41,0x34,0x16,0x01,0x04,0x48,0x35,0xc7, +0x30,0x06,0x64,0xe0,0x98,0x47,0xd0,0xc0,0x80,0xa0,0x89,0x01,0x41,0x23,0x03,0x82, +0x63,0x21,0x40,0x70,0x50,0x66,0x70,0x06,0x68,0x90,0x06,0x58,0x06,0xe1,0x40,0x00, +0x25,0x00,0x00,0x00,0x56,0x52,0x4c,0xcc,0x73,0xd3,0x56,0x41,0x4c,0xcc,0x53,0xdb, +0x05,0x31,0x31,0xcf,0x6d,0x19,0xc4,0xc4,0x3c,0xba,0x6d,0x10,0x13,0xf3,0xf4,0xd6, +0x41,0x08,0xc0,0xb2,0x18,0x46,0x21,0x38,0x4d,0x85,0x9b,0x46,0x21,0x38,0x4d,0xb5, +0x9b,0x8a,0x21,0x00,0xcb,0x82,0xdf,0x66,0x62,0x08,0x4e,0x53,0xdd,0xb7,0x9d,0x18, +0x82,0xd3,0x54,0xb7,0x6e,0x28,0x86,0xe0,0x34,0xd5,0xdd,0xdb,0x47,0x31,0x31,0x4f, +0x9d,0x9b,0x87,0x21,0x00,0xcb,0x52,0xdf,0x06,0x62,0x08,0xc0,0xb2,0xd4,0xbc,0x59, +0x10,0x82,0xd3,0x54,0x96,0x62,0x08,0x4e,0x53,0xe1,0xb6,0x85,0x14,0x13,0xf3,0xd8, +0xb4,0x8d,0x14,0x13,0xf3,0xd8,0xb9,0x89,0x18,0x02,0xb0,0x2c,0xf6,0x6d,0x24,0x86, +0x00,0x2c,0x8b,0xcd,0x1b,0x87,0x21,0x38,0x4d,0x55,0xd3,0xd6,0x30,0x54,0xc0,0x72, +0x00,0x05,0xd1,0x4c,0x11,0x06,0x00,0x00,0x00,0x00,0x00,0x00,0x61,0x20,0x00,0x00, +0x19,0x00,0x00,0x00,0x13,0x04,0x41,0x2c,0x10,0x00,0x00,0x00,0x03,0x00,0x00,0x00, +0x24,0x4a,0x60,0x04,0x80,0xc2,0x0c,0x00,0x00,0x00,0x00,0x00,0x63,0x08,0xcd,0x33, +0x16,0x01,0x04,0x48,0x34,0xc7,0x00,0x49,0xcf,0x58,0x04,0x10,0x28,0xd1,0x1c,0xc3, +0x44,0x39,0x58,0x85,0x03,0x01,0x00,0x00,0x0a,0x00,0x00,0x00,0x26,0x41,0x08,0xc0, +0xb2,0x18,0x45,0x21,0x00,0xcb,0xb2,0x5b,0x04,0x31,0x31,0x8f,0x6d,0x13,0xc4,0xc4, +0x3c,0xb9,0x35,0x0c,0x15,0xb0,0x58,0x05,0x31,0x31,0x4f,0x7f,0x00,0x05,0xd1,0x4c, +0x11,0x06,0x00,0x00,0x00,0x00,0x00,0x00,0x61,0x20,0x00,0x00,0x1b,0x00,0x00,0x00, +0x13,0x04,0x41,0x2c,0x10,0x00,0x00,0x00,0x03,0x00,0x00,0x00,0x24,0xca,0x60,0x04, +0xa0,0x04,0x8a,0x80,0xc2,0x0c,0x00,0x00,0x63,0x08,0xcd,0x33,0x16,0x01,0x04,0xca, +0x34,0xc7,0x20,0x51,0xcf,0x1c,0x43,0x45,0x41,0x73,0x0c,0x16,0x15,0xcd,0x31,0x5c, +0x94,0x83,0x58,0x38,0x10,0x00,0x00,0x00,0x0b,0x00,0x00,0x00,0x76,0x51,0x4c,0xcc, +0x53,0xdb,0x86,0x51,0x4c,0xcc,0x53,0xe7,0x36,0x41,0x4c,0xcc,0x63,0x5b,0x05,0x31, +0x31,0x8f,0x6e,0x16,0xc4,0xc4,0x3c,0xbd,0x51,0x10,0x02,0xb0,0x2c,0xd6,0x30,0x54, +0xc0,0x72,0x00,0x05,0xd1,0x4c,0x11,0x06,0x00,0x00,0x00,0x00,0x61,0x20,0x00,0x00, +0x2c,0x00,0x00,0x00,0x13,0x04,0x45,0x2c,0x10,0x00,0x00,0x00,0x03,0x00,0x00,0x00, +0x24,0xca,0xa0,0x04,0x46,0x00,0x8a,0x80,0xc0,0x08,0x00,0x00,0x63,0x08,0x0d,0x34, +0xdc,0x30,0x49,0xc4,0x2c,0x03,0x11,0x50,0x63,0x08,0xcd,0x33,0xdc,0x50,0x49,0xc4, +0x2c,0x03,0x21,0x58,0x63,0x08,0x4d,0x34,0xdc,0x70,0x49,0xc4,0x2c,0x03,0x31,0x60, +0x63,0x08,0x8d,0x33,0xdc,0x90,0x49,0x84,0x69,0x22,0x70,0xc3,0x27,0x1c,0x08,0x00, +0x17,0x00,0x00,0x00,0x96,0x51,0x4c,0xcc,0x53,0xdf,0x66,0x41,0x08,0xcc,0x83,0xdb, +0x04,0x31,0x31,0x4f,0x6d,0x15,0xc4,0xc4,0x3c,0xb7,0x61,0x10,0x02,0xf3,0xf0,0x76, +0x41,0x4c,0xcc,0xb3,0x1f,0x81,0x11,0x11,0x13,0x15,0x35,0x37,0x90,0x2c,0x4e,0xf4, +0x47,0x87,0x54,0xd7,0x17,0x70,0x2c,0x4e,0xf4,0x47,0x87,0x74,0x02,0xc8,0xe2,0x44, +0x7f,0x74,0x48,0xb9,0x69,0x14,0x02,0xf3,0xd4,0xb8,0x6d,0x18,0x11,0x31,0x55,0xc0, +0x62,0x0d,0x43,0x05,0x2c,0x07,0x50,0x10,0xcd,0x14,0x61,0x46,0x31,0x08,0xcc,0x03, +0x00,0x00,0x00,0x00,0x71,0x20,0x00,0x00,0x12,0x00,0x00,0x00,0x66,0x40,0x54,0x82, +0x23,0x19,0xc3,0xa0,0x20,0x8b,0x1d,0x18,0x4f,0x84,0x34,0x53,0x61,0x03,0xc4,0xe3, +0x58,0x85,0x05,0x14,0xbe,0x34,0x45,0xb5,0x21,0x10,0x82,0x23,0x15,0x46,0x30,0x2c, +0xc8,0x64,0x02,0x06,0xf0,0x3c,0x91,0x73,0x19,0x00,0xe1,0x4b,0x53,0x64,0x0a,0x84, +0x84,0x34,0x85,0x25,0x0c,0x92,0x20,0x59,0xc1,0x20,0x30,0x8f,0x2d,0x10,0x95,0x84, +0x34,0x00,0x00,0x00,0x00,0x00,0x00,0x00}; diff --git a/src/gallium/auxiliary/gallivm/gallivm_cpu.cpp b/src/gallium/auxiliary/gallivm/gallivm_cpu.cpp index e64bfb1c6cb..3a2f2878a30 100644 --- a/src/gallium/auxiliary/gallivm/gallivm_cpu.cpp +++ b/src/gallium/auxiliary/gallivm/gallivm_cpu.cpp @@ -46,6 +46,7 @@ #include "tgsi/tgsi_dump.h" #include "util/u_memory.h" +#include "util/u_math.h" #include <llvm/Module.h> #include <llvm/CallingConv.h> @@ -157,8 +158,8 @@ void gallivm_cpu_jit_compile(struct gallivm_cpu_engine *cpu, struct gallivm_prog llvm::ExistingModuleProvider *mp = new llvm::ExistingModuleProvider(mod); llvm::ExecutionEngine *ee = cpu->engine; assert(ee); - /*FIXME : remove */ - ee->DisableLazyCompilation(); + /*FIXME : why was this disabled ? we need it for pow/sqrt/... */ + ee->DisableLazyCompilation(false); ee->addModuleProvider(mp); llvm::Function *func = func_for_shader(prog); @@ -201,7 +202,6 @@ int gallivm_cpu_vs_exec(struct gallivm_prog *prog, unsigned int i, j; unsigned slot; vertex_shader_runner runner = reinterpret_cast<vertex_shader_runner>(prog->function); - assert(runner); for (i = 0; i < count; i += MAX_TGSI_VERTICES) { diff --git a/src/gallium/auxiliary/gallivm/instructions.cpp b/src/gallium/auxiliary/gallivm/instructions.cpp index a82dc30306d..599975d5add 100644 --- a/src/gallium/auxiliary/gallivm/instructions.cpp +++ b/src/gallium/auxiliary/gallivm/instructions.cpp @@ -83,6 +83,7 @@ Instructions::Instructions(llvm::Module *mod, llvm::Function *func, llvm::BasicB m_llvmPow = 0; m_llvmFloor = 0; m_llvmFlog = 0; + m_llvmFexp = 0; m_llvmLit = 0; m_fmtPtr = 0; @@ -92,194 +93,271 @@ Instructions::Instructions(llvm::Module *mod, llvm::Function *func, llvm::BasicB m_mod = ParseBitcodeFile(buffer); } +llvm::BasicBlock * Instructions::currentBlock() const +{ + return m_builder.GetInsertBlock(); +} + +llvm::Value * Instructions::abs(llvm::Value *in) +{ + std::vector<llvm::Value*> vec = extractVector(in); + Value *xabs = callFAbs(vec[0]); + Value *yabs = callFAbs(vec[1]); + Value *zabs = callFAbs(vec[2]); + Value *wabs = callFAbs(vec[3]); + return vectorFromVals(xabs, yabs, zabs, wabs); +} + llvm::Value * Instructions::add(llvm::Value *in1, llvm::Value *in2) { return m_builder.CreateAdd(in1, in2, name("add")); } -llvm::Value * Instructions::madd(llvm::Value *in1, llvm::Value *in2, - llvm::Value *in3) +llvm::Value * Instructions::arl(llvm::Value *in) { - Value *mulRes = mul(in1, in2); - return add(mulRes, in3); + return floor(in); } - -llvm::Value * Instructions::mul(llvm::Value *in1, llvm::Value *in2) + +void Instructions::beginLoop() { - return m_builder.CreateMul(in1, in2, name("mul")); + BasicBlock *begin = BasicBlock::Create(name("loop"), m_func,0); + BasicBlock *end = BasicBlock::Create(name("endloop"), m_func,0); + + m_builder.CreateBr(begin); + Loop loop; + loop.begin = begin; + loop.end = end; + m_builder.SetInsertPoint(begin); + m_loopStack.push(loop); } -const char * Instructions::name(const char *prefix) +void Instructions::bgnSub(unsigned label) { - ++m_idx; - snprintf(m_name, 32, "%s%d", prefix, m_idx); - return m_name; + llvm::Function *func = findFunction(label); + + Function::arg_iterator args = func->arg_begin(); + Value *ptr_INPUT = args++; + ptr_INPUT->setName("INPUT"); + m_storage->pushArguments(ptr_INPUT); + + llvm::BasicBlock *entry = BasicBlock::Create("entry", func, 0); + + m_func = func; + m_builder.SetInsertPoint(entry); } -llvm::Value * Instructions::dp3(llvm::Value *in1, llvm::Value *in2) +void Instructions::brk() { - Value *mulRes = mul(in1, in2); - Value *x = m_builder.CreateExtractElement(mulRes, - m_storage->constantInt(0), - name("extractx")); - Value *y = m_builder.CreateExtractElement(mulRes, - m_storage->constantInt(1), - name("extracty")); - Value *z = m_builder.CreateExtractElement(mulRes, - m_storage->constantInt(2), - name("extractz")); - Value *xy = m_builder.CreateAdd(x, y,name("xy")); - Value *dot3 = m_builder.CreateAdd(xy, z, name("dot3")); - return vectorFromVals(dot3, dot3, dot3, dot3); + assert(!m_loopStack.empty()); + BasicBlock *unr = BasicBlock::Create(name("unreachable"), m_func,0); + m_builder.CreateBr(m_loopStack.top().end); + m_builder.SetInsertPoint(unr); } -llvm::Value *Instructions::callFSqrt(llvm::Value *val) +void Instructions::cal(int label, llvm::Value *input) { - if (!m_llvmFSqrt) { - // predeclare the intrinsic - std::vector<const Type*> fsqrtArgs; - fsqrtArgs.push_back(Type::FloatTy); - PAListPtr fsqrtPal; - FunctionType* fsqrtType = FunctionType::get( - /*Result=*/Type::FloatTy, - /*Params=*/fsqrtArgs, - /*isVarArg=*/false); - m_llvmFSqrt = Function::Create( - /*Type=*/fsqrtType, - /*Linkage=*/GlobalValue::ExternalLinkage, - /*Name=*/"llvm.sqrt.f32", m_mod); - m_llvmFSqrt->setCallingConv(CallingConv::C); - m_llvmFSqrt->setParamAttrs(fsqrtPal); - } - CallInst *call = m_builder.CreateCall(m_llvmFSqrt, val, - name("sqrt")); - call->setCallingConv(CallingConv::C); - call->setTailCall(false); - return call; + std::vector<Value*> params; + params.push_back(input); + llvm::Function *func = findFunction(label); + + m_builder.CreateCall(func, params.begin(), params.end()); } -llvm::Value * Instructions::rsq(llvm::Value *in1) +llvm::Value * Instructions::ceil(llvm::Value *in) { - Value *x = m_builder.CreateExtractElement(in1, - m_storage->constantInt(0), - name("extractx")); - Value *abs = callFAbs(x); - Value *sqrt = callFSqrt(abs); - - Value *rsqrt = m_builder.CreateFDiv(ConstantFP::get(APFloat(1.f)), - sqrt, - name("rsqrt")); - return vectorFromVals(rsqrt, rsqrt, rsqrt, rsqrt); + std::vector<llvm::Value*> vec = extractVector(in); + return vectorFromVals(callCeil(vec[0]), callCeil(vec[1]), + callCeil(vec[2]), callCeil(vec[3])); } -llvm::Value * Instructions::vectorFromVals(llvm::Value *x, llvm::Value *y, - llvm::Value *z, llvm::Value *w) +llvm::Value * Instructions::clamp(llvm::Value *in1) { - Constant *const_vec = Constant::getNullValue(m_floatVecType); - Value *res = m_builder.CreateInsertElement(const_vec, x, - m_storage->constantInt(0), - name("vecx")); - res = m_builder.CreateInsertElement(res, y, m_storage->constantInt(1), - name("vecxy")); - res = m_builder.CreateInsertElement(res, z, m_storage->constantInt(2), - name("vecxyz")); - if (w) - res = m_builder.CreateInsertElement(res, w, m_storage->constantInt(3), - name("vecxyzw")); - return res; + llvm::Value *zero = constVector(0.0f, 0.0f, 0.0f, 0.0f); + llvm::Value *one = constVector(1.0f, 1.0f, 1.0f, 1.0f); + return min( max(zero, in1), one); } -llvm::Value *Instructions::callFAbs(llvm::Value *val) +llvm::Value * Instructions::cmp(llvm::Value *in1, llvm::Value *in2, llvm::Value *in3) { - if (!m_llvmFAbs) { - // predeclare the intrinsic - std::vector<const Type*> fabsArgs; - fabsArgs.push_back(Type::FloatTy); - PAListPtr fabsPal; - FunctionType* fabsType = FunctionType::get( - /*Result=*/Type::FloatTy, - /*Params=*/fabsArgs, - /*isVarArg=*/false); - m_llvmFAbs = Function::Create( - /*Type=*/fabsType, - /*Linkage=*/GlobalValue::ExternalLinkage, - /*Name=*/"fabs", m_mod); - m_llvmFAbs->setCallingConv(CallingConv::C); - m_llvmFAbs->setParamAttrs(fabsPal); - } - CallInst *call = m_builder.CreateCall(m_llvmFAbs, val, - name("fabs")); - call->setCallingConv(CallingConv::C); + llvm::Function *func = m_mod->getFunction("cmp"); + assert(func); + + std::vector<Value*> params; + params.push_back(in1); + params.push_back(in2); + params.push_back(in3); + CallInst *call = m_builder.CreateCall(func, params.begin(), params.end(), name("cmpres")); call->setTailCall(false); return call; } -llvm::Value * Instructions::lit(llvm::Value *in) +llvm::Value * Instructions::cnd(llvm::Value *in1, llvm::Value *in2, llvm::Value *in3) { - if (!m_llvmLit) { - m_llvmLit = m_mod->getFunction("lit"); - } - CallInst *call = m_builder.CreateCall(m_llvmLit, in, name("litres")); - call->setCallingConv(CallingConv::C); - call->setTailCall(false); - return call; + std::vector<llvm::Value*> vec1 = extractVector(in1); + std::vector<llvm::Value*> vec2 = extractVector(in2); + std::vector<llvm::Value*> vec3 = extractVector(in3); + Constant *half = ConstantFP::get(APFloat(0.5f)); + + Value *xcmp = m_builder.CreateFCmpOGT(vec1[0], half, name("xcmp")); + Value *selx = m_builder.CreateSelect(xcmp, vec2[0], vec3[0], + name("selx")); + + Value *ycmp = m_builder.CreateFCmpOGT(vec1[1], half, name("ycmp")); + Value *sely = m_builder.CreateSelect(ycmp, vec2[1], vec3[1], + name("sely")); + + Value *zcmp = m_builder.CreateFCmpOGT(vec1[2], half, name("zcmp")); + Value *selz = m_builder.CreateSelect(zcmp, vec2[2], vec3[2], + name("selz")); + + Value *wcmp = m_builder.CreateFCmpOGT(vec1[3], half, name("wcmp")); + Value *selw = m_builder.CreateSelect(wcmp, vec2[3], vec3[3], + name("selw")); + + return vectorFromVals(selx, sely, selz, selw); } -llvm::Value * Instructions::sub(llvm::Value *in1, llvm::Value *in2) +llvm::Value * Instructions::cnd0(llvm::Value *in1, llvm::Value *in2, llvm::Value *in3) { - Value *res = m_builder.CreateSub(in1, in2, name("sub")); - return res; + std::vector<llvm::Value*> vec1 = extractVector(in1); + std::vector<llvm::Value*> vec2 = extractVector(in2); + std::vector<llvm::Value*> vec3 = extractVector(in3); + Constant *zero = Constant::getNullValue(Type::FloatTy); + + Value *xcmp = m_builder.CreateFCmpOGE(vec1[0], zero, name("xcmp")); + Value *selx = m_builder.CreateSelect(xcmp, vec2[0], vec3[0], + name("selx")); + + Value *ycmp = m_builder.CreateFCmpOGE(vec1[1], zero, name("ycmp")); + Value *sely = m_builder.CreateSelect(ycmp, vec2[1], vec3[1], + name("sely")); + + Value *zcmp = m_builder.CreateFCmpOGE(vec1[2], zero, name("zcmp")); + Value *selz = m_builder.CreateSelect(zcmp, vec2[2], vec3[2], + name("selz")); + + Value *wcmp = m_builder.CreateFCmpOGE(vec1[3], zero, name("wcmp")); + Value *selw = m_builder.CreateSelect(wcmp, vec2[3], vec3[3], + name("selw")); + + return vectorFromVals(selx, sely, selz, selw); } -llvm::Value * Instructions::callPow(llvm::Value *val1, llvm::Value *val2) +llvm::Value * Instructions::cos(llvm::Value *in) { - if (!m_llvmPow) { - // predeclare the intrinsic - std::vector<const Type*> powArgs; - powArgs.push_back(Type::FloatTy); - powArgs.push_back(Type::FloatTy); - PAListPtr powPal; - FunctionType* powType = FunctionType::get( - /*Result=*/Type::FloatTy, - /*Params=*/powArgs, - /*isVarArg=*/false); - m_llvmPow = Function::Create( - /*Type=*/powType, - /*Linkage=*/GlobalValue::ExternalLinkage, - /*Name=*/"llvm.pow.f32", m_mod); - m_llvmPow->setCallingConv(CallingConv::C); - m_llvmPow->setParamAttrs(powPal); - } - std::vector<Value*> params; - params.push_back(val1); - params.push_back(val2); - CallInst *call = m_builder.CreateCall(m_llvmPow, params.begin(), params.end(), - name("pow")); - call->setCallingConv(CallingConv::C); +#if 0 + llvm::Function *func = m_mod->getFunction("vcos"); + assert(func); + + CallInst *call = m_builder.CreateCall(func, in, name("cosres")); call->setTailCall(false); return call; +#else + std::vector<llvm::Value*> elems = extractVector(in); + Function *func = m_mod->getFunction("cosf"); + assert(func); + CallInst *cos = m_builder.CreateCall(func, elems[0], name("cosres")); + cos->setCallingConv(CallingConv::C); + cos->setTailCall(true); + return vectorFromVals(cos, cos, cos, cos); +#endif } -llvm::Value * Instructions::pow(llvm::Value *in1, llvm::Value *in2) +llvm::Value * Instructions::cross(llvm::Value *in1, llvm::Value *in2) { Value *x1 = m_builder.CreateExtractElement(in1, m_storage->constantInt(0), name("x1")); + Value *y1 = m_builder.CreateExtractElement(in1, + m_storage->constantInt(1), + name("y1")); + Value *z1 = m_builder.CreateExtractElement(in1, + m_storage->constantInt(2), + name("z1")); + Value *x2 = m_builder.CreateExtractElement(in2, m_storage->constantInt(0), name("x2")); - llvm::Value *val = callPow(x1, x2); - return vectorFromVals(val, val, val, val); + Value *y2 = m_builder.CreateExtractElement(in2, + m_storage->constantInt(1), + name("y2")); + Value *z2 = m_builder.CreateExtractElement(in2, + m_storage->constantInt(2), + name("z2")); + Value *y1z2 = mul(y1, z2); + Value *z1y2 = mul(z1, y2); + + Value *z1x2 = mul(z1, x2); + Value *x1z2 = mul(x1, z2); + + Value *x1y2 = mul(x1, y2); + Value *y1x2 = mul(y1, x2); + + return vectorFromVals(sub(y1z2, z1y2), sub(z1x2, x1z2), sub(x1y2, y1x2)); } -llvm::Value * Instructions::rcp(llvm::Value *in1) +llvm::Value * Instructions::ddx(llvm::Value *in) { - Value *x1 = m_builder.CreateExtractElement(in1, - m_storage->constantInt(0), - name("x1")); - Value *res = m_builder.CreateFDiv(ConstantFP::get(APFloat(1.f)), - x1, name("rcp")); - return vectorFromVals(res, res, res, res); + // FIXME + assert(0); +} + +llvm::Value * Instructions::ddy(llvm::Value *in) +{ + // FIXME + assert(0); +} + +llvm::Value * Instructions::div(llvm::Value *in1, llvm::Value *in2) +{ + return m_builder.CreateFDiv(in1, in2, name("div")); +} + +llvm::Value * Instructions::dot2add(llvm::Value *in1, llvm::Value *in2, llvm::Value *in3) +{ + Value *mulRes = mul(in1, in2); + Value *x = m_builder.CreateExtractElement(mulRes, + m_storage->constantInt(0), + name("extractx")); + Value *y = m_builder.CreateExtractElement(mulRes, + m_storage->constantInt(1), + name("extracty")); + Value *z = m_builder.CreateExtractElement(in3, + m_storage->constantInt(2), + name("extractz")); + Value *xy = m_builder.CreateAdd(x, y,name("xy")); + Value *dot2add = m_builder.CreateAdd(xy, z, name("dot2add")); + return vectorFromVals(dot2add, dot2add, dot2add, dot2add); +} + +llvm::Value * Instructions::dp2(llvm::Value *in1, llvm::Value *in2) +{ + Value *mulRes = mul(in1, in2); + Value *x = m_builder.CreateExtractElement(mulRes, + m_storage->constantInt(0), + name("extractx")); + Value *y = m_builder.CreateExtractElement(mulRes, + m_storage->constantInt(1), + name("extracty")); + Value *xy = m_builder.CreateAdd(x, y,name("xy")); + return vectorFromVals(xy, xy, xy, xy); +} + +llvm::Value * Instructions::dp3(llvm::Value *in1, llvm::Value *in2) +{ + Value *mulRes = mul(in1, in2); + Value *x = m_builder.CreateExtractElement(mulRes, + m_storage->constantInt(0), + name("extractx")); + Value *y = m_builder.CreateExtractElement(mulRes, + m_storage->constantInt(1), + name("extracty")); + Value *z = m_builder.CreateExtractElement(mulRes, + m_storage->constantInt(2), + name("extractz")); + Value *xy = m_builder.CreateAdd(x, y,name("xy")); + Value *dot3 = m_builder.CreateAdd(xy, z, name("dot3")); + return vectorFromVals(dot3, dot3, dot3, dot3); } llvm::Value * Instructions::dp4(llvm::Value *in1, llvm::Value *in2) @@ -321,6 +399,53 @@ llvm::Value * Instructions::dst(llvm::Value *in1, llvm::Value *in2) ry, z, w); } +void Instructions::elseop() +{ + assert(!m_ifStack.empty()); + BasicBlock *ifend = BasicBlock::Create(name("ifend"), m_func,0); + m_builder.CreateBr(ifend); + m_builder.SetInsertPoint(m_ifStack.top()); + currentBlock()->setName(name("ifelse")); + m_ifStack.pop(); + m_ifStack.push(ifend); +} + +void Instructions::endif() +{ + assert(!m_ifStack.empty()); + m_builder.CreateBr(m_ifStack.top()); + m_builder.SetInsertPoint(m_ifStack.top()); + m_ifStack.pop(); +} + +void Instructions::endLoop() +{ + assert(!m_loopStack.empty()); + Loop loop = m_loopStack.top(); + m_builder.CreateBr(loop.begin); + loop.end->moveAfter(currentBlock()); + m_builder.SetInsertPoint(loop.end); + m_loopStack.pop(); +} + +void Instructions::end() +{ + m_builder.CreateRetVoid(); +} + +void Instructions::endSub() +{ + m_func = 0; + m_builder.SetInsertPoint(0); +} + +llvm::Value * Instructions::exp(llvm::Value *in) +{ + std::vector<llvm::Value*> vec = extractVector(in); + return vectorFromVals(callFExp(vec[0]), callFExp(vec[1]), + callFExp(vec[2]), callFExp(vec[3])); +} + llvm::Value * Instructions::ex2(llvm::Value *in) { llvm::Value *val = callPow(ConstantFP::get(APFloat(2.f)), @@ -330,31 +455,6 @@ llvm::Value * Instructions::ex2(llvm::Value *in) return vectorFromVals(val, val, val, val); } -llvm::Value * Instructions::callFloor(llvm::Value *val) -{ - if (!m_llvmFloor) { - // predeclare the intrinsic - std::vector<const Type*> floorArgs; - floorArgs.push_back(Type::FloatTy); - PAListPtr floorPal; - FunctionType* floorType = FunctionType::get( - /*Result=*/Type::FloatTy, - /*Params=*/floorArgs, - /*isVarArg=*/false); - m_llvmFloor = Function::Create( - /*Type=*/floorType, - /*Linkage=*/GlobalValue::ExternalLinkage, - /*Name=*/"floorf", m_mod); - m_llvmFloor->setCallingConv(CallingConv::C); - m_llvmFloor->setParamAttrs(floorPal); - } - CallInst *call = m_builder.CreateCall(m_llvmFloor, val, - name("floorf")); - call->setCallingConv(CallingConv::C); - call->setTailCall(false); - return call; -} - llvm::Value * Instructions::floor(llvm::Value *in) { std::vector<llvm::Value*> vec = extractVector(in); @@ -362,42 +462,52 @@ llvm::Value * Instructions::floor(llvm::Value *in) callFloor(vec[2]), callFloor(vec[3])); } -llvm::Value * Instructions::arl(llvm::Value *in) -{ - return floor(in); -} - llvm::Value * Instructions::frc(llvm::Value *in) { llvm::Value *flr = floor(in); return sub(in, flr); } -llvm::Value * Instructions::callFLog(llvm::Value *val) +void Instructions::ifop(llvm::Value *in) { - if (!m_llvmFlog) { - // predeclare the intrinsic - std::vector<const Type*> flogArgs; - flogArgs.push_back(Type::FloatTy); - PAListPtr flogPal; - FunctionType* flogType = FunctionType::get( - /*Result=*/Type::FloatTy, - /*Params=*/flogArgs, - /*isVarArg=*/false); - m_llvmFlog = Function::Create( - /*Type=*/flogType, - /*Linkage=*/GlobalValue::ExternalLinkage, - /*Name=*/"logf", m_mod); - m_llvmFlog->setCallingConv(CallingConv::C); - m_llvmFlog->setParamAttrs(flogPal); - } - CallInst *call = m_builder.CreateCall(m_llvmFlog, val, - name("logf")); - call->setCallingConv(CallingConv::C); + BasicBlock *ifthen = BasicBlock::Create(name("ifthen"), m_func,0); + BasicBlock *ifend = BasicBlock::Create(name("ifthenend"), m_func,0); + + //BasicBlock *yblock = new BasicBlock(name("yblock"), m_func,0); + //BasicBlock *zblock = new BasicBlock(name("zblock"), m_func,0); + //BasicBlock *wblock = new BasicBlock(name("wblock"), m_func,0); + + Constant *float0 = Constant::getNullValue(Type::FloatTy); + + Value *x = m_builder.CreateExtractElement(in, m_storage->constantInt(0), + name("extractx")); + Value *xcmp = m_builder.CreateFCmpUNE(x, float0, name("xcmp")); + m_builder.CreateCondBr(xcmp, ifthen, ifend); + //m_builder.SetInsertPoint(yblock); + + m_builder.SetInsertPoint(ifthen); + m_ifStack.push(ifend); +} + +llvm::Value * Instructions::kil(llvm::Value *in) +{ + llvm::Function *func = m_mod->getFunction("kil"); + assert(func); + + CallInst *call = m_builder.CreateCall(func, in, name("kilpres")); call->setTailCall(false); return call; } +llvm::Value * Instructions::lerp(llvm::Value *in1, llvm::Value *in2, + llvm::Value *in3) +{ + llvm::Value *m = mul(in1, in2); + llvm::Value *vec1 = constVector(1.f, 1.f, 1.f, 1.f); + llvm::Value *s = sub(vec1, in1); + return add(m, mul(s, in3)); +} + llvm::Value * Instructions::lg2(llvm::Value *in) { std::vector<llvm::Value*> vec = extractVector(in); @@ -407,142 +517,176 @@ llvm::Value * Instructions::lg2(llvm::Value *in) callFLog(vec[2]), callFLog(vec[3])), const_vec); } -llvm::Value * Instructions::min(llvm::Value *in1, llvm::Value *in2) +llvm::Value * Instructions::lit(llvm::Value *in) +{ + if (!m_llvmLit) { + m_llvmLit = m_mod->getFunction("lit"); + } + CallInst *call = m_builder.CreateCall(m_llvmLit, in, name("litres")); + call->setCallingConv(CallingConv::C); + call->setTailCall(false); + return call; +} + +llvm::Value * Instructions::log(llvm::Value *in) +{ + std::vector<llvm::Value*> vec = extractVector(in); + return vectorFromVals(callFLog(vec[0]), callFLog(vec[1]), + callFLog(vec[2]), callFLog(vec[3])); +} + +llvm::Value * Instructions::madd(llvm::Value *in1, llvm::Value *in2, + llvm::Value *in3) +{ + Value *mulRes = mul(in1, in2); + return add(mulRes, in3); +} + +llvm::Value * Instructions::max(llvm::Value *in1, llvm::Value *in2) { std::vector<llvm::Value*> vec1 = extractVector(in1); std::vector<llvm::Value*> vec2 = extractVector(in2); - Value *xcmp = m_builder.CreateFCmpOLT(vec1[0], vec2[0], name("xcmp")); + Value *xcmp = m_builder.CreateFCmpOGT(vec1[0], vec2[0], + name("xcmp")); Value *selx = m_builder.CreateSelect(xcmp, vec1[0], vec2[0], name("selx")); - Value *ycmp = m_builder.CreateFCmpOLT(vec1[1], vec2[1], name("ycmp")); + Value *ycmp = m_builder.CreateFCmpOGT(vec1[1], vec2[1], + name("ycmp")); Value *sely = m_builder.CreateSelect(ycmp, vec1[1], vec2[1], name("sely")); - Value *zcmp = m_builder.CreateFCmpOLT(vec1[2], vec2[2], name("zcmp")); + Value *zcmp = m_builder.CreateFCmpOGT(vec1[2], vec2[2], + name("zcmp")); Value *selz = m_builder.CreateSelect(zcmp, vec1[2], vec2[2], name("selz")); - Value *wcmp = m_builder.CreateFCmpOLT(vec1[3], vec2[3], name("wcmp")); + Value *wcmp = m_builder.CreateFCmpOGT(vec1[3], vec2[3], + name("wcmp")); Value *selw = m_builder.CreateSelect(wcmp, vec1[3], vec2[3], name("selw")); return vectorFromVals(selx, sely, selz, selw); } -llvm::Value * Instructions::max(llvm::Value *in1, llvm::Value *in2) +llvm::Value * Instructions::min(llvm::Value *in1, llvm::Value *in2) { std::vector<llvm::Value*> vec1 = extractVector(in1); std::vector<llvm::Value*> vec2 = extractVector(in2); - Value *xcmp = m_builder.CreateFCmpOGT(vec1[0], vec2[0], - name("xcmp")); + Value *xcmp = m_builder.CreateFCmpOLT(vec1[0], vec2[0], name("xcmp")); Value *selx = m_builder.CreateSelect(xcmp, vec1[0], vec2[0], name("selx")); - Value *ycmp = m_builder.CreateFCmpOGT(vec1[1], vec2[1], - name("ycmp")); + Value *ycmp = m_builder.CreateFCmpOLT(vec1[1], vec2[1], name("ycmp")); Value *sely = m_builder.CreateSelect(ycmp, vec1[1], vec2[1], name("sely")); - Value *zcmp = m_builder.CreateFCmpOGT(vec1[2], vec2[2], - name("zcmp")); + Value *zcmp = m_builder.CreateFCmpOLT(vec1[2], vec2[2], name("zcmp")); Value *selz = m_builder.CreateSelect(zcmp, vec1[2], vec2[2], name("selz")); - Value *wcmp = m_builder.CreateFCmpOGT(vec1[3], vec2[3], - name("wcmp")); + Value *wcmp = m_builder.CreateFCmpOLT(vec1[3], vec2[3], name("wcmp")); Value *selw = m_builder.CreateSelect(wcmp, vec1[3], vec2[3], name("selw")); return vectorFromVals(selx, sely, selz, selw); } -void Instructions::printVector(llvm::Value *val) +llvm::Value * Instructions::mul(llvm::Value *in1, llvm::Value *in2) { - static const char *frmt = "Vector is [%f, %f, %f, %f]\x0A"; + return m_builder.CreateMul(in1, in2, name("mul")); +} - if (!m_fmtPtr) { - Constant *format = ConstantArray::get(frmt, true); - ArrayType *arrayTy = ArrayType::get(IntegerType::get(8), strlen(frmt) + 1); - GlobalVariable* globalFormat = new GlobalVariable( - /*Type=*/arrayTy, - /*isConstant=*/true, - /*Linkage=*/GlobalValue::InternalLinkage, - /*Initializer=*/0, // has initializer, specified below - /*Name=*/name(".str"), - m_mod); - globalFormat->setInitializer(format); +llvm::Value * Instructions::neg(llvm::Value *in) +{ + Value *neg = m_builder.CreateNeg(in, name("neg")); + return neg; +} - Constant* const_int0 = Constant::getNullValue(IntegerType::get(32)); - std::vector<Constant*> const_ptr_21_indices; - const_ptr_21_indices.push_back(const_int0); - const_ptr_21_indices.push_back(const_int0); - m_fmtPtr = ConstantExpr::getGetElementPtr(globalFormat, - &const_ptr_21_indices[0], const_ptr_21_indices.size()); - } +llvm::Value * Instructions::nrm(llvm::Value *in) +{ + llvm::Value *v = rsq(in); + return mul(v, in); +} - Function *func_printf = m_mod->getFunction("printf"); - if (!func_printf) - func_printf = declarePrintf(); - assert(func_printf); - std::vector<llvm::Value*> vec = extractVector(val); - Value *dx = m_builder.CreateFPExt(vec[0], Type::DoubleTy, name("dx")); - Value *dy = m_builder.CreateFPExt(vec[1], Type::DoubleTy, name("dy")); - Value *dz = m_builder.CreateFPExt(vec[2], Type::DoubleTy, name("dz")); - Value *dw = m_builder.CreateFPExt(vec[3], Type::DoubleTy, name("dw")); - std::vector<Value*> params; - params.push_back(m_fmtPtr); - params.push_back(dx); - params.push_back(dy); - params.push_back(dz); - params.push_back(dw); - CallInst *call = m_builder.CreateCall(func_printf, params.begin(), params.end(), - name("printf")); - call->setCallingConv(CallingConv::C); - call->setTailCall(true); +llvm::Value * Instructions::pow(llvm::Value *in1, llvm::Value *in2) +{ + Value *x1 = m_builder.CreateExtractElement(in1, + m_storage->constantInt(0), + name("x1")); + Value *x2 = m_builder.CreateExtractElement(in2, + m_storage->constantInt(0), + name("x2")); + llvm::Value *val = callPow(x1, x2); + return vectorFromVals(val, val, val, val); } -llvm::Function * Instructions::declarePrintf() +llvm::Value * Instructions::rcp(llvm::Value *in1) { - std::vector<const Type*> args; - PAListPtr params; - FunctionType* funcTy = FunctionType::get( - /*Result=*/IntegerType::get(32), - /*Params=*/args, - /*isVarArg=*/true); - Function* func_printf = Function::Create( - /*Type=*/funcTy, - /*Linkage=*/GlobalValue::ExternalLinkage, - /*Name=*/"printf", m_mod); - func_printf->setCallingConv(CallingConv::C); - func_printf->setParamAttrs(params); - return func_printf; + Value *x1 = m_builder.CreateExtractElement(in1, + m_storage->constantInt(0), + name("x1")); + Value *res = m_builder.CreateFDiv(ConstantFP::get(APFloat(1.f)), + x1, name("rcp")); + return vectorFromVals(res, res, res, res); +} + +llvm::Value * Instructions::rsq(llvm::Value *in1) +{ + Value *x = m_builder.CreateExtractElement(in1, + m_storage->constantInt(0), + name("extractx")); + Value *abs = callFAbs(x); + Value *sqrt = callFSqrt(abs); + + Value *rsqrt = m_builder.CreateFDiv(ConstantFP::get(APFloat(1.f)), + sqrt, + name("rsqrt")); + return vectorFromVals(rsqrt, rsqrt, rsqrt, rsqrt); } +llvm::Value * Instructions::scs(llvm::Value *in) +{ + llvm::Function *func = m_mod->getFunction("scs"); + assert(func); -llvm::Value * Instructions::sgt(llvm::Value *in1, llvm::Value *in2) + CallInst *call = m_builder.CreateCall(func, in, name("scsres")); + call->setTailCall(false); + return call; +} + +llvm::Value * Instructions::seq(llvm::Value *in1, llvm::Value *in2) { Constant *const1f = ConstantFP::get(APFloat(1.000000e+00f)); Constant *const0f = Constant::getNullValue(Type::FloatTy); std::vector<llvm::Value*> vec1 = extractVector(in1); std::vector<llvm::Value*> vec2 = extractVector(in2); - Value *xcmp = m_builder.CreateFCmpOGT(vec1[0], vec2[0], name("xcmp")); + + Value *xcmp = m_builder.CreateFCmpOEQ(vec1[0], vec2[0], name("xcmp")); Value *x = m_builder.CreateSelect(xcmp, const1f, const0f, name("xsel")); - Value *ycmp = m_builder.CreateFCmpOGT(vec1[1], vec2[1], name("ycmp")); + Value *ycmp = m_builder.CreateFCmpOEQ(vec1[1], vec2[1], name("ycmp")); Value *y = m_builder.CreateSelect(ycmp, const1f, const0f, name("ysel")); - Value *zcmp = m_builder.CreateFCmpOGT(vec1[2], vec2[2], name("zcmp")); + Value *zcmp = m_builder.CreateFCmpOEQ(vec1[2], vec2[2], name("zcmp")); Value *z = m_builder.CreateSelect(zcmp, const1f, const0f, name("zsel")); - Value *wcmp = m_builder.CreateFCmpOGT(vec1[3], vec2[3], name("wcmp")); + Value *wcmp = m_builder.CreateFCmpOEQ(vec1[3], vec2[3], name("wcmp")); Value *w = m_builder.CreateSelect(wcmp, const1f, const0f, name("wsel")); return vectorFromVals(x, y, z, w); } + +llvm::Value * Instructions::sfl(llvm::Value *in1, llvm::Value *in2) +{ + Constant *const0f = Constant::getNullValue(Type::FloatTy); + + return vectorFromVals(const0f, const0f, const0f, const0f); +} + llvm::Value * Instructions::sge(llvm::Value *in1, llvm::Value *in2) { Constant *const1f = ConstantFP::get(APFloat(1.000000e+00f)); @@ -566,157 +710,118 @@ llvm::Value * Instructions::sge(llvm::Value *in1, llvm::Value *in2) return vectorFromVals(x, y, z, w); } - -llvm::Value * Instructions::slt(llvm::Value *in1, llvm::Value *in2) +llvm::Value * Instructions::sgt(llvm::Value *in1, llvm::Value *in2) { Constant *const1f = ConstantFP::get(APFloat(1.000000e+00f)); Constant *const0f = Constant::getNullValue(Type::FloatTy); std::vector<llvm::Value*> vec1 = extractVector(in1); std::vector<llvm::Value*> vec2 = extractVector(in2); - - Value *xcmp = m_builder.CreateFCmpOLT(vec1[0], vec2[0], name("xcmp")); + Value *xcmp = m_builder.CreateFCmpOGT(vec1[0], vec2[0], name("xcmp")); Value *x = m_builder.CreateSelect(xcmp, const1f, const0f, name("xsel")); - Value *ycmp = m_builder.CreateFCmpOLT(vec1[1], vec2[1], name("ycmp")); + Value *ycmp = m_builder.CreateFCmpOGT(vec1[1], vec2[1], name("ycmp")); Value *y = m_builder.CreateSelect(ycmp, const1f, const0f, name("ysel")); - Value *zcmp = m_builder.CreateFCmpOLT(vec1[2], vec2[2], name("zcmp")); + Value *zcmp = m_builder.CreateFCmpOGT(vec1[2], vec2[2], name("zcmp")); Value *z = m_builder.CreateSelect(zcmp, const1f, const0f, name("zsel")); - Value *wcmp = m_builder.CreateFCmpOLT(vec1[3], vec2[3], name("wcmp")); + Value *wcmp = m_builder.CreateFCmpOGT(vec1[3], vec2[3], name("wcmp")); Value *w = m_builder.CreateSelect(wcmp, const1f, const0f, name("wsel")); return vectorFromVals(x, y, z, w); } -llvm::Value * Instructions::cross(llvm::Value *in1, llvm::Value *in2) +llvm::Value * Instructions::sin(llvm::Value *in) { - Value *x1 = m_builder.CreateExtractElement(in1, - m_storage->constantInt(0), - name("x1")); - Value *y1 = m_builder.CreateExtractElement(in1, - m_storage->constantInt(1), - name("y1")); - Value *z1 = m_builder.CreateExtractElement(in1, - m_storage->constantInt(2), - name("z1")); + llvm::Function *func = m_mod->getFunction("vsin"); + assert(func); - Value *x2 = m_builder.CreateExtractElement(in2, - m_storage->constantInt(0), - name("x2")); - Value *y2 = m_builder.CreateExtractElement(in2, - m_storage->constantInt(1), - name("y2")); - Value *z2 = m_builder.CreateExtractElement(in2, - m_storage->constantInt(2), - name("z2")); - Value *y1z2 = mul(y1, z2); - Value *z1y2 = mul(z1, y2); + CallInst *call = m_builder.CreateCall(func, in, name("sinres")); + call->setTailCall(false); + return call; +} - Value *z1x2 = mul(z1, x2); - Value *x1z2 = mul(x1, z2); +llvm::Value * Instructions::sle(llvm::Value *in1, llvm::Value *in2) +{ + Constant *const1f = ConstantFP::get(APFloat(1.000000e+00f)); + Constant *const0f = Constant::getNullValue(Type::FloatTy); - Value *x1y2 = mul(x1, y2); - Value *y1x2 = mul(y1, x2); + std::vector<llvm::Value*> vec1 = extractVector(in1); + std::vector<llvm::Value*> vec2 = extractVector(in2); - return vectorFromVals(sub(y1z2, z1y2), sub(z1x2, x1z2), sub(x1y2, y1x2)); -} + Value *xcmp = m_builder.CreateFCmpOLE(vec1[0], vec2[0], name("xcmp")); + Value *x = m_builder.CreateSelect(xcmp, const1f, const0f, name("xsel")); + Value *ycmp = m_builder.CreateFCmpOLE(vec1[1], vec2[1], name("ycmp")); + Value *y = m_builder.CreateSelect(ycmp, const1f, const0f, name("ysel")); -llvm::Value * Instructions::abs(llvm::Value *in) -{ - std::vector<llvm::Value*> vec = extractVector(in); - Value *xabs = callFAbs(vec[0]); - Value *yabs = callFAbs(vec[1]); - Value *zabs = callFAbs(vec[2]); - Value *wabs = callFAbs(vec[3]); - return vectorFromVals(xabs, yabs, zabs, wabs); + Value *zcmp = m_builder.CreateFCmpOLE(vec1[2], vec2[2], name("zcmp")); + Value *z = m_builder.CreateSelect(zcmp, const1f, const0f, name("zsel")); + + Value *wcmp = m_builder.CreateFCmpOLE(vec1[3], vec2[3], name("wcmp")); + Value *w = m_builder.CreateSelect(wcmp, const1f, const0f, name("wsel")); + + return vectorFromVals(x, y, z, w); } -void Instructions::ifop(llvm::Value *in) +llvm::Value * Instructions::slt(llvm::Value *in1, llvm::Value *in2) { - BasicBlock *ifthen = BasicBlock::Create(name("ifthen"), m_func,0); - BasicBlock *ifend = BasicBlock::Create(name("ifthenend"), m_func,0); + Constant *const1f = ConstantFP::get(APFloat(1.000000e+00f)); + Constant *const0f = Constant::getNullValue(Type::FloatTy); - //BasicBlock *yblock = new BasicBlock(name("yblock"), m_func,0); - //BasicBlock *zblock = new BasicBlock(name("zblock"), m_func,0); - //BasicBlock *wblock = new BasicBlock(name("wblock"), m_func,0); + std::vector<llvm::Value*> vec1 = extractVector(in1); + std::vector<llvm::Value*> vec2 = extractVector(in2); - Constant *float0 = Constant::getNullValue(Type::FloatTy); + Value *xcmp = m_builder.CreateFCmpOLT(vec1[0], vec2[0], name("xcmp")); + Value *x = m_builder.CreateSelect(xcmp, const1f, const0f, name("xsel")); - Value *x = m_builder.CreateExtractElement(in, m_storage->constantInt(0), - name("extractx")); - Value *xcmp = m_builder.CreateFCmpUNE(x, float0, name("xcmp")); - m_builder.CreateCondBr(xcmp, ifthen, ifend); - //m_builder.SetInsertPoint(yblock); + Value *ycmp = m_builder.CreateFCmpOLT(vec1[1], vec2[1], name("ycmp")); + Value *y = m_builder.CreateSelect(ycmp, const1f, const0f, name("ysel")); - m_builder.SetInsertPoint(ifthen); - m_ifStack.push(ifend); -} + Value *zcmp = m_builder.CreateFCmpOLT(vec1[2], vec2[2], name("zcmp")); + Value *z = m_builder.CreateSelect(zcmp, const1f, const0f, name("zsel")); -llvm::BasicBlock * Instructions::currentBlock() const -{ - return m_builder.GetInsertBlock(); -} + Value *wcmp = m_builder.CreateFCmpOLT(vec1[3], vec2[3], name("wcmp")); + Value *w = m_builder.CreateSelect(wcmp, const1f, const0f, name("wsel")); -void Instructions::elseop() -{ - assert(!m_ifStack.empty()); - BasicBlock *ifend = BasicBlock::Create(name("ifend"), m_func,0); - m_builder.CreateBr(ifend); - m_builder.SetInsertPoint(m_ifStack.top()); - currentBlock()->setName(name("ifelse")); - m_ifStack.pop(); - m_ifStack.push(ifend); + return vectorFromVals(x, y, z, w); } -void Instructions::endif() +llvm::Value * Instructions::sne(llvm::Value *in1, llvm::Value *in2) { - assert(!m_ifStack.empty()); - m_builder.CreateBr(m_ifStack.top()); - m_builder.SetInsertPoint(m_ifStack.top()); - m_ifStack.pop(); -} + Constant *const1f = ConstantFP::get(APFloat(1.000000e+00f)); + Constant *const0f = Constant::getNullValue(Type::FloatTy); -llvm::Value * Instructions::lerp(llvm::Value *in1, llvm::Value *in2, - llvm::Value *in3) -{ - llvm::Value *m = mul(in1, in2); - llvm::Value *vec1 = constVector(1.f, 1.f, 1.f, 1.f); - llvm::Value *s = sub(vec1, in1); - return add(m, mul(s, in3)); -} + std::vector<llvm::Value*> vec1 = extractVector(in1); + std::vector<llvm::Value*> vec2 = extractVector(in2); -void Instructions::beginLoop() -{ - BasicBlock *begin = BasicBlock::Create(name("loop"), m_func,0); - BasicBlock *end = BasicBlock::Create(name("endloop"), m_func,0); + Value *xcmp = m_builder.CreateFCmpONE(vec1[0], vec2[0], name("xcmp")); + Value *x = m_builder.CreateSelect(xcmp, const1f, const0f, name("xsel")); - m_builder.CreateBr(begin); - Loop loop; - loop.begin = begin; - loop.end = end; - m_builder.SetInsertPoint(begin); - m_loopStack.push(loop); + Value *ycmp = m_builder.CreateFCmpONE(vec1[1], vec2[1], name("ycmp")); + Value *y = m_builder.CreateSelect(ycmp, const1f, const0f, name("ysel")); + + Value *zcmp = m_builder.CreateFCmpONE(vec1[2], vec2[2], name("zcmp")); + Value *z = m_builder.CreateSelect(zcmp, const1f, const0f, name("zsel")); + + Value *wcmp = m_builder.CreateFCmpONE(vec1[3], vec2[3], name("wcmp")); + Value *w = m_builder.CreateSelect(wcmp, const1f, const0f, name("wsel")); + + return vectorFromVals(x, y, z, w); } -void Instructions::endLoop() +llvm::Value * Instructions::str(llvm::Value *in1, llvm::Value *in2) { - assert(!m_loopStack.empty()); - Loop loop = m_loopStack.top(); - m_builder.CreateBr(loop.begin); - loop.end->moveAfter(currentBlock()); - m_builder.SetInsertPoint(loop.end); - m_loopStack.pop(); + Constant *const1f = ConstantFP::get(APFloat(1.000000e+00f)); + + return vectorFromVals(const1f, const1f, const1f, const1f); } -void Instructions::brk() +llvm::Value * Instructions::sub(llvm::Value *in1, llvm::Value *in2) { - assert(!m_loopStack.empty()); - BasicBlock *unr = BasicBlock::Create(name("unreachable"), m_func,0); - m_builder.CreateBr(m_loopStack.top().end); - m_builder.SetInsertPoint(unr); + Value *res = m_builder.CreateSub(in1, in2, name("sub")); + return res; } llvm::Value * Instructions::trunc(llvm::Value *in) @@ -741,18 +846,298 @@ llvm::Value * Instructions::trunc(llvm::Value *in) return vectorFromVals(fx, fy, fz, fw); } -void Instructions::end() +llvm::Value * Instructions::x2d(llvm::Value *in1, llvm::Value *in2, llvm::Value *in3) { - m_builder.CreateRetVoid(); + std::vector<llvm::Value*> vec1 = extractVector(in1); + std::vector<llvm::Value*> vec2 = extractVector(in2); + std::vector<llvm::Value*> vec3 = extractVector(in3); + + Value *x2x3 = m_builder.CreateMul( vec2[0], vec3[0], name("x2x3")); + Value *y2y3 = m_builder.CreateMul( vec2[1], vec3[1], name("y2y3")); + Value *x1px2x3 = m_builder.CreateAdd (vec1[0], x2x3, name("x1 + x2x3")); + Value *x1px2x3py2y3 = m_builder.CreateAdd (x1px2x3, y2y3, name("x1 + x2x3 + y2y3")); + + Value *x2z3 = m_builder.CreateMul( vec2[0], vec3[2], name("x2z3")); + Value *y2w3 = m_builder.CreateMul( vec2[1], vec3[3], name("y2w3")); + Value *y1px2z3 = m_builder.CreateAdd (vec1[1], x2z3, name("y1 + x2z3")); + Value *y1px2z3py2w3 = m_builder.CreateAdd (y1px2z3, y2w3, name("y1 + x2z3 + y2w3")); + + return vectorFromVals(x1px2x3py2y3, y1px2z3py2w3, x1px2x3py2y3, y1px2z3py2w3); } -void Instructions::cal(int label, llvm::Value *input) +void Instructions::printVector(llvm::Value *val) { + static const char *frmt = "Vector is [%f, %f, %f, %f]\x0A"; + + if (!m_fmtPtr) { + Constant *format = ConstantArray::get(frmt, true); + ArrayType *arrayTy = ArrayType::get(IntegerType::get(8), strlen(frmt) + 1); + GlobalVariable* globalFormat = new GlobalVariable( + /*Type=*/arrayTy, + /*isConstant=*/true, + /*Linkage=*/GlobalValue::InternalLinkage, + /*Initializer=*/0, // has initializer, specified below + /*Name=*/name(".str"), + m_mod); + globalFormat->setInitializer(format); + + Constant* const_int0 = Constant::getNullValue(IntegerType::get(32)); + std::vector<Constant*> const_ptr_21_indices; + const_ptr_21_indices.push_back(const_int0); + const_ptr_21_indices.push_back(const_int0); + m_fmtPtr = ConstantExpr::getGetElementPtr(globalFormat, + &const_ptr_21_indices[0], const_ptr_21_indices.size()); + } + + Function *func_printf = m_mod->getFunction("printf"); + if (!func_printf) + func_printf = declarePrintf(); + assert(func_printf); + std::vector<llvm::Value*> vec = extractVector(val); + Value *dx = m_builder.CreateFPExt(vec[0], Type::DoubleTy, name("dx")); + Value *dy = m_builder.CreateFPExt(vec[1], Type::DoubleTy, name("dy")); + Value *dz = m_builder.CreateFPExt(vec[2], Type::DoubleTy, name("dz")); + Value *dw = m_builder.CreateFPExt(vec[3], Type::DoubleTy, name("dw")); std::vector<Value*> params; - params.push_back(input); - llvm::Function *func = findFunction(label); + params.push_back(m_fmtPtr); + params.push_back(dx); + params.push_back(dy); + params.push_back(dz); + params.push_back(dw); + CallInst *call = m_builder.CreateCall(func_printf, params.begin(), params.end(), + name("printf")); + call->setCallingConv(CallingConv::C); + call->setTailCall(true); +} - m_builder.CreateCall(func, params.begin(), params.end()); +const char * Instructions::name(const char *prefix) +{ + ++m_idx; + snprintf(m_name, 32, "%s%d", prefix, m_idx); + return m_name; +} + +llvm::Value * Instructions::callCeil(llvm::Value *val) +{ + if (!m_llvmCeil) { + // predeclare the intrinsic + std::vector<const Type*> ceilArgs; + ceilArgs.push_back(Type::FloatTy); + AttrListPtr ceilPal; + FunctionType* ceilType = FunctionType::get( + /*Result=*/Type::FloatTy, + /*Params=*/ceilArgs, + /*isVarArg=*/false); + m_llvmCeil = Function::Create( + /*Type=*/ceilType, + /*Linkage=*/GlobalValue::ExternalLinkage, + /*Name=*/"ceilf", m_mod); + m_llvmCeil->setCallingConv(CallingConv::C); + m_llvmCeil->setAttributes(ceilPal); + } + CallInst *call = m_builder.CreateCall(m_llvmCeil, val, + name("ceilf")); + call->setCallingConv(CallingConv::C); + call->setTailCall(false); + return call; +} + +llvm::Value *Instructions::callFAbs(llvm::Value *val) +{ + if (!m_llvmFAbs) { + // predeclare the intrinsic + std::vector<const Type*> fabsArgs; + fabsArgs.push_back(Type::FloatTy); + AttrListPtr fabsPal; + FunctionType* fabsType = FunctionType::get( + /*Result=*/Type::FloatTy, + /*Params=*/fabsArgs, + /*isVarArg=*/false); + m_llvmFAbs = Function::Create( + /*Type=*/fabsType, + /*Linkage=*/GlobalValue::ExternalLinkage, + /*Name=*/"fabs", m_mod); + m_llvmFAbs->setCallingConv(CallingConv::C); + m_llvmFAbs->setAttributes(fabsPal); + } + CallInst *call = m_builder.CreateCall(m_llvmFAbs, val, + name("fabs")); + call->setCallingConv(CallingConv::C); + call->setTailCall(false); + return call; +} + +llvm::Value * Instructions::callFExp(llvm::Value *val) +{ + if (!m_llvmFexp) { + // predeclare the intrinsic + std::vector<const Type*> fexpArgs; + fexpArgs.push_back(Type::FloatTy); + AttrListPtr fexpPal; + FunctionType* fexpType = FunctionType::get( + /*Result=*/Type::FloatTy, + /*Params=*/fexpArgs, + /*isVarArg=*/false); + m_llvmFexp = Function::Create( + /*Type=*/fexpType, + /*Linkage=*/GlobalValue::ExternalLinkage, + /*Name=*/"expf", m_mod); + m_llvmFexp->setCallingConv(CallingConv::C); + m_llvmFexp->setAttributes(fexpPal); + } + CallInst *call = m_builder.CreateCall(m_llvmFexp, val, + name("expf")); + call->setCallingConv(CallingConv::C); + call->setTailCall(false); + return call; +} + +llvm::Value * Instructions::callFLog(llvm::Value *val) +{ + if (!m_llvmFlog) { + // predeclare the intrinsic + std::vector<const Type*> flogArgs; + flogArgs.push_back(Type::FloatTy); + AttrListPtr flogPal; + FunctionType* flogType = FunctionType::get( + /*Result=*/Type::FloatTy, + /*Params=*/flogArgs, + /*isVarArg=*/false); + m_llvmFlog = Function::Create( + /*Type=*/flogType, + /*Linkage=*/GlobalValue::ExternalLinkage, + /*Name=*/"logf", m_mod); + m_llvmFlog->setCallingConv(CallingConv::C); + m_llvmFlog->setAttributes(flogPal); + } + CallInst *call = m_builder.CreateCall(m_llvmFlog, val, + name("logf")); + call->setCallingConv(CallingConv::C); + call->setTailCall(false); + return call; +} + +llvm::Value * Instructions::callFloor(llvm::Value *val) +{ + if (!m_llvmFloor) { + // predeclare the intrinsic + std::vector<const Type*> floorArgs; + floorArgs.push_back(Type::FloatTy); + AttrListPtr floorPal; + FunctionType* floorType = FunctionType::get( + /*Result=*/Type::FloatTy, + /*Params=*/floorArgs, + /*isVarArg=*/false); + m_llvmFloor = Function::Create( + /*Type=*/floorType, + /*Linkage=*/GlobalValue::ExternalLinkage, + /*Name=*/"floorf", m_mod); + m_llvmFloor->setCallingConv(CallingConv::C); + m_llvmFloor->setAttributes(floorPal); + } + CallInst *call = m_builder.CreateCall(m_llvmFloor, val, + name("floorf")); + call->setCallingConv(CallingConv::C); + call->setTailCall(false); + return call; +} + +llvm::Value *Instructions::callFSqrt(llvm::Value *val) +{ + if (!m_llvmFSqrt) { + // predeclare the intrinsic + std::vector<const Type*> fsqrtArgs; + fsqrtArgs.push_back(Type::FloatTy); + AttrListPtr fsqrtPal; + FunctionType* fsqrtType = FunctionType::get( + /*Result=*/Type::FloatTy, + /*Params=*/fsqrtArgs, + /*isVarArg=*/false); + m_llvmFSqrt = Function::Create( + /*Type=*/fsqrtType, + /*Linkage=*/GlobalValue::ExternalLinkage, + /*Name=*/"llvm.sqrt.f32", m_mod); + m_llvmFSqrt->setCallingConv(CallingConv::C); + m_llvmFSqrt->setAttributes(fsqrtPal); + } + CallInst *call = m_builder.CreateCall(m_llvmFSqrt, val, + name("sqrt")); + call->setCallingConv(CallingConv::C); + call->setTailCall(false); + return call; +} + +llvm::Value * Instructions::callPow(llvm::Value *val1, llvm::Value *val2) +{ + if (!m_llvmPow) { + // predeclare the intrinsic + std::vector<const Type*> powArgs; + powArgs.push_back(Type::FloatTy); + powArgs.push_back(Type::FloatTy); + AttrListPtr powPal; + FunctionType* powType = FunctionType::get( + /*Result=*/Type::FloatTy, + /*Params=*/powArgs, + /*isVarArg=*/false); + m_llvmPow = Function::Create( + /*Type=*/powType, + /*Linkage=*/GlobalValue::ExternalLinkage, + /*Name=*/"llvm.pow.f32", m_mod); + m_llvmPow->setCallingConv(CallingConv::C); + m_llvmPow->setAttributes(powPal); + } + std::vector<Value*> params; + params.push_back(val1); + params.push_back(val2); + CallInst *call = m_builder.CreateCall(m_llvmPow, params.begin(), params.end(), + name("pow")); + call->setCallingConv(CallingConv::C); + call->setTailCall(false); + return call; +} + +llvm::Value * Instructions::vectorFromVals(llvm::Value *x, llvm::Value *y, + llvm::Value *z, llvm::Value *w) +{ + Constant *const_vec = Constant::getNullValue(m_floatVecType); + Value *res = m_builder.CreateInsertElement(const_vec, x, + m_storage->constantInt(0), + name("vecx")); + res = m_builder.CreateInsertElement(res, y, m_storage->constantInt(1), + name("vecxy")); + res = m_builder.CreateInsertElement(res, z, m_storage->constantInt(2), + name("vecxyz")); + if (w) + res = m_builder.CreateInsertElement(res, w, m_storage->constantInt(3), + name("vecxyzw")); + return res; +} + +llvm::Value * Instructions::constVector(float x, float y, float z, float w) +{ + std::vector<Constant*> vec(4); + vec[0] = ConstantFP::get(APFloat(x)); + vec[1] = ConstantFP::get(APFloat(y)); + vec[2] = ConstantFP::get(APFloat(z)); + vec[3] = ConstantFP::get(APFloat(w)); + return ConstantVector::get(m_floatVecType, vec); +} + +llvm::Function * Instructions::declarePrintf() +{ + std::vector<const Type*> args; + AttrListPtr params; + FunctionType* funcTy = FunctionType::get( + /*Result=*/IntegerType::get(32), + /*Params=*/args, + /*isVarArg=*/true); + Function* func_printf = Function::Create( + /*Type=*/funcTy, + /*Linkage=*/GlobalValue::ExternalLinkage, + /*Name=*/"printf", m_mod); + func_printf->setCallingConv(CallingConv::C); + func_printf->setAttributes(params); + return func_printf; } llvm::Function * Instructions::declareFunc(int label) @@ -763,7 +1148,7 @@ llvm::Function * Instructions::declareFunc(int label) args.push_back(vecPtr); args.push_back(vecPtr); args.push_back(vecPtr); - PAListPtr params; + AttrListPtr params; FunctionType *funcType = FunctionType::get( /*Result=*/Type::VoidTy, /*Params=*/args, @@ -774,31 +1159,10 @@ llvm::Function * Instructions::declareFunc(int label) /*Linkage=*/GlobalValue::ExternalLinkage, /*Name=*/name.c_str(), m_mod); func->setCallingConv(CallingConv::C); - func->setParamAttrs(params); + func->setAttributes(params); return func; } -void Instructions::bgnSub(unsigned label) -{ - llvm::Function *func = findFunction(label); - - Function::arg_iterator args = func->arg_begin(); - Value *ptr_INPUT = args++; - ptr_INPUT->setName("INPUT"); - m_storage->pushArguments(ptr_INPUT); - - llvm::BasicBlock *entry = BasicBlock::Create("entry", func, 0); - - m_func = func; - m_builder.SetInsertPoint(entry); -} - -void Instructions::endSub() -{ - m_func = 0; - m_builder.SetInsertPoint(0); -} - llvm::Function * Instructions::findFunction(int label) { llvm::Function *func = m_functions[label]; @@ -809,17 +1173,6 @@ llvm::Function * Instructions::findFunction(int label) return func; } -llvm::Value * Instructions::constVector(float x, float y, float z, float w) -{ - std::vector<Constant*> vec(4); - vec[0] = ConstantFP::get(APFloat(x)); - vec[1] = ConstantFP::get(APFloat(y)); - vec[2] = ConstantFP::get(APFloat(z)); - vec[3] = ConstantFP::get(APFloat(w)); - return ConstantVector::get(m_floatVecType, vec); -} - - std::vector<llvm::Value*> Instructions::extractVector(llvm::Value *vec) { std::vector<llvm::Value*> elems(4); @@ -834,69 +1187,7 @@ std::vector<llvm::Value*> Instructions::extractVector(llvm::Value *vec) return elems; } -llvm::Value * Instructions::cmp(llvm::Value *in1, llvm::Value *in2, llvm::Value *in3) -{ - llvm::Function *func = m_mod->getFunction("cmp"); - assert(func); - - std::vector<Value*> params; - params.push_back(in1); - params.push_back(in2); - params.push_back(in3); - CallInst *call = m_builder.CreateCall(func, params.begin(), params.end(), name("cmpres")); - call->setTailCall(false); - return call; -} - -llvm::Value * Instructions::cos(llvm::Value *in) -{ -#if 0 - llvm::Function *func = m_mod->getFunction("vcos"); - assert(func); - CallInst *call = m_builder.CreateCall(func, in, name("cosres")); - call->setTailCall(false); - return call; -#else - std::vector<llvm::Value*> elems = extractVector(in); - Function *func = m_mod->getFunction("cosf"); - assert(func); - CallInst *cos = m_builder.CreateCall(func, elems[0], name("cosres")); - cos->setCallingConv(CallingConv::C); - cos->setTailCall(true); - return vectorFromVals(cos, cos, cos, cos); -#endif -} - -llvm::Value * Instructions::scs(llvm::Value *in) -{ - llvm::Function *func = m_mod->getFunction("scs"); - assert(func); - - CallInst *call = m_builder.CreateCall(func, in, name("scsres")); - call->setTailCall(false); - return call; -} - -llvm::Value * Instructions::kil(llvm::Value *in) -{ - llvm::Function *func = m_mod->getFunction("kil"); - assert(func); - - CallInst *call = m_builder.CreateCall(func, in, name("kilpres")); - call->setTailCall(false); - return call; -} - -llvm::Value * Instructions::sin(llvm::Value *in) -{ - llvm::Function *func = m_mod->getFunction("vsin"); - assert(func); - - CallInst *call = m_builder.CreateCall(func, in, name("sinres")); - call->setTailCall(false); - return call; -} #endif //MESA_LLVM diff --git a/src/gallium/auxiliary/gallivm/instructions.h b/src/gallium/auxiliary/gallivm/instructions.h index d286ce80c78..e18571251ee 100644 --- a/src/gallium/auxiliary/gallivm/instructions.h +++ b/src/gallium/auxiliary/gallivm/instructions.h @@ -57,15 +57,24 @@ public: llvm::BasicBlock *currentBlock() const; llvm::Value *abs(llvm::Value *in1); - llvm::Value *arl(llvm::Value *in1); llvm::Value *add(llvm::Value *in1, llvm::Value *in2); + llvm::Value *arl(llvm::Value *in1); void beginLoop(); void bgnSub(unsigned); void brk(); void cal(int label, llvm::Value *input); + llvm::Value *ceil(llvm::Value *in); + llvm::Value *clamp(llvm::Value *in); llvm::Value *cmp(llvm::Value *in1, llvm::Value *in2, llvm::Value *in3); + llvm::Value *cnd(llvm::Value *in1, llvm::Value *in2, llvm::Value *in3); + llvm::Value *cnd0(llvm::Value *in1, llvm::Value *in2, llvm::Value *in3); llvm::Value *cos(llvm::Value *in); llvm::Value *cross(llvm::Value *in1, llvm::Value *in2); + llvm::Value *ddx(llvm::Value *in); + llvm::Value *ddy(llvm::Value *in); + llvm::Value *div(llvm::Value *in1, llvm::Value *in2); + llvm::Value *dot2add(llvm::Value *in, llvm::Value *in2, llvm::Value *in3); + llvm::Value *dp2(llvm::Value *in1, llvm::Value *in2); llvm::Value *dp3(llvm::Value *in1, llvm::Value *in2); llvm::Value *dp4(llvm::Value *in1, llvm::Value *in2); llvm::Value *dph(llvm::Value *in1, llvm::Value *in2); @@ -75,6 +84,7 @@ public: void endLoop(); void end(); void endSub(); + llvm::Value *exp(llvm::Value *in); llvm::Value *ex2(llvm::Value *in); llvm::Value *floor(llvm::Value *in); llvm::Value *frc(llvm::Value *in); @@ -82,32 +92,43 @@ public: llvm::Value *kil(llvm::Value *in); llvm::Value *lerp(llvm::Value *in1, llvm::Value *in2, llvm::Value *in3); - llvm::Value *lit(llvm::Value *in); llvm::Value *lg2(llvm::Value *in); + llvm::Value *lit(llvm::Value *in); + llvm::Value *log(llvm::Value *in); llvm::Value *madd(llvm::Value *in1, llvm::Value *in2, llvm::Value *in3); - llvm::Value *min(llvm::Value *in1, llvm::Value *in2); llvm::Value *max(llvm::Value *in1, llvm::Value *in2); + llvm::Value *min(llvm::Value *in1, llvm::Value *in2); llvm::Value *mul(llvm::Value *in1, llvm::Value *in2); + llvm::Value *neg(llvm::Value *in); + llvm::Value *nrm(llvm::Value *in); llvm::Value *pow(llvm::Value *in1, llvm::Value *in2); llvm::Value *rcp(llvm::Value *in); llvm::Value *rsq(llvm::Value *in); llvm::Value *scs(llvm::Value *in); + llvm::Value *seq(llvm::Value *in1, llvm::Value *in2); + llvm::Value *sfl(llvm::Value *in1, llvm::Value *in2); llvm::Value *sge(llvm::Value *in1, llvm::Value *in2); llvm::Value *sgt(llvm::Value *in1, llvm::Value *in2); llvm::Value *sin(llvm::Value *in); + llvm::Value *sle(llvm::Value *in1, llvm::Value *in2); llvm::Value *slt(llvm::Value *in1, llvm::Value *in2); + llvm::Value *sne(llvm::Value *in1, llvm::Value *in2); + llvm::Value *str(llvm::Value *in1, llvm::Value *in2); llvm::Value *sub(llvm::Value *in1, llvm::Value *in2); llvm::Value *trunc(llvm::Value *in); + llvm::Value *x2d(llvm::Value *in1, llvm::Value *in2, llvm::Value *in3); void printVector(llvm::Value *val); private: const char *name(const char *prefix); + llvm::Value *callCeil(llvm::Value *val); llvm::Value *callFAbs(llvm::Value *val); + llvm::Value *callFExp(llvm::Value *val); + llvm::Value *callFLog(llvm::Value *val); llvm::Value *callFloor(llvm::Value *val); llvm::Value *callFSqrt(llvm::Value *val); - llvm::Value *callFLog(llvm::Value *val); llvm::Value *callPow(llvm::Value *val1, llvm::Value *val2); llvm::Value *vectorFromVals(llvm::Value *x, llvm::Value *y, @@ -125,16 +146,18 @@ private: llvm::Module *m_mod; llvm::Function *m_func; char m_name[32]; - llvm::IRBuilder m_builder; + llvm::IRBuilder<> m_builder; int m_idx; llvm::VectorType *m_floatVecType; + llvm::Function *m_llvmCeil; llvm::Function *m_llvmFSqrt; llvm::Function *m_llvmFAbs; llvm::Function *m_llvmPow; llvm::Function *m_llvmFloor; llvm::Function *m_llvmFlog; + llvm::Function *m_llvmFexp; llvm::Function *m_llvmLit; llvm::Constant *m_fmtPtr; diff --git a/src/gallium/auxiliary/gallivm/instructionssoa.cpp b/src/gallium/auxiliary/gallivm/instructionssoa.cpp index efddc04e818..d5600fd22da 100644 --- a/src/gallium/auxiliary/gallivm/instructionssoa.cpp +++ b/src/gallium/auxiliary/gallivm/instructionssoa.cpp @@ -90,68 +90,11 @@ llvm::Value * InstructionsSoa::vectorFromVals(llvm::Value *x, llvm::Value *y, return res; } -std::vector<llvm::Value*> InstructionsSoa::arl(const std::vector<llvm::Value*> in) -{ - std::vector<llvm::Value*> res(4); - - //Extract x's - llvm::Value *x1 = m_builder.CreateExtractElement(in[0], - m_storage->constantInt(0), - name("extractX")); - //cast it to an unsigned int - x1 = m_builder.CreateFPToUI(x1, IntegerType::get(32), name("x1IntCast")); - - res[0] = x1;//vectorFromVals(x1, x2, x3, x4); - //only x is valid. the others shouldn't be necessary - /* - res[1] = Constant::getNullValue(m_floatVecType); - res[2] = Constant::getNullValue(m_floatVecType); - res[3] = Constant::getNullValue(m_floatVecType); - */ - - return res; -} - - -std::vector<llvm::Value*> InstructionsSoa::add(const std::vector<llvm::Value*> in1, - const std::vector<llvm::Value*> in2) -{ - std::vector<llvm::Value*> res(4); - - res[0] = m_builder.CreateAdd(in1[0], in2[0], name("addx")); - res[1] = m_builder.CreateAdd(in1[1], in2[1], name("addy")); - res[2] = m_builder.CreateAdd(in1[2], in2[2], name("addz")); - res[3] = m_builder.CreateAdd(in1[3], in2[3], name("addw")); - - return res; -} - -std::vector<llvm::Value*> InstructionsSoa::mul(const std::vector<llvm::Value*> in1, - const std::vector<llvm::Value*> in2) -{ - std::vector<llvm::Value*> res(4); - - res[0] = m_builder.CreateMul(in1[0], in2[0], name("mulx")); - res[1] = m_builder.CreateMul(in1[1], in2[1], name("muly")); - res[2] = m_builder.CreateMul(in1[2], in2[2], name("mulz")); - res[3] = m_builder.CreateMul(in1[3], in2[3], name("mulw")); - - return res; -} - void InstructionsSoa::end() { m_builder.CreateRetVoid(); } -std::vector<llvm::Value*> InstructionsSoa::madd(const std::vector<llvm::Value*> in1, - const std::vector<llvm::Value*> in2, - const std::vector<llvm::Value*> in3) -{ - std::vector<llvm::Value*> res = mul(in1, in2); - return add(res, in3); -} - std::vector<llvm::Value*> InstructionsSoa::extractVector(llvm::Value *vector) { std::vector<llvm::Value*> res(4); @@ -171,6 +114,11 @@ std::vector<llvm::Value*> InstructionsSoa::extractVector(llvm::Value *vector) return res; } +llvm::IRBuilder<>* InstructionsSoa::getIRBuilder() +{ + return &m_builder; +} + void InstructionsSoa::createFunctionMap() { m_functionsMap[TGSI_OPCODE_ABS] = "abs"; @@ -181,6 +129,7 @@ void InstructionsSoa::createFunctionMap() m_functionsMap[TGSI_OPCODE_POWER] = "pow"; m_functionsMap[TGSI_OPCODE_LIT] = "lit"; m_functionsMap[TGSI_OPCODE_RSQ] = "rsq"; + m_functionsMap[TGSI_OPCODE_SLT] = "slt"; } void InstructionsSoa::createDependencies() @@ -273,6 +222,41 @@ std::vector<llvm::Value*> InstructionsSoa::abs(const std::vector<llvm::Value*> i return callBuiltin(func, in1); } +std::vector<llvm::Value*> InstructionsSoa::add(const std::vector<llvm::Value*> in1, + const std::vector<llvm::Value*> in2) +{ + std::vector<llvm::Value*> res(4); + + res[0] = m_builder.CreateAdd(in1[0], in2[0], name("addx")); + res[1] = m_builder.CreateAdd(in1[1], in2[1], name("addy")); + res[2] = m_builder.CreateAdd(in1[2], in2[2], name("addz")); + res[3] = m_builder.CreateAdd(in1[3], in2[3], name("addw")); + + return res; +} + +std::vector<llvm::Value*> InstructionsSoa::arl(const std::vector<llvm::Value*> in) +{ + std::vector<llvm::Value*> res(4); + + //Extract x's + llvm::Value *x1 = m_builder.CreateExtractElement(in[0], + m_storage->constantInt(0), + name("extractX")); + //cast it to an unsigned int + x1 = m_builder.CreateFPToUI(x1, IntegerType::get(32), name("x1IntCast")); + + res[0] = x1;//vectorFromVals(x1, x2, x3, x4); + //only x is valid. the others shouldn't be necessary + /* + res[1] = Constant::getNullValue(m_floatVecType); + res[2] = Constant::getNullValue(m_floatVecType); + res[3] = Constant::getNullValue(m_floatVecType); + */ + + return res; +} + std::vector<llvm::Value*> InstructionsSoa::dp3(const std::vector<llvm::Value*> in1, const std::vector<llvm::Value*> in2) { @@ -280,6 +264,98 @@ std::vector<llvm::Value*> InstructionsSoa::dp3(const std::vector<llvm::Value*> i return callBuiltin(func, in1, in2); } +std::vector<llvm::Value*> InstructionsSoa::lit(const std::vector<llvm::Value*> in) +{ + llvm::Function *func = function(TGSI_OPCODE_LIT); + return callBuiltin(func, in); +} + +std::vector<llvm::Value*> InstructionsSoa::madd(const std::vector<llvm::Value*> in1, + const std::vector<llvm::Value*> in2, + const std::vector<llvm::Value*> in3) +{ + std::vector<llvm::Value*> res = mul(in1, in2); + return add(res, in3); +} + +std::vector<llvm::Value*> InstructionsSoa::max(const std::vector<llvm::Value*> in1, + const std::vector<llvm::Value*> in2) +{ + llvm::Function *func = function(TGSI_OPCODE_MAX); + return callBuiltin(func, in1, in2); +} + +std::vector<llvm::Value*> InstructionsSoa::min(const std::vector<llvm::Value*> in1, + const std::vector<llvm::Value*> in2) +{ + llvm::Function *func = function(TGSI_OPCODE_MIN); + return callBuiltin(func, in1, in2); +} + +std::vector<llvm::Value*> InstructionsSoa::mul(const std::vector<llvm::Value*> in1, + const std::vector<llvm::Value*> in2) +{ + std::vector<llvm::Value*> res(4); + + res[0] = m_builder.CreateMul(in1[0], in2[0], name("mulx")); + res[1] = m_builder.CreateMul(in1[1], in2[1], name("muly")); + res[2] = m_builder.CreateMul(in1[2], in2[2], name("mulz")); + res[3] = m_builder.CreateMul(in1[3], in2[3], name("mulw")); + + return res; +} + +std::vector<llvm::Value*> InstructionsSoa::pow(const std::vector<llvm::Value*> in1, + const std::vector<llvm::Value*> in2) +{ + llvm::Function *func = function(TGSI_OPCODE_POWER); + return callBuiltin(func, in1, in2); +} + +std::vector<llvm::Value*> InstructionsSoa::rsq(const std::vector<llvm::Value*> in) +{ + llvm::Function *func = function(TGSI_OPCODE_RSQ); + return callBuiltin(func, in); +} + +std::vector<llvm::Value*> InstructionsSoa::slt(const std::vector<llvm::Value*> in1, + const std::vector<llvm::Value*> in2) +{ + llvm::Function *func = function(TGSI_OPCODE_SLT); + return callBuiltin(func, in1, in2); +} + +std::vector<llvm::Value*> InstructionsSoa::sub(const std::vector<llvm::Value*> in1, + const std::vector<llvm::Value*> in2) +{ + std::vector<llvm::Value*> res(4); + + res[0] = m_builder.CreateSub(in1[0], in2[0], name("subx")); + res[1] = m_builder.CreateSub(in1[1], in2[1], name("suby")); + res[2] = m_builder.CreateSub(in1[2], in2[2], name("subz")); + res[3] = m_builder.CreateSub(in1[3], in2[3], name("subw")); + + return res; +} + +void checkFunction(Function *func) +{ + for (Function::const_iterator BI = func->begin(), BE = func->end(); + BI != BE; ++BI) { + const BasicBlock &BB = *BI; + for (BasicBlock::const_iterator II = BB.begin(), IE = BB.end(); + II != IE; ++II) { + const Instruction &I = *II; + std::cout<< "Instr = "<<I; + for (unsigned op = 0, E = I.getNumOperands(); op != E; ++op) { + const Value *Op = I.getOperand(op); + std::cout<< "\top = "<<Op<<"("<<op<<")"<<std::endl; + //I->setOperand(op, V); + } + } + } +} + llvm::Value * InstructionsSoa::allocaTemp() { VectorType *vector = VectorType::get(Type::FloatTy, 4); @@ -399,46 +475,6 @@ std::vector<Value*> InstructionsSoa::callBuiltin(llvm::Function *func, const std return allocaToResult(allocaPtr); } -std::vector<llvm::Value*> InstructionsSoa::pow(const std::vector<llvm::Value*> in1, - const std::vector<llvm::Value*> in2) -{ - llvm::Function *func = function(TGSI_OPCODE_POWER); - return callBuiltin(func, in1, in2); -} - -std::vector<llvm::Value*> InstructionsSoa::min(const std::vector<llvm::Value*> in1, - const std::vector<llvm::Value*> in2) -{ - llvm::Function *func = function(TGSI_OPCODE_MIN); - return callBuiltin(func, in1, in2); -} - - -std::vector<llvm::Value*> InstructionsSoa::max(const std::vector<llvm::Value*> in1, - const std::vector<llvm::Value*> in2) -{ - llvm::Function *func = function(TGSI_OPCODE_MAX); - return callBuiltin(func, in1, in2); -} - -void checkFunction(Function *func) -{ - for (Function::const_iterator BI = func->begin(), BE = func->end(); - BI != BE; ++BI) { - const BasicBlock &BB = *BI; - for (BasicBlock::const_iterator II = BB.begin(), IE = BB.end(); - II != IE; ++II) { - const Instruction &I = *II; - std::cout<< "Instr = "<<I; - for (unsigned op = 0, E = I.getNumOperands(); op != E; ++op) { - const Value *Op = I.getOperand(op); - std::cout<< "\top = "<<Op<<"("<<op<<")"<<std::endl; - //I->setOperand(op, V); - } - } - } -} - void InstructionsSoa::injectFunction(llvm::Function *originalFunc, int op) { assert(originalFunc); @@ -458,8 +494,8 @@ void InstructionsSoa::injectFunction(llvm::Function *originalFunc, int op) func = Function::Create(originalFunc->getFunctionType(), GlobalValue::ExternalLinkage, originalFunc->getName(), currentModule()); func->setCallingConv(CallingConv::C); - const PAListPtr pal; - func->setParamAttrs(pal); + const AttrListPtr pal; + func->setAttributes(pal); currentModule()->dump(); } else { DenseMap<const Value*, Value *> val; @@ -483,28 +519,4 @@ void InstructionsSoa::injectFunction(llvm::Function *originalFunc, int op) } } -std::vector<llvm::Value*> InstructionsSoa::sub(const std::vector<llvm::Value*> in1, - const std::vector<llvm::Value*> in2) -{ - std::vector<llvm::Value*> res(4); - - res[0] = m_builder.CreateSub(in1[0], in2[0], name("subx")); - res[1] = m_builder.CreateSub(in1[1], in2[1], name("suby")); - res[2] = m_builder.CreateSub(in1[2], in2[2], name("subz")); - res[3] = m_builder.CreateSub(in1[3], in2[3], name("subw")); - - return res; -} - -std::vector<llvm::Value*> InstructionsSoa::lit(const std::vector<llvm::Value*> in) -{ - llvm::Function *func = function(TGSI_OPCODE_LIT); - return callBuiltin(func, in); -} - -std::vector<llvm::Value*> InstructionsSoa::rsq(const std::vector<llvm::Value*> in) -{ - llvm::Function *func = function(TGSI_OPCODE_RSQ); - return callBuiltin(func, in); -} diff --git a/src/gallium/auxiliary/gallivm/instructionssoa.h b/src/gallium/auxiliary/gallivm/instructionssoa.h index 3e20b652dd3..d6831e0a6b9 100644 --- a/src/gallium/auxiliary/gallivm/instructionssoa.h +++ b/src/gallium/auxiliary/gallivm/instructionssoa.h @@ -69,11 +69,14 @@ public: std::vector<llvm::Value*> pow(const std::vector<llvm::Value*> in1, const std::vector<llvm::Value*> in2); std::vector<llvm::Value*> rsq(const std::vector<llvm::Value*> in1); + std::vector<llvm::Value*> slt(const std::vector<llvm::Value*> in1, + const std::vector<llvm::Value*> in2); std::vector<llvm::Value*> sub(const std::vector<llvm::Value*> in1, const std::vector<llvm::Value*> in2); void end(); std::vector<llvm::Value*> extractVector(llvm::Value *vector); + llvm::IRBuilder<>* getIRBuilder(); private: const char * name(const char *prefix) const; llvm::Value *vectorFromVals(llvm::Value *x, llvm::Value *y, @@ -96,7 +99,7 @@ private: const std::vector<llvm::Value*> in3); void injectFunction(llvm::Function *originalFunc, int op = TGSI_OPCODE_LAST); private: - llvm::IRBuilder m_builder; + llvm::IRBuilder<> m_builder; StorageSoa *m_storage; std::map<int, std::string> m_functionsMap; diff --git a/src/gallium/auxiliary/gallivm/soabuiltins.c b/src/gallium/auxiliary/gallivm/soabuiltins.c index 78f84510e29..cb85e1734ec 100644 --- a/src/gallium/auxiliary/gallivm/soabuiltins.c +++ b/src/gallium/auxiliary/gallivm/soabuiltins.c @@ -36,6 +36,8 @@ typedef __attribute__(( ext_vector_type(4) )) float float4; extern float fabsf(float val); +/* helpers */ + float4 absvec(float4 vec) { float4 res; @@ -47,6 +49,58 @@ float4 absvec(float4 vec) return res; } +float4 maxvec(float4 a, float4 b) +{ + return (float4){(a.x > b.x) ? a.x : b.x, + (a.y > b.y) ? a.y : b.y, + (a.z > b.z) ? a.z : b.z, + (a.w > b.w) ? a.w : b.w}; +} + +float4 minvec(float4 a, float4 b) +{ + return (float4){(a.x < b.x) ? a.x : b.x, + (a.y < b.y) ? a.y : b.y, + (a.z < b.z) ? a.z : b.z, + (a.w < b.w) ? a.w : b.w}; +} + +extern float powf(float num, float p); +extern float sqrtf(float x); + +float4 powvec(float4 vec, float4 q) +{ + float4 p; + p.x = powf(vec.x, q.x); + p.y = powf(vec.y, q.y); + p.z = powf(vec.z, q.z); + p.w = powf(vec.w, q.w); + return p; +} + +float4 sqrtvec(float4 vec) +{ + float4 p; + p.x = sqrtf(vec.x); + p.y = sqrtf(vec.y); + p.z = sqrtf(vec.z); + p.w = sqrtf(vec.w); + return p; +} + +float4 sltvec(float4 v1, float4 v2) +{ + float4 p; + p.x = (v1.x < v2.x) ? 1.0 : 0.0; + p.y = (v1.y < v2.y) ? 1.0 : 0.0; + p.z = (v1.z < v2.z) ? 1.0 : 0.0; + p.w = (v1.w < v2.w) ? 1.0 : 0.0; + return p; +} + + +/* instructions */ + void abs(float4 *res, float4 tmp0x, float4 tmp0y, float4 tmp0z, float4 tmp0w) { @@ -69,7 +123,6 @@ void dp3(float4 *res, res[3] = dot; } - void dp4(float4 *res, float4 tmp0x, float4 tmp0y, float4 tmp0z, float4 tmp0w, float4 tmp1x, float4 tmp1y, float4 tmp1z, float4 tmp1w) @@ -83,35 +136,25 @@ void dp4(float4 *res, res[3] = dot; } -extern float powf(float num, float p); -extern float sqrtf(float x); - -float4 powvec(float4 vec, float4 q) -{ - float4 p; - p.x = powf(vec.x, q.x); - p.y = powf(vec.y, q.y); - p.z = powf(vec.z, q.z); - p.w = powf(vec.w, q.w); - return p; -} - -void pow(float4 *res, - float4 tmp0x, float4 tmp0y, float4 tmp0z, float4 tmp0w, - float4 tmp1x, float4 tmp1y, float4 tmp1z, float4 tmp1w) +void lit(float4 *res, + float4 tmp0x, float4 tmp0y, float4 tmp0z, float4 tmp0w) { - res[0] = powvec(tmp0x, tmp1x); - res[1] = res[0]; - res[2] = res[0]; - res[3] = res[0]; -} + const float4 zerovec = (float4) {0.0, 0.0, 0.0, 0.0}; + const float4 min128 = (float4) {-128.f, -128.f, -128.f, -128.f}; + const float4 plus128 = (float4) {128.f, 128.f, 128.f, 128.f}; -float4 minvec(float4 a, float4 b) -{ - return (float4){(a.x < b.x) ? a.x : b.x, - (a.y < b.y) ? a.y : b.y, - (a.z < b.z) ? a.z : b.z, - (a.w < b.w) ? a.w : b.w}; + res[0] = (float4){1.0, 1.0, 1.0, 1.0}; + if (tmp0x.x > 0) { + float4 tmpy = maxvec(tmp0y, zerovec); + float4 tmpw = minvec(tmp0w, plus128); + tmpw = maxvec(tmpw, min128); + res[1] = tmp0x; + res[2] = powvec(tmpy, tmpw); + } else { + res[1] = zerovec; + res[2] = zerovec; + } + res[3] = (float4){1.0, 1.0, 1.0, 1.0}; } void min(float4 *res, @@ -125,14 +168,6 @@ void min(float4 *res, } -float4 maxvec(float4 a, float4 b) -{ - return (float4){(a.x > b.x) ? a.x : b.x, - (a.y > b.y) ? a.y : b.y, - (a.z > b.z) ? a.z : b.z, - (a.w > b.w) ? a.w : b.w}; -} - void max(float4 *res, float4 tmp0x, float4 tmp0y, float4 tmp0z, float4 tmp0w, float4 tmp1x, float4 tmp1y, float4 tmp1z, float4 tmp1w) @@ -143,37 +178,14 @@ void max(float4 *res, res[3] = maxvec(tmp0w, tmp1w); } - -void lit(float4 *res, - float4 tmp0x, float4 tmp0y, float4 tmp0z, float4 tmp0w) -{ - const float4 zerovec = (float4) {0.0, 0.0, 0.0, 0.0}; - const float4 min128 = (float4) {-128.f, -128.f, -128.f, -128.f}; - const float4 plus128 = (float4) {128.f, 128.f, 128.f, 128.f}; - - res[0] = (float4){1.0, 1.0, 1.0, 1.0}; - if (tmp0x.x > 0) { - float4 tmpy = maxvec(tmp0y, zerovec); - float4 tmpw = minvec(tmp0w, plus128); - tmpw = maxvec(tmpw, min128); - res[1] = tmp0x; - res[2] = powvec(tmpy, tmpw); - } else { - res[1] = zerovec; - res[2] = zerovec; - } - res[3] = (float4){1.0, 1.0, 1.0, 1.0}; -} - - -float4 sqrtvec(float4 vec) +void pow(float4 *res, + float4 tmp0x, float4 tmp0y, float4 tmp0z, float4 tmp0w, + float4 tmp1x, float4 tmp1y, float4 tmp1z, float4 tmp1w) { - float4 p; - p.x = sqrtf(vec.x); - p.y = sqrtf(vec.y); - p.z = sqrtf(vec.z); - p.w = sqrtf(vec.w); - return p; + res[0] = powvec(tmp0x, tmp1x); + res[1] = res[0]; + res[2] = res[0]; + res[3] = res[0]; } void rsq(float4 *res, @@ -185,3 +197,14 @@ void rsq(float4 *res, res[2] = onevec/sqrtvec(absvec(tmp0z)); res[3] = onevec/sqrtvec(absvec(tmp0w)); } + +void slt(float4 *res, + float4 tmp0x, float4 tmp0y, float4 tmp0z, float4 tmp0w, + float4 tmp1x, float4 tmp1y, float4 tmp1z, float4 tmp1w) +{ + res[0] = sltvec(tmp0x, tmp1x); + res[1] = sltvec(tmp0y, tmp1y); + res[2] = sltvec(tmp0z, tmp1z); + res[3] = sltvec(tmp0w, tmp1w); +} + diff --git a/src/gallium/auxiliary/gallivm/storagesoa.cpp b/src/gallium/auxiliary/gallivm/storagesoa.cpp index 78d754371f0..4fc075cf6d4 100644 --- a/src/gallium/auxiliary/gallivm/storagesoa.cpp +++ b/src/gallium/auxiliary/gallivm/storagesoa.cpp @@ -93,7 +93,7 @@ void StorageSoa::declareImmediates() std::vector<float> vals(4); std::vector<Constant*> channelArray; - vals[0] = vec[0]; vals[1] = vec[0]; vals[2] = vec[0]; vals[3] = vec[0]; + vals[0] = vec[0]; vals[1] = vec[1]; vals[2] = vec[2]; vals[3] = vec[3]; llvm::Constant *xChannel = createConstGlobalVector(vals); vals[0] = vec[1]; vals[1] = vec[1]; vals[2] = vec[1]; vals[3] = vec[1]; @@ -144,22 +144,43 @@ std::vector<llvm::Value*> StorageSoa::inputElement(llvm::Value *idx) return res; } -std::vector<llvm::Value*> StorageSoa::constElement(llvm::Value *idx) +llvm::Value* StorageSoa::unpackConstElement(llvm::IRBuilder<>* m_builder, llvm::Value* vector, int cc) { - std::vector<llvm::Value*> res(4); + std::vector<llvm::Value*> x(4); + x[0] = m_builder->CreateExtractElement(vector, + constantInt(cc), + name("x")); + + VectorType *vectorType = VectorType::get(Type::FloatTy, 4); + Constant *constVector = Constant::getNullValue(vectorType); + Value *res = m_builder->CreateInsertElement(constVector, x[0], + constantInt(0), + name("vecx")); + res = m_builder->CreateInsertElement(res, x[0], constantInt(1), + name("vecxx")); + res = m_builder->CreateInsertElement(res, x[0], constantInt(2), + name("vecxxx")); + res = m_builder->CreateInsertElement(res, x[0], constantInt(3), + name("vecxxxx")); + return res; +} + +std::vector<llvm::Value*> StorageSoa::constElement(llvm::IRBuilder<>* m_builder, llvm::Value *idx) +{ + llvm::Value* res; + std::vector<llvm::Value*> res2(4); llvm::Value *xChannel, *yChannel, *zChannel, *wChannel; xChannel = elementPointer(m_consts, idx, 0); - yChannel = elementPointer(m_consts, idx, 1); - zChannel = elementPointer(m_consts, idx, 2); - wChannel = elementPointer(m_consts, idx, 3); - res[0] = alignedArrayLoad(xChannel); - res[1] = alignedArrayLoad(yChannel); - res[2] = alignedArrayLoad(zChannel); - res[3] = alignedArrayLoad(wChannel); + res = alignedArrayLoad(xChannel); - return res; + res2[0]=unpackConstElement(m_builder, res,0); + res2[1]=unpackConstElement(m_builder, res,1); + res2[2]=unpackConstElement(m_builder, res,2); + res2[3]=unpackConstElement(m_builder, res,3); + + return res2; } std::vector<llvm::Value*> StorageSoa::outputElement(llvm::Value *idx) @@ -260,6 +281,12 @@ llvm::Module * StorageSoa::currentModule() const return m_block->getParent()->getParent(); } +llvm::Constant * StorageSoa::createConstGlobalFloat(const float val) +{ + Constant*c = ConstantFP::get(APFloat(val)); + return c; +} + llvm::Constant * StorageSoa::createConstGlobalVector(const std::vector<float> &vec) { VectorType *vectorType = VectorType::get(Type::FloatTy, 4); @@ -278,7 +305,7 @@ llvm::Constant * StorageSoa::createConstGlobalVector(const std::vector<float> &v } std::vector<llvm::Value*> StorageSoa::load(enum tgsi_file_type type, int idx, int swizzle, - llvm::Value *indIdx) + llvm::IRBuilder<>* m_builder,llvm::Value *indIdx) { std::vector<llvm::Value*> val(4); @@ -302,7 +329,7 @@ std::vector<llvm::Value*> StorageSoa::load(enum tgsi_file_type type, int idx, in val = tempElement(realIndex); break; case TGSI_FILE_CONSTANT: - val = constElement(realIndex); + val = constElement(m_builder, realIndex); break; case TGSI_FILE_IMMEDIATE: val = immediateElement(realIndex); diff --git a/src/gallium/auxiliary/gallivm/storagesoa.h b/src/gallium/auxiliary/gallivm/storagesoa.h index ae2fc7c6aee..f21ca6ec433 100644 --- a/src/gallium/auxiliary/gallivm/storagesoa.h +++ b/src/gallium/auxiliary/gallivm/storagesoa.h @@ -29,6 +29,7 @@ #define STORAGESOA_H #include <pipe/p_shader_tokens.h> +#include <llvm/Support/IRBuilder.h> #include <vector> #include <list> @@ -56,7 +57,7 @@ public: std::vector<llvm::Value*> load(enum tgsi_file_type type, int idx, int swizzle, - llvm::Value *indIdx =0); + llvm::IRBuilder<>* m_builder, llvm::Value *indIdx =0); void store(enum tgsi_file_type type, int idx, const std::vector<llvm::Value*> &val, int mask); @@ -76,10 +77,12 @@ private: const char *name(const char *prefix) const; llvm::Value *alignedArrayLoad(llvm::Value *val); llvm::Module *currentModule() const; + llvm::Constant *createConstGlobalFloat(const float val); llvm::Constant *createConstGlobalVector(const std::vector<float> &vec); std::vector<llvm::Value*> inputElement(llvm::Value *indIdx); - std::vector<llvm::Value*> constElement(llvm::Value *indIdx); + llvm::Value* unpackConstElement(llvm::IRBuilder<>* m_builder, llvm::Value *indIdx, int cc); + std::vector<llvm::Value*> constElement(llvm::IRBuilder<>* m_builder, llvm::Value *indIdx); std::vector<llvm::Value*> outputElement(llvm::Value *indIdx); std::vector<llvm::Value*> tempElement(llvm::Value *indIdx); std::vector<llvm::Value*> immediateElement(llvm::Value *indIdx); diff --git a/src/gallium/auxiliary/gallivm/tgsitollvm.cpp b/src/gallium/auxiliary/gallivm/tgsitollvm.cpp index cc1516a45e0..1191a6cae97 100644 --- a/src/gallium/auxiliary/gallivm/tgsitollvm.cpp +++ b/src/gallium/auxiliary/gallivm/tgsitollvm.cpp @@ -52,7 +52,7 @@ static inline FunctionType *vertexShaderFunctionType() // pass are castable to the following: // [4 x <4 x float>] inputs, // [4 x <4 x float>] output, - // [4 x [4 x float]] consts, + // [4 x [1 x float]] consts, // [4 x <4 x float>] temps std::vector<const Type*> funcArgs; @@ -61,7 +61,7 @@ static inline FunctionType *vertexShaderFunctionType() PointerType *vectorArrayPtr = PointerType::get(vectorArray, 0); ArrayType *floatArray = ArrayType::get(Type::FloatTy, 4); - ArrayType *constsArray = ArrayType::get(floatArray, 4); + ArrayType *constsArray = ArrayType::get(floatArray, 1); PointerType *constsArrayPtr = PointerType::get(constsArray, 0); funcArgs.push_back(vectorArrayPtr);//inputs @@ -246,6 +246,7 @@ translate_instruction(llvm::Module *module, val = storage->constElement(src->SrcRegister.Index, indIdx); } else if (src->SrcRegister.File == TGSI_FILE_INPUT) { val = storage->inputElement(src->SrcRegister.Index, indIdx); + // FIXME we should not be generating elements for temporaries, this creates useless memory writes } else if (src->SrcRegister.File == TGSI_FILE_TEMPORARY) { val = storage->tempElement(src->SrcRegister.Index); } else if (src->SrcRegister.File == TGSI_FILE_OUTPUT) { @@ -286,9 +287,13 @@ translate_instruction(llvm::Module *module, out = instr->rsq(inputs[0]); } break; - case TGSI_OPCODE_EXP: + case TGSI_OPCODE_EXP: { + out = instr->exp(inputs[0]); + } break; - case TGSI_OPCODE_LOG: + case TGSI_OPCODE_LOG: { + out = instr->log(inputs[0]); + } break; case TGSI_OPCODE_MUL: { out = instr->mul(inputs[0], inputs[1]); @@ -338,21 +343,31 @@ translate_instruction(llvm::Module *module, out = instr->lerp(inputs[0], inputs[1], inputs[2]); } break; - case TGSI_OPCODE_CND: + case TGSI_OPCODE_CND: { + out = instr->cnd(inputs[0], inputs[1], inputs[2]); + } break; - case TGSI_OPCODE_CND0: + case TGSI_OPCODE_CND0: { + out = instr->cnd0(inputs[0], inputs[1], inputs[2]); + } break; - case TGSI_OPCODE_DOT2ADD: + case TGSI_OPCODE_DOT2ADD: { + out = instr->dot2add(inputs[0], inputs[1], inputs[2]); + } break; case TGSI_OPCODE_INDEX: break; - case TGSI_OPCODE_NEGATE: + case TGSI_OPCODE_NEGATE: { + out = instr->neg(inputs[0]); + } break; case TGSI_OPCODE_FRAC: { out = instr->frc(inputs[0]); } break; - case TGSI_OPCODE_CLAMP: + case TGSI_OPCODE_CLAMP: { + out = instr->clamp(inputs[0]); + } break; case TGSI_OPCODE_FLOOR: { out = instr->floor(inputs[0]); @@ -392,9 +407,13 @@ translate_instruction(llvm::Module *module, out = instr->cos(inputs[0]); } break; - case TGSI_OPCODE_DDX: + case TGSI_OPCODE_DDX: { + out = instr->ddx(inputs[0]); + } break; - case TGSI_OPCODE_DDY: + case TGSI_OPCODE_DDY: { + out = instr->ddy(inputs[0]); + } break; case TGSI_OPCODE_KILP: break; @@ -408,9 +427,13 @@ translate_instruction(llvm::Module *module, break; case TGSI_OPCODE_RFL: break; - case TGSI_OPCODE_SEQ: + case TGSI_OPCODE_SEQ: { + out = instr->seq(inputs[0], inputs[1]); + } break; - case TGSI_OPCODE_SFL: + case TGSI_OPCODE_SFL: { + out = instr->sfl(inputs[0], inputs[1]); + } break; case TGSI_OPCODE_SGT: { out = instr->sgt(inputs[0], inputs[1]); @@ -420,11 +443,17 @@ translate_instruction(llvm::Module *module, out = instr->sin(inputs[0]); } break; - case TGSI_OPCODE_SLE: + case TGSI_OPCODE_SLE: { + out = instr->sle(inputs[0], inputs[1]); + } break; - case TGSI_OPCODE_SNE: + case TGSI_OPCODE_SNE: { + out = instr->sne(inputs[0], inputs[1]); + } break; - case TGSI_OPCODE_STR: + case TGSI_OPCODE_STR: { + out = instr->str(inputs[0], inputs[1]); + } break; case TGSI_OPCODE_TEX: break; @@ -438,7 +467,9 @@ translate_instruction(llvm::Module *module, break; case TGSI_OPCODE_UP4UB: break; - case TGSI_OPCODE_X2D: + case TGSI_OPCODE_X2D: { + out = instr->x2d(inputs[0], inputs[1], inputs[2]); + } break; case TGSI_OPCODE_ARA: break; @@ -468,11 +499,18 @@ translate_instruction(llvm::Module *module, break; case TGSI_OPCODE_TXB: break; - case TGSI_OPCODE_NRM: + case TGSI_OPCODE_NRM4: + case TGSI_OPCODE_NRM: { + out = instr->nrm(inputs[0]); + } break; - case TGSI_OPCODE_DIV: + case TGSI_OPCODE_DIV: { + out = instr->div(inputs[0], inputs[1]); + } break; - case TGSI_OPCODE_DP2: + case TGSI_OPCODE_DP2: { + out = instr->dp2(inputs[0], inputs[1]); + } break; case TGSI_OPCODE_TXL: break; @@ -590,8 +628,6 @@ translate_instruction(llvm::Module *module, break; case TGSI_OPCODE_M3X2: break; - case TGSI_OPCODE_NRM4: - break; case TGSI_OPCODE_CALLNZ: break; case TGSI_OPCODE_IFC: @@ -641,6 +677,7 @@ translate_instruction(llvm::Module *module, if (dst->DstRegister.File == TGSI_FILE_OUTPUT) { storage->setOutputElement(dst->DstRegister.Index, out, dst->DstRegister.WriteMask); + // FIXME we should not be generating elements for temporaries, this creates useless memory writes } else if (dst->DstRegister.File == TGSI_FILE_TEMPORARY) { storage->setTempElement(dst->DstRegister.Index, out, dst->DstRegister.WriteMask); } else if (dst->DstRegister.File == TGSI_FILE_ADDRESS) { @@ -672,9 +709,8 @@ translate_instructionir(llvm::Module *module, if (src->SrcRegister.Indirect) { indIdx = storage->addrElement(src->SrcRegisterInd.Index); } - val = storage->load((enum tgsi_file_type)src->SrcRegister.File, - src->SrcRegister.Index, swizzle, indIdx); + src->SrcRegister.Index, swizzle, instr->getIRBuilder(), indIdx); inputs[i] = val; } @@ -732,6 +768,7 @@ translate_instructionir(llvm::Module *module, } break; case TGSI_OPCODE_SLT: { + out = instr->slt(inputs[0], inputs[1]); } break; case TGSI_OPCODE_SGE: { @@ -989,7 +1026,6 @@ translate_instructionir(llvm::Module *module, /* store results */ for (int i = 0; i < inst->Instruction.NumDstRegs; ++i) { struct tgsi_full_dst_register *dst = &inst->FullDstRegisters[i]; - storage->store((enum tgsi_file_type)dst->DstRegister.File, dst->DstRegister.Index, out, dst->DstRegister.WriteMask); } diff --git a/src/gallium/auxiliary/rtasm/rtasm_ppc_spe.c b/src/gallium/auxiliary/rtasm/rtasm_ppc_spe.c index 491141f1908..dea1aed0320 100644 --- a/src/gallium/auxiliary/rtasm/rtasm_ppc_spe.c +++ b/src/gallium/auxiliary/rtasm/rtasm_ppc_spe.c @@ -164,6 +164,27 @@ rem_prefix(const char *longname) } +static const char * +reg_name(int reg) +{ + switch (reg) { + case SPE_REG_SP: + return "$sp"; + case SPE_REG_RA: + return "$lr"; + default: + { + /* cycle through four buffers to handle multiple calls per printf */ + static char buf[4][10]; + static int b = 0; + b = (b + 1) % 4; + sprintf(buf[b], "$%d", reg); + return buf[b]; + } + } +} + + static void emit_RR(struct spe_function *p, unsigned op, unsigned rT, unsigned rA, unsigned rB, const char *name) { @@ -176,7 +197,8 @@ static void emit_RR(struct spe_function *p, unsigned op, unsigned rT, assert(p->num_inst <= p->max_inst); if (p->print) { indent(p); - printf("%s\t$%d, $%d, $%d\n", rem_prefix(name), rT, rA, rB); + printf("%s\t%s, %s, %s\n", + rem_prefix(name), reg_name(rT), reg_name(rA), reg_name(rB)); } } @@ -194,7 +216,8 @@ static void emit_RRR(struct spe_function *p, unsigned op, unsigned rT, assert(p->num_inst <= p->max_inst); if (p->print) { indent(p); - printf("%s\t$%d, $%d, $%d, $%d\n", rem_prefix(name), rT, rA, rB, rC); + printf("%s\t%s, %s, %s, %s\n", rem_prefix(name), reg_name(rT), + reg_name(rA), reg_name(rB), reg_name(rC)); } } @@ -211,7 +234,8 @@ static void emit_RI7(struct spe_function *p, unsigned op, unsigned rT, assert(p->num_inst <= p->max_inst); if (p->print) { indent(p); - printf("%s\t$%d, $%d, 0x%x\n", rem_prefix(name), rT, rA, imm); + printf("%s\t%s, %s, 0x%x\n", + rem_prefix(name), reg_name(rT), reg_name(rA), imm); } } @@ -229,7 +253,8 @@ static void emit_RI8(struct spe_function *p, unsigned op, unsigned rT, assert(p->num_inst <= p->max_inst); if (p->print) { indent(p); - printf("%s\t$%d, $%d, 0x%x\n", rem_prefix(name), rT, rA, imm); + printf("%s\t%s, %s, 0x%x\n", + rem_prefix(name), reg_name(rT), reg_name(rA), imm); } } @@ -247,15 +272,22 @@ static void emit_RI10(struct spe_function *p, unsigned op, unsigned rT, assert(p->num_inst <= p->max_inst); if (p->print) { indent(p); - if (strcmp(name, "spe_lqd") == 0 || - strcmp(name, "spe_stqd") == 0) - printf("%s\t$%d, 0x%x($%d)\n", rem_prefix(name), rT, imm, rA); - else - printf("%s\t$%d, $%d, 0x%x\n", rem_prefix(name), rT, rA, imm); + printf("%s\t%s, %s, 0x%x\n", + rem_prefix(name), reg_name(rT), reg_name(rA), imm); } } +/** As above, but do range checking on signed immediate value */ +static void emit_RI10s(struct spe_function *p, unsigned op, unsigned rT, + unsigned rA, int imm, const char *name) +{ + assert(imm <= 511); + assert(imm >= -512); + emit_RI10(p, op, rT, rA, imm, name); +} + + static void emit_RI16(struct spe_function *p, unsigned op, unsigned rT, int imm, const char *name) { @@ -267,7 +299,7 @@ static void emit_RI16(struct spe_function *p, unsigned op, unsigned rT, assert(p->num_inst <= p->max_inst); if (p->print) { indent(p); - printf("%s\t$%d, 0x%x\n", rem_prefix(name), rT, imm); + printf("%s\t%s, 0x%x\n", rem_prefix(name), reg_name(rT), imm); } } @@ -283,7 +315,7 @@ static void emit_RI18(struct spe_function *p, unsigned op, unsigned rT, assert(p->num_inst <= p->max_inst); if (p->print) { indent(p); - printf("%s\t$%d, 0x%x\n", rem_prefix(name), rT, imm); + printf("%s\t%s, 0x%x\n", rem_prefix(name), reg_name(rT), imm); } } @@ -332,6 +364,12 @@ void _name (struct spe_function *p, unsigned rT, unsigned rA, int imm) \ emit_RI10(p, _op, rT, rA, imm, __FUNCTION__); \ } +#define EMIT_RI10s(_name, _op) \ +void _name (struct spe_function *p, unsigned rT, unsigned rA, int imm) \ +{ \ + emit_RI10s(p, _op, rT, rA, imm, __FUNCTION__); \ +} + #define EMIT_RI16(_name, _op) \ void _name (struct spe_function *p, unsigned rT, int imm) \ { \ @@ -353,20 +391,28 @@ void _name (struct spe_function *p, int imm) \ #include "rtasm_ppc_spe.h" + /** * Initialize an spe_function. * \param code_size size of instruction buffer to allocate, in bytes. */ void spe_init_func(struct spe_function *p, unsigned code_size) { + unsigned int i; + p->store = align_malloc(code_size, 16); p->num_inst = 0; p->max_inst = code_size / SPE_INST_SIZE; + p->set_count = 0; + memset(p->regs, 0, SPE_NUM_REGS * sizeof(p->regs[0])); + /* Conservatively treat R0 - R2 and R80 - R127 as non-volatile. */ - p->regs[0] = ~7; - p->regs[1] = (1U << (80 - 64)) - 1; + p->regs[0] = p->regs[1] = p->regs[2] = 1; + for (i = 80; i <= 127; i++) { + p->regs[i] = 1; + } p->print = false; p->indent = 0; @@ -398,12 +444,8 @@ int spe_allocate_available_register(struct spe_function *p) { unsigned i; for (i = 0; i < SPE_NUM_REGS; i++) { - const uint64_t mask = (1ULL << (i % 64)); - const unsigned idx = i / 64; - - assert(idx < 2); - if ((p->regs[idx] & mask) != 0) { - p->regs[idx] &= ~mask; + if (p->regs[i] == 0) { + p->regs[i] = 1; return i; } } @@ -417,31 +459,84 @@ int spe_allocate_available_register(struct spe_function *p) */ int spe_allocate_register(struct spe_function *p, int reg) { - const unsigned idx = reg / 64; - const unsigned bit = reg % 64; - assert(reg < SPE_NUM_REGS); - assert((p->regs[idx] & (1ULL << bit)) != 0); - - p->regs[idx] &= ~(1ULL << bit); + assert(p->regs[reg] == 0); + p->regs[reg] = 1; return reg; } /** - * Mark the given SPE register as "unallocated". + * Mark the given SPE register as "unallocated". Note that this should + * only be used on registers allocated in the current register set; an + * assertion will fail if an attempt is made to deallocate a register + * allocated in an earlier register set. */ void spe_release_register(struct spe_function *p, int reg) { - const unsigned idx = reg / 64; - const unsigned bit = reg % 64; + assert(reg < SPE_NUM_REGS); + assert(p->regs[reg] == 1); - assert(idx < 2); + p->regs[reg] = 0; +} - assert(reg < SPE_NUM_REGS); - assert((p->regs[idx] & (1ULL << bit)) == 0); +/** + * Start a new set of registers. This can be called if + * it will be difficult later to determine exactly what + * registers were actually allocated during a code generation + * sequence, and you really just want to deallocate all of them. + */ +void spe_allocate_register_set(struct spe_function *p) +{ + unsigned int i; - p->regs[idx] |= (1ULL << bit); + /* Keep track of the set count. If it ever wraps around to 0, + * we're in trouble. + */ + p->set_count++; + assert(p->set_count > 0); + + /* Increment the allocation count of all registers currently + * allocated. Then any registers that are allocated in this set + * will be the only ones with a count of 1; they'll all be released + * when the register set is released. + */ + for (i = 0; i < SPE_NUM_REGS; i++) { + if (p->regs[i] > 0) + p->regs[i]++; + } +} + +void spe_release_register_set(struct spe_function *p) +{ + unsigned int i; + + /* If the set count drops below zero, we're in trouble. */ + assert(p->set_count > 0); + p->set_count--; + + /* Drop the allocation level of all registers. Any allocated + * during this register set will drop to 0 and then become + * available. + */ + for (i = 0; i < SPE_NUM_REGS; i++) { + if (p->regs[i] > 0) + p->regs[i]--; + } +} + + +unsigned +spe_get_registers_used(const struct spe_function *p, ubyte used[]) +{ + unsigned i, num = 0; + /* only count registers in the range available to callers */ + for (i = 2; i < 80; i++) { + if (p->regs[i]) { + used[num++] = i; + } + } + return num; } @@ -459,7 +554,7 @@ spe_indent(struct spe_function *p, int spaces) } -extern void +void spe_comment(struct spe_function *p, int rel_indent, const char *s) { if (p->print) { @@ -472,6 +567,56 @@ spe_comment(struct spe_function *p, int rel_indent, const char *s) /** + * Load quad word. + * NOTE: offset is in bytes and the least significant 4 bits must be zero! + */ +void spe_lqd(struct spe_function *p, unsigned rT, unsigned rA, int offset) +{ + const boolean pSave = p->print; + + /* offset must be a multiple of 16 */ + assert(offset % 16 == 0); + /* offset must fit in 10-bit signed int field, after shifting */ + assert((offset >> 4) <= 511); + assert((offset >> 4) >= -512); + + p->print = FALSE; + emit_RI10(p, 0x034, rT, rA, offset >> 4, "spe_lqd"); + p->print = pSave; + + if (p->print) { + indent(p); + printf("lqd\t%s, %d(%s)\n", reg_name(rT), offset, reg_name(rA)); + } +} + + +/** + * Store quad word. + * NOTE: offset is in bytes and the least significant 4 bits must be zero! + */ +void spe_stqd(struct spe_function *p, unsigned rT, unsigned rA, int offset) +{ + const boolean pSave = p->print; + + /* offset must be a multiple of 16 */ + assert(offset % 16 == 0); + /* offset must fit in 10-bit signed int field, after shifting */ + assert((offset >> 4) <= 511); + assert((offset >> 4) >= -512); + + p->print = FALSE; + emit_RI10(p, 0x024, rT, rA, offset >> 4, "spe_stqd"); + p->print = pSave; + + if (p->print) { + indent(p); + printf("stqd\t%s, %d(%s)\n", reg_name(rT), offset, reg_name(rA)); + } +} + + +/** * For branch instructions: * \param d if 1, disable interupts if branch is taken * \param e if 1, enable interupts if branch is taken @@ -603,22 +748,187 @@ void spe_load_uint(struct spe_function *p, unsigned rT, unsigned int ui) { /* If the whole value is in the lower 18 bits, use ila, which * doesn't sign-extend. Otherwise, if the two halfwords of - * the constant are identical, use ilh. Otherwise, we have - * to use ilhu followed by iohl. + * the constant are identical, use ilh. Otherwise, if every byte of + * the desired value is 0x00 or 0xff, we can use Form Select Mask for + * Bytes Immediate (fsmbi) to load the value in a single instruction. + * Otherwise, in the general case, we have to use ilhu followed by iohl. */ - if ((ui & 0xfffc0000) == ui) { + if ((ui & 0x0003ffff) == ui) { spe_ila(p, rT, ui); } else if ((ui >> 16) == (ui & 0xffff)) { spe_ilh(p, rT, ui & 0xffff); } + else if ( + ((ui & 0x000000ff) == 0 || (ui & 0x000000ff) == 0x000000ff) && + ((ui & 0x0000ff00) == 0 || (ui & 0x0000ff00) == 0x0000ff00) && + ((ui & 0x00ff0000) == 0 || (ui & 0x00ff0000) == 0x00ff0000) && + ((ui & 0xff000000) == 0 || (ui & 0xff000000) == 0xff000000) + ) { + unsigned int mask = 0; + /* fsmbi duplicates each bit in the given mask eight times, + * using a 16-bit value to initialize a 16-byte quadword. + * Each 4-bit nybble of the mask corresponds to a full word + * of the result; look at the value and figure out the mask + * (replicated for each word in the quadword), and then + * form the "select mask" to get the value. + */ + if ((ui & 0x000000ff) == 0x000000ff) mask |= 0x1111; + if ((ui & 0x0000ff00) == 0x0000ff00) mask |= 0x2222; + if ((ui & 0x00ff0000) == 0x00ff0000) mask |= 0x4444; + if ((ui & 0xff000000) == 0xff000000) mask |= 0x8888; + spe_fsmbi(p, rT, mask); + } else { + /* The general case: this usually uses two instructions, but + * may use only one if the low-order 16 bits of each word are 0. + */ spe_ilhu(p, rT, ui >> 16); if (ui & 0xffff) spe_iohl(p, rT, ui & 0xffff); } } +/** + * This function is constructed identically to spe_xor_uint() below. + * Changes to one should be made in the other. + */ +void +spe_and_uint(struct spe_function *p, unsigned rT, unsigned rA, unsigned int ui) +{ + /* If we can, emit a single instruction, either And Byte Immediate + * (which uses the same constant across each byte), And Halfword Immediate + * (which sign-extends a 10-bit immediate to 16 bits and uses that + * across each halfword), or And Word Immediate (which sign-extends + * a 10-bit immediate to 32 bits). + * + * Otherwise, we'll need to use a temporary register. + */ + unsigned int tmp; + + /* If the upper 23 bits are all 0s or all 1s, sign extension + * will work and we can use And Word Immediate + */ + tmp = ui & 0xfffffe00; + if (tmp == 0xfffffe00 || tmp == 0) { + spe_andi(p, rT, rA, ui & 0x000003ff); + return; + } + + /* If the ui field is symmetric along halfword boundaries and + * the upper 7 bits of each halfword are all 0s or 1s, we + * can use And Halfword Immediate + */ + tmp = ui & 0xfe00fe00; + if ((tmp == 0xfe00fe00 || tmp == 0) && ((ui >> 16) == (ui & 0x0000ffff))) { + spe_andhi(p, rT, rA, ui & 0x000003ff); + return; + } + + /* If the ui field is symmetric in each byte, then we can use + * the And Byte Immediate instruction. + */ + tmp = ui & 0x000000ff; + if ((ui >> 24) == tmp && ((ui >> 16) & 0xff) == tmp && ((ui >> 8) & 0xff) == tmp) { + spe_andbi(p, rT, rA, tmp); + return; + } + + /* Otherwise, we'll have to use a temporary register. */ + unsigned int tmp_reg = spe_allocate_available_register(p); + spe_load_uint(p, tmp_reg, ui); + spe_and(p, rT, rA, tmp_reg); + spe_release_register(p, tmp_reg); +} + + +/** + * This function is constructed identically to spe_and_uint() above. + * Changes to one should be made in the other. + */ +void +spe_xor_uint(struct spe_function *p, unsigned rT, unsigned rA, unsigned int ui) +{ + /* If we can, emit a single instruction, either Exclusive Or Byte + * Immediate (which uses the same constant across each byte), Exclusive + * Or Halfword Immediate (which sign-extends a 10-bit immediate to + * 16 bits and uses that across each halfword), or Exclusive Or Word + * Immediate (which sign-extends a 10-bit immediate to 32 bits). + * + * Otherwise, we'll need to use a temporary register. + */ + unsigned int tmp; + + /* If the upper 23 bits are all 0s or all 1s, sign extension + * will work and we can use Exclusive Or Word Immediate + */ + tmp = ui & 0xfffffe00; + if (tmp == 0xfffffe00 || tmp == 0) { + spe_xori(p, rT, rA, ui & 0x000003ff); + return; + } + + /* If the ui field is symmetric along halfword boundaries and + * the upper 7 bits of each halfword are all 0s or 1s, we + * can use Exclusive Or Halfword Immediate + */ + tmp = ui & 0xfe00fe00; + if ((tmp == 0xfe00fe00 || tmp == 0) && ((ui >> 16) == (ui & 0x0000ffff))) { + spe_xorhi(p, rT, rA, ui & 0x000003ff); + return; + } + + /* If the ui field is symmetric in each byte, then we can use + * the Exclusive Or Byte Immediate instruction. + */ + tmp = ui & 0x000000ff; + if ((ui >> 24) == tmp && ((ui >> 16) & 0xff) == tmp && ((ui >> 8) & 0xff) == tmp) { + spe_xorbi(p, rT, rA, tmp); + return; + } + + /* Otherwise, we'll have to use a temporary register. */ + unsigned int tmp_reg = spe_allocate_available_register(p); + spe_load_uint(p, tmp_reg, ui); + spe_xor(p, rT, rA, tmp_reg); + spe_release_register(p, tmp_reg); +} + +void +spe_compare_equal_uint(struct spe_function *p, unsigned rT, unsigned rA, unsigned int ui) +{ + /* If the comparison value is 9 bits or less, it fits inside a + * Compare Equal Word Immediate instruction. + */ + if ((ui & 0x000001ff) == ui) { + spe_ceqi(p, rT, rA, ui); + } + /* Otherwise, we're going to have to load a word first. */ + else { + unsigned int tmp_reg = spe_allocate_available_register(p); + spe_load_uint(p, tmp_reg, ui); + spe_ceq(p, rT, rA, tmp_reg); + spe_release_register(p, tmp_reg); + } +} + +void +spe_compare_greater_uint(struct spe_function *p, unsigned rT, unsigned rA, unsigned int ui) +{ + /* If the comparison value is 10 bits or less, it fits inside a + * Compare Logical Greater Than Word Immediate instruction. + */ + if ((ui & 0x000003ff) == ui) { + spe_clgti(p, rT, rA, ui); + } + /* Otherwise, we're going to have to load a word first. */ + else { + unsigned int tmp_reg = spe_allocate_available_register(p); + spe_load_uint(p, tmp_reg, ui); + spe_clgt(p, rT, rA, tmp_reg); + spe_release_register(p, tmp_reg); + } +} void spe_splat(struct spe_function *p, unsigned rT, unsigned rA) diff --git a/src/gallium/auxiliary/rtasm/rtasm_ppc_spe.h b/src/gallium/auxiliary/rtasm/rtasm_ppc_spe.h index 61c7edeb604..d6a3c02f20c 100644 --- a/src/gallium/auxiliary/rtasm/rtasm_ppc_spe.h +++ b/src/gallium/auxiliary/rtasm/rtasm_ppc_spe.h @@ -53,17 +53,26 @@ struct spe_function uint num_inst; uint max_inst; - /** - * Mask of used / unused registers - * - * Each set bit corresponds to an available register. Each cleared bit - * corresponds to an allocated register. + /** + * The "set count" reflects the number of nested register sets + * are allowed. In the unlikely case that we exceed the set count, + * register allocation will start to be confused, which is critical + * enough that we check for it. + */ + unsigned char set_count; + + /** + * Flags for used and unused registers. Each byte corresponds to a + * register; a 0 in that byte means that the register is available. + * A value of 1 means that the register was allocated in the current + * register set. Any other value N means that the register was allocated + * N register sets ago. * * \sa * spe_allocate_register, spe_allocate_available_register, - * spe_release_register + * spe_allocate_register_set, spe_release_register_set, spe_release_register, */ - uint64_t regs[SPE_NUM_REGS / 64]; + unsigned char regs[SPE_NUM_REGS]; boolean print; /**< print/dump instructions as they're emitted? */ int indent; /**< number of spaces to indent */ @@ -77,6 +86,11 @@ extern unsigned spe_code_size(const struct spe_function *p); extern int spe_allocate_available_register(struct spe_function *p); extern int spe_allocate_register(struct spe_function *p, int reg); extern void spe_release_register(struct spe_function *p, int reg); +extern void spe_allocate_register_set(struct spe_function *p); +extern void spe_release_register_set(struct spe_function *p); + +extern unsigned +spe_get_registers_used(const struct spe_function *p, ubyte used[]); extern void spe_print_code(struct spe_function *p, boolean enable); extern void spe_indent(struct spe_function *p, int spaces); @@ -105,6 +119,9 @@ extern void spe_comment(struct spe_function *p, int rel_indent, const char *s); #define EMIT_RI10(_name, _op) \ extern void _name (struct spe_function *p, unsigned rT, unsigned rA, \ int imm) +#define EMIT_RI10s(_name, _op) \ + extern void _name (struct spe_function *p, unsigned rT, unsigned rA, \ + int imm) #define EMIT_RI16(_name, _op) \ extern void _name (struct spe_function *p, unsigned rT, int imm) #define EMIT_RI18(_name, _op) \ @@ -117,11 +134,9 @@ extern void spe_comment(struct spe_function *p, int rel_indent, const char *s); /* Memory load / store instructions */ -EMIT_RI10(spe_lqd, 0x034); EMIT_RR (spe_lqx, 0x1c4); EMIT_RI16(spe_lqa, 0x061); EMIT_RI16(spe_lqr, 0x067); -EMIT_RI10(spe_stqd, 0x024); EMIT_RR (spe_stqx, 0x144); EMIT_RI16(spe_stqa, 0x041); EMIT_RI16(spe_stqr, 0x047); @@ -151,7 +166,7 @@ EMIT_RI16(spe_fsmbi, 0x065); EMIT_RR (spe_ah, 0x0c8); EMIT_RI10(spe_ahi, 0x01d); EMIT_RR (spe_a, 0x0c0); -EMIT_RI10(spe_ai, 0x01c); +EMIT_RI10s(spe_ai, 0x01c); EMIT_RR (spe_sfh, 0x048); EMIT_RI10(spe_sfhi, 0x00d); EMIT_RR (spe_sf, 0x040); @@ -189,19 +204,19 @@ EMIT_R (spe_xshw, 0x2ae); EMIT_R (spe_xswd, 0x2a6); EMIT_RR (spe_and, 0x0c1); EMIT_RR (spe_andc, 0x2c1); -EMIT_RI10(spe_andbi, 0x016); -EMIT_RI10(spe_andhi, 0x015); -EMIT_RI10(spe_andi, 0x014); +EMIT_RI10s(spe_andbi, 0x016); +EMIT_RI10s(spe_andhi, 0x015); +EMIT_RI10s(spe_andi, 0x014); EMIT_RR (spe_or, 0x041); EMIT_RR (spe_orc, 0x2c9); -EMIT_RI10(spe_orbi, 0x006); -EMIT_RI10(spe_orhi, 0x005); -EMIT_RI10(spe_ori, 0x004); +EMIT_RI10s(spe_orbi, 0x006); +EMIT_RI10s(spe_orhi, 0x005); +EMIT_RI10s(spe_ori, 0x004); EMIT_R (spe_orx, 0x1f0); EMIT_RR (spe_xor, 0x241); -EMIT_RI10(spe_xorbi, 0x026); -EMIT_RI10(spe_xorhi, 0x025); -EMIT_RI10(spe_xori, 0x024); +EMIT_RI10s(spe_xorbi, 0x026); +EMIT_RI10s(spe_xorhi, 0x025); +EMIT_RI10s(spe_xori, 0x024); EMIT_RR (spe_nand, 0x0c9); EMIT_RR (spe_nor, 0x049); EMIT_RR (spe_eqv, 0x249); @@ -279,6 +294,12 @@ EMIT_RI16(spe_brz, 0x040); EMIT_RI16(spe_brhnz, 0x046); EMIT_RI16(spe_brhz, 0x044); +extern void +spe_lqd(struct spe_function *p, unsigned rT, unsigned rA, int offset); + +extern void +spe_stqd(struct spe_function *p, unsigned rT, unsigned rA, int offset); + extern void spe_bi(struct spe_function *p, unsigned rA, int d, int e); extern void spe_iret(struct spe_function *p, unsigned rA, int d, int e); extern void spe_bisled(struct spe_function *p, unsigned rT, unsigned rA, @@ -307,6 +328,22 @@ spe_load_int(struct spe_function *p, unsigned rT, int i); extern void spe_load_uint(struct spe_function *p, unsigned rT, unsigned int ui); +/** And immediate value into rT. */ +extern void +spe_and_uint(struct spe_function *p, unsigned rT, unsigned rA, unsigned int ui); + +/** Xor immediate value into rT. */ +extern void +spe_xor_uint(struct spe_function *p, unsigned rT, unsigned rA, unsigned int ui); + +/** Compare equal with immediate value. */ +extern void +spe_compare_equal_uint(struct spe_function *p, unsigned rT, unsigned rA, unsigned int ui); + +/** Compare greater with immediate value. */ +extern void +spe_compare_greater_uint(struct spe_function *p, unsigned rT, unsigned rA, unsigned int ui); + /** Replicate word 0 of rA across rT. */ extern void spe_splat(struct spe_function *p, unsigned rT, unsigned rA); @@ -388,6 +425,7 @@ EMIT_R (spe_wrch, 0x10d); #undef EMIT_RI7 #undef EMIT_RI8 #undef EMIT_RI10 +#undef EMIT_RI10s #undef EMIT_RI16 #undef EMIT_RI18 #undef EMIT_I16 diff --git a/src/gallium/auxiliary/rtasm/rtasm_x86sse.c b/src/gallium/auxiliary/rtasm/rtasm_x86sse.c index 3bba9dcc076..99ee74cf14b 100644 --- a/src/gallium/auxiliary/rtasm/rtasm_x86sse.c +++ b/src/gallium/auxiliary/rtasm/rtasm_x86sse.c @@ -371,7 +371,11 @@ void x86_jcc( struct x86_function *p, DUMP_I(cc); if (offset < 0) { - assert(p->csr - p->store > -offset); + /*assert(p->csr - p->store > -offset);*/ + if (p->csr - p->store <= -offset) { + /* probably out of memory (using the error_overflow buffer) */ + return; + } } if (offset <= 127 && offset >= -128) { @@ -675,6 +679,44 @@ void x86_and( struct x86_function *p, * SSE instructions */ +void sse_prefetchnta( struct x86_function *p, struct x86_reg ptr) +{ + DUMP_R( ptr ); + assert(ptr.mod != mod_REG); + emit_2ub(p, 0x0f, 0x18); + emit_modrm_noreg(p, 0, ptr); +} + +void sse_prefetch0( struct x86_function *p, struct x86_reg ptr) +{ + DUMP_R( ptr ); + assert(ptr.mod != mod_REG); + emit_2ub(p, 0x0f, 0x18); + emit_modrm_noreg(p, 1, ptr); +} + +void sse_prefetch1( struct x86_function *p, struct x86_reg ptr) +{ + DUMP_R( ptr ); + assert(ptr.mod != mod_REG); + emit_2ub(p, 0x0f, 0x18); + emit_modrm_noreg(p, 2, ptr); +} + +void sse_movntps( struct x86_function *p, + struct x86_reg dst, + struct x86_reg src) +{ + DUMP_RR( dst, src ); + + assert(dst.mod != mod_REG); + assert(src.mod == mod_REG); + emit_2ub(p, 0x0f, 0x2b); + emit_modrm(p, src, dst); +} + + + void sse_movss( struct x86_function *p, struct x86_reg dst, diff --git a/src/gallium/auxiliary/rtasm/rtasm_x86sse.h b/src/gallium/auxiliary/rtasm/rtasm_x86sse.h index 510aa1b0dec..1b5eaaca850 100644 --- a/src/gallium/auxiliary/rtasm/rtasm_x86sse.h +++ b/src/gallium/auxiliary/rtasm/rtasm_x86sse.h @@ -185,6 +185,13 @@ void sse2_pshufd( struct x86_function *p, struct x86_reg dest, struct x86_reg ar void sse2_rcpps( struct x86_function *p, struct x86_reg dst, struct x86_reg src ); void sse2_rcpss( struct x86_function *p, struct x86_reg dst, struct x86_reg src ); + +void sse_prefetchnta( struct x86_function *p, struct x86_reg ptr); +void sse_prefetch0( struct x86_function *p, struct x86_reg ptr); +void sse_prefetch1( struct x86_function *p, struct x86_reg ptr); + +void sse_movntps( struct x86_function *p, struct x86_reg dst, struct x86_reg src); + void sse_addps( struct x86_function *p, struct x86_reg dst, struct x86_reg src ); void sse_addss( struct x86_function *p, struct x86_reg dst, struct x86_reg src ); void sse_cvtps2pi( struct x86_function *p, struct x86_reg dst, struct x86_reg src ); diff --git a/src/gallium/auxiliary/tgsi/tgsi_sse2.c b/src/gallium/auxiliary/tgsi/tgsi_sse2.c index 4fdad3a5c78..f79170b9d65 100644 --- a/src/gallium/auxiliary/tgsi/tgsi_sse2.c +++ b/src/gallium/auxiliary/tgsi/tgsi_sse2.c @@ -25,6 +25,10 @@ * **************************************************************************/ +#include "pipe/p_config.h" + +#if defined(PIPE_ARCH_X86) && defined(PIPE_ARCH_SSE) + #include "pipe/p_debug.h" #include "pipe/p_shader_tokens.h" #include "util/u_math.h" @@ -36,8 +40,6 @@ #include "rtasm/rtasm_x86sse.h" -#ifdef PIPE_ARCH_X86 - /* for 1/sqrt() * * This costs about 100fps (close to 10%) in gears: diff --git a/src/gallium/auxiliary/util/Makefile b/src/gallium/auxiliary/util/Makefile index d3951e4e7d7..b3d1045a8f4 100644 --- a/src/gallium/auxiliary/util/Makefile +++ b/src/gallium/auxiliary/util/Makefile @@ -10,6 +10,7 @@ C_SOURCES = \ u_gen_mipmap.c \ u_handle_table.c \ u_hash_table.c \ + u_keymap.c \ u_math.c \ u_mm.c \ u_rect.c \ diff --git a/src/gallium/auxiliary/util/SConscript b/src/gallium/auxiliary/util/SConscript index e65c17b1cc8..8a04955a16e 100644 --- a/src/gallium/auxiliary/util/SConscript +++ b/src/gallium/auxiliary/util/SConscript @@ -11,13 +11,14 @@ util = env.ConvenienceLibrary( 'u_gen_mipmap.c', 'u_handle_table.c', 'u_hash_table.c', + 'u_keymap.c', 'u_math.c', 'u_mm.c', 'u_rect.c', 'u_simple_shaders.c', 'u_snprintf.c', - 'u_stream_stdc.c', - 'u_stream_wd.c', + 'u_stream_stdc.c', + 'u_stream_wd.c', 'u_tile.c', 'u_time.c', ]) diff --git a/src/gallium/auxiliary/util/u_keymap.c b/src/gallium/auxiliary/util/u_keymap.c new file mode 100644 index 00000000000..01b17ddb1b3 --- /dev/null +++ b/src/gallium/auxiliary/util/u_keymap.c @@ -0,0 +1,309 @@ +/************************************************************************** + * + * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +/** + * Key lookup/associative container. + * + * Like Jose's u_hash_table, based on CSO cache code for now. + * + * Author: Brian Paul + */ + + +#include "pipe/p_compiler.h" +#include "pipe/p_debug.h" +#include "pipe/p_error.h" + +#include "cso_cache/cso_hash.h" + +#include "util/u_memory.h" +#include "util/u_keymap.h" + + +struct keymap +{ + struct cso_hash *cso; + unsigned key_size; + unsigned max_entries; /* XXX not obeyed net */ + unsigned num_entries; + keymap_delete_func delete_func; +}; + + +struct keymap_item +{ + void *key, *value; +}; + + +/** + * This the default key-delete function used when the client doesn't + * provide one. + */ +static void +default_delete_func(const struct keymap *map, + const void *key, void *data, void *user) +{ + FREE((void*) data); +} + + +static INLINE struct keymap_item * +hash_table_item(struct cso_hash_iter iter) +{ + return (struct keymap_item *) cso_hash_iter_data(iter); +} + + +/** + * Return 4-byte hash key for a block of bytes. + */ +static unsigned +hash(const void *key, unsigned keySize) +{ + unsigned i, hash; + + keySize /= 4; /* convert from bytes to uints */ + + hash = 0; + for (i = 0; i < keySize; i++) { + hash ^= (i + 1) * ((const unsigned *) key)[i]; + } + + /*hash = hash ^ (hash >> 11) ^ (hash >> 22);*/ + + return hash; +} + + +/** + * Create a new map. + * \param keySize size of the keys in bytes + * \param maxEntries max number of entries to allow (~0 = infinity) + * \param deleteFunc optional callback to call when entries + * are deleted/replaced + */ +struct keymap * +util_new_keymap(unsigned keySize, unsigned maxEntries, + keymap_delete_func deleteFunc) +{ + struct keymap *map = MALLOC_STRUCT(keymap); + if (!map) + return NULL; + + map->cso = cso_hash_create(); + if (!map->cso) { + FREE(map); + return NULL; + } + + map->max_entries = maxEntries; + map->num_entries = 0; + map->key_size = keySize; + map->delete_func = deleteFunc ? deleteFunc : default_delete_func; + + return map; +} + + +/** + * Delete/free a keymap and all entries. The deleteFunc that was given at + * create time will be called for each entry. + * \param user user-provided pointer passed through to the delete callback + */ +void +util_delete_keymap(struct keymap *map, void *user) +{ + util_keymap_remove_all(map, user); + cso_hash_delete(map->cso); + FREE(map); +} + + +static INLINE struct cso_hash_iter +hash_table_find_iter(const struct keymap *map, const void *key, + unsigned key_hash) +{ + struct cso_hash_iter iter; + struct keymap_item *item; + + iter = cso_hash_find(map->cso, key_hash); + while (!cso_hash_iter_is_null(iter)) { + item = (struct keymap_item *) cso_hash_iter_data(iter); + if (!memcmp(item->key, key, map->key_size)) + break; + iter = cso_hash_iter_next(iter); + } + + return iter; +} + + +static INLINE struct keymap_item * +hash_table_find_item(const struct keymap *map, const void *key, + unsigned key_hash) +{ + struct cso_hash_iter iter = hash_table_find_iter(map, key, key_hash); + if (cso_hash_iter_is_null(iter)) { + return NULL; + } + else { + return hash_table_item(iter); + } +} + + +/** + * Insert a new key + data pointer into the table. + * Note: we create a copy of the key, but not the data! + * If the key is already present in the table, replace the existing + * entry (calling the delete callback on the previous entry). + * If the maximum capacity of the map is reached an old entry + * will be deleted (the delete callback will be called). + */ +boolean +util_keymap_insert(struct keymap *map, const void *key, + const void *data, void *user) +{ + unsigned key_hash; + struct keymap_item *item; + struct cso_hash_iter iter; + + assert(map); + + key_hash = hash(key, map->key_size); + + item = hash_table_find_item(map, key, key_hash); + if (item) { + /* call delete callback for old entry/item */ + map->delete_func(map, item->key, item->value, user); + item->value = (void *) data; + return TRUE; + } + + item = MALLOC_STRUCT(keymap_item); + if (!item) + return FALSE; + + item->key = mem_dup(key, map->key_size); + item->value = (void *) data; + + iter = cso_hash_insert(map->cso, key_hash, item); + if (cso_hash_iter_is_null(iter)) { + FREE(item); + return FALSE; + } + + map->num_entries++; + + return TRUE; +} + + +/** + * Look up a key in the map and return the associated data pointer. + */ +const void * +util_keymap_lookup(const struct keymap *map, const void *key) +{ + unsigned key_hash; + struct keymap_item *item; + + assert(map); + + key_hash = hash(key, map->key_size); + + item = hash_table_find_item(map, key, key_hash); + if (!item) + return NULL; + + return item->value; +} + + +/** + * Remove an entry from the map. + * The delete callback will be called if the given key/entry is found. + * \param user passed to the delete callback as the last param. + */ +void +util_keymap_remove(struct keymap *map, const void *key, void *user) +{ + unsigned key_hash; + struct cso_hash_iter iter; + struct keymap_item *item; + + assert(map); + + key_hash = hash(key, map->key_size); + + iter = hash_table_find_iter(map, key, key_hash); + if (cso_hash_iter_is_null(iter)) + return; + + item = hash_table_item(iter); + assert(item); + map->delete_func(map, item->key, item->value, user); + FREE(item->key); + FREE(item); + + map->num_entries--; + + cso_hash_erase(map->cso, iter); +} + + +/** + * Remove all entries from the map, calling the delete callback for each. + * \param user passed to the delete callback as the last param. + */ +void +util_keymap_remove_all(struct keymap *map, void *user) +{ + struct cso_hash_iter iter; + struct keymap_item *item; + + assert(map); + + iter = cso_hash_first_node(map->cso); + while (!cso_hash_iter_is_null(iter)) { + item = (struct keymap_item *) + cso_hash_take(map->cso, cso_hash_iter_key(iter)); + map->delete_func(map, item->key, item->value, user); + FREE(item->key); + FREE(item); + iter = cso_hash_first_node(map->cso); + } +} + + +extern void +util_keymap_info(const struct keymap *map) +{ + debug_printf("Keymap %p: %u of max %u entries\n", + (void *) map, map->num_entries, map->max_entries); +} diff --git a/src/gallium/auxiliary/util/u_keymap.h b/src/gallium/auxiliary/util/u_keymap.h new file mode 100644 index 00000000000..8d60a76fc3c --- /dev/null +++ b/src/gallium/auxiliary/util/u_keymap.h @@ -0,0 +1,68 @@ +/************************************************************************** + * + * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#ifndef U_KEYMAP_H +#define U_KEYMAP_H + +#include "pipe/p_compiler.h" + + +/** opaque keymap type */ +struct keymap; + + +/** Delete/callback function type */ +typedef void (*keymap_delete_func)(const struct keymap *map, + const void *key, void *data, + void *user); + + +extern struct keymap * +util_new_keymap(unsigned keySize, unsigned maxEntries, + keymap_delete_func deleteFunc); + +extern void +util_delete_keymap(struct keymap *map, void *user); + +extern boolean +util_keymap_insert(struct keymap *map, const void *key, + const void *data, void *user); + +extern const void * +util_keymap_lookup(const struct keymap *map, const void *key); + +extern void +util_keymap_remove(struct keymap *map, const void *key, void *user); + +extern void +util_keymap_remove_all(struct keymap *map, void *user); + +extern void +util_keymap_info(const struct keymap *map); + + +#endif /* U_KEYMAP_H */ diff --git a/src/gallium/auxiliary/util/u_math.c b/src/gallium/auxiliary/util/u_math.c index 0729114d6a1..5b3cab4642a 100644 --- a/src/gallium/auxiliary/util/u_math.c +++ b/src/gallium/auxiliary/util/u_math.c @@ -30,7 +30,7 @@ #include "util/u_math.h" - +/** 2^x, for x in [-1.0, 1.0[ */ float pow2_table[POW2_TABLE_SIZE]; @@ -38,9 +38,21 @@ static void init_pow2_table(void) { int i; - for (i = 0; i < POW2_TABLE_SIZE; i++) { - pow2_table[i] = (float) pow(2.0, i / POW2_TABLE_SCALE); - } + for (i = 0; i < POW2_TABLE_SIZE; i++) + pow2_table[i] = (float) pow(2.0, (i - POW2_TABLE_OFFSET) / POW2_TABLE_SCALE); +} + + +/** log2(x), for x in [1.0, 2.0[ */ +float log2_table[LOG2_TABLE_SIZE]; + + +static void +init_log2_table(void) +{ + unsigned i; + for (i = 0; i < LOG2_TABLE_SIZE; i++) + log2_table[i] = (float) log2(1.0 + i * (1.0 / LOG2_TABLE_SIZE)); } @@ -53,6 +65,7 @@ util_init_math(void) static boolean initialized = FALSE; if (!initialized) { init_pow2_table(); + init_log2_table(); initialized = TRUE; } } diff --git a/src/gallium/auxiliary/util/u_math.h b/src/gallium/auxiliary/util/u_math.h index 084655e6ddc..be7303e5503 100644 --- a/src/gallium/auxiliary/util/u_math.h +++ b/src/gallium/auxiliary/util/u_math.h @@ -174,8 +174,10 @@ static INLINE float logf( float f ) -#define POW2_TABLE_SIZE 256 -#define POW2_TABLE_SCALE ((float) (POW2_TABLE_SIZE-1)) +#define POW2_TABLE_SIZE_LOG2 9 +#define POW2_TABLE_SIZE (1 << POW2_TABLE_SIZE_LOG2) +#define POW2_TABLE_OFFSET (POW2_TABLE_SIZE/2) +#define POW2_TABLE_SCALE ((float)(POW2_TABLE_SIZE/2)) extern float pow2_table[POW2_TABLE_SIZE]; @@ -186,98 +188,78 @@ util_init_math(void); union fi { float f; - int i; - unsigned ui; + int32_t i; + uint32_t ui; }; /** - * Fast approximation to exp(x). - * Compute with base 2 exponents: exp(x) = exp2(log2(e) * x) - * Note: log2(e) is a constant, k = 1.44269 - * So, exp(x) = exp2(k * x); + * Fast version of 2^x * Identity: exp2(a + b) = exp2(a) * exp2(b) - * Let ipart = int(k*x) - * Let fpart = k*x - ipart; - * So, exp2(k*x) = exp2(ipart) * exp2(fpart) + * Let ipart = int(x) + * Let fpart = x - ipart; + * So, exp2(x) = exp2(ipart) * exp2(fpart) * Compute exp2(ipart) with i << ipart * Compute exp2(fpart) with lookup table. */ static INLINE float -util_fast_exp(float x) +util_fast_exp2(float x) { - if (x >= 0.0f) { - float k = 1.44269f; /* = log2(e) */ - float kx = k * x; - int ipart = (int) kx; - float fpart = kx - (float) ipart; - float y = (float) (1 << ipart) - * pow2_table[(int) (fpart * POW2_TABLE_SCALE)]; - return y; - } - else { - /* exp(-x) = 1.0 / exp(x) */ - float k = -1.44269f; - float kx = k * x; - int ipart = (int) kx; - float fpart = kx - (float) ipart; - float y = (float) (1 << ipart) - * pow2_table[(int) (fpart * POW2_TABLE_SCALE)]; - return 1.0f / y; - } + int32_t ipart; + float fpart, mpart; + union fi epart; + + if(x > 129.00000f) + return 3.402823466e+38f; + + if(x < -126.99999f) + return 0.0f; + + ipart = (int32_t) x; + fpart = x - (float) ipart; + + /* same as + * epart.f = (float) (1 << ipart) + * but faster and without integer overflow for ipart > 31 */ + epart.i = (ipart + 127 ) << 23; + + mpart = pow2_table[POW2_TABLE_OFFSET + (int)(fpart * POW2_TABLE_SCALE)]; + + return epart.f * mpart; } /** - * Fast version of 2^x - * XXX the above function could be implemented in terms of this one. + * Fast approximation to exp(x). */ static INLINE float -util_fast_exp2(float x) +util_fast_exp(float x) { - if (x >= 0.0f) { - int ipart = (int) x; - float fpart = x - (float) ipart; - float y = (float) (1 << ipart) - * pow2_table[(int) (fpart * POW2_TABLE_SCALE)]; - return y; - } - else { - /* exp(-x) = 1.0 / exp(x) */ - int ipart = (int) -x; - float fpart = -x - (float) ipart; - float y = (float) (1 << ipart) - * pow2_table[(int) (fpart * POW2_TABLE_SCALE)]; - return 1.0f / y; - } + const float k = 1.44269f; /* = log2(e) */ + return util_fast_exp2(k * x); } -/** - * Based on code from http://www.flipcode.com/archives/Fast_log_Function.shtml - */ +#define LOG2_TABLE_SIZE_LOG2 8 +#define LOG2_TABLE_SIZE (1 << LOG2_TABLE_SIZE_LOG2) +extern float log2_table[LOG2_TABLE_SIZE]; + + static INLINE float -util_fast_log2(float val) +util_fast_log2(float x) { union fi num; - int log_2; - num.f = val; - log_2 = ((num.i >> 23) & 255) - 128; - num.i &= ~(255 << 23); - num.i += 127 << 23; - num.f = ((-1.0f/3) * num.f + 2) * num.f - 2.0f/3; - return num.f + log_2; + float epart, mpart; + num.f = x; + epart = (float)(((num.i & 0x7f800000) >> 23) - 127); + mpart = log2_table[(num.i & 0x007fffff) >> (23 - LOG2_TABLE_SIZE_LOG2)]; + return epart + mpart; } static INLINE float util_fast_pow(float x, float y) { - /* XXX these tests may need adjustment */ - if (y >= 3.0f && (-0.02f <= x && x <= 0.02f)) - return 0.0f; - if (y >= 50.0f && (-0.9f <= x && x <= 0.9f)) - return 0.0f; return util_fast_exp2(util_fast_log2(x) * y); } diff --git a/src/gallium/auxiliary/util/u_sse.h b/src/gallium/auxiliary/util/u_sse.h index 0c8356cd057..e2a8491e62c 100644 --- a/src/gallium/auxiliary/util/u_sse.h +++ b/src/gallium/auxiliary/util/u_sse.h @@ -37,6 +37,10 @@ #ifndef U_SSE_H_ #define U_SSE_H_ +#include "pipe/p_config.h" + +#if defined(PIPE_ARCH_SSE) + #include <xmmintrin.h> #include <emmintrin.h> @@ -66,7 +70,8 @@ _mm_castps_si128(__m128 a) return u.m128i; } -#endif +#endif /* defined(_MSC_VER) && _MSC_VER < 1500 */ +#endif /* PIPE_ARCH_X86 || PIPE_ARCH_X86_64 */ #endif /* U_SSE_H_ */ diff --git a/src/gallium/drivers/cell/common.h b/src/gallium/drivers/cell/common.h index 99329fd8e22..b0169b8e329 100644 --- a/src/gallium/drivers/cell/common.h +++ b/src/gallium/drivers/cell/common.h @@ -67,6 +67,7 @@ #define CELL_MAX_SPUS 6 #define CELL_MAX_SAMPLERS 4 +#define CELL_MAX_TEXTURE_LEVELS 12 /* 2k x 2k */ #define TILE_SIZE 32 @@ -94,6 +95,7 @@ #define CELL_CMD_STATE_BIND_VS 18 #define CELL_CMD_STATE_FRAGMENT_PROGRAM 19 #define CELL_CMD_STATE_ATTRIB_FETCH 20 +#define CELL_CMD_STATE_FS_CONSTANTS 21 #define CELL_CMD_VS_EXECUTE 22 #define CELL_CMD_FLUSH_BUFFER_RANGE 23 @@ -127,7 +129,7 @@ struct cell_command_fragment_ops /** Max instructions for fragment programs */ -#define SPU_MAX_FRAGMENT_PROGRAM_INSTS 128 +#define SPU_MAX_FRAGMENT_PROGRAM_INSTS 512 /** * Command to send a fragment program to SPUs. @@ -227,6 +229,7 @@ struct cell_command_render float xmin, ymin, xmax, ymax; /* XXX another dummy field */ uint min_index; boolean inline_verts; + uint front_winding; /* the rasterizer needs to be able to determine facing to apply front/back-facing stencil */ }; @@ -248,9 +251,12 @@ struct cell_command_sampler struct cell_command_texture { uint64_t opcode; /**< CELL_CMD_STATE_TEXTURE */ + uint target; /**< PIPE_TEXTURE_x */ uint unit; - void *start; /**< Address in main memory */ - ushort width, height; + void *start[CELL_MAX_TEXTURE_LEVELS]; /**< Address in main memory */ + ushort width[CELL_MAX_TEXTURE_LEVELS]; + ushort height[CELL_MAX_TEXTURE_LEVELS]; + ushort depth[CELL_MAX_TEXTURE_LEVELS]; }; diff --git a/src/gallium/drivers/cell/ppu/cell_context.c b/src/gallium/drivers/cell/ppu/cell_context.c index 62e213ea354..b66aa9c9d99 100644 --- a/src/gallium/drivers/cell/ppu/cell_context.c +++ b/src/gallium/drivers/cell/ppu/cell_context.c @@ -62,6 +62,8 @@ cell_destroy_context( struct pipe_context *pipe ) { struct cell_context *cell = cell_context(pipe); + util_delete_keymap(cell->fragment_ops_cache, NULL); + cell_spu_exit(cell); align_free(cell); @@ -126,11 +128,14 @@ cell_create_context(struct pipe_screen *screen, cell_init_state_functions(cell); cell_init_shader_functions(cell); cell_init_surface_functions(cell); - cell_init_texture_functions(cell); cell_init_vertex_functions(cell); cell->draw = cell_draw_create(cell); + /* Create cache of fragment ops generated code */ + cell->fragment_ops_cache = + util_new_keymap(sizeof(struct cell_fragment_ops_key), ~0, NULL); + cell_init_vbuf(cell); draw_set_rasterize_stage(cell->draw, cell->vbuf); @@ -156,5 +161,8 @@ cell_create_context(struct pipe_screen *screen, cell_init_batch_buffers(cell); + /* make sure SPU initializations are done before proceeding */ + cell_flush_int(cell, CELL_FLUSH_WAIT); + return &cell->pipe; } diff --git a/src/gallium/drivers/cell/ppu/cell_context.h b/src/gallium/drivers/cell/ppu/cell_context.h index 3dc15c9233c..80a9b3d7e13 100644 --- a/src/gallium/drivers/cell/ppu/cell_context.h +++ b/src/gallium/drivers/cell/ppu/cell_context.h @@ -38,6 +38,7 @@ #include "cell/common.h" #include "rtasm/rtasm_ppc_spe.h" #include "tgsi/tgsi_scan.h" +#include "util/u_keymap.h" struct cell_vbuf_render; @@ -67,6 +68,19 @@ struct cell_fragment_shader_state /** + * Key for mapping per-fragment state to cached SPU machine code. + * keymap(cell_fragment_ops_key) => cell_command_fragment_ops + */ +struct cell_fragment_ops_key +{ + struct pipe_blend_state blend; + struct pipe_depth_stencil_alpha_state dsa; + enum pipe_format color_format; + enum pipe_format zs_format; +}; + + +/** * Per-context state, subclass of pipe_context. */ struct cell_context @@ -107,6 +121,9 @@ struct cell_context uint dirty; + /** Cache of code generated for per-fragment ops */ + struct keymap *fragment_ops_cache; + /** The primitive drawing context */ struct draw_context *draw; struct draw_stage *render_stage; diff --git a/src/gallium/drivers/cell/ppu/cell_gen_fp.c b/src/gallium/drivers/cell/ppu/cell_gen_fp.c index 8d2d4f2a0f2..3dfd5f673dd 100644 --- a/src/gallium/drivers/cell/ppu/cell_gen_fp.c +++ b/src/gallium/drivers/cell/ppu/cell_gen_fp.c @@ -77,13 +77,15 @@ struct codegen /** Per-instruction temps / intermediate temps */ int num_itemps; - int itemps[10]; + int itemps[12]; /** Current IF/ELSE/ENDIF nesting level */ int if_nesting; /** Index of execution mask register */ int exec_mask_reg; + int frame_size; /**< Stack frame size, in words */ + struct spe_function *f; boolean error; }; @@ -165,6 +167,37 @@ get_exec_mask_reg(struct codegen *gen) } +static boolean +is_register_src(struct codegen *gen, int channel, + const struct tgsi_full_src_register *src) +{ + int swizzle = tgsi_util_get_full_src_register_extswizzle(src, channel); + int sign_op = tgsi_util_get_full_src_register_sign_mode(src, channel); + + if (swizzle > TGSI_SWIZZLE_W || sign_op != TGSI_UTIL_SIGN_KEEP) { + return FALSE; + } + if (src->SrcRegister.File == TGSI_FILE_TEMPORARY || + src->SrcRegister.File == TGSI_FILE_IMMEDIATE) { + return TRUE; + } + return FALSE; +} + + +static boolean +is_memory_dst(struct codegen *gen, int channel, + const struct tgsi_full_dst_register *dst) +{ + if (dst->DstRegister.File == TGSI_FILE_OUTPUT) { + return TRUE; + } + else { + return FALSE; + } +} + + /** * Return the index of the SPU temporary containing the named TGSI * source register. If the TGSI register is a TGSI_FILE_TEMPORARY we @@ -185,41 +218,48 @@ get_src_reg(struct codegen *gen, assert(swizzle >= TGSI_SWIZZLE_X); assert(swizzle <= TGSI_EXTSWIZZLE_ONE); - switch (src->SrcRegister.File) { - case TGSI_FILE_TEMPORARY: - reg = gen->temp_regs[src->SrcRegister.Index][swizzle]; - break; - case TGSI_FILE_INPUT: - { - if(swizzle == TGSI_EXTSWIZZLE_ONE) - { - /* Load const one float and early out */ - reg = get_const_one_reg(gen); - } - else if(swizzle == TGSI_EXTSWIZZLE_ZERO) + if (swizzle == TGSI_EXTSWIZZLE_ONE) { + /* Load const one float and early out */ + reg = get_const_one_reg(gen); + } + else if (swizzle == TGSI_EXTSWIZZLE_ZERO) { + /* Load const zero float and early out */ + reg = get_itemp(gen); + spe_xor(gen->f, reg, reg, reg); + } + else { + assert(swizzle < 4); + + switch (src->SrcRegister.File) { + case TGSI_FILE_TEMPORARY: + reg = gen->temp_regs[src->SrcRegister.Index][swizzle]; + break; + case TGSI_FILE_INPUT: { - /* Load const zero float and early out */ + /* offset is measured in quadwords, not bytes */ + int offset = src->SrcRegister.Index * 4 + swizzle; reg = get_itemp(gen); - spe_xor(gen->f, reg, reg, reg); + reg_is_itemp = TRUE; + /* Load: reg = memory[(machine_reg) + offset] */ + spe_lqd(gen->f, reg, gen->inputs_reg, offset * 16); } - else + break; + case TGSI_FILE_IMMEDIATE: + reg = gen->imm_regs[src->SrcRegister.Index][swizzle]; + break; + case TGSI_FILE_CONSTANT: { /* offset is measured in quadwords, not bytes */ int offset = src->SrcRegister.Index * 4 + swizzle; reg = get_itemp(gen); reg_is_itemp = TRUE; /* Load: reg = memory[(machine_reg) + offset] */ - spe_lqd(gen->f, reg, gen->inputs_reg, offset); + spe_lqd(gen->f, reg, gen->constants_reg, offset * 16); } + break; + default: + assert(0); } - break; - case TGSI_FILE_IMMEDIATE: - reg = gen->imm_regs[src->SrcRegister.Index][swizzle]; - break; - case TGSI_FILE_CONSTANT: - /* xxx fall-through for now / fix */ - default: - assert(0); } /* @@ -243,7 +283,7 @@ get_src_reg(struct codegen *gen, } /* mask with bit 31 set, the rest cleared */ - spe_load_int(gen->f, bit31mask_reg, (1 << 31)); + spe_load_uint(gen->f, bit31mask_reg, (1 << 31)); if (sign_op == TGSI_UTIL_SIGN_CLEAR) { spe_andc(gen->f, result_reg, reg, bit31mask_reg); @@ -318,6 +358,7 @@ store_dest_reg(struct codegen *gen, } else { /* we're not inside a condition or loop: do nothing special */ + } break; case TGSI_FILE_OUTPUT: @@ -330,17 +371,17 @@ store_dest_reg(struct codegen *gen, /* First read the current value from memory: * Load: curval = memory[(machine_reg) + offset] */ - spe_lqd(gen->f, curval_reg, gen->outputs_reg, offset); + spe_lqd(gen->f, curval_reg, gen->outputs_reg, offset * 16); /* Mix curval with newvalue according to exec mask: * d[i] = mask_reg[i] ? value_reg : d_reg */ spe_selb(gen->f, curval_reg, curval_reg, value_reg, exec_reg); /* Store: memory[(machine_reg) + offset] = curval */ - spe_stqd(gen->f, curval_reg, gen->outputs_reg, offset); + spe_stqd(gen->f, curval_reg, gen->outputs_reg, offset * 16); } else { /* Store: memory[(machine_reg) + offset] = reg */ - spe_stqd(gen->f, value_reg, gen->outputs_reg, offset); + spe_stqd(gen->f, value_reg, gen->outputs_reg, offset * 16); } } break; @@ -350,18 +391,95 @@ store_dest_reg(struct codegen *gen, } + +static void +emit_prologue(struct codegen *gen) +{ + gen->frame_size = 1024; /* XXX temporary, should be dynamic */ + + spe_comment(gen->f, -4, "Function prologue:"); + + /* save $lr on stack # stqd $lr,16($sp) */ + spe_stqd(gen->f, SPE_REG_RA, SPE_REG_SP, 16); + + if (gen->frame_size >= 512) { + /* offset is too large for ai instruction */ + int offset_reg = spe_allocate_available_register(gen->f); + int sp_reg = spe_allocate_available_register(gen->f); + /* offset = -framesize */ + spe_load_int(gen->f, offset_reg, -gen->frame_size); + /* sp = $sp */ + spe_move(gen->f, sp_reg, SPE_REG_SP); + /* $sp = $sp + offset_reg */ + spe_a(gen->f, SPE_REG_SP, SPE_REG_SP, offset_reg); + /* save $sp in stack frame */ + spe_stqd(gen->f, sp_reg, SPE_REG_SP, 0); + /* clean up */ + spe_release_register(gen->f, offset_reg); + spe_release_register(gen->f, sp_reg); + } + else { + /* save stack pointer # stqd $sp,-frameSize($sp) */ + spe_stqd(gen->f, SPE_REG_SP, SPE_REG_SP, -gen->frame_size); + + /* adjust stack pointer # ai $sp,$sp,-frameSize */ + spe_ai(gen->f, SPE_REG_SP, SPE_REG_SP, -gen->frame_size); + } +} + + +static void +emit_epilogue(struct codegen *gen) +{ + spe_comment(gen->f, -4, "Function epilogue:"); + + if (gen->frame_size >= 512) { + /* offset is too large for ai instruction */ + int offset_reg = spe_allocate_available_register(gen->f); + /* offset = framesize */ + spe_load_int(gen->f, offset_reg, gen->frame_size); + /* $sp = $sp + offset */ + spe_a(gen->f, SPE_REG_SP, SPE_REG_SP, offset_reg); + /* clean up */ + spe_release_register(gen->f, offset_reg); + } + else { + /* restore stack pointer # ai $sp,$sp,frameSize */ + spe_ai(gen->f, SPE_REG_SP, SPE_REG_SP, gen->frame_size); + } + + /* restore $lr # lqd $lr,16($sp) */ + spe_lqd(gen->f, SPE_REG_RA, SPE_REG_SP, 16); + + /* return from function call */ + spe_bi(gen->f, SPE_REG_RA, 0, 0); +} + + static boolean emit_MOV(struct codegen *gen, const struct tgsi_full_instruction *inst) { - int ch; + int ch, src_reg[4], dst_reg[4]; + spe_comment(gen->f, -4, "MOV:"); for (ch = 0; ch < 4; ch++) { if (inst->FullDstRegisters[0].DstRegister.WriteMask & (1 << ch)) { - int src_reg = get_src_reg(gen, ch, &inst->FullSrcRegisters[0]); - int dst_reg = get_dst_reg(gen, ch, &inst->FullDstRegisters[0]); - /* XXX we don't always need to actually emit a mov instruction here */ - spe_move(gen->f, dst_reg, src_reg); - store_dest_reg(gen, dst_reg, ch, &inst->FullDstRegisters[0]); + src_reg[ch] = get_src_reg(gen, ch, &inst->FullSrcRegisters[0]); + dst_reg[ch] = get_dst_reg(gen, ch, &inst->FullDstRegisters[0]); + } + } + + for (ch = 0; ch < 4; ch++) { + if (inst->FullDstRegisters[0].DstRegister.WriteMask & (1 << ch)) { + if (is_register_src(gen, ch, &inst->FullSrcRegisters[0]) && + is_memory_dst(gen, ch, &inst->FullDstRegisters[0])) { + /* special-case: register to memory store */ + store_dest_reg(gen, src_reg[ch], ch, &inst->FullDstRegisters[0]); + } + else { + spe_move(gen->f, dst_reg[ch], src_reg[ch]); + store_dest_reg(gen, dst_reg[ch], ch, &inst->FullDstRegisters[0]); + } free_itemps(gen); } } @@ -376,22 +494,25 @@ emit_MOV(struct codegen *gen, const struct tgsi_full_instruction *inst) static boolean emit_ADD(struct codegen *gen, const struct tgsi_full_instruction *inst) { - int ch; + int ch, s1_reg[4], s2_reg[4], d_reg[4]; + spe_comment(gen->f, -4, "ADD:"); - /* Loop over Red/Green/Blue/Alpha channels */ + /* Loop over Red/Green/Blue/Alpha channels, fetch src operands */ for (ch = 0; ch < 4; ch++) { /* If the dest R, G, B or A writemask is enabled... */ if (inst->FullDstRegisters[0].DstRegister.WriteMask & (1 << ch)) { - /* get indexes of the two src, one dest SPE registers */ - int s1_reg = get_src_reg(gen, ch, &inst->FullSrcRegisters[0]); - int s2_reg = get_src_reg(gen, ch, &inst->FullSrcRegisters[1]); - int d_reg = get_dst_reg(gen, ch, &inst->FullDstRegisters[0]); - + s1_reg[ch] = get_src_reg(gen, ch, &inst->FullSrcRegisters[0]); + s2_reg[ch] = get_src_reg(gen, ch, &inst->FullSrcRegisters[1]); + d_reg[ch] = get_dst_reg(gen, ch, &inst->FullDstRegisters[0]); + } + } + /* Loop over Red/Green/Blue/Alpha channels, do the add, store results */ + for (ch = 0; ch < 4; ch++) { + if (inst->FullDstRegisters[0].DstRegister.WriteMask & (1 << ch)) { /* Emit actual SPE instruction: d = s1 + s2 */ - spe_fa(gen->f, d_reg, s1_reg, s2_reg); - + spe_fa(gen->f, d_reg[ch], s1_reg[ch], s2_reg[ch]); /* Store the result (a no-op for TGSI_FILE_TEMPORARY dests) */ - store_dest_reg(gen, d_reg, ch, &inst->FullDstRegisters[0]); + store_dest_reg(gen, d_reg[ch], ch, &inst->FullDstRegisters[0]); /* Free any intermediate temps we allocated */ free_itemps(gen); } @@ -405,23 +526,20 @@ emit_ADD(struct codegen *gen, const struct tgsi_full_instruction *inst) static boolean emit_SUB(struct codegen *gen, const struct tgsi_full_instruction *inst) { - int ch; + int ch, s1_reg[4], s2_reg[4], d_reg[4]; spe_comment(gen->f, -4, "SUB:"); - /* Loop over Red/Green/Blue/Alpha channels */ for (ch = 0; ch < 4; ch++) { - /* If the dest R, G, B or A writemask is enabled... */ if (inst->FullDstRegisters[0].DstRegister.WriteMask & (1 << ch)) { - /* get indexes of the two src, one dest SPE registers */ - int s1_reg = get_src_reg(gen, ch, &inst->FullSrcRegisters[0]); - int s2_reg = get_src_reg(gen, ch, &inst->FullSrcRegisters[1]); - int d_reg = get_dst_reg(gen, ch, &inst->FullDstRegisters[0]); - - /* Emit actual SPE instruction: d = s1 - s2 */ - spe_fs(gen->f, d_reg, s1_reg, s2_reg); - - /* Store the result (a no-op for TGSI_FILE_TEMPORARY dests) */ - store_dest_reg(gen, d_reg, ch, &inst->FullDstRegisters[0]); - /* Free any intermediate temps we allocated */ + s1_reg[ch] = get_src_reg(gen, ch, &inst->FullSrcRegisters[0]); + s2_reg[ch] = get_src_reg(gen, ch, &inst->FullSrcRegisters[1]); + d_reg[ch] = get_dst_reg(gen, ch, &inst->FullDstRegisters[0]); + } + } + for (ch = 0; ch < 4; ch++) { + if (inst->FullDstRegisters[0].DstRegister.WriteMask & (1 << ch)) { + /* d = s1 - s2 */ + spe_fs(gen->f, d_reg[ch], s1_reg[ch], s2_reg[ch]); + store_dest_reg(gen, d_reg[ch], ch, &inst->FullDstRegisters[0]); free_itemps(gen); } } @@ -434,17 +552,21 @@ emit_SUB(struct codegen *gen, const struct tgsi_full_instruction *inst) static boolean emit_MAD(struct codegen *gen, const struct tgsi_full_instruction *inst) { - int ch; + int ch, s1_reg[4], s2_reg[4], s3_reg[4], d_reg[4]; spe_comment(gen->f, -4, "MAD:"); for (ch = 0; ch < 4; ch++) { if (inst->FullDstRegisters[0].DstRegister.WriteMask & (1 << ch)) { - int s1_reg = get_src_reg(gen, ch, &inst->FullSrcRegisters[0]); - int s2_reg = get_src_reg(gen, ch, &inst->FullSrcRegisters[1]); - int s3_reg = get_src_reg(gen, ch, &inst->FullSrcRegisters[2]); - int d_reg = get_dst_reg(gen, ch, &inst->FullDstRegisters[0]); + s1_reg[ch] = get_src_reg(gen, ch, &inst->FullSrcRegisters[0]); + s2_reg[ch] = get_src_reg(gen, ch, &inst->FullSrcRegisters[1]); + s3_reg[ch] = get_src_reg(gen, ch, &inst->FullSrcRegisters[2]); + d_reg[ch] = get_dst_reg(gen, ch, &inst->FullDstRegisters[0]); + } + } + for (ch = 0; ch < 4; ch++) { + if (inst->FullDstRegisters[0].DstRegister.WriteMask & (1 << ch)) { /* d = s1 * s2 + s3 */ - spe_fma(gen->f, d_reg, s1_reg, s2_reg, s3_reg); - store_dest_reg(gen, d_reg, ch, &inst->FullDstRegisters[0]); + spe_fma(gen->f, d_reg[ch], s1_reg[ch], s2_reg[ch], s3_reg[ch]); + store_dest_reg(gen, d_reg[ch], ch, &inst->FullDstRegisters[0]); free_itemps(gen); } } @@ -458,21 +580,37 @@ emit_MAD(struct codegen *gen, const struct tgsi_full_instruction *inst) static boolean emit_LERP(struct codegen *gen, const struct tgsi_full_instruction *inst) { - int ch; + int ch, s1_reg[4], s2_reg[4], s3_reg[4], d_reg[4], tmp_reg[4]; spe_comment(gen->f, -4, "LERP:"); + /* setup/get src/dst/temp regs */ for (ch = 0; ch < 4; ch++) { if (inst->FullDstRegisters[0].DstRegister.WriteMask & (1 << ch)) { - int s1_reg = get_src_reg(gen, ch, &inst->FullSrcRegisters[0]); - int s2_reg = get_src_reg(gen, ch, &inst->FullSrcRegisters[1]); - int s3_reg = get_src_reg(gen, ch, &inst->FullSrcRegisters[2]); - int d_reg = get_dst_reg(gen, ch, &inst->FullDstRegisters[0]); - /* d = s3 + s1(s2 - s3) */ - spe_fs(gen->f, d_reg, s2_reg, s3_reg); - spe_fma(gen->f, d_reg, d_reg, s1_reg, s3_reg); - store_dest_reg(gen, d_reg, ch, &inst->FullDstRegisters[0]); - free_itemps(gen); + s1_reg[ch] = get_src_reg(gen, ch, &inst->FullSrcRegisters[0]); + s2_reg[ch] = get_src_reg(gen, ch, &inst->FullSrcRegisters[1]); + s3_reg[ch] = get_src_reg(gen, ch, &inst->FullSrcRegisters[2]); + d_reg[ch] = get_dst_reg(gen, ch, &inst->FullDstRegisters[0]); + tmp_reg[ch] = get_itemp(gen); + } + } + + /* d = s3 + s1(s2 - s3) */ + /* do all subtracts, then all fma, then all stores to better pipeline */ + for (ch = 0; ch < 4; ch++) { + if (inst->FullDstRegisters[0].DstRegister.WriteMask & (1 << ch)) { + spe_fs(gen->f, tmp_reg[ch], s2_reg[ch], s3_reg[ch]); + } + } + for (ch = 0; ch < 4; ch++) { + if (inst->FullDstRegisters[0].DstRegister.WriteMask & (1 << ch)) { + spe_fma(gen->f, d_reg[ch], tmp_reg[ch], s1_reg[ch], s3_reg[ch]); } } + for (ch = 0; ch < 4; ch++) { + if (inst->FullDstRegisters[0].DstRegister.WriteMask & (1 << ch)) { + store_dest_reg(gen, d_reg[ch], ch, &inst->FullDstRegisters[0]); + } + } + free_itemps(gen); return true; } @@ -482,16 +620,20 @@ emit_LERP(struct codegen *gen, const struct tgsi_full_instruction *inst) static boolean emit_MUL(struct codegen *gen, const struct tgsi_full_instruction *inst) { - int ch; + int ch, s1_reg[4], s2_reg[4], d_reg[4]; spe_comment(gen->f, -4, "MUL:"); for (ch = 0; ch < 4; ch++) { if (inst->FullDstRegisters[0].DstRegister.WriteMask & (1 << ch)) { - int s1_reg = get_src_reg(gen, ch, &inst->FullSrcRegisters[0]); - int s2_reg = get_src_reg(gen, ch, &inst->FullSrcRegisters[1]); - int d_reg = get_dst_reg(gen, ch, &inst->FullDstRegisters[0]); + s1_reg[ch] = get_src_reg(gen, ch, &inst->FullSrcRegisters[0]); + s2_reg[ch] = get_src_reg(gen, ch, &inst->FullSrcRegisters[1]); + d_reg[ch] = get_dst_reg(gen, ch, &inst->FullDstRegisters[0]); + } + } + for (ch = 0; ch < 4; ch++) { + if (inst->FullDstRegisters[0].DstRegister.WriteMask & (1 << ch)) { /* d = s1 * s2 */ - spe_fm(gen->f, d_reg, s1_reg, s2_reg); - store_dest_reg(gen, d_reg, ch, &inst->FullDstRegisters[0]); + spe_fm(gen->f, d_reg[ch], s1_reg[ch], s2_reg[ch]); + store_dest_reg(gen, d_reg[ch], ch, &inst->FullDstRegisters[0]); free_itemps(gen); } } @@ -557,7 +699,7 @@ emit_ABS(struct codegen *gen, const struct tgsi_full_instruction *inst) const int bit31mask_reg = get_itemp(gen); /* mask with bit 31 set, the rest cleared */ - spe_load_int(gen->f, bit31mask_reg, (1 << 31)); + spe_load_uint(gen->f, bit31mask_reg, (1 << 31)); /* d = sign bit cleared in s1 */ spe_andc(gen->f, d_reg, s1_reg, bit31mask_reg); @@ -576,27 +718,36 @@ static boolean emit_DP3(struct codegen *gen, const struct tgsi_full_instruction *inst) { int ch; + int s1x_reg, s1y_reg, s1z_reg; + int s2x_reg, s2y_reg, s2z_reg; + int t0_reg = get_itemp(gen), t1_reg = get_itemp(gen); + spe_comment(gen->f, -4, "DP3:"); - int s1_reg = get_src_reg(gen, CHAN_X, &inst->FullSrcRegisters[0]); - int s2_reg = get_src_reg(gen, CHAN_X, &inst->FullSrcRegisters[1]); - int tmp_reg = get_itemp(gen); - /* t = x0 * x1 */ - spe_fm(gen->f, tmp_reg, s1_reg, s2_reg); + s1x_reg = get_src_reg(gen, CHAN_X, &inst->FullSrcRegisters[0]); + s2x_reg = get_src_reg(gen, CHAN_X, &inst->FullSrcRegisters[1]); + s1y_reg = get_src_reg(gen, CHAN_Y, &inst->FullSrcRegisters[0]); + s2y_reg = get_src_reg(gen, CHAN_Y, &inst->FullSrcRegisters[1]); + s1z_reg = get_src_reg(gen, CHAN_Z, &inst->FullSrcRegisters[0]); + s2z_reg = get_src_reg(gen, CHAN_Z, &inst->FullSrcRegisters[1]); - s1_reg = get_src_reg(gen, CHAN_Y, &inst->FullSrcRegisters[0]); - s2_reg = get_src_reg(gen, CHAN_Y, &inst->FullSrcRegisters[1]); - /* t = y0 * y1 + t */ - spe_fma(gen->f, tmp_reg, s1_reg, s2_reg, tmp_reg); + /* t0 = x0 * x1 */ + spe_fm(gen->f, t0_reg, s1x_reg, s2x_reg); - s1_reg = get_src_reg(gen, CHAN_Z, &inst->FullSrcRegisters[0]); - s2_reg = get_src_reg(gen, CHAN_Z, &inst->FullSrcRegisters[1]); - /* t = z0 * z1 + t */ - spe_fma(gen->f, tmp_reg, s1_reg, s2_reg, tmp_reg); + /* t1 = y0 * y1 */ + spe_fm(gen->f, t1_reg, s1y_reg, s2y_reg); + + /* t0 = z0 * z1 + t0 */ + spe_fma(gen->f, t0_reg, s1z_reg, s2z_reg, t0_reg); + + /* t0 = t0 + t1 */ + spe_fa(gen->f, t0_reg, t0_reg, t1_reg); for (ch = 0; ch < 4; ch++) { if (inst->FullDstRegisters[0].DstRegister.WriteMask & (1 << ch)) { - store_dest_reg(gen, tmp_reg, ch, &inst->FullDstRegisters[0]); + int d_reg = get_dst_reg(gen, ch, &inst->FullDstRegisters[0]); + spe_move(gen->f, d_reg, t0_reg); + store_dest_reg(gen, d_reg, ch, &inst->FullDstRegisters[0]); } } @@ -611,32 +762,41 @@ static boolean emit_DP4(struct codegen *gen, const struct tgsi_full_instruction *inst) { int ch; + int s0x_reg, s0y_reg, s0z_reg, s0w_reg; + int s1x_reg, s1y_reg, s1z_reg, s1w_reg; + int t0_reg = get_itemp(gen), t1_reg = get_itemp(gen); + spe_comment(gen->f, -4, "DP4:"); - int s1_reg = get_src_reg(gen, CHAN_X, &inst->FullSrcRegisters[0]); - int s2_reg = get_src_reg(gen, CHAN_X, &inst->FullSrcRegisters[1]); - int tmp_reg = get_itemp(gen); - /* t = x0 * x1 */ - spe_fm(gen->f, tmp_reg, s1_reg, s2_reg); + s0x_reg = get_src_reg(gen, CHAN_X, &inst->FullSrcRegisters[0]); + s1x_reg = get_src_reg(gen, CHAN_X, &inst->FullSrcRegisters[1]); + s0y_reg = get_src_reg(gen, CHAN_Y, &inst->FullSrcRegisters[0]); + s1y_reg = get_src_reg(gen, CHAN_Y, &inst->FullSrcRegisters[1]); + s0z_reg = get_src_reg(gen, CHAN_Z, &inst->FullSrcRegisters[0]); + s1z_reg = get_src_reg(gen, CHAN_Z, &inst->FullSrcRegisters[1]); + s0w_reg = get_src_reg(gen, CHAN_W, &inst->FullSrcRegisters[0]); + s1w_reg = get_src_reg(gen, CHAN_W, &inst->FullSrcRegisters[1]); - s1_reg = get_src_reg(gen, CHAN_Y, &inst->FullSrcRegisters[0]); - s2_reg = get_src_reg(gen, CHAN_Y, &inst->FullSrcRegisters[1]); - /* t = y0 * y1 + t */ - spe_fma(gen->f, tmp_reg, s1_reg, s2_reg, tmp_reg); + /* t0 = x0 * x1 */ + spe_fm(gen->f, t0_reg, s0x_reg, s1x_reg); - s1_reg = get_src_reg(gen, CHAN_Z, &inst->FullSrcRegisters[0]); - s2_reg = get_src_reg(gen, CHAN_Z, &inst->FullSrcRegisters[1]); - /* t = z0 * z1 + t */ - spe_fma(gen->f, tmp_reg, s1_reg, s2_reg, tmp_reg); + /* t1 = y0 * y1 */ + spe_fm(gen->f, t1_reg, s0y_reg, s1y_reg); - s1_reg = get_src_reg(gen, CHAN_W, &inst->FullSrcRegisters[0]); - s2_reg = get_src_reg(gen, CHAN_W, &inst->FullSrcRegisters[1]); - /* t = w0 * w1 + t */ - spe_fma(gen->f, tmp_reg, s1_reg, s2_reg, tmp_reg); + /* t0 = z0 * z1 + t0 */ + spe_fma(gen->f, t0_reg, s0z_reg, s1z_reg, t0_reg); + + /* t1 = w0 * w1 + t1 */ + spe_fma(gen->f, t1_reg, s0w_reg, s1w_reg, t1_reg); + + /* t0 = t0 + t1 */ + spe_fa(gen->f, t0_reg, t0_reg, t1_reg); for (ch = 0; ch < 4; ch++) { if (inst->FullDstRegisters[0].DstRegister.WriteMask & (1 << ch)) { - store_dest_reg(gen, tmp_reg, ch, &inst->FullDstRegisters[0]); + int d_reg = get_dst_reg(gen, ch, &inst->FullDstRegisters[0]); + spe_move(gen->f, d_reg, t0_reg); + store_dest_reg(gen, d_reg, ch, &inst->FullDstRegisters[0]); } } @@ -650,6 +810,7 @@ emit_DP4(struct codegen *gen, const struct tgsi_full_instruction *inst) static boolean emit_DPH(struct codegen *gen, const struct tgsi_full_instruction *inst) { + /* XXX rewrite this function to look more like DP3/DP4 */ int ch; spe_comment(gen->f, -4, "DPH:"); @@ -676,6 +837,8 @@ emit_DPH(struct codegen *gen, const struct tgsi_full_instruction *inst) for (ch = 0; ch < 4; ch++) { if (inst->FullDstRegisters[0].DstRegister.WriteMask & (1 << ch)) { + int d_reg = get_dst_reg(gen, ch, &inst->FullDstRegisters[0]); + spe_move(gen->f, d_reg, tmp_reg); store_dest_reg(gen, tmp_reg, ch, &inst->FullDstRegisters[0]); } } @@ -1016,15 +1179,15 @@ emit_FLR(struct codegen *gen, const struct tgsi_full_instruction *inst) int tmp_reg = get_itemp(gen); /* If negative, subtract 1.0 */ - spe_fcgt(gen->f, d_reg, zero_reg, s1_reg); - spe_selb(gen->f, tmp_reg, zero_reg, get_const_one_reg(gen), d_reg); - spe_fs(gen->f, d_reg, s1_reg, tmp_reg); + spe_fcgt(gen->f, tmp_reg, zero_reg, s1_reg); + spe_selb(gen->f, tmp_reg, zero_reg, get_const_one_reg(gen), tmp_reg); + spe_fs(gen->f, tmp_reg, s1_reg, tmp_reg); /* Convert float to int */ - spe_cflts(gen->f, d_reg, d_reg, 0); + spe_cflts(gen->f, tmp_reg, tmp_reg, 0); /* Convert int to float */ - spe_csflt(gen->f, d_reg, d_reg, 0); + spe_csflt(gen->f, d_reg, tmp_reg, 0); store_dest_reg(gen, d_reg, ch, &inst->FullDstRegisters[0]); free_itemps(gen); @@ -1035,15 +1198,14 @@ emit_FLR(struct codegen *gen, const struct tgsi_full_instruction *inst) } /** - * Emit frac. - * Input - FLR(Input) + * Compute frac = Input - FLR(Input) */ static boolean emit_FRC(struct codegen *gen, const struct tgsi_full_instruction *inst) { int ch; - spe_comment(gen->f, -4, "FLR:"); + spe_comment(gen->f, -4, "FRC:"); int zero_reg = get_itemp(gen); spe_xor(gen->f, zero_reg, zero_reg, zero_reg); @@ -1055,18 +1217,18 @@ emit_FRC(struct codegen *gen, const struct tgsi_full_instruction *inst) int tmp_reg = get_itemp(gen); /* If negative, subtract 1.0 */ - spe_fcgt(gen->f, d_reg, zero_reg, s1_reg); - spe_selb(gen->f, tmp_reg, zero_reg, get_const_one_reg(gen), d_reg); - spe_fs(gen->f, d_reg, s1_reg, tmp_reg); + spe_fcgt(gen->f, tmp_reg, zero_reg, s1_reg); + spe_selb(gen->f, tmp_reg, zero_reg, get_const_one_reg(gen), tmp_reg); + spe_fs(gen->f, tmp_reg, s1_reg, tmp_reg); /* Convert float to int */ - spe_cflts(gen->f, d_reg, d_reg, 0); + spe_cflts(gen->f, tmp_reg, tmp_reg, 0); /* Convert int to float */ - spe_csflt(gen->f, d_reg, d_reg, 0); + spe_csflt(gen->f, tmp_reg, tmp_reg, 0); /* d = s1 - FLR(s1) */ - spe_fs(gen->f, d_reg, s1_reg, d_reg); + spe_fs(gen->f, d_reg, s1_reg, tmp_reg); store_dest_reg(gen, d_reg, ch, &inst->FullDstRegisters[0]); free_itemps(gen); @@ -1091,6 +1253,21 @@ print_functions(struct cell_context *cell) #endif +static uint +lookup_function(struct cell_context *cell, const char *funcname) +{ + const struct cell_spu_function_info *funcs = &cell->spu_functions; + uint i, addr = 0; + for (i = 0; i < funcs->num; i++) { + if (strcmp(funcs->names[i], funcname) == 0) { + addr = funcs->addrs[i]; + } + } + assert(addr && "spu function not found"); + return addr / 4; /* discard 2 least significant bits */ +} + + /** * Emit code to call a SPU function. * Used to implement instructions like SIN/COS/POW/TEX/etc. @@ -1100,77 +1277,56 @@ emit_function_call(struct codegen *gen, const struct tgsi_full_instruction *inst, char *funcname, uint num_args) { - const struct cell_spu_function_info *funcs = &gen->cell->spu_functions; + const uint addr = lookup_function(gen->cell, funcname); char comment[100]; - uint addr; int ch; - /* XXX temporary value */ - const int frameSize = 64; /* stack frame (activation record) size */ - assert(num_args <= 3); - /* lookup function address */ - { - uint i; - addr = 0; - for (i = 0; i < funcs->num; i++) { - if (strcmp(funcs->names[i], funcname) == 0) { - addr = funcs->addrs[i]; - } - } - assert(addr && "spu function not found"); - } - - addr /= 4; /* discard 2 least significant bits */ - snprintf(comment, sizeof(comment), "CALL %s:", funcname); spe_comment(gen->f, -4, comment); for (ch = 0; ch < 4; ch++) { if (inst->FullDstRegisters[0].DstRegister.WriteMask & (1 << ch)) { - int d_reg = get_dst_reg(gen, ch, &inst->FullDstRegisters[0]); - int s_regs[3]; - uint a; + int s_regs[3], d_reg; + ubyte usedRegs[SPE_NUM_REGS]; + uint a, i, numUsed; + for (a = 0; a < num_args; a++) { s_regs[a] = get_src_reg(gen, ch, &inst->FullSrcRegisters[a]); } + d_reg = get_dst_reg(gen, ch, &inst->FullDstRegisters[0]); - /* Basically: - * save registers on stack - * move parameters to registers 3, 4, 5... - * call function - * save return value (reg 3) - * restore registers from stack - */ - - /* XXX hack: load first function param */ - spe_move(gen->f, 3, s_regs[0]); + numUsed = spe_get_registers_used(gen->f, usedRegs); + assert(numUsed < gen->frame_size / 16 - 2); - /* save $lr on stack # stqd $lr,16($sp) */ - spe_stqd(gen->f, SPE_REG_RA, SPE_REG_SP, 16); - /* save stack pointer # stqd $sp,-frameSize($sp) */ - spe_stqd(gen->f, SPE_REG_SP, SPE_REG_SP, -frameSize); - - /* XXX save registers to stack here */ + /* save registers to stack */ + for (i = 0; i < numUsed; i++) { + uint reg = usedRegs[i]; + int offset = 2 + i; + spe_stqd(gen->f, reg, SPE_REG_SP, 16 * offset); + } - /* adjust stack pointer # ai $sp,$sp,-frameSize */ - spe_ai(gen->f, SPE_REG_SP, SPE_REG_SP, -frameSize); + /* setup function arguments */ + for (a = 0; a < num_args; a++) { + spe_move(gen->f, 3 + a, s_regs[a]); + } /* branch to function, save return addr */ spe_brasl(gen->f, SPE_REG_RA, addr); - /* restore stack pointer # ai $sp,$sp,frameSize */ - spe_ai(gen->f, SPE_REG_SP, SPE_REG_SP, frameSize); - - /* XXX restore registers from stack here */ - - /* restore $lr # lqd $lr,16($sp) */ - spe_lqd(gen->f, SPE_REG_RA, SPE_REG_SP, 16); - - /* XXX hack: save function's return value */ + /* save function's return value */ spe_move(gen->f, d_reg, 3); + /* restore registers from stack */ + for (i = 0; i < numUsed; i++) { + uint reg = usedRegs[i]; + if (reg != d_reg) { + int offset = 2 + i; + spe_lqd(gen->f, reg, SPE_REG_SP, 16 * offset); + } + } + store_dest_reg(gen, d_reg, ch, &inst->FullDstRegisters[0]); free_itemps(gen); } @@ -1180,31 +1336,114 @@ emit_function_call(struct codegen *gen, } +static boolean +emit_TXP(struct codegen *gen, const struct tgsi_full_instruction *inst) +{ + const uint addr = lookup_function(gen->cell, "spu_txp"); + const uint unit = inst->FullSrcRegisters[1].SrcRegister.Index; + int ch; + int coord_regs[4], d_regs[4]; + + assert(inst->FullSrcRegisters[1].SrcRegister.File == TGSI_FILE_SAMPLER); + + spe_comment(gen->f, -4, "CALL txp:"); + + /* get src/dst reg info */ + for (ch = 0; ch < 4; ch++) { + coord_regs[ch] = get_src_reg(gen, ch, &inst->FullSrcRegisters[0]); + d_regs[ch] = get_dst_reg(gen, ch, &inst->FullDstRegisters[0]); + } + + { + ubyte usedRegs[SPE_NUM_REGS]; + uint i, numUsed; + + numUsed = spe_get_registers_used(gen->f, usedRegs); + assert(numUsed < gen->frame_size / 16 - 2); + + /* save registers to stack */ + for (i = 0; i < numUsed; i++) { + uint reg = usedRegs[i]; + int offset = 2 + i; + spe_stqd(gen->f, reg, SPE_REG_SP, 16 * offset); + } + + /* setup function arguments */ + for (i = 0; i < 4; i++) { + spe_move(gen->f, 3 + i, coord_regs[i]); + } + spe_load_uint(gen->f, 7, unit); /* sampler unit */ + + /* branch to function, save return addr */ + spe_brasl(gen->f, SPE_REG_RA, addr); + + /* save function's return values (four pixel's colors) */ + for (i = 0; i < 4; i++) { + spe_move(gen->f, d_regs[i], 3 + i); + } + + /* restore registers from stack */ + for (i = 0; i < numUsed; i++) { + uint reg = usedRegs[i]; + if (reg != d_regs[0] && + reg != d_regs[1] && + reg != d_regs[2] && + reg != d_regs[3]) { + int offset = 2 + i; + spe_lqd(gen->f, reg, SPE_REG_SP, 16 * offset); + } + } + } + + for (ch = 0; ch < 4; ch++) { + if (inst->FullDstRegisters[0].DstRegister.WriteMask & (1 << ch)) { + store_dest_reg(gen, d_regs[ch], ch, &inst->FullDstRegisters[0]); + free_itemps(gen); + } + } + + return TRUE; +} + + /** * Emit max. See emit_SGT for comments. */ static boolean emit_MAX(struct codegen *gen, const struct tgsi_full_instruction *inst) { - int ch; + int ch, s0_reg[4], s1_reg[4], d_reg[4], tmp_reg[4]; spe_comment(gen->f, -4, "MAX:"); for (ch = 0; ch < 4; ch++) { if (inst->FullDstRegisters[0].DstRegister.WriteMask & (1 << ch)) { - int s1_reg = get_src_reg(gen, ch, &inst->FullSrcRegisters[0]); - int s2_reg = get_src_reg(gen, ch, &inst->FullSrcRegisters[1]); - int d_reg = get_dst_reg(gen, ch, &inst->FullDstRegisters[0]); + s0_reg[ch] = get_src_reg(gen, ch, &inst->FullSrcRegisters[0]); + s1_reg[ch] = get_src_reg(gen, ch, &inst->FullSrcRegisters[1]); + d_reg[ch] = get_dst_reg(gen, ch, &inst->FullDstRegisters[0]); + tmp_reg[ch] = get_itemp(gen); + } + } - /* d = (s1 > s2) ? s1 : s2 */ - spe_fcgt(gen->f, d_reg, s1_reg, s2_reg); - spe_selb(gen->f, d_reg, s2_reg, s1_reg, d_reg); + /* d = (s0 > s1) ? s0 : s1 */ + for (ch = 0; ch < 4; ch++) { + if (inst->FullDstRegisters[0].DstRegister.WriteMask & (1 << ch)) { + spe_fcgt(gen->f, tmp_reg[ch], s0_reg[ch], s1_reg[ch]); + } + } + for (ch = 0; ch < 4; ch++) { + if (inst->FullDstRegisters[0].DstRegister.WriteMask & (1 << ch)) { + spe_selb(gen->f, d_reg[ch], s1_reg[ch], s0_reg[ch], tmp_reg[ch]); + } + } - store_dest_reg(gen, d_reg, ch, &inst->FullDstRegisters[0]); - free_itemps(gen); + for (ch = 0; ch < 4; ch++) { + if (inst->FullDstRegisters[0].DstRegister.WriteMask & (1 << ch)) { + store_dest_reg(gen, d_reg[ch], ch, &inst->FullDstRegisters[0]); } } + free_itemps(gen); return true; } @@ -1214,25 +1453,38 @@ emit_MAX(struct codegen *gen, const struct tgsi_full_instruction *inst) static boolean emit_MIN(struct codegen *gen, const struct tgsi_full_instruction *inst) { - int ch; + int ch, s0_reg[4], s1_reg[4], d_reg[4], tmp_reg[4]; spe_comment(gen->f, -4, "MIN:"); for (ch = 0; ch < 4; ch++) { if (inst->FullDstRegisters[0].DstRegister.WriteMask & (1 << ch)) { - int s1_reg = get_src_reg(gen, ch, &inst->FullSrcRegisters[0]); - int s2_reg = get_src_reg(gen, ch, &inst->FullSrcRegisters[1]); - int d_reg = get_dst_reg(gen, ch, &inst->FullDstRegisters[0]); + s0_reg[ch] = get_src_reg(gen, ch, &inst->FullSrcRegisters[0]); + s1_reg[ch] = get_src_reg(gen, ch, &inst->FullSrcRegisters[1]); + d_reg[ch] = get_dst_reg(gen, ch, &inst->FullDstRegisters[0]); + tmp_reg[ch] = get_itemp(gen); + } + } - /* d = (s2 > s1) ? s1 : s2 */ - spe_fcgt(gen->f, d_reg, s2_reg, s1_reg); - spe_selb(gen->f, d_reg, s2_reg, s1_reg, d_reg); + /* d = (s1 > s0) ? s0 : s1 */ + for (ch = 0; ch < 4; ch++) { + if (inst->FullDstRegisters[0].DstRegister.WriteMask & (1 << ch)) { + spe_fcgt(gen->f, tmp_reg[ch], s1_reg[ch], s0_reg[ch]); + } + } + for (ch = 0; ch < 4; ch++) { + if (inst->FullDstRegisters[0].DstRegister.WriteMask & (1 << ch)) { + spe_selb(gen->f, d_reg[ch], s1_reg[ch], s0_reg[ch], tmp_reg[ch]); + } + } - store_dest_reg(gen, d_reg, ch, &inst->FullDstRegisters[0]); - free_itemps(gen); + for (ch = 0; ch < 4; ch++) { + if (inst->FullDstRegisters[0].DstRegister.WriteMask & (1 << ch)) { + store_dest_reg(gen, d_reg[ch], ch, &inst->FullDstRegisters[0]); } } + free_itemps(gen); return true; } @@ -1339,8 +1591,7 @@ static boolean emit_END(struct codegen *gen) { spe_comment(gen->f, -4, "END:"); - /* return from function call */ - spe_bi(gen->f, SPE_REG_RA, 0, 0); + emit_epilogue(gen); return true; } @@ -1413,6 +1664,18 @@ emit_instruction(struct codegen *gen, return emit_function_call(gen, inst, "spu_sin", 1); case TGSI_OPCODE_POW: return emit_function_call(gen, inst, "spu_pow", 2); + case TGSI_OPCODE_EXPBASE2: + return emit_function_call(gen, inst, "spu_exp2", 1); + case TGSI_OPCODE_LOGBASE2: + return emit_function_call(gen, inst, "spu_log2", 1); + case TGSI_OPCODE_TEX: + /* fall-through for now */ + case TGSI_OPCODE_TXD: + /* fall-through for now */ + case TGSI_OPCODE_TXB: + /* fall-through for now */ + case TGSI_OPCODE_TXP: + return emit_TXP(gen, inst); case TGSI_OPCODE_IF: return emit_IF(gen, inst); @@ -1456,16 +1719,23 @@ emit_immediate(struct codegen *gen, const struct tgsi_full_immediate *immed) for (ch = 0; ch < 4; ch++) { float val = immed->u.ImmediateFloat32[ch].Float; - int reg = spe_allocate_available_register(gen->f); - if (reg < 0) - return false; + if (ch > 0 && val == immed->u.ImmediateFloat32[ch - 1].Float) { + /* re-use previous register */ + gen->imm_regs[gen->num_imm][ch] = gen->imm_regs[gen->num_imm][ch - 1]; + } + else { + int reg = spe_allocate_available_register(gen->f); + + if (reg < 0) + return false; - /* update immediate map */ - gen->imm_regs[gen->num_imm][ch] = reg; + /* update immediate map */ + gen->imm_regs[gen->num_imm][ch] = reg; - /* emit initializer instruction */ - spe_load_float(gen->f, reg, val); + /* emit initializer instruction */ + spe_load_float(gen->f, reg, val); + } } gen->num_imm++; @@ -1488,12 +1758,6 @@ emit_declaration(struct cell_context *cell, switch (decl->Declaration.File) { case TGSI_FILE_TEMPORARY: - if (cell->debug_flags & CELL_DEBUG_ASM) { - printf("Declare temp reg %d .. %d\n", - decl->DeclarationRange.First, - decl->DeclarationRange.Last); - } - for (i = decl->DeclarationRange.First; i <= decl->DeclarationRange.Last; i++) { @@ -1508,12 +1772,12 @@ emit_declaration(struct cell_context *cell, * to SPU memory. someday... */ - if (cell->debug_flags & CELL_DEBUG_ASM) { - printf(" SPE regs: %d %d %d %d\n", - gen->temp_regs[i][0], - gen->temp_regs[i][1], - gen->temp_regs[i][2], - gen->temp_regs[i][3]); + { + char buf[100]; + sprintf(buf, "TGSI temp[%d] maps to SPU regs [$%d $%d $%d $%d]", i, + gen->temp_regs[i][0], gen->temp_regs[i][1], + gen->temp_regs[i][2], gen->temp_regs[i][3]); + spe_comment(gen->f, -4, buf); } } break; @@ -1525,6 +1789,7 @@ emit_declaration(struct cell_context *cell, } + /** * Translate TGSI shader code to SPE instructions. This is done when * the state tracker gives us a new shader (via pipe->create_fs_state()). @@ -1564,12 +1829,14 @@ cell_gen_fragment_program(struct cell_context *cell, tgsi_parse_init(&parse, tokens); + emit_prologue(&gen); + while (!tgsi_parse_end_of_tokens(&parse) && !gen.error) { tgsi_parse_token(&parse); switch (parse.FullToken.Token.Type) { case TGSI_TOKEN_TYPE_IMMEDIATE: - if (!emit_immediate(&gen, &parse.FullToken.FullImmediate)) + if (!emit_immediate(&gen, &parse.FullToken.FullImmediate)) gen.error = true; break; @@ -1588,7 +1855,6 @@ cell_gen_fragment_program(struct cell_context *cell, } } - if (gen.error) { /* terminate the SPE code */ return emit_END(&gen); diff --git a/src/gallium/drivers/cell/ppu/cell_gen_fragment.c b/src/gallium/drivers/cell/ppu/cell_gen_fragment.c index 653afc235df..4e1e53ecdc7 100644 --- a/src/gallium/drivers/cell/ppu/cell_gen_fragment.c +++ b/src/gallium/drivers/cell/ppu/cell_gen_fragment.c @@ -54,10 +54,12 @@ * \param ifragZ_reg register containing integer fragment Z values (in) * \param ifbZ_reg register containing integer frame buffer Z values (in/out) * \param zmask_reg register containing result of Z test/comparison (out) + * + * Returns true if the Z-buffer needs to be updated. */ -static void -gen_depth_test(const struct pipe_depth_stencil_alpha_state *dsa, - struct spe_function *f, +static boolean +gen_depth_test(struct spe_function *f, + const struct pipe_depth_stencil_alpha_state *dsa, int mask_reg, int ifragZ_reg, int ifbZ_reg, int zmask_reg) { /* NOTE: we use clgt below, not cgt, because we want to compare _unsigned_ @@ -132,7 +134,10 @@ gen_depth_test(const struct pipe_depth_stencil_alpha_state *dsa, * framebufferZ = (ztest_passed ? fragmentZ : framebufferZ; */ spe_selb(f, ifbZ_reg, ifbZ_reg, ifragZ_reg, mask_reg); + return true; } + + return false; } @@ -238,22 +243,35 @@ gen_alpha_test(const struct pipe_depth_stencil_alpha_state *dsa, * it and have to allocate and load it again unnecessarily. */ static inline void -setup_const_register(struct spe_function *f, boolean *is_already_set, unsigned int *r, float value) +setup_optional_register(struct spe_function *f, boolean *is_already_set, unsigned int *r) { if (*is_already_set) return; *r = spe_allocate_available_register(f); - spe_load_float(f, *r, value); *is_already_set = true; } static inline void -release_const_register(struct spe_function *f, boolean *is_already_set, unsigned int r) +release_optional_register(struct spe_function *f, boolean *is_already_set, unsigned int r) { if (!*is_already_set) return; spe_release_register(f, r); *is_already_set = false; } +static inline void +setup_const_register(struct spe_function *f, boolean *is_already_set, unsigned int *r, float value) +{ + if (*is_already_set) return; + setup_optional_register(f, is_already_set, r); + spe_load_float(f, *r, value); +} + +static inline void +release_const_register(struct spe_function *f, boolean *is_already_set, unsigned int r) +{ + release_optional_register(f, is_already_set, r); +} + /** * Generate SPE code to implement the given blend mode for a quad of pixels. * \param f SPE function to append instruction onto. @@ -1117,6 +1135,666 @@ gen_colormask(struct spe_function *f, spe_release_register(f, colormask_reg); } +/* This function is annoyingly similar to gen_depth_test(), above, except + * that instead of comparing two varying values (i.e. fragment and buffer), + * we're comparing a varying value with a static value. As such, we have + * access to the Compare Immediate instructions where we don't in + * gen_depth_test(), which is what makes us very different. + * + * The return value in the stencil_pass_reg is a bitmask of valid + * fragments that also passed the stencil test. The bitmask of valid + * fragments that failed would be found in (mask_reg & ~stencil_pass_reg). + */ +static void +gen_stencil_test(struct spe_function *f, const struct pipe_stencil_state *state, + unsigned int mask_reg, unsigned int fbS_reg, + unsigned int stencil_pass_reg) +{ + /* Generate code that puts the set of passing fragments into the stencil_pass_reg + * register, taking into account whether each fragment was active to begin with. + */ + switch (state->func) { + case PIPE_FUNC_EQUAL: + /* stencil_pass = mask & (s == reference) */ + spe_compare_equal_uint(f, stencil_pass_reg, fbS_reg, state->ref_value); + spe_and(f, stencil_pass_reg, mask_reg, stencil_pass_reg); + break; + + case PIPE_FUNC_NOTEQUAL: + /* stencil_pass = mask & ~(s == reference) */ + spe_compare_equal_uint(f, stencil_pass_reg, fbS_reg, state->ref_value); + spe_andc(f, stencil_pass_reg, mask_reg, stencil_pass_reg); + break; + + case PIPE_FUNC_GREATER: + /* stencil_pass = mask & (s > reference) */ + spe_compare_greater_uint(f, stencil_pass_reg, fbS_reg, state->ref_value); + spe_and(f, stencil_pass_reg, mask_reg, stencil_pass_reg); + break; + + case PIPE_FUNC_LESS: { + /* stencil_pass = mask & (reference > s) */ + /* There's no convenient Compare Less Than Immediate instruction, so + * we'll have to do this one the harder way, by loading a register and + * comparing directly. Compare Logical Greater Than Word (clgt) + * treats its operands as unsigned - no sign extension. + */ + unsigned int tmp_reg = spe_allocate_available_register(f); + spe_load_uint(f, tmp_reg, state->ref_value); + spe_clgt(f, stencil_pass_reg, tmp_reg, fbS_reg); + spe_and(f, stencil_pass_reg, mask_reg, stencil_pass_reg); + spe_release_register(f, tmp_reg); + break; + } + + case PIPE_FUNC_LEQUAL: + /* stencil_pass = mask & (s <= reference) = mask & ~(s > reference) */ + spe_compare_greater_uint(f, stencil_pass_reg, fbS_reg, state->ref_value); + spe_andc(f, stencil_pass_reg, mask_reg, stencil_pass_reg); + break; + + case PIPE_FUNC_GEQUAL: { + /* stencil_pass = mask & (s >= reference) = mask & ~(reference > s) */ + /* As above, we have to do this by loading a register */ + unsigned int tmp_reg = spe_allocate_available_register(f); + spe_load_uint(f, tmp_reg, state->ref_value); + spe_clgt(f, stencil_pass_reg, tmp_reg, fbS_reg); + spe_andc(f, stencil_pass_reg, mask_reg, stencil_pass_reg); + spe_release_register(f, tmp_reg); + break; + } + + case PIPE_FUNC_NEVER: + /* stencil_pass = mask & 0 = 0 */ + spe_load_uint(f, stencil_pass_reg, 0); + break; + + case PIPE_FUNC_ALWAYS: + /* stencil_pass = mask & 1 = mask */ + spe_move(f, stencil_pass_reg, mask_reg); + break; + } + + /* The fragments that passed the stencil test are now in stencil_pass_reg. + * The fragments that failed would be (mask_reg & ~stencil_pass_reg). + */ +} + +/* This function generates code that calculates a set of new stencil values + * given the earlier values and the operation to apply. It does not + * apply any tests. It is intended to be called up to 3 times + * (for the stencil fail operation, for the stencil pass-z fail operation, + * and for the stencil pass-z pass operation) to collect up to three + * possible sets of values, and for the caller to combine them based + * on the result of the tests. + * + * stencil_max_value should be (2^n - 1) where n is the number of bits + * in the stencil buffer - in other words, it should be usable as a mask. + */ +static void +gen_stencil_values(struct spe_function *f, unsigned int stencil_op, + unsigned int stencil_ref_value, unsigned int stencil_max_value, + unsigned int fbS_reg, unsigned int newS_reg) +{ + /* The code below assumes that newS_reg and fbS_reg are not the same + * register; if they can be, the calculations below will have to use + * an additional temporary register. For now, mark the assumption + * with an assertion that will fail if they are the same. + */ + ASSERT(fbS_reg != newS_reg); + + /* The code also assumes the the stencil_max_value is of the form + * 2^n-1 and can therefore be used as a mask for the valid bits in + * addition to a maximum. Make sure this is the case as well. + * The clever math below exploits the fact that incrementing a + * binary number serves to flip all the bits of a number starting at + * the LSB and continuing to (and including) the first zero bit + * found. That means that a number and its increment will always + * have at least one bit in common (the high order bit, if nothing + * else) *unless* the number is zero, *or* the number is of a form + * consisting of some number of 1s in the low-order bits followed + * by nothing but 0s in the high-order bits. The latter case + * implies it's of the form 2^n-1. + */ + ASSERT(stencil_max_value > 0 && ((stencil_max_value + 1) & stencil_max_value) == 0); + + switch(stencil_op) { + case PIPE_STENCIL_OP_KEEP: + /* newS = S */ + spe_move(f, newS_reg, fbS_reg); + break; + + case PIPE_STENCIL_OP_ZERO: + /* newS = 0 */ + spe_zero(f, newS_reg); + break; + + case PIPE_STENCIL_OP_REPLACE: + /* newS = stencil reference value */ + spe_load_uint(f, newS_reg, stencil_ref_value); + break; + + case PIPE_STENCIL_OP_INCR: { + /* newS = (s == max ? max : s + 1) */ + unsigned int equals_reg = spe_allocate_available_register(f); + + spe_compare_equal_uint(f, equals_reg, fbS_reg, stencil_max_value); + /* Add Word Immediate computes rT = rA + 10-bit signed immediate */ + spe_ai(f, newS_reg, fbS_reg, 1); + /* Select from the current value or the new value based on the equality test */ + spe_selb(f, newS_reg, fbS_reg, newS_reg, equals_reg); + + spe_release_register(f, equals_reg); + break; + } + case PIPE_STENCIL_OP_DECR: { + /* newS = (s == 0 ? 0 : s - 1) */ + unsigned int equals_reg = spe_allocate_available_register(f); + + spe_compare_equal_uint(f, equals_reg, fbS_reg, 0); + /* Add Word Immediate with a (-1) value works */ + spe_ai(f, newS_reg, fbS_reg, -1); + /* Select from the current value or the new value based on the equality test */ + spe_selb(f, newS_reg, fbS_reg, newS_reg, equals_reg); + + spe_release_register(f, equals_reg); + break; + } + case PIPE_STENCIL_OP_INCR_WRAP: + /* newS = (s == max ? 0 : s + 1), but since max is 2^n-1, we can + * do a normal add and mask off the correct bits + */ + spe_ai(f, newS_reg, fbS_reg, 1); + spe_and_uint(f, newS_reg, newS_reg, stencil_max_value); + break; + + case PIPE_STENCIL_OP_DECR_WRAP: + /* newS = (s == 0 ? max : s - 1), but we'll pull the same mask trick as above */ + spe_ai(f, newS_reg, fbS_reg, -1); + spe_and_uint(f, newS_reg, newS_reg, stencil_max_value); + break; + + case PIPE_STENCIL_OP_INVERT: + /* newS = ~s. We take advantage of the mask/max value to invert only + * the valid bits for the field so we don't have to do an extra "and". + */ + spe_xor_uint(f, newS_reg, fbS_reg, stencil_max_value); + break; + + default: + ASSERT(0); + } +} + + +/* This function generates code to get all the necessary possible + * stencil values. For each of the output registers (fail_reg, + * zfail_reg, and zpass_reg), it either allocates a new register + * and calculates a new set of values based on the stencil operation, + * or it reuses a register allocation and calculation done for an + * earlier (matching) operation, or it reuses the fbS_reg register + * (if the stencil operation is KEEP, which doesn't change the + * stencil buffer). + * + * Since this function allocates a variable number of registers, + * to avoid incurring complex logic to free them, they should + * be allocated after a spe_allocate_register_set() call + * and released by the corresponding spe_release_register_set() call. + */ +static void +gen_get_stencil_values(struct spe_function *f, const struct pipe_depth_stencil_alpha_state *dsa, + unsigned int fbS_reg, + unsigned int *fail_reg, unsigned int *zfail_reg, + unsigned int *zpass_reg, unsigned int *back_fail_reg, + unsigned int *back_zfail_reg, unsigned int *back_zpass_reg) +{ + unsigned zfail_op, back_zfail_op; + + /* Stenciling had better be enabled here */ + ASSERT(dsa->stencil[0].enabled); + + /* If the depth test is not enabled, it is treated as though it always + * passes. In particular, that means that the "zfail_op" (and the backfacing + * counterpart, if active) are not considered - a failing stencil test will + * trigger the "fail_op", and a passing stencil test will trigger the + * "zpass_op". + * + * By overriding the operations in this case to be PIPE_STENCIL_OP_KEEP, + * we keep them from being calculated. + */ + if (dsa->depth.enabled) { + zfail_op = dsa->stencil[0].zfail_op; + back_zfail_op = dsa->stencil[1].zfail_op; + } + else { + zfail_op = PIPE_STENCIL_OP_KEEP; + back_zfail_op = PIPE_STENCIL_OP_KEEP; + } + + /* One-sided or front-facing stencil */ + if (dsa->stencil[0].fail_op == PIPE_STENCIL_OP_KEEP) { + *fail_reg = fbS_reg; + } + else { + *fail_reg = spe_allocate_available_register(f); + gen_stencil_values(f, dsa->stencil[0].fail_op, dsa->stencil[0].ref_value, + 0xff, fbS_reg, *fail_reg); + } + + if (zfail_op == PIPE_STENCIL_OP_KEEP) { + *zfail_reg = fbS_reg; + } + else if (zfail_op == dsa->stencil[0].fail_op) { + *zfail_reg = *fail_reg; + } + else { + *zfail_reg = spe_allocate_available_register(f); + gen_stencil_values(f, dsa->stencil[0].zfail_op, dsa->stencil[0].ref_value, + 0xff, fbS_reg, *zfail_reg); + } + + if (dsa->stencil[0].zpass_op == PIPE_STENCIL_OP_KEEP) { + *zpass_reg = fbS_reg; + } + else if (dsa->stencil[0].zpass_op == dsa->stencil[0].fail_op) { + *zpass_reg = *fail_reg; + } + else if (dsa->stencil[0].zpass_op == zfail_op) { + *zpass_reg = *zfail_reg; + } + else { + *zpass_reg = spe_allocate_available_register(f); + gen_stencil_values(f, dsa->stencil[0].zpass_op, dsa->stencil[0].ref_value, + 0xff, fbS_reg, *zpass_reg); + } + + /* If two-sided stencil is enabled, we have more work to do. */ + if (!dsa->stencil[1].enabled) { + /* This just flags that the registers need not be deallocated later */ + *back_fail_reg = fbS_reg; + *back_zfail_reg = fbS_reg; + *back_zpass_reg = fbS_reg; + } + else { + /* Same calculations as above, but for the back stencil */ + if (dsa->stencil[1].fail_op == PIPE_STENCIL_OP_KEEP) { + *back_fail_reg = fbS_reg; + } + else if (dsa->stencil[1].fail_op == dsa->stencil[0].fail_op) { + *back_fail_reg = *fail_reg; + } + else if (dsa->stencil[1].fail_op == zfail_op) { + *back_fail_reg = *zfail_reg; + } + else if (dsa->stencil[1].fail_op == dsa->stencil[0].zpass_op) { + *back_fail_reg = *zpass_reg; + } + else { + *back_fail_reg = spe_allocate_available_register(f); + gen_stencil_values(f, dsa->stencil[1].fail_op, dsa->stencil[1].ref_value, + 0xff, fbS_reg, *back_fail_reg); + } + + if (back_zfail_op == PIPE_STENCIL_OP_KEEP) { + *back_zfail_reg = fbS_reg; + } + else if (back_zfail_op == dsa->stencil[0].fail_op) { + *back_zfail_reg = *fail_reg; + } + else if (back_zfail_op == zfail_op) { + *back_zfail_reg = *zfail_reg; + } + else if (back_zfail_op == dsa->stencil[0].zpass_op) { + *back_zfail_reg = *zpass_reg; + } + else if (back_zfail_op == dsa->stencil[1].fail_op) { + *back_zfail_reg = *back_fail_reg; + } + else { + *back_zfail_reg = spe_allocate_available_register(f); + gen_stencil_values(f, dsa->stencil[1].zfail_op, dsa->stencil[1].ref_value, + 0xff, fbS_reg, *back_zfail_reg); + } + + if (dsa->stencil[1].zpass_op == PIPE_STENCIL_OP_KEEP) { + *back_zpass_reg = fbS_reg; + } + else if (dsa->stencil[1].zpass_op == dsa->stencil[0].fail_op) { + *back_zpass_reg = *fail_reg; + } + else if (dsa->stencil[1].zpass_op == zfail_op) { + *back_zpass_reg = *zfail_reg; + } + else if (dsa->stencil[1].zpass_op == dsa->stencil[0].zpass_op) { + *back_zpass_reg = *zpass_reg; + } + else if (dsa->stencil[1].zpass_op == dsa->stencil[1].fail_op) { + *back_zpass_reg = *back_fail_reg; + } + else if (dsa->stencil[1].zpass_op == back_zfail_op) { + *back_zpass_reg = *back_zfail_reg; + } + else { + *back_zfail_reg = spe_allocate_available_register(f); + gen_stencil_values(f, dsa->stencil[1].zpass_op, dsa->stencil[1].ref_value, + 0xff, fbS_reg, *back_zpass_reg); + } + } /* End of calculations for back-facing stencil */ +} + +/* Note that fbZ_reg may *not* be set on entry, if in fact + * the depth test is not enabled. This function must not use + * the register if depth is not enabled. + */ +static boolean +gen_stencil_depth_test(struct spe_function *f, + const struct pipe_depth_stencil_alpha_state *dsa, + const int const facing_reg, + const int mask_reg, const int fragZ_reg, + const int fbZ_reg, const int fbS_reg) +{ + /* True if we've generated code that could require writeback to the + * depth and/or stencil buffers + */ + boolean modified_buffers = false; + + boolean need_to_calculate_stencil_values; + boolean need_to_writemask_stencil_values; + + /* Registers. We may or may not actually allocate these, depending + * on whether the state values indicate that we need them. + */ + unsigned int stencil_pass_reg, stencil_fail_reg; + unsigned int stencil_fail_values, stencil_pass_depth_fail_values, stencil_pass_depth_pass_values; + unsigned int stencil_writemask_reg; + unsigned int zmask_reg; + unsigned int newS_reg; + + /* Stenciling is quite complex: up to six different configurable stencil + * operations/calculations can be required (three each for front-facing + * and back-facing fragments). Many of those operations will likely + * be identical, so there's good reason to try to avoid calculating + * the same values more than once (which unfortunately makes the code less + * straightforward). + * + * To make register management easier, we start a new + * register set; we can release all the registers in the set at + * once, and avoid having to keep track of exactly which registers + * we allocate. We can still allocate and free registers as + * desired (if we know we no longer need a register), but we don't + * have to spend the complexity to track the more difficult variant + * register usage scenarios. + */ + spe_comment(f, 0, "Allocating stencil register set"); + spe_allocate_register_set(f); + + /* Calculate the writemask. If the writemask is trivial (either + * all 0s, meaning that we don't need to calculate any stencil values + * because they're not going to change the stencil anyway, or all 1s, + * meaning that we have to calculate the stencil values but do not + * need to mask them), we can avoid generating code. Don't forget + * that we need to consider backfacing stencil, if enabled. + */ + if (dsa->stencil[0].write_mask == 0x0 && (!dsa->stencil[1].enabled || dsa->stencil[1].write_mask == 0x00)) { + /* Trivial: don't need to calculate stencil values, and don't need to + * write them back to the framebuffer. + */ + need_to_calculate_stencil_values = false; + need_to_writemask_stencil_values = false; + } + else if (dsa->stencil[0].write_mask == 0xff && (!dsa->stencil[1].enabled || dsa->stencil[1].write_mask == 0xff)) { + /* Still trivial, but a little less so. We need to write the stencil + * values, but we don't need to mask them. + */ + need_to_calculate_stencil_values = true; + need_to_writemask_stencil_values = false; + } + else { + /* The general case: calculate, mask, and write */ + need_to_calculate_stencil_values = true; + need_to_writemask_stencil_values = true; + + /* While we're here, generate code that calculates what the + * writemask should be. If backface stenciling is enabled, + * and the backface writemask is not the same as the frontface + * writemask, we'll have to generate code that merges the + * two masks into a single effective mask based on fragment facing. + */ + spe_comment(f, 0, "Computing stencil writemask"); + stencil_writemask_reg = spe_allocate_available_register(f); + spe_load_uint(f, stencil_writemask_reg, dsa->stencil[0].write_mask); + if (dsa->stencil[1].enabled && dsa->stencil[0].write_mask != dsa->stencil[1].write_mask) { + unsigned int back_write_mask_reg = spe_allocate_available_register(f); + spe_comment(f, 0, "Resolving two-sided stencil writemask"); + spe_load_uint(f, back_write_mask_reg, dsa->stencil[1].write_mask); + spe_selb(f, stencil_writemask_reg, stencil_writemask_reg, back_write_mask_reg, facing_reg); + spe_release_register(f, back_write_mask_reg); + } + } + + /* At least one-sided stenciling must be on. Generate code that + * runs the stencil test on the basic/front-facing stencil, leaving + * the mask of passing stencil bits in stencil_pass_reg. This mask will + * be used both to mask the set of active pixels, and also to + * determine how the stencil buffer changes. + * + * This test will *not* change the value in mask_reg (because we don't + * yet know whether to apply the two-sided stencil or one-sided stencil). + */ + spe_comment(f, 0, "Running basic stencil test"); + stencil_pass_reg = spe_allocate_available_register(f); + gen_stencil_test(f, &dsa->stencil[0], mask_reg, fbS_reg, stencil_pass_reg); + + /* If two-sided stenciling is on, generate code to run the stencil + * test on the backfacing stencil as well, and combine the two results + * into the one correct result based on facing. + */ + if (dsa->stencil[1].enabled) { + unsigned int temp_reg = spe_allocate_available_register(f); + spe_comment(f, 0, "Running backface stencil test"); + gen_stencil_test(f, &dsa->stencil[1], mask_reg, fbS_reg, temp_reg); + spe_selb(f, stencil_pass_reg, stencil_pass_reg, temp_reg, facing_reg); + spe_release_register(f, temp_reg); + } + + /* Generate code that, given the mask of valid fragments and the + * mask of valid fragments that passed the stencil test, computes + * the mask of valid fragments that failed the stencil test. We + * have to do this before we run a depth test (because the + * depth test should not be performed on fragments that failed the + * stencil test, and because the depth test will update the + * mask of valid fragments based on the results of the depth test). + */ + spe_comment(f, 0, "Computing stencil fail mask and updating fragment mask"); + stencil_fail_reg = spe_allocate_available_register(f); + spe_andc(f, stencil_fail_reg, mask_reg, stencil_pass_reg); + /* Now remove the stenciled-out pixels from the valid fragment mask, + * so we can later use the valid fragment mask in the depth test. + */ + spe_and(f, mask_reg, mask_reg, stencil_pass_reg); + + /* We may not need to calculate stencil values, if the writemask is off */ + if (need_to_calculate_stencil_values) { + unsigned int back_stencil_fail_values, back_stencil_pass_depth_fail_values, back_stencil_pass_depth_pass_values; + unsigned int front_stencil_fail_values, front_stencil_pass_depth_fail_values, front_stencil_pass_depth_pass_values; + + /* Generate code that calculates exactly which stencil values we need, + * without calculating the same value twice (say, if two different + * stencil ops have the same value). This code will work for one-sided + * and two-sided stenciling (so that we take into account that operations + * may match between front and back stencils), and will also take into + * account whether the depth test is enabled (if the depth test is off, + * we don't need any of the zfail results, because the depth test always + * is considered to pass if it is disabled). Any register value that + * does not need to be calculated will come back with the same value + * that's in fbS_reg. + * + * This function will allocate a variant number of registers that + * will be released as part of the register set. + */ + spe_comment(f, 0, "Computing stencil values"); + gen_get_stencil_values(f, dsa, fbS_reg, + &front_stencil_fail_values, &front_stencil_pass_depth_fail_values, + &front_stencil_pass_depth_pass_values, &back_stencil_fail_values, + &back_stencil_pass_depth_fail_values, &back_stencil_pass_depth_pass_values); + + /* Tricky, tricky, tricky - the things we do to create optimal + * code... + * + * The various stencil values registers may overlap with each other + * and with fbS_reg arbitrarily (as any particular operation is + * only calculated once and stored in one register, no matter + * how many times it is used). So we can't change the values + * within those registers directly - if we change a value in a + * register that's being referenced by two different calculations, + * we've just unwittingly changed the second value as well... + * + * Avoid this by allocating new registers to hold the results + * (there may be 2, if the depth test is off, or 3, if it is on). + * These will be released as part of the register set. + */ + if (!dsa->stencil[1].enabled) { + /* The easy case: if two-sided stenciling is *not* enabled, we + * just use the front-sided values. + */ + stencil_fail_values = front_stencil_fail_values; + stencil_pass_depth_fail_values = front_stencil_pass_depth_fail_values; + stencil_pass_depth_pass_values = front_stencil_pass_depth_pass_values; + } + else { /* two-sided stencil enabled */ + spe_comment(f, 0, "Resolving backface stencil values"); + /* Allocate new registers for the needed merged values */ + stencil_fail_values = spe_allocate_available_register(f); + spe_selb(f, stencil_fail_values, front_stencil_fail_values, back_stencil_fail_values, facing_reg); + if (dsa->depth.enabled) { + stencil_pass_depth_fail_values = spe_allocate_available_register(f); + spe_selb(f, stencil_pass_depth_fail_values, front_stencil_pass_depth_fail_values, back_stencil_pass_depth_fail_values, facing_reg); + } + else { + stencil_pass_depth_fail_values = fbS_reg; + } + stencil_pass_depth_pass_values = spe_allocate_available_register(f); + spe_selb(f, stencil_pass_depth_pass_values, front_stencil_pass_depth_pass_values, back_stencil_pass_depth_pass_values, facing_reg); + } + } + + /* We now have all the stencil values we need. We also need + * the results of the depth test to figure out which + * stencil values will become the new stencil values. (Even if + * we aren't actually calculating stencil values, we need to apply + * the depth test if it's enabled.) + * + * The code generated by gen_depth_test() returns the results of the + * test in the given register, but also alters the mask_reg based + * on the results of the test. + */ + if (dsa->depth.enabled) { + spe_comment(f, 0, "Running stencil depth test"); + zmask_reg = spe_allocate_available_register(f); + modified_buffers |= gen_depth_test(f, dsa, mask_reg, fragZ_reg, fbZ_reg, zmask_reg); + } + + if (need_to_calculate_stencil_values) { + + /* If we need to writemask the stencil values before going into + * the stencil buffer, we'll have to use a new register to + * hold the new values. If not, we can just keep using the + * current register. + */ + if (need_to_writemask_stencil_values) { + newS_reg = spe_allocate_available_register(f); + spe_comment(f, 0, "Saving current stencil values for writemasking"); + spe_move(f, newS_reg, fbS_reg); + } + else { + newS_reg = fbS_reg; + } + + /* Merge in the selected stencil fail values */ + if (stencil_fail_values != fbS_reg) { + spe_comment(f, 0, "Loading stencil fail values"); + spe_selb(f, newS_reg, newS_reg, stencil_fail_values, stencil_fail_reg); + modified_buffers = true; + } + + /* Same for the stencil pass/depth fail values. If this calculation + * is not needed (say, if depth test is off), then the + * stencil_pass_depth_fail_values register will be equal to fbS_reg + * and we'll skip the calculation. + */ + if (stencil_pass_depth_fail_values != fbS_reg) { + /* We don't actually have a stencil pass/depth fail mask yet. + * Calculate it here from the stencil passing mask and the + * depth passing mask. Note that zmask_reg *must* have been + * set above if we're here. + */ + unsigned int stencil_pass_depth_fail_mask = spe_allocate_available_register(f); + spe_comment(f, 0, "Loading stencil pass/depth fail values"); + spe_andc(f, stencil_pass_depth_fail_mask, stencil_pass_reg, zmask_reg); + + spe_selb(f, newS_reg, newS_reg, stencil_pass_depth_fail_values, stencil_pass_depth_fail_mask); + + spe_release_register(f, stencil_pass_depth_fail_mask); + modified_buffers = true; + } + + /* Same for the stencil pass/depth pass mask. Note that we + * *can* get here with zmask_reg being unset (if the depth + * test is off but the stencil test is on). In this case, + * we assume the depth test passes, and don't need to mask + * the stencil pass mask with the Z mask. + */ + if (stencil_pass_depth_pass_values != fbS_reg) { + if (dsa->depth.enabled) { + unsigned int stencil_pass_depth_pass_mask = spe_allocate_available_register(f); + /* We'll need a separate register */ + spe_comment(f, 0, "Loading stencil pass/depth pass values"); + spe_and(f, stencil_pass_depth_pass_mask, stencil_pass_reg, zmask_reg); + spe_selb(f, newS_reg, newS_reg, stencil_pass_depth_pass_values, stencil_pass_depth_pass_mask); + spe_release_register(f, stencil_pass_depth_pass_mask); + } + else { + /* We can use the same stencil-pass register */ + spe_comment(f, 0, "Loading stencil pass values"); + spe_selb(f, newS_reg, newS_reg, stencil_pass_depth_pass_values, stencil_pass_reg); + } + modified_buffers = true; + } + + /* Almost done. If we need to writemask, do it now, leaving the + * results in the fbS_reg register passed in. If we don't need + * to writemask, then the results are *already* in the fbS_reg, + * so there's nothing more to do. + */ + + if (need_to_writemask_stencil_values && modified_buffers) { + /* The Select Bytes command makes a fine writemask. Where + * the mask is 0, the first (original) values are retained, + * effectively masking out changes. Where the mask is 1, the + * second (new) values are retained, incorporating changes. + */ + spe_comment(f, 0, "Writemasking new stencil values"); + spe_selb(f, fbS_reg, fbS_reg, newS_reg, stencil_writemask_reg); + } + + } /* done calculating stencil values */ + + /* The stencil and/or depth values have been applied, and the + * mask_reg, fbS_reg, and fbZ_reg values have been updated. + * We're all done, except that we've allocated a fair number + * of registers that we didn't bother tracking. Release all + * those registers as part of the register set, and go home. + */ + spe_comment(f, 0, "Releasing stencil register set"); + spe_release_register_set(f); + + /* Return true if we could have modified the stencil and/or + * depth buffers. + */ + return modified_buffers; +} + + /** * Generate SPE code to implement the fragment operations (alpha test, * depth test, stencil test, blending, colormask, and final @@ -1156,6 +1834,7 @@ cell_gen_fragment_function(struct cell_context *cell, struct spe_function *f) const int fragB_reg = 10; /* vector float */ const int fragA_reg = 11; /* vector float */ const int mask_reg = 12; /* vector uint */ + const int facing_reg = 13; /* uint */ /* offset of quad from start of tile * XXX assuming 4-byte pixels for color AND Z/stencil!!!! @@ -1183,6 +1862,7 @@ cell_gen_fragment_function(struct cell_context *cell, struct spe_function *f) spe_allocate_register(f, fragB_reg); spe_allocate_register(f, fragA_reg); spe_allocate_register(f, mask_reg); + spe_allocate_register(f, facing_reg); quad_offset_reg = spe_allocate_available_register(f); fbRGBA_reg = spe_allocate_available_register(f); @@ -1195,6 +1875,7 @@ cell_gen_fragment_function(struct cell_context *cell, struct spe_function *f) ASSERT(TILE_SIZE == 32); + spe_comment(f, 0, "Compute quad offset within tile"); spe_rotmi(f, y2_reg, y_reg, -1); /* y2 = y / 2 */ spe_rotmi(f, x2_reg, x_reg, -1); /* x2 = x / 2 */ spe_shli(f, y2_reg, y2_reg, 4); /* y2 *= 16 */ @@ -1205,130 +1886,188 @@ cell_gen_fragment_function(struct cell_context *cell, struct spe_function *f) spe_release_register(f, y2_reg); } - if (dsa->alpha.enabled) { gen_alpha_test(dsa, f, mask_reg, fragA_reg); } + /* If we need the stencil buffers (because one- or two-sided stencil is + * enabled) or the depth buffer (because the depth test is enabled), + * go grab them. Note that if either one- or two-sided stencil is + * enabled, dsa->stencil[0].enabled will be true. + */ if (dsa->depth.enabled || dsa->stencil[0].enabled) { const enum pipe_format zs_format = cell->framebuffer.zsbuf->format; boolean write_depth_stencil; - int fbZ_reg = spe_allocate_available_register(f); /* Z values */ - int fbS_reg = spe_allocate_available_register(f); /* Stencil values */ + /* We may or may not need to allocate a register for Z or stencil values */ + boolean fbS_reg_set = false, fbZ_reg_set = false; + unsigned int fbS_reg, fbZ_reg = 0; + + spe_comment(f, 0, "Fetching Z/stencil quad from tile"); /* fetch quad of depth/stencil values from tile at (x,y) */ /* Load: fbZS_reg = memory[depth_tile_reg + offset_reg] */ + /* XXX Not sure this is allowed if we've only got a 16-bit Z buffer... */ spe_lqx(f, fbZS_reg, depth_tile_reg, quad_offset_reg); - if (dsa->depth.enabled) { - /* Extract Z bits from fbZS_reg into fbZ_reg */ - if (zs_format == PIPE_FORMAT_S8Z24_UNORM || - zs_format == PIPE_FORMAT_X8Z24_UNORM) { - int mask_reg = spe_allocate_available_register(f); - spe_fsmbi(f, mask_reg, 0x7777); /* mask[0,1,2,3] = 0x00ffffff */ - spe_and(f, fbZ_reg, fbZS_reg, mask_reg); /* fbZ = fbZS & mask */ - spe_release_register(f, mask_reg); - /* OK, fbZ_reg has four 24-bit Z values now */ - } - else if (zs_format == PIPE_FORMAT_Z24S8_UNORM || - zs_format == PIPE_FORMAT_Z24X8_UNORM) { - spe_rotmi(f, fbZ_reg, fbZS_reg, -8); /* fbZ = fbZS >> 8 */ - /* OK, fbZ_reg has four 24-bit Z values now */ - } - else if (zs_format == PIPE_FORMAT_Z32_UNORM) { - spe_move(f, fbZ_reg, fbZS_reg); - /* OK, fbZ_reg has four 32-bit Z values now */ - } - else if (zs_format == PIPE_FORMAT_Z16_UNORM) { - spe_move(f, fbZ_reg, fbZS_reg); - /* OK, fbZ_reg has four 16-bit Z values now */ - } - else { - ASSERT(0); /* invalid format */ - } - - /* Convert fragZ values from float[4] to 16, 24 or 32-bit uint[4] */ - if (zs_format == PIPE_FORMAT_S8Z24_UNORM || - zs_format == PIPE_FORMAT_X8Z24_UNORM || - zs_format == PIPE_FORMAT_Z24S8_UNORM || - zs_format == PIPE_FORMAT_Z24X8_UNORM) { - /* scale/convert fragZ from float in [0,1] to uint in [0, ~0] */ - spe_cfltu(f, fragZ_reg, fragZ_reg, 32); - /* fragZ = fragZ >> 8 */ - spe_rotmi(f, fragZ_reg, fragZ_reg, -8); - } - else if (zs_format == PIPE_FORMAT_Z32_UNORM) { - /* scale/convert fragZ from float in [0,1] to uint in [0, ~0] */ - spe_cfltu(f, fragZ_reg, fragZ_reg, 32); - } - else if (zs_format == PIPE_FORMAT_Z16_UNORM) { - /* scale/convert fragZ from float in [0,1] to uint in [0, ~0] */ - spe_cfltu(f, fragZ_reg, fragZ_reg, 32); - /* fragZ = fragZ >> 16 */ - spe_rotmi(f, fragZ_reg, fragZ_reg, -16); - } - } - else { - /* no Z test, but set Z to zero so we don't OR-in garbage below */ - spe_load_uint(f, fbZ_reg, 0); /* XXX set to zero for now */ + /* From the Z/stencil buffer format, pull out the bits we need for + * Z and/or stencil. We'll also convert the incoming fragment Z + * value in fragZ_reg from a floating point value in [0.0..1.0] to + * an unsigned integer value with the appropriate resolution. + */ + switch(zs_format) { + + case PIPE_FORMAT_S8Z24_UNORM: /* fall through */ + case PIPE_FORMAT_X8Z24_UNORM: + if (dsa->depth.enabled) { + /* We need the Z part at least */ + setup_optional_register(f, &fbZ_reg_set, &fbZ_reg); + /* four 24-bit Z values in the low-order bits */ + spe_and_uint(f, fbZ_reg, fbZS_reg, 0x00ffffff); + + /* Incoming fragZ_reg value is a float in 0.0...1.0; convert + * to a 24-bit unsigned integer + */ + spe_cfltu(f, fragZ_reg, fragZ_reg, 32); + spe_rotmi(f, fragZ_reg, fragZ_reg, -8); + } + if (dsa->stencil[0].enabled) { + setup_optional_register(f, &fbS_reg_set, &fbS_reg); + /* four 8-bit Z values in the high-order bits */ + spe_rotmi(f, fbS_reg, fbZS_reg, -24); + } + break; + + case PIPE_FORMAT_Z24S8_UNORM: /* fall through */ + case PIPE_FORMAT_Z24X8_UNORM: + if (dsa->depth.enabled) { + setup_optional_register(f, &fbZ_reg_set, &fbZ_reg); + /* shift by 8 to get the upper 24-bit values */ + spe_rotmi(f, fbS_reg, fbZS_reg, -8); + + /* Incoming fragZ_reg value is a float in 0.0...1.0; convert + * to a 24-bit unsigned integer + */ + spe_cfltu(f, fragZ_reg, fragZ_reg, 32); + spe_rotmi(f, fragZ_reg, fragZ_reg, -8); + } + if (dsa->stencil[0].enabled) { + setup_optional_register(f, &fbS_reg_set, &fbS_reg); + /* 8-bit stencil in the low-order bits - mask them out */ + spe_and_uint(f, fbS_reg, fbZS_reg, 0x000000ff); + } + break; + + case PIPE_FORMAT_Z32_UNORM: + if (dsa->depth.enabled) { + setup_optional_register(f, &fbZ_reg_set, &fbZ_reg); + /* Copy over 4 32-bit values */ + spe_move(f, fbZ_reg, fbZS_reg); + + /* Incoming fragZ_reg value is a float in 0.0...1.0; convert + * to a 32-bit unsigned integer + */ + spe_cfltu(f, fragZ_reg, fragZ_reg, 32); + } + /* No stencil, so can't do anything there */ + break; + + case PIPE_FORMAT_Z16_UNORM: + if (dsa->depth.enabled) { + /* XXX Not sure this is correct, but it was here before, so we're + * going with it for now + */ + setup_optional_register(f, &fbZ_reg_set, &fbZ_reg); + /* Copy over 4 32-bit values */ + spe_move(f, fbZ_reg, fbZS_reg); + + /* Incoming fragZ_reg value is a float in 0.0...1.0; convert + * to a 16-bit unsigned integer + */ + spe_cfltu(f, fragZ_reg, fragZ_reg, 32); + spe_rotmi(f, fragZ_reg, fragZ_reg, -16); + } + /* No stencil */ + break; + + default: + ASSERT(0); /* invalid format */ } - + /* If stencil is enabled, use the stencil-specific code + * generator to generate both the stencil and depth (if needed) + * tests. Otherwise, if only depth is enabled, generate + * a quick depth test. The test generators themselves will + * report back whether the depth/stencil buffer has to be + * written back. + */ if (dsa->stencil[0].enabled) { - /* Extract Stencil bit sfrom fbZS_reg into fbS_reg */ - if (zs_format == PIPE_FORMAT_S8Z24_UNORM || - zs_format == PIPE_FORMAT_X8Z24_UNORM) { - /* XXX extract with a shift */ - ASSERT(0); - } - else if (zs_format == PIPE_FORMAT_Z24S8_UNORM || - zs_format == PIPE_FORMAT_Z24X8_UNORM) { - /* XXX extract with a mask */ - ASSERT(0); - } - } - else { - /* no stencil test, but set to zero so we don't OR-in garbage below */ - spe_load_uint(f, fbS_reg, 0); /* XXX set to zero for now */ - } + /* This will perform the stencil and depth tests, and update + * the mask_reg, fbZ_reg, and fbS_reg as required by the + * tests. + */ + ASSERT(fbS_reg_set); + spe_comment(f, 0, "Perform stencil test"); - if (dsa->stencil[0].enabled) { - /* XXX this may involve depth testing too */ - // gen_stencil_test(dsa, f, ... ); - ASSERT(0); + /* Note that fbZ_reg may not be set on entry, if stenciling + * is enabled but there's no Z-buffer. The + * gen_stencil_depth_test() function must ignore the + * fbZ_reg register if depth is not enabled. + */ + write_depth_stencil = gen_stencil_depth_test(f, dsa, facing_reg, mask_reg, fragZ_reg, fbZ_reg, fbS_reg); } else if (dsa->depth.enabled) { int zmask_reg = spe_allocate_available_register(f); - gen_depth_test(dsa, f, mask_reg, fragZ_reg, fbZ_reg, zmask_reg); + ASSERT(fbZ_reg_set); + spe_comment(f, 0, "Perform depth test"); + write_depth_stencil = gen_depth_test(f, dsa, mask_reg, fragZ_reg, fbZ_reg, zmask_reg); spe_release_register(f, zmask_reg); } - - /* do we need to write Z and/or Stencil back into framebuffer? */ - write_depth_stencil = (dsa->depth.writemask | - dsa->stencil[0].write_mask | - dsa->stencil[1].write_mask); + else { + write_depth_stencil = false; + } if (write_depth_stencil) { /* Merge latest Z and Stencil values into fbZS_reg. * fbZ_reg has four Z vals in bits [23..0] or bits [15..0]. * fbS_reg has four 8-bit Z values in bits [7..0]. */ + spe_comment(f, 0, "Store quad's depth/stencil values in tile"); if (zs_format == PIPE_FORMAT_S8Z24_UNORM || zs_format == PIPE_FORMAT_X8Z24_UNORM) { - spe_shli(f, fbS_reg, fbS_reg, 24); /* fbS = fbS << 24 */ - spe_or(f, fbZS_reg, fbS_reg, fbZ_reg); /* fbZS = fbS | fbZ */ + if (fbS_reg_set && fbZ_reg_set) { + spe_shli(f, fbS_reg, fbS_reg, 24); /* fbS = fbS << 24 */ + spe_or(f, fbZS_reg, fbS_reg, fbZ_reg); /* fbZS = fbS | fbZ */ + } + else if (fbS_reg_set) { + spe_shli(f, fbZS_reg, fbS_reg, 24); /* fbS = fbS << 24 */ + } + else { + spe_move(f, fbZS_reg, fbZ_reg); + } } else if (zs_format == PIPE_FORMAT_Z24S8_UNORM || zs_format == PIPE_FORMAT_Z24X8_UNORM) { - spe_shli(f, fbZ_reg, fbZ_reg, 8); /* fbZ = fbZ << 8 */ - spe_or(f, fbZS_reg, fbS_reg, fbZ_reg); /* fbZS = fbS | fbZ */ + if (fbS_reg_set && fbZ_reg_set) { + spe_shli(f, fbZ_reg, fbZ_reg, 8); /* fbZ = fbZ << 8 */ + spe_or(f, fbZS_reg, fbS_reg, fbZ_reg); /* fbZS = fbS | fbZ */ + } + else if (fbS_reg_set) { + spe_move(f, fbZS_reg, fbS_reg); + } + else { + spe_shli(f, fbZ_reg, fbZ_reg, 8); /* fbZ = fbZ << 8 */ + } } else if (zs_format == PIPE_FORMAT_Z32_UNORM) { - spe_move(f, fbZS_reg, fbZ_reg); /* fbZS = fbZ */ + if (fbZ_reg_set) { + spe_move(f, fbZS_reg, fbZ_reg); /* fbZS = fbZ */ + } } else if (zs_format == PIPE_FORMAT_Z16_UNORM) { - spe_move(f, fbZS_reg, fbZ_reg); /* fbZS = fbZ */ + if (fbZ_reg_set) { + spe_move(f, fbZS_reg, fbZ_reg); /* fbZS = fbZ */ + } } else if (zs_format == PIPE_FORMAT_S8_UNORM) { ASSERT(0); /* XXX to do */ @@ -1341,21 +2080,21 @@ cell_gen_fragment_function(struct cell_context *cell, struct spe_function *f) spe_stqx(f, fbZS_reg, depth_tile_reg, quad_offset_reg); } - spe_release_register(f, fbZ_reg); - spe_release_register(f, fbS_reg); + release_optional_register(f, &fbZ_reg_set, fbZ_reg); + release_optional_register(f, &fbS_reg_set, fbS_reg); } - /* Get framebuffer quad/colors. We'll need these for blending, * color masking, and to obey the quad/pixel mask. * Load: fbRGBA_reg = memory[color_tile + quad_offset] * Note: if mask={~0,~0,~0,~0} and we're not blending or colormasking * we could skip this load. */ + spe_comment(f, 0, "Fetch quad colors from tile"); spe_lqx(f, fbRGBA_reg, color_tile_reg, quad_offset_reg); - if (blend->blend_enable) { + spe_comment(f, 0, "Perform blending"); gen_blend(blend, blend_color, f, color_format, fragR_reg, fragG_reg, fragB_reg, fragA_reg, fbRGBA_reg); } @@ -1369,19 +2108,21 @@ cell_gen_fragment_function(struct cell_context *cell, struct spe_function *f) int rgba_reg = spe_allocate_available_register(f); /* Pack four float colors as four 32-bit int colors */ + spe_comment(f, 0, "Convert float quad colors to packed int framebuffer colors"); gen_pack_colors(f, color_format, fragR_reg, fragG_reg, fragB_reg, fragA_reg, rgba_reg); if (blend->logicop_enable) { + spe_comment(f, 0, "Compute logic op"); gen_logicop(blend, f, rgba_reg, fbRGBA_reg); } if (blend->colormask != PIPE_MASK_RGBA) { + spe_comment(f, 0, "Compute color mask"); gen_colormask(f, blend->colormask, color_format, rgba_reg, fbRGBA_reg); } - /* Mix fragment colors with framebuffer colors using the quad/pixel mask: * if (mask[i]) * rgba[i] = rgba[i]; @@ -1393,6 +2134,7 @@ cell_gen_fragment_function(struct cell_context *cell, struct spe_function *f) /* Store updated quad in tile: * memory[color_tile + quad_offset] = rgba_reg; */ + spe_comment(f, 0, "Store quad colors into color tile"); spe_stqx(f, rgba_reg, color_tile_reg, quad_offset_reg); spe_release_register(f, rgba_reg); diff --git a/src/gallium/drivers/cell/ppu/cell_pipe_state.c b/src/gallium/drivers/cell/ppu/cell_pipe_state.c index 8c55b8e0933..2e3086c4fae 100644 --- a/src/gallium/drivers/cell/ppu/cell_pipe_state.c +++ b/src/gallium/drivers/cell/ppu/cell_pipe_state.c @@ -258,8 +258,6 @@ cell_set_sampler_textures(struct pipe_context *pipe, } cell->num_textures = num; - cell_update_texture_mapping(cell); - cell->dirty |= CELL_NEW_TEXTURE; } diff --git a/src/gallium/drivers/cell/ppu/cell_render.c b/src/gallium/drivers/cell/ppu/cell_render.c index dd25ae880e5..79cb8df82fa 100644 --- a/src/gallium/drivers/cell/ppu/cell_render.c +++ b/src/gallium/drivers/cell/ppu/cell_render.c @@ -152,6 +152,7 @@ cell_flush_prim_buffer(struct cell_context *cell) struct cell_command_render *render = &cell_global.command[i].render; render->prim_type = PIPE_PRIM_TRIANGLES; render->num_verts = cell->prim_buffer.num_verts; + render->front_winding = cell->rasterizer->front_winding; render->vertex_size = cell->vertex_info->size * 4; render->xmin = cell->prim_buffer.xmin; render->ymin = cell->prim_buffer.ymin; diff --git a/src/gallium/drivers/cell/ppu/cell_screen.c b/src/gallium/drivers/cell/ppu/cell_screen.c index 47ba6fa2909..d2235579507 100644 --- a/src/gallium/drivers/cell/ppu/cell_screen.c +++ b/src/gallium/drivers/cell/ppu/cell_screen.c @@ -76,11 +76,11 @@ cell_get_param(struct pipe_screen *screen, int param) case PIPE_CAP_TEXTURE_SHADOW_MAP: return 10; case PIPE_CAP_MAX_TEXTURE_2D_LEVELS: - return 12; /* max 2Kx2K */ + return CELL_MAX_TEXTURE_LEVELS; case PIPE_CAP_MAX_TEXTURE_3D_LEVELS: return 8; /* max 128x128x128 */ case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS: - return 12; /* max 2Kx2K */ + return CELL_MAX_TEXTURE_LEVELS; default: return 10; } diff --git a/src/gallium/drivers/cell/ppu/cell_state.h b/src/gallium/drivers/cell/ppu/cell_state.h index a7771a55a31..b193170f9ce 100644 --- a/src/gallium/drivers/cell/ppu/cell_state.h +++ b/src/gallium/drivers/cell/ppu/cell_state.h @@ -44,8 +44,9 @@ #define CELL_NEW_TEXTURE 0x800 #define CELL_NEW_VERTEX 0x1000 #define CELL_NEW_VS 0x2000 -#define CELL_NEW_CONSTANTS 0x4000 -#define CELL_NEW_VERTEX_INFO 0x8000 +#define CELL_NEW_VS_CONSTANTS 0x4000 +#define CELL_NEW_FS_CONSTANTS 0x8000 +#define CELL_NEW_VERTEX_INFO 0x10000 extern void diff --git a/src/gallium/drivers/cell/ppu/cell_state_emit.c b/src/gallium/drivers/cell/ppu/cell_state_emit.c index f35893537bf..bb694aa1073 100644 --- a/src/gallium/drivers/cell/ppu/cell_state_emit.c +++ b/src/gallium/drivers/cell/ppu/cell_state_emit.c @@ -25,6 +25,7 @@ * **************************************************************************/ +#include "pipe/p_inlines.h" #include "util/u_memory.h" #include "cell_context.h" #include "cell_gen_fragment.h" @@ -36,6 +37,79 @@ #include "draw/draw_private.h" +/** + * Find/create a cell_command_fragment_ops object corresponding to the + * current blend/stencil/z/colormask/etc. state. + */ +static struct cell_command_fragment_ops * +lookup_fragment_ops(struct cell_context *cell) +{ + struct cell_fragment_ops_key key; + struct cell_command_fragment_ops *ops; + + /* + * Build key + */ + memset(&key, 0, sizeof(key)); + key.blend = *cell->blend; + key.dsa = *cell->depth_stencil; + + if (cell->framebuffer.cbufs[0]) + key.color_format = cell->framebuffer.cbufs[0]->format; + else + key.color_format = PIPE_FORMAT_NONE; + + if (cell->framebuffer.zsbuf) + key.zs_format = cell->framebuffer.zsbuf->format; + else + key.zs_format = PIPE_FORMAT_NONE; + + /* + * Look up key in cache. + */ + ops = (struct cell_command_fragment_ops *) + util_keymap_lookup(cell->fragment_ops_cache, &key); + + /* + * If not found, create/save new fragment ops command. + */ + if (!ops) { + struct spe_function spe_code; + + if (0) + debug_printf("**** Create New Fragment Ops\n"); + + /* Prepare the buffer that will hold the generated code. */ + spe_init_func(&spe_code, SPU_MAX_FRAGMENT_OPS_INSTS * SPE_INST_SIZE); + + /* generate new code */ + cell_gen_fragment_function(cell, &spe_code); + + /* alloc new fragment ops command */ + ops = CALLOC_STRUCT(cell_command_fragment_ops); + + /* populate the new cell_command_fragment_ops object */ + ops->opcode = CELL_CMD_STATE_FRAGMENT_OPS; + memcpy(ops->code, spe_code.store, spe_code_size(&spe_code)); + ops->dsa = *cell->depth_stencil; + ops->blend = *cell->blend; + + /* insert cell_command_fragment_ops object into keymap/cache */ + util_keymap_insert(cell->fragment_ops_cache, &key, ops, NULL); + + /* release rtasm buffer */ + spe_release_func(&spe_code); + } + else { + if (0) + debug_printf("**** Re-use Fragment Ops\n"); + } + + return ops; +} + + + static void emit_state_cmd(struct cell_context *cell, uint cmd, const void *state, uint state_size) @@ -89,31 +163,31 @@ cell_emit_state(struct cell_context *cell) } } + if (cell->dirty & (CELL_NEW_FS_CONSTANTS)) { + const uint shader = PIPE_SHADER_FRAGMENT; + const uint num_const = cell->constants[shader].size / sizeof(float); + uint i, j; + float *buf = cell_batch_alloc(cell, 16 + num_const * sizeof(float)); + uint64_t *ibuf = (uint64_t *) buf; + const float *constants = pipe_buffer_map(cell->pipe.screen, + cell->constants[shader].buffer, + PIPE_BUFFER_USAGE_CPU_READ); + ibuf[0] = CELL_CMD_STATE_FS_CONSTANTS; + ibuf[1] = num_const; + j = 4; + for (i = 0; i < num_const; i++) { + buf[j++] = constants[i]; + } + pipe_buffer_unmap(cell->pipe.screen, cell->constants[shader].buffer); + } + if (cell->dirty & (CELL_NEW_FRAMEBUFFER | CELL_NEW_DEPTH_STENCIL | CELL_NEW_BLEND)) { - /* XXX we don't want to always do codegen here. We should have - * a hash/lookup table to cache previous results... - */ - struct cell_command_fragment_ops *fops - = cell_batch_alloc(cell, sizeof(*fops)); - struct spe_function spe_code; - - /* Prepare the buffer that will hold the generated code. */ - spe_init_func(&spe_code, SPU_MAX_FRAGMENT_OPS_INSTS * SPE_INST_SIZE); - - /* generate new code */ - cell_gen_fragment_function(cell, &spe_code); - - /* put the new code into the batch buffer */ - fops->opcode = CELL_CMD_STATE_FRAGMENT_OPS; - memcpy(&fops->code, spe_code.store, - SPU_MAX_FRAGMENT_OPS_INSTS * SPE_INST_SIZE); - fops->dsa = *cell->depth_stencil; - fops->blend = *cell->blend; - - /* free codegen buffer */ - spe_release_func(&spe_code); + struct cell_command_fragment_ops *fops, *fops_cmd; + fops_cmd = cell_batch_alloc(cell, sizeof(*fops_cmd)); + fops = lookup_fragment_ops(cell); + memcpy(fops_cmd, fops, sizeof(*fops)); } if (cell->dirty & CELL_NEW_SAMPLER) { @@ -137,14 +211,24 @@ cell_emit_state(struct cell_context *cell) texture->opcode = CELL_CMD_STATE_TEXTURE; texture->unit = i; if (cell->texture[i]) { - texture->start = cell->texture[i]->tiled_data; - texture->width = cell->texture[i]->base.width[0]; - texture->height = cell->texture[i]->base.height[0]; + uint level; + for (level = 0; level < CELL_MAX_TEXTURE_LEVELS; level++) { + texture->start[level] = cell->texture[i]->tiled_data[level]; + texture->width[level] = cell->texture[i]->base.width[level]; + texture->height[level] = cell->texture[i]->base.height[level]; + texture->depth[level] = cell->texture[i]->base.depth[level]; + } + texture->target = cell->texture[i]->base.target; } else { - texture->start = NULL; - texture->width = 1; - texture->height = 1; + uint level; + for (level = 0; level < CELL_MAX_TEXTURE_LEVELS; level++) { + texture->start[level] = NULL; + texture->width[level] = 0; + texture->height[level] = 0; + texture->depth[level] = 0; + } + texture->target = 0; } } } diff --git a/src/gallium/drivers/cell/ppu/cell_state_shader.c b/src/gallium/drivers/cell/ppu/cell_state_shader.c index 3a0d066da2a..54a17eaf2b7 100644 --- a/src/gallium/drivers/cell/ppu/cell_state_shader.c +++ b/src/gallium/drivers/cell/ppu/cell_state_shader.c @@ -197,7 +197,10 @@ cell_set_constant_buffer(struct pipe_context *pipe, buf->buffer); cell->constants[shader].size = buf->size; - cell->dirty |= CELL_NEW_CONSTANTS; + if (shader == PIPE_SHADER_VERTEX) + cell->dirty |= CELL_NEW_VS_CONSTANTS; + else if (shader == PIPE_SHADER_FRAGMENT) + cell->dirty |= CELL_NEW_FS_CONSTANTS; } diff --git a/src/gallium/drivers/cell/ppu/cell_texture.c b/src/gallium/drivers/cell/ppu/cell_texture.c index b6590dfb86e..230e1925733 100644 --- a/src/gallium/drivers/cell/ppu/cell_texture.c +++ b/src/gallium/drivers/cell/ppu/cell_texture.c @@ -52,20 +52,22 @@ static unsigned minify( unsigned d ) static void -cell_texture_layout(struct cell_texture * spt) +cell_texture_layout(struct cell_texture *ct) { - struct pipe_texture *pt = &spt->base; + struct pipe_texture *pt = &ct->base; unsigned level; unsigned width = pt->width[0]; unsigned height = pt->height[0]; unsigned depth = pt->depth[0]; - spt->buffer_size = 0; + ct->buffer_size = 0; for ( level = 0 ; level <= pt->last_level ; level++ ) { unsigned size; unsigned w_tile, h_tile; + assert(level < CELL_MAX_TEXTURE_LEVELS); + /* width, height, rounded up to tile size */ w_tile = align(width, TILE_SIZE); h_tile = align(height, TILE_SIZE); @@ -76,9 +78,9 @@ cell_texture_layout(struct cell_texture * spt) pt->nblocksx[level] = pf_get_nblocksx(&pt->block, w_tile); pt->nblocksy[level] = pf_get_nblocksy(&pt->block, h_tile); - spt->stride[level] = pt->nblocksx[level] * pt->block.size; + ct->stride[level] = pt->nblocksx[level] * pt->block.size; - spt->level_offset[level] = spt->buffer_size; + ct->level_offset[level] = ct->buffer_size; size = pt->nblocksx[level] * pt->nblocksy[level] * pt->block.size; if (pt->target == PIPE_TEXTURE_CUBE) @@ -86,7 +88,7 @@ cell_texture_layout(struct cell_texture * spt) else size *= depth; - spt->buffer_size += size; + ct->buffer_size += size; width = minify(width); height = minify(height); @@ -100,26 +102,25 @@ cell_texture_create(struct pipe_screen *screen, const struct pipe_texture *templat) { struct pipe_winsys *ws = screen->winsys; - struct cell_texture *spt = CALLOC_STRUCT(cell_texture); - if (!spt) + struct cell_texture *ct = CALLOC_STRUCT(cell_texture); + if (!ct) return NULL; - spt->base = *templat; - spt->base.refcount = 1; - spt->base.screen = screen; + ct->base = *templat; + ct->base.refcount = 1; + ct->base.screen = screen; - cell_texture_layout(spt); + cell_texture_layout(ct); - spt->buffer = ws->buffer_create(ws, 32, - PIPE_BUFFER_USAGE_PIXEL, - spt->buffer_size); + ct->buffer = ws->buffer_create(ws, 32, PIPE_BUFFER_USAGE_PIXEL, + ct->buffer_size); - if (!spt->buffer) { - FREE(spt); + if (!ct->buffer) { + FREE(ct); return NULL; } - return &spt->base; + return &ct->base; } @@ -135,29 +136,116 @@ cell_texture_release(struct pipe_screen *screen, __FUNCTION__, (void *) *pt, (*pt)->refcount - 1); */ if (--(*pt)->refcount <= 0) { - struct cell_texture *spt = cell_texture(*pt); + struct cell_texture *ct = cell_texture(*pt); + uint i; /* - DBG("%s deleting %p\n", __FUNCTION__, (void *) spt); + DBG("%s deleting %p\n", __FUNCTION__, (void *) ct); */ - pipe_buffer_reference(screen, &spt->buffer, NULL); + pipe_buffer_reference(screen, &ct->buffer, NULL); + + for (i = 0; i < CELL_MAX_TEXTURE_LEVELS; i++) { + if (ct->tiled_data[i]) { + align_free(ct->tiled_data[i]); + } + } - FREE(spt); + FREE(ct); } *pt = NULL; } -#if 0 + +/** + * Convert image from linear layout to tiled layout. 4-byte pixels. + */ +static void +twiddle_image_uint(uint w, uint h, uint tile_size, uint *dst, + uint src_stride, const uint *src) +{ + const uint tile_size2 = tile_size * tile_size; + const uint h_t = (h + tile_size - 1) / tile_size; + const uint w_t = (w + tile_size - 1) / tile_size; + + uint it, jt; /* tile counters */ + uint i, j; /* intra-tile counters */ + + src_stride /= 4; /* convert from bytes to pixels */ + + /* loop over dest tiles */ + for (it = 0; it < h_t; it++) { + for (jt = 0; jt < w_t; jt++) { + /* start of dest tile: */ + uint *tdst = dst + (it * w_t + jt) * tile_size2; + + /* compute size of this tile (may be smaller than tile_size) */ + /* XXX note: a compiler bug was found here. That's why the code + * looks as it does. + */ + uint tile_width = w - jt * tile_size; + tile_width = MIN2(tile_width, tile_size); + uint tile_height = h - it * tile_size; + tile_height = MIN2(tile_height, tile_size); + + /* loop over texels in the tile */ + for (i = 0; i < tile_height; i++) { + for (j = 0; j < tile_width; j++) { + const uint srci = it * tile_size + i; + const uint srcj = jt * tile_size + j; + ASSERT(srci < h); + ASSERT(srcj < w); + tdst[i * tile_size + j] = src[srci * src_stride + srcj]; + } + } + } + } +} + + +/** + * Convert linear texture image data to tiled format for SPU usage. + */ static void -cell_texture_update(struct pipe_context *pipe, struct pipe_texture *texture, - uint face, uint levelsMask) +cell_twiddle_texture(struct pipe_screen *screen, + struct pipe_surface *surface) { - /* XXX TO DO: re-tile the texture data ... */ + struct cell_texture *ct = cell_texture(surface->texture); + const uint level = surface->level; + const uint texWidth = ct->base.width[level]; + const uint texHeight = ct->base.height[level]; + const uint bufWidth = align(texWidth, TILE_SIZE); + const uint bufHeight = align(texHeight, TILE_SIZE); + const void *map = pipe_buffer_map(screen, surface->buffer, + PIPE_BUFFER_USAGE_CPU_READ); + const uint *src = (const uint *) ((const ubyte *) map + surface->offset); + + switch (ct->base.format) { + case PIPE_FORMAT_A8R8G8B8_UNORM: + { + int numFaces = ct->base.target == PIPE_TEXTURE_CUBE ? 6 : 1; + int offset = bufWidth * bufHeight * 4 * surface->face; + uint *dst; + + if (!ct->tiled_data[level]) { + ct->tiled_data[level] = + align_malloc(bufWidth * bufHeight * 4 * numFaces, 16); + } + + dst = (uint *) ((ubyte *) ct->tiled_data[level] + offset); + twiddle_image_uint(texWidth, texHeight, TILE_SIZE, dst, + surface->stride, src); + } + break; + default: + printf("Cell: twiddle unsupported texture format\n"); + ; + } + + pipe_buffer_unmap(screen, surface->buffer); } -#endif static struct pipe_surface * @@ -167,22 +255,22 @@ cell_get_tex_surface(struct pipe_screen *screen, unsigned usage) { struct pipe_winsys *ws = screen->winsys; - struct cell_texture *spt = cell_texture(pt); + struct cell_texture *ct = cell_texture(pt); struct pipe_surface *ps; ps = ws->surface_alloc(ws); if (ps) { assert(ps->refcount); assert(ps->winsys); - winsys_buffer_reference(ws, &ps->buffer, spt->buffer); + winsys_buffer_reference(ws, &ps->buffer, ct->buffer); ps->format = pt->format; ps->block = pt->block; ps->width = pt->width[level]; ps->height = pt->height[level]; ps->nblocksx = pt->nblocksx[level]; ps->nblocksy = pt->nblocksy[level]; - ps->stride = spt->stride[level]; - ps->offset = spt->level_offset[level]; + ps->stride = ct->stride[level]; + ps->offset = ct->level_offset[level]; ps->usage = usage; /* XXX may need to override usage flags (see sp_texture.c) */ @@ -206,118 +294,12 @@ cell_get_tex_surface(struct pipe_screen *screen, } - -/** - * Copy tile data from linear layout to tiled layout. - * XXX this should be rolled into the future surface-creation code. - * XXX also need "untile" code... - */ -static void -tile_copy_data(uint w, uint h, uint tile_size, uint *dst, const uint *src) -{ - const uint tile_size2 = tile_size * tile_size; - const uint h_t = h / tile_size, w_t = w / tile_size; - - uint it, jt; /* tile counters */ - uint i, j; /* intra-tile counters */ - - /* loop over dest tiles */ - for (it = 0; it < h_t; it++) { - for (jt = 0; jt < w_t; jt++) { - /* start of dest tile: */ - uint *tdst = dst + (it * w_t + jt) * tile_size2; - /* loop over texels in the tile */ - for (i = 0; i < tile_size; i++) { - for (j = 0; j < tile_size; j++) { - const uint srci = it * tile_size + i; - const uint srcj = jt * tile_size + j; - *tdst++ = src[srci * w + srcj]; - } - } - } - } -} - - - -/** - * Convert linear texture image data to tiled format for SPU usage. - * XXX recast this in terms of pipe_surfaces (aka texture views). - */ -static void -cell_tile_texture(struct cell_context *cell, - struct cell_texture *texture) -{ - struct pipe_screen *screen = cell->pipe.screen; - uint face = 0, level = 0, zslice = 0; - struct pipe_surface *surf; - const uint w = texture->base.width[0], h = texture->base.height[0]; - const uint *src; - - /* temporary restrictions: */ - assert(w >= TILE_SIZE); - assert(h >= TILE_SIZE); - assert(w % TILE_SIZE == 0); - assert(h % TILE_SIZE == 0); - - surf = screen->get_tex_surface(screen, &texture->base, face, level, zslice, - PIPE_BUFFER_USAGE_CPU_WRITE); - ASSERT(surf); - - src = (const uint *) pipe_surface_map(surf, PIPE_BUFFER_USAGE_CPU_WRITE); - - if (texture->tiled_data) { - align_free(texture->tiled_data); - } - texture->tiled_data = align_malloc(w * h * 4, 16); - - tile_copy_data(w, h, TILE_SIZE, texture->tiled_data, src); - - pipe_surface_unmap(surf); - - pipe_surface_reference(&surf, NULL); -} - - -void -cell_update_texture_mapping(struct cell_context *cell) -{ -#if 0 - uint face = 0, level = 0, zslice = 0; -#endif - uint i; - - for (i = 0; i < CELL_MAX_SAMPLERS; i++) { - if (cell->texture[i]) - cell_tile_texture(cell, cell->texture[i]); - } - -#if 0 - if (cell->tex_surf && cell->tex_map) { - pipe_surface_unmap(cell->tex_surf); - cell->tex_map = NULL; - } - - /* XXX free old surface */ - - cell->tex_surf = cell_get_tex_surface(&cell->pipe, - &cell->texture[0]->base, - face, level, zslice); - - cell->tex_map = pipe_surface_map(cell->tex_surf); -#endif -} - - static void cell_tex_surface_release(struct pipe_screen *screen, struct pipe_surface **s) { - /* Effectively do the texture_update work here - if texture images - * needed post-processing to put them into hardware layout, this is - * where it would happen. For softpipe, nothing to do. - */ - assert ((*s)->texture); + /* XXX if done rendering to teximage, re-tile */ + pipe_texture_reference(&(*s)->texture, NULL); screen->winsys->surface_release(screen->winsys, s); @@ -325,9 +307,9 @@ cell_tex_surface_release(struct pipe_screen *screen, static void * -cell_surface_map( struct pipe_screen *screen, - struct pipe_surface *surface, - unsigned flags ) +cell_surface_map(struct pipe_screen *screen, + struct pipe_surface *surface, + unsigned flags) { ubyte *map; @@ -339,22 +321,8 @@ cell_surface_map( struct pipe_screen *screen, map = pipe_buffer_map( screen, surface->buffer, flags ); if (map == NULL) return NULL; - - /* May want to different things here depending on read/write nature - * of the map: - */ - if (surface->texture && - (flags & PIPE_BUFFER_USAGE_CPU_WRITE)) - { - /* Do something to notify sharing contexts of a texture change. - * In softpipe, that would mean flushing the texture cache. - */ -#if 0 - cell_screen(screen)->timestamp++; -#endif - } - - return map + surface->offset; + else + return (void *) (map + surface->offset); } @@ -362,17 +330,21 @@ static void cell_surface_unmap(struct pipe_screen *screen, struct pipe_surface *surface) { - pipe_buffer_unmap( screen, surface->buffer ); -} + struct cell_texture *ct = cell_texture(surface->texture); + assert(ct); -void -cell_init_texture_functions(struct cell_context *cell) -{ - /*cell->pipe.texture_update = cell_texture_update;*/ + if ((ct->base.tex_usage & PIPE_TEXTURE_USAGE_SAMPLER) && + (surface->usage & PIPE_BUFFER_USAGE_CPU_WRITE)) { + /* convert from linear to tiled layout */ + cell_twiddle_texture(screen, surface); + } + + pipe_buffer_unmap( screen, surface->buffer ); } + void cell_init_screen_texture_funcs(struct pipe_screen *screen) { diff --git a/src/gallium/drivers/cell/ppu/cell_texture.h b/src/gallium/drivers/cell/ppu/cell_texture.h index 6d37e95ebce..a0757091b07 100644 --- a/src/gallium/drivers/cell/ppu/cell_texture.h +++ b/src/gallium/drivers/cell/ppu/cell_texture.h @@ -40,15 +40,15 @@ struct cell_texture { struct pipe_texture base; - unsigned long level_offset[PIPE_MAX_TEXTURE_LEVELS]; - unsigned long stride[PIPE_MAX_TEXTURE_LEVELS]; + unsigned long level_offset[CELL_MAX_TEXTURE_LEVELS]; + unsigned long stride[CELL_MAX_TEXTURE_LEVELS]; /* The data is held here: */ struct pipe_buffer *buffer; unsigned long buffer_size; - void *tiled_data; /* XXX this may be temporary */ /*ALIGN16*/ + void *tiled_data[CELL_MAX_TEXTURE_LEVELS]; /* XXX this may be temporary */ /*ALIGN16*/ }; @@ -62,14 +62,6 @@ cell_texture(struct pipe_texture *pt) extern void -cell_update_texture_mapping(struct cell_context *cell); - - -extern void -cell_init_texture_functions(struct cell_context *cell); - - -extern void cell_init_screen_texture_funcs(struct pipe_screen *screen); diff --git a/src/gallium/drivers/cell/ppu/cell_vbuf.c b/src/gallium/drivers/cell/ppu/cell_vbuf.c index aa63435b934..578ddf62dcb 100644 --- a/src/gallium/drivers/cell/ppu/cell_vbuf.c +++ b/src/gallium/drivers/cell/ppu/cell_vbuf.c @@ -214,6 +214,7 @@ cell_vbuf_draw(struct vbuf_render *vbr, render->opcode = CELL_CMD_RENDER; render->prim_type = cvbr->prim; + render->front_winding = cell->rasterizer->front_winding; render->num_indexes = nr_indices; render->min_index = min_index; diff --git a/src/gallium/drivers/cell/ppu/cell_vertex_fetch.c b/src/gallium/drivers/cell/ppu/cell_vertex_fetch.c index 566df7f59e3..18969005b02 100644 --- a/src/gallium/drivers/cell/ppu/cell_vertex_fetch.c +++ b/src/gallium/drivers/cell/ppu/cell_vertex_fetch.c @@ -73,8 +73,8 @@ emit_matrix_transpose(struct spe_function *p, int col3; - spe_lqd(p, shuf_hi, shuf_ptr, 3); - spe_lqd(p, shuf_lo, shuf_ptr, 4); + spe_lqd(p, shuf_hi, shuf_ptr, 3*16); + spe_lqd(p, shuf_lo, shuf_ptr, 4*16); spe_shufb(p, t1, row0, row2, shuf_hi); spe_shufb(p, t2, row0, row2, shuf_lo); @@ -122,13 +122,13 @@ emit_matrix_transpose(struct spe_function *p, */ switch (count) { case 4: - spe_stqd(p, col3, dest_ptr, 3); + spe_stqd(p, col3, dest_ptr, 3 * 16); case 3: - spe_stqd(p, col2, dest_ptr, 2); + spe_stqd(p, col2, dest_ptr, 2 * 16); case 2: - spe_stqd(p, col1, dest_ptr, 1); + spe_stqd(p, col1, dest_ptr, 1 * 16); case 1: - spe_stqd(p, col0, dest_ptr, 0); + spe_stqd(p, col0, dest_ptr, 0 * 16); } @@ -166,17 +166,17 @@ emit_fetch(struct spe_function *p, float scale_signed = 0.0; float scale_unsigned = 0.0; - spe_lqd(p, v0, in_ptr, 0 + offset[0]); - spe_lqd(p, v1, in_ptr, 1 + offset[0]); - spe_lqd(p, v2, in_ptr, 2 + offset[0]); - spe_lqd(p, v3, in_ptr, 3 + offset[0]); + spe_lqd(p, v0, in_ptr, (0 + offset[0]) * 16); + spe_lqd(p, v1, in_ptr, (1 + offset[0]) * 16); + spe_lqd(p, v2, in_ptr, (2 + offset[0]) * 16); + spe_lqd(p, v3, in_ptr, (3 + offset[0]) * 16); offset[0] += 4; switch (bytes) { case 1: scale_signed = 1.0f / 127.0f; scale_unsigned = 1.0f / 255.0f; - spe_lqd(p, tmp, shuf_ptr, 1); + spe_lqd(p, tmp, shuf_ptr, 1 * 16); spe_shufb(p, v0, v0, v0, tmp); spe_shufb(p, v1, v1, v1, tmp); spe_shufb(p, v2, v2, v2, tmp); @@ -185,7 +185,7 @@ emit_fetch(struct spe_function *p, case 2: scale_signed = 1.0f / 32767.0f; scale_unsigned = 1.0f / 65535.0f; - spe_lqd(p, tmp, shuf_ptr, 2); + spe_lqd(p, tmp, shuf_ptr, 2 * 16); spe_shufb(p, v0, v0, v0, tmp); spe_shufb(p, v1, v1, v1, tmp); spe_shufb(p, v2, v2, v2, tmp); @@ -241,11 +241,11 @@ emit_fetch(struct spe_function *p, switch (count) { case 1: - spe_stqd(p, float_zero, out_ptr, 1); + spe_stqd(p, float_zero, out_ptr, 1 * 16); case 2: - spe_stqd(p, float_zero, out_ptr, 2); + spe_stqd(p, float_zero, out_ptr, 2 * 16); case 3: - spe_stqd(p, float_one, out_ptr, 3); + spe_stqd(p, float_one, out_ptr, 3 * 16); } if (float_zero != -1) { diff --git a/src/gallium/drivers/cell/spu/spu_colorpack.h b/src/gallium/drivers/cell/spu/spu_colorpack.h index fd8dc6ded3e..d7ce0055248 100644 --- a/src/gallium/drivers/cell/spu/spu_colorpack.h +++ b/src/gallium/drivers/cell/spu/spu_colorpack.h @@ -31,6 +31,7 @@ #define SPU_COLORPACK_H +#include <transpose_matrix4x4.h> #include <spu_intrinsics.h> @@ -84,10 +85,10 @@ spu_unpack_B8G8R8A8(uint color) vector unsigned int color_u4 = spu_splats(color); color_u4 = spu_shuffle(color_u4, color_u4, ((vector unsigned char) { - 10, 10, 10, 10, - 5, 5, 5, 5, + 2, 2, 2, 2, + 1, 1, 1, 1, 0, 0, 0, 0, - 15, 15, 15, 15}) ); + 3, 3, 3, 3}) ); return spu_convtf(color_u4, 32); } @@ -98,13 +99,47 @@ spu_unpack_A8R8G8B8(uint color) vector unsigned int color_u4 = spu_splats(color); color_u4 = spu_shuffle(color_u4, color_u4, ((vector unsigned char) { - 5, 5, 5, 5, - 10, 10, 10, 10, - 15, 15, 15, 15, + 1, 1, 1, 1, + 2, 2, 2, 2, + 3, 3, 3, 3, 0, 0, 0, 0}) ); - return spu_convtf(color_u4, 32); } +/** + * \param color_in - array of 32-bit packed ARGB colors + * \param color_out - returns float colors in RRRR, GGGG, BBBB, AAAA order + */ +static INLINE void +spu_unpack_A8R8G8B8_transpose4(const vector unsigned int color_in[4], + vector float color_out[4]) +{ + vector unsigned int c0; + + c0 = spu_shuffle(color_in[0], color_in[0], + ((vector unsigned char) { + 1, 1, 1, 1, 2, 2, 2, 2, 3, 3, 3, 3, 0, 0, 0, 0}) ); + color_out[0] = spu_convtf(c0, 32); + + c0 = spu_shuffle(color_in[1], color_in[1], + ((vector unsigned char) { + 1, 1, 1, 1, 2, 2, 2, 2, 3, 3, 3, 3, 0, 0, 0, 0}) ); + color_out[1] = spu_convtf(c0, 32); + + c0 = spu_shuffle(color_in[2], color_in[2], + ((vector unsigned char) { + 1, 1, 1, 1, 2, 2, 2, 2, 3, 3, 3, 3, 0, 0, 0, 0}) ); + color_out[2] = spu_convtf(c0, 32); + + c0 = spu_shuffle(color_in[3], color_in[3], + ((vector unsigned char) { + 1, 1, 1, 1, 2, 2, 2, 2, 3, 3, 3, 3, 0, 0, 0, 0}) ); + color_out[3] = spu_convtf(c0, 32); + + _transpose_matrix4x4(color_out, color_out); +} + + + #endif /* SPU_COLORPACK_H */ diff --git a/src/gallium/drivers/cell/spu/spu_command.c b/src/gallium/drivers/cell/spu/spu_command.c index ec9da5d8870..c28677ebf87 100644 --- a/src/gallium/drivers/cell/spu/spu_command.c +++ b/src/gallium/drivers/cell/spu/spu_command.c @@ -59,6 +59,14 @@ static unsigned char attribute_fetch_code_buffer[136 * PIPE_MAX_ATTRIBS] +static INLINE int +align(int value, int alignment) +{ + return (value + alignment - 1) & ~(alignment - 1); +} + + + /** * Tell the PPU that this SPU has finished copying a buffer to * local store and that it may be reused by the PPU. @@ -231,6 +239,25 @@ cmd_state_fragment_program(const struct cell_command_fragment_program *fp) } +static uint +cmd_state_fs_constants(const uint64_t *buffer, uint pos) +{ + const uint num_const = buffer[pos + 1]; + const float *constants = (const float *) &buffer[pos + 2]; + uint i; + + DEBUG_PRINTF("CMD_STATE_FS_CONSTANTS (%u)\n", num_const); + + /* Expand each float to float[4] for SOA execution */ + for (i = 0; i < num_const; i++) { + spu.constants[i] = spu_splats(constants[i]); + } + + /* return new buffer pos (in 8-byte words) */ + return pos + 2 + num_const / 2; +} + + static void cmd_state_framebuffer(const struct cell_command_framebuffer *cmd) { @@ -276,16 +303,96 @@ cmd_state_framebuffer(const struct cell_command_framebuffer *cmd) } +/** + * Tex texture mask_s/t and scale_s/t fields depend on the texture size and + * sampler wrap modes. + */ +static void +update_tex_masks(struct spu_texture *texture, + const struct pipe_sampler_state *sampler, + uint unit) +{ + uint i; + + for (i = 0; i < CELL_MAX_TEXTURE_LEVELS; i++) { + int width = texture->level[i].width; + int height = texture->level[i].height; + + if (sampler->wrap_s == PIPE_TEX_WRAP_REPEAT) + texture->level[i].mask_s = spu_splats(width - 1); + else + texture->level[i].mask_s = spu_splats(~0); + + if (sampler->wrap_t == PIPE_TEX_WRAP_REPEAT) + texture->level[i].mask_t = spu_splats(height - 1); + else + texture->level[i].mask_t = spu_splats(~0); + + if (sampler->normalized_coords) { + texture->level[i].scale_s = spu_splats((float) width); + texture->level[i].scale_t = spu_splats((float) height); + } + else { + texture->level[i].scale_s = spu_splats(1.0f); + texture->level[i].scale_t = spu_splats(1.0f); + } + } + + /* XXX temporary hack */ + if (texture->target == PIPE_TEXTURE_CUBE) { + spu.sample_texture4[unit] = sample_texture4_cube; + } +} + + static void cmd_state_sampler(const struct cell_command_sampler *sampler) { - DEBUG_PRINTF("SAMPLER [%u]\n", sampler->unit); + uint unit = sampler->unit; + + DEBUG_PRINTF("SAMPLER [%u]\n", unit); + + spu.sampler[unit] = sampler->state; + + switch (spu.sampler[unit].min_img_filter) { + case PIPE_TEX_FILTER_LINEAR: + spu.min_sample_texture4[unit] = sample_texture4_bilinear; + break; + case PIPE_TEX_FILTER_ANISO: + /* fall-through, for now */ + case PIPE_TEX_FILTER_NEAREST: + spu.min_sample_texture4[unit] = sample_texture4_nearest; + break; + default: + ASSERT(0); + } + + switch (spu.sampler[sampler->unit].mag_img_filter) { + case PIPE_TEX_FILTER_LINEAR: + spu.mag_sample_texture4[unit] = sample_texture4_bilinear; + break; + case PIPE_TEX_FILTER_ANISO: + /* fall-through, for now */ + case PIPE_TEX_FILTER_NEAREST: + spu.mag_sample_texture4[unit] = sample_texture4_nearest; + break; + default: + ASSERT(0); + } + + switch (spu.sampler[sampler->unit].min_mip_filter) { + case PIPE_TEX_MIPFILTER_NEAREST: + case PIPE_TEX_MIPFILTER_LINEAR: + spu.sample_texture4[unit] = sample_texture4_lod; + break; + case PIPE_TEX_MIPFILTER_NONE: + spu.sample_texture4[unit] = spu.mag_sample_texture4[unit]; + break; + default: + ASSERT(0); + } - spu.sampler[sampler->unit] = sampler->state; - if (spu.sampler[sampler->unit].min_img_filter == PIPE_TEX_FILTER_LINEAR) - spu.sample_texture[sampler->unit] = sample_texture_bilinear; - else - spu.sample_texture[sampler->unit] = sample_texture_nearest; + update_tex_masks(&spu.texture[unit], &spu.sampler[unit], unit); } @@ -293,24 +400,44 @@ static void cmd_state_texture(const struct cell_command_texture *texture) { const uint unit = texture->unit; - const uint width = texture->width; - const uint height = texture->height; + uint i; + + //if (spu.init.id==0) Debug=1; + + DEBUG_PRINTF("TEXTURE [%u]\n", texture->unit); + + spu.texture[unit].max_level = 0; + spu.texture[unit].target = texture->target; - DEBUG_PRINTF("TEXTURE [%u] at %p size %u x %u\n", - texture->unit, texture->start, - texture->width, texture->height); + for (i = 0; i < CELL_MAX_TEXTURE_LEVELS; i++) { + uint width = texture->width[i]; + uint height = texture->height[i]; + uint depth = texture->depth[i]; - spu.texture[unit].start = texture->start; - spu.texture[unit].width = width; - spu.texture[unit].height = height; + DEBUG_PRINTF(" LEVEL %u: at %p size[0] %u x %u\n", i, + texture->start[i], texture->width[i], texture->height[i]); - spu.texture[unit].tiles_per_row = width / TILE_SIZE; + spu.texture[unit].level[i].start = texture->start[i]; + spu.texture[unit].level[i].width = width; + spu.texture[unit].level[i].height = height; + spu.texture[unit].level[i].depth = depth; - spu.texture[unit].tex_size = (vector float) { width, height, 0.0, 0.0}; - spu.texture[unit].tex_size_mask = (vector unsigned int) - { width - 1, height - 1, 0, 0 }; - spu.texture[unit].tex_size_x_mask = spu_splats(width - 1); - spu.texture[unit].tex_size_y_mask = spu_splats(height - 1); + spu.texture[unit].level[i].tiles_per_row = + (width + TILE_SIZE - 1) / TILE_SIZE; + + spu.texture[unit].level[i].bytes_per_image = + 4 * align(width, TILE_SIZE) * align(height, TILE_SIZE) * depth; + + spu.texture[unit].level[i].max_s = spu_splats((int) width - 1); + spu.texture[unit].level[i].max_t = spu_splats((int) height - 1); + + if (texture->start[i]) + spu.texture[unit].max_level = i; + } + + update_tex_masks(&spu.texture[unit], &spu.sampler[unit], unit); + + //Debug=0; } @@ -456,6 +583,9 @@ cmd_batch(uint opcode) pos += sizeof(*fp) / 8; } break; + case CELL_CMD_STATE_FS_CONSTANTS: + pos = cmd_state_fs_constants(buffer, pos); + break; case CELL_CMD_STATE_SAMPLER: { struct cell_command_sampler *sampler diff --git a/src/gallium/drivers/cell/spu/spu_funcs.c b/src/gallium/drivers/cell/spu/spu_funcs.c index b57ad3f3b81..5c3ee305d48 100644 --- a/src/gallium/drivers/cell/spu/spu_funcs.c +++ b/src/gallium/drivers/cell/spu/spu_funcs.c @@ -35,53 +35,96 @@ #include <string.h> #include <libmisc.h> -#include <cos8_v.h> -#include <sin8_v.h> +#include <math.h> +#include <cos14_v.h> +#include <sin14_v.h> +#include <transpose_matrix4x4.h> #include "cell/common.h" #include "spu_main.h" #include "spu_funcs.h" -#define M_PI 3.1415926 +/** For "return"-ing four vectors */ +struct vec_4x4 +{ + vector float v[4]; +}; static vector float spu_cos(vector float x) { -#if 0 - static const float scale = 1.0 / (2.0 * M_PI); - x = x * spu_splats(scale); /* normalize */ - return _cos8_v(x); -#else - /* just pass-through to avoid trashing caller's stack */ - return x; -#endif + return _cos14_v(x); } static vector float spu_sin(vector float x) { -#if 0 - static const float scale = 1.0 / (2.0 * M_PI); - x = x * spu_splats(scale); /* normalize */ - return _sin8_v(x); /* 8-bit accuracy enough?? */ -#else - /* just pass-through to avoid trashing caller's stack */ - return x; -#endif + return _sin14_v(x); +} + +static vector float +spu_pow(vector float x, vector float y) +{ + float z0 = powf(spu_extract(x,0), spu_extract(y,0)); + float z1 = powf(spu_extract(x,1), spu_extract(y,1)); + float z2 = powf(spu_extract(x,2), spu_extract(y,2)); + float z3 = powf(spu_extract(x,3), spu_extract(y,3)); + return (vector float) {z0, z1, z2, z3}; } +static vector float +spu_exp2(vector float x) +{ + float z0 = powf(2.0f, spu_extract(x,0)); + float z1 = powf(2.0f, spu_extract(x,1)); + float z2 = powf(2.0f, spu_extract(x,2)); + float z3 = powf(2.0f, spu_extract(x,3)); + return (vector float) {z0, z1, z2, z3}; +} +static vector float +spu_log2(vector float x) +{ + /* + * log_base_2(x) = log(x) / log(2) + * 1.442695 = 1/log(2). + */ + static const vector float k = {1.442695F, 1.442695F, 1.442695F, 1.442695F}; + float z0 = logf(spu_extract(x,0)); + float z1 = logf(spu_extract(x,1)); + float z2 = logf(spu_extract(x,2)); + float z3 = logf(spu_extract(x,3)); + vector float v = (vector float) {z0, z1, z2, z3}; + return spu_mul(v, k); +} + + +static struct vec_4x4 +spu_txp(vector float s, vector float t, vector float r, vector float q, + unsigned unit) +{ + struct vec_4x4 colors; + spu.sample_texture4[unit](s, t, r, q, unit, 0, 0, colors.v); + return colors; +} + + +/** + * Add named function to list of "exported" functions that will be + * made available to the PPU-hosted code generator. + */ static void -add_func(struct cell_spu_function_info *spu_functions, - const char *name, void *addr) +export_func(struct cell_spu_function_info *spu_functions, + const char *name, void *addr) { uint n = spu_functions->num; ASSERT(strlen(name) < 16); strcpy(spu_functions->names[n], name); spu_functions->addrs[n] = (uint) addr; spu_functions->num++; + ASSERT(spu_functions->num <= 16); } @@ -99,8 +142,12 @@ return_function_info(void) ASSERT(sizeof(funcs) == 256); /* must be multiple of 16 bytes */ funcs.num = 0; - add_func(&funcs, "spu_cos", &spu_cos); - add_func(&funcs, "spu_sin", &spu_sin); + export_func(&funcs, "spu_cos", &spu_cos); + export_func(&funcs, "spu_sin", &spu_sin); + export_func(&funcs, "spu_pow", &spu_pow); + export_func(&funcs, "spu_exp2", &spu_exp2); + export_func(&funcs, "spu_log2", &spu_log2); + export_func(&funcs, "spu_txp", &spu_txp); /* Send the function info back to the PPU / main memory */ mfc_put((void *) &funcs, /* src in local store */ diff --git a/src/gallium/drivers/cell/spu/spu_main.h b/src/gallium/drivers/cell/spu/spu_main.h index 29a305232ec..eff43b870ca 100644 --- a/src/gallium/drivers/cell/spu/spu_main.h +++ b/src/gallium/drivers/cell/spu/spu_main.h @@ -41,6 +41,9 @@ #define MAX_HEIGHT 1024 +#define CELL_MAX_CONSTANTS 32 /**< number of float[4] constants */ + + /** * A tile is basically a TILE_SIZE x TILE_SIZE block of 4-byte pixels. * The data may be addressed through several different types. @@ -61,8 +64,13 @@ typedef union { /** Function for sampling textures */ -typedef vector float (*spu_sample_texture_func)(uint unit, - vector float texcoord); +typedef void (*spu_sample_texture4_func)(vector float s, + vector float t, + vector float r, + vector float q, + uint unit, uint level, uint face, + vector float colors[4]); + /** Function for performing per-fragment ops */ typedef void (*spu_fragment_ops_func)(uint x, uint y, @@ -73,7 +81,8 @@ typedef void (*spu_fragment_ops_func)(uint x, uint y, vector float fragGreen, vector float fragBlue, vector float fragAlpha, - vector unsigned int mask); + vector unsigned int mask, + uint facing); /** Function for running fragment program */ typedef void (*spu_fragment_program_func)(vector float *inputs, @@ -98,15 +107,27 @@ struct spu_framebuffer } ALIGN16_ATTRIB; -struct spu_texture +/** per-texture level info */ +struct spu_texture_level { void *start; - ushort width, height; + ushort width, height, depth; ushort tiles_per_row; - vector float tex_size; - vector unsigned int tex_size_mask; /**< == int(size - 1) */ - vector unsigned int tex_size_x_mask; /**< == int(size - 1) */ - vector unsigned int tex_size_y_mask; /**< == int(size - 1) */ + uint bytes_per_image; + /** texcoord scale factors */ + vector float scale_s, scale_t, scale_r; + /** texcoord masks (if REPEAT then size-1, else ~0) */ + vector signed int mask_s, mask_t, mask_r; + /** texcoord clamp limits */ + vector signed int max_s, max_t, max_r; +} ALIGN16_ATTRIB; + + +struct spu_texture +{ + struct spu_texture_level level[CELL_MAX_TEXTURE_LEVELS]; + uint max_level; + uint target; /**< PIPE_TEXTURE_x */ } ALIGN16_ATTRIB; @@ -154,11 +175,12 @@ struct spu_global spu_fragment_program_func fragment_program; /** Current texture sampler function */ - spu_sample_texture_func sample_texture[CELL_MAX_SAMPLERS]; + spu_sample_texture4_func sample_texture4[CELL_MAX_SAMPLERS]; + spu_sample_texture4_func min_sample_texture4[CELL_MAX_SAMPLERS]; + spu_sample_texture4_func mag_sample_texture4[CELL_MAX_SAMPLERS]; - /** Fragment program constants (XXX preliminary/used) */ -#define MAX_CONSTANTS 32 - vector float constants[MAX_CONSTANTS]; + /** Fragment program constants */ + vector float constants[4 * CELL_MAX_CONSTANTS]; } ALIGN16_ATTRIB; diff --git a/src/gallium/drivers/cell/spu/spu_per_fragment_op.c b/src/gallium/drivers/cell/spu/spu_per_fragment_op.c index f107764fb28..d252fa6dc18 100644 --- a/src/gallium/drivers/cell/spu/spu_per_fragment_op.c +++ b/src/gallium/drivers/cell/spu/spu_per_fragment_op.c @@ -57,7 +57,8 @@ spu_fallback_fragment_ops(uint x, uint y, vector float fragG, vector float fragB, vector float fragA, - vector unsigned int mask) + vector unsigned int mask, + uint facing) { vector float frag_aos[4]; unsigned int fbc0, fbc1, fbc2, fbc3 ; /* framebuffer/tile colors */ @@ -433,23 +434,23 @@ spu_fallback_fragment_ops(uint x, uint y, /* Form bitmask depending on color buffer format and colormask bits */ switch (spu.fb.color_format) { case PIPE_FORMAT_A8R8G8B8_UNORM: - if (spu.blend.colormask & (1<<0)) + if (spu.blend.colormask & PIPE_MASK_R) cmask |= 0x00ff0000; /* red */ - if (spu.blend.colormask & (1<<1)) + if (spu.blend.colormask & PIPE_MASK_G) cmask |= 0x0000ff00; /* green */ - if (spu.blend.colormask & (1<<2)) + if (spu.blend.colormask & PIPE_MASK_B) cmask |= 0x000000ff; /* blue */ - if (spu.blend.colormask & (1<<3)) + if (spu.blend.colormask & PIPE_MASK_A) cmask |= 0xff000000; /* alpha */ break; case PIPE_FORMAT_B8G8R8A8_UNORM: - if (spu.blend.colormask & (1<<0)) + if (spu.blend.colormask & PIPE_MASK_R) cmask |= 0x0000ff00; /* red */ - if (spu.blend.colormask & (1<<1)) + if (spu.blend.colormask & PIPE_MASK_G) cmask |= 0x00ff0000; /* green */ - if (spu.blend.colormask & (1<<2)) + if (spu.blend.colormask & PIPE_MASK_B) cmask |= 0xff000000; /* blue */ - if (spu.blend.colormask & (1<<3)) + if (spu.blend.colormask & PIPE_MASK_A) cmask |= 0x000000ff; /* alpha */ break; default: diff --git a/src/gallium/drivers/cell/spu/spu_per_fragment_op.h b/src/gallium/drivers/cell/spu/spu_per_fragment_op.h index f817abf0463..a61689c83ab 100644 --- a/src/gallium/drivers/cell/spu/spu_per_fragment_op.h +++ b/src/gallium/drivers/cell/spu/spu_per_fragment_op.h @@ -38,7 +38,8 @@ spu_fallback_fragment_ops(uint x, uint y, vector float fragGreen, vector float fragBlue, vector float fragAlpha, - vector unsigned int mask); + vector unsigned int mask, + uint facing); #endif /* SPU_PER_FRAGMENT_OP */ diff --git a/src/gallium/drivers/cell/spu/spu_render.c b/src/gallium/drivers/cell/spu/spu_render.c index 305dc988810..82dbeb26b76 100644 --- a/src/gallium/drivers/cell/spu/spu_render.c +++ b/src/gallium/drivers/cell/spu/spu_render.c @@ -279,7 +279,7 @@ cmd_render(const struct cell_command_render *render, uint *pos_incr) v1 = (const float *) (vertices + indexes[j+1] * vertex_size); v2 = (const float *) (vertices + indexes[j+2] * vertex_size); - drawn += tri_draw(v0, v1, v2, tx, ty); + drawn += tri_draw(v0, v1, v2, tx, ty, render->front_winding); } //printf("SPU %u: drew %u of %u\n", spu.init.id, drawn, render->num_indexes/3); @@ -297,5 +297,3 @@ cmd_render(const struct cell_command_render *render, uint *pos_incr) printf("SPU %u: RENDER done\n", spu.init.id); } - - diff --git a/src/gallium/drivers/cell/spu/spu_texture.c b/src/gallium/drivers/cell/spu/spu_texture.c index 117b8a36f80..42eb06a3625 100644 --- a/src/gallium/drivers/cell/spu/spu_texture.c +++ b/src/gallium/drivers/cell/spu/spu_texture.c @@ -26,6 +26,8 @@ **************************************************************************/ +#include <math.h> + #include "pipe/p_compiler.h" #include "spu_main.h" #include "spu_texture.h" @@ -40,37 +42,19 @@ void invalidate_tex_cache(void) { - uint unit = 0; - uint bytes = 4 * spu.texture[unit].width - * spu.texture[unit].height; - - spu_dcache_mark_dirty((unsigned) spu.texture[unit].start, bytes); -} + uint lvl; + for (lvl = 0; lvl < CELL_MAX_TEXTURE_LEVELS; lvl++) { + uint unit = 0; + uint bytes = 4 * spu.texture[unit].level[lvl].width + * spu.texture[unit].level[lvl].height; + if (spu.texture[unit].target == PIPE_TEXTURE_CUBE) + bytes *= 6; + else if (spu.texture[unit].target == PIPE_TEXTURE_3D) + bytes *= spu.texture[unit].level[lvl].depth; -/** - * XXX look into getting texels for all four pixels in a quad at once. - */ -static uint -get_texel(uint unit, vec_uint4 coordinate) -{ - /* - * XXX we could do the "/ TILE_SIZE" and "% TILE_SIZE" operations as - * SIMD since X and Y are already in a SIMD register. - */ - const unsigned texture_ea = (uintptr_t) spu.texture[unit].start; - ushort x = spu_extract(coordinate, 0); - ushort y = spu_extract(coordinate, 1); - unsigned tile_offset = sizeof(tile_t) - * ((y / TILE_SIZE * spu.texture[unit].tiles_per_row) + (x / TILE_SIZE)); - ushort texel_offset = (ushort) 4 - * (ushort) (((ushort) (y % TILE_SIZE) * (ushort) TILE_SIZE) + (x % TILE_SIZE)); - vec_uint4 tmp; - - spu_dcache_fetch_unaligned((qword *) & tmp, - texture_ea + tile_offset + texel_offset, - 4); - return spu_extract(tmp, 0); + spu_dcache_mark_dirty((unsigned) spu.texture[unit].level[lvl].start, bytes); + } } @@ -88,15 +72,17 @@ get_texel(uint unit, vec_uint4 coordinate) * a time. */ static void -get_four_texels(uint unit, vec_uint4 x, vec_uint4 y, vec_uint4 *texels) +get_four_texels(uint unit, uint level, uint face, vec_int4 x, vec_int4 y, + vec_uint4 *texels) { - const unsigned texture_ea = (uintptr_t) spu.texture[unit].start; - vec_uint4 tile_x = spu_rlmask(x, -5); - vec_uint4 tile_y = spu_rlmask(y, -5); - const qword offset_x = si_andi((qword) x, 0x1f); - const qword offset_y = si_andi((qword) y, 0x1f); + const struct spu_texture_level *tlevel = &spu.texture[unit].level[level]; + unsigned texture_ea = (uintptr_t) tlevel->start; + const vec_int4 tile_x = spu_rlmask(x, -5); /* tile_x = x / 32 */ + const vec_int4 tile_y = spu_rlmask(y, -5); /* tile_y = y / 32 */ + const qword offset_x = si_andi((qword) x, 0x1f); /* offset_x = x & 0x1f */ + const qword offset_y = si_andi((qword) y, 0x1f); /* offset_y = y & 0x1f */ - const qword tiles_per_row = (qword) spu_splats(spu.texture[unit].tiles_per_row); + const qword tiles_per_row = (qword) spu_splats(tlevel->tiles_per_row); const qword tile_size = (qword) spu_splats((unsigned) sizeof(tile_t)); qword tile_offset = si_mpya((qword) tile_y, tiles_per_row, (qword) tile_x); @@ -107,6 +93,8 @@ get_four_texels(uint unit, vec_uint4 x, vec_uint4 y, vec_uint4 *texels) vec_uint4 offset = (vec_uint4) si_a(tile_offset, texel_offset); + texture_ea = texture_ea + face * tlevel->bytes_per_image; + spu_dcache_fetch_unaligned((qword *) & texels[0], texture_ea + spu_extract(offset, 0), 4); spu_dcache_fetch_unaligned((qword *) & texels[1], @@ -118,83 +106,496 @@ get_four_texels(uint unit, vec_uint4 x, vec_uint4 y, vec_uint4 *texels) } +/** clamp vec to [0, max] */ +static INLINE vector signed int +spu_clamp(vector signed int vec, vector signed int max) +{ + static const vector signed int zero = {0,0,0,0}; + vector unsigned int c; + c = spu_cmpgt(vec, zero); /* c = vec > zero ? ~0 : 0 */ + vec = spu_sel(zero, vec, c); + c = spu_cmpgt(vec, max); /* c = vec > max ? ~0 : 0 */ + vec = spu_sel(vec, max, c); + return vec; +} + + + /** - * Get texture sample at texcoord. + * Do nearest texture sampling for four pixels. + * \param colors returned colors in SOA format (rrrr, gggg, bbbb, aaaa). */ -vector float -sample_texture_nearest(uint unit, vector float texcoord) +void +sample_texture4_nearest(vector float s, vector float t, + vector float r, vector float q, + uint unit, uint level, uint face, + vector float colors[4]) { - vector float tc = spu_mul(texcoord, spu.texture[unit].tex_size); - vector unsigned int itc = spu_convtu(tc, 0); /* convert to int */ - itc = spu_and(itc, spu.texture[unit].tex_size_mask); /* mask (GL_REPEAT) */ - uint texel = get_texel(unit, itc); - return spu_unpack_A8R8G8B8(texel); + const struct spu_texture_level *tlevel = &spu.texture[unit].level[level]; + vector float ss = spu_mul(s, tlevel->scale_s); + vector float tt = spu_mul(t, tlevel->scale_t); + vector signed int is = spu_convts(ss, 0); + vector signed int it = spu_convts(tt, 0); + vec_uint4 texels[4]; + + /* PIPE_TEX_WRAP_REPEAT */ + is = spu_and(is, tlevel->mask_s); + it = spu_and(it, tlevel->mask_t); + + /* PIPE_TEX_WRAP_CLAMP */ + is = spu_clamp(is, tlevel->max_s); + it = spu_clamp(it, tlevel->max_t); + + get_four_texels(unit, level, face, is, it, texels); + + /* convert four packed ARGBA pixels to float RRRR,GGGG,BBBB,AAAA */ + spu_unpack_A8R8G8B8_transpose4(texels, colors); } -vector float -sample_texture_bilinear(uint unit, vector float texcoord) +/** + * Do bilinear texture sampling for four pixels. + * \param colors returned colors in SOA format (rrrr, gggg, bbbb, aaaa). + */ +void +sample_texture4_bilinear(vector float s, vector float t, + vector float r, vector float q, + uint unit, uint level, uint face, + vector float colors[4]) { - static const vec_uint4 offset_x = {0, 0, 1, 1}; - static const vec_uint4 offset_y = {0, 1, 0, 1}; + const struct spu_texture_level *tlevel = &spu.texture[unit].level[level]; + static const vector float half = {-0.5f, -0.5f, -0.5f, -0.5f}; - vector float tc = spu_mul(texcoord, spu.texture[unit].tex_size); - tc = spu_add(tc, spu_splats(-0.5f)); /* half texel bias */ + vector float ss = spu_madd(s, tlevel->scale_s, half); + vector float tt = spu_madd(t, tlevel->scale_t, half); - /* integer texcoords S,T: */ - vec_uint4 itc = spu_convtu(tc, 0); /* convert to int */ + vector signed int is0 = spu_convts(ss, 0); + vector signed int it0 = spu_convts(tt, 0); - vec_uint4 texels[4]; - - /* setup texcoords for quad: - * +-----+-----+ - * |x0,y0|x1,y1| - * +-----+-----+ - * |x2,y2|x3,y3| - * +-----+-----+ - */ - vec_uint4 x = spu_splats(spu_extract(itc, 0)); - vec_uint4 y = spu_splats(spu_extract(itc, 1)); - x = spu_add(x, offset_x); - y = spu_add(y, offset_y); + /* is + 1, it + 1 */ + vector signed int is1 = spu_add(is0, 1); + vector signed int it1 = spu_add(it0, 1); + + /* PIPE_TEX_WRAP_REPEAT */ + is0 = spu_and(is0, tlevel->mask_s); + it0 = spu_and(it0, tlevel->mask_t); + is1 = spu_and(is1, tlevel->mask_s); + it1 = spu_and(it1, tlevel->mask_t); - /* GL_REPEAT wrap mode: */ - x = spu_and(x, spu.texture[unit].tex_size_x_mask); - y = spu_and(y, spu.texture[unit].tex_size_y_mask); + /* PIPE_TEX_WRAP_CLAMP */ + is0 = spu_clamp(is0, tlevel->max_s); + it0 = spu_clamp(it0, tlevel->max_t); + is1 = spu_clamp(is1, tlevel->max_s); + it1 = spu_clamp(it1, tlevel->max_t); - get_four_texels(unit, x, y, texels); + /* get packed int texels */ + vector unsigned int texels[16]; + get_four_texels(unit, level, face, is0, it0, texels + 0); /* upper-left */ + get_four_texels(unit, level, face, is1, it0, texels + 4); /* upper-right */ + get_four_texels(unit, level, face, is0, it1, texels + 8); /* lower-left */ + get_four_texels(unit, level, face, is1, it1, texels + 12); /* lower-right */ - /* integer A8R8G8B8 to float texel conversion */ - vector float texel00 = spu_unpack_A8R8G8B8(spu_extract(texels[0], 0)); - vector float texel01 = spu_unpack_A8R8G8B8(spu_extract(texels[1], 0)); - vector float texel10 = spu_unpack_A8R8G8B8(spu_extract(texels[2], 0)); - vector float texel11 = spu_unpack_A8R8G8B8(spu_extract(texels[3], 0)); + /* XXX possibly rework following code to compute the weighted sample + * colors with integer arithmetic for fewer int->float conversions. + */ + /* convert packed int texels to float colors */ + vector float ftexels[16]; + spu_unpack_A8R8G8B8_transpose4(texels + 0, ftexels + 0); + spu_unpack_A8R8G8B8_transpose4(texels + 4, ftexels + 4); + spu_unpack_A8R8G8B8_transpose4(texels + 8, ftexels + 8); + spu_unpack_A8R8G8B8_transpose4(texels + 12, ftexels + 12); /* Compute weighting factors in [0,1] * Multiply texcoord by 1024, AND with 1023, convert back to float. */ - vector float tc1024 = spu_mul(tc, spu_splats(1024.0f)); - vector signed int itc1024 = spu_convts(tc1024, 0); - itc1024 = spu_and(itc1024, spu_splats((1 << 10) - 1)); - vector float weight = spu_convtf(itc1024, 10); - - /* smeared frac and 1-frac */ - vector float sfrac = spu_splats(spu_extract(weight, 0)); - vector float tfrac = spu_splats(spu_extract(weight, 1)); - vector float sfrac1 = spu_sub(spu_splats(1.0f), sfrac); - vector float tfrac1 = spu_sub(spu_splats(1.0f), tfrac); - - /* multiply the samples (colors) by the S/T weights */ - texel00 = spu_mul(spu_mul(texel00, sfrac1), tfrac1); - texel10 = spu_mul(spu_mul(texel10, sfrac ), tfrac1); - texel01 = spu_mul(spu_mul(texel01, sfrac1), tfrac ); - texel11 = spu_mul(spu_mul(texel11, sfrac ), tfrac ); - - /* compute sum of weighted samples */ - vector float texel_sum = spu_add(texel00, texel01); - texel_sum = spu_add(texel_sum, texel10); - texel_sum = spu_add(texel_sum, texel11); - - return texel_sum; + vector float ss1024 = spu_mul(ss, spu_splats(1024.0f)); + vector signed int iss1024 = spu_convts(ss1024, 0); + iss1024 = spu_and(iss1024, 1023); + vector float sWeights0 = spu_convtf(iss1024, 10); + + vector float tt1024 = spu_mul(tt, spu_splats(1024.0f)); + vector signed int itt1024 = spu_convts(tt1024, 0); + itt1024 = spu_and(itt1024, 1023); + vector float tWeights0 = spu_convtf(itt1024, 10); + + /* 1 - sWeight and 1 - tWeight */ + vector float sWeights1 = spu_sub(spu_splats(1.0f), sWeights0); + vector float tWeights1 = spu_sub(spu_splats(1.0f), tWeights0); + + /* reds, for four pixels */ + ftexels[ 0] = spu_mul(ftexels[ 0], spu_mul(sWeights1, tWeights1)); /*ul*/ + ftexels[ 4] = spu_mul(ftexels[ 4], spu_mul(sWeights0, tWeights1)); /*ur*/ + ftexels[ 8] = spu_mul(ftexels[ 8], spu_mul(sWeights1, tWeights0)); /*ll*/ + ftexels[12] = spu_mul(ftexels[12], spu_mul(sWeights0, tWeights0)); /*lr*/ + colors[0] = spu_add(spu_add(ftexels[0], ftexels[4]), + spu_add(ftexels[8], ftexels[12])); + + /* greens, for four pixels */ + ftexels[ 1] = spu_mul(ftexels[ 1], spu_mul(sWeights1, tWeights1)); /*ul*/ + ftexels[ 5] = spu_mul(ftexels[ 5], spu_mul(sWeights0, tWeights1)); /*ur*/ + ftexels[ 9] = spu_mul(ftexels[ 9], spu_mul(sWeights1, tWeights0)); /*ll*/ + ftexels[13] = spu_mul(ftexels[13], spu_mul(sWeights0, tWeights0)); /*lr*/ + colors[1] = spu_add(spu_add(ftexels[1], ftexels[5]), + spu_add(ftexels[9], ftexels[13])); + + /* blues, for four pixels */ + ftexels[ 2] = spu_mul(ftexels[ 2], spu_mul(sWeights1, tWeights1)); /*ul*/ + ftexels[ 6] = spu_mul(ftexels[ 6], spu_mul(sWeights0, tWeights1)); /*ur*/ + ftexels[10] = spu_mul(ftexels[10], spu_mul(sWeights1, tWeights0)); /*ll*/ + ftexels[14] = spu_mul(ftexels[14], spu_mul(sWeights0, tWeights0)); /*lr*/ + colors[2] = spu_add(spu_add(ftexels[2], ftexels[6]), + spu_add(ftexels[10], ftexels[14])); + + /* alphas, for four pixels */ + ftexels[ 3] = spu_mul(ftexels[ 3], spu_mul(sWeights1, tWeights1)); /*ul*/ + ftexels[ 7] = spu_mul(ftexels[ 7], spu_mul(sWeights0, tWeights1)); /*ur*/ + ftexels[11] = spu_mul(ftexels[11], spu_mul(sWeights1, tWeights0)); /*ll*/ + ftexels[15] = spu_mul(ftexels[15], spu_mul(sWeights0, tWeights0)); /*lr*/ + colors[3] = spu_add(spu_add(ftexels[3], ftexels[7]), + spu_add(ftexels[11], ftexels[15])); +} + + + +/** + * Adapted from /opt/cell/sdk/usr/spu/include/transpose_matrix4x4.h + */ +static INLINE void +transpose(vector unsigned int *mOut0, + vector unsigned int *mOut1, + vector unsigned int *mOut2, + vector unsigned int *mOut3, + vector unsigned int *mIn) +{ + vector unsigned int abcd, efgh, ijkl, mnop; /* input vectors */ + vector unsigned int aeim, bfjn, cgko, dhlp; /* output vectors */ + vector unsigned int aibj, ckdl, emfn, gohp; /* intermediate vectors */ + + vector unsigned char shufflehi = ((vector unsigned char) { + 0x00, 0x01, 0x02, 0x03, + 0x10, 0x11, 0x12, 0x13, + 0x04, 0x05, 0x06, 0x07, + 0x14, 0x15, 0x16, 0x17}); + vector unsigned char shufflelo = ((vector unsigned char) { + 0x08, 0x09, 0x0A, 0x0B, + 0x18, 0x19, 0x1A, 0x1B, + 0x0C, 0x0D, 0x0E, 0x0F, + 0x1C, 0x1D, 0x1E, 0x1F}); + abcd = *(mIn+0); + efgh = *(mIn+1); + ijkl = *(mIn+2); + mnop = *(mIn+3); + + aibj = spu_shuffle(abcd, ijkl, shufflehi); + ckdl = spu_shuffle(abcd, ijkl, shufflelo); + emfn = spu_shuffle(efgh, mnop, shufflehi); + gohp = spu_shuffle(efgh, mnop, shufflelo); + + aeim = spu_shuffle(aibj, emfn, shufflehi); + bfjn = spu_shuffle(aibj, emfn, shufflelo); + cgko = spu_shuffle(ckdl, gohp, shufflehi); + dhlp = spu_shuffle(ckdl, gohp, shufflelo); + + *mOut0 = aeim; + *mOut1 = bfjn; + *mOut2 = cgko; + *mOut3 = dhlp; +} + + +/** + * Bilinear filtering, using int intead of float arithmetic + */ +void +sample_texture4_bilinear_2(vector float s, vector float t, + vector float r, vector float q, + uint unit, uint level, uint face, + vector float colors[4]) +{ + const struct spu_texture_level *tlevel = &spu.texture[unit].level[level]; + static const vector float half = {-0.5f, -0.5f, -0.5f, -0.5f}; + + /* Scale texcoords by size of texture, and add half pixel bias */ + vector float ss = spu_madd(s, tlevel->scale_s, half); + vector float tt = spu_madd(t, tlevel->scale_t, half); + + /* convert float coords to fixed-pt coords with 8 fraction bits */ + vector signed int is = spu_convts(ss, 8); + vector signed int it = spu_convts(tt, 8); + + /* compute integer texel weights in [0, 255] */ + vector signed int sWeights0 = spu_and(is, 255); + vector signed int tWeights0 = spu_and(it, 255); + vector signed int sWeights1 = spu_sub(255, sWeights0); + vector signed int tWeights1 = spu_sub(255, tWeights0); + + /* texel coords: is0 = is / 256, it0 = is / 256 */ + vector signed int is0 = spu_rlmask(is, -8); + vector signed int it0 = spu_rlmask(it, -8); + + /* texel coords: i1 = is0 + 1, it1 = it0 + 1 */ + vector signed int is1 = spu_add(is0, 1); + vector signed int it1 = spu_add(it0, 1); + + /* PIPE_TEX_WRAP_REPEAT */ + is0 = spu_and(is0, tlevel->mask_s); + it0 = spu_and(it0, tlevel->mask_t); + is1 = spu_and(is1, tlevel->mask_s); + it1 = spu_and(it1, tlevel->mask_t); + + /* PIPE_TEX_WRAP_CLAMP */ + is0 = spu_clamp(is0, tlevel->max_s); + it0 = spu_clamp(it0, tlevel->max_t); + is1 = spu_clamp(is1, tlevel->max_s); + it1 = spu_clamp(it1, tlevel->max_t); + + /* get packed int texels */ + vector unsigned int texels[16]; + get_four_texels(unit, level, face, is0, it0, texels + 0); /* upper-left */ + get_four_texels(unit, level, face, is1, it0, texels + 4); /* upper-right */ + get_four_texels(unit, level, face, is0, it1, texels + 8); /* lower-left */ + get_four_texels(unit, level, face, is1, it1, texels + 12); /* lower-right */ + + /* twiddle packed 32-bit BGRA pixels into RGBA as four unsigned ints */ + { + static const unsigned char ZERO = 0x80; + int i; + for (i = 0; i < 16; i++) { + texels[i] = spu_shuffle(texels[i], texels[i], + ((vector unsigned char) { + ZERO, ZERO, ZERO, 1, + ZERO, ZERO, ZERO, 2, + ZERO, ZERO, ZERO, 3, + ZERO, ZERO, ZERO, 0})); + } + } + + /* convert RGBA,RGBA,RGBA,RGBA to RRRR,GGGG,BBBB,AAAA */ + vector unsigned int texel0, texel1, texel2, texel3, texel4, texel5, texel6, texel7, + texel8, texel9, texel10, texel11, texel12, texel13, texel14, texel15; + transpose(&texel0, &texel1, &texel2, &texel3, texels + 0); + transpose(&texel4, &texel5, &texel6, &texel7, texels + 4); + transpose(&texel8, &texel9, &texel10, &texel11, texels + 8); + transpose(&texel12, &texel13, &texel14, &texel15, texels + 12); + + /* computed weighted colors */ + vector unsigned int c0, c1, c2, c3, cSum; + + /* red */ + c0 = (vector unsigned int) si_mpyu((qword) texel0, si_mpyu((qword) sWeights1, (qword) tWeights1)); /*ul*/ + c1 = (vector unsigned int) si_mpyu((qword) texel4, si_mpyu((qword) sWeights0, (qword) tWeights1)); /*ur*/ + c2 = (vector unsigned int) si_mpyu((qword) texel8, si_mpyu((qword) sWeights1, (qword) tWeights0)); /*ll*/ + c3 = (vector unsigned int) si_mpyu((qword) texel12, si_mpyu((qword) sWeights0, (qword) tWeights0)); /*lr*/ + cSum = spu_add(spu_add(c0, c1), spu_add(c2, c3)); + colors[0] = spu_convtf(cSum, 24); + + /* green */ + c0 = (vector unsigned int) si_mpyu((qword) texel1, si_mpyu((qword) sWeights1, (qword) tWeights1)); /*ul*/ + c1 = (vector unsigned int) si_mpyu((qword) texel5, si_mpyu((qword) sWeights0, (qword) tWeights1)); /*ur*/ + c2 = (vector unsigned int) si_mpyu((qword) texel9, si_mpyu((qword) sWeights1, (qword) tWeights0)); /*ll*/ + c3 = (vector unsigned int) si_mpyu((qword) texel13, si_mpyu((qword) sWeights0, (qword) tWeights0)); /*lr*/ + cSum = spu_add(spu_add(c0, c1), spu_add(c2, c3)); + colors[1] = spu_convtf(cSum, 24); + + /* blue */ + c0 = (vector unsigned int) si_mpyu((qword) texel2, si_mpyu((qword) sWeights1, (qword) tWeights1)); /*ul*/ + c1 = (vector unsigned int) si_mpyu((qword) texel6, si_mpyu((qword) sWeights0, (qword) tWeights1)); /*ur*/ + c2 = (vector unsigned int) si_mpyu((qword) texel10, si_mpyu((qword) sWeights1, (qword) tWeights0)); /*ll*/ + c3 = (vector unsigned int) si_mpyu((qword) texel14, si_mpyu((qword) sWeights0, (qword) tWeights0)); /*lr*/ + cSum = spu_add(spu_add(c0, c1), spu_add(c2, c3)); + colors[2] = spu_convtf(cSum, 24); + + /* alpha */ + c0 = (vector unsigned int) si_mpyu((qword) texel3, si_mpyu((qword) sWeights1, (qword) tWeights1)); /*ul*/ + c1 = (vector unsigned int) si_mpyu((qword) texel7, si_mpyu((qword) sWeights0, (qword) tWeights1)); /*ur*/ + c2 = (vector unsigned int) si_mpyu((qword) texel11, si_mpyu((qword) sWeights1, (qword) tWeights0)); /*ll*/ + c3 = (vector unsigned int) si_mpyu((qword) texel15, si_mpyu((qword) sWeights0, (qword) tWeights0)); /*lr*/ + cSum = spu_add(spu_add(c0, c1), spu_add(c2, c3)); + colors[3] = spu_convtf(cSum, 24); +} + + + +/** + * Compute level of detail factor from texcoords. + */ +static float +compute_lambda(uint unit, vector float s, vector float t) +{ + uint baseLevel = 0; + float width = spu.texture[unit].level[baseLevel].width; + float height = spu.texture[unit].level[baseLevel].width; + float dsdx = width * (spu_extract(s, 1) - spu_extract(s, 0)); + float dsdy = width * (spu_extract(s, 2) - spu_extract(s, 0)); + float dtdx = height * (spu_extract(t, 1) - spu_extract(t, 0)); + float dtdy = height * (spu_extract(t, 2) - spu_extract(t, 0)); + float x = dsdx * dsdx + dtdx * dtdx; + float y = dsdy * dsdy + dtdy * dtdy; + float rho = x > y ? x : y; + rho = sqrtf(rho); + float lambda = logf(rho) * 1.442695f; + return lambda; +} + + + +/** + * Texture sampling with level of detail selection. + */ +void +sample_texture4_lod(vector float s, vector float t, + vector float r, vector float q, + uint unit, uint level_ignored, uint face, + vector float colors[4]) +{ + /* + * Note that we're computing a lambda/lod here that's used for all + * four pixels in the quad. + */ + float lambda = compute_lambda(unit, s, t); + + /* apply lod bias */ + lambda += spu.sampler[unit].lod_bias; + + /* clamp */ + if (lambda < spu.sampler[unit].min_lod) + lambda = spu.sampler[unit].min_lod; + else if (lambda > spu.sampler[unit].max_lod) + lambda = spu.sampler[unit].max_lod; + + if (lambda <= 0.0f) { + /* magnify */ + spu.mag_sample_texture4[unit](s, t, r, q, unit, 0, 0, colors); + } + else { + /* minify */ + int level = (int) (lambda + 0.5f); + if (level > (int) spu.texture[unit].max_level) + level = spu.texture[unit].max_level; + spu.min_sample_texture4[unit](s, t, r, q, unit, level, 0, colors); + /* XXX to do: mipmap level interpolation */ + } +} + + +/** XXX need a SIMD version of this */ +static unsigned +choose_cube_face(float rx, float ry, float rz, float *newS, float *newT) +{ + /* + major axis + direction target sc tc ma + ---------- ------------------------------- --- --- --- + +rx TEXTURE_CUBE_MAP_POSITIVE_X_EXT -rz -ry rx + -rx TEXTURE_CUBE_MAP_NEGATIVE_X_EXT +rz -ry rx + +ry TEXTURE_CUBE_MAP_POSITIVE_Y_EXT +rx +rz ry + -ry TEXTURE_CUBE_MAP_NEGATIVE_Y_EXT +rx -rz ry + +rz TEXTURE_CUBE_MAP_POSITIVE_Z_EXT +rx -ry rz + -rz TEXTURE_CUBE_MAP_NEGATIVE_Z_EXT -rx -ry rz + */ + const float arx = fabsf(rx); + const float ary = fabsf(ry); + const float arz = fabsf(rz); + unsigned face; + float sc, tc, ma; + + if (arx > ary && arx > arz) { + if (rx >= 0.0F) { + face = PIPE_TEX_FACE_POS_X; + sc = -rz; + tc = -ry; + ma = arx; + } + else { + face = PIPE_TEX_FACE_NEG_X; + sc = rz; + tc = -ry; + ma = arx; + } + } + else if (ary > arx && ary > arz) { + if (ry >= 0.0F) { + face = PIPE_TEX_FACE_POS_Y; + sc = rx; + tc = rz; + ma = ary; + } + else { + face = PIPE_TEX_FACE_NEG_Y; + sc = rx; + tc = -rz; + ma = ary; + } + } + else { + if (rz > 0.0F) { + face = PIPE_TEX_FACE_POS_Z; + sc = rx; + tc = -ry; + ma = arz; + } + else { + face = PIPE_TEX_FACE_NEG_Z; + sc = -rx; + tc = -ry; + ma = arz; + } + } + + *newS = (sc / ma + 1.0F) * 0.5F; + *newT = (tc / ma + 1.0F) * 0.5F; + + return face; +} + + + +void +sample_texture4_cube(vector float s, vector float t, + vector float r, vector float q, + uint unit, uint level, uint face_ignored, + vector float colors[4]) +{ + static const vector float zero = {0.0f, 0.0f, 0.0f, 0.0f}; + uint p, faces[4]; + float newS[4], newT[4]; + + /* Compute cube face referenced by the four sets of texcoords. + * XXX we should SIMD-ize this. + */ + for (p = 0; p < 4; p++) { + float rx = spu_extract(s, p); + float ry = spu_extract(t, p); + float rz = spu_extract(r, p); + faces[p] = choose_cube_face(rx, ry, rz, &newS[p], &newT[p]); + } + + if (faces[0] == faces[1] && + faces[0] == faces[2] && + faces[0] == faces[3]) { + /* GOOD! All four texcoords refer to the same cube face */ + s = (vector float) {newS[0], newS[1], newS[2], newS[3]}; + t = (vector float) {newT[0], newT[1], newT[2], newT[3]}; + sample_texture4_nearest(s, t, zero, zero, unit, level, faces[0], colors); + } + else { + /* BAD! The four texcoords refer to different faces */ + for (p = 0; p < 4; p++) { + vector float c[4]; + + sample_texture4_nearest(spu_splats(newS[p]), spu_splats(newT[p]), + zero, zero, unit, level, faces[p], c); + + float red = spu_extract(c[0], p); + float green = spu_extract(c[1], p); + float blue = spu_extract(c[2], p); + float alpha = spu_extract(c[3], p); + + colors[0] = spu_insert(red, colors[0], p); + colors[1] = spu_insert(green, colors[1], p); + colors[2] = spu_insert(blue, colors[2], p); + colors[3] = spu_insert(alpha, colors[3], p); + } + } } diff --git a/src/gallium/drivers/cell/spu/spu_texture.h b/src/gallium/drivers/cell/spu/spu_texture.h index f7c9738be88..387484c3ad3 100644 --- a/src/gallium/drivers/cell/spu/spu_texture.h +++ b/src/gallium/drivers/cell/spu/spu_texture.h @@ -36,12 +36,38 @@ extern void invalidate_tex_cache(void); -extern vector float -sample_texture_nearest(uint unit, vector float texcoord); +extern void +sample_texture4_nearest(vector float s, vector float t, + vector float r, vector float q, + uint unit, uint level, uint face, + vector float colors[4]); + + +extern void +sample_texture4_bilinear(vector float s, vector float t, + vector float r, vector float q, + uint unit, uint level, uint face, + vector float colors[4]); + +extern void +sample_texture4_bilinear_2(vector float s, vector float t, + vector float r, vector float q, + uint unit, uint level, uint face, + vector float colors[4]); + +extern void +sample_texture4_lod(vector float s, vector float t, + vector float r, vector float q, + uint unit, uint level, uint face, + vector float colors[4]); -extern vector float -sample_texture_bilinear(uint unit, vector float texcoord); + +extern void +sample_texture4_cube(vector float s, vector float t, + vector float r, vector float q, + uint unit, uint level_ignored, uint face_ignored, + vector float colors[4]); #endif /* SPU_TEXTURE_H */ diff --git a/src/gallium/drivers/cell/spu/spu_tri.c b/src/gallium/drivers/cell/spu/spu_tri.c index 0a8fb56a62c..03f094373df 100644 --- a/src/gallium/drivers/cell/spu/spu_tri.c +++ b/src/gallium/drivers/cell/spu/spu_tri.c @@ -116,21 +116,15 @@ struct setup_stage { struct edge etop; struct edge emaj; - float oneoverarea; + float oneOverArea; - uint tx, ty; + uint facing; + + uint tx, ty; /**< position of current tile (x, y) */ int cliprect_minx, cliprect_maxx, cliprect_miny, cliprect_maxy; -#if 0 - struct tgsi_interp_coef coef[PIPE_MAX_SHADER_INPUTS]; -#else struct interp_coef coef[PIPE_MAX_SHADER_INPUTS]; -#endif - -#if 0 - struct quad_header quad; -#endif struct { int left[2]; /**< [0] = row0, [1] = row1 */ @@ -142,101 +136,61 @@ struct setup_stage { }; - static struct setup_stage setup; - - -#if 0 -/** - * Basically a cast wrapper. - */ -static INLINE struct setup_stage *setup_stage( struct draw_stage *stage ) -{ - return (struct setup_stage *)stage; -} -#endif - -#if 0 -/** - * Clip setup.quad against the scissor/surface bounds. - */ -static INLINE void -quad_clip(struct setup_stage *setup) -{ - const struct pipe_scissor_state *cliprect = &setup.softpipe->cliprect; - const int minx = (int) cliprect->minx; - const int maxx = (int) cliprect->maxx; - const int miny = (int) cliprect->miny; - const int maxy = (int) cliprect->maxy; - - if (setup.quad.x0 >= maxx || - setup.quad.y0 >= maxy || - setup.quad.x0 + 1 < minx || - setup.quad.y0 + 1 < miny) { - /* totally clipped */ - setup.quad.mask = 0x0; - return; - } - if (setup.quad.x0 < minx) - setup.quad.mask &= (MASK_BOTTOM_RIGHT | MASK_TOP_RIGHT); - if (setup.quad.y0 < miny) - setup.quad.mask &= (MASK_BOTTOM_LEFT | MASK_BOTTOM_RIGHT); - if (setup.quad.x0 == maxx - 1) - setup.quad.mask &= (MASK_BOTTOM_LEFT | MASK_TOP_LEFT); - if (setup.quad.y0 == maxy - 1) - setup.quad.mask &= (MASK_TOP_LEFT | MASK_TOP_RIGHT); -} -#endif - -#if 0 -/** - * Emit a quad (pass to next stage) with clipping. - */ -static INLINE void -clip_emit_quad(struct setup_stage *setup) -{ - quad_clip(setup); - if (setup.quad.mask) { - struct softpipe_context *sp = setup.softpipe; - sp->quad.first->run(sp->quad.first, &setup.quad); - } -} -#endif - /** * Evaluate attribute coefficients (plane equations) to compute * attribute values for the four fragments in a quad. * Eg: four colors will be computed (in AoS format). */ static INLINE void -eval_coeff(uint slot, float x, float y, vector float result[4]) +eval_coeff(uint slot, float x, float y, vector float w, vector float result[4]) { - switch (spu.vertex_info.interp_mode[slot]) { + switch (spu.vertex_info.attrib[slot].interp_mode) { case INTERP_CONSTANT: result[QUAD_TOP_LEFT] = result[QUAD_TOP_RIGHT] = result[QUAD_BOTTOM_LEFT] = result[QUAD_BOTTOM_RIGHT] = setup.coef[slot].a0.v; break; - case INTERP_LINEAR: - /* fall-through, for now */ - default: { - register vector float dadx = setup.coef[slot].dadx.v; - register vector float dady = setup.coef[slot].dady.v; - register vector float topLeft - = spu_add(setup.coef[slot].a0.v, - spu_add(spu_mul(spu_splats(x), dadx), - spu_mul(spu_splats(y), dady))); + vector float dadx = setup.coef[slot].dadx.v; + vector float dady = setup.coef[slot].dady.v; + vector float topLeft = + spu_add(setup.coef[slot].a0.v, + spu_add(spu_mul(spu_splats(x), dadx), + spu_mul(spu_splats(y), dady))); result[QUAD_TOP_LEFT] = topLeft; result[QUAD_TOP_RIGHT] = spu_add(topLeft, dadx); result[QUAD_BOTTOM_LEFT] = spu_add(topLeft, dady); result[QUAD_BOTTOM_RIGHT] = spu_add(spu_add(topLeft, dadx), dady); } + break; + case INTERP_PERSPECTIVE: + { + vector float dadx = setup.coef[slot].dadx.v; + vector float dady = setup.coef[slot].dady.v; + vector float topLeft = + spu_add(setup.coef[slot].a0.v, + spu_add(spu_mul(spu_splats(x), dadx), + spu_mul(spu_splats(y), dady))); + + vector float wInv = spu_re(w); /* 1.0 / w */ + + result[QUAD_TOP_LEFT] = spu_mul(topLeft, wInv); + result[QUAD_TOP_RIGHT] = spu_mul(spu_add(topLeft, dadx), wInv); + result[QUAD_BOTTOM_LEFT] = spu_mul(spu_add(topLeft, dady), wInv); + result[QUAD_BOTTOM_RIGHT] = spu_mul(spu_add(spu_add(topLeft, dadx), dady), wInv); + } + break; + case INTERP_POS: + case INTERP_NONE: + break; + default: + ASSERT(0); } } @@ -246,14 +200,14 @@ eval_coeff(uint slot, float x, float y, vector float result[4]) * XXX this will all be re-written someday. */ static INLINE void -eval_coeff_soa(uint slot, float x, float y, vector float result[4]) +eval_coeff_soa(uint slot, float x, float y, vector float w, vector float result[4]) { - eval_coeff(slot, x, y, result); + eval_coeff(slot, x, y, w, result); _transpose_matrix4x4(result, result); } - +/** Evalute coefficients to get Z for four pixels in a quad */ static INLINE vector float eval_z(float x, float y) { @@ -267,6 +221,20 @@ eval_z(float x, float y) } +/** Evalute coefficients to get W for four pixels in a quad */ +static INLINE vector float +eval_w(float x, float y) +{ + const uint slot = 0; + const float dwdx = setup.coef[slot].dadx.f[3]; + const float dwdy = setup.coef[slot].dady.f[3]; + const float topLeft = setup.coef[slot].a0.f[3] + x * dwdx + y * dwdy; + const vector float topLeftv = spu_splats(topLeft); + const vector float derivs = (vector float) { 0.0, dwdx, dwdy, dwdx + dwdy }; + return spu_add(topLeftv, derivs); +} + + /** * Emit a quad (pass to next stage). No clipping is done. * Note: about 1/5 to 1/7 of the time, mask is zero and this function @@ -274,7 +242,7 @@ eval_z(float x, float y) * overall. */ static INLINE void -emit_quad( int x, int y, mask_t mask ) +emit_quad( int x, int y, mask_t mask) { /* If any bits in mask are set... */ if (spu_extract(spu_orx(mask), 0)) { @@ -284,84 +252,21 @@ emit_quad( int x, int y, mask_t mask ) spu.cur_ctile_status = TILE_STATUS_DIRTY; spu.cur_ztile_status = TILE_STATUS_DIRTY; - if (spu.texture[0].start) { - /* - * Temporary texture mapping path - * This will go away when fragment programs support TEX inst. - */ - const uint unit = 0; - vector float colors[4]; - vector float texcoords[4]; - eval_coeff(2, (float) x, (float) y, texcoords); - - if (spu_extract(mask, 0)) - colors[0] = spu.sample_texture[unit](unit, texcoords[0]); - if (spu_extract(mask, 1)) - colors[1] = spu.sample_texture[unit](unit, texcoords[1]); - if (spu_extract(mask, 2)) - colors[2] = spu.sample_texture[unit](unit, texcoords[2]); - if (spu_extract(mask, 3)) - colors[3] = spu.sample_texture[unit](unit, texcoords[3]); - - - if (spu.texture[1].start) { - /* multi-texture mapping */ - const uint unit = 1; - vector float colors1[4]; - - eval_coeff(2, (float) x, (float) y, texcoords); - - if (spu_extract(mask, 0)) - colors1[0] = spu.sample_texture[unit](unit, texcoords[0]); - if (spu_extract(mask, 1)) - colors1[1] = spu.sample_texture[unit](unit, texcoords[1]); - if (spu_extract(mask, 2)) - colors1[2] = spu.sample_texture[unit](unit, texcoords[2]); - if (spu_extract(mask, 3)) - colors1[3] = spu.sample_texture[unit](unit, texcoords[3]); - - /* hack: modulate first texture by second */ - colors[0] = spu_mul(colors[0], colors1[0]); - colors[1] = spu_mul(colors[1], colors1[1]); - colors[2] = spu_mul(colors[2], colors1[2]); - colors[3] = spu_mul(colors[3], colors1[3]); - } - - { - /* Convert fragment data from AoS to SoA format. - * I.e. (RGBA,RGBA,RGBA,RGBA) -> (RRRR,GGGG,BBBB,AAAA) - * This is temporary! - */ - vector float soa_frag[4]; - _transpose_matrix4x4(soa_frag, colors); - - vector float fragZ = eval_z((float) x, (float) y); - - /* Do all per-fragment/quad operations here, including: - * alpha test, z test, stencil test, blend and framebuffer writing. - */ - spu.fragment_ops(ix, iy, &spu.ctile, &spu.ztile, - fragZ, - soa_frag[0], soa_frag[1], - soa_frag[2], soa_frag[3], - mask); - } - - } - else { + { /* * Run fragment shader, execute per-fragment ops, update fb/tile. */ vector float inputs[4*4], outputs[2*4]; vector float fragZ = eval_z((float) x, (float) y); + vector float fragW = eval_w((float) x, (float) y); /* setup inputs */ #if 0 - eval_coeff_soa(1, (float) x, (float) y, inputs); + eval_coeff_soa(1, (float) x, (float) y, fragW, inputs); #else uint i; for (i = 0; i < spu.vertex_info.num_attribs; i++) { - eval_coeff_soa(i+1, (float) x, (float) y, inputs + i * 4); + eval_coeff_soa(i+1, (float) x, (float) y, fragW, inputs + i * 4); } #endif ASSERT(spu.fragment_program); @@ -379,7 +284,8 @@ emit_quad( int x, int y, mask_t mask ) outputs[0*4+1], outputs[0*4+2], outputs[0*4+3], - mask); + mask, + setup.facing); } } } @@ -389,7 +295,8 @@ emit_quad( int x, int y, mask_t mask ) * Given an X or Y coordinate, return the block/quad coordinate that it * belongs to. */ -static INLINE int block( int x ) +static INLINE int +block(int x) { return x & ~1; } @@ -400,7 +307,8 @@ static INLINE int block( int x ) * the triangle's bounds. * The mask is a uint4 vector and each element will be 0 or 0xffffffff. */ -static INLINE mask_t calculate_mask( int x ) +static INLINE mask_t +calculate_mask(int x) { /* This is a little tricky. * Use & instead of && to avoid branches. @@ -418,7 +326,8 @@ static INLINE mask_t calculate_mask( int x ) /** * Render a horizontal span of quads */ -static void flush_spans( void ) +static void +flush_spans(void) { int minleft, maxright; int x; @@ -446,7 +355,6 @@ static void flush_spans( void ) return; } - /* OK, we're very likely to need the tile data now. * clear or finish waiting if needed. */ @@ -482,9 +390,7 @@ static void flush_spans( void ) * calculate_mask() could be simplified a bit... */ for (x = block(minleft); x <= block(maxright); x += 2) { -#if 1 - emit_quad( x, setup.span.y, calculate_mask( x ) ); -#endif + emit_quad( x, setup.span.y, calculate_mask( x )); } setup.span.y = 0; @@ -493,8 +399,10 @@ static void flush_spans( void ) setup.span.right[1] = 0; } + #if DEBUG_VERTS -static void print_vertex(const struct vertex_header *v) +static void +print_vertex(const struct vertex_header *v) { int i; fprintf(stderr, "Vertex: (%p)\n", v); @@ -506,11 +414,11 @@ static void print_vertex(const struct vertex_header *v) #endif -static boolean setup_sort_vertices(const struct vertex_header *v0, - const struct vertex_header *v1, - const struct vertex_header *v2) +static boolean +setup_sort_vertices(const struct vertex_header *v0, + const struct vertex_header *v1, + const struct vertex_header *v2) { - #if DEBUG_VERTS fprintf(stderr, "Triangle:\n"); print_vertex(v0); @@ -599,13 +507,13 @@ static boolean setup_sort_vertices(const struct vertex_header *v0, * use the prim->det value because its sign is correct. */ { - const float area = (setup.emaj.dx * setup.ebot.dy - - setup.ebot.dx * setup.emaj.dy); + const float area = (setup.emaj.dx * setup.ebot.dy - + setup.ebot.dx * setup.emaj.dy); - setup.oneoverarea = 1.0f / area; + setup.oneOverArea = 1.0f / area; /* _mesa_printf("%s one-over-area %f area %f det %f\n", - __FUNCTION__, setup.oneoverarea, area, prim->det ); + __FUNCTION__, setup.oneOverArea, area, prim->det ); */ } @@ -628,7 +536,7 @@ static boolean setup_sort_vertices(const struct vertex_header *v0, * \param slot which attribute slot */ static INLINE void -const_coeff(uint slot) +const_coeff4(uint slot) { setup.coef[slot].dadx.v = (vector float) {0.0, 0.0, 0.0, 0.0}; setup.coef[slot].dady.v = (vector float) {0.0, 0.0, 0.0, 0.0}; @@ -637,58 +545,6 @@ const_coeff(uint slot) /** - * Compute a0, dadx and dady for a linearly interpolated coefficient, - * for a triangle. - */ -static INLINE void -tri_linear_coeff(uint slot, uint firstComp, uint lastComp) -{ - uint i; - const float *vmin_d = (float *) &setup.vmin->data[slot]; - const float *vmid_d = (float *) &setup.vmid->data[slot]; - const float *vmax_d = (float *) &setup.vmax->data[slot]; - const float x = spu_extract(setup.vmin->data[0], 0) - 0.5f; - const float y = spu_extract(setup.vmin->data[0], 1) - 0.5f; - - for (i = firstComp; i < lastComp; i++) { - float botda = vmid_d[i] - vmin_d[i]; - float majda = vmax_d[i] - vmin_d[i]; - float a = setup.ebot.dy * majda - botda * setup.emaj.dy; - float b = setup.emaj.dx * botda - majda * setup.ebot.dx; - - ASSERT(slot < PIPE_MAX_SHADER_INPUTS); - - setup.coef[slot].dadx.f[i] = a * setup.oneoverarea; - setup.coef[slot].dady.f[i] = b * setup.oneoverarea; - - /* calculate a0 as the value which would be sampled for the - * fragment at (0,0), taking into account that we want to sample at - * pixel centers, in other words (0.5, 0.5). - * - * this is neat but unfortunately not a good way to do things for - * triangles with very large values of dadx or dady as it will - * result in the subtraction and re-addition from a0 of a very - * large number, which means we'll end up loosing a lot of the - * fractional bits and precision from a0. the way to fix this is - * to define a0 as the sample at a pixel center somewhere near vmin - * instead - i'll switch to this later. - */ - setup.coef[slot].a0.f[i] = (vmin_d[i] - - (setup.coef[slot].dadx.f[i] * x + - setup.coef[slot].dady.f[i] * y)); - } - - /* - _mesa_printf("attr[%d].%c: %f dx:%f dy:%f\n", - slot, "xyzw"[i], - setup.coef[slot].a0[i], - setup.coef[slot].dadx.f[i], - setup.coef[slot].dady.f[i]); - */ -} - - -/** * As above, but interp setup all four vector components. */ static INLINE void @@ -708,8 +564,8 @@ tri_linear_coeff4(uint slot) vector float b = spu_sub(spu_mul(spu_splats(setup.emaj.dx), botda), spu_mul(majda, spu_splats(setup.ebot.dx))); - setup.coef[slot].dadx.v = spu_mul(a, spu_splats(setup.oneoverarea)); - setup.coef[slot].dady.v = spu_mul(b, spu_splats(setup.oneoverarea)); + setup.coef[slot].dadx.v = spu_mul(a, spu_splats(setup.oneOverArea)); + setup.coef[slot].dady.v = spu_mul(b, spu_splats(setup.oneOverArea)); vector float tempx = spu_mul(setup.coef[slot].dadx.v, xxxx); vector float tempy = spu_mul(setup.coef[slot].dady.v, yyyy); @@ -718,8 +574,6 @@ tri_linear_coeff4(uint slot) } - -#if 0 /** * Compute a0, dadx and dady for a perspective-corrected interpolant, * for a triangle. @@ -728,82 +582,76 @@ tri_linear_coeff4(uint slot) * Later, when we compute the value at a particular fragment position we'll * divide the interpolated value by the interpolated W at that fragment. */ -static void tri_persp_coeff( unsigned slot, - unsigned i ) +static void +tri_persp_coeff4(uint slot) { - /* premultiply by 1/w: - */ - float mina = setup.vmin->data[slot][i] * setup.vmin->data[0][3]; - float mida = setup.vmid->data[slot][i] * setup.vmid->data[0][3]; - float maxa = setup.vmax->data[slot][i] * setup.vmax->data[0][3]; - - float botda = mida - mina; - float majda = maxa - mina; - float a = setup.ebot.dy * majda - botda * setup.emaj.dy; - float b = setup.emaj.dx * botda - majda * setup.ebot.dx; - - /* - printf("tri persp %d,%d: %f %f %f\n", slot, i, - setup.vmin->data[slot][i], - setup.vmid->data[slot][i], - setup.vmax->data[slot][i] - ); - */ + const vector float xxxx = spu_splats(spu_extract(setup.vmin->data[0], 0) - 0.5f); + const vector float yyyy = spu_splats(spu_extract(setup.vmin->data[0], 1) - 0.5f); + + const vector float vmin_w = spu_splats(spu_extract(setup.vmin->data[0], 3)); + const vector float vmid_w = spu_splats(spu_extract(setup.vmid->data[0], 3)); + const vector float vmax_w = spu_splats(spu_extract(setup.vmax->data[0], 3)); + + vector float vmin_d = setup.vmin->data[slot]; + vector float vmid_d = setup.vmid->data[slot]; + vector float vmax_d = setup.vmax->data[slot]; - assert(slot < PIPE_MAX_SHADER_INPUTS); - assert(i <= 3); + vmin_d = spu_mul(vmin_d, vmin_w); + vmid_d = spu_mul(vmid_d, vmid_w); + vmax_d = spu_mul(vmax_d, vmax_w); - setup.coef[slot].dadx.f[i] = a * setup.oneoverarea; - setup.coef[slot].dady.f[i] = b * setup.oneoverarea; - setup.coef[slot].a0.f[i] = (mina - - (setup.coef[slot].dadx.f[i] * (setup.vmin->data[0][0] - 0.5f) + - setup.coef[slot].dady.f[i] * (setup.vmin->data[0][1] - 0.5f))); + vector float botda = vmid_d - vmin_d; + vector float majda = vmax_d - vmin_d; + + vector float a = spu_sub(spu_mul(spu_splats(setup.ebot.dy), majda), + spu_mul(botda, spu_splats(setup.emaj.dy))); + vector float b = spu_sub(spu_mul(spu_splats(setup.emaj.dx), botda), + spu_mul(majda, spu_splats(setup.ebot.dx))); + + setup.coef[slot].dadx.v = spu_mul(a, spu_splats(setup.oneOverArea)); + setup.coef[slot].dady.v = spu_mul(b, spu_splats(setup.oneOverArea)); + + vector float tempx = spu_mul(setup.coef[slot].dadx.v, xxxx); + vector float tempy = spu_mul(setup.coef[slot].dady.v, yyyy); + + setup.coef[slot].a0.v = spu_sub(vmin_d, spu_add(tempx, tempy)); } -#endif + /** * Compute the setup.coef[] array dadx, dady, a0 values. * Must be called after setup.vmin,vmid,vmax,vprovoke are initialized. */ -static void setup_tri_coefficients(void) +static void +setup_tri_coefficients(void) { -#if 1 uint i; for (i = 0; i < spu.vertex_info.num_attribs; i++) { - switch (spu.vertex_info.interp_mode[i]) { + switch (spu.vertex_info.attrib[i].interp_mode) { case INTERP_NONE: break; - case INTERP_POS: - /*tri_linear_coeff(i, 2, 3);*/ - /* XXX interp W if PERSPECTIVE... */ - tri_linear_coeff4(i); - break; case INTERP_CONSTANT: - const_coeff(i); + const_coeff4(i); break; + case INTERP_POS: + /* fall-through */ case INTERP_LINEAR: tri_linear_coeff4(i); break; case INTERP_PERSPECTIVE: - tri_linear_coeff4(i); /* temporary */ + tri_persp_coeff4(i); break; default: ASSERT(0); } } -#else - ASSERT(spu.vertex_info.interp_mode[0] == INTERP_POS); - ASSERT(spu.vertex_info.interp_mode[1] == INTERP_LINEAR || - spu.vertex_info.interp_mode[1] == INTERP_CONSTANT); - tri_linear_coeff(0, 2, 3); /* slot 0, z */ - tri_linear_coeff(1, 0, 4); /* slot 1, color */ -#endif } -static void setup_tri_edges(void) +static void +setup_tri_edges(void) { float vmin_x = spu_extract(setup.vmin->data[0], 0) + 0.5f; float vmid_x = spu_extract(setup.vmid->data[0], 0) + 0.5f; @@ -833,9 +681,8 @@ static void setup_tri_edges(void) * Render the upper or lower half of a triangle. * Scissoring/cliprect is applied here too. */ -static void subtriangle( struct edge *eleft, - struct edge *eright, - unsigned lines ) +static void +subtriangle(struct edge *eleft, struct edge *eright, unsigned lines) { const int minx = setup.cliprect_minx; const int maxx = setup.cliprect_maxx; @@ -903,12 +750,27 @@ static void subtriangle( struct edge *eleft, } +static float +determinant(const float *v0, const float *v1, const float *v2) +{ + /* edge vectors e = v0 - v2, f = v1 - v2 */ + const float ex = v0[0] - v2[0]; + const float ey = v0[1] - v2[1]; + const float fx = v1[0] - v2[0]; + const float fy = v1[1] - v2[1]; + + /* det = cross(e,f).z */ + return ex * fy - ey * fx; +} + + /** * Draw triangle into tile at (tx, ty) (tile coords) * The tile data should have already been fetched. */ boolean -tri_draw(const float *v0, const float *v1, const float *v2, uint tx, uint ty) +tri_draw(const float *v0, const float *v1, const float *v2, + uint tx, uint ty, uint front_winding) { setup.tx = tx; setup.ty = ty; @@ -919,6 +781,12 @@ tri_draw(const float *v0, const float *v1, const float *v2, uint tx, uint ty) setup.cliprect_maxx = (tx + 1) * TILE_SIZE; setup.cliprect_maxy = (ty + 1) * TILE_SIZE; + /* Before we sort vertices, determine the facing of the triangle, + * which will be needed for front/back-face stencil application + */ + float det = determinant(v0, v1, v2); + setup.facing = (det > 0.0) ^ (front_winding == PIPE_WINDING_CW); + if (!setup_sort_vertices((struct vertex_header *) v0, (struct vertex_header *) v1, (struct vertex_header *) v2)) { @@ -932,19 +800,14 @@ tri_draw(const float *v0, const float *v1, const float *v2, uint tx, uint ty) setup.span.y_flags = 0; setup.span.right[0] = 0; setup.span.right[1] = 0; - /* setup.span.z_mode = tri_z_mode( setup.ctx ); */ - /* init_constant_attribs( setup ); */ - - if (setup.oneoverarea < 0.0) { - /* emaj on left: - */ + if (setup.oneOverArea < 0.0) { + /* emaj on left */ subtriangle( &setup.emaj, &setup.ebot, setup.ebot.lines ); subtriangle( &setup.emaj, &setup.etop, setup.etop.lines ); } else { - /* emaj on right: - */ + /* emaj on right */ subtriangle( &setup.ebot, &setup.emaj, setup.ebot.lines ); subtriangle( &setup.etop, &setup.emaj, setup.etop.lines ); } diff --git a/src/gallium/drivers/cell/spu/spu_tri.h b/src/gallium/drivers/cell/spu/spu_tri.h index aa694dd7c93..abc3d35160e 100644 --- a/src/gallium/drivers/cell/spu/spu_tri.h +++ b/src/gallium/drivers/cell/spu/spu_tri.h @@ -31,7 +31,7 @@ extern boolean -tri_draw(const float *v0, const float *v1, const float *v2, uint tx, uint ty); +tri_draw(const float *v0, const float *v1, const float *v2, uint tx, uint ty, uint front_winding); #endif /* SPU_TRI_H */ diff --git a/src/gallium/drivers/i915simple/i915_prim_emit.c b/src/gallium/drivers/i915simple/i915_prim_emit.c index d194c2fb158..8f1f58b2dd1 100644 --- a/src/gallium/drivers/i915simple/i915_prim_emit.c +++ b/src/gallium/drivers/i915simple/i915_prim_emit.c @@ -77,9 +77,9 @@ emit_hw_vertex( struct i915_context *i915, assert(!i915->dirty); for (i = 0; i < vinfo->num_attribs; i++) { - const uint j = vinfo->src_index[i]; + const uint j = vinfo->attrib[i].src_index; const float *attrib = vertex->data[j]; - switch (vinfo->emit[i]) { + switch (vinfo->attrib[i].emit) { case EMIT_1F: OUT_BATCH( fui(attrib[0]) ); count++; diff --git a/src/gallium/drivers/i915simple/i915_state_derived.c b/src/gallium/drivers/i915simple/i915_state_derived.c index 488615067c5..178d4e8781d 100644 --- a/src/gallium/drivers/i915simple/i915_state_derived.c +++ b/src/gallium/drivers/i915simple/i915_state_derived.c @@ -88,12 +88,12 @@ static void calculate_vertex_layout( struct i915_context *i915 ) if (needW) { draw_emit_vertex_attr(&vinfo, EMIT_4F, INTERP_LINEAR, src); vinfo.hwfmt[0] |= S4_VFMT_XYZW; - vinfo.emit[0] = EMIT_4F; + vinfo.attrib[0].emit = EMIT_4F; } else { draw_emit_vertex_attr(&vinfo, EMIT_3F, INTERP_LINEAR, src); vinfo.hwfmt[0] |= S4_VFMT_XYZ; - vinfo.emit[0] = EMIT_3F; + vinfo.attrib[0].emit = EMIT_3F; } /* hardware point size */ diff --git a/src/gallium/drivers/softpipe/sp_fs_sse.c b/src/gallium/drivers/softpipe/sp_fs_sse.c index 496ed43df26..0111469405f 100644 --- a/src/gallium/drivers/softpipe/sp_fs_sse.c +++ b/src/gallium/drivers/softpipe/sp_fs_sse.c @@ -40,7 +40,7 @@ #include "tgsi/tgsi_sse2.h" -#ifdef PIPE_ARCH_X86 +#if defined(PIPE_ARCH_X86) && defined(PIPE_ARCH_SSE) #include "rtasm/rtasm_x86sse.h" diff --git a/src/gallium/drivers/softpipe/sp_setup.c b/src/gallium/drivers/softpipe/sp_setup.c index bc8263c33e3..13d80173937 100644 --- a/src/gallium/drivers/softpipe/sp_setup.c +++ b/src/gallium/drivers/softpipe/sp_setup.c @@ -773,10 +773,10 @@ static void setup_tri_coefficients( struct setup_context *setup ) /* setup interpolation for all the remaining attributes: */ for (fragSlot = 0; fragSlot < spfs->info.num_inputs; fragSlot++) { - const uint vertSlot = vinfo->src_index[fragSlot]; + const uint vertSlot = vinfo->attrib[fragSlot].src_index; uint j; - switch (vinfo->interp_mode[fragSlot]) { + switch (vinfo->attrib[fragSlot].interp_mode) { case INTERP_CONSTANT: for (j = 0; j < NUM_CHANNELS; j++) const_coeff(setup, &setup->coef[fragSlot], vertSlot, j); @@ -1084,10 +1084,10 @@ setup_line_coefficients(struct setup_context *setup, /* setup interpolation for all the remaining attributes: */ for (fragSlot = 0; fragSlot < spfs->info.num_inputs; fragSlot++) { - const uint vertSlot = vinfo->src_index[fragSlot]; + const uint vertSlot = vinfo->attrib[fragSlot].src_index; uint j; - switch (vinfo->interp_mode[fragSlot]) { + switch (vinfo->attrib[fragSlot].interp_mode) { case INTERP_CONSTANT: for (j = 0; j < NUM_CHANNELS; j++) const_coeff(setup, &setup->coef[fragSlot], vertSlot, j); @@ -1331,10 +1331,10 @@ setup_point( struct setup_context *setup, const_coeff(setup, &setup->posCoef, 0, 3); for (fragSlot = 0; fragSlot < spfs->info.num_inputs; fragSlot++) { - const uint vertSlot = vinfo->src_index[fragSlot]; + const uint vertSlot = vinfo->attrib[fragSlot].src_index; uint j; - switch (vinfo->interp_mode[fragSlot]) { + switch (vinfo->attrib[fragSlot].interp_mode) { case INTERP_CONSTANT: /* fall-through */ case INTERP_LINEAR: diff --git a/src/gallium/include/pipe/p_config.h b/src/gallium/include/pipe/p_config.h index af3746c0265..ef05547819d 100644 --- a/src/gallium/include/pipe/p_config.h +++ b/src/gallium/include/pipe/p_config.h @@ -85,6 +85,14 @@ #define PIPE_ARCH_X86_64 #endif +#if defined(PIPE_ARCH_X86) || defined(PIPE_ARCH_X86_64) +#if defined(PIPE_CC_GCC) && !defined(__SSE2__) +/* #warning SSE2 support requires -msse -msse2 compiler options */ +#else +#define PIPE_ARCH_SSE +#endif +#endif + #if 0 /* FIXME */ #define PIPE_ARCH_PPC #endif diff --git a/src/mesa/drivers/common/descrip.mms b/src/mesa/drivers/common/descrip.mms index 99a2ae6c37d..d5bbc69dfd5 100644 --- a/src/mesa/drivers/common/descrip.mms +++ b/src/mesa/drivers/common/descrip.mms @@ -1,6 +1,6 @@ # Makefile for core library for VMS # contributed by Jouk Jansen [email protected] -# Last revision : 3 October 2007 +# Last revision : 29 September 2008 .first define gl [----.include.gl] @@ -19,7 +19,8 @@ VPATH = RCS INCDIR = [----.include],[--.main],[--.glapi],[--.shader] LIBDIR = [----.lib] -CFLAGS = /include=($(INCDIR),[])/define=(PTHREADS=1)/name=(as_is,short)/float=ieee/ieee=denorm +CFLAGS = /include=($(INCDIR),[])/define=(PTHREADS=1)/name=(as_is,short)\ + /float=ieee/ieee=denorm/warn=disable=(PTRMISMATCH) SOURCES = driverfuncs.c diff --git a/src/mesa/drivers/dri/common/dri_util.c b/src/mesa/drivers/dri/common/dri_util.c index ceedd914fb3..a16cb504c73 100644 --- a/src/mesa/drivers/dri/common/dri_util.c +++ b/src/mesa/drivers/dri/common/dri_util.c @@ -315,9 +315,6 @@ static void driSwapBuffers(__DRIdrawable *dPriv) { __DRIscreen *psp = dPriv->driScreenPriv; - if (!dPriv->numClipRects) - return; - psp->DriverAPI.SwapBuffers(dPriv); driReportDamage(dPriv, dPriv->pClipRects, dPriv->numClipRects); diff --git a/src/mesa/drivers/dri/i915/i830_vtbl.c b/src/mesa/drivers/dri/i915/i830_vtbl.c index 0ab27704d52..773a8b4dd01 100644 --- a/src/mesa/drivers/dri/i915/i830_vtbl.c +++ b/src/mesa/drivers/dri/i915/i830_vtbl.c @@ -566,6 +566,13 @@ i830_destroy_context(struct intel_context *intel) GLuint i; struct i830_context *i830 = i830_context(&intel->ctx); + intel_region_release(&i830->state.draw_region); + intel_region_release(&i830->state.depth_region); + intel_region_release(&i830->meta.draw_region); + intel_region_release(&i830->meta.depth_region); + intel_region_release(&i830->initial.draw_region); + intel_region_release(&i830->initial.depth_region); + for (i = 0; i < I830_TEX_UNITS; i++) { if (i830->state.tex_buffer[i] != NULL) { dri_bo_unreference(i830->state.tex_buffer[i]); diff --git a/src/mesa/drivers/dri/i915/i915_context.c b/src/mesa/drivers/dri/i915/i915_context.c index efcac911aa6..e0ddc7fd613 100644 --- a/src/mesa/drivers/dri/i915/i915_context.c +++ b/src/mesa/drivers/dri/i915/i915_context.c @@ -56,8 +56,6 @@ static const struct dri_extension i915_extensions[] = { {"GL_ARB_shadow", NULL}, {"GL_ARB_texture_non_power_of_two", NULL}, {"GL_EXT_shadow_funcs", NULL}, - /* ARB extn won't work if not enabled */ - {"GL_SGIX_depth_texture", NULL}, {NULL, NULL} }; diff --git a/src/mesa/drivers/dri/i915/i915_context.h b/src/mesa/drivers/dri/i915/i915_context.h index c6958dd8d4b..a2376e50e15 100644 --- a/src/mesa/drivers/dri/i915/i915_context.h +++ b/src/mesa/drivers/dri/i915/i915_context.h @@ -125,6 +125,9 @@ struct i915_fragment_program GLboolean on_hardware; GLboolean error; /* If program is malformed for any reason. */ + /** Record of which phases R registers were last written in. */ + GLuint register_phases[16]; + GLuint indirections; GLuint nr_tex_indirect; GLuint nr_tex_insn; GLuint nr_alu_insn; diff --git a/src/mesa/drivers/dri/i915/i915_program.c b/src/mesa/drivers/dri/i915/i915_program.c index 49193297a8c..350da5e1697 100644 --- a/src/mesa/drivers/dri/i915/i915_program.c +++ b/src/mesa/drivers/dri/i915/i915_program.c @@ -190,6 +190,9 @@ i915_emit_arith(struct i915_fragment_program * p, *(p->csr++) = (A1_SRC0(src0) | A1_SRC1(src1)); *(p->csr++) = (A2_SRC1(src1) | A2_SRC2(src2)); + if (GET_UREG_TYPE(dest) == REG_TYPE_R) + p->register_phases[GET_UREG_NR(dest)] = p->nr_tex_indirect; + p->nr_alu_insn++; return dest; } @@ -237,10 +240,22 @@ GLuint i915_emit_texld( struct i915_fragment_program *p, else { assert(GET_UREG_TYPE(dest) != REG_TYPE_CONST); assert(dest = UREG(GET_UREG_TYPE(dest), GET_UREG_NR(dest))); + /* Can't use unsaved temps for coords, as the phase boundary would result + * in the contents becoming undefined. + */ + assert(GET_UREG_TYPE(coord) != REG_TYPE_U); + + /* Output register being oC or oD defines a phase boundary */ + if (GET_UREG_TYPE(dest) == REG_TYPE_OC || + GET_UREG_TYPE(dest) == REG_TYPE_OD) + p->nr_tex_indirect++; - if (GET_UREG_TYPE(coord) != REG_TYPE_T) { + /* Reading from an r# register whose contents depend on output of the + * current phase defines a phase boundary. + */ + if (GET_UREG_TYPE(coord) == REG_TYPE_R && + p->register_phases[GET_UREG_NR(coord)] == p->nr_tex_indirect) p->nr_tex_indirect++; - } *(p->csr++) = (op | T0_DEST( dest ) | @@ -249,6 +264,9 @@ GLuint i915_emit_texld( struct i915_fragment_program *p, *(p->csr++) = T1_ADDRESS_REG( coord ); *(p->csr++) = T2_MBZ; + if (GET_UREG_TYPE(dest) == REG_TYPE_R) + p->register_phases[GET_UREG_NR(dest)] = p->nr_tex_indirect; + p->nr_tex_insn++; return dest; } @@ -413,7 +431,8 @@ i915_init_program(struct i915_context *i915, struct i915_fragment_program *p) p->on_hardware = 0; p->error = 0; - p->nr_tex_indirect = 1; /* correct? */ + memset(&p->register_phases, 0, sizeof(p->register_phases)); + p->nr_tex_indirect = 1; p->nr_tex_insn = 0; p->nr_alu_insn = 0; p->nr_decl_insn = 0; diff --git a/src/mesa/drivers/dri/i915/i915_texstate.c b/src/mesa/drivers/dri/i915/i915_texstate.c index ae42b102db8..d1b0dcdf319 100644 --- a/src/mesa/drivers/dri/i915/i915_texstate.c +++ b/src/mesa/drivers/dri/i915/i915_texstate.c @@ -307,10 +307,21 @@ i915_update_tex_unit(struct intel_context *intel, GLuint unit, GLuint ss3) } - state[I915_TEXREG_SS4] = INTEL_PACKCOLOR8888(tObj->_BorderChan[0], - tObj->_BorderChan[1], - tObj->_BorderChan[2], - tObj->_BorderChan[3]); + if (firstImage->_BaseFormat == GL_DEPTH_COMPONENT) { + /* GL specs that border color for depth textures is taken from the + * R channel, while the hardware uses A. Spam R into all the channels + * for safety. + */ + state[I915_TEXREG_SS4] = INTEL_PACKCOLOR8888(tObj->_BorderChan[0], + tObj->_BorderChan[0], + tObj->_BorderChan[0], + tObj->_BorderChan[0]); + } else { + state[I915_TEXREG_SS4] = INTEL_PACKCOLOR8888(tObj->_BorderChan[0], + tObj->_BorderChan[1], + tObj->_BorderChan[2], + tObj->_BorderChan[3]); + } I915_ACTIVESTATE(i915, I915_UPLOAD_TEX(unit), GL_TRUE); diff --git a/src/mesa/drivers/dri/i915/i915_vtbl.c b/src/mesa/drivers/dri/i915/i915_vtbl.c index edbbe23e094..7431a9cf76d 100644 --- a/src/mesa/drivers/dri/i915/i915_vtbl.c +++ b/src/mesa/drivers/dri/i915/i915_vtbl.c @@ -490,6 +490,13 @@ i915_destroy_context(struct intel_context *intel) GLuint i; struct i915_context *i915 = i915_context(&intel->ctx); + intel_region_release(&i915->state.draw_region); + intel_region_release(&i915->state.depth_region); + intel_region_release(&i915->meta.draw_region); + intel_region_release(&i915->meta.depth_region); + intel_region_release(&i915->initial.draw_region); + intel_region_release(&i915->initial.depth_region); + for (i = 0; i < I915_TEX_UNITS; i++) { if (i915->state.tex_buffer[i] != NULL) { dri_bo_unreference(i915->state.tex_buffer[i]); diff --git a/src/mesa/drivers/dri/i965/Makefile b/src/mesa/drivers/dri/i965/Makefile index c2d555cd0cf..005460f3547 100644 --- a/src/mesa/drivers/dri/i965/Makefile +++ b/src/mesa/drivers/dri/i965/Makefile @@ -18,8 +18,9 @@ DRIVER_SOURCES = \ intel_screen.c \ intel_span.c \ intel_pixel.c \ - intel_pixel_copy.c \ intel_pixel_bitmap.c \ + intel_pixel_copy.c \ + intel_pixel_draw.c \ intel_state.c \ intel_tex.c \ intel_tex_copy.c \ @@ -51,6 +52,7 @@ DRIVER_SOURCES = \ brw_metaops.c \ brw_misc_state.c \ brw_program.c \ + brw_queryobj.c \ brw_sf.c \ brw_sf_emit.c \ brw_sf_state.c \ diff --git a/src/mesa/drivers/dri/i965/brw_context.c b/src/mesa/drivers/dri/i965/brw_context.c index 92629016d98..474158b484b 100644 --- a/src/mesa/drivers/dri/i965/brw_context.c +++ b/src/mesa/drivers/dri/i965/brw_context.c @@ -33,6 +33,7 @@ #include "main/imports.h" #include "main/api_noop.h" #include "main/vtxfmt.h" +#include "main/simple_list.h" #include "shader/shader_api.h" #include "brw_context.h" @@ -68,6 +69,7 @@ static void brwInitDriverFunctions( struct dd_function_table *functions ) brwInitFragProgFuncs( functions ); brwInitProgFuncs( functions ); + brw_init_queryobj_functions(functions); } @@ -150,9 +152,9 @@ GLboolean brwCreateContext( const __GLcontextModes *mesaVis, ctx->VertexProgram._MaintainTnlProgram = GL_TRUE; ctx->FragmentProgram._MaintainTexEnvProgram = GL_TRUE; - brw_draw_init( brw ); + make_empty_list(&brw->query.active_head); - brw_ProgramCacheInit( ctx ); + brw_draw_init( brw ); return GL_TRUE; } diff --git a/src/mesa/drivers/dri/i965/brw_context.h b/src/mesa/drivers/dri/i965/brw_context.h index b04487ecee9..1c6a0dede0b 100644 --- a/src/mesa/drivers/dri/i965/brw_context.h +++ b/src/mesa/drivers/dri/i965/brw_context.h @@ -130,7 +130,6 @@ struct brw_context; #define BRW_NEW_CONTEXT 0x80 #define BRW_NEW_WM_INPUT_DIMENSIONS 0x100 #define BRW_NEW_INPUT_VARYING 0x200 -#define BRW_NEW_TNL_PROGRAM 0x400 #define BRW_NEW_PSP 0x800 #define BRW_NEW_METAOPS 0x1000 #define BRW_NEW_FENCE 0x2000 @@ -411,7 +410,22 @@ struct brw_tnl_cache { GLuint size, n_items; }; +struct brw_query_object { + struct gl_query_object Base; + /** Doubly linked list of active query objects in the context. */ + struct brw_query_object *prev, *next; + + /** Last query BO associated with this query. */ + dri_bo *bo; + /** First index in bo with query data for this object. */ + int first_index; + /** Last index in bo with query data for this object. */ + int last_index; + + /* Total count of pixels from previous BOs */ + unsigned int count; +}; struct brw_context { @@ -488,10 +502,6 @@ struct brw_context GLboolean active; } metaops; - /* Track fixed function t&l in a vertex program: - */ - struct gl_vertex_program *tnl_program; - struct brw_tnl_cache tnl_program_cache; /* Active vertex program: */ @@ -631,7 +641,12 @@ struct brw_context dri_bo *vp_bo; } cc; - + struct { + struct brw_query_object active_head; + dri_bo *bo; + int index; + GLboolean active; + } query; /* Used to give every program string a unique id */ GLuint program_id; @@ -656,7 +671,13 @@ GLboolean brwCreateContext( const __GLcontextModes *mesaVis, __DRIcontextPrivate *driContextPriv, void *sharedContextPrivate); - +/*====================================================================== + * brw_queryobj.c + */ +void brw_init_queryobj_functions(struct dd_function_table *functions); +void brw_prepare_query_begin(struct brw_context *brw); +void brw_emit_query_begin(struct brw_context *brw); +void brw_emit_query_end(struct brw_context *brw); /*====================================================================== * brw_state.c diff --git a/src/mesa/drivers/dri/i965/brw_draw.c b/src/mesa/drivers/dri/i965/brw_draw.c index 9a353fc7b66..6c71b4abcf0 100644 --- a/src/mesa/drivers/dri/i965/brw_draw.c +++ b/src/mesa/drivers/dri/i965/brw_draw.c @@ -382,7 +382,6 @@ void brw_draw_prims( GLcontext *ctx, return; } - /* Make a first attempt at drawing: */ retval = brw_try_draw_prims(ctx, arrays, prim, nr_prims, ib, min_index, max_index); @@ -395,6 +394,7 @@ void brw_draw_prims( GLcontext *ctx, _swsetup_Wakeup(ctx); _tnl_draw_prims(ctx, arrays, prim, nr_prims, ib, min_index, max_index); } + } void brw_draw_init( struct brw_context *brw ) @@ -409,8 +409,18 @@ void brw_draw_init( struct brw_context *brw ) void brw_draw_destroy( struct brw_context *brw ) { + int i; + if (brw->vb.upload.bo != NULL) { dri_bo_unreference(brw->vb.upload.bo); brw->vb.upload.bo = NULL; } + + for (i = 0; i < VERT_ATTRIB_MAX; i++) { + dri_bo_unreference(brw->vb.inputs[i].bo); + brw->vb.inputs[i].bo = NULL; + } + + dri_bo_unreference(brw->ib.bo); + brw->ib.bo = NULL; } diff --git a/src/mesa/drivers/dri/i965/brw_draw_upload.c b/src/mesa/drivers/dri/i965/brw_draw_upload.c index 303eaac5cf9..7b88b5eaa1e 100644 --- a/src/mesa/drivers/dri/i965/brw_draw_upload.c +++ b/src/mesa/drivers/dri/i965/brw_draw_upload.c @@ -365,8 +365,10 @@ static void brw_prepare_vertices(struct brw_context *brw) if (i == 0) { /* Position array not properly enabled: */ - if (input->glarray->StrideB == 0) - return; + if (input->glarray->StrideB == 0) { + intel->Fallback = 1; + return; + } interleave = input->glarray->StrideB; ptr = input->glarray->Ptr; @@ -413,6 +415,8 @@ static void brw_prepare_vertices(struct brw_context *brw) copy_array_to_vbo_array(brw, upload[i], upload[i]->element_size); } } + + brw_prepare_query_begin(brw); } static void brw_emit_vertices(struct brw_context *brw) @@ -433,6 +437,7 @@ static void brw_emit_vertices(struct brw_context *brw) enabled[nr_enabled++] = input; } + brw_emit_query_begin(brw); /* Now emit VB and VEP state packets. * diff --git a/src/mesa/drivers/dri/i965/brw_fallback.c b/src/mesa/drivers/dri/i965/brw_fallback.c index 2f6b7febbde..4ea660a51a3 100644 --- a/src/mesa/drivers/dri/i965/brw_fallback.c +++ b/src/mesa/drivers/dri/i965/brw_fallback.c @@ -74,10 +74,7 @@ static GLboolean do_check_fallback(struct brw_context *brw) if (texUnit->_ReallyEnabled) { struct intel_texture_object *intelObj = intel_texture_object(texUnit->_Current); struct gl_texture_image *texImage = intelObj->base.Image[0][intelObj->firstLevel]; - if (texImage->Border || - ((texImage->_BaseFormat == GL_DEPTH_COMPONENT) && - ((texImage->TexObject->WrapS == GL_CLAMP_TO_BORDER) || - (texImage->TexObject->WrapT == GL_CLAMP_TO_BORDER)))) { + if (texImage->Border) { DBG("FALLBACK: texture border\n"); return GL_TRUE; } diff --git a/src/mesa/drivers/dri/i965/brw_queryobj.c b/src/mesa/drivers/dri/i965/brw_queryobj.c new file mode 100644 index 00000000000..a1a1353dee7 --- /dev/null +++ b/src/mesa/drivers/dri/i965/brw_queryobj.c @@ -0,0 +1,263 @@ +/* + * Copyright © 2008 Intel Corporation + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING + * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS + * IN THE SOFTWARE. + * + * Authors: + * Eric Anholt <[email protected]> + * + */ + +/** @file support for ARB_query_object + * + * ARB_query_object is implemented by using the PIPE_CONTROL command to stall + * execution on the completion of previous depth tests, and write the + * current PS_DEPTH_COUNT to a buffer object. + * + * We use before and after counts when drawing during a query so that + * we don't pick up other clients' query data in ours. To reduce overhead, + * a single BO is used to record the query data for all active queries at + * once. This also gives us a simple bound on how much batchbuffer space is + * required for handling queries, so that we can be sure that we won't + * have to emit a batchbuffer without getting the ending PS_DEPTH_COUNT. + */ +#include "main/simple_list.h" +#include "main/imports.h" + +#include "brw_context.h" +#include "intel_batchbuffer.h" +#include "intel_reg.h" + +/** Waits on the query object's BO and totals the results for this query */ +static void +brw_queryobj_get_results(struct brw_query_object *query) +{ + int i; + uint64_t *results; + + if (query->bo == NULL) + return; + + /* Map and count the pixels from the current query BO */ + dri_bo_map(query->bo, GL_FALSE); + results = query->bo->virtual; + for (i = query->first_index; i <= query->last_index; i++) { + query->Base.Result += results[i * 2 + 1] - results[i * 2]; + } + dri_bo_unmap(query->bo); + + dri_bo_unreference(query->bo); + query->bo = NULL; +} + +static struct gl_query_object * +brw_new_query_object(GLcontext *ctx, GLuint id) +{ + struct brw_query_object *query; + + query = _mesa_calloc(sizeof(struct brw_query_object)); + + query->Base.Id = id; + query->Base.Result = 0; + query->Base.Active = GL_FALSE; + query->Base.Ready = GL_TRUE; + + return &query->Base; +} + +static void +brw_delete_query(GLcontext *ctx, struct gl_query_object *q) +{ + struct brw_query_object *query = (struct brw_query_object *)q; + + dri_bo_unreference(query->bo); + _mesa_free(query); +} + +static void +brw_begin_query(GLcontext *ctx, struct gl_query_object *q) +{ + struct brw_context *brw = brw_context(ctx); + struct intel_context *intel = intel_context(ctx); + struct brw_query_object *query = (struct brw_query_object *)q; + + /* Reset our driver's tracking of query state. */ + dri_bo_unreference(query->bo); + query->bo = NULL; + query->first_index = -1; + query->last_index = -1; + + insert_at_head(&brw->query.active_head, query); + intel->stats_wm++; +} + +/** + * Begin the ARB_occlusion_query query on a query object. + */ +static void +brw_end_query(GLcontext *ctx, struct gl_query_object *q) +{ + struct brw_context *brw = brw_context(ctx); + struct intel_context *intel = intel_context(ctx); + struct brw_query_object *query = (struct brw_query_object *)q; + + /* Flush the batchbuffer in case it has writes to our query BO. + * Have later queries write to a new query BO so that further rendering + * doesn't delay the collection of our results. + */ + if (query->bo) { + brw_emit_query_end(brw); + intel_batchbuffer_flush(intel->batch); + + dri_bo_unreference(brw->query.bo); + brw->query.bo = NULL; + } + + remove_from_list(query); + + intel->stats_wm--; +} + +static void brw_wait_query(GLcontext *ctx, struct gl_query_object *q) +{ + struct brw_query_object *query = (struct brw_query_object *)q; + + brw_queryobj_get_results(query); + query->Base.Ready = GL_TRUE; +} + +static void brw_check_query(GLcontext *ctx, struct gl_query_object *q) +{ + /* XXX: Need to expose dri_bo_is_idle from bufmgr. */ +#if 0 + struct brw_query_object *query = (struct brw_query_object *)q; + + if (dri_bo_is_idle(query->bo)) { + brw_queryobj_get_results(query); + query->Base.Ready = GL_TRUE; + } +#else + brw_wait_query(ctx, q); +#endif +} + +/** Called to set up the query BO and account for its aperture space */ +void +brw_prepare_query_begin(struct brw_context *brw) +{ + struct intel_context *intel = &brw->intel; + dri_bo *aper_array[] = { + intel->batch->buf, + brw->query.bo, + }; + + /* Skip if we're not doing any queries. */ + if (is_empty_list(&brw->query.active_head)) + return; + + /* Get a new query BO if we're going to need it. */ + if (brw->query.bo == NULL || + brw->query.index * 2 + 1 >= 4096 / sizeof(uint64_t)) { + dri_bo_unreference(brw->query.bo); + brw->query.bo = NULL; + + brw->query.bo = dri_bo_alloc(intel->bufmgr, "query", 4096, 1); + brw->query.index = 0; + } + + if (dri_bufmgr_check_aperture_space(aper_array, ARRAY_SIZE(aper_array))) + intel_batchbuffer_flush(intel->batch); +} + +/** Called just before primitive drawing to get a beginning PS_DEPTH_COUNT. */ +void +brw_emit_query_begin(struct brw_context *brw) +{ + struct intel_context *intel = &brw->intel; + struct brw_query_object *query; + + /* Skip if we're not doing any queries, or we've emitted the start. */ + if (brw->query.active || is_empty_list(&brw->query.active_head)) + return; + + BEGIN_BATCH(4, IGNORE_CLIPRECTS); + OUT_BATCH(_3DSTATE_PIPE_CONTROL | + PIPE_CONTROL_DEPTH_STALL | + PIPE_CONTROL_WRITE_DEPTH_COUNT); + /* This object could be mapped cacheable, but we don't have an exposed + * mechanism to support that. Since it's going uncached, tell GEM that + * we're writing to it. The usual clflush should be all that's required + * to pick up the results. + */ + OUT_RELOC(brw->query.bo, + I915_GEM_DOMAIN_INSTRUCTION, I915_GEM_DOMAIN_INSTRUCTION, + PIPE_CONTROL_GLOBAL_GTT_WRITE | + ((brw->query.index * 2) * sizeof(uint64_t))); + OUT_BATCH(0); + OUT_BATCH(0); + ADVANCE_BATCH(); + + foreach(query, &brw->query.active_head) { + if (query->bo != brw->query.bo) { + if (query->bo != NULL) + brw_queryobj_get_results(query); + dri_bo_reference(brw->query.bo); + query->bo = brw->query.bo; + query->first_index = brw->query.index; + } + query->last_index = brw->query.index; + } + brw->query.active = GL_TRUE; +} + +/** Called at batchbuffer flush to get an ending PS_DEPTH_COUNT */ +void +brw_emit_query_end(struct brw_context *brw) +{ + struct intel_context *intel = &brw->intel; + + if (!brw->query.active) + return; + + BEGIN_BATCH(4, IGNORE_CLIPRECTS); + OUT_BATCH(_3DSTATE_PIPE_CONTROL | + PIPE_CONTROL_DEPTH_STALL | + PIPE_CONTROL_WRITE_DEPTH_COUNT); + OUT_RELOC(brw->query.bo, + I915_GEM_DOMAIN_INSTRUCTION, I915_GEM_DOMAIN_INSTRUCTION, + PIPE_CONTROL_GLOBAL_GTT_WRITE | + ((brw->query.index * 2 + 1) * sizeof(uint64_t))); + OUT_BATCH(0); + OUT_BATCH(0); + ADVANCE_BATCH(); + + brw->query.active = GL_FALSE; + brw->query.index++; +} + +void brw_init_queryobj_functions(struct dd_function_table *functions) +{ + functions->NewQueryObject = brw_new_query_object; + functions->DeleteQuery = brw_delete_query; + functions->BeginQuery = brw_begin_query; + functions->EndQuery = brw_end_query; + functions->CheckQuery = brw_check_query; + functions->WaitQuery = brw_wait_query; +} diff --git a/src/mesa/drivers/dri/i965/brw_state.h b/src/mesa/drivers/dri/i965/brw_state.h index 3ea6151ae95..4c04036ef08 100644 --- a/src/mesa/drivers/dri/i965/brw_state.h +++ b/src/mesa/drivers/dri/i965/brw_state.h @@ -74,7 +74,6 @@ const struct brw_tracked_state brw_wm_unit; const struct brw_tracked_state brw_psp_urb_cbs; const struct brw_tracked_state brw_active_vertprog; -const struct brw_tracked_state brw_tnl_vertprog; const struct brw_tracked_state brw_pipe_control; const struct brw_tracked_state brw_clear_surface_cache; diff --git a/src/mesa/drivers/dri/i965/brw_state_cache.c b/src/mesa/drivers/dri/i965/brw_state_cache.c index 1318dea5940..d5b51664066 100644 --- a/src/mesa/drivers/dri/i965/brw_state_cache.c +++ b/src/mesa/drivers/dri/i965/brw_state_cache.c @@ -497,9 +497,10 @@ void brw_destroy_cache( struct brw_context *brw ) GLuint i; brw_clear_cache(brw); - for (i = 0; i < BRW_MAX_CACHE; i++) + for (i = 0; i < BRW_MAX_CACHE; i++) { + dri_bo_unreference(brw->cache.last_bo[i]); free(brw->cache.name[i]); - + } free(brw->cache.items); brw->cache.items = NULL; brw->cache.size = 0; diff --git a/src/mesa/drivers/dri/i965/brw_state_upload.c b/src/mesa/drivers/dri/i965/brw_state_upload.c index 7d4fd467b1c..b6a52843a81 100644 --- a/src/mesa/drivers/dri/i965/brw_state_upload.c +++ b/src/mesa/drivers/dri/i965/brw_state_upload.c @@ -45,7 +45,6 @@ const struct brw_tracked_state *atoms[] = { &brw_check_fallback, - &brw_tnl_vertprog, &brw_active_vertprog, &brw_wm_input_sizes, &brw_vs_prog, diff --git a/src/mesa/drivers/dri/i965/brw_vs.h b/src/mesa/drivers/dri/i965/brw_vs.h index 41a33ffe387..22388ec99d0 100644 --- a/src/mesa/drivers/dri/i965/brw_vs.h +++ b/src/mesa/drivers/dri/i965/brw_vs.h @@ -80,8 +80,4 @@ struct brw_vs_compile { void brw_vs_emit( struct brw_vs_compile *c ); - -void brw_ProgramCacheDestroy( GLcontext *ctx ); -void brw_ProgramCacheInit( GLcontext *ctx ); - #endif diff --git a/src/mesa/drivers/dri/i965/brw_vs_tnl.c b/src/mesa/drivers/dri/i965/brw_vs_tnl.c index 9b04f19112a..eacc289f1f1 100644 --- a/src/mesa/drivers/dri/i965/brw_vs_tnl.c +++ b/src/mesa/drivers/dri/i965/brw_vs_tnl.c @@ -33,1635 +33,15 @@ #include "main/glheader.h" #include "main/macros.h" #include "main/enums.h" -#include "shader/prog_parameter.h" -#include "shader/prog_print.h" -#include "shader/prog_statevars.h" #include "brw_vs.h" #include "brw_state.h" -struct state_key { - unsigned light_global_enabled:1; - unsigned light_local_viewer:1; - unsigned light_twoside:1; - unsigned light_color_material:1; - unsigned light_color_material_mask:12; - unsigned light_material_mask:12; - unsigned normalize:1; - unsigned rescale_normals:1; - unsigned fog_source_is_depth:1; - unsigned tnl_do_vertex_fog:1; - unsigned separate_specular:1; - unsigned fog_option:2; - unsigned point_attenuated:1; - unsigned texture_enabled_global:1; - unsigned fragprog_inputs_read:12; - - struct { - unsigned light_enabled:1; - unsigned light_eyepos3_is_zero:1; - unsigned light_spotcutoff_is_180:1; - unsigned light_attenuated:1; - unsigned texunit_really_enabled:1; - unsigned texmat_enabled:1; - unsigned texgen_enabled:4; - unsigned texgen_mode0:4; - unsigned texgen_mode1:4; - unsigned texgen_mode2:4; - unsigned texgen_mode3:4; - } unit[8]; -}; - - - -#define FOG_NONE 0 -#define FOG_LINEAR 1 -#define FOG_EXP 2 -#define FOG_EXP2 3 - -static GLuint translate_fog_mode( GLenum mode ) -{ - switch (mode) { - case GL_LINEAR: return FOG_LINEAR; - case GL_EXP: return FOG_EXP; - case GL_EXP2: return FOG_EXP2; - default: return FOG_NONE; - } -} - -#define TXG_NONE 0 -#define TXG_OBJ_LINEAR 1 -#define TXG_EYE_LINEAR 2 -#define TXG_SPHERE_MAP 3 -#define TXG_REFLECTION_MAP 4 -#define TXG_NORMAL_MAP 5 - -static GLuint translate_texgen( GLboolean enabled, GLenum mode ) -{ - if (!enabled) - return TXG_NONE; - - switch (mode) { - case GL_OBJECT_LINEAR: return TXG_OBJ_LINEAR; - case GL_EYE_LINEAR: return TXG_EYE_LINEAR; - case GL_SPHERE_MAP: return TXG_SPHERE_MAP; - case GL_REFLECTION_MAP_NV: return TXG_REFLECTION_MAP; - case GL_NORMAL_MAP_NV: return TXG_NORMAL_MAP; - default: return TXG_NONE; - } -} - -static void make_state_key( GLcontext *ctx, struct state_key *key ) -{ - struct brw_context *brw = brw_context(ctx); - const struct gl_fragment_program *fp = brw->fragment_program; - GLuint i; - - /* This now relies on texenvprogram.c being active: - */ - assert(fp); - - memset(key, 0, sizeof(*key)); - - /* BRW_NEW_FRAGMENT_PROGRAM */ - key->fragprog_inputs_read = fp->Base.InputsRead; - - /* _NEW_LIGHT */ - key->separate_specular = (brw->attribs.Light->Model.ColorControl == - GL_SEPARATE_SPECULAR_COLOR); - - /* _NEW_LIGHT */ - if (brw->attribs.Light->Enabled) { - key->light_global_enabled = 1; - - if (brw->attribs.Light->Model.LocalViewer) - key->light_local_viewer = 1; - - if (brw->attribs.Light->Model.TwoSide) - key->light_twoside = 1; - - if (brw->attribs.Light->ColorMaterialEnabled) { - key->light_color_material = 1; - key->light_color_material_mask = brw->attribs.Light->ColorMaterialBitmask; - } - - /* BRW_NEW_INPUT_VARYING */ - - /* For these programs, material values are stuffed into the - * generic slots: - */ - for (i = 0 ; i < MAT_ATTRIB_MAX ; i++) - if (brw->vb.info.varying & (1<<(VERT_ATTRIB_GENERIC0 + i))) - key->light_material_mask |= 1<<i; - - for (i = 0; i < MAX_LIGHTS; i++) { - struct gl_light *light = &brw->attribs.Light->Light[i]; - - if (light->Enabled) { - key->unit[i].light_enabled = 1; - - if (light->EyePosition[3] == 0.0) - key->unit[i].light_eyepos3_is_zero = 1; - - if (light->SpotCutoff == 180.0) - key->unit[i].light_spotcutoff_is_180 = 1; - - if (light->ConstantAttenuation != 1.0 || - light->LinearAttenuation != 0.0 || - light->QuadraticAttenuation != 0.0) - key->unit[i].light_attenuated = 1; - } - } - } - - /* _NEW_TRANSFORM */ - if (brw->attribs.Transform->Normalize) - key->normalize = 1; - - if (brw->attribs.Transform->RescaleNormals) - key->rescale_normals = 1; - - /* BRW_NEW_FRAGMENT_PROGRAM */ - key->fog_option = translate_fog_mode(fp->FogOption); - if (key->fog_option) - key->fragprog_inputs_read |= FRAG_BIT_FOGC; - - /* _NEW_FOG */ - if (brw->attribs.Fog->FogCoordinateSource == GL_FRAGMENT_DEPTH_EXT) - key->fog_source_is_depth = 1; - - /* _NEW_HINT, ??? */ - if (1) - key->tnl_do_vertex_fog = 1; - - /* _NEW_POINT */ - if (brw->attribs.Point->_Attenuated) - key->point_attenuated = 1; - - /* _NEW_TEXTURE */ - if (brw->attribs.Texture->_TexGenEnabled || - brw->attribs.Texture->_TexMatEnabled || - brw->attribs.Texture->_EnabledUnits) - key->texture_enabled_global = 1; - - for (i = 0; i < MAX_TEXTURE_UNITS; i++) { - struct gl_texture_unit *texUnit = &brw->attribs.Texture->Unit[i]; - - if (texUnit->_ReallyEnabled) - key->unit[i].texunit_really_enabled = 1; - - if (brw->attribs.Texture->_TexMatEnabled & ENABLE_TEXMAT(i)) - key->unit[i].texmat_enabled = 1; - - if (texUnit->TexGenEnabled) { - key->unit[i].texgen_enabled = 1; - - key->unit[i].texgen_mode0 = - translate_texgen( texUnit->TexGenEnabled & (1<<0), - texUnit->GenModeS ); - key->unit[i].texgen_mode1 = - translate_texgen( texUnit->TexGenEnabled & (1<<1), - texUnit->GenModeT ); - key->unit[i].texgen_mode2 = - translate_texgen( texUnit->TexGenEnabled & (1<<2), - texUnit->GenModeR ); - key->unit[i].texgen_mode3 = - translate_texgen( texUnit->TexGenEnabled & (1<<3), - texUnit->GenModeQ ); - } - } -} - - - -/* Very useful debugging tool - produces annotated listing of - * generated program with line/function references for each - * instruction back into this file: - */ -#define DISASSEM 0 - -/* Should be tunable by the driver - do we want to do matrix - * multiplications with DP4's or with MUL/MAD's? SSE works better - * with the latter, drivers may differ. - */ -#define PREFER_DP4 1 - - -/* Use uregs to represent registers internally, translate to Mesa's - * expected formats on emit. - * - * NOTE: These are passed by value extensively in this file rather - * than as usual by pointer reference. If this disturbs you, try - * remembering they are just 32bits in size. - * - * GCC is smart enough to deal with these dword-sized structures in - * much the same way as if I had defined them as dwords and was using - * macros to access and set the fields. This is much nicer and easier - * to evolve. - */ -struct ureg { - GLuint file:4; - GLint idx:8; /* relative addressing may be negative */ - GLuint negate:1; - GLuint swz:12; - GLuint pad:7; -}; - - -struct tnl_program { - const struct state_key *state; - struct gl_vertex_program *program; - - GLuint nr_instructions; - GLuint temp_in_use; - GLuint temp_reserved; - - struct ureg eye_position; - struct ureg eye_position_normalized; - struct ureg eye_normal; - struct ureg identity; - - GLuint materials; - GLuint color_materials; -}; - - -const static struct ureg undef = { - PROGRAM_UNDEFINED, - ~0, - 0, - 0, - 0 -}; - -/* Local shorthand: - */ -#define X SWIZZLE_X -#define Y SWIZZLE_Y -#define Z SWIZZLE_Z -#define W SWIZZLE_W - - -/* Construct a ureg: - */ -static struct ureg make_ureg(GLuint file, GLint idx) -{ - struct ureg reg; - reg.file = file; - reg.idx = idx; - reg.negate = 0; - reg.swz = SWIZZLE_NOOP; - reg.pad = 0; - return reg; -} - - - -static struct ureg ureg_negate( struct ureg reg ) -{ - reg.negate ^= 1; - return reg; -} - - -static struct ureg swizzle( struct ureg reg, int x, int y, int z, int w ) -{ - reg.swz = MAKE_SWIZZLE4(GET_SWZ(reg.swz, x), - GET_SWZ(reg.swz, y), - GET_SWZ(reg.swz, z), - GET_SWZ(reg.swz, w)); - - return reg; -} - -static struct ureg swizzle1( struct ureg reg, int x ) -{ - return swizzle(reg, x, x, x, x); -} - -static struct ureg get_temp( struct tnl_program *p ) -{ - int bit = ffs( ~p->temp_in_use ); - if (!bit) { - fprintf(stderr, "%s: out of temporaries\n", __FILE__); - assert(0); - } - - if (bit > p->program->Base.NumTemporaries) - p->program->Base.NumTemporaries = bit; - - p->temp_in_use |= 1<<(bit-1); - return make_ureg(PROGRAM_TEMPORARY, bit-1); -} - -static struct ureg reserve_temp( struct tnl_program *p ) -{ - struct ureg temp = get_temp( p ); - p->temp_reserved |= 1<<temp.idx; - return temp; -} - -static void release_temp( struct tnl_program *p, struct ureg reg ) -{ - if (reg.file == PROGRAM_TEMPORARY) { - p->temp_in_use &= ~(1<<reg.idx); - p->temp_in_use |= p->temp_reserved; /* can't release reserved temps */ - } -} - -static void release_temps( struct tnl_program *p ) -{ - p->temp_in_use = p->temp_reserved; -} - - - -static struct ureg register_input( struct tnl_program *p, GLuint input ) -{ - assert(input < 32); - - p->program->Base.InputsRead |= (1<<input); - return make_ureg(PROGRAM_INPUT, input); -} - -static struct ureg register_output( struct tnl_program *p, GLuint output ) -{ - p->program->Base.OutputsWritten |= (1<<output); - return make_ureg(PROGRAM_OUTPUT, output); -} - -static struct ureg register_const4f( struct tnl_program *p, - GLfloat s0, - GLfloat s1, - GLfloat s2, - GLfloat s3) -{ - GLfloat values[4]; - GLint idx; - GLuint swizzle; - values[0] = s0; - values[1] = s1; - values[2] = s2; - values[3] = s3; - idx = _mesa_add_unnamed_constant( p->program->Base.Parameters, values, 4, - &swizzle); - assert(swizzle == SWIZZLE_NOOP); /* Need to handle swizzle in reg setup */ - return make_ureg(PROGRAM_STATE_VAR, idx); -} - -#define register_const1f(p, s0) register_const4f(p, s0, 0, 0, 1) -#define register_scalar_const(p, s0) register_const4f(p, s0, s0, s0, s0) -#define register_const2f(p, s0, s1) register_const4f(p, s0, s1, 0, 1) -#define register_const3f(p, s0, s1, s2) register_const4f(p, s0, s1, s2, 1) - -static GLboolean is_undef( struct ureg reg ) -{ - return reg.file == PROGRAM_UNDEFINED; -} - -static struct ureg get_identity_param( struct tnl_program *p ) -{ - if (is_undef(p->identity)) - p->identity = register_const4f(p, 0,0,0,1); - - return p->identity; -} - -static struct ureg register_param5( struct tnl_program *p, - GLint s0, - GLint s1, - GLint s2, - GLint s3, - GLint s4) -{ - gl_state_index tokens[STATE_LENGTH]; - GLint idx; - tokens[0] = s0; - tokens[1] = s1; - tokens[2] = s2; - tokens[3] = s3; - tokens[4] = s4; - idx = _mesa_add_state_reference( p->program->Base.Parameters, tokens ); - return make_ureg(PROGRAM_STATE_VAR, idx); -} - - -#define register_param1(p,s0) register_param5(p,s0,0,0,0,0) -#define register_param2(p,s0,s1) register_param5(p,s0,s1,0,0,0) -#define register_param3(p,s0,s1,s2) register_param5(p,s0,s1,s2,0,0) -#define register_param4(p,s0,s1,s2,s3) register_param5(p,s0,s1,s2,s3,0) - - -static void register_matrix_param5( struct tnl_program *p, - GLint s0, /* matrix name */ - GLint s1, /* texture matrix number */ - GLint s2, /* first row */ - GLint s3, /* last row */ - GLint s4, /* modifier */ - struct ureg *matrix ) -{ - GLint i; - - /* This is a bit sad as the support is there to pull the whole - * matrix out in one go: - */ - for (i = 0; i <= s3 - s2; i++) - matrix[i] = register_param5( p, s0, s1, i, i, s4 ); -} - - -static void emit_arg( struct prog_src_register *src, - struct ureg reg ) -{ - src->File = reg.file; - src->Index = reg.idx; - src->Swizzle = reg.swz; - src->RelAddr = 0; - src->NegateBase = reg.negate; - src->Abs = 0; - src->NegateAbs = 0; -} - -static void emit_dst( struct prog_dst_register *dst, - struct ureg reg, GLuint mask ) -{ - dst->File = reg.file; - dst->Index = reg.idx; - /* allow zero as a shorthand for xyzw */ - dst->WriteMask = mask ? mask : WRITEMASK_XYZW; - dst->CondMask = 0; - dst->CondSwizzle = 0; - dst->CondSrc = 0; - dst->pad = 0; -} - -static void debug_insn( struct prog_instruction *inst, const char *fn, - GLuint line ) -{ - if (DISASSEM) { - static const char *last_fn; - - if (fn != last_fn) { - last_fn = fn; - _mesa_printf("%s:\n", fn); - } - - _mesa_printf("%d:\t", line); - _mesa_print_instruction(inst); - } -} - - -static void emit_op3fn(struct tnl_program *p, - GLuint op, - struct ureg dest, - GLuint mask, - struct ureg src0, - struct ureg src1, - struct ureg src2, - const char *fn, - GLuint line) -{ - GLuint nr = p->program->Base.NumInstructions++; - - if (nr >= p->nr_instructions) { - int new_nr_instructions = p->nr_instructions * 2; - - p->program->Base.Instructions = - _mesa_realloc(p->program->Base.Instructions, - sizeof(struct prog_instruction) * p->nr_instructions, - sizeof(struct prog_instruction) * new_nr_instructions); - p->nr_instructions = new_nr_instructions; - } - - { - struct prog_instruction *inst = &p->program->Base.Instructions[nr]; - memset(inst, 0, sizeof(*inst)); - inst->Opcode = op; - inst->StringPos = 0; - inst->Data = 0; - - emit_arg( &inst->SrcReg[0], src0 ); - emit_arg( &inst->SrcReg[1], src1 ); - emit_arg( &inst->SrcReg[2], src2 ); - - emit_dst( &inst->DstReg, dest, mask ); - - debug_insn(inst, fn, line); - } -} - - - -#define emit_op3(p, op, dst, mask, src0, src1, src2) \ - emit_op3fn(p, op, dst, mask, src0, src1, src2, __FUNCTION__, __LINE__) - -#define emit_op2(p, op, dst, mask, src0, src1) \ - emit_op3fn(p, op, dst, mask, src0, src1, undef, __FUNCTION__, __LINE__) - -#define emit_op1(p, op, dst, mask, src0) \ - emit_op3fn(p, op, dst, mask, src0, undef, undef, __FUNCTION__, __LINE__) - - -static struct ureg make_temp( struct tnl_program *p, struct ureg reg ) -{ - if (reg.file == PROGRAM_TEMPORARY && - !(p->temp_reserved & (1<<reg.idx))) - return reg; - else { - struct ureg temp = get_temp(p); - emit_op1(p, OPCODE_MOV, temp, 0, reg); - return temp; - } -} - - -/* Currently no tracking performed of input/output/register size or - * active elements. Could be used to reduce these operations, as - * could the matrix type. - */ -static void emit_matrix_transform_vec4( struct tnl_program *p, - struct ureg dest, - const struct ureg *mat, - struct ureg src) -{ - emit_op2(p, OPCODE_DP4, dest, WRITEMASK_X, src, mat[0]); - emit_op2(p, OPCODE_DP4, dest, WRITEMASK_Y, src, mat[1]); - emit_op2(p, OPCODE_DP4, dest, WRITEMASK_Z, src, mat[2]); - emit_op2(p, OPCODE_DP4, dest, WRITEMASK_W, src, mat[3]); -} - -/* This version is much easier to implement if writemasks are not - * supported natively on the target or (like SSE), the target doesn't - * have a clean/obvious dotproduct implementation. - */ -static void emit_transpose_matrix_transform_vec4( struct tnl_program *p, - struct ureg dest, - const struct ureg *mat, - struct ureg src) -{ - struct ureg tmp; - - if (dest.file != PROGRAM_TEMPORARY) - tmp = get_temp(p); - else - tmp = dest; - - emit_op2(p, OPCODE_MUL, tmp, 0, swizzle1(src,X), mat[0]); - emit_op3(p, OPCODE_MAD, tmp, 0, swizzle1(src,Y), mat[1], tmp); - emit_op3(p, OPCODE_MAD, tmp, 0, swizzle1(src,Z), mat[2], tmp); - emit_op3(p, OPCODE_MAD, dest, 0, swizzle1(src,W), mat[3], tmp); - - if (dest.file != PROGRAM_TEMPORARY) - release_temp(p, tmp); -} - -static void emit_matrix_transform_vec3( struct tnl_program *p, - struct ureg dest, - const struct ureg *mat, - struct ureg src) -{ - emit_op2(p, OPCODE_DP3, dest, WRITEMASK_X, src, mat[0]); - emit_op2(p, OPCODE_DP3, dest, WRITEMASK_Y, src, mat[1]); - emit_op2(p, OPCODE_DP3, dest, WRITEMASK_Z, src, mat[2]); -} - - -static void emit_normalize_vec3( struct tnl_program *p, - struct ureg dest, - struct ureg src ) -{ - emit_op2(p, OPCODE_DP3, dest, WRITEMASK_W, src, src); - emit_op1(p, OPCODE_RSQ, dest, WRITEMASK_W, swizzle1(dest,W)); - emit_op2(p, OPCODE_MUL, dest, WRITEMASK_XYZ, src, swizzle1(dest,W)); -} - -static void emit_passthrough( struct tnl_program *p, - GLuint input, - GLuint output ) -{ - struct ureg out = register_output(p, output); - emit_op1(p, OPCODE_MOV, out, 0, register_input(p, input)); -} - -static struct ureg get_eye_position( struct tnl_program *p ) -{ - if (is_undef(p->eye_position)) { - struct ureg pos = register_input( p, VERT_ATTRIB_POS ); - struct ureg modelview[4]; - - p->eye_position = reserve_temp(p); - - if (PREFER_DP4) { - register_matrix_param5( p, STATE_MODELVIEW_MATRIX, 0, 0, 3, - 0, modelview ); - - emit_matrix_transform_vec4(p, p->eye_position, modelview, pos); - } - else { - register_matrix_param5( p, STATE_MODELVIEW_MATRIX, 0, 0, 3, - STATE_MATRIX_TRANSPOSE, modelview ); - - emit_transpose_matrix_transform_vec4(p, p->eye_position, modelview, pos); - } - } - - return p->eye_position; -} - - -#if 0 -static struct ureg get_eye_z( struct tnl_program *p ) -{ - if (!is_undef(p->eye_position)) { - return swizzle1(p->eye_position, Z); - } - else if (!is_undef(p->eye_z)) { - struct ureg pos = register_input( p, BRW_ATTRIB_POS ); - struct ureg modelview2; - - p->eye_z = reserve_temp(p); - - register_matrix_param6( p, STATE_MATRIX, STATE_MODELVIEW, 0, 2, 1, - STATE_MATRIX, &modelview2 ); - - emit_matrix_transform_vec4(p, p->eye_position, modelview, pos); - emit_op2(p, OPCODE_DP4, p->eye_z, WRITEMASK_Z, pos, modelview2); - } - - return swizzle1(p->eye_z, Z) -} -#endif - - - -static struct ureg get_eye_position_normalized( struct tnl_program *p ) -{ - if (is_undef(p->eye_position_normalized)) { - struct ureg eye = get_eye_position(p); - p->eye_position_normalized = reserve_temp(p); - emit_normalize_vec3(p, p->eye_position_normalized, eye); - } - - return p->eye_position_normalized; -} - - -static struct ureg get_eye_normal( struct tnl_program *p ) -{ - if (is_undef(p->eye_normal)) { - struct ureg normal = register_input(p, VERT_ATTRIB_NORMAL ); - struct ureg mvinv[3]; - - register_matrix_param5( p, STATE_MODELVIEW_MATRIX, 0, 0, 2, - STATE_MATRIX_INVTRANS, mvinv ); - - p->eye_normal = reserve_temp(p); - - /* Transform to eye space: - */ - emit_matrix_transform_vec3( p, p->eye_normal, mvinv, normal ); - - /* Normalize/Rescale: - */ - if (p->state->normalize) { - emit_normalize_vec3( p, p->eye_normal, p->eye_normal ); - } - else if (p->state->rescale_normals) { - struct ureg rescale = register_param2(p, STATE_INTERNAL, - STATE_NORMAL_SCALE); - - emit_op2( p, OPCODE_MUL, p->eye_normal, 0, p->eye_normal, - swizzle1(rescale, X)); - } - } - - return p->eye_normal; -} - - - -static void build_hpos( struct tnl_program *p ) -{ - struct ureg pos = register_input( p, VERT_ATTRIB_POS ); - struct ureg hpos = register_output( p, VERT_RESULT_HPOS ); - struct ureg mvp[4]; - - if (PREFER_DP4) { - register_matrix_param5( p, STATE_MVP_MATRIX, 0, 0, 3, - 0, mvp ); - emit_matrix_transform_vec4( p, hpos, mvp, pos ); - } - else { - register_matrix_param5( p, STATE_MVP_MATRIX, 0, 0, 3, - STATE_MATRIX_TRANSPOSE, mvp ); - emit_transpose_matrix_transform_vec4( p, hpos, mvp, pos ); - } -} - - -static GLuint material_attrib( GLuint side, GLuint property ) -{ - return (property - STATE_AMBIENT) * 2 + side; -} - -/* Get a bitmask of which material values vary on a per-vertex basis. - */ -static void set_material_flags( struct tnl_program *p ) -{ - p->color_materials = 0; - p->materials = 0; - - if (p->state->light_color_material) { - p->materials = - p->color_materials = p->state->light_color_material_mask; - } - - p->materials |= p->state->light_material_mask; -} - - -static struct ureg get_material( struct tnl_program *p, GLuint side, - GLuint property ) -{ - GLuint attrib = material_attrib(side, property); - - if (p->color_materials & (1<<attrib)) - return register_input(p, VERT_ATTRIB_COLOR0); - else if (p->materials & (1<<attrib)) - return register_input( p, attrib + _TNL_ATTRIB_MAT_FRONT_AMBIENT ); - else - return register_param3( p, STATE_MATERIAL, side, property ); -} - -#define SCENE_COLOR_BITS(side) ((MAT_BIT_FRONT_EMISSION | \ - MAT_BIT_FRONT_AMBIENT | \ - MAT_BIT_FRONT_DIFFUSE) << (side)) - -/* Either return a precalculated constant value or emit code to - * calculate these values dynamically in the case where material calls - * are present between begin/end pairs. - * - * Probably want to shift this to the program compilation phase - if - * we always emitted the calculation here, a smart compiler could - * detect that it was constant (given a certain set of inputs), and - * lift it out of the main loop. That way the programs created here - * would be independent of the vertex_buffer details. - */ -static struct ureg get_scenecolor( struct tnl_program *p, GLuint side ) -{ - if (p->materials & SCENE_COLOR_BITS(side)) { - struct ureg lm_ambient = register_param1(p, STATE_LIGHTMODEL_AMBIENT); - struct ureg material_emission = get_material(p, side, STATE_EMISSION); - struct ureg material_ambient = get_material(p, side, STATE_AMBIENT); - struct ureg material_diffuse = get_material(p, side, STATE_DIFFUSE); - struct ureg tmp = make_temp(p, material_diffuse); - emit_op3(p, OPCODE_MAD, tmp, WRITEMASK_XYZ, lm_ambient, - material_ambient, material_emission); - return tmp; - } - else - return register_param2( p, STATE_LIGHTMODEL_SCENECOLOR, side ); -} - - -static struct ureg get_lightprod( struct tnl_program *p, GLuint light, - GLuint side, GLuint property ) -{ - GLuint attrib = material_attrib(side, property); - if (p->materials & (1<<attrib)) { - struct ureg light_value = - register_param3(p, STATE_LIGHT, light, property); - struct ureg material_value = get_material(p, side, property); - struct ureg tmp = get_temp(p); - emit_op2(p, OPCODE_MUL, tmp, 0, light_value, material_value); - return tmp; - } - else - return register_param4(p, STATE_LIGHTPROD, light, side, property); -} - -static struct ureg calculate_light_attenuation( struct tnl_program *p, - GLuint i, - struct ureg VPpli, - struct ureg dist ) -{ - struct ureg attenuation = register_param3(p, STATE_LIGHT, i, - STATE_ATTENUATION); - struct ureg att = get_temp(p); - - /* Calculate spot attenuation: - */ - if (!p->state->unit[i].light_spotcutoff_is_180) { - struct ureg spot_dir_norm = register_param3(p, STATE_INTERNAL, - STATE_LIGHT_SPOT_DIR_NORMALIZED, i); - struct ureg spot = get_temp(p); - struct ureg slt = get_temp(p); - - emit_op2(p, OPCODE_DP3, spot, 0, ureg_negate(VPpli), spot_dir_norm); - emit_op2(p, OPCODE_SLT, slt, 0, swizzle1(spot_dir_norm,W), spot); - emit_op2(p, OPCODE_POW, spot, 0, spot, swizzle1(attenuation, W)); - emit_op2(p, OPCODE_MUL, att, 0, slt, spot); - - release_temp(p, spot); - release_temp(p, slt); - } - - /* Calculate distance attenuation: - */ - if (p->state->unit[i].light_attenuated) { - - /* 1/d,d,d,1/d */ - emit_op1(p, OPCODE_RCP, dist, WRITEMASK_YZ, dist); - /* 1,d,d*d,1/d */ - emit_op2(p, OPCODE_MUL, dist, WRITEMASK_XZ, dist, swizzle1(dist,Y)); - /* 1/dist-atten */ - emit_op2(p, OPCODE_DP3, dist, 0, attenuation, dist); - - if (!p->state->unit[i].light_spotcutoff_is_180) { - /* dist-atten */ - emit_op1(p, OPCODE_RCP, dist, 0, dist); - /* spot-atten * dist-atten */ - emit_op2(p, OPCODE_MUL, att, 0, dist, att); - } else { - /* dist-atten */ - emit_op1(p, OPCODE_RCP, att, 0, dist); - } - } - - return att; -} - - - - - -/* Need to add some addtional parameters to allow lighting in object - * space - STATE_SPOT_DIRECTION and STATE_HALF_VECTOR implicitly assume eye - * space lighting. - */ -static void build_lighting( struct tnl_program *p ) -{ - const GLboolean twoside = p->state->light_twoside; - const GLboolean separate = p->state->separate_specular; - GLuint nr_lights = 0, count = 0; - struct ureg normal = get_eye_normal(p); - struct ureg lit = get_temp(p); - struct ureg dots = get_temp(p); - struct ureg _col0 = undef, _col1 = undef; - struct ureg _bfc0 = undef, _bfc1 = undef; - GLuint i; - - for (i = 0; i < MAX_LIGHTS; i++) - if (p->state->unit[i].light_enabled) - nr_lights++; - - set_material_flags(p); - - { - struct ureg shininess = get_material(p, 0, STATE_SHININESS); - emit_op1(p, OPCODE_MOV, dots, WRITEMASK_W, swizzle1(shininess,X)); - release_temp(p, shininess); - - _col0 = make_temp(p, get_scenecolor(p, 0)); - if (separate) - _col1 = make_temp(p, get_identity_param(p)); - else - _col1 = _col0; - - } - - if (twoside) { - struct ureg shininess = get_material(p, 1, STATE_SHININESS); - emit_op1(p, OPCODE_MOV, dots, WRITEMASK_Z, - ureg_negate(swizzle1(shininess,X))); - release_temp(p, shininess); - - _bfc0 = make_temp(p, get_scenecolor(p, 1)); - if (separate) - _bfc1 = make_temp(p, get_identity_param(p)); - else - _bfc1 = _bfc0; - } - - - /* If no lights, still need to emit the scenecolor. - */ - /* KW: changed to do this always - v1.17 "Fix lighting alpha result"? - */ - if (p->state->fragprog_inputs_read & FRAG_BIT_COL0) - { - struct ureg res0 = register_output( p, VERT_RESULT_COL0 ); - emit_op1(p, OPCODE_MOV, res0, 0, _col0); - - if (twoside) { - struct ureg res0 = register_output( p, VERT_RESULT_BFC0 ); - emit_op1(p, OPCODE_MOV, res0, 0, _bfc0); - } - } - - if (separate && (p->state->fragprog_inputs_read & FRAG_BIT_COL1)) { - - struct ureg res1 = register_output( p, VERT_RESULT_COL1 ); - emit_op1(p, OPCODE_MOV, res1, 0, _col1); - - if (twoside) { - struct ureg res1 = register_output( p, VERT_RESULT_BFC1 ); - emit_op1(p, OPCODE_MOV, res1, 0, _bfc1); - } - } - - if (nr_lights == 0) { - release_temps(p); - return; - } - - - for (i = 0; i < MAX_LIGHTS; i++) { - if (p->state->unit[i].light_enabled) { - struct ureg half = undef; - struct ureg att = undef, VPpli = undef; - - count++; - - if (p->state->unit[i].light_eyepos3_is_zero) { - /* Can used precomputed constants in this case. - * Attenuation never applies to infinite lights. - */ - VPpli = register_param3(p, STATE_LIGHT, i, - STATE_LIGHT_POSITION_NORMALIZED); - if (p->state->light_local_viewer) { - struct ureg eye_hat = get_eye_position_normalized(p); - half = get_temp(p); - emit_op2(p, OPCODE_SUB, half, 0, VPpli, eye_hat); - emit_normalize_vec3(p, half, half); - } else { - half = register_param3(p, STATE_LIGHT, i, STATE_HALF_VECTOR); - } - } - else { - struct ureg Ppli = register_param3(p, STATE_LIGHT, i, - STATE_POSITION); - struct ureg V = get_eye_position(p); - struct ureg dist = get_temp(p); - struct ureg tmpPpli = get_temp(p); - - VPpli = get_temp(p); - half = get_temp(p); - - /* In homogeneous object coordinates - */ - emit_op1(p, OPCODE_RCP, dist, 0, swizzle1(Ppli, W)); - emit_op2(p, OPCODE_MUL, tmpPpli, 0, Ppli, dist); - - /* Calulate VPpli vector - */ - emit_op2(p, OPCODE_SUB, VPpli, 0, tmpPpli, V); - - /* Normalize VPpli. The dist value also used in - * attenuation below. - */ - emit_op2(p, OPCODE_DP3, dist, 0, VPpli, VPpli); - emit_op1(p, OPCODE_RSQ, dist, 0, dist); - emit_op2(p, OPCODE_MUL, VPpli, 0, VPpli, dist); - - - /* Calculate attenuation: - */ - if (!p->state->unit[i].light_spotcutoff_is_180 || - p->state->unit[i].light_attenuated) { - att = calculate_light_attenuation(p, i, VPpli, dist); - } - - - /* Calculate viewer direction, or use infinite viewer: - */ - if (p->state->light_local_viewer) { - struct ureg eye_hat = get_eye_position_normalized(p); - emit_op2(p, OPCODE_SUB, half, 0, VPpli, eye_hat); - } - else { - struct ureg z_dir = swizzle(get_identity_param(p),X,Y,W,Z); - emit_op2(p, OPCODE_ADD, half, 0, VPpli, z_dir); - } - - emit_normalize_vec3(p, half, half); - - release_temp(p, dist); - release_temp(p, tmpPpli); - } - - /* Calculate dot products: - */ - emit_op2(p, OPCODE_DP3, dots, WRITEMASK_X, normal, VPpli); - emit_op2(p, OPCODE_DP3, dots, WRITEMASK_Y, normal, half); - - - /* Front face lighting: - */ - { - struct ureg ambient = get_lightprod(p, i, 0, STATE_AMBIENT); - struct ureg diffuse = get_lightprod(p, i, 0, STATE_DIFFUSE); - struct ureg specular = get_lightprod(p, i, 0, STATE_SPECULAR); - struct ureg res0, res1; - GLuint mask0, mask1; - - emit_op1(p, OPCODE_LIT, lit, 0, dots); - - if (!is_undef(att)) - emit_op2(p, OPCODE_MUL, lit, 0, lit, att); - - - mask0 = 0; - mask1 = 0; - res0 = _col0; - res1 = _col1; - - if (count == nr_lights) { - if (separate) { - mask0 = WRITEMASK_XYZ; - mask1 = WRITEMASK_XYZ; - - if (p->state->fragprog_inputs_read & FRAG_BIT_COL0) - res0 = register_output( p, VERT_RESULT_COL0 ); - - if (p->state->fragprog_inputs_read & FRAG_BIT_COL1) - res1 = register_output( p, VERT_RESULT_COL1 ); - } - else { - mask1 = WRITEMASK_XYZ; - - if (p->state->fragprog_inputs_read & FRAG_BIT_COL0) - res1 = register_output( p, VERT_RESULT_COL0 ); - } - } - - emit_op3(p, OPCODE_MAD, _col0, 0, swizzle1(lit,X), ambient, _col0); - emit_op3(p, OPCODE_MAD, res0, mask0, swizzle1(lit,Y), diffuse, _col0); - emit_op3(p, OPCODE_MAD, res1, mask1, swizzle1(lit,Z), specular, _col1); - - release_temp(p, ambient); - release_temp(p, diffuse); - release_temp(p, specular); - } - - /* Back face lighting: - */ - if (twoside) { - struct ureg ambient = get_lightprod(p, i, 1, STATE_AMBIENT); - struct ureg diffuse = get_lightprod(p, i, 1, STATE_DIFFUSE); - struct ureg specular = get_lightprod(p, i, 1, STATE_SPECULAR); - struct ureg res0, res1; - GLuint mask0, mask1; - - emit_op1(p, OPCODE_LIT, lit, 0, ureg_negate(swizzle(dots,X,Y,W,Z))); - - if (!is_undef(att)) - emit_op2(p, OPCODE_MUL, lit, 0, lit, att); - - mask0 = 0; - mask1 = 0; - res0 = _bfc0; - res1 = _bfc1; - - if (count == nr_lights) { - if (separate) { - mask0 = WRITEMASK_XYZ; - mask1 = WRITEMASK_XYZ; - if (p->state->fragprog_inputs_read & FRAG_BIT_COL0) - res0 = register_output( p, VERT_RESULT_BFC0 ); - - if (p->state->fragprog_inputs_read & FRAG_BIT_COL1) - res1 = register_output( p, VERT_RESULT_BFC1 ); - } - else { - mask1 = WRITEMASK_XYZ; - - if (p->state->fragprog_inputs_read & FRAG_BIT_COL0) - res1 = register_output( p, VERT_RESULT_BFC0 ); - } - } - - emit_op3(p, OPCODE_MAD, _bfc0, 0, swizzle1(lit,X), ambient, _bfc0); - emit_op3(p, OPCODE_MAD, res0, mask0, swizzle1(lit,Y), diffuse, _bfc0); - emit_op3(p, OPCODE_MAD, res1, mask1, swizzle1(lit,Z), specular, _bfc1); - - release_temp(p, ambient); - release_temp(p, diffuse); - release_temp(p, specular); - } - - release_temp(p, half); - release_temp(p, VPpli); - release_temp(p, att); - } - } - - release_temps( p ); -} - - -static void build_fog( struct tnl_program *p ) -{ - struct ureg fog = register_output(p, VERT_RESULT_FOGC); - struct ureg input; - GLuint useabs = p->state->fog_source_is_depth && p->state->fog_option && - (p->state->fog_option != FOG_EXP2); - - if (p->state->fog_source_is_depth) { - input = swizzle1(get_eye_position(p), Z); - } - else { - input = swizzle1(register_input(p, VERT_ATTRIB_FOG), X); - if (p->state->fog_option && - p->state->tnl_do_vertex_fog) - input = swizzle1(register_input(p, VERT_ATTRIB_FOG), X); - else - input = register_input(p, VERT_ATTRIB_FOG); - } - - if (p->state->fog_option && - p->state->tnl_do_vertex_fog) { - struct ureg params = register_param2(p, STATE_INTERNAL, - STATE_FOG_PARAMS_OPTIMIZED); - struct ureg tmp = get_temp(p); - struct ureg id = get_identity_param(p); - - emit_op1(p, OPCODE_MOV, fog, 0, id); - - if (useabs) { - emit_op1(p, OPCODE_ABS, tmp, 0, input); - } - - switch (p->state->fog_option) { - case FOG_LINEAR: { - emit_op3(p, OPCODE_MAD, tmp, 0, useabs ? tmp : input, - swizzle1(params,X), swizzle1(params,Y)); - emit_op2(p, OPCODE_MAX, tmp, 0, tmp, swizzle1(id,X)); /* saturate */ - emit_op2(p, OPCODE_MIN, fog, WRITEMASK_X, tmp, swizzle1(id,W)); - break; - } - case FOG_EXP: - emit_op2(p, OPCODE_MUL, tmp, 0, useabs ? tmp : input, - swizzle1(params,Z)); - emit_op1(p, OPCODE_EX2, fog, WRITEMASK_X, ureg_negate(tmp)); - break; - case FOG_EXP2: - emit_op2(p, OPCODE_MUL, tmp, 0, input, swizzle1(params,W)); - emit_op2(p, OPCODE_MUL, tmp, 0, tmp, tmp); - emit_op1(p, OPCODE_EX2, fog, WRITEMASK_X, ureg_negate(tmp)); - break; - } - - release_temp(p, tmp); - } - else { - /* results = incoming fog coords (compute fog per-fragment later) - * - * KW: Is it really necessary to do anything in this case? - */ - emit_op1(p, useabs ? OPCODE_ABS : OPCODE_MOV, fog, 0, input); - } -} - -static void build_reflect_texgen( struct tnl_program *p, - struct ureg dest, - GLuint writemask ) -{ - struct ureg normal = get_eye_normal(p); - struct ureg eye_hat = get_eye_position_normalized(p); - struct ureg tmp = get_temp(p); - - /* n.u */ - emit_op2(p, OPCODE_DP3, tmp, 0, normal, eye_hat); - /* 2n.u */ - emit_op2(p, OPCODE_ADD, tmp, 0, tmp, tmp); - /* (-2n.u)n + u */ - emit_op3(p, OPCODE_MAD, dest, writemask, ureg_negate(tmp), normal, eye_hat); - - release_temp(p, tmp); -} - -static void build_sphere_texgen( struct tnl_program *p, - struct ureg dest, - GLuint writemask ) -{ - struct ureg normal = get_eye_normal(p); - struct ureg eye_hat = get_eye_position_normalized(p); - struct ureg tmp = get_temp(p); - struct ureg half = register_scalar_const(p, .5); - struct ureg r = get_temp(p); - struct ureg inv_m = get_temp(p); - struct ureg id = get_identity_param(p); - - /* Could share the above calculations, but it would be - * a fairly odd state for someone to set (both sphere and - * reflection active for different texture coordinate - * components. Of course - if two texture units enable - * reflect and/or sphere, things start to tilt in favour - * of seperating this out: - */ - - /* n.u */ - emit_op2(p, OPCODE_DP3, tmp, 0, normal, eye_hat); - /* 2n.u */ - emit_op2(p, OPCODE_ADD, tmp, 0, tmp, tmp); - /* (-2n.u)n + u */ - emit_op3(p, OPCODE_MAD, r, 0, ureg_negate(tmp), normal, eye_hat); - /* r + 0,0,1 */ - emit_op2(p, OPCODE_ADD, tmp, 0, r, swizzle(id,X,Y,W,Z)); - /* rx^2 + ry^2 + (rz+1)^2 */ - emit_op2(p, OPCODE_DP3, tmp, 0, tmp, tmp); - /* 2/m */ - emit_op1(p, OPCODE_RSQ, tmp, 0, tmp); - /* 1/m */ - emit_op2(p, OPCODE_MUL, inv_m, 0, tmp, half); - /* r/m + 1/2 */ - emit_op3(p, OPCODE_MAD, dest, writemask, r, inv_m, half); - - release_temp(p, tmp); - release_temp(p, r); - release_temp(p, inv_m); -} - - -static void build_texture_transform( struct tnl_program *p ) -{ - GLuint i, j; - - for (i = 0; i < MAX_TEXTURE_UNITS; i++) { - - if (!(p->state->fragprog_inputs_read & (FRAG_BIT_TEX0<<i))) - continue; - - if (p->state->unit[i].texgen_enabled || - p->state->unit[i].texmat_enabled) { - - GLuint texmat_enabled = p->state->unit[i].texmat_enabled; - struct ureg out = register_output(p, VERT_RESULT_TEX0 + i); - struct ureg out_texgen = undef; - - if (p->state->unit[i].texgen_enabled) { - GLuint copy_mask = 0; - GLuint sphere_mask = 0; - GLuint reflect_mask = 0; - GLuint normal_mask = 0; - GLuint modes[4]; - - if (texmat_enabled) - out_texgen = get_temp(p); - else - out_texgen = out; - - modes[0] = p->state->unit[i].texgen_mode0; - modes[1] = p->state->unit[i].texgen_mode1; - modes[2] = p->state->unit[i].texgen_mode2; - modes[3] = p->state->unit[i].texgen_mode3; - - for (j = 0; j < 4; j++) { - switch (modes[j]) { - case TXG_OBJ_LINEAR: { - struct ureg obj = register_input(p, VERT_ATTRIB_POS); - struct ureg plane = - register_param3(p, STATE_TEXGEN, i, - STATE_TEXGEN_OBJECT_S + j); - - emit_op2(p, OPCODE_DP4, out_texgen, WRITEMASK_X << j, - obj, plane ); - break; - } - case TXG_EYE_LINEAR: { - struct ureg eye = get_eye_position(p); - struct ureg plane = - register_param3(p, STATE_TEXGEN, i, - STATE_TEXGEN_EYE_S + j); - - emit_op2(p, OPCODE_DP4, out_texgen, WRITEMASK_X << j, - eye, plane ); - break; - } - case TXG_SPHERE_MAP: - sphere_mask |= WRITEMASK_X << j; - break; - case TXG_REFLECTION_MAP: - reflect_mask |= WRITEMASK_X << j; - break; - case TXG_NORMAL_MAP: - normal_mask |= WRITEMASK_X << j; - break; - case TXG_NONE: - copy_mask |= WRITEMASK_X << j; - } - - } - - - if (sphere_mask) { - build_sphere_texgen(p, out_texgen, sphere_mask); - } - - if (reflect_mask) { - build_reflect_texgen(p, out_texgen, reflect_mask); - } - - if (normal_mask) { - struct ureg normal = get_eye_normal(p); - emit_op1(p, OPCODE_MOV, out_texgen, normal_mask, normal ); - } - - if (copy_mask) { - struct ureg in = register_input(p, VERT_ATTRIB_TEX0+i); - emit_op1(p, OPCODE_MOV, out_texgen, copy_mask, in ); - } - } - - if (texmat_enabled) { - struct ureg texmat[4]; - struct ureg in = (!is_undef(out_texgen) ? - out_texgen : - register_input(p, VERT_ATTRIB_TEX0+i)); - if (PREFER_DP4) { - register_matrix_param5( p, STATE_TEXTURE_MATRIX, i, 0, 3, - 0, texmat ); - emit_matrix_transform_vec4( p, out, texmat, in ); - } - else { - register_matrix_param5( p, STATE_TEXTURE_MATRIX, i, 0, 3, - STATE_MATRIX_TRANSPOSE, texmat ); - emit_transpose_matrix_transform_vec4( p, out, texmat, in ); - } - } - - release_temps(p); - } - else { - emit_passthrough(p, VERT_ATTRIB_TEX0+i, VERT_RESULT_TEX0+i); - } - } -} - - -/* Seems like it could be tighter: - */ -static void build_pointsize( struct tnl_program *p ) -{ - struct ureg eye = get_eye_position(p); - struct ureg state_size = register_param1(p, STATE_POINT_SIZE); - struct ureg state_attenuation = register_param1(p, STATE_POINT_ATTENUATION); - struct ureg out = register_output(p, VERT_RESULT_PSIZ); - struct ureg ut = get_temp(p); - - /* 1, Z, Z * Z, 1 */ - emit_op1(p, OPCODE_MOV, ut, WRITEMASK_XW, swizzle1(get_identity_param(p), W)); - emit_op1(p, OPCODE_ABS, ut, WRITEMASK_YZ, swizzle1(eye, Z)); - emit_op2(p, OPCODE_MUL, ut, WRITEMASK_Z, ut, ut); - - - /* p1 + p2 * dist + p3 * dist * dist, 0 */ - emit_op2(p, OPCODE_DP3, ut, WRITEMASK_X, ut, state_attenuation); - - /* 1 / sqrt(factor) */ - emit_op1(p, OPCODE_RSQ, ut, WRITEMASK_X, ut ); - - /* ut = pointSize / factor */ - emit_op2(p, OPCODE_MUL, ut, WRITEMASK_X, ut, state_size); - - /* Clamp to min/max - state_size.[yz] - */ - emit_op2(p, OPCODE_MAX, ut, WRITEMASK_X, ut, swizzle1(state_size, Y)); - emit_op2(p, OPCODE_MIN, out, 0, swizzle1(ut, X), swizzle1(state_size, Z)); - - release_temp(p, ut); -} - -static void build_tnl_program( struct tnl_program *p ) -{ - /* Emit the program, starting with modelviewproject: - */ - build_hpos(p); - - /* Lighting calculations: - */ - if (p->state->fragprog_inputs_read & (FRAG_BIT_COL0|FRAG_BIT_COL1)) { - if (p->state->light_global_enabled) - build_lighting(p); - else { - if (p->state->fragprog_inputs_read & FRAG_BIT_COL0) - emit_passthrough(p, VERT_ATTRIB_COLOR0, VERT_RESULT_COL0); - - if (p->state->fragprog_inputs_read & FRAG_BIT_COL1) - emit_passthrough(p, VERT_ATTRIB_COLOR1, VERT_RESULT_COL1); - } - } - - if ((p->state->fragprog_inputs_read & FRAG_BIT_FOGC) || - p->state->fog_option != FOG_NONE) - build_fog(p); - - if (p->state->fragprog_inputs_read & FRAG_BITS_TEX_ANY) - build_texture_transform(p); - - if (p->state->point_attenuated) - build_pointsize(p); - - /* Finish up: - */ - emit_op1(p, OPCODE_END, undef, 0, undef); - - /* Disassemble: - */ - if (DISASSEM) { - _mesa_printf ("\n"); - } -} - - -static void build_new_tnl_program( const struct state_key *key, - struct gl_vertex_program *program, - GLuint max_temps) -{ - struct tnl_program p; - - _mesa_memset(&p, 0, sizeof(p)); - p.state = key; - p.program = program; - p.eye_position = undef; - p.eye_position_normalized = undef; - p.eye_normal = undef; - p.identity = undef; - p.temp_in_use = 0; - p.nr_instructions = 16; - - if (max_temps >= sizeof(int) * 8) - p.temp_reserved = 0; - else - p.temp_reserved = ~((1<<max_temps)-1); - - p.program->Base.Instructions = - _mesa_malloc(sizeof(struct prog_instruction) * p.nr_instructions); - p.program->Base.String = 0; - p.program->Base.NumInstructions = - p.program->Base.NumTemporaries = - p.program->Base.NumParameters = - p.program->Base.NumAttributes = p.program->Base.NumAddressRegs = 0; - p.program->Base.Parameters = _mesa_new_parameter_list(); - p.program->Base.InputsRead = 0; - p.program->Base.OutputsWritten = 0; - - build_tnl_program( &p ); -} - -static void *search_cache( struct brw_tnl_cache *cache, - GLuint hash, - const void *key, - GLuint keysize) -{ - struct brw_tnl_cache_item *c; - - for (c = cache->items[hash % cache->size]; c; c = c->next) { - if (c->hash == hash && memcmp(c->key, key, keysize) == 0) - return c->data; - } - - return NULL; -} - -static void rehash( struct brw_tnl_cache *cache ) -{ - struct brw_tnl_cache_item **items; - struct brw_tnl_cache_item *c, *next; - GLuint size, i; - - size = cache->size * 3; - items = (struct brw_tnl_cache_item**) _mesa_malloc(size * sizeof(*items)); - _mesa_memset(items, 0, size * sizeof(*items)); - - for (i = 0; i < cache->size; i++) - for (c = cache->items[i]; c; c = next) { - next = c->next; - c->next = items[c->hash % size]; - items[c->hash % size] = c; - } - - FREE(cache->items); - cache->items = items; - cache->size = size; -} - -static void cache_item( struct brw_tnl_cache *cache, - GLuint hash, - const struct state_key *key, - void *data ) -{ - struct brw_tnl_cache_item *c = MALLOC(sizeof(*c)); - c->hash = hash; - - c->key = malloc(sizeof(*key)); - memcpy(c->key, key, sizeof(*key)); - - c->data = data; - - if (++cache->n_items > cache->size * 1.5) - rehash(cache); - - c->next = cache->items[hash % cache->size]; - cache->items[hash % cache->size] = c; -} - - -static GLuint hash_key( struct state_key *key ) -{ - GLuint *ikey = (GLuint *)key; - GLuint hash = 0, i; - - /* I'm sure this can be improved on, but speed is important: - */ - for (i = 0; i < sizeof(*key)/sizeof(GLuint); i++) - hash += ikey[i]; - - return hash; -} - -static void prepare_tnl_program( struct brw_context *brw ) -{ - GLcontext *ctx = &brw->intel.ctx; - struct state_key key; - GLuint hash; - struct gl_vertex_program *old = brw->tnl_program; - - /* _NEW_PROGRAM */ - if (brw->attribs.VertexProgram->_Current) - return; - - /* Grab all the relevent state and put it in a single structure: - */ - make_state_key(ctx, &key); - hash = hash_key(&key); - - /* Look for an already-prepared program for this state: - */ - brw->tnl_program = (struct gl_vertex_program *) - search_cache( &brw->tnl_program_cache, hash, &key, sizeof(key) ); - - /* OK, we'll have to build a new one: - */ - if (!brw->tnl_program) { - brw->tnl_program = (struct gl_vertex_program *) - ctx->Driver.NewProgram(ctx, GL_VERTEX_PROGRAM_ARB, 0); - - build_new_tnl_program( &key, brw->tnl_program, -/* ctx->Const.MaxVertexProgramTemps */ - 32 - ); - - if (ctx->Driver.ProgramStringNotify) - ctx->Driver.ProgramStringNotify( ctx, GL_VERTEX_PROGRAM_ARB, - &brw->tnl_program->Base ); - - cache_item( &brw->tnl_program_cache, - hash, &key, brw->tnl_program ); - } - - if (old != brw->tnl_program) - brw->state.dirty.brw |= BRW_NEW_TNL_PROGRAM; - return; -} - -/* Note: See brw_draw.c - the vertex program must not rely on - * brw->primitive or brw->reduced_prim. - */ -const struct brw_tracked_state brw_tnl_vertprog = { - .dirty = { - .mesa = (_NEW_PROGRAM | - _NEW_LIGHT | - _NEW_TRANSFORM | - _NEW_FOG | - _NEW_HINT | - _NEW_POINT | - _NEW_TEXTURE | - _NEW_TEXTURE_MATRIX), - .brw = (BRW_NEW_FRAGMENT_PROGRAM | - BRW_NEW_INPUT_VARYING), - .cache = 0 - }, - .prepare = prepare_tnl_program -}; - - - - static void prepare_active_vertprog( struct brw_context *brw ) { const struct gl_vertex_program *prev = brw->vertex_program; - /* NEW_PROGRAM */ - if (brw->attribs.VertexProgram->_Current) { - brw->vertex_program = brw->attribs.VertexProgram->_Current; - } - else { - /* BRW_NEW_TNL_PROGRAM */ - brw->vertex_program = brw->tnl_program; - } + brw->vertex_program = brw->attribs.VertexProgram->_Current; if (brw->vertex_program != prev) brw->state.dirty.brw |= BRW_NEW_VERTEX_PROGRAM; @@ -1672,37 +52,8 @@ static void prepare_active_vertprog( struct brw_context *brw ) const struct brw_tracked_state brw_active_vertprog = { .dirty = { .mesa = _NEW_PROGRAM, - .brw = BRW_NEW_TNL_PROGRAM, + .brw = 0, .cache = 0 }, .prepare = prepare_active_vertprog }; - - -void brw_ProgramCacheInit( GLcontext *ctx ) -{ - struct brw_context *brw = brw_context(ctx); - - brw->tnl_program_cache.size = 17; - brw->tnl_program_cache.n_items = 0; - brw->tnl_program_cache.items = (struct brw_tnl_cache_item **) - _mesa_calloc(brw->tnl_program_cache.size * - sizeof(struct brw_tnl_cache_item)); -} - -void brw_ProgramCacheDestroy( GLcontext *ctx ) -{ - struct brw_context *brw = brw_context(ctx); - struct brw_tnl_cache_item *c, *next; - GLuint i; - - for (i = 0; i < brw->tnl_program_cache.size; i++) - for (c = brw->tnl_program_cache.items[i]; c; c = next) { - next = c->next; - FREE(c->key); - FREE(c->data); - FREE(c); - } - - FREE(brw->tnl_program_cache.items); -} diff --git a/src/mesa/drivers/dri/i965/brw_vtbl.c b/src/mesa/drivers/dri/i965/brw_vtbl.c index 2a03fc59f3e..a64e437860f 100644 --- a/src/mesa/drivers/dri/i965/brw_vtbl.c +++ b/src/mesa/drivers/dri/i965/brw_vtbl.c @@ -51,6 +51,12 @@ #include "brw_vs.h" #include <stdarg.h> +static void +dri_bo_release(dri_bo **bo) +{ + dri_bo_unreference(*bo); + *bo = NULL; +} /* called from intelDestroyContext() */ @@ -58,13 +64,40 @@ static void brw_destroy_context( struct intel_context *intel ) { GLcontext *ctx = &intel->ctx; struct brw_context *brw = brw_context(&intel->ctx); + int i; brw_destroy_metaops(brw); brw_destroy_state(brw); brw_draw_destroy( brw ); - brw_ProgramCacheDestroy( ctx ); brw_FrameBufferTexDestroy( brw ); + + for (i = 0; i < brw->state.nr_draw_regions; i++) + intel_region_release(&brw->state.draw_regions[i]); + brw->state.nr_draw_regions = 0; + intel_region_release(&brw->state.depth_region); + + dri_bo_release(&brw->curbe.curbe_bo); + dri_bo_release(&brw->vs.prog_bo); + dri_bo_release(&brw->vs.state_bo); + dri_bo_release(&brw->gs.prog_bo); + dri_bo_release(&brw->gs.state_bo); + dri_bo_release(&brw->clip.prog_bo); + dri_bo_release(&brw->clip.state_bo); + dri_bo_release(&brw->clip.vp_bo); + dri_bo_release(&brw->sf.prog_bo); + dri_bo_release(&brw->sf.state_bo); + dri_bo_release(&brw->sf.vp_bo); + for (i = 0; i < BRW_MAX_TEX_UNIT; i++) + dri_bo_release(&brw->wm.sdc_bo[i]); + dri_bo_release(&brw->wm.bind_bo); + for (i = 0; i < BRW_WM_MAX_SURF; i++) + dri_bo_release(&brw->wm.surf_bo[i]); + dri_bo_release(&brw->wm.prog_bo); + dri_bo_release(&brw->wm.state_bo); + dri_bo_release(&brw->cc.prog_bo); + dri_bo_release(&brw->cc.state_bo); + dri_bo_release(&brw->cc.vp_bo); } /* called from intelDrawBuffer() @@ -87,6 +120,15 @@ static void brw_set_draw_region( struct intel_context *intel, brw->state.nr_draw_regions = num_regions; } +/* called from intel_batchbuffer_flush and children before sending a + * batchbuffer off. + */ +static void brw_finish_batch(struct intel_context *intel) +{ + struct brw_context *brw = brw_context(&intel->ctx); + + brw_emit_query_end(brw); +} /* called from intelFlushBatchLocked */ @@ -185,6 +227,7 @@ void brwInitVtbl( struct brw_context *brw ) brw->intel.vtbl.note_fence = brw_note_fence; brw->intel.vtbl.note_unlock = brw_note_unlock; brw->intel.vtbl.new_batch = brw_new_batch; + brw->intel.vtbl.finish_batch = brw_finish_batch; brw->intel.vtbl.destroy = brw_destroy_context; brw->intel.vtbl.set_draw_region = brw_set_draw_region; brw->intel.vtbl.flush_cmd = brw_flush_cmd; diff --git a/src/mesa/drivers/dri/i965/brw_wm_sampler_state.c b/src/mesa/drivers/dri/i965/brw_wm_sampler_state.c index e1db31ec08c..f12ef47a7d7 100644 --- a/src/mesa/drivers/dri/i965/brw_wm_sampler_state.c +++ b/src/mesa/drivers/dri/i965/brw_wm_sampler_state.c @@ -229,6 +229,9 @@ brw_wm_sampler_populate_key(struct brw_context *brw, struct wm_sampler_entry *entry = &key->sampler[unit]; struct gl_texture_unit *texUnit = &brw->attribs.Texture->Unit[unit]; struct gl_texture_object *texObj = texUnit->_Current; + struct intel_texture_object *intelObj = intel_texture_object(texObj); + struct gl_texture_image *firstImage = + texObj->Image[0][intelObj->firstLevel]; entry->wrap_r = texObj->WrapR; entry->wrap_s = texObj->WrapS; @@ -244,8 +247,22 @@ brw_wm_sampler_populate_key(struct brw_context *brw, entry->comparefunc = texObj->CompareFunc; dri_bo_unreference(brw->wm.sdc_bo[unit]); - brw->wm.sdc_bo[unit] = upload_default_color(brw, texObj->BorderColor); - + if (firstImage->_BaseFormat == GL_DEPTH_COMPONENT) { + float bordercolor[4] = { + texObj->BorderColor[0], + texObj->BorderColor[0], + texObj->BorderColor[0], + texObj->BorderColor[0] + }; + /* GL specs that border color for depth textures is taken from the + * R channel, while the hardware uses A. Spam R into all the + * channels for safety. + */ + brw->wm.sdc_bo[unit] = upload_default_color(brw, bordercolor); + } else { + brw->wm.sdc_bo[unit] = upload_default_color(brw, + texObj->BorderColor); + } key->sampler_count = unit + 1; } } @@ -304,7 +321,7 @@ static void upload_wm_samplers( struct brw_context *brw ) continue; dri_bo_emit_reloc(brw->wm.sampler_bo, - I915_GEM_DOMAIN_INSTRUCTION, 0, + I915_GEM_DOMAIN_SAMPLER, 0, 0, i * sizeof(struct brw_sampler_state) + offsetof(struct brw_sampler_state, ss2), diff --git a/src/mesa/drivers/dri/i965/brw_wm_surface_state.c b/src/mesa/drivers/dri/i965/brw_wm_surface_state.c index 47127c04210..3790b50c976 100644 --- a/src/mesa/drivers/dri/i965/brw_wm_surface_state.c +++ b/src/mesa/drivers/dri/i965/brw_wm_surface_state.c @@ -274,7 +274,6 @@ brw_update_texture_surface( GLcontext *ctx, GLuint unit ) key.width = firstImage->Width; key.height = firstImage->Height; key.cpp = intelObj->mt->cpp; - key.depth = firstImage->Depth; key.tiling = intelObj->mt->region->tiling; dri_bo_unreference(brw->wm.surf_bo[unit + MAX_DRAW_BUFFERS]); diff --git a/src/mesa/drivers/dri/i965/intel_pixel_draw.c b/src/mesa/drivers/dri/i965/intel_pixel_draw.c new file mode 120000 index 00000000000..8431a24edfc --- /dev/null +++ b/src/mesa/drivers/dri/i965/intel_pixel_draw.c @@ -0,0 +1 @@ +../intel/intel_pixel_draw.c
\ No newline at end of file diff --git a/src/mesa/drivers/dri/intel/intel_buffers.c b/src/mesa/drivers/dri/intel/intel_buffers.c index defa5b173d9..f5eaf765f38 100644 --- a/src/mesa/drivers/dri/intel/intel_buffers.c +++ b/src/mesa/drivers/dri/intel/intel_buffers.c @@ -135,6 +135,14 @@ intel_readbuf_region(struct intel_context *intel) static void intelSetRenderbufferClipRects(struct intel_context *intel) { + /* If the batch contents require looping over cliprects, flush them before + * we go changing which cliprects get referenced when that happens. + */ + if (intel->batch->cliprect_mode == LOOP_CLIPRECTS && + (intel->fboRect.x2 != intel->ctx.DrawBuffer->Width || + intel->fboRect.x2 != intel->ctx.DrawBuffer->Height)) + intel_batchbuffer_flush(intel->batch); + assert(intel->ctx.DrawBuffer->Width > 0); assert(intel->ctx.DrawBuffer->Height > 0); intel->fboRect.x1 = 0; @@ -160,6 +168,12 @@ intelSetFrontClipRects(struct intel_context *intel) if (!dPriv) return; + /* If the batch contents require looping over cliprects, flush them before + * we go changing which cliprects get referenced when that happens. + */ + if (intel->batch->cliprect_mode == LOOP_CLIPRECTS && + intel->pClipRects != dPriv->pClipRects) + intel_batchbuffer_flush(intel->batch); intel->numClipRects = dPriv->numClipRects; intel->pClipRects = dPriv->pClipRects; intel->drawX = dPriv->x; @@ -183,6 +197,10 @@ intelSetBackClipRects(struct intel_context *intel) if (intel_fb->pf_active || dPriv->numBackClipRects == 0) { /* use the front clip rects */ + if (intel->batch->cliprect_mode == LOOP_CLIPRECTS && + intel->pClipRects != dPriv->pClipRects) + intel_batchbuffer_flush(intel->batch); + intel->numClipRects = dPriv->numClipRects; intel->pClipRects = dPriv->pClipRects; intel->drawX = dPriv->x; @@ -190,6 +208,10 @@ intelSetBackClipRects(struct intel_context *intel) } else { /* use the back clip rects */ + if (intel->batch->cliprect_mode == LOOP_CLIPRECTS && + intel->pClipRects != dPriv->pBackClipRects) + intel_batchbuffer_flush(intel->batch); + intel->numClipRects = dPriv->numBackClipRects; intel->pClipRects = dPriv->pBackClipRects; intel->drawX = dPriv->backX; @@ -900,12 +922,6 @@ intel_draw_buffer(GLcontext * ctx, struct gl_framebuffer *fb) if (fb->Name) intel_validate_paired_depth_stencil(ctx, fb); - /* If the batch contents require looping over cliprects, flush them before - * we go changing which cliprects get referenced when that happens. - */ - if (intel->batch->cliprect_mode == LOOP_CLIPRECTS) - intel_batchbuffer_flush(intel->batch); - /* * How many color buffers are we drawing into? */ diff --git a/src/mesa/drivers/dri/intel/intel_context.c b/src/mesa/drivers/dri/intel/intel_context.c index 57e574447a6..2b3a9b9d371 100644 --- a/src/mesa/drivers/dri/intel/intel_context.c +++ b/src/mesa/drivers/dri/intel/intel_context.c @@ -68,14 +68,15 @@ int INTEL_DEBUG = (0); #endif -#define need_GL_NV_point_sprite #define need_GL_ARB_multisample +#define need_GL_ARB_occlusion_query #define need_GL_ARB_point_parameters +#define need_GL_ARB_shader_objects #define need_GL_ARB_texture_compression #define need_GL_ARB_vertex_buffer_object #define need_GL_ARB_vertex_program +#define need_GL_ARB_vertex_shader #define need_GL_ARB_window_pos -#define need_GL_ARB_occlusion_query #define need_GL_EXT_blend_color #define need_GL_EXT_blend_equation_separate #define need_GL_EXT_blend_func_separate @@ -84,14 +85,13 @@ int INTEL_DEBUG = (0); #define need_GL_EXT_fog_coord #define need_GL_EXT_framebuffer_object #define need_GL_EXT_multi_draw_arrays +#define need_GL_EXT_point_parameters #define need_GL_EXT_secondary_color -#define need_GL_NV_vertex_program #define need_GL_ATI_separate_stencil -#define need_GL_EXT_point_parameters +#define need_GL_NV_point_sprite +#define need_GL_NV_vertex_program #define need_GL_VERSION_2_0 #define need_GL_VERSION_2_1 -#define need_GL_ARB_shader_objects -#define need_GL_ARB_vertex_shader #include "extension_helper.h" @@ -344,88 +344,82 @@ intel_viewport(GLcontext *ctx, GLint x, GLint y, GLsizei w, GLsizei h) * i965_dri. */ static const struct dri_extension card_extensions[] = { - {"GL_ARB_multisample", GL_ARB_multisample_functions}, - {"GL_ARB_multitexture", NULL}, - {"GL_ARB_point_parameters", GL_ARB_point_parameters_functions}, - {"GL_NV_point_sprite", GL_NV_point_sprite_functions}, - {"GL_ARB_texture_border_clamp", NULL}, - {"GL_ARB_texture_compression", GL_ARB_texture_compression_functions}, - {"GL_ARB_texture_cube_map", NULL}, - {"GL_ARB_texture_env_add", NULL}, - {"GL_ARB_texture_env_combine", NULL}, - {"GL_ARB_texture_env_crossbar", NULL}, - {"GL_ARB_texture_env_dot3", NULL}, - {"GL_ARB_texture_mirrored_repeat", NULL}, - {"GL_ARB_texture_non_power_of_two", NULL }, - {"GL_ARB_texture_rectangle", NULL}, - {"GL_NV_texture_rectangle", NULL}, - {"GL_EXT_texture_rectangle", NULL}, - {"GL_ARB_point_parameters", NULL}, - {"GL_ARB_vertex_buffer_object", GL_ARB_vertex_buffer_object_functions}, - {"GL_ARB_vertex_program", GL_ARB_vertex_program_functions}, - {"GL_ARB_window_pos", GL_ARB_window_pos_functions}, - {"GL_EXT_blend_color", GL_EXT_blend_color_functions}, - {"GL_EXT_blend_equation_separate", - GL_EXT_blend_equation_separate_functions}, - {"GL_EXT_blend_func_separate", GL_EXT_blend_func_separate_functions}, - {"GL_EXT_blend_minmax", GL_EXT_blend_minmax_functions}, - {"GL_EXT_blend_logic_op", NULL}, - {"GL_EXT_blend_subtract", NULL}, - {"GL_EXT_cull_vertex", GL_EXT_cull_vertex_functions}, - {"GL_EXT_fog_coord", GL_EXT_fog_coord_functions}, - {"GL_EXT_multi_draw_arrays", GL_EXT_multi_draw_arrays_functions}, - {"GL_ATI_separate_stencil", GL_ATI_separate_stencil_functions}, -#if 1 /* XXX FBO temporary? */ - {"GL_EXT_packed_depth_stencil", NULL}, -#endif - {"GL_EXT_secondary_color", GL_EXT_secondary_color_functions}, - {"GL_EXT_stencil_wrap", NULL}, - {"GL_EXT_texture_edge_clamp", NULL}, - {"GL_EXT_texture_env_combine", NULL}, - {"GL_EXT_texture_env_dot3", NULL}, - {"GL_EXT_texture_filter_anisotropic", NULL}, - {"GL_EXT_texture_lod_bias", NULL}, - {"GL_3DFX_texture_compression_FXT1", NULL}, - {"GL_APPLE_client_storage", NULL}, - {"GL_MESA_pack_invert", NULL}, - {"GL_MESA_ycbcr_texture", NULL}, - {"GL_NV_blend_square", NULL}, - {"GL_NV_vertex_program", GL_NV_vertex_program_functions}, - {"GL_NV_vertex_program1_1", NULL}, - { "GL_SGIS_generate_mipmap", NULL }, - {NULL, NULL} + { "GL_ARB_multisample", GL_ARB_multisample_functions }, + { "GL_ARB_multitexture", NULL }, + { "GL_ARB_point_parameters", GL_ARB_point_parameters_functions }, + { "GL_ARB_texture_border_clamp", NULL }, + { "GL_ARB_texture_compression", GL_ARB_texture_compression_functions }, + { "GL_ARB_texture_cube_map", NULL }, + { "GL_ARB_texture_env_add", NULL }, + { "GL_ARB_texture_env_combine", NULL }, + { "GL_ARB_texture_env_crossbar", NULL }, + { "GL_ARB_texture_env_dot3", NULL }, + { "GL_ARB_texture_mirrored_repeat", NULL }, + { "GL_ARB_texture_rectangle", NULL }, + { "GL_ARB_vertex_buffer_object", GL_ARB_vertex_buffer_object_functions }, + { "GL_ARB_vertex_program", GL_ARB_vertex_program_functions }, + { "GL_ARB_window_pos", GL_ARB_window_pos_functions }, + { "GL_EXT_blend_color", GL_EXT_blend_color_functions }, + { "GL_EXT_blend_equation_separate", GL_EXT_blend_equation_separate_functions }, + { "GL_EXT_blend_func_separate", GL_EXT_blend_func_separate_functions }, + { "GL_EXT_blend_minmax", GL_EXT_blend_minmax_functions }, + { "GL_EXT_blend_logic_op", NULL }, + { "GL_EXT_blend_subtract", NULL }, + { "GL_EXT_cull_vertex", GL_EXT_cull_vertex_functions }, + { "GL_EXT_fog_coord", GL_EXT_fog_coord_functions }, + { "GL_EXT_multi_draw_arrays", GL_EXT_multi_draw_arrays_functions }, + { "GL_EXT_packed_depth_stencil", NULL }, + { "GL_EXT_secondary_color", GL_EXT_secondary_color_functions }, + { "GL_EXT_stencil_wrap", NULL }, + { "GL_EXT_texture_edge_clamp", NULL }, + { "GL_EXT_texture_env_combine", NULL }, + { "GL_EXT_texture_env_dot3", NULL }, + { "GL_EXT_texture_filter_anisotropic", NULL }, + { "GL_EXT_texture_lod_bias", NULL }, + { "GL_3DFX_texture_compression_FXT1", NULL }, + { "GL_APPLE_client_storage", NULL }, + { "GL_MESA_pack_invert", NULL }, + { "GL_MESA_ycbcr_texture", NULL }, + { "GL_NV_blend_square", NULL }, + { "GL_NV_point_sprite", GL_NV_point_sprite_functions }, + { "GL_NV_vertex_program", GL_NV_vertex_program_functions }, + { "GL_NV_vertex_program1_1", NULL }, + { "GL_SGIS_generate_mipmap", NULL }, + { NULL, NULL } }; static const struct dri_extension brw_extensions[] = { - { "GL_ARB_shading_language_100", GL_VERSION_2_0_functions}, - { "GL_ARB_shading_language_120", GL_VERSION_2_1_functions}, - { "GL_ARB_shader_objects", GL_ARB_shader_objects_functions}, - { "GL_ARB_vertex_shader", GL_ARB_vertex_shader_functions}, - { "GL_ARB_point_sprite", NULL}, - { "GL_ARB_fragment_shader", NULL }, - { "GL_ARB_draw_buffers", NULL }, { "GL_ARB_depth_texture", NULL }, + { "GL_ARB_draw_buffers", NULL }, { "GL_ARB_fragment_program", NULL }, + { "GL_ARB_fragment_program_shadow", NULL }, + { "GL_ARB_fragment_shader", NULL }, + { "GL_ARB_occlusion_query", GL_ARB_occlusion_query_functions }, + { "GL_ARB_point_sprite", NULL }, + { "GL_ARB_shader_objects", GL_ARB_shader_objects_functions }, + { "GL_ARB_shading_language_100", GL_VERSION_2_0_functions }, +#if 0 + /* Support for GLSL 1.20 is currently broken in core Mesa. + */ + { "GL_ARB_shading_language_120", GL_VERSION_2_1_functions }, +#endif { "GL_ARB_shadow", NULL }, + { "GL_ARB_texture_non_power_of_two", NULL }, + { "GL_ARB_vertex_shader", GL_ARB_vertex_shader_functions }, { "GL_EXT_shadow_funcs", NULL }, - { "GL_ARB_fragment_program_shadow", NULL }, - /* ARB extn won't work if not enabled */ - { "GL_SGIX_depth_texture", NULL }, - { "GL_EXT_texture_sRGB", NULL}, + { "GL_EXT_texture_sRGB", NULL }, + { "GL_ATI_separate_stencil", GL_ATI_separate_stencil_functions }, { NULL, NULL } }; -#ifdef I915_MMIO_READ -static const struct dri_extension arb_oc_extensions[] = { - {"GL_ARB_occlusion_query", GL_ARB_occlusion_query_functions}, - {NULL, NULL} +static const struct dri_extension arb_oq_extensions[] = { + { NULL, NULL } }; -#endif static const struct dri_extension ttm_extensions[] = { - {"GL_EXT_framebuffer_object", GL_EXT_framebuffer_object_functions}, - {"GL_ARB_pixel_buffer_object", NULL}, - {NULL, NULL} + { "GL_ARB_pixel_buffer_object", NULL }, + { "GL_EXT_framebuffer_object", GL_EXT_framebuffer_object_functions }, + { NULL, NULL } }; /** @@ -445,13 +439,6 @@ void intelInitExtensions(GLcontext *ctx, GLboolean enable_imaging) if (intel == NULL || intel->ttm) driInitExtensions(ctx, ttm_extensions, GL_FALSE); -#ifdef I915_MMIO_READ - if (intel == NULL || - (IS_965(intel->intelScreen->deviceID) && - intel->intelScreen->drmMinor >= 8)) - driInitExtensions(ctx, arb_oc_extensions, GL_FALSE); -#endif - if (intel == NULL || IS_965(intel->intelScreen->deviceID)) driInitExtensions(ctx, brw_extensions, GL_FALSE); } @@ -548,39 +535,6 @@ intelFinish(GLcontext * ctx) } } -#ifdef I915_MMIO_READ -static void -intelBeginQuery(GLcontext *ctx, struct gl_query_object *q) -{ - struct intel_context *intel = intel_context( ctx ); - struct drm_i915_mmio io = { - .read_write = I915_MMIO_READ, - .reg = MMIO_REGS_PS_DEPTH_COUNT, - .data = &q->Result - }; - intel->stats_wm++; - intelFinish(&intel->ctx); - drmCommandWrite(intel->driFd, DRM_I915_MMIO, &io, sizeof(io)); -} - -static void -intelEndQuery(GLcontext *ctx, struct gl_query_object *q) -{ - struct intel_context *intel = intel_context( ctx ); - GLuint64EXT tmp; - struct drm_i915_mmio io = { - .read_write = I915_MMIO_READ, - .reg = MMIO_REGS_PS_DEPTH_COUNT, - .data = &tmp - }; - intelFinish(&intel->ctx); - drmCommandWrite(intel->driFd, DRM_I915_MMIO, &io, sizeof(io)); - q->Result = tmp - q->Result; - q->Ready = GL_TRUE; - intel->stats_wm--; -} -#endif - void intelInitDriverFunctions(struct dd_function_table *functions) { @@ -597,11 +551,6 @@ intelInitDriverFunctions(struct dd_function_table *functions) functions->CopyConvolutionFilter1D = _swrast_CopyConvolutionFilter1D; functions->CopyConvolutionFilter2D = _swrast_CopyConvolutionFilter2D; -#ifdef I915_MMIO_READ - functions->BeginQuery = intelBeginQuery; - functions->EndQuery = intelEndQuery; -#endif - intelInitTextureFuncs(functions); intelInitStateFuncs(functions); intelInitBufferFuncs(functions); @@ -810,7 +759,12 @@ intelDestroyContext(__DRIcontextPrivate * driContextPriv) intel->Fallback = 0; /* don't call _swrast_Flush later */ intel_batchbuffer_free(intel->batch); + intel->batch = NULL; + free(intel->prim.vb); + intel->prim.vb = NULL; + dri_bo_unreference(intel->prim.vb_bo); + intel->prim.vb_bo = NULL; if (release_texture_heaps) { /* This share group is about to go away, free our private @@ -820,6 +774,13 @@ intelDestroyContext(__DRIcontextPrivate * driContextPriv) fprintf(stderr, "do something to free texture heaps\n"); } + intel_region_release(&intel->front_region); + intel_region_release(&intel->back_region); + intel_region_release(&intel->third_region); + intel_region_release(&intel->depth_region); + + driDestroyOptionCache(&intel->optionCache); + /* free the Mesa context */ _mesa_free_context_data(&intel->ctx); } diff --git a/src/mesa/drivers/dri/intel/intel_context.h b/src/mesa/drivers/dri/intel/intel_context.h index 4af4cb9c96f..554159ac441 100644 --- a/src/mesa/drivers/dri/intel/intel_context.h +++ b/src/mesa/drivers/dri/intel/intel_context.h @@ -226,7 +226,6 @@ struct intel_context GLenum reduced_primitive; GLuint vertex_size; GLubyte *verts; /* points to tnl->clipspace.vertex_buf */ - struct intel_region *draw_region; /* Fallback rasterization functions */ diff --git a/src/mesa/drivers/dri/intel/intel_pixel.c b/src/mesa/drivers/dri/intel/intel_pixel.c index f39fac13cf5..b267ffd890c 100644 --- a/src/mesa/drivers/dri/intel/intel_pixel.c +++ b/src/mesa/drivers/dri/intel/intel_pixel.c @@ -181,9 +181,9 @@ intelInitPixelFuncs(struct dd_function_table *functions) if (!getenv("INTEL_NO_BLIT")) { functions->Bitmap = intelBitmap; functions->CopyPixels = intelCopyPixels; + functions->DrawPixels = intelDrawPixels; #ifdef I915 functions->ReadPixels = intelReadPixels; - functions->DrawPixels = intelDrawPixels; #endif } } diff --git a/src/mesa/drivers/dri/intel/intel_pixel_draw.c b/src/mesa/drivers/dri/intel/intel_pixel_draw.c index be213e7b961..b60dad74601 100644 --- a/src/mesa/drivers/dri/intel/intel_pixel_draw.c +++ b/src/mesa/drivers/dri/intel/intel_pixel_draw.c @@ -51,7 +51,6 @@ #include "intel_regions.h" #include "intel_pixel.h" #include "intel_buffer_objects.h" -#include "intel_tris.h" static GLboolean diff --git a/src/mesa/drivers/dri/intel/intel_reg.h b/src/mesa/drivers/dri/intel/intel_reg.h index 96af7e1a030..c21f4080935 100644 --- a/src/mesa/drivers/dri/intel/intel_reg.h +++ b/src/mesa/drivers/dri/intel/intel_reg.h @@ -45,6 +45,25 @@ #define I1_LOAD_S(n) (1<<(4+n)) /** @{ + * + * PIPE_CONTROL operation, a combination MI_FLUSH and register write with + * additional flushing control. + */ +#define _3DSTATE_PIPE_CONTROL (CMD_3D | (3 << 27) | (2 << 24) | 2) +#define PIPE_CONTROL_NO_WRITE (0 << 14) +#define PIPE_CONTROL_WRITE_IMMEDIATE (1 << 14) +#define PIPE_CONTROL_WRITE_DEPTH_COUNT (2 << 14) +#define PIPE_CONTROL_WRITE_TIMESTAMP (3 << 14) +#define PIPE_CONTROL_DEPTH_STALL (1 << 13) +#define PIPE_CONTROL_WRITE_FLUSH (1 << 12) +#define PIPE_CONTROL_INSTRUCTION_FLUSH (1 << 11) +#define PIPE_CONTROL_INTERRUPT_ENABLE (1 << 8) +#define PIPE_CONTROL_PPGTT_WRITE (0 << 2) +#define PIPE_CONTROL_GLOBAL_GTT_WRITE (1 << 2) + +/** @} */ + +/** @{ * 915 definitions */ #define S0_VB_OFFSET_MASK 0xffffffc diff --git a/src/mesa/drivers/dri/intel/intel_regions.c b/src/mesa/drivers/dri/intel/intel_regions.c index cb0f4ba083b..45faf64c713 100644 --- a/src/mesa/drivers/dri/intel/intel_regions.c +++ b/src/mesa/drivers/dri/intel/intel_regions.c @@ -478,6 +478,11 @@ intel_recreate_static(struct intel_context *intel, region->pitch = intelScreen->pitch; region->height = intelScreen->height; /* needed? */ + if (region->buffer != NULL) { + dri_bo_unreference(region->buffer); + region->buffer = NULL; + } + if (intel->ttm) { assert(region_desc->bo_handle != -1); region->buffer = intel_bo_gem_create_from_name(intel->bufmgr, @@ -486,6 +491,11 @@ intel_recreate_static(struct intel_context *intel, intel_set_region_tiling_gem(intel, region, region_desc->bo_handle); } else { + if (region->classic_map != NULL) { + drmUnmap(region->classic_map, + region->pitch * region->cpp * region->height); + region->classic_map = NULL; + } ret = drmMap(intel->driFd, region_desc->handle, region->pitch * region->cpp * region->height, ®ion->classic_map); diff --git a/src/mesa/drivers/dri/intel/intel_tex_copy.c b/src/mesa/drivers/dri/intel/intel_tex_copy.c index a7b88b39c04..f4cb4a781c6 100644 --- a/src/mesa/drivers/dri/intel/intel_tex_copy.c +++ b/src/mesa/drivers/dri/intel/intel_tex_copy.c @@ -60,7 +60,7 @@ get_teximage_source(struct intel_context *intel, GLenum internalFormat) switch (internalFormat) { case GL_DEPTH_COMPONENT: - case GL_DEPTH_COMPONENT16_ARB: + case GL_DEPTH_COMPONENT16: irb = intel_get_renderbuffer(intel->ctx.ReadBuffer, BUFFER_DEPTH); if (irb && irb->region && irb->region->cpp == 2) return irb->region; diff --git a/src/mesa/drivers/dri/intel/intel_tex_validate.c b/src/mesa/drivers/dri/intel/intel_tex_validate.c index 3dae738ac2e..820683d42eb 100644 --- a/src/mesa/drivers/dri/intel/intel_tex_validate.c +++ b/src/mesa/drivers/dri/intel/intel_tex_validate.c @@ -141,10 +141,7 @@ intel_finalize_mipmap_tree(struct intel_context *intel, GLuint unit) /* Fallback case: */ - if (firstImage->base.Border || - ((firstImage->base._BaseFormat == GL_DEPTH_COMPONENT) && - ((tObj->WrapS == GL_CLAMP_TO_BORDER) || - (tObj->WrapT == GL_CLAMP_TO_BORDER)))) { + if (firstImage->base.Border) { if (intelObj->mt) { intel_miptree_release(intel, &intelObj->mt); } diff --git a/src/mesa/drivers/dri/r300/r300_context.c b/src/mesa/drivers/dri/r300/r300_context.c index ee4a69dce36..37436275e34 100644 --- a/src/mesa/drivers/dri/r300/r300_context.c +++ b/src/mesa/drivers/dri/r300/r300_context.c @@ -138,7 +138,6 @@ const struct dri_extension card_extensions[] = { {"GL_NV_blend_square", NULL}, {"GL_NV_vertex_program", GL_NV_vertex_program_functions}, {"GL_SGIS_generate_mipmap", NULL}, - {"GL_SGIX_depth_texture", NULL}, {NULL, NULL} /* *INDENT-ON* */ }; diff --git a/src/mesa/glapi/glapi.c b/src/mesa/glapi/glapi.c index 4b23b422233..9b5144a88b6 100644 --- a/src/mesa/glapi/glapi.c +++ b/src/mesa/glapi/glapi.c @@ -294,7 +294,31 @@ _glapi_get_context(void) #endif } +#ifdef USE_X86_ASM +#if defined( GLX_USE_TLS ) +extern GLubyte gl_dispatch_functions_start[]; +extern GLubyte gl_dispatch_functions_end[]; +#else +extern const GLubyte gl_dispatch_functions_start[]; +#endif + +#endif /* USE_X86_ASM */ + + +#if defined(USE_X64_64_ASM) && defined(GLX_USE_TLS) +# define DISPATCH_FUNCTION_SIZE 16 +#elif defined(USE_X86_ASM) +# if defined(THREADS) && !defined(GLX_USE_TLS) +# define DISPATCH_FUNCTION_SIZE 32 +# else +# define DISPATCH_FUNCTION_SIZE 16 +# endif +#endif + +#if !defined(DISPATCH_FUNCTION_SIZE) && !defined(XFree86Server) && !defined(XGLServer) +# define NEED_FUNCTION_POINTER +#endif #if defined(PTHREADS) || defined(GLX_USE_TLS) /** diff --git a/src/mesa/main/api_exec.c b/src/mesa/main/api_exec.c index 0c3c9c4de49..bae3bf11cb5 100644 --- a/src/mesa/main/api_exec.c +++ b/src/mesa/main/api_exec.c @@ -58,7 +58,7 @@ #include "colortab.h" #endif #include "context.h" -#if FEATURE_convolution +#if FEATURE_convolve #include "convolve.h" #endif #include "depth.h" @@ -402,7 +402,7 @@ _mesa_init_exec_table(struct _glapi_table *exec) SET_GetColorTableParameteriv(exec, _mesa_GetColorTableParameteriv); #endif -#if FEATURE_convolution +#if FEATURE_convolve SET_ConvolutionFilter1D(exec, _mesa_ConvolutionFilter1D); SET_ConvolutionFilter2D(exec, _mesa_ConvolutionFilter2D); SET_ConvolutionParameterf(exec, _mesa_ConvolutionParameterf); diff --git a/src/mesa/main/api_loopback.c b/src/mesa/main/api_loopback.c index 924d7134a27..0e3f5ff9570 100644 --- a/src/mesa/main/api_loopback.c +++ b/src/mesa/main/api_loopback.c @@ -31,7 +31,6 @@ #include "glheader.h" #include "macros.h" -#include "colormac.h" #include "api_loopback.h" #include "mtypes.h" #include "glapi/glapi.h" diff --git a/src/mesa/main/blend.c b/src/mesa/main/blend.c index 4d4a8971419..39cf6153e28 100644 --- a/src/mesa/main/blend.c +++ b/src/mesa/main/blend.c @@ -31,7 +31,6 @@ #include "glheader.h" #include "blend.h" -#include "colormac.h" #include "context.h" #include "enums.h" #include "macros.h" diff --git a/src/mesa/main/colormac.h b/src/mesa/main/colormac.h index a19521fc85b..a34bd2ed388 100644 --- a/src/mesa/main/colormac.h +++ b/src/mesa/main/colormac.h @@ -1,8 +1,8 @@ /* * Mesa 3-D graphics library - * Version: 6.1 + * Version: 7.3 * - * Copyright (C) 1999-2004 Brian Paul All Rights Reserved. + * Copyright (C) 1999-2008 Brian Paul All Rights Reserved. * * Permission is hereby granted, free of charge, to any person obtaining a * copy of this software and associated documentation files (the "Software"), @@ -180,20 +180,20 @@ do { \ */ /*@{*/ -#define PACK_COLOR_8888( R, G, B, A ) \ - (((R) << 24) | ((G) << 16) | ((B) << 8) | (A)) +#define PACK_COLOR_8888( X, Y, Z, W ) \ + (((X) << 24) | ((Y) << 16) | ((Z) << 8) | (W)) -#define PACK_COLOR_8888_REV( R, G, B, A ) \ - (((A) << 24) | ((B) << 16) | ((G) << 8) | (R)) +#define PACK_COLOR_8888_REV( X, Y, Z, W ) \ + (((W) << 24) | ((Z) << 16) | ((Y) << 8) | (X)) -#define PACK_COLOR_888( R, G, B ) \ - (((R) << 16) | ((G) << 8) | (B)) +#define PACK_COLOR_888( X, Y, Z ) \ + (((X) << 16) | ((Y) << 8) | (Z)) -#define PACK_COLOR_565( R, G, B ) \ - ((((R) & 0xf8) << 8) | (((G) & 0xfc) << 3) | (((B) & 0xf8) >> 3)) +#define PACK_COLOR_565( X, Y, Z ) \ + ((((X) & 0xf8) << 8) | (((Y) & 0xfc) << 3) | (((Z) & 0xf8) >> 3)) -#define PACK_COLOR_565_REV( R, G, B ) \ - (((R) & 0xf8) | ((G) & 0xe0) >> 5 | (((G) & 0x1c) << 11) | (((B) & 0xf8) << 5)) +#define PACK_COLOR_565_REV( X, Y, Z ) \ + (((X) & 0xf8) | ((Y) & 0xe0) >> 5 | (((Y) & 0x1c) << 11) | (((Z) & 0xf8) << 5)) #define PACK_COLOR_1555( A, B, G, R ) \ ((((B) & 0xf8) << 7) | (((G) & 0xf8) << 2) | (((R) & 0xf8) >> 3) | \ diff --git a/src/mesa/main/depthstencil.c b/src/mesa/main/depthstencil.c index fb54d6184d9..9d208e2997d 100644 --- a/src/mesa/main/depthstencil.c +++ b/src/mesa/main/depthstencil.c @@ -282,8 +282,8 @@ _mesa_new_z24_renderbuffer_wrapper(GLcontext *ctx, z24rb->RefCount = 1; z24rb->Width = dsrb->Width; z24rb->Height = dsrb->Height; - z24rb->InternalFormat = GL_DEPTH_COMPONENT24_ARB; - z24rb->_ActualFormat = GL_DEPTH_COMPONENT24_ARB; + z24rb->InternalFormat = GL_DEPTH_COMPONENT24; + z24rb->_ActualFormat = GL_DEPTH_COMPONENT24; z24rb->_BaseFormat = GL_DEPTH_COMPONENT; z24rb->DataType = GL_UNSIGNED_INT; z24rb->DepthBits = 24; diff --git a/src/mesa/main/descrip.mms b/src/mesa/main/descrip.mms index 3ef215f47fa..e49ec65d42d 100644 --- a/src/mesa/main/descrip.mms +++ b/src/mesa/main/descrip.mms @@ -1,6 +1,6 @@ # Makefile for core library for VMS # contributed by Jouk Jansen [email protected] -# Last revision : 2 October 2007 +# Last revision : 29 September 2008 .first define gl [---.include.gl] @@ -21,6 +21,7 @@ CFLAGS = /include=($(INCDIR),[])/define=(PTHREADS=1)/name=(as_is,short)/float=ie SOURCES =accum.c \ api_arrayelt.c \ + api_exec.c \ api_loopback.c \ api_noop.c \ api_validate.c \ @@ -29,6 +30,7 @@ SOURCES =accum.c \ blend.c \ bufferobj.c \ buffers.c \ + clear.c \ clip.c \ colortab.c \ context.c \ @@ -46,6 +48,7 @@ SOURCES =accum.c \ extensions.c \ fbobject.c \ feedback.c \ + ffvertex_prog.c \ fog.c \ framebuffer.c \ get.c \ @@ -60,22 +63,29 @@ SOURCES =accum.c \ matrix.c \ mipmap.c \ mm.c \ + multisample.c \ pixel.c \ + pixelstore.c \ points.c \ polygon.c \ rastpos.c \ rbadaptors.c \ + readpix.c \ renderbuffer.c \ + scissor.c \ shaders.c \ state.c \ stencil.c \ texcompress.c \ texcompress_fxt1.c \ texcompress_s3tc.c \ + texenv.c \ texenvprogram.c \ texformat.c \ + texgen.c \ teximage.c \ texobj.c \ + texparam.c \ texrender.c \ texstate.c \ texstore.c \ @@ -86,6 +96,7 @@ SOURCES =accum.c \ OBJECTS=accum.obj,\ api_arrayelt.obj,\ +api_exec.obj,\ api_loopback.obj,\ api_noop.obj,\ api_validate.obj,\ @@ -94,6 +105,7 @@ attrib.obj,\ blend.obj,\ bufferobj.obj,\ buffers.obj,\ +clear.obj,\ clip.obj,\ colortab.obj,\ context.obj,\ @@ -111,6 +123,7 @@ execmem.obj,\ extensions.obj,\ fbobject.obj,\ feedback.obj,\ +ffvertex_prog.obj,\ fog.obj,\ framebuffer.obj,\ get.obj,\ @@ -125,21 +138,28 @@ lines.obj,\ matrix.obj,\ mipmap.obj,\ mm.obj,\ +multisample.obj,\ pixel.obj,\ +pixelstore.obj,\ points.obj,\ polygon.obj,\ rastpos.obj,\ +readpix.obj,\ renderbuffer.obj,\ +scissor.obj,\ shaders.obj,\ state.obj,\ stencil.obj,\ texcompress.obj,\ texcompress_fxt1.obj,\ texcompress_s3tc.obj,\ +texenv.obj,\ texenvprogram.obj,\ texformat.obj,\ +texgen.obj,\ teximage.obj,\ texobj.obj,\ +texparam.obj,\ texrender.obj,\ texstate.obj,\ texstore.obj,\ @@ -226,3 +246,13 @@ vtxfmt.obj : vtxfmt.c shaders.obj : shaders.c queryobj.obj : queryobj.c rbadaptors.obj : rbadaptors.c +clear.obj : clear.c +multisample.obj : multisample.c +scissor.obj : scissor.c +texenv.obj : texenv.c +texgen.obj : texgen.c +texparam.obj : texparam.c +readpix.obj : readpix.c +ffvertex_prog.obj : ffvertex_prog.c +api_exec.obj : api_exec.c +pixelstore.obj : pixelstore.c diff --git a/src/mesa/main/dlist.c b/src/mesa/main/dlist.c index f7660930a9c..c7db435506e 100644 --- a/src/mesa/main/dlist.c +++ b/src/mesa/main/dlist.c @@ -41,7 +41,6 @@ #endif #include "arrayobj.h" #include "clip.h" -#include "colormac.h" #include "colortab.h" #include "context.h" #include "convolve.h" diff --git a/src/mesa/main/extensions.c b/src/mesa/main/extensions.c index 8dfbfeb62e4..95bf1165f45 100644 --- a/src/mesa/main/extensions.c +++ b/src/mesa/main/extensions.c @@ -165,7 +165,7 @@ static const struct { { OFF, "GL_SGIS_texture_border_clamp", F(ARB_texture_border_clamp) }, { ON, "GL_SGIS_texture_edge_clamp", F(SGIS_texture_edge_clamp) }, { ON, "GL_SGIS_texture_lod", F(SGIS_texture_lod) }, - { OFF, "GL_SGIX_depth_texture", F(SGIX_depth_texture) }, + { OFF, "GL_SGIX_depth_texture", F(ARB_depth_texture) }, { OFF, "GL_SGIX_shadow", F(SGIX_shadow) }, { OFF, "GL_SGIX_shadow_ambient", F(SGIX_shadow_ambient) }, { OFF, "GL_SUN_multi_draw_arrays", F(EXT_multi_draw_arrays) }, @@ -292,7 +292,6 @@ _mesa_enable_sw_extensions(GLcontext *ctx) ctx->Extensions.SGI_texture_color_table = GL_TRUE; ctx->Extensions.SGIS_generate_mipmap = GL_TRUE; ctx->Extensions.SGIS_texture_edge_clamp = GL_TRUE; - ctx->Extensions.SGIX_depth_texture = GL_TRUE; ctx->Extensions.SGIX_shadow = GL_TRUE; ctx->Extensions.SGIX_shadow_ambient = GL_TRUE; #if FEATURE_ARB_vertex_program || FEATURE_ARB_fragment_program diff --git a/src/mesa/main/ffvertex_prog.c b/src/mesa/main/ffvertex_prog.c index 787672be9f1..308b4ef7115 100644 --- a/src/mesa/main/ffvertex_prog.c +++ b/src/mesa/main/ffvertex_prog.c @@ -178,12 +178,12 @@ static GLboolean check_active_shininess( GLcontext *ctx, -static struct state_key *make_state_key( GLcontext *ctx ) +static void make_state_key( GLcontext *ctx, struct state_key *key ) { const struct gl_fragment_program *fp; - struct state_key *key = CALLOC_STRUCT(state_key); GLuint i; + memset(key, 0, sizeof(struct state_key)); fp = ctx->FragmentProgram._Current; /* This now relies on texenvprogram.c being active: @@ -301,8 +301,6 @@ static struct state_key *make_state_key( GLcontext *ctx ) texUnit->GenModeQ ); } } - - return key; } @@ -1714,16 +1712,16 @@ struct gl_vertex_program * _mesa_get_fixed_func_vertex_program(GLcontext *ctx) { struct gl_vertex_program *prog; - struct state_key *key; + struct state_key key; /* Grab all the relevent state and put it in a single structure: */ - key = make_state_key(ctx); + make_state_key(ctx, &key); /* Look for an already-prepared program for this state: */ prog = (struct gl_vertex_program *) - _mesa_search_program_cache(ctx->VertexProgram.Cache, key, sizeof(*key)); + _mesa_search_program_cache(ctx->VertexProgram.Cache, &key, sizeof(key)); if (!prog) { /* OK, we'll have to build a new one */ @@ -1735,7 +1733,7 @@ _mesa_get_fixed_func_vertex_program(GLcontext *ctx) if (!prog) return NULL; - create_new_program( key, prog, + create_new_program( &key, prog, ctx->Const.VertexProgram.MaxTemps ); #if 0 @@ -1744,10 +1742,8 @@ _mesa_get_fixed_func_vertex_program(GLcontext *ctx) &prog->Base ); #endif _mesa_program_cache_insert(ctx, ctx->VertexProgram.Cache, - key, sizeof(*key), &prog->Base); + &key, sizeof(key), &prog->Base); } - _mesa_free(key); - return prog; } diff --git a/src/mesa/main/getstring.c b/src/mesa/main/getstring.c index 1a82ccce59a..94bf5de1e8b 100644 --- a/src/mesa/main/getstring.c +++ b/src/mesa/main/getstring.c @@ -25,7 +25,6 @@ #include "glheader.h" -#include "colormac.h" #include "context.h" #include "get.h" #include "version.h" diff --git a/src/mesa/main/mfeatures.h b/src/mesa/main/mfeatures.h index ed78f57edf6..3819da3d680 100644 --- a/src/mesa/main/mfeatures.h +++ b/src/mesa/main/mfeatures.h @@ -39,7 +39,7 @@ #define FEATURE_accum _HAVE_FULL_GL #define FEATURE_attrib_stack _HAVE_FULL_GL #define FEATURE_colortable _HAVE_FULL_GL -#define FEATURE_convolution _HAVE_FULL_GL +#define FEATURE_convolve _HAVE_FULL_GL #define FEATURE_dispatch _HAVE_FULL_GL #define FEATURE_dlist _HAVE_FULL_GL #define FEATURE_draw_read_buffer _HAVE_FULL_GL diff --git a/src/mesa/main/mtypes.h b/src/mesa/main/mtypes.h index 8e4f6a2e663..052da2c18e1 100644 --- a/src/mesa/main/mtypes.h +++ b/src/mesa/main/mtypes.h @@ -2628,7 +2628,6 @@ struct gl_extensions GLboolean SGIS_generate_mipmap; GLboolean SGIS_texture_edge_clamp; GLboolean SGIS_texture_lod; - GLboolean SGIX_depth_texture; GLboolean SGIX_shadow; GLboolean SGIX_shadow_ambient; /* or GL_ARB_shadow_ambient */ GLboolean TDFX_texture_compression_FXT1; @@ -2722,6 +2721,7 @@ struct gl_matrix_stack #define _NEW_MULTISAMPLE 0x2000000 /**< __GLcontextRec::Multisample */ #define _NEW_TRACK_MATRIX 0x4000000 /**< __GLcontextRec::VertexProgram */ #define _NEW_PROGRAM 0x8000000 /**< __GLcontextRec::VertexProgram */ +#define _NEW_CURRENT_ATTRIB 0x10000000 /**< __GLcontextRec::Current */ #define _NEW_ALL ~0 /*@}*/ @@ -3049,6 +3049,8 @@ struct __GLcontextRec GLenum RenderMode; /**< either GL_RENDER, GL_SELECT, GL_FEEDBACK */ GLbitfield NewState; /**< bitwise-or of _NEW_* flags */ + GLbitfield varying_vp_inputs; /**< mask of VERT_BIT_* flags */ + /** \name Derived state */ /*@{*/ /** Bitwise-or of DD_* flags. Note that this bitfield may be used before diff --git a/src/mesa/main/queryobj.c b/src/mesa/main/queryobj.c index a1e32e70ba9..554e0b0d181 100644 --- a/src/mesa/main/queryobj.c +++ b/src/mesa/main/queryobj.c @@ -90,12 +90,11 @@ _mesa_wait_query(GLcontext *ctx, struct gl_query_object *q) /** - * Delete an occlusion query object. + * Delete a query object. Called via ctx->Driver.DeleteQuery(). * Not removed from hash table here. - * XXX maybe add Delete() method to gl_query_object class and call that instead */ void -_mesa_delete_query(struct gl_query_object *q) +_mesa_delete_query(GLcontext *ctx, struct gl_query_object *q) { _mesa_free(q); } @@ -546,6 +545,6 @@ delete_queryobj_cb(GLuint id, void *data, void *userData) void _mesa_free_query_data(GLcontext *ctx) { - _mesa_HashDeleteAll(ctx->Query.QueryObjects, delete_queryobj_cb, NULL); + _mesa_HashDeleteAll(ctx->Query.QueryObjects, delete_queryobj_cb, ctx); _mesa_DeleteHashTable(ctx->Query.QueryObjects); } diff --git a/src/mesa/main/queryobj.h b/src/mesa/main/queryobj.h index c05a1f3da82..9a9774641bb 100644 --- a/src/mesa/main/queryobj.h +++ b/src/mesa/main/queryobj.h @@ -37,7 +37,7 @@ extern void _mesa_free_query_data(GLcontext *ctx); extern void -_mesa_delete_query(struct gl_query_object *q); +_mesa_delete_query(GLcontext *ctx, struct gl_query_object *q); extern void _mesa_begin_query(GLcontext *ctx, struct gl_query_object *q); diff --git a/src/mesa/main/shaders.c b/src/mesa/main/shaders.c index f0db0d2a818..e5c54bb10d7 100644 --- a/src/mesa/main/shaders.c +++ b/src/mesa/main/shaders.c @@ -233,13 +233,32 @@ _mesa_GetObjectParameterivARB(GLhandleARB object, GLenum pname, GLint *params) GET_CURRENT_CONTEXT(ctx); /* Implement in terms of GetProgramiv, GetShaderiv */ if (ctx->Driver.IsProgram(ctx, object)) { - ctx->Driver.GetProgramiv(ctx, object, pname, params); + if (pname == GL_OBJECT_TYPE_ARB) { + *params = GL_PROGRAM_OBJECT_ARB; + } else { + ctx->Driver.GetProgramiv(ctx, object, pname, params); + } } else if (ctx->Driver.IsShader(ctx, object)) { - ctx->Driver.GetShaderiv(ctx, object, pname, params); + if (pname == GL_OBJECT_TYPE_ARB) { + *params = GL_SHADER_OBJECT_ARB; + } else { + ctx->Driver.GetShaderiv(ctx, object, pname, params); + } } else { - _mesa_error(ctx, GL_INVALID_OPERATION, "glGetObjectParameterivARB"); + /* error code depends on pname */ + GLenum err; + switch (pname) { + case GL_OBJECT_TYPE_ARB: + case GL_OBJECT_DELETE_STATUS_ARB: + case GL_OBJECT_INFO_LOG_LENGTH_ARB: + err = GL_INVALID_OPERATION; + break; + default: + err = GL_INVALID_VALUE; + } + _mesa_error(ctx, err, "glGetObjectParameterivARB"); } } diff --git a/src/mesa/main/state.c b/src/mesa/main/state.c index 5913019bc12..2f0e7cc368e 100644 --- a/src/mesa/main/state.c +++ b/src/mesa/main/state.c @@ -258,12 +258,6 @@ update_program(GLcontext *ctx) } } - if (ctx->VertexProgram._Current) - assert(ctx->VertexProgram._Current->Base.Parameters); - if (ctx->FragmentProgram._Current) - assert(ctx->FragmentProgram._Current->Base.Parameters); - - /* XXX: get rid of _Active flag. */ #if 1 @@ -447,6 +441,9 @@ _mesa_update_state_locked( GLcontext *ctx ) GLbitfield new_state = ctx->NewState; GLbitfield prog_flags = _NEW_PROGRAM; + if (new_state == _NEW_CURRENT_ATTRIB) + goto out; + if (MESA_VERBOSE & VERBOSE_STATE) _mesa_print_state("_mesa_update_state", new_state); @@ -510,7 +507,8 @@ _mesa_update_state_locked( GLcontext *ctx ) _mesa_update_tnl_spaces( ctx, new_state ); if (ctx->FragmentProgram._MaintainTexEnvProgram) { - prog_flags |= (_NEW_TEXTURE | _NEW_FOG | _DD_NEW_SEPARATE_SPECULAR); + prog_flags |= (_NEW_ARRAY | _NEW_TEXTURE_MATRIX | _NEW_LIGHT | + _NEW_TEXTURE | _NEW_FOG | _DD_NEW_SEPARATE_SPECULAR); } if (ctx->VertexProgram._MaintainTnlProgram) { prog_flags |= (_NEW_ARRAY | _NEW_TEXTURE | _NEW_TEXTURE_MATRIX | @@ -532,6 +530,7 @@ _mesa_update_state_locked( GLcontext *ctx ) * Set ctx->NewState to zero to avoid recursion if * Driver.UpdateState() has to call FLUSH_VERTICES(). (fixed?) */ + out: new_state = ctx->NewState; ctx->NewState = 0; ctx->Driver.UpdateState(ctx, new_state); @@ -548,3 +547,38 @@ _mesa_update_state( GLcontext *ctx ) _mesa_update_state_locked(ctx); _mesa_unlock_context_textures(ctx); } + + + + +/* Want to figure out which fragment program inputs are actually + * constant/current values from ctx->Current. These should be + * referenced as a tracked state variable rather than a fragment + * program input, to save the overhead of putting a constant value in + * every submitted vertex, transferring it to hardware, interpolating + * it across the triangle, etc... + * + * When there is a VP bound, just use vp->outputs. But when we're + * generating vp from fixed function state, basically want to + * calculate: + * + * vp_out_2_fp_in( vp_in_2_vp_out( varying_inputs ) | + * potential_vp_outputs ) + * + * Where potential_vp_outputs is calculated by looking at enabled + * texgen, etc. + * + * The generated fragment program should then only declare inputs that + * may vary or otherwise differ from the ctx->Current values. + * Otherwise, the fp should track them as state values instead. + */ +void +_mesa_set_varying_vp_inputs( GLcontext *ctx, + GLbitfield varying_inputs ) +{ + if (ctx->varying_vp_inputs != varying_inputs) { + ctx->varying_vp_inputs = varying_inputs; + ctx->NewState |= _NEW_ARRAY; + //_mesa_printf("%s %x\n", __FUNCTION__, varying_inputs); + } +} diff --git a/src/mesa/main/state.h b/src/mesa/main/state.h index bb7cb8f32a3..79f2f6beb0c 100644 --- a/src/mesa/main/state.h +++ b/src/mesa/main/state.h @@ -37,5 +37,8 @@ _mesa_update_state( GLcontext *ctx ); extern void _mesa_update_state_locked( GLcontext *ctx ); +void +_mesa_set_varying_vp_inputs( GLcontext *ctx, + GLbitfield varying_inputs ); #endif diff --git a/src/mesa/main/texenvprogram.c b/src/mesa/main/texenvprogram.c index 2bce93eef1d..c23173014de 100644 --- a/src/mesa/main/texenvprogram.c +++ b/src/mesa/main/texenvprogram.c @@ -55,15 +55,17 @@ struct texenvprog_cache_item #define DISASSEM (MESA_VERBOSE & VERBOSE_DISASSEM) struct mode_opt { - GLuint Source:4; - GLuint Operand:3; + GLubyte Source:4; + GLubyte Operand:3; }; struct state_key { - GLbitfield enabled_units; + GLuint nr_enabled_units:8; + GLuint enabled_units:8; GLuint separate_specular:1; GLuint fog_enabled:1; GLuint fog_mode:2; + GLuint inputs_available:12; struct { GLuint enabled:1; @@ -74,10 +76,10 @@ struct state_key { GLuint NumArgsRGB:2; GLuint ModeRGB:4; - struct mode_opt OptRGB[3]; - GLuint NumArgsA:2; GLuint ModeA:4; + + struct mode_opt OptRGB[3]; struct mode_opt OptA[3]; } unit[8]; }; @@ -199,6 +201,66 @@ static GLuint translate_tex_src_bit( GLbitfield bit ) } } +#define VERT_BIT_TEX_ANY (0xff << VERT_ATTRIB_TEX0) +#define VERT_RESULT_TEX_ANY (0xff << VERT_RESULT_TEX0) + +/** + * Identify all possible varying inputs. The fragment program will + * never reference non-varying inputs, but will track them via state + * constants instead. + * + * This function figures out all the inputs that the fragment program + * has access to. The bitmask is later reduced to just those which + * are actually referenced. + */ +static GLbitfield get_fp_input_mask( GLcontext *ctx ) +{ + GLbitfield fp_inputs = 0x0; + + if (!ctx->VertexProgram._Enabled || + !ctx->VertexProgram._Current) { + + /* Fixed function logic */ + GLbitfield varying_inputs = ctx->varying_vp_inputs; + + /* First look at what values may be computed by the generated + * vertex program: + */ + if (ctx->Light.Enabled) { + fp_inputs |= FRAG_BIT_COL0; + + if (ctx->_TriangleCaps & DD_SEPARATE_SPECULAR) + fp_inputs |= FRAG_BIT_COL1; + } + + fp_inputs |= (ctx->Texture._TexGenEnabled | + ctx->Texture._TexMatEnabled) << FRAG_ATTRIB_TEX0; + + /* Then look at what might be varying as a result of enabled + * arrays, etc: + */ + if (varying_inputs & VERT_BIT_COLOR0) fp_inputs |= FRAG_BIT_COL0; + if (varying_inputs & VERT_BIT_COLOR1) fp_inputs |= FRAG_BIT_COL1; + + fp_inputs |= (((varying_inputs & VERT_BIT_TEX_ANY) >> VERT_ATTRIB_TEX0) + << FRAG_ATTRIB_TEX0); + + } + else { + /* calculate from vp->outputs */ + GLbitfield vp_outputs = ctx->VertexProgram._Current->Base.OutputsWritten; + + if (vp_outputs & (1 << VERT_RESULT_COL0)) fp_inputs |= FRAG_BIT_COL0; + if (vp_outputs & (1 << VERT_RESULT_COL1)) fp_inputs |= FRAG_BIT_COL1; + + fp_inputs |= (((vp_outputs & VERT_RESULT_TEX_ANY) >> VERT_RESULT_TEX0) + << FRAG_ATTRIB_TEX0); + } + + return fp_inputs; +} + + /** * Examine current texture environment state and generate a unique * key to identify it. @@ -206,7 +268,9 @@ static GLuint translate_tex_src_bit( GLbitfield bit ) static void make_state_key( GLcontext *ctx, struct state_key *key ) { GLuint i, j; - + GLbitfield inputs_referenced = FRAG_BIT_COL0; + GLbitfield inputs_available = get_fp_input_mask( ctx ); + memset(key, 0, sizeof(*key)); for (i=0;i<MAX_TEXTURE_UNITS;i++) { @@ -217,6 +281,8 @@ static void make_state_key( GLcontext *ctx, struct state_key *key ) key->unit[i].enabled = 1; key->enabled_units |= (1<<i); + key->nr_enabled_units = i+1; + inputs_referenced |= FRAG_BIT_TEX(i); key->unit[i].source_index = translate_tex_src_bit(texUnit->_ReallyEnabled); @@ -245,16 +311,22 @@ static void make_state_key( GLcontext *ctx, struct state_key *key ) } } - if (ctx->_TriangleCaps & DD_SEPARATE_SPECULAR) + if (ctx->_TriangleCaps & DD_SEPARATE_SPECULAR) { key->separate_specular = 1; + inputs_referenced |= FRAG_BIT_COL1; + } if (ctx->Fog.Enabled) { key->fog_enabled = 1; key->fog_mode = translate_fog_mode(ctx->Fog.Mode); + inputs_referenced |= FRAG_BIT_FOGC; /* maybe */ } + + key->inputs_available = (inputs_available & inputs_referenced); } -/* Use uregs to represent registers internally, translate to Mesa's +/** + * Use uregs to represent registers internally, translate to Mesa's * expected formats on emit. * * NOTE: These are passed by value extensively in this file rather @@ -287,16 +359,16 @@ static const struct ureg undef = { }; -/* State used to build the fragment program: +/** State used to build the fragment program: */ struct texenv_fragment_program { struct gl_fragment_program *program; GLcontext *ctx; struct state_key *state; - GLbitfield alu_temps; /* Track texture indirections, see spec. */ - GLbitfield temps_output; /* Track texture indirections, see spec. */ - GLbitfield temp_in_use; /* Tracks temporary regs which are in use. */ + GLbitfield alu_temps; /**< Track texture indirections, see spec. */ + GLbitfield temps_output; /**< Track texture indirections, see spec. */ + GLbitfield temp_in_use; /**< Tracks temporary regs which are in use. */ GLboolean error; struct ureg src_texture[MAX_TEXTURE_UNITS]; @@ -304,11 +376,11 @@ struct texenv_fragment_program { * else undef. */ - struct ureg src_previous; /* Reg containing color from previous + struct ureg src_previous; /**< Reg containing color from previous * stage. May need to be decl'd. */ - GLuint last_tex_stage; /* Number of last enabled texture unit */ + GLuint last_tex_stage; /**< Number of last enabled texture unit */ struct ureg half; struct ureg one; @@ -411,6 +483,14 @@ static struct ureg get_tex_temp( struct texenv_fragment_program *p ) } +/** Mark a temp reg as being no longer allocatable. */ +static void reserve_temp( struct texenv_fragment_program *p, struct ureg r ) +{ + if (r.file == PROGRAM_TEMPORARY) + p->temps_output |= (1 << r.idx); +} + + static void release_temps(GLcontext *ctx, struct texenv_fragment_program *p ) { GLuint max_temp = ctx->Const.FragmentProgram.MaxTemps; @@ -449,11 +529,29 @@ static struct ureg register_param5( struct texenv_fragment_program *p, #define register_param3(p,s0,s1,s2) register_param5(p,s0,s1,s2,0,0) #define register_param4(p,s0,s1,s2,s3) register_param5(p,s0,s1,s2,s3,0) +static GLuint frag_to_vert_attrib( GLuint attrib ) +{ + switch (attrib) { + case FRAG_ATTRIB_COL0: return VERT_ATTRIB_COLOR0; + case FRAG_ATTRIB_COL1: return VERT_ATTRIB_COLOR1; + default: + assert(attrib >= FRAG_ATTRIB_TEX0); + assert(attrib <= FRAG_ATTRIB_TEX7); + return attrib - FRAG_ATTRIB_TEX0 + VERT_ATTRIB_TEX0; + } +} + static struct ureg register_input( struct texenv_fragment_program *p, GLuint input ) { - p->program->Base.InputsRead |= (1 << input); - return make_ureg(PROGRAM_INPUT, input); + if (p->state->inputs_available & (1<<input)) { + p->program->Base.InputsRead |= (1 << input); + return make_ureg(PROGRAM_INPUT, input); + } + else { + GLuint idx = frag_to_vert_attrib( input ); + return register_param3( p, STATE_INTERNAL, STATE_CURRENT_ATTRIB, idx ); + } } @@ -504,10 +602,12 @@ emit_op(struct texenv_fragment_program *p, emit_dst( &inst->DstReg, dest, mask ); +#if 0 /* Accounting for indirection tracking: */ if (dest.file == PROGRAM_TEMPORARY) p->temps_output |= 1 << dest.idx; +#endif return inst; } @@ -562,6 +662,10 @@ static struct ureg emit_texld( struct texenv_fragment_program *p, p->program->Base.NumTexInstructions++; + /* Accounting for indirection tracking: + */ + reserve_temp(p, dest); + /* Is this a texture indirection? */ if ((coord.file == PROGRAM_TEMPORARY && @@ -1079,6 +1183,7 @@ create_new_program(GLcontext *ctx, struct state_key *key, for (unit = 0 ; unit < ctx->Const.MaxTextureUnits; unit++) if (key->enabled_units & (1<<unit)) { p.src_previous = emit_texenv( &p, unit ); + reserve_temp(&p, p.src_previous); /* don't re-use this temp reg */ release_temps(ctx, &p); /* release all temps */ } } diff --git a/src/mesa/main/texformat.c b/src/mesa/main/texformat.c index 60f36c4a87f..ce2772c2992 100644 --- a/src/mesa/main/texformat.c +++ b/src/mesa/main/texformat.c @@ -1420,14 +1420,13 @@ _mesa_choose_tex_format( GLcontext *ctx, GLint internalFormat, ; /* fallthrough */ } - if (ctx->Extensions.SGIX_depth_texture || - ctx->Extensions.ARB_depth_texture) { + if (ctx->Extensions.ARB_depth_texture) { switch (internalFormat) { case GL_DEPTH_COMPONENT: - case GL_DEPTH_COMPONENT24_SGIX: - case GL_DEPTH_COMPONENT32_SGIX: + case GL_DEPTH_COMPONENT24: + case GL_DEPTH_COMPONENT32: return &_mesa_texformat_z32; - case GL_DEPTH_COMPONENT16_SGIX: + case GL_DEPTH_COMPONENT16: return &_mesa_texformat_z16; default: ; /* fallthrough */ diff --git a/src/mesa/main/teximage.c b/src/mesa/main/teximage.c index 91c41f38c30..bf23a1f290d 100644 --- a/src/mesa/main/teximage.c +++ b/src/mesa/main/teximage.c @@ -241,13 +241,12 @@ _mesa_base_tex_format( GLcontext *ctx, GLint internalFormat ) } } - if (ctx->Extensions.SGIX_depth_texture || - ctx->Extensions.ARB_depth_texture) { + if (ctx->Extensions.ARB_depth_texture) { switch (internalFormat) { case GL_DEPTH_COMPONENT: - case GL_DEPTH_COMPONENT16_SGIX: - case GL_DEPTH_COMPONENT24_SGIX: - case GL_DEPTH_COMPONENT32_SGIX: + case GL_DEPTH_COMPONENT16: + case GL_DEPTH_COMPONENT24: + case GL_DEPTH_COMPONENT32: return GL_DEPTH_COMPONENT; default: ; /* fallthrough */ @@ -526,9 +525,9 @@ static GLboolean is_depth_format(GLenum format) { switch (format) { - case GL_DEPTH_COMPONENT16_ARB: - case GL_DEPTH_COMPONENT24_ARB: - case GL_DEPTH_COMPONENT32_ARB: + case GL_DEPTH_COMPONENT16: + case GL_DEPTH_COMPONENT24: + case GL_DEPTH_COMPONENT32: case GL_DEPTH_COMPONENT: return GL_TRUE; default: @@ -2308,8 +2307,7 @@ _mesa_GetTexImage( GLenum target, GLint level, GLenum format, return; } - if (!ctx->Extensions.SGIX_depth_texture && - !ctx->Extensions.ARB_depth_texture && is_depth_format(format)) { + if (!ctx->Extensions.ARB_depth_texture && is_depth_format(format)) { _mesa_error(ctx, GL_INVALID_ENUM, "glGetTexImage(format)"); return; } diff --git a/src/mesa/main/texparam.c b/src/mesa/main/texparam.c index 664adadfb9c..a9e752a6371 100644 --- a/src/mesa/main/texparam.c +++ b/src/mesa/main/texparam.c @@ -642,8 +642,7 @@ _mesa_GetTexLevelParameteriv( GLenum target, GLint level, *params = 0; break; case GL_TEXTURE_DEPTH_SIZE_ARB: - if (ctx->Extensions.SGIX_depth_texture || - ctx->Extensions.ARB_depth_texture) + if (ctx->Extensions.ARB_depth_texture) *params = img->TexFormat->DepthBits; else _mesa_error(ctx, GL_INVALID_ENUM, @@ -903,9 +902,9 @@ _mesa_GetTexParameterfv( GLenum target, GLenum pname, GLfloat *params ) #ifdef FEATURE_OES_draw_texture case GL_TEXTURE_CROP_RECT_OES: params[0] = obj->CropRect[0]; - params[0] = obj->CropRect[1]; - params[0] = obj->CropRect[2]; - params[0] = obj->CropRect[3]; + params[1] = obj->CropRect[1]; + params[2] = obj->CropRect[2]; + params[3] = obj->CropRect[3]; break; #endif default: @@ -1053,9 +1052,9 @@ _mesa_GetTexParameteriv( GLenum target, GLenum pname, GLint *params ) #ifdef FEATURE_OES_draw_texture case GL_TEXTURE_CROP_RECT_OES: params[0] = obj->CropRect[0]; - params[0] = obj->CropRect[1]; - params[0] = obj->CropRect[2]; - params[0] = obj->CropRect[3]; + params[1] = obj->CropRect[1]; + params[2] = obj->CropRect[2]; + params[3] = obj->CropRect[3]; break; #endif default: diff --git a/src/mesa/main/texstore.c b/src/mesa/main/texstore.c index 76b785ab0e0..3639a914c4c 100644 --- a/src/mesa/main/texstore.c +++ b/src/mesa/main/texstore.c @@ -1,6 +1,6 @@ /* * Mesa 3-D graphics library - * Version: 7.1 + * Version: 7.3 * * Copyright (C) 1999-2008 Brian Paul All Rights Reserved. * @@ -1536,10 +1536,10 @@ _mesa_texstore_argb8888(TEXSTORE_PARAMS) for (row = 0; row < srcHeight; row++) { GLuint *d4 = (GLuint *) dstRow; for (col = 0; col < srcWidth; col++) { - d4[col] = ((0xff << 24) | - (srcRow[col * 3 + RCOMP] << 16) | - (srcRow[col * 3 + GCOMP] << 8) | - (srcRow[col * 3 + BCOMP] << 0)); + d4[col] = PACK_COLOR_8888(0xff, + srcRow[col * 3 + RCOMP], + srcRow[col * 3 + GCOMP], + srcRow[col * 3 + BCOMP]); } dstRow += dstRowStride; srcRow += srcRowStride; @@ -1551,8 +1551,7 @@ _mesa_texstore_argb8888(TEXSTORE_PARAMS) dstFormat == &_mesa_texformat_argb8888 && srcFormat == GL_RGBA && baseInternalFormat == GL_RGBA && - srcType == GL_UNSIGNED_BYTE && - littleEndian) { + srcType == GL_UNSIGNED_BYTE) { /* same as above case, but src data has alpha too */ GLint img, row, col; /* For some reason, streaming copies to write-combined regions @@ -1573,39 +1572,10 @@ _mesa_texstore_argb8888(TEXSTORE_PARAMS) for (row = 0; row < srcHeight; row++) { GLuint *d4 = (GLuint *) dstRow; for (col = 0; col < srcWidth; col++) { - d4[col] = ((srcRow[col * 4 + ACOMP] << 24) | - (srcRow[col * 4 + RCOMP] << 16) | - (srcRow[col * 4 + GCOMP] << 8) | - (srcRow[col * 4 + BCOMP] << 0)); - } - dstRow += dstRowStride; - srcRow += srcRowStride; - } - } - } - else if (!ctx->_ImageTransferState && - !srcPacking->SwapBytes && - dstFormat == &_mesa_texformat_argb8888 && - srcFormat == GL_RGBA && - baseInternalFormat == GL_RGBA && - srcType == GL_UNSIGNED_BYTE) { - - GLint img, row, col; - for (img = 0; img < srcDepth; img++) { - const GLint srcRowStride = _mesa_image_row_stride(srcPacking, - srcWidth, srcFormat, srcType); - GLubyte *srcRow = (GLubyte *) _mesa_image_address(dims, srcPacking, - srcAddr, srcWidth, srcHeight, srcFormat, srcType, img, 0, 0); - GLubyte *dstRow = (GLubyte *) dstAddr - + dstImageOffsets[dstZoffset + img] * dstFormat->TexelBytes - + dstYoffset * dstRowStride - + dstXoffset * dstFormat->TexelBytes; - for (row = 0; row < srcHeight; row++) { - for (col = 0; col < srcWidth; col++) { - dstRow[col * 4 + 0] = srcRow[col * 4 + BCOMP]; - dstRow[col * 4 + 1] = srcRow[col * 4 + GCOMP]; - dstRow[col * 4 + 2] = srcRow[col * 4 + RCOMP]; - dstRow[col * 4 + 3] = srcRow[col * 4 + ACOMP]; + d4[col] = PACK_COLOR_8888(srcRow[col * 4 + ACOMP], + srcRow[col * 4 + RCOMP], + srcRow[col * 4 + GCOMP], + srcRow[col * 4 + BCOMP]); } dstRow += dstRowStride; srcRow += srcRowStride; diff --git a/src/mesa/shader/arbprogparse.c b/src/mesa/shader/arbprogparse.c index f499499eb3a..536404bf978 100644 --- a/src/mesa/shader/arbprogparse.c +++ b/src/mesa/shader/arbprogparse.c @@ -64,12 +64,6 @@ having three separate program parameter arrays. #include "prog_statevars.h" #include "prog_instruction.h" - -/* For ARB programs, use the NV instruction limits */ -#define MAX_INSTRUCTIONS MAX2(MAX_NV_FRAGMENT_PROGRAM_INSTRUCTIONS, \ - MAX_NV_VERTEX_PROGRAM_INSTRUCTIONS) - - /** * This is basically a union of the vertex_program and fragment_program * structs that we can use to parse the program into @@ -2609,7 +2603,7 @@ parse_src_reg (GLcontext * ctx, const GLubyte ** inst, /* If we're referencing the Program->Parameters[] array, check if the * parameter is really a constant/literal. If so, set File to CONSTANT. */ - assert(*Index < Program->Base.Parameters->NumParameters); + assert(*Index < (GLint) Program->Base.Parameters->NumParameters); file = Program->Base.Parameters->Parameters[*Index].Type; if (file == PROGRAM_CONSTANT) *File = PROGRAM_CONSTANT; @@ -3443,7 +3437,7 @@ parse_instructions(GLcontext * ctx, const GLubyte * inst, : ctx->Const.VertexProgram.MaxInstructions; GLint err = 0; - ASSERT(MAX_INSTRUCTIONS >= maxInst); + ASSERT(MAX_PROGRAM_INSTRUCTIONS >= maxInst); Program->MajorVersion = (GLuint) * inst++; Program->MinorVersion = (GLuint) * inst++; @@ -3798,7 +3792,7 @@ _mesa_parse_arb_program(GLcontext *ctx, GLenum target, /* Initialize the arb_program struct */ program->Base.String = strz; - program->Base.Instructions = _mesa_alloc_instructions(MAX_INSTRUCTIONS); + program->Base.Instructions = _mesa_alloc_instructions(MAX_PROGRAM_INSTRUCTIONS); program->Base.NumInstructions = program->Base.NumTemporaries = program->Base.NumParameters = @@ -3843,12 +3837,12 @@ _mesa_parse_arb_program(GLcontext *ctx, GLenum target, _mesa_free (parsed); - /* Reallocate the instruction array from size [MAX_INSTRUCTIONS] + /* Reallocate the instruction array from size [MAX_PROGRAM_INSTRUCTIONS] * to size [ap.Base.NumInstructions]. */ program->Base.Instructions = _mesa_realloc_instructions(program->Base.Instructions, - MAX_INSTRUCTIONS, + MAX_PROGRAM_INSTRUCTIONS, program->Base.NumInstructions); return !err; @@ -3901,6 +3895,9 @@ _mesa_parse_arb_fragment_program(GLcontext* ctx, GLenum target, program->FogOption = ap.FogOption; program->UsesKill = ap.UsesKill; + if (program->FogOption) + program->Base.InputsRead |= FRAG_BIT_FOGC; + if (program->Base.Instructions) _mesa_free(program->Base.Instructions); program->Base.Instructions = ap.Base.Instructions; diff --git a/src/mesa/shader/descrip.mms b/src/mesa/shader/descrip.mms index bdac946efe2..19bafd48302 100644 --- a/src/mesa/shader/descrip.mms +++ b/src/mesa/shader/descrip.mms @@ -1,6 +1,6 @@ # Makefile for core library for VMS # contributed by Jouk Jansen [email protected] -# Last revision : 27 May 2008 +# Last revision : 29 September 2008 .first define gl [---.include.gl] define math [-.math] @@ -34,6 +34,7 @@ SOURCES = \ prog_instruction.c \ prog_parameter.c \ prog_print.c \ + prog_cache.c \ prog_statevars.c \ shader_api.c prog_uniform.c @@ -52,7 +53,7 @@ OBJECTS = \ prog_parameter.obj,\ prog_print.obj,\ prog_statevars.obj,\ - shader_api.obj,prog_uniform.obj + shader_api.obj,prog_uniform.obj,prog_cache.obj ##### RULES ##### @@ -91,3 +92,4 @@ prog_print.obj : prog_print.c prog_statevars.obj : prog_statevars.c shader_api.obj : shader_api.c prog_uniform.obj : prog_uniform.c +prog_cache.obj : prog_cache.c diff --git a/src/mesa/shader/prog_cache.c b/src/mesa/shader/prog_cache.c index 36a25377c55..9437e596138 100644 --- a/src/mesa/shader/prog_cache.c +++ b/src/mesa/shader/prog_cache.c @@ -44,6 +44,7 @@ struct cache_item struct gl_program_cache { struct cache_item **items; + struct cache_item *last; GLuint size, n_items; }; @@ -83,6 +84,8 @@ rehash(struct gl_program_cache *cache) struct cache_item *c, *next; GLuint size, i; + cache->last = NULL; + size = cache->size * 3; items = (struct cache_item**) _mesa_malloc(size * sizeof(*items)); _mesa_memset(items, 0, size * sizeof(*items)); @@ -105,6 +108,8 @@ clear_cache(GLcontext *ctx, struct gl_program_cache *cache) { struct cache_item *c, *next; GLuint i; + + cache->last = NULL; for (i = 0; i < cache->size; i++) { for (c = cache->items[i]; c; c = next) { @@ -149,18 +154,26 @@ _mesa_delete_program_cache(GLcontext *ctx, struct gl_program_cache *cache) struct gl_program * -_mesa_search_program_cache(const struct gl_program_cache *cache, +_mesa_search_program_cache(struct gl_program_cache *cache, const void *key, GLuint keysize) { - const GLuint hash = hash_key(key, keysize); - struct cache_item *c; - - for (c = cache->items[hash % cache->size]; c; c = c->next) { - if (c->hash == hash && memcmp(c->key, key, keysize) == 0) - return c->program; + if (cache->last && + memcmp(cache->last->key, key, keysize) == 0) { + return cache->last->program; } + else { + const GLuint hash = hash_key(key, keysize); + struct cache_item *c; + + for (c = cache->items[hash % cache->size]; c; c = c->next) { + if (c->hash == hash && memcmp(c->key, key, keysize) == 0) { + cache->last = c; + return c->program; + } + } - return NULL; + return NULL; + } } diff --git a/src/mesa/shader/prog_cache.h b/src/mesa/shader/prog_cache.h index a8c91fba011..4e1ccac03ff 100644 --- a/src/mesa/shader/prog_cache.h +++ b/src/mesa/shader/prog_cache.h @@ -42,7 +42,7 @@ _mesa_delete_program_cache(GLcontext *ctx, struct gl_program_cache *pc); extern struct gl_program * -_mesa_search_program_cache(const struct gl_program_cache *cache, +_mesa_search_program_cache(struct gl_program_cache *cache, const void *key, GLuint keysize); extern void diff --git a/src/mesa/shader/prog_statevars.c b/src/mesa/shader/prog_statevars.c index d4e31207e84..34c47413506 100644 --- a/src/mesa/shader/prog_statevars.c +++ b/src/mesa/shader/prog_statevars.c @@ -395,6 +395,12 @@ _mesa_fetch_state(GLcontext *ctx, const gl_state_index state[], case STATE_INTERNAL: switch (state[1]) { + case STATE_CURRENT_ATTRIB: { + const GLuint idx = (GLuint) state[2]; + COPY_4V(value, ctx->Current.Attrib[idx]); + return; + } + case STATE_NORMAL_SCALE: ASSIGN_4V(value, ctx->_ModelViewInvScale, @@ -501,6 +507,9 @@ _mesa_fetch_state(GLcontext *ctx, const gl_state_index state[], } return; + /* XXX: make sure new tokens added here are also handled in the + * _mesa_program_state_flags() switch, below. + */ default: /* unknown state indexes are silently ignored * should be handled by the driver. @@ -574,11 +583,29 @@ _mesa_program_state_flags(const gl_state_index state[STATE_LENGTH]) case STATE_INTERNAL: switch (state[1]) { + case STATE_CURRENT_ATTRIB: + return _NEW_CURRENT_ATTRIB; + + case STATE_NORMAL_SCALE: + return _NEW_MODELVIEW; + case STATE_TEXRECT_SCALE: case STATE_SHADOW_AMBIENT: return _NEW_TEXTURE; case STATE_FOG_PARAMS_OPTIMIZED: return _NEW_FOG; + case STATE_LIGHT_SPOT_DIR_NORMALIZED: + case STATE_LIGHT_POSITION: + case STATE_LIGHT_POSITION_NORMALIZED: + case STATE_LIGHT_HALF_VECTOR: + return _NEW_LIGHT; + + case STATE_PT_SCALE: + case STATE_PT_BIAS: + case STATE_PCM_SCALE: + case STATE_PCM_BIAS: + return _NEW_PIXEL; + default: /* unknown state indexes are silently ignored and * no flag set, since it is handled by the driver. diff --git a/src/mesa/shader/prog_statevars.h b/src/mesa/shader/prog_statevars.h index 20643ca7947..72e51f40314 100644 --- a/src/mesa/shader/prog_statevars.h +++ b/src/mesa/shader/prog_statevars.h @@ -104,6 +104,7 @@ typedef enum gl_state_index_ { STATE_LOCAL, STATE_INTERNAL, /* Mesa additions */ + STATE_CURRENT_ATTRIB, /* ctx->Current vertex attrib value */ STATE_NORMAL_SCALE, STATE_TEXRECT_SCALE, STATE_FOG_PARAMS_OPTIMIZED, /* for faster fog calc */ diff --git a/src/mesa/shader/program.c b/src/mesa/shader/program.c index f120c20bdf2..37962f0e9bb 100644 --- a/src/mesa/shader/program.c +++ b/src/mesa/shader/program.c @@ -372,7 +372,11 @@ _mesa_reference_program(GLcontext *ctx, assert(ptr); if (*ptr && prog) { /* sanity check */ - ASSERT((*ptr)->Target == prog->Target); + if ((*ptr)->Target == GL_VERTEX_PROGRAM_ARB) + ASSERT(prog->Target == GL_VERTEX_PROGRAM_ARB); + else if ((*ptr)->Target == GL_FRAGMENT_PROGRAM_ARB) + ASSERT(prog->Target == GL_FRAGMENT_PROGRAM_ARB || + prog->Target == GL_FRAGMENT_PROGRAM_NV); } if (*ptr == prog) { return; /* no change */ @@ -686,17 +690,47 @@ _mesa_combine_programs(GLcontext *ctx, if (newProg->Target == GL_FRAGMENT_PROGRAM_ARB) { struct gl_fragment_program *fprogA, *fprogB, *newFprog; + GLbitfield progB_inputsRead = progB->InputsRead; + GLint progB_colorFile, progB_colorIndex; + fprogA = (struct gl_fragment_program *) progA; fprogB = (struct gl_fragment_program *) progB; newFprog = (struct gl_fragment_program *) newProg; newFprog->UsesKill = fprogA->UsesKill || fprogB->UsesKill; + /* We'll do a search and replace for instances + * of progB_colorFile/progB_colorIndex below... + */ + progB_colorFile = PROGRAM_INPUT; + progB_colorIndex = FRAG_ATTRIB_COL0; + + /* + * The fragment program may get color from a state var rather than + * a fragment input (vertex output) if it's constant. + * See the texenvprogram.c code. + * So, search the program's parameter list now to see if the program + * gets color from a state var instead of a conventional fragment + * input register. + */ + for (i = 0; i < progB->Parameters->NumParameters; i++) { + struct gl_program_parameter *p = &progB->Parameters->Parameters[i]; + if (p->Type == PROGRAM_STATE_VAR && + p->StateIndexes[0] == STATE_INTERNAL && + p->StateIndexes[1] == STATE_CURRENT_ATTRIB && + p->StateIndexes[2] == VERT_ATTRIB_COLOR0) { + progB_inputsRead |= FRAG_BIT_COL0; + progB_colorFile = PROGRAM_STATE_VAR; + progB_colorIndex = i; + break; + } + } + /* Connect color outputs of fprogA to color inputs of fprogB, via a * new temporary register. */ if ((progA->OutputsWritten & (1 << FRAG_RESULT_COLR)) && - (progB->InputsRead & (1 << FRAG_ATTRIB_COL0))) { + (progB_inputsRead & FRAG_BIT_COL0)) { GLint tempReg = _mesa_find_free_register(newProg, PROGRAM_TEMPORARY); if (tempReg < 0) { _mesa_problem(ctx, "No free temp regs found in " @@ -707,13 +741,14 @@ _mesa_combine_programs(GLcontext *ctx, replace_registers(newInst, lenA, PROGRAM_OUTPUT, FRAG_RESULT_COLR, PROGRAM_TEMPORARY, tempReg); - /* replace reads from input.color[0] with tempReg */ + /* replace reads from the input color with tempReg */ replace_registers(newInst + lenA, lenB, - PROGRAM_INPUT, FRAG_ATTRIB_COL0, - PROGRAM_TEMPORARY, tempReg); + progB_colorFile, progB_colorIndex, /* search for */ + PROGRAM_TEMPORARY, tempReg /* replace with */ ); } - inputsB = progB->InputsRead; + /* compute combined program's InputsRead */ + inputsB = progB_inputsRead; if (progA->OutputsWritten & (1 << FRAG_RESULT_COLR)) { inputsB &= ~(1 << FRAG_ATTRIB_COL0); } diff --git a/src/mesa/shader/shader_api.c b/src/mesa/shader/shader_api.c index 854f8bfdaa6..266ecc4ef21 100644 --- a/src/mesa/shader/shader_api.c +++ b/src/mesa/shader/shader_api.c @@ -47,7 +47,7 @@ #include "shader/shader_api.h" #include "shader/slang/slang_compile.h" #include "shader/slang/slang_link.h" - +#include "glapi/dispatch.h" #ifndef GL_PROGRAM_BINARY_LENGTH_OES @@ -455,7 +455,13 @@ _mesa_attach_shader(GLcontext *ctx, GLuint program, GLuint shader) n = shProg->NumShaders; for (i = 0; i < n; i++) { if (shProg->Shaders[i] == sh) { - /* already attached */ + /* The shader is already attched to this program. The + * GL_ARB_shader_objects spec says: + * + * "The error INVALID_OPERATION is generated by AttachObjectARB + * if <obj> is already attached to <containerObj>." + */ + _mesa_error(ctx, GL_INVALID_OPERATION, "glAttachShader"); return; } } @@ -919,24 +925,15 @@ _mesa_get_attached_shaders(GLcontext *ctx, GLuint program, GLsizei maxCount, static GLuint _mesa_get_handle(GLcontext *ctx, GLenum pname) { -#if 0 - GET_CURRENT_CONTEXT(ctx); - - switch (pname) { - case GL_PROGRAM_OBJECT_ARB: - { - struct gl2_program_intf **pro = ctx->Shader.CurrentProgram; - - if (pro != NULL) - return (**pro)._container._generic. - GetName((struct gl2_generic_intf **) (pro)); - } - break; - default: + GLint handle = 0; + + if (pname == GL_PROGRAM_OBJECT_ARB) { + CALL_GetIntegerv(ctx->Exec, (GL_CURRENT_PROGRAM, &handle)); + } else { _mesa_error(ctx, GL_INVALID_ENUM, "glGetHandleARB"); } -#endif - return 0; + + return handle; } diff --git a/src/mesa/shader/slang/slang_emit.c b/src/mesa/shader/slang/slang_emit.c index 9e8daa10517..f3c3fa6c5b3 100644 --- a/src/mesa/shader/slang/slang_emit.c +++ b/src/mesa/shader/slang/slang_emit.c @@ -1579,13 +1579,17 @@ emit_array_element(slang_emit_info *emitInfo, slang_ir_node *n) else { /* Variable array index */ struct prog_instruction *inst; + slang_ir_storage dstStore = *n->Store; /* do codegen for array index expression */ emit(emitInfo, n->Children[1]); inst = new_instruction(emitInfo, OPCODE_ARL); - storage_to_dst_reg(&inst->DstReg, n->Store, n->Writemask); + if (dstStore.Size > 4) + dstStore.Size = 4; /* only emit one instruction */ + + storage_to_dst_reg(&inst->DstReg, &dstStore, n->Writemask); storage_to_src_reg(&inst->SrcReg[0], n->Children[1]->Store); inst->DstReg.File = PROGRAM_ADDRESS; diff --git a/src/mesa/shader/slang/slang_link.c b/src/mesa/shader/slang/slang_link.c index d884be2a75d..00e89537685 100644 --- a/src/mesa/shader/slang/slang_link.c +++ b/src/mesa/shader/slang/slang_link.c @@ -408,7 +408,7 @@ _slang_update_inputs_outputs(struct gl_program *prog) } } else if (inst->SrcReg[j].File == PROGRAM_ADDRESS) { - maxAddrReg = MAX2(maxAddrReg, inst->SrcReg[j].Index + 1); + maxAddrReg = MAX2(maxAddrReg, (GLuint) (inst->SrcReg[j].Index + 1)); } } if (inst->DstReg.File == PROGRAM_OUTPUT) { diff --git a/src/mesa/state_tracker/st_atom_rasterizer.c b/src/mesa/state_tracker/st_atom_rasterizer.c index fc47896c242..5eef4ebe92e 100644 --- a/src/mesa/state_tracker/st_atom_rasterizer.c +++ b/src/mesa/state_tracker/st_atom_rasterizer.c @@ -215,6 +215,9 @@ static void update_raster_state( struct st_context *st ) raster->sprite_coord_mode[i] = PIPE_SPRITE_COORD_NONE; } } + + /* ST_NEW_VERTEX_PROGRAM + */ if (vertProg) { if (vertProg->Base.Id == 0) { if (vertProg->Base.OutputsWritten & (1 << VERT_RESULT_PSIZ)) { @@ -277,7 +280,7 @@ const struct st_tracked_state st_update_rasterizer = { _NEW_POLYGON | _NEW_PROGRAM | _NEW_SCISSOR), /* mesa state dependencies*/ - 0, /* state tracker dependencies */ + ST_NEW_VERTEX_PROGRAM, /* state tracker dependencies */ }, update_raster_state /* update function */ }; diff --git a/src/mesa/state_tracker/st_cb_texture.c b/src/mesa/state_tracker/st_cb_texture.c index e545e00eb2a..acaf1de8823 100644 --- a/src/mesa/state_tracker/st_cb_texture.c +++ b/src/mesa/state_tracker/st_cb_texture.c @@ -26,7 +26,7 @@ **************************************************************************/ #include "main/imports.h" -#if FEATURE_convolution +#if FEATURE_convolve #include "main/convolve.h" #endif #include "main/enums.h" @@ -409,7 +409,7 @@ st_TexImage(GLcontext * ctx, stImage->face = _mesa_tex_target_to_face(target); stImage->level = level; -#if FEATURE_convolution +#if FEATURE_convolve if (ctx->_ImageTransferState & IMAGE_CONVOLUTION_BIT) { _mesa_adjust_image_for_convolution(ctx, dims, &postConvWidth, &postConvHeight); diff --git a/src/mesa/state_tracker/st_draw.c b/src/mesa/state_tracker/st_draw.c index bdf8648ef7c..61949a93884 100644 --- a/src/mesa/state_tracker/st_draw.c +++ b/src/mesa/state_tracker/st_draw.c @@ -525,6 +525,8 @@ st_draw_vbo(GLcontext *ctx, vbuffer, velements); num_vbuffers = 1; num_velements = vp->num_inputs; + if (num_velements == 0) + num_vbuffers = 0; } else { /*printf("Draw non-interleaved\n");*/ diff --git a/src/mesa/swrast/s_depth.c b/src/mesa/swrast/s_depth.c index 293eb8628e6..26e23f02d59 100644 --- a/src/mesa/swrast/s_depth.c +++ b/src/mesa/swrast/s_depth.c @@ -1,8 +1,8 @@ /* * Mesa 3-D graphics library - * Version: 6.5.1 + * Version: 7.2.1 * - * Copyright (C) 1999-2006 Brian Paul All Rights Reserved. + * Copyright (C) 1999-2008 Brian Paul All Rights Reserved. * * Permission is hereby granted, free of charge, to any person obtaining a * copy of this software and associated documentation files (the "Software"), @@ -534,15 +534,15 @@ depth_test_span( GLcontext *ctx, SWspan *span) if (rb->DataType == GL_UNSIGNED_SHORT) { GLushort zbuffer[MAX_WIDTH]; rb->GetRow(ctx, rb, count, x, y, zbuffer); - passed = depth_test_span16(ctx, count, zbuffer, zValues, mask ); - rb->PutRow(ctx, rb, count, x, y, zbuffer, NULL); + passed = depth_test_span16(ctx, count, zbuffer, zValues, mask); + rb->PutRow(ctx, rb, count, x, y, zbuffer, mask); } else { GLuint zbuffer[MAX_WIDTH]; ASSERT(rb->DataType == GL_UNSIGNED_INT); rb->GetRow(ctx, rb, count, x, y, zbuffer); - passed = depth_test_span32(ctx, count, zbuffer, zValues, mask ); - rb->PutRow(ctx, rb, count, x, y, zbuffer, NULL); + passed = depth_test_span32(ctx, count, zbuffer, zValues, mask); + rb->PutRow(ctx, rb, count, x, y, zbuffer, mask); } } @@ -1080,15 +1080,15 @@ depth_test_pixels( GLcontext *ctx, SWspan *span ) if (rb->DataType == GL_UNSIGNED_SHORT) { GLushort zbuffer[MAX_WIDTH]; _swrast_get_values(ctx, rb, count, x, y, zbuffer, sizeof(GLushort)); - depth_test_span16(ctx, count, zbuffer, z, mask ); - rb->PutValues(ctx, rb, count, x, y, zbuffer, NULL); + depth_test_span16(ctx, count, zbuffer, z, mask); + rb->PutValues(ctx, rb, count, x, y, zbuffer, mask); } else { GLuint zbuffer[MAX_WIDTH]; ASSERT(rb->DataType == GL_UNSIGNED_INT); _swrast_get_values(ctx, rb, count, x, y, zbuffer, sizeof(GLuint)); - depth_test_span32(ctx, count, zbuffer, z, mask ); - rb->PutValues(ctx, rb, count, x, y, zbuffer, NULL); + depth_test_span32(ctx, count, zbuffer, z, mask); + rb->PutValues(ctx, rb, count, x, y, zbuffer, mask); } } diff --git a/src/mesa/swrast/s_points.c b/src/mesa/swrast/s_points.c index 846c485f15a..61ff4d0b84c 100644 --- a/src/mesa/swrast/s_points.c +++ b/src/mesa/swrast/s_points.c @@ -129,13 +129,13 @@ sprite_point(GLcontext *ctx, const SWvertex *vert) s = 0.0; dsdx = 1.0 / size; if (ctx->Point.SpriteOrigin == GL_LOWER_LEFT) { - t0 = 0.0; dtdy = 1.0 / size; + t0 = 0.5 * dtdy; } else { /* GL_UPPER_LEFT */ - t0 = 1.0; dtdy = -1.0 / size; + t0 = 1.0 + 0.5 * dtdy; } ATTRIB_LOOP_BEGIN diff --git a/src/mesa/swrast/s_texcombine.c b/src/mesa/swrast/s_texcombine.c index 4e3d329075c..632d650007e 100644 --- a/src/mesa/swrast/s_texcombine.c +++ b/src/mesa/swrast/s_texcombine.c @@ -26,6 +26,7 @@ #include "main/glheader.h" #include "main/context.h" #include "main/colormac.h" +#include "main/image.h" #include "main/imports.h" #include "main/macros.h" #include "main/pixel.h" diff --git a/src/mesa/swrast/s_texstore.c b/src/mesa/swrast/s_texstore.c index 15d52aa587c..16b00b9fa1c 100644 --- a/src/mesa/swrast/s_texstore.c +++ b/src/mesa/swrast/s_texstore.c @@ -216,9 +216,9 @@ is_depth_format(GLenum format) { switch (format) { case GL_DEPTH_COMPONENT: - case GL_DEPTH_COMPONENT16_SGIX: - case GL_DEPTH_COMPONENT24_SGIX: - case GL_DEPTH_COMPONENT32_SGIX: + case GL_DEPTH_COMPONENT16: + case GL_DEPTH_COMPONENT24: + case GL_DEPTH_COMPONENT32: return GL_TRUE; default: return GL_FALSE; diff --git a/src/mesa/swrast_setup/ss_context.c b/src/mesa/swrast_setup/ss_context.c index f4d90c514bb..61172f9979b 100644 --- a/src/mesa/swrast_setup/ss_context.c +++ b/src/mesa/swrast_setup/ss_context.c @@ -112,22 +112,25 @@ setup_vertex_format(GLcontext *ctx) { TNLcontext *tnl = TNL_CONTEXT(ctx); SScontext *swsetup = SWSETUP_CONTEXT(ctx); + GLboolean intColors = !ctx->FragmentProgram._Current + && !ctx->ATIFragmentShader._Enabled + && ctx->RenderMode == GL_RENDER + && CHAN_TYPE == GL_UNSIGNED_BYTE; - if (!RENDERINPUTS_EQUAL(tnl->render_inputs_bitset, + if (intColors != swsetup->intColors || + !RENDERINPUTS_EQUAL(tnl->render_inputs_bitset, swsetup->last_index_bitset)) { DECLARE_RENDERINPUTS(index_bitset); struct tnl_attr_map map[_TNL_ATTRIB_MAX]; int i, e = 0; + swsetup->intColors = intColors; + RENDERINPUTS_COPY( index_bitset, tnl->render_inputs_bitset ); EMIT_ATTR( _TNL_ATTRIB_POS, EMIT_4F_VIEWPORT, attrib[FRAG_ATTRIB_WPOS] ); if (RENDERINPUTS_TEST( index_bitset, _TNL_ATTRIB_COLOR0 )) { - swsetup->intColors = !ctx->FragmentProgram._Current - && !ctx->ATIFragmentShader._Enabled - && ctx->RenderMode == GL_RENDER - && CHAN_TYPE == GL_UNSIGNED_BYTE; if (swsetup->intColors) EMIT_ATTR( _TNL_ATTRIB_COLOR0, EMIT_4CHAN_4F_RGBA, color ); else diff --git a/src/mesa/tnl/descrip.mms b/src/mesa/tnl/descrip.mms index f77f672dc8f..25dd1aecb13 100644 --- a/src/mesa/tnl/descrip.mms +++ b/src/mesa/tnl/descrip.mms @@ -1,6 +1,6 @@ # Makefile for core library for VMS # contributed by Jouk Jansen [email protected] -# Last revision : 30 November 2007 +# Last revision : 39 September 2008 .first define gl [---.include.gl] @@ -27,13 +27,13 @@ SOURCES = t_context.c t_draw.c \ t_pipeline.c t_vb_fog.c \ t_vb_light.c t_vb_normals.c t_vb_points.c t_vb_program.c \ t_vb_render.c t_vb_texgen.c t_vb_texmat.c t_vb_vertex.c \ - t_vertex.c \ + t_vertex.c t_rasterpos.c\ t_vertex_generic.c t_vp_build.c OBJECTS = t_context.obj,t_draw.obj,\ t_pipeline.obj,t_vb_fog.obj,t_vb_light.obj,t_vb_normals.obj,\ t_vb_points.obj,t_vb_program.obj,t_vb_render.obj,t_vb_texgen.obj,\ - t_vb_texmat.obj,t_vb_vertex.obj,\ + t_vb_texmat.obj,t_vb_vertex.obj,t_rasterpos.obj,\ t_vertex.obj,t_vertex_generic.obj,\ t_vp_build.obj @@ -65,3 +65,4 @@ t_vb_vertex.obj : t_vb_vertex.c t_vertex.obj : t_vertex.c t_vertex_generic.obj : t_vertex_generic.c t_vp_build.obj : t_vp_build.c +t_rasterpos.obj : t_rasterpos.c diff --git a/src/mesa/tnl/t_vertex_generic.c b/src/mesa/tnl/t_vertex_generic.c index db70ad4dad5..f763522f91f 100644 --- a/src/mesa/tnl/t_vertex_generic.c +++ b/src/mesa/tnl/t_vertex_generic.c @@ -34,6 +34,12 @@ #include "t_vertex.h" +#if 0 +#define DEBUG_INSERT printf("%s\n", __FUNCTION__) +#else +#define DEBUG_INSERT +#endif + /* * These functions take the NDC coordinates pointed to by 'in', apply the @@ -45,7 +51,7 @@ static INLINE void insert_4f_viewport_4( const struct tnl_clipspace_attr *a, GLu { GLfloat *out = (GLfloat *)v; const GLfloat * const vp = a->vp; - + DEBUG_INSERT; out[0] = vp[0] * in[0] + vp[12]; out[1] = vp[5] * in[1] + vp[13]; out[2] = vp[10] * in[2] + vp[14]; @@ -57,7 +63,7 @@ static INLINE void insert_4f_viewport_3( const struct tnl_clipspace_attr *a, GLu { GLfloat *out = (GLfloat *)v; const GLfloat * const vp = a->vp; - + DEBUG_INSERT; out[0] = vp[0] * in[0] + vp[12]; out[1] = vp[5] * in[1] + vp[13]; out[2] = vp[10] * in[2] + vp[14]; @@ -69,7 +75,7 @@ static INLINE void insert_4f_viewport_2( const struct tnl_clipspace_attr *a, GLu { GLfloat *out = (GLfloat *)v; const GLfloat * const vp = a->vp; - + DEBUG_INSERT; out[0] = vp[0] * in[0] + vp[12]; out[1] = vp[5] * in[1] + vp[13]; out[2] = vp[14]; @@ -81,7 +87,7 @@ static INLINE void insert_4f_viewport_1( const struct tnl_clipspace_attr *a, GLu { GLfloat *out = (GLfloat *)v; const GLfloat * const vp = a->vp; - + DEBUG_INSERT; out[0] = vp[0] * in[0] + vp[12]; out[1] = vp[13]; out[2] = vp[14]; @@ -93,7 +99,7 @@ static INLINE void insert_3f_viewport_3( const struct tnl_clipspace_attr *a, GLu { GLfloat *out = (GLfloat *)v; const GLfloat * const vp = a->vp; - + DEBUG_INSERT; out[0] = vp[0] * in[0] + vp[12]; out[1] = vp[5] * in[1] + vp[13]; out[2] = vp[10] * in[2] + vp[14]; @@ -104,7 +110,7 @@ static INLINE void insert_3f_viewport_2( const struct tnl_clipspace_attr *a, GLu { GLfloat *out = (GLfloat *)v; const GLfloat * const vp = a->vp; - + DEBUG_INSERT; out[0] = vp[0] * in[0] + vp[12]; out[1] = vp[5] * in[1] + vp[13]; out[2] = vp[10] * in[2] + vp[14]; @@ -115,7 +121,7 @@ static INLINE void insert_3f_viewport_1( const struct tnl_clipspace_attr *a, GLu { GLfloat *out = (GLfloat *)v; const GLfloat * const vp = a->vp; - + DEBUG_INSERT; out[0] = vp[0] * in[0] + vp[12]; out[1] = vp[13]; out[2] = vp[14]; @@ -126,7 +132,7 @@ static INLINE void insert_2f_viewport_2( const struct tnl_clipspace_attr *a, GLu { GLfloat *out = (GLfloat *)v; const GLfloat * const vp = a->vp; - + DEBUG_INSERT; out[0] = vp[0] * in[0] + vp[12]; out[1] = vp[5] * in[1] + vp[13]; } @@ -136,7 +142,7 @@ static INLINE void insert_2f_viewport_1( const struct tnl_clipspace_attr *a, GLu { GLfloat *out = (GLfloat *)v; const GLfloat * const vp = a->vp; - + DEBUG_INSERT; out[0] = vp[0] * in[0] + vp[12]; out[1] = vp[13]; } @@ -150,7 +156,7 @@ static INLINE void insert_4f_4( const struct tnl_clipspace_attr *a, GLubyte *v, { GLfloat *out = (GLfloat *)(v); (void) a; - + DEBUG_INSERT; out[0] = in[0]; out[1] = in[1]; out[2] = in[2]; @@ -161,7 +167,7 @@ static INLINE void insert_4f_3( const struct tnl_clipspace_attr *a, GLubyte *v, { GLfloat *out = (GLfloat *)(v); (void) a; - + DEBUG_INSERT; out[0] = in[0]; out[1] = in[1]; out[2] = in[2]; @@ -172,7 +178,7 @@ static INLINE void insert_4f_2( const struct tnl_clipspace_attr *a, GLubyte *v, { GLfloat *out = (GLfloat *)(v); (void) a; - + DEBUG_INSERT; out[0] = in[0]; out[1] = in[1]; out[2] = 0; @@ -183,7 +189,7 @@ static INLINE void insert_4f_1( const struct tnl_clipspace_attr *a, GLubyte *v, { GLfloat *out = (GLfloat *)(v); (void) a; - + DEBUG_INSERT; out[0] = in[0]; out[1] = 0; out[2] = 0; @@ -194,7 +200,7 @@ static INLINE void insert_3f_xyw_4( const struct tnl_clipspace_attr *a, GLubyte { GLfloat *out = (GLfloat *)(v); (void) a; - + DEBUG_INSERT; out[0] = in[0]; out[1] = in[1]; out[2] = in[3]; @@ -203,6 +209,7 @@ static INLINE void insert_3f_xyw_4( const struct tnl_clipspace_attr *a, GLubyte static INLINE void insert_3f_xyw_err( const struct tnl_clipspace_attr *a, GLubyte *v, const GLfloat *in ) { (void) a; (void) v; (void) in; + DEBUG_INSERT; _mesa_exit(1); } @@ -210,7 +217,7 @@ static INLINE void insert_3f_3( const struct tnl_clipspace_attr *a, GLubyte *v, { GLfloat *out = (GLfloat *)(v); (void) a; - + DEBUG_INSERT; out[0] = in[0]; out[1] = in[1]; out[2] = in[2]; @@ -220,7 +227,7 @@ static INLINE void insert_3f_2( const struct tnl_clipspace_attr *a, GLubyte *v, { GLfloat *out = (GLfloat *)(v); (void) a; - + DEBUG_INSERT; out[0] = in[0]; out[1] = in[1]; out[2] = 0; @@ -230,7 +237,7 @@ static INLINE void insert_3f_1( const struct tnl_clipspace_attr *a, GLubyte *v, { GLfloat *out = (GLfloat *)(v); (void) a; - + DEBUG_INSERT; out[0] = in[0]; out[1] = 0; out[2] = 0; @@ -241,7 +248,7 @@ static INLINE void insert_2f_2( const struct tnl_clipspace_attr *a, GLubyte *v, { GLfloat *out = (GLfloat *)(v); (void) a; - + DEBUG_INSERT; out[0] = in[0]; out[1] = in[1]; } @@ -250,7 +257,7 @@ static INLINE void insert_2f_1( const struct tnl_clipspace_attr *a, GLubyte *v, { GLfloat *out = (GLfloat *)(v); (void) a; - + DEBUG_INSERT; out[0] = in[0]; out[1] = 0; } @@ -259,12 +266,13 @@ static INLINE void insert_1f_1( const struct tnl_clipspace_attr *a, GLubyte *v, { GLfloat *out = (GLfloat *)(v); (void) a; - + DEBUG_INSERT; out[0] = in[0]; } static INLINE void insert_null( const struct tnl_clipspace_attr *a, GLubyte *v, const GLfloat *in ) { + DEBUG_INSERT; (void) a; (void) v; (void) in; } @@ -272,6 +280,7 @@ static INLINE void insert_4chan_4f_rgba_4( const struct tnl_clipspace_attr *a, G const GLfloat *in ) { GLchan *c = (GLchan *)v; + DEBUG_INSERT; (void) a; UNCLAMPED_FLOAT_TO_CHAN(c[0], in[0]); UNCLAMPED_FLOAT_TO_CHAN(c[1], in[1]); @@ -283,6 +292,7 @@ static INLINE void insert_4chan_4f_rgba_3( const struct tnl_clipspace_attr *a, G const GLfloat *in ) { GLchan *c = (GLchan *)v; + DEBUG_INSERT; (void) a; UNCLAMPED_FLOAT_TO_CHAN(c[0], in[0]); UNCLAMPED_FLOAT_TO_CHAN(c[1], in[1]); @@ -294,6 +304,7 @@ static INLINE void insert_4chan_4f_rgba_2( const struct tnl_clipspace_attr *a, G const GLfloat *in ) { GLchan *c = (GLchan *)v; + DEBUG_INSERT; (void) a; UNCLAMPED_FLOAT_TO_CHAN(c[0], in[0]); UNCLAMPED_FLOAT_TO_CHAN(c[1], in[1]); @@ -305,6 +316,7 @@ static INLINE void insert_4chan_4f_rgba_1( const struct tnl_clipspace_attr *a, G const GLfloat *in ) { GLchan *c = (GLchan *)v; + DEBUG_INSERT; (void) a; UNCLAMPED_FLOAT_TO_CHAN(c[0], in[0]); c[1] = 0; @@ -315,6 +327,7 @@ static INLINE void insert_4chan_4f_rgba_1( const struct tnl_clipspace_attr *a, G static INLINE void insert_4ub_4f_rgba_4( const struct tnl_clipspace_attr *a, GLubyte *v, const GLfloat *in ) { + DEBUG_INSERT; (void) a; UNCLAMPED_FLOAT_TO_UBYTE(v[0], in[0]); UNCLAMPED_FLOAT_TO_UBYTE(v[1], in[1]); @@ -325,6 +338,7 @@ static INLINE void insert_4ub_4f_rgba_4( const struct tnl_clipspace_attr *a, GLu static INLINE void insert_4ub_4f_rgba_3( const struct tnl_clipspace_attr *a, GLubyte *v, const GLfloat *in ) { + DEBUG_INSERT; (void) a; UNCLAMPED_FLOAT_TO_UBYTE(v[0], in[0]); UNCLAMPED_FLOAT_TO_UBYTE(v[1], in[1]); @@ -335,6 +349,7 @@ static INLINE void insert_4ub_4f_rgba_3( const struct tnl_clipspace_attr *a, GLu static INLINE void insert_4ub_4f_rgba_2( const struct tnl_clipspace_attr *a, GLubyte *v, const GLfloat *in ) { + DEBUG_INSERT; (void) a; UNCLAMPED_FLOAT_TO_UBYTE(v[0], in[0]); UNCLAMPED_FLOAT_TO_UBYTE(v[1], in[1]); @@ -345,6 +360,7 @@ static INLINE void insert_4ub_4f_rgba_2( const struct tnl_clipspace_attr *a, GLu static INLINE void insert_4ub_4f_rgba_1( const struct tnl_clipspace_attr *a, GLubyte *v, const GLfloat *in ) { + DEBUG_INSERT; (void) a; UNCLAMPED_FLOAT_TO_UBYTE(v[0], in[0]); v[1] = 0; @@ -355,6 +371,7 @@ static INLINE void insert_4ub_4f_rgba_1( const struct tnl_clipspace_attr *a, GLu static INLINE void insert_4ub_4f_bgra_4( const struct tnl_clipspace_attr *a, GLubyte *v, const GLfloat *in ) { + DEBUG_INSERT; (void) a; UNCLAMPED_FLOAT_TO_UBYTE(v[2], in[0]); UNCLAMPED_FLOAT_TO_UBYTE(v[1], in[1]); @@ -365,6 +382,7 @@ static INLINE void insert_4ub_4f_bgra_4( const struct tnl_clipspace_attr *a, GLu static INLINE void insert_4ub_4f_bgra_3( const struct tnl_clipspace_attr *a, GLubyte *v, const GLfloat *in ) { + DEBUG_INSERT; (void) a; UNCLAMPED_FLOAT_TO_UBYTE(v[2], in[0]); UNCLAMPED_FLOAT_TO_UBYTE(v[1], in[1]); @@ -375,6 +393,7 @@ static INLINE void insert_4ub_4f_bgra_3( const struct tnl_clipspace_attr *a, GLu static INLINE void insert_4ub_4f_bgra_2( const struct tnl_clipspace_attr *a, GLubyte *v, const GLfloat *in ) { + DEBUG_INSERT; (void) a; UNCLAMPED_FLOAT_TO_UBYTE(v[2], in[0]); UNCLAMPED_FLOAT_TO_UBYTE(v[1], in[1]); @@ -385,6 +404,7 @@ static INLINE void insert_4ub_4f_bgra_2( const struct tnl_clipspace_attr *a, GLu static INLINE void insert_4ub_4f_bgra_1( const struct tnl_clipspace_attr *a, GLubyte *v, const GLfloat *in ) { + DEBUG_INSERT; (void) a; UNCLAMPED_FLOAT_TO_UBYTE(v[2], in[0]); v[1] = 0; @@ -395,6 +415,7 @@ static INLINE void insert_4ub_4f_bgra_1( const struct tnl_clipspace_attr *a, GLu static INLINE void insert_4ub_4f_argb_4( const struct tnl_clipspace_attr *a, GLubyte *v, const GLfloat *in ) { + DEBUG_INSERT; (void) a; UNCLAMPED_FLOAT_TO_UBYTE(v[1], in[0]); UNCLAMPED_FLOAT_TO_UBYTE(v[2], in[1]); @@ -405,6 +426,7 @@ static INLINE void insert_4ub_4f_argb_4( const struct tnl_clipspace_attr *a, GLu static INLINE void insert_4ub_4f_argb_3( const struct tnl_clipspace_attr *a, GLubyte *v, const GLfloat *in ) { + DEBUG_INSERT; (void) a; UNCLAMPED_FLOAT_TO_UBYTE(v[1], in[0]); UNCLAMPED_FLOAT_TO_UBYTE(v[2], in[1]); @@ -415,6 +437,7 @@ static INLINE void insert_4ub_4f_argb_3( const struct tnl_clipspace_attr *a, GLu static INLINE void insert_4ub_4f_argb_2( const struct tnl_clipspace_attr *a, GLubyte *v, const GLfloat *in ) { + DEBUG_INSERT; (void) a; UNCLAMPED_FLOAT_TO_UBYTE(v[1], in[0]); UNCLAMPED_FLOAT_TO_UBYTE(v[2], in[1]); @@ -425,6 +448,7 @@ static INLINE void insert_4ub_4f_argb_2( const struct tnl_clipspace_attr *a, GLu static INLINE void insert_4ub_4f_argb_1( const struct tnl_clipspace_attr *a, GLubyte *v, const GLfloat *in ) { + DEBUG_INSERT; (void) a; UNCLAMPED_FLOAT_TO_UBYTE(v[1], in[0]); v[2] = 0x00; @@ -435,6 +459,7 @@ static INLINE void insert_4ub_4f_argb_1( const struct tnl_clipspace_attr *a, GLu static INLINE void insert_4ub_4f_abgr_4( const struct tnl_clipspace_attr *a, GLubyte *v, const GLfloat *in ) { + DEBUG_INSERT; (void) a; UNCLAMPED_FLOAT_TO_UBYTE(v[3], in[0]); UNCLAMPED_FLOAT_TO_UBYTE(v[2], in[1]); @@ -445,6 +470,7 @@ static INLINE void insert_4ub_4f_abgr_4( const struct tnl_clipspace_attr *a, GLu static INLINE void insert_4ub_4f_abgr_3( const struct tnl_clipspace_attr *a, GLubyte *v, const GLfloat *in ) { + DEBUG_INSERT; (void) a; UNCLAMPED_FLOAT_TO_UBYTE(v[3], in[0]); UNCLAMPED_FLOAT_TO_UBYTE(v[2], in[1]); @@ -455,6 +481,7 @@ static INLINE void insert_4ub_4f_abgr_3( const struct tnl_clipspace_attr *a, GLu static INLINE void insert_4ub_4f_abgr_2( const struct tnl_clipspace_attr *a, GLubyte *v, const GLfloat *in ) { + DEBUG_INSERT; (void) a; UNCLAMPED_FLOAT_TO_UBYTE(v[3], in[0]); UNCLAMPED_FLOAT_TO_UBYTE(v[2], in[1]); @@ -465,6 +492,7 @@ static INLINE void insert_4ub_4f_abgr_2( const struct tnl_clipspace_attr *a, GLu static INLINE void insert_4ub_4f_abgr_1( const struct tnl_clipspace_attr *a, GLubyte *v, const GLfloat *in ) { + DEBUG_INSERT; (void) a; UNCLAMPED_FLOAT_TO_UBYTE(v[3], in[0]); v[2] = 0x00; @@ -475,6 +503,7 @@ static INLINE void insert_4ub_4f_abgr_1( const struct tnl_clipspace_attr *a, GLu static INLINE void insert_3ub_3f_rgb_3( const struct tnl_clipspace_attr *a, GLubyte *v, const GLfloat *in ) { + DEBUG_INSERT; (void) a; UNCLAMPED_FLOAT_TO_UBYTE(v[0], in[0]); UNCLAMPED_FLOAT_TO_UBYTE(v[1], in[1]); @@ -484,6 +513,7 @@ static INLINE void insert_3ub_3f_rgb_3( const struct tnl_clipspace_attr *a, GLub static INLINE void insert_3ub_3f_rgb_2( const struct tnl_clipspace_attr *a, GLubyte *v, const GLfloat *in ) { + DEBUG_INSERT; (void) a; UNCLAMPED_FLOAT_TO_UBYTE(v[0], in[0]); UNCLAMPED_FLOAT_TO_UBYTE(v[1], in[1]); @@ -493,6 +523,7 @@ static INLINE void insert_3ub_3f_rgb_2( const struct tnl_clipspace_attr *a, GLub static INLINE void insert_3ub_3f_rgb_1( const struct tnl_clipspace_attr *a, GLubyte *v, const GLfloat *in ) { + DEBUG_INSERT; (void) a; UNCLAMPED_FLOAT_TO_UBYTE(v[0], in[0]); v[1] = 0; @@ -502,6 +533,7 @@ static INLINE void insert_3ub_3f_rgb_1( const struct tnl_clipspace_attr *a, GLub static INLINE void insert_3ub_3f_bgr_3( const struct tnl_clipspace_attr *a, GLubyte *v, const GLfloat *in ) { + DEBUG_INSERT; (void) a; UNCLAMPED_FLOAT_TO_UBYTE(v[2], in[0]); UNCLAMPED_FLOAT_TO_UBYTE(v[1], in[1]); @@ -511,6 +543,7 @@ static INLINE void insert_3ub_3f_bgr_3( const struct tnl_clipspace_attr *a, GLub static INLINE void insert_3ub_3f_bgr_2( const struct tnl_clipspace_attr *a, GLubyte *v, const GLfloat *in ) { + DEBUG_INSERT; (void) a; UNCLAMPED_FLOAT_TO_UBYTE(v[2], in[0]); UNCLAMPED_FLOAT_TO_UBYTE(v[1], in[1]); @@ -520,6 +553,7 @@ static INLINE void insert_3ub_3f_bgr_2( const struct tnl_clipspace_attr *a, GLub static INLINE void insert_3ub_3f_bgr_1( const struct tnl_clipspace_attr *a, GLubyte *v, const GLfloat *in ) { + DEBUG_INSERT; (void) a; UNCLAMPED_FLOAT_TO_UBYTE(v[2], in[0]); v[1] = 0; @@ -530,6 +564,7 @@ static INLINE void insert_3ub_3f_bgr_1( const struct tnl_clipspace_attr *a, GLub static INLINE void insert_1ub_1f_1( const struct tnl_clipspace_attr *a, GLubyte *v, const GLfloat *in ) { + DEBUG_INSERT; (void) a; UNCLAMPED_FLOAT_TO_UBYTE(v[0], in[0]); } @@ -551,6 +586,7 @@ static void extract_4f_viewport( const struct tnl_clipspace_attr *a, GLfloat *ou /* Although included for completeness, the position coordinate is * usually handled differently during clipping. */ + DEBUG_INSERT; out[0] = (in[0] - vp[12]) / vp[0]; out[1] = (in[1] - vp[13]) / vp[5]; out[2] = (in[2] - vp[14]) / vp[10]; @@ -562,7 +598,7 @@ static void extract_3f_viewport( const struct tnl_clipspace_attr *a, GLfloat *ou { const GLfloat *in = (const GLfloat *)v; const GLfloat * const vp = a->vp; - + DEBUG_INSERT; out[0] = (in[0] - vp[12]) / vp[0]; out[1] = (in[1] - vp[13]) / vp[5]; out[2] = (in[2] - vp[14]) / vp[10]; @@ -575,7 +611,7 @@ static void extract_2f_viewport( const struct tnl_clipspace_attr *a, GLfloat *ou { const GLfloat *in = (const GLfloat *)v; const GLfloat * const vp = a->vp; - + DEBUG_INSERT; out[0] = (in[0] - vp[12]) / vp[0]; out[1] = (in[1] - vp[13]) / vp[5]; out[2] = 0; diff --git a/src/mesa/vbo/vbo_exec_api.c b/src/mesa/vbo/vbo_exec_api.c index fdb0c5a9a4d..a6ce26ffed8 100644 --- a/src/mesa/vbo/vbo_exec_api.c +++ b/src/mesa/vbo/vbo_exec_api.c @@ -143,29 +143,37 @@ static void vbo_exec_copy_to_current( struct vbo_exec_context *exec ) for (i = VBO_ATTRIB_POS+1 ; i < VBO_ATTRIB_MAX ; i++) { if (exec->vtx.attrsz[i]) { - GLfloat *current = (GLfloat *)vbo->currval[i].Ptr; - /* Note: the exec->vtx.current[i] pointers point into the * ctx->Current.Attrib and ctx->Light.Material.Attrib arrays. */ - COPY_CLEAN_4V(current, - exec->vtx.attrsz[i], - exec->vtx.attrptr[i]); + GLfloat *current = (GLfloat *)vbo->currval[i].Ptr; + GLfloat tmp[4]; + + COPY_CLEAN_4V(tmp, + exec->vtx.attrsz[i], + exec->vtx.attrptr[i]); + + if (memcmp(current, tmp, sizeof(tmp)) != 0) + { + memcpy(current, tmp, sizeof(tmp)); - /* Given that we explicitly state size here, there is no need - * for the COPY_CLEAN above, could just copy 16 bytes and be - * done. The only problem is when Mesa accesses ctx->Current - * directly. - */ - vbo->currval[i].Size = exec->vtx.attrsz[i]; - - /* This triggers rather too much recalculation of Mesa state - * that doesn't get used (eg light positions). - */ - if (i >= VBO_ATTRIB_MAT_FRONT_AMBIENT && - i <= VBO_ATTRIB_MAT_BACK_INDEXES) - ctx->NewState |= _NEW_LIGHT; + /* Given that we explicitly state size here, there is no need + * for the COPY_CLEAN above, could just copy 16 bytes and be + * done. The only problem is when Mesa accesses ctx->Current + * directly. + */ + vbo->currval[i].Size = exec->vtx.attrsz[i]; + + /* This triggers rather too much recalculation of Mesa state + * that doesn't get used (eg light positions). + */ + if (i >= VBO_ATTRIB_MAT_FRONT_AMBIENT && + i <= VBO_ATTRIB_MAT_BACK_INDEXES) + ctx->NewState |= _NEW_LIGHT; + + ctx->NewState |= _NEW_CURRENT_ATTRIB; + } } } diff --git a/src/mesa/vbo/vbo_exec_array.c b/src/mesa/vbo/vbo_exec_array.c index 0f9d8da3568..8871e10cf60 100644 --- a/src/mesa/vbo/vbo_exec_array.c +++ b/src/mesa/vbo/vbo_exec_array.c @@ -127,6 +127,7 @@ static void recalculate_input_bindings( GLcontext *ctx ) struct vbo_context *vbo = vbo_context(ctx); struct vbo_exec_context *exec = &vbo->exec; const struct gl_client_array **inputs = &exec->array.inputs[0]; + GLbitfield const_inputs = 0x0; GLuint i; exec->array.program_mode = get_program_mode(ctx); @@ -141,19 +142,24 @@ static void recalculate_input_bindings( GLcontext *ctx ) for (i = 0; i <= VERT_ATTRIB_TEX7; i++) { if (exec->array.legacy_array[i]->Enabled) inputs[i] = exec->array.legacy_array[i]; - else + else { inputs[i] = &vbo->legacy_currval[i]; + const_inputs |= 1 << i; + } } for (i = 0; i < MAT_ATTRIB_MAX; i++) { inputs[VERT_ATTRIB_GENERIC0 + i] = &vbo->mat_currval[i]; + const_inputs |= 1 << (VERT_ATTRIB_GENERIC0 + i); } /* Could use just about anything, just to fill in the empty * slots: */ - for (i = MAT_ATTRIB_MAX; i < VERT_ATTRIB_MAX - VERT_ATTRIB_GENERIC0; i++) + for (i = MAT_ATTRIB_MAX; i < VERT_ATTRIB_MAX - VERT_ATTRIB_GENERIC0; i++) { inputs[VERT_ATTRIB_GENERIC0 + i] = &vbo->generic_currval[i]; + const_inputs |= 1 << (VERT_ATTRIB_GENERIC0 + i); + } break; case VP_NV: @@ -166,15 +172,19 @@ static void recalculate_input_bindings( GLcontext *ctx ) inputs[i] = exec->array.generic_array[i]; else if (exec->array.legacy_array[i]->Enabled) inputs[i] = exec->array.legacy_array[i]; - else + else { inputs[i] = &vbo->legacy_currval[i]; + const_inputs |= 1 << i; + } } /* Could use just about anything, just to fill in the empty * slots: */ - for (i = VERT_ATTRIB_GENERIC0; i < VERT_ATTRIB_MAX; i++) + for (i = VERT_ATTRIB_GENERIC0; i < VERT_ATTRIB_MAX; i++) { inputs[i] = &vbo->generic_currval[i - VERT_ATTRIB_GENERIC0]; + const_inputs |= 1 << i; + } break; case VP_ARB: @@ -189,25 +199,34 @@ static void recalculate_input_bindings( GLcontext *ctx ) inputs[0] = exec->array.generic_array[0]; else if (exec->array.legacy_array[0]->Enabled) inputs[0] = exec->array.legacy_array[0]; - else + else { inputs[0] = &vbo->legacy_currval[0]; + const_inputs |= 1 << 0; + } for (i = 1; i <= VERT_ATTRIB_TEX7; i++) { if (exec->array.legacy_array[i]->Enabled) inputs[i] = exec->array.legacy_array[i]; - else + else { inputs[i] = &vbo->legacy_currval[i]; + const_inputs |= 1 << i; + } } for (i = 0; i < 16; i++) { if (exec->array.generic_array[i]->Enabled) inputs[VERT_ATTRIB_GENERIC0 + i] = exec->array.generic_array[i]; - else + else { inputs[VERT_ATTRIB_GENERIC0 + i] = &vbo->generic_currval[i]; + const_inputs |= 1 << (VERT_ATTRIB_GENERIC0 + i); + } + } break; } + + _mesa_set_varying_vp_inputs( ctx, ~const_inputs ); } static void bind_arrays( GLcontext *ctx ) @@ -257,6 +276,11 @@ vbo_exec_DrawArrays(GLenum mode, GLint start, GLsizei count) bind_arrays( ctx ); + /* Again... + */ + if (ctx->NewState) + _mesa_update_state( ctx ); + prim[0].begin = 1; prim[0].end = 1; prim[0].weak = 0; @@ -297,6 +321,9 @@ vbo_exec_DrawRangeElements(GLenum mode, bind_arrays( ctx ); + if (ctx->NewState) + _mesa_update_state( ctx ); + ib.count = count; ib.type = type; ib.obj = ctx->Array.ElementArrayBufferObj; diff --git a/src/mesa/vbo/vbo_exec_draw.c b/src/mesa/vbo/vbo_exec_draw.c index 92356ba9773..5bf3d836db5 100644 --- a/src/mesa/vbo/vbo_exec_draw.c +++ b/src/mesa/vbo/vbo_exec_draw.c @@ -150,6 +150,7 @@ static void vbo_exec_bind_arrays( GLcontext *ctx ) GLubyte *data = exec->vtx.buffer_map; const GLuint *map; GLuint attr; + GLbitfield varying_inputs = 0x0; /* Install the default (ie Current) attributes first, then overlay * all active ones. @@ -211,8 +212,11 @@ static void vbo_exec_bind_arrays( GLcontext *ctx ) arrays[attr]._MaxElement = count; /* ??? */ data += exec->vtx.attrsz[src] * sizeof(GLfloat); + varying_inputs |= 1<<attr; } } + + _mesa_set_varying_vp_inputs( ctx, varying_inputs ); } @@ -242,6 +246,9 @@ void vbo_exec_vtx_flush( struct vbo_exec_context *exec ) */ vbo_exec_bind_arrays( ctx ); + if (ctx->NewState) + _mesa_update_state( ctx ); + /* if using a real VBO, unmap it before drawing */ if (exec->vtx.bufferobj->Name) { ctx->Driver.UnmapBuffer(ctx, target, exec->vtx.bufferobj); diff --git a/src/mesa/vbo/vbo_save_draw.c b/src/mesa/vbo/vbo_save_draw.c index ed82f09958d..0488c5d7182 100644 --- a/src/mesa/vbo/vbo_save_draw.c +++ b/src/mesa/vbo/vbo_save_draw.c @@ -64,18 +64,26 @@ static void _playback_copy_to_current( GLcontext *ctx, for (i = VBO_ATTRIB_POS+1 ; i < VBO_ATTRIB_MAX ; i++) { if (node->attrsz[i]) { GLfloat *current = (GLfloat *)vbo->currval[i].Ptr; + GLfloat tmp[4]; - COPY_CLEAN_4V(current, - node->attrsz[i], - data); + COPY_CLEAN_4V(tmp, + node->attrsz[i], + data); + + if (memcmp(current, tmp, 4 * sizeof(GLfloat)) != 0) + { + memcpy(current, tmp, 4 * sizeof(GLfloat)); - vbo->currval[i].Size = node->attrsz[i]; + vbo->currval[i].Size = node->attrsz[i]; - data += node->attrsz[i]; + if (i >= VBO_ATTRIB_FIRST_MATERIAL && + i <= VBO_ATTRIB_LAST_MATERIAL) + ctx->NewState |= _NEW_LIGHT; + + ctx->NewState |= _NEW_CURRENT_ATTRIB; + } - if (i >= VBO_ATTRIB_FIRST_MATERIAL && - i <= VBO_ATTRIB_LAST_MATERIAL) - ctx->NewState |= _NEW_LIGHT; + data += node->attrsz[i]; } } @@ -110,6 +118,7 @@ static void vbo_bind_vertex_list( GLcontext *ctx, GLuint data = node->buffer_offset; const GLuint *map; GLuint attr; + GLbitfield varying_inputs = 0x0; /* Install the default (ie Current) attributes first, then overlay * all active ones. @@ -159,8 +168,11 @@ static void vbo_bind_vertex_list( GLcontext *ctx, assert(arrays[attr].BufferObj->Name); data += node->attrsz[src] * sizeof(GLfloat); + varying_inputs |= 1<<attr; } } + + _mesa_set_varying_vp_inputs( ctx, varying_inputs ); } static void vbo_save_loopback_vertex_list( GLcontext *ctx, @@ -229,6 +241,11 @@ void vbo_save_playback_vertex_list( GLcontext *ctx, void *data ) vbo_bind_vertex_list( ctx, node ); + /* Again... + */ + if (ctx->NewState) + _mesa_update_state( ctx ); + vbo_context(ctx)->draw_prims( ctx, save->inputs, node->prim, diff --git a/src/mesa/x86-64/x86-64.c b/src/mesa/x86-64/x86-64.c index 9ec43c841d3..96f8da87f05 100644 --- a/src/mesa/x86-64/x86-64.c +++ b/src/mesa/x86-64/x86-64.c @@ -41,7 +41,10 @@ #include "math/m_debug.h" #endif +extern void _mesa_x86_64_cpuid(unsigned int *regs); + DECLARE_XFORM_GROUP( x86_64, 4 ) +DECLARE_XFORM_GROUP( 3dnow, 4 ) #else /* just to silence warning below */ @@ -81,6 +84,7 @@ static void message( const char *msg ) void _mesa_init_all_x86_64_transform_asm(void) { #ifdef USE_X86_64_ASM + unsigned int regs[4]; if ( _mesa_getenv( "MESA_NO_ASM" ) ) { return; @@ -88,24 +92,32 @@ void _mesa_init_all_x86_64_transform_asm(void) message("Initializing x86-64 optimizations\n"); - ASSIGN_XFORM_GROUP( x86_64, 4 ); - /* _mesa_transform_tab[4][MATRIX_GENERAL] = _mesa_x86_64_transform_points4_general; _mesa_transform_tab[4][MATRIX_IDENTITY] = _mesa_x86_64_transform_points4_identity; _mesa_transform_tab[4][MATRIX_3D] = _mesa_x86_64_transform_points4_3d; - _mesa_transform_tab[4][MATRIX_3D_NO_ROT] = - _mesa_x86_64_transform_points4_3d_no_rot; - _mesa_transform_tab[4][MATRIX_PERSPECTIVE] = - _mesa_x86_64_transform_points4_perspective; - _mesa_transform_tab[4][MATRIX_2D_NO_ROT] = - _mesa_x86_64_transform_points4_2d_no_rot; - _mesa_transform_tab[4][MATRIX_2D] = - _mesa_x86_64_transform_points4_2d; - */ + + regs[0] = 0x80000001; + regs[1] = 0x00000000; + regs[2] = 0x00000000; + regs[3] = 0x00000000; + _mesa_x86_64_cpuid(regs); + if (regs[3] & (1U << 31)) { + message("3Dnow! detected\n"); + _mesa_transform_tab[4][MATRIX_3D_NO_ROT] = + _mesa_3dnow_transform_points4_3d_no_rot; + _mesa_transform_tab[4][MATRIX_PERSPECTIVE] = + _mesa_3dnow_transform_points4_perspective; + _mesa_transform_tab[4][MATRIX_2D_NO_ROT] = + _mesa_3dnow_transform_points4_2d_no_rot; + _mesa_transform_tab[4][MATRIX_2D] = + _mesa_3dnow_transform_points4_2d; + + } + #ifdef DEBUG_MATH _math_test_all_transform_functions("x86_64"); diff --git a/src/mesa/x86-64/xform4.S b/src/mesa/x86-64/xform4.S index 3f9c9d56ab7..805969127db 100644 --- a/src/mesa/x86-64/xform4.S +++ b/src/mesa/x86-64/xform4.S @@ -29,7 +29,22 @@ .text .align 16 +.globl _mesa_x86_64_cpuid +_mesa_x86_64_cpuid: + pushq %rbx + movl (%rdi), %eax + movl 8(%rdi), %ecx + + cpuid + + movl %ebx, 4(%rdi) + movl %eax, (%rdi) + movl %ecx, 8(%rdi) + movl %edx, 12(%rdi) + popq %rbx + ret +.align 16 .globl _mesa_x86_64_transform_points4_general _mesa_x86_64_transform_points4_general: /* @@ -204,8 +219,8 @@ p4_identity_done: .align 16 -.globl _mesa_x86_64_transform_points4_3d_no_rot -_mesa_x86_64_transform_points4_3d_no_rot: +.globl _mesa_3dnow_transform_points4_3d_no_rot +_mesa_3dnow_transform_points4_3d_no_rot: movl V4F_COUNT(%rdx), %ecx /* count */ movzx V4F_STRIDE(%rdx), %eax /* stride */ @@ -268,8 +283,8 @@ p4_3d_no_rot_done: .align 16 -.globl _mesa_x86_64_transform_points4_perspective -_mesa_x86_64_transform_points4_perspective: +.globl _mesa_3dnow_transform_points4_perspective +_mesa_3dnow_transform_points4_perspective: movl V4F_COUNT(%rdx), %ecx /* count */ movzx V4F_STRIDE(%rdx), %eax /* stride */ @@ -334,8 +349,8 @@ p4_perspective_done: ret .align 16 -.globl _mesa_x86_64_transform_points4_2d_no_rot -_mesa_x86_64_transform_points4_2d_no_rot: +.globl _mesa_3dnow_transform_points4_2d_no_rot +_mesa_3dnow_transform_points4_2d_no_rot: movl V4F_COUNT(%rdx), %ecx /* count */ movzx V4F_STRIDE(%rdx), %eax /* stride */ @@ -389,8 +404,8 @@ p4_2d_no_rot_done: .align 16 -.globl _mesa_x86_64_transform_points4_2d -_mesa_x86_64_transform_points4_2d: +.globl _mesa_3dnow_transform_points4_2d +_mesa_3dnow_transform_points4_2d: movl V4F_COUNT(%rdx), %ecx /* count */ movzx V4F_STRIDE(%rdx), %eax /* stride */ |