diff options
Diffstat (limited to 'src/gallium/drivers/r300')
-rw-r--r-- | src/gallium/drivers/r300/r300_context.c | 97 | ||||
-rw-r--r-- | src/gallium/drivers/r300/r300_context.h | 10 | ||||
-rw-r--r-- | src/gallium/drivers/r300/r300_emit.c | 42 | ||||
-rw-r--r-- | src/gallium/drivers/r300/r300_flush.c | 5 | ||||
-rw-r--r-- | src/gallium/drivers/r300/r300_fs.c | 23 | ||||
-rw-r--r-- | src/gallium/drivers/r300/r300_hyperz.c | 28 | ||||
-rw-r--r-- | src/gallium/drivers/r300/r300_hyperz.h | 2 | ||||
-rw-r--r-- | src/gallium/drivers/r300/r300_reg.h | 1 | ||||
-rw-r--r-- | src/gallium/drivers/r300/r300_render.c | 55 | ||||
-rw-r--r-- | src/gallium/drivers/r300/r300_state.c | 64 | ||||
-rw-r--r-- | src/gallium/drivers/r300/r300_state_derived.c | 26 | ||||
-rw-r--r-- | src/gallium/drivers/r300/r300_texture.c | 49 | ||||
-rw-r--r-- | src/gallium/drivers/r300/r300_texture.h | 7 | ||||
-rw-r--r-- | src/gallium/drivers/r300/r300_texture_desc.c | 6 | ||||
-rw-r--r-- | src/gallium/drivers/r300/r300_texture_desc.h | 1 | ||||
-rw-r--r-- | src/gallium/drivers/r300/r300_tgsi_to_rc.c | 4 | ||||
-rw-r--r-- | src/gallium/drivers/r300/r300_vs.c | 17 | ||||
-rw-r--r-- | src/gallium/drivers/r300/r300_winsys.h | 1 |
18 files changed, 274 insertions, 164 deletions
diff --git a/src/gallium/drivers/r300/r300_context.c b/src/gallium/drivers/r300/r300_context.c index e8b6c4f7af8..624dadd07d7 100644 --- a/src/gallium/drivers/r300/r300_context.c +++ b/src/gallium/drivers/r300/r300_context.c @@ -65,7 +65,7 @@ static void r300_release_referenced_objects(struct r300_context *r300) unsigned i; /* Framebuffer state. */ - util_assign_framebuffer_state(fb, NULL); + util_unreference_framebuffer_state(fb); /* Textures. */ for (i = 0; i < textures->sampler_view_count; i++) @@ -99,8 +99,10 @@ static void r300_destroy_context(struct pipe_context* context) struct r300_context* r300 = r300_context(context); struct r300_atom *atom; - util_blitter_destroy(r300->blitter); - draw_destroy(r300->draw); + if (r300->blitter) + util_blitter_destroy(r300->blitter); + if (r300->draw) + draw_destroy(r300->draw); /* Print stats, if enabled. */ if (SCREEN_DBG_ON(r300->screen, DBG_STATS)) { @@ -112,40 +114,48 @@ static void r300_destroy_context(struct pipe_context* context) } } - u_upload_destroy(r300->upload_vb); - u_upload_destroy(r300->upload_ib); + if (r300->upload_vb) + u_upload_destroy(r300->upload_vb); + if (r300->upload_ib) + u_upload_destroy(r300->upload_ib); - /* setup hyper-z mm */ - if (r300->rws->get_value(r300->rws, R300_CAN_HYPERZ)) - r300_hyperz_destroy_mm(r300); - - translate_cache_destroy(r300->tran.translate_cache); + if (r300->tran.translate_cache) + translate_cache_destroy(r300->tran.translate_cache); + /* XXX: This function assumes r300->query_list was initialized */ r300_release_referenced_objects(r300); - r300->rws->cs_destroy(r300->cs); + if (r300->zmask_mm) + r300_hyperz_destroy_mm(r300); + + if (r300->cs) + r300->rws->cs_destroy(r300->cs); + /* XXX: No way to tell if this was initialized or not? */ util_mempool_destroy(&r300->pool_transfers); r300_update_num_contexts(r300->screen, -1); - FREE(r300->aa_state.state); - FREE(r300->blend_color_state.state); - FREE(r300->clip_state.state); - FREE(r300->fb_state.state); - FREE(r300->gpu_flush.state); - FREE(r300->hyperz_state.state); - FREE(r300->invariant_state.state); - FREE(r300->rs_block_state.state); - FREE(r300->scissor_state.state); - FREE(r300->textures_state.state); - FREE(r300->vap_invariant_state.state); - FREE(r300->viewport_state.state); - FREE(r300->ztop_state.state); - FREE(r300->fs_constants.state); - FREE(r300->vs_constants.state); - if (!r300->screen->caps.has_tcl) { - FREE(r300->vertex_stream_state.state); + /* Free the structs allocated in r300_setup_atoms() */ + if (r300->aa_state.state) { + FREE(r300->aa_state.state); + FREE(r300->blend_color_state.state); + FREE(r300->clip_state.state); + FREE(r300->fb_state.state); + FREE(r300->gpu_flush.state); + FREE(r300->hyperz_state.state); + FREE(r300->invariant_state.state); + FREE(r300->rs_block_state.state); + FREE(r300->scissor_state.state); + FREE(r300->textures_state.state); + FREE(r300->vap_invariant_state.state); + FREE(r300->viewport_state.state); + FREE(r300->ztop_state.state); + FREE(r300->fs_constants.state); + FREE(r300->vs_constants.state); + if (!r300->screen->caps.has_tcl) { + FREE(r300->vertex_stream_state.state); + } } FREE(r300); } @@ -158,12 +168,14 @@ void r300_flush_cb(void *data) } #define R300_INIT_ATOM(atomname, atomsize) \ + do { \ r300->atomname.name = #atomname; \ r300->atomname.state = NULL; \ r300->atomname.size = atomsize; \ r300->atomname.emit = r300_emit_##atomname; \ r300->atomname.dirty = FALSE; \ - insert_at_tail(&r300->atom_list, &r300->atomname); + insert_at_tail(&r300->atom_list, &r300->atomname); \ + } while (0) static void r300_setup_atoms(struct r300_context* r300) { @@ -404,19 +416,21 @@ struct pipe_context* r300_create_context(struct pipe_screen* screen, r300->context.destroy = r300_destroy_context; - r300->cs = rws->cs_create(rws); + make_empty_list(&r300->query_list); util_mempool_create(&r300->pool_transfers, sizeof(struct pipe_transfer), 64, UTIL_MEMPOOL_SINGLETHREADED); + r300->cs = rws->cs_create(rws); + if (r300->cs == NULL) + goto fail; + if (!r300screen->caps.has_tcl) { /* Create a Draw. This is used for SW TCL. */ r300->draw = draw_create(&r300->context); /* Enable our renderer. */ draw_set_rasterize_stage(r300->draw, r300_draw_stage(r300)); - /* Enable Draw's clipping. */ - draw_set_driver_clipping(r300->draw, FALSE); /* Disable converting points/lines to triangles. */ draw_wide_line_threshold(r300->draw, 10000000.f); draw_wide_point_threshold(r300->draw, 10000000.f); @@ -424,8 +438,6 @@ struct pipe_context* r300_create_context(struct pipe_screen* screen, r300_setup_atoms(r300); - make_empty_list(&r300->query_list); - r300_init_blit_functions(r300); r300_init_flush_functions(r300); r300_init_query_functions(r300); @@ -433,6 +445,8 @@ struct pipe_context* r300_create_context(struct pipe_screen* screen, r300_init_resource_functions(r300); r300->blitter = util_blitter_create(&r300->context); + if (r300->blitter == NULL) + goto fail; /* Render functions must be initialized after blitter. */ r300_init_render_functions(r300); @@ -441,22 +455,25 @@ struct pipe_context* r300_create_context(struct pipe_screen* screen, /* setup hyper-z mm */ if (r300->rws->get_value(r300->rws, R300_CAN_HYPERZ)) - r300_hyperz_init_mm(r300); + if (!r300_hyperz_init_mm(r300)) + goto fail; r300->upload_ib = u_upload_create(&r300->context, 32 * 1024, 16, PIPE_BIND_INDEX_BUFFER); if (r300->upload_ib == NULL) - goto no_upload_ib; + goto fail; r300->upload_vb = u_upload_create(&r300->context, 128 * 1024, 16, PIPE_BIND_VERTEX_BUFFER); if (r300->upload_vb == NULL) - goto no_upload_vb; + goto fail; r300->tran.translate_cache = translate_cache_create(); + if (r300->tran.translate_cache == NULL) + goto fail; r300_init_states(&r300->context); @@ -486,10 +503,8 @@ struct pipe_context* r300_create_context(struct pipe_screen* screen, return &r300->context; - no_upload_ib: - u_upload_destroy(r300->upload_ib); - no_upload_vb: - FREE(r300); + fail: + r300_destroy_context(&r300->context); return NULL; } diff --git a/src/gallium/drivers/r300/r300_context.h b/src/gallium/drivers/r300/r300_context.h index 6fa7f470f98..8f0e86fd378 100644 --- a/src/gallium/drivers/r300/r300_context.h +++ b/src/gallium/drivers/r300/r300_context.h @@ -254,8 +254,8 @@ struct r300_ztop_state { struct r300_constant_buffer { /* Buffer of constants */ uint32_t *ptr; - /* Total number of vec4s */ - unsigned count; + /* Remapping table. */ + unsigned *remap_table; }; /* Query object. @@ -449,6 +449,7 @@ struct r300_context { struct r300_screen *screen; /* Draw module. Used mostly for SW TCL. */ struct draw_context* draw; + size_t draw_vbo_size; /* Accelerated blit support. */ struct blitter_context* blitter; /* Stencil two-sided reference value fallback. */ @@ -649,6 +650,11 @@ void r300_translate_index_buffer(struct r300_context *r300, /* r300_render_stencilref.c */ void r300_plug_in_stencil_ref_fallback(struct r300_context *r300); +/* r300_render.c */ +void r300_draw_flush_vbuf(struct r300_context *r300); +boolean r500_index_bias_supported(struct r300_context *r300); +void r500_emit_index_bias(struct r300_context *r300, int index_bias); + /* r300_state.c */ enum r300_fb_state_change { R300_CHANGED_FB_STATE = 0, diff --git a/src/gallium/drivers/r300/r300_emit.c b/src/gallium/drivers/r300/r300_emit.c index d0fd45349e3..232259e21d1 100644 --- a/src/gallium/drivers/r300/r300_emit.c +++ b/src/gallium/drivers/r300/r300_emit.c @@ -180,9 +180,18 @@ void r300_emit_fs_constants(struct r300_context* r300, unsigned size, void *stat BEGIN_CS(size); OUT_CS_REG_SEQ(R300_PFS_PARAM_0_X, count * 4); - for (i = 0; i < count; i++) - for (j = 0; j < 4; j++) - OUT_CS(pack_float24(*(float*)&buf->ptr[i*4+j])); + if (buf->remap_table){ + for (i = 0; i < count; i++) { + float *data = (float*)&buf->ptr[buf->remap_table[i]*4]; + for (j = 0; j < 4; j++) + OUT_CS(pack_float24(data[j])); + } + } else { + for (i = 0; i < count; i++) + for (j = 0; j < 4; j++) + OUT_CS(pack_float24(*(float*)&buf->ptr[i*4+j])); + } + END_CS; } @@ -226,7 +235,7 @@ void r500_emit_fs_constants(struct r300_context* r300, unsigned size, void *stat { struct r300_fragment_shader *fs = r300_fs(r300); struct r300_constant_buffer *buf = (struct r300_constant_buffer*)state; - unsigned count = fs->shader->externals_count * 4; + unsigned count = fs->shader->externals_count; CS_LOCALS(r300); if (count == 0) @@ -234,8 +243,15 @@ void r500_emit_fs_constants(struct r300_context* r300, unsigned size, void *stat BEGIN_CS(size); OUT_CS_REG(R500_GA_US_VECTOR_INDEX, R500_GA_US_VECTOR_INDEX_TYPE_CONST); - OUT_CS_ONE_REG(R500_GA_US_VECTOR_DATA, count); - OUT_CS_TABLE(buf->ptr, count); + OUT_CS_ONE_REG(R500_GA_US_VECTOR_DATA, count * 4); + if (buf->remap_table){ + for (unsigned i = 0; i < count; i++) { + uint32_t *data = &buf->ptr[buf->remap_table[i]*4]; + OUT_CS_TABLE(data, 4); + } + } else { + OUT_CS_TABLE(buf->ptr, count * 4); + } END_CS; } @@ -893,7 +909,7 @@ void r300_emit_vs_state(struct r300_context* r300, unsigned size, void* state) unsigned pvs_num_slots = MIN3(vtx_mem_size / input_count, vtx_mem_size / output_count, 10); - unsigned pvs_num_controllers = MIN2(vtx_mem_size / temp_count, 6); + unsigned pvs_num_controllers = MIN2(vtx_mem_size / temp_count, 5); unsigned imm_first = vs->externals_count; unsigned imm_end = vs->code.constants.Count; @@ -961,6 +977,7 @@ void r300_emit_vs_constants(struct r300_context* r300, unsigned count = ((struct r300_vertex_shader*)r300->vs_state.state)->externals_count; struct r300_constant_buffer *buf = (struct r300_constant_buffer*)state; + unsigned i; CS_LOCALS(r300); if (!count) @@ -971,7 +988,14 @@ void r300_emit_vs_constants(struct r300_context* r300, (r300->screen->caps.is_r500 ? R500_PVS_CONST_START : R300_PVS_CONST_START)); OUT_CS_ONE_REG(R300_VAP_PVS_UPLOAD_DATA, count * 4); - OUT_CS_TABLE(buf->ptr, count * 4); + if (buf->remap_table){ + for (i = 0; i < count; i++) { + uint32_t *data = &buf->ptr[buf->remap_table[i]*4]; + OUT_CS_TABLE(data, 4); + } + } else { + OUT_CS_TABLE(buf->ptr, count * 4); + } END_CS; } @@ -1219,6 +1243,8 @@ unsigned r300_get_num_cs_end_dwords(struct r300_context *r300) /* Emitted in flush. */ dwords += 26; /* emit_query_end */ dwords += r300->hyperz_state.size + 2; /* emit_hyperz_end + zcache flush */ + if (r500_index_bias_supported(r300)) + dwords += 2; return dwords; } diff --git a/src/gallium/drivers/r300/r300_flush.c b/src/gallium/drivers/r300/r300_flush.c index fe182b6615b..2b5d2e42ba5 100644 --- a/src/gallium/drivers/r300/r300_flush.c +++ b/src/gallium/drivers/r300/r300_flush.c @@ -43,9 +43,14 @@ static void r300_flush(struct pipe_context* pipe, u_upload_flush(r300->upload_vb); u_upload_flush(r300->upload_ib); + if (r300->draw) + r300_draw_flush_vbuf(r300); + if (r300->dirty_hw) { r300_emit_hyperz_end(r300); r300_emit_query_end(r300); + if (r500_index_bias_supported(r300)) + r500_emit_index_bias(r300, 0); r300->flush_counter++; r300->rws->cs_flush(r300->cs); diff --git a/src/gallium/drivers/r300/r300_fs.c b/src/gallium/drivers/r300/r300_fs.c index 2a0c30620ad..9845e546109 100644 --- a/src/gallium/drivers/r300/r300_fs.c +++ b/src/gallium/drivers/r300/r300_fs.c @@ -257,17 +257,17 @@ static void r300_emit_fs_code_to_buffer( shader->cb_code_size = 19 + ((code->inst_end + 1) * 6) + imm_count * 7 + - code->int_constant_count * 2; + code->int_constant_count * 2; NEW_CB(shader->cb_code, shader->cb_code_size); OUT_CB_REG(R500_US_CONFIG, R500_ZERO_TIMES_ANYTHING_EQUALS_ZERO); OUT_CB_REG(R500_US_PIXSIZE, code->max_temp_idx); OUT_CB_REG(R500_US_FC_CTRL, code->us_fc_ctrl); for(i = 0; i < code->int_constant_count; i++){ - OUT_CB_REG(R500_US_FC_INT_CONST_0 + (i * 4), - code->int_constants[i]); - } - OUT_CB_REG(R500_US_CODE_RANGE, + OUT_CB_REG(R500_US_FC_INT_CONST_0 + (i * 4), + code->int_constants[i]); + } + OUT_CB_REG(R500_US_CODE_RANGE, R500_US_CODE_RANGE_ADDR(0) | R500_US_CODE_RANGE_SIZE(code->inst_end)); OUT_CB_REG(R500_US_CODE_OFFSET, 0); OUT_CB_REG(R500_US_CODE_ADDR, @@ -386,6 +386,7 @@ static void r300_translate_fragment_shader( compiler.state = shader->compare_state; compiler.Base.is_r500 = r300->screen->caps.is_r500; compiler.Base.max_temp_regs = compiler.Base.is_r500 ? 128 : 32; + compiler.Base.remove_unused_constants = TRUE; compiler.AllocateHwInputs = &allocate_hardware_inputs; compiler.UserData = &shader->inputs; @@ -431,9 +432,8 @@ static void r300_translate_fragment_shader( } if (compiler.Base.Error) { - DBG(r300, DBG_FP, "r300 FP: Compiler Error:\n%sUsing a dummy shader" - " instead.\nIf there's an 'unknown opcode' message, please" - " file a bug report and attach this log.\n", compiler.Base.ErrorMsg); + fprintf(stderr, "r300 FP: Compiler Error:\n%sUsing a dummy shader" + " instead.\n", compiler.Base.ErrorMsg); if (shader->dummy) { fprintf(stderr, "r300 FP: Cannot compile the dummy shader! " @@ -447,7 +447,12 @@ static void r300_translate_fragment_shader( } /* Initialize numbers of constants for each type. */ - shader->externals_count = ttr.immediate_offset; + shader->externals_count = 0; + for (i = 0; + i < shader->code.constants.Count && + shader->code.constants.Constants[i].Type == RC_CONSTANT_EXTERNAL; i++) { + shader->externals_count = i+1; + } shader->immediates_count = 0; shader->rc_state_count = 0; diff --git a/src/gallium/drivers/r300/r300_hyperz.c b/src/gallium/drivers/r300/r300_hyperz.c index 811b5646e16..eb5b0c36f8f 100644 --- a/src/gallium/drivers/r300/r300_hyperz.c +++ b/src/gallium/drivers/r300/r300_hyperz.c @@ -354,7 +354,12 @@ void r300_zmask_alloc_block(struct r300_context *r300, struct r300_surface *surf /* We currently don't handle decompression for 3D textures and cubemaps * correctly. */ if (tex->desc.b.b.target != PIPE_TEXTURE_1D && - tex->desc.b.b.target != PIPE_TEXTURE_2D) + tex->desc.b.b.target != PIPE_TEXTURE_2D && + tex->desc.b.b.target != PIPE_TEXTURE_RECT) + return; + + /* Cannot flush zmask of 16-bit zbuffers. */ + if (util_format_get_blocksizebits(tex->desc.b.b.format) == 16) return; if (tex->zmask_mem[level]) @@ -373,23 +378,36 @@ void r300_zmask_alloc_block(struct r300_context *r300, struct r300_surface *surf return; } -void r300_hyperz_init_mm(struct r300_context *r300) +boolean r300_hyperz_init_mm(struct r300_context *r300) { struct r300_screen* r300screen = r300->screen; int frag_pipes = r300screen->caps.num_frag_pipes; - if (r300screen->caps.hiz_ram) + r300->zmask_mm = u_mmInit(0, r300screen->caps.zmask_ram * frag_pipes); + if (!r300->zmask_mm) + return FALSE; + + if (r300screen->caps.hiz_ram) { r300->hiz_mm = u_mmInit(0, r300screen->caps.hiz_ram * frag_pipes); + if (!r300->hiz_mm) { + u_mmDestroy(r300->zmask_mm); + r300->zmask_mm = NULL; + return FALSE; + } + } - r300->zmask_mm = u_mmInit(0, r300screen->caps.zmask_ram * frag_pipes); + return TRUE; } void r300_hyperz_destroy_mm(struct r300_context *r300) { struct r300_screen* r300screen = r300->screen; - if (r300screen->caps.hiz_ram) + if (r300screen->caps.hiz_ram) { u_mmDestroy(r300->hiz_mm); + r300->hiz_mm = NULL; + } u_mmDestroy(r300->zmask_mm); + r300->zmask_mm = NULL; } diff --git a/src/gallium/drivers/r300/r300_hyperz.h b/src/gallium/drivers/r300/r300_hyperz.h index 09e1ff6625c..30a23ec6493 100644 --- a/src/gallium/drivers/r300/r300_hyperz.h +++ b/src/gallium/drivers/r300/r300_hyperz.h @@ -30,6 +30,6 @@ void r300_update_hyperz_state(struct r300_context* r300); void r300_hiz_alloc_block(struct r300_context *r300, struct r300_surface *surf); void r300_zmask_alloc_block(struct r300_context *r300, struct r300_surface *surf, int compress); -void r300_hyperz_init_mm(struct r300_context *r300); +boolean r300_hyperz_init_mm(struct r300_context *r300); void r300_hyperz_destroy_mm(struct r300_context *r300); #endif diff --git a/src/gallium/drivers/r300/r300_reg.h b/src/gallium/drivers/r300/r300_reg.h index 60d3b600cb7..6bea783f697 100644 --- a/src/gallium/drivers/r300/r300_reg.h +++ b/src/gallium/drivers/r300/r300_reg.h @@ -1607,6 +1607,7 @@ USE OR OTHER DEALINGS IN THE SOFTWARE. # define R300_TX_FORMAT_3D (1 << 25) # define R300_TX_FORMAT_CUBIC_MAP (2 << 25) +# define R300_TX_FORMAT_TEX_COORD_TYPE_MASK (0x3 << 25) /* alpha modes, convenience mostly */ /* if you have alpha, pick constant appropriate to the diff --git a/src/gallium/drivers/r300/r300_render.c b/src/gallium/drivers/r300/r300_render.c index 86b11ca0458..20bad2c56f5 100644 --- a/src/gallium/drivers/r300/r300_render.c +++ b/src/gallium/drivers/r300/r300_render.c @@ -118,13 +118,13 @@ static uint32_t r300_provoking_vertex_fixes(struct r300_context *r300, return color_control; } -static boolean index_bias_supported(struct r300_context *r300) +boolean r500_index_bias_supported(struct r300_context *r300) { return r300->screen->caps.is_r500 && r300->rws->get_value(r300->rws, R300_VID_DRM_2_3_0); } -static void r500_emit_index_bias(struct r300_context *r300, int index_bias) +void r500_emit_index_bias(struct r300_context *r300, int index_bias) { CS_LOCALS(r300); @@ -199,7 +199,7 @@ static void r300_prepare_for_rendering(struct r300_context *r300, boolean emit_aos = flags & PREP_EMIT_AOS; boolean emit_aos_swtcl = flags & PREP_EMIT_AOS_SWTCL; boolean indexed = flags & PREP_INDEXED; - boolean hw_index_bias = index_bias_supported(r300); + boolean hw_index_bias = r500_index_bias_supported(r300); /* Add dirty state, index offset, and AOS. */ if (first_draw) { @@ -506,7 +506,7 @@ static void r300_draw_range_elements(struct pipe_context* pipe, translate = TRUE; } - if (indexBias && !index_bias_supported(r300)) { + if (indexBias && !r500_index_bias_supported(r300)) { r300_split_index_bias(r300, indexBias, &buffer_offset, &index_offset); } @@ -680,18 +680,11 @@ static void r300_swtcl_draw_vbo(struct pipe_context* pipe, if (info->indexed && r300->index_buffer.buffer) { indices = pipe_buffer_map(pipe, r300->index_buffer.buffer, PIPE_TRANSFER_READ, &ib_transfer); - if (indices) - indices = (void *) ((char *) indices + r300->index_buffer.offset); } - draw_set_mapped_element_buffer_range(r300->draw, (indices) ? - r300->index_buffer.index_size : 0, - info->index_bias, - info->min_index, - info->max_index, - indices); + draw_set_mapped_index_buffer(r300->draw, indices); - draw_arrays(r300->draw, info->mode, info->start, count); + draw_vbo(r300->draw, info); /* XXX Not sure whether this is the best fix. * It prevents CS from being rejected and weird assertion failures. */ @@ -707,8 +700,7 @@ static void r300_swtcl_draw_vbo(struct pipe_context* pipe, if (ib_transfer) { pipe_buffer_unmap(pipe, r300->index_buffer.buffer, ib_transfer); - draw_set_mapped_element_buffer_range(r300->draw, 0, 0, info->start, - info->start + count - 1, NULL); + draw_set_mapped_index_buffer(r300->draw, NULL); } } @@ -726,8 +718,6 @@ struct r300_render { unsigned hwprim; /* VBO */ - struct pipe_resource* vbo; - size_t vbo_size; size_t vbo_offset; size_t vbo_max_used; void * vbo_ptr; @@ -759,31 +749,31 @@ static boolean r300_render_allocate_vertices(struct vbuf_render* render, struct pipe_screen* screen = r300->context.screen; size_t size = (size_t)vertex_size * (size_t)count; - if (size + r300render->vbo_offset > r300render->vbo_size) + if (size + r300render->vbo_offset > r300->draw_vbo_size) { - pipe_resource_reference(&r300->vbo, NULL); - r300render->vbo = pipe_buffer_create(screen, - PIPE_BIND_VERTEX_BUFFER, - R300_MAX_DRAW_VBO_SIZE); + pipe_resource_reference(&r300->vbo, NULL); + r300->vbo = pipe_buffer_create(screen, + PIPE_BIND_VERTEX_BUFFER, + R300_MAX_DRAW_VBO_SIZE); r300render->vbo_offset = 0; - r300render->vbo_size = R300_MAX_DRAW_VBO_SIZE; + r300->draw_vbo_size = R300_MAX_DRAW_VBO_SIZE; } r300render->vertex_size = vertex_size; - r300->vbo = r300render->vbo; r300->vbo_offset = r300render->vbo_offset; - return (r300render->vbo) ? TRUE : FALSE; + return (r300->vbo) ? TRUE : FALSE; } static void* r300_render_map_vertices(struct vbuf_render* render) { struct r300_render* r300render = r300_render(render); + struct r300_context* r300 = r300render->r300; assert(!r300render->vbo_transfer); r300render->vbo_ptr = pipe_buffer_map(&r300render->r300->context, - r300render->vbo, + r300->vbo, PIPE_TRANSFER_WRITE, &r300render->vbo_transfer); @@ -798,12 +788,13 @@ static void r300_render_unmap_vertices(struct vbuf_render* render, { struct r300_render* r300render = r300_render(render); struct pipe_context* context = &r300render->r300->context; + struct r300_context* r300 = r300render->r300; assert(r300render->vbo_transfer); r300render->vbo_max_used = MAX2(r300render->vbo_max_used, r300render->vertex_size * (max + 1)); - pipe_buffer_unmap(context, r300render->vbo, r300render->vbo_transfer); + pipe_buffer_unmap(context, r300->vbo, r300render->vbo_transfer); r300render->vbo_transfer = NULL; } @@ -880,7 +871,7 @@ static void r300_render_draw_elements(struct vbuf_render* render, struct r300_context* r300 = r300render->r300; int i; unsigned end_cs_dwords; - unsigned max_index = (r300render->vbo_size - r300render->vbo_offset) / + unsigned max_index = (r300->draw_vbo_size - r300render->vbo_offset) / (r300render->r300->vertex_info.size * 4) - 1; unsigned short_count; unsigned free_dwords; @@ -956,8 +947,6 @@ static struct vbuf_render* r300_render_create(struct r300_context* r300) r300render->base.release_vertices = r300_render_release_vertices; r300render->base.destroy = r300_render_destroy; - r300render->vbo = NULL; - r300render->vbo_size = 0; r300render->vbo_offset = 0; return &r300render->base; @@ -986,6 +975,12 @@ struct draw_stage* r300_draw_stage(struct r300_context* r300) return stage; } +void r300_draw_flush_vbuf(struct r300_context *r300) +{ + pipe_resource_reference(&r300->vbo, NULL); + r300->draw_vbo_size = 0; +} + /**************************************************************************** * End of SW TCL functions * ***************************************************************************/ diff --git a/src/gallium/drivers/r300/r300_state.c b/src/gallium/drivers/r300/r300_state.c index 239edd98e32..8ccb63964e7 100644 --- a/src/gallium/drivers/r300/r300_state.c +++ b/src/gallium/drivers/r300/r300_state.c @@ -23,7 +23,7 @@ #include "draw/draw_context.h" -#include "util/u_blitter.h" +#include "util/u_framebuffer.h" #include "util/u_math.h" #include "util/u_mm.h" #include "util/u_memory.h" @@ -748,7 +748,7 @@ static void /* The tiling flags are dependent on the surface miplevel, unfortunately. */ r300_fb_set_tiling_flags(r300, state); - util_assign_framebuffer_state(r300->fb_state.state, state); + util_copy_framebuffer_state(r300->fb_state.state, state); r300_mark_fb_state_dirty(r300, R300_CHANGED_FB_STATE); @@ -865,6 +865,9 @@ void r300_mark_fs_code_dirty(struct r300_context *r300) r300->fs_rc_constant_state.size = fs->shader->rc_state_count * 5; r300->fs_constants.size = fs->shader->externals_count * 4 + 1; } + + ((struct r300_constant_buffer*)r300->fs_constants.state)->remap_table = + fs->shader->code.constants_remap_table; } /* Bind fragment shader state. */ @@ -937,9 +940,9 @@ static void* r300_create_rs_state(struct pipe_context* pipe, uint32_t stuffing_enable; /* R300_GB_ENABLE: 0x4008 */ /* Point sprites texture coordinates, 0: lower left, 1: upper right */ - float point_texcoord_left; /* R300_GA_POINT_S0: 0x4200 */ + float point_texcoord_left = 0; /* R300_GA_POINT_S0: 0x4200 */ float point_texcoord_bottom = 0;/* R300_GA_POINT_T0: 0x4204 */ - float point_texcoord_right; /* R300_GA_POINT_S1: 0x4208 */ + float point_texcoord_right = 1; /* R300_GA_POINT_S1: 0x4208 */ float point_texcoord_top = 0; /* R300_GA_POINT_T1: 0x420c */ CB_LOCALS; @@ -947,6 +950,11 @@ static void* r300_create_rs_state(struct pipe_context* pipe, rs->rs = *state; rs->rs_draw = *state; + /* Generate point sprite texture coordinates in GENERIC0 + * if point_quad_rasterization is TRUE. */ + rs->rs.sprite_coord_enable = state->point_quad_rasterization * + (state->sprite_coord_enable | 1); + /* Override some states for Draw. */ rs->rs_draw.sprite_coord_enable = 0; /* We can do this in HW. */ @@ -1048,16 +1056,13 @@ static void* r300_create_rs_state(struct pipe_context* pipe, /* Point sprites */ stuffing_enable = 0; - if (state->sprite_coord_enable) { + if (rs->rs.sprite_coord_enable) { stuffing_enable = R300_GB_POINT_STUFF_ENABLE; - for (i = 0; i < 8; i++) { - if (state->sprite_coord_enable & (1 << i)) + for (i = 0; i < 8; i++) { + if (rs->rs.sprite_coord_enable & (1 << i)) stuffing_enable |= R300_GB_TEX_ST << (R300_GB_TEX0_SOURCE_SHIFT + (i*2)); - } - - point_texcoord_left = 0.0f; - point_texcoord_right = 1.0f; + } switch (state->sprite_coord_mode) { case PIPE_SPRITE_COORD_UPPER_LEFT: @@ -1208,8 +1213,8 @@ static void* /* Unfortunately, r300-r500 don't support floating-point mipmap lods. */ /* We must pass these to the merge function to clamp them properly. */ - sampler->min_lod = MAX2((unsigned)state->min_lod, 0); - sampler->max_lod = MAX2((unsigned)ceilf(state->max_lod), 0); + sampler->min_lod = (unsigned)MAX2(state->min_lod, 0); + sampler->max_lod = (unsigned)MAX2(ceilf(state->max_lod), 0); lod_bias = CLAMP((int)(state->lod_bias * 32 + 1), -(1 << 9), (1 << 9) - 1); @@ -1548,7 +1553,12 @@ static void r300_set_index_buffer(struct pipe_context* pipe, memset(&r300->index_buffer, 0, sizeof(r300->index_buffer)); } - /* TODO make this more like a state */ + if (r300->screen->caps.has_tcl) { + /* TODO make this more like a state */ + } + else { + draw_set_index_buffer(r300->draw, ib); + } } /* Initialize the PSC tables. */ @@ -1765,6 +1775,9 @@ static void r300_bind_vs_state(struct pipe_context* pipe, void* shader) r300->vs_constants.size = 0; } + ((struct r300_constant_buffer*)r300->vs_constants.state)->remap_table = + vs->code.constants_remap_table; + r300->pvs_flush.dirty = TRUE; } else { draw_bind_vertex_shader(r300->draw, @@ -1779,6 +1792,8 @@ static void r300_delete_vs_state(struct pipe_context* pipe, void* shader) if (r300->screen->caps.has_tcl) { rc_constants_destroy(&vs->code.constants); + if (vs->code.constants_remap_table) + FREE(vs->code.constants_remap_table); } else { draw_delete_vertex_shader(r300->draw, (struct draw_vertex_shader*)vs->draw_vs); @@ -1795,47 +1810,28 @@ static void r300_set_constant_buffer(struct pipe_context *pipe, struct r300_context* r300 = r300_context(pipe); struct r300_constant_buffer *cbuf; uint32_t *mapped = r300_buffer(buf)->user_buffer; - int max_size = 0, max_size_bytes = 0, clamped_size = 0; switch (shader) { case PIPE_SHADER_VERTEX: cbuf = (struct r300_constant_buffer*)r300->vs_constants.state; - max_size = 256; break; case PIPE_SHADER_FRAGMENT: cbuf = (struct r300_constant_buffer*)r300->fs_constants.state; - if (r300->screen->caps.is_r500) { - max_size = 256; - } else { - max_size = 32; - } break; default: assert(0); return; } - max_size_bytes = max_size * 4 * sizeof(float); if (buf == NULL || buf->width0 == 0 || (mapped = r300_buffer(buf)->constant_buffer) == NULL) { - cbuf->count = 0; return; } if (shader == PIPE_SHADER_FRAGMENT || (shader == PIPE_SHADER_VERTEX && r300->screen->caps.has_tcl)) { assert((buf->width0 % (4 * sizeof(float))) == 0); - - /* Check the size of the constant buffer. */ - /* XXX Subtract immediates and RC_STATE_* variables. */ - if (buf->width0 > max_size_bytes) { - fprintf(stderr, "r300: Max size of the constant buffer is " - "%i*4 floats.\n", max_size); - } - - clamped_size = MIN2(buf->width0, max_size_bytes); - cbuf->count = clamped_size / (4 * sizeof(float)); - cbuf->ptr = mapped; + cbuf->ptr = mapped + index*4; } if (shader == PIPE_SHADER_VERTEX) { diff --git a/src/gallium/drivers/r300/r300_state_derived.c b/src/gallium/drivers/r300/r300_state_derived.c index 4a63ed7fc13..960dfdbaf03 100644 --- a/src/gallium/drivers/r300/r300_state_derived.c +++ b/src/gallium/drivers/r300/r300_state_derived.c @@ -211,7 +211,7 @@ static void r300_rs_col(struct r300_rs_block* rs, int id, int ptr, static void r300_rs_col_write(struct r300_rs_block* rs, int id, int fp_offset, enum r300_rs_col_write_type type) { - assert(type != WRITE_COLOR); + assert(type == WRITE_COLOR); rs->inst[id] |= R300_RS_INST_COL_CN_WRITE | R300_RS_INST_COL_ADDR(fp_offset); } @@ -592,6 +592,25 @@ static void r300_merge_textures_and_samplers(struct r300_context* r300) texstate->filter1 = sampler->filter1; texstate->border_color = sampler->border_color; + /* determine min/max levels */ + max_level = MIN3(sampler->max_lod + view->base.first_level, + tex->desc.b.b.last_level, view->base.last_level); + min_level = MIN2(sampler->min_lod + view->base.first_level, + max_level); + + if (tex->desc.is_npot && min_level > 0) { + /* Even though we do not implement mipmapping for NPOT + * textures, we should at least honor the minimum level + * which is allowed to be displayed. We do this by setting up + * an i-th mipmap level as the zero level. */ + r300_texture_setup_format_state(r300->screen, &tex->desc, + min_level, + &texstate->format); + texstate->format.tile_config |= + tex->desc.offset_in_bytes[min_level] & 0xffffffe0; + assert((tex->desc.offset_in_bytes[min_level] & 0x1f) == 0); + } + /* Assign a texture cache region. */ texstate->format.format1 |= view->texcache_region; @@ -654,12 +673,7 @@ static void r300_merge_textures_and_samplers(struct r300_context* r300) texstate->filter0 |= R300_TX_WRAP_T(R300_TX_CLAMP_TO_EDGE); } } else { - /* determine min/max levels */ /* the MAX_MIP level is the largest (finest) one */ - max_level = MIN3(sampler->max_lod + view->base.first_level, - tex->desc.b.b.last_level, view->base.last_level); - min_level = MIN2(sampler->min_lod + view->base.first_level, - max_level); texstate->format.format0 |= R300_TX_NUM_LEVELS(max_level); texstate->filter0 |= R300_TX_MAX_MIP_LEVEL(min_level); } diff --git a/src/gallium/drivers/r300/r300_texture.c b/src/gallium/drivers/r300/r300_texture.c index da8eadd3b53..66f6d80bd0c 100644 --- a/src/gallium/drivers/r300/r300_texture.c +++ b/src/gallium/drivers/r300/r300_texture.c @@ -541,48 +541,55 @@ boolean r300_is_sampler_format_supported(enum pipe_format format) return r300_translate_texformat(format, 0, TRUE) != ~0; } -static void r300_texture_setup_immutable_state(struct r300_screen* screen, - struct r300_texture* tex) +void r300_texture_setup_format_state(struct r300_screen *screen, + struct r300_texture_desc *desc, + unsigned level, + struct r300_texture_format_state *out) { - struct r300_texture_format_state* f = &tex->tx_format; - struct pipe_resource *pt = &tex->desc.b.b; + struct pipe_resource *pt = &desc->b.b; boolean is_r500 = screen->caps.is_r500; + /* Mask out all the fields we change. */ + out->format0 = 0; + out->format1 &= ~R300_TX_FORMAT_TEX_COORD_TYPE_MASK; + out->format2 &= R500_TXFORMAT_MSB; + out->tile_config = 0; + /* Set sampler state. */ - f->format0 = R300_TX_WIDTH((pt->width0 - 1) & 0x7ff) | - R300_TX_HEIGHT((pt->height0 - 1) & 0x7ff); + out->format0 = R300_TX_WIDTH((u_minify(pt->width0, level) - 1) & 0x7ff) | + R300_TX_HEIGHT((u_minify(pt->height0, level) - 1) & 0x7ff); - if (tex->desc.uses_stride_addressing) { + if (desc->uses_stride_addressing) { /* rectangles love this */ - f->format0 |= R300_TX_PITCH_EN; - f->format2 = (tex->desc.stride_in_pixels[0] - 1) & 0x1fff; + out->format0 |= R300_TX_PITCH_EN; + out->format2 = (desc->stride_in_pixels[level] - 1) & 0x1fff; } else { /* Power of two textures (3D, mipmaps, and no pitch), * also NPOT textures with a width being POT. */ - f->format0 |= R300_TX_DEPTH(util_logbase2(pt->depth0) & 0xf); + out->format0 |= + R300_TX_DEPTH(util_logbase2(u_minify(pt->depth0, level)) & 0xf); } - f->format1 = 0; if (pt->target == PIPE_TEXTURE_CUBE) { - f->format1 |= R300_TX_FORMAT_CUBIC_MAP; + out->format1 |= R300_TX_FORMAT_CUBIC_MAP; } if (pt->target == PIPE_TEXTURE_3D) { - f->format1 |= R300_TX_FORMAT_3D; + out->format1 |= R300_TX_FORMAT_3D; } /* large textures on r500 */ if (is_r500) { if (pt->width0 > 2048) { - f->format2 |= R500_TXWIDTH_BIT11; + out->format2 |= R500_TXWIDTH_BIT11; } if (pt->height0 > 2048) { - f->format2 |= R500_TXHEIGHT_BIT11; + out->format2 |= R500_TXHEIGHT_BIT11; } } - f->tile_config = R300_TXO_MACRO_TILE(tex->desc.macrotile[0]) | - R300_TXO_MICRO_TILE(tex->desc.microtile); + out->tile_config = R300_TXO_MACRO_TILE(desc->macrotile[level]) | + R300_TXO_MICRO_TILE(desc->microtile); } static void r300_texture_setup_fb_state(struct r300_screen* screen, @@ -716,7 +723,7 @@ r300_texture_create_object(struct r300_screen *rscreen, return NULL; } /* Initialize the hardware state. */ - r300_texture_setup_immutable_state(rscreen, tex); + r300_texture_setup_format_state(rscreen, &tex->desc, 0, &tex->tx_format); r300_texture_setup_fb_state(rscreen, tex); tex->desc.b.vtbl = &r300_texture_vtbl; @@ -754,7 +761,8 @@ struct pipe_resource *r300_texture_create(struct pipe_screen *screen, /* Refuse to create a texture with size 0. */ if (!base->width0 || (!base->height0 && (base->target == PIPE_TEXTURE_2D || - base->target == PIPE_TEXTURE_CUBE)) || + base->target == PIPE_TEXTURE_CUBE || + base->target == PIPE_TEXTURE_RECT)) || (!base->depth0 && base->target == PIPE_TEXTURE_3D)) { fprintf(stderr, "r300: texture_create: " "Got invalid texture dimensions: %ix%ix%i\n", @@ -787,7 +795,8 @@ struct pipe_resource *r300_texture_from_handle(struct pipe_screen *screen, unsigned stride, size; /* Support only 2D textures without mipmaps */ - if (base->target != PIPE_TEXTURE_2D || + if ((base->target != PIPE_TEXTURE_2D && + base->target != PIPE_TEXTURE_RECT) || base->depth0 != 1 || base->last_level != 0) { return NULL; diff --git a/src/gallium/drivers/r300/r300_texture.h b/src/gallium/drivers/r300/r300_texture.h index a4524320fda..c4588a0c90b 100644 --- a/src/gallium/drivers/r300/r300_texture.h +++ b/src/gallium/drivers/r300/r300_texture.h @@ -23,11 +23,14 @@ #ifndef R300_TEXTURE_H #define R300_TEXTURE_H +#include "pipe/p_compiler.h" #include "pipe/p_format.h" struct pipe_screen; struct pipe_resource; struct winsys_handle; +struct r300_texture_format_state; +struct r300_texture_desc; struct r300_texture; struct r300_screen; @@ -50,6 +53,10 @@ boolean r300_is_zs_format_supported(enum pipe_format format); boolean r300_is_sampler_format_supported(enum pipe_format format); +void r300_texture_setup_format_state(struct r300_screen *screen, + struct r300_texture_desc *desc, + unsigned level, + struct r300_texture_format_state *out); struct pipe_resource* r300_texture_from_handle(struct pipe_screen* screen, diff --git a/src/gallium/drivers/r300/r300_texture_desc.c b/src/gallium/drivers/r300/r300_texture_desc.c index 5d690e8c332..2fe5d721881 100644 --- a/src/gallium/drivers/r300/r300_texture_desc.c +++ b/src/gallium/drivers/r300/r300_texture_desc.c @@ -184,7 +184,8 @@ static unsigned r300_texture_get_nblocksy(struct r300_texture_desc *desc, /* This is needed for the kernel checker, unfortunately. */ if ((desc->b.b.target != PIPE_TEXTURE_1D && - desc->b.b.target != PIPE_TEXTURE_2D) || + desc->b.b.target != PIPE_TEXTURE_2D && + desc->b.b.target != PIPE_TEXTURE_RECT) || desc->b.b.last_level != 0) { height = util_next_power_of_two(height); } @@ -202,7 +203,8 @@ static unsigned r300_texture_get_nblocksy(struct r300_texture_desc *desc, * Do so for 3 or more macrotiles in the Y direction. */ if (level == 0 && desc->b.b.last_level == 0 && (desc->b.b.target == PIPE_TEXTURE_1D || - desc->b.b.target == PIPE_TEXTURE_2D) && + desc->b.b.target == PIPE_TEXTURE_2D || + desc->b.b.target == PIPE_TEXTURE_RECT) && height >= tile_height * 3) { height = align(height, tile_height * 2); } diff --git a/src/gallium/drivers/r300/r300_texture_desc.h b/src/gallium/drivers/r300/r300_texture_desc.h index 95de66f6549..3d7fe1fb473 100644 --- a/src/gallium/drivers/r300/r300_texture_desc.h +++ b/src/gallium/drivers/r300/r300_texture_desc.h @@ -24,6 +24,7 @@ #ifndef R300_TEXTURE_DESC_H #define R300_TEXTURE_DESC_H +#include "pipe/p_format.h" #include "r300_defines.h" struct pipe_resource; diff --git a/src/gallium/drivers/r300/r300_tgsi_to_rc.c b/src/gallium/drivers/r300/r300_tgsi_to_rc.c index dd697b9c374..a4911b9a2a6 100644 --- a/src/gallium/drivers/r300/r300_tgsi_to_rc.c +++ b/src/gallium/drivers/r300/r300_tgsi_to_rc.c @@ -97,13 +97,13 @@ static unsigned translate_opcode(unsigned opcode) /* case TGSI_OPCODE_BRA: return RC_OPCODE_BRA; */ /* case TGSI_OPCODE_CAL: return RC_OPCODE_CAL; */ /* case TGSI_OPCODE_RET: return RC_OPCODE_RET; */ - /* case TGSI_OPCODE_SSG: return RC_OPCODE_SSG; */ + case TGSI_OPCODE_SSG: return RC_OPCODE_SSG; case TGSI_OPCODE_CMP: return RC_OPCODE_CMP; case TGSI_OPCODE_SCS: return RC_OPCODE_SCS; case TGSI_OPCODE_TXB: return RC_OPCODE_TXB; /* case TGSI_OPCODE_NRM: return RC_OPCODE_NRM; */ /* case TGSI_OPCODE_DIV: return RC_OPCODE_DIV; */ - /* case TGSI_OPCODE_DP2: return RC_OPCODE_DP2; */ + case TGSI_OPCODE_DP2: return RC_OPCODE_DP2; case TGSI_OPCODE_TXL: return RC_OPCODE_TXL; case TGSI_OPCODE_BRK: return RC_OPCODE_BRK; case TGSI_OPCODE_IF: return RC_OPCODE_IF; diff --git a/src/gallium/drivers/r300/r300_vs.c b/src/gallium/drivers/r300/r300_vs.c index 54c8de12419..5f8dbb28d0c 100644 --- a/src/gallium/drivers/r300/r300_vs.c +++ b/src/gallium/drivers/r300/r300_vs.c @@ -196,6 +196,7 @@ void r300_translate_vertex_shader(struct r300_context *r300, { struct r300_vertex_program_compiler compiler; struct tgsi_to_rc ttr; + unsigned i; /* Setup the compiler */ rc_init(&compiler.Base); @@ -205,6 +206,7 @@ void r300_translate_vertex_shader(struct r300_context *r300, compiler.UserData = vs; compiler.Base.is_r500 = r300->screen->caps.is_r500; compiler.Base.max_temp_regs = 32; + compiler.Base.remove_unused_constants = TRUE; if (compiler.Base.Debug) { DBG(r300, DBG_VP, "r300: Initial vertex program\n"); @@ -227,9 +229,8 @@ void r300_translate_vertex_shader(struct r300_context *r300, /* Invoke the compiler */ r3xx_compile_vertex_program(&compiler); if (compiler.Base.Error) { - DBG(r300, DBG_VP, "r300 VP: Compiler error:\n%sUsing a dummy shader" - " instead.\nIf there's an 'unknown opcode' message, please" - " file a bug report and attach this log.\n", compiler.Base.ErrorMsg); + fprintf(stderr, "r300 VP: Compiler error:\n%sUsing a dummy shader" + " instead.\n", compiler.Base.ErrorMsg); if (vs->dummy) { fprintf(stderr, "r300 VP: Cannot compile the dummy shader! " @@ -243,7 +244,15 @@ void r300_translate_vertex_shader(struct r300_context *r300, } /* Initialize numbers of constants for each type. */ - vs->externals_count = ttr.immediate_offset; + vs->externals_count = 0; + for (i = 0; + i < vs->code.constants.Count && + vs->code.constants.Constants[i].Type == RC_CONSTANT_EXTERNAL; i++) { + vs->externals_count = i+1; + } + for (; i < vs->code.constants.Count; i++) { + assert(vs->code.constants.Constants[i].Type == RC_CONSTANT_IMMEDIATE); + } vs->immediates_count = vs->code.constants.Count - vs->externals_count; /* And, finally... */ diff --git a/src/gallium/drivers/r300/r300_winsys.h b/src/gallium/drivers/r300/r300_winsys.h index 187780750fa..4597332399a 100644 --- a/src/gallium/drivers/r300/r300_winsys.h +++ b/src/gallium/drivers/r300/r300_winsys.h @@ -33,6 +33,7 @@ #include "r300_defines.h" +struct winsys_handle; struct r300_winsys_screen; struct r300_winsys_buffer; |