diff options
Diffstat (limited to 'src/gallium/drivers/r300/r300_emit.c')
| -rw-r--r-- | src/gallium/drivers/r300/r300_emit.c | 399 |
1 files changed, 286 insertions, 113 deletions
diff --git a/src/gallium/drivers/r300/r300_emit.c b/src/gallium/drivers/r300/r300_emit.c index 258c38fefdc..98a39390bf9 100644 --- a/src/gallium/drivers/r300/r300_emit.c +++ b/src/gallium/drivers/r300/r300_emit.c @@ -31,16 +31,18 @@ #include "r300_screen.h" #include "r300_state_derived.h" #include "r300_state_inlines.h" +#include "r300_texture.h" #include "r300_vs.h" void r300_emit_blend_state(struct r300_context* r300, struct r300_blend_state* blend) { CS_LOCALS(r300); - BEGIN_CS(7); - OUT_CS_REG_SEQ(R300_RB3D_CBLEND, 2); + BEGIN_CS(8); + OUT_CS_REG_SEQ(R300_RB3D_CBLEND, 3); OUT_CS(blend->blend_control); OUT_CS(blend->alpha_blend_control); + OUT_CS(blend->color_channel_mask); OUT_CS_REG(R300_RB3D_ROPCNTL, blend->rop); OUT_CS_REG(R300_RB3D_DITHER_CTL, blend->dither); END_CS; @@ -101,19 +103,23 @@ void r300_emit_dsa_state(struct r300_context* r300, struct r300_screen* r300screen = r300_screen(r300->context.screen); CS_LOCALS(r300); - BEGIN_CS(r300screen->caps->is_r500 ? 8 : 8); + BEGIN_CS(r300screen->caps->is_r500 ? 10 : 8); OUT_CS_REG(R300_FG_ALPHA_FUNC, dsa->alpha_function); - /* XXX figure out the r300 counterpart for this */ - if (r300screen->caps->is_r500) { - /* OUT_CS_REG(R500_FG_ALPHA_VALUE, dsa->alpha_reference); */ - } + + /* not needed since we use the 8bit alpha ref */ + /*if (r300screen->caps->is_r500) { + OUT_CS_REG(R500_FG_ALPHA_VALUE, dsa->alpha_reference); + }*/ + OUT_CS_REG_SEQ(R300_ZB_CNTL, 3); OUT_CS(dsa->z_buffer_control); OUT_CS(dsa->z_stencil_control); OUT_CS(dsa->stencil_ref_mask); OUT_CS_REG(R300_ZB_ZTOP, r300->ztop_state.z_buffer_top); + + /* XXX it seems r3xx doesn't support STENCILREFMASK_BF */ if (r300screen->caps->is_r500) { - /* OUT_CS_REG(R500_ZB_STENCILREFMASK_BF, dsa->stencil_ref_bf); */ + OUT_CS_REG(R500_ZB_STENCILREFMASK_BF, dsa->stencil_ref_bf); } END_CS; } @@ -123,7 +129,9 @@ static const float * get_shader_constant( struct rc_constant * constant, struct r300_constant_buffer * externals) { - static const float zero[4] = { 0.0, 0.0, 0.0, 0.0 }; + static float vec[4] = { 0.0, 0.0, 0.0, 1.0 }; + struct pipe_texture *tex; + switch(constant->Type) { case RC_CONSTANT_EXTERNAL: return externals->constants[constant->u.External]; @@ -131,11 +139,31 @@ static const float * get_shader_constant( case RC_CONSTANT_IMMEDIATE: return constant->u.Immediate; + case RC_CONSTANT_STATE: + switch (constant->u.State[0]) { + /* Factor for converting rectangle coords to + * normalized coords. Should only show up on non-r500. */ + case RC_STATE_R300_TEXRECT_FACTOR: + tex = &r300->textures[constant->u.State[1]]->tex; + vec[0] = 1.0 / tex->width0; + vec[1] = 1.0 / tex->height0; + break; + + default: + debug_printf("r300: Implementation error: " + "Unknown RC_CONSTANT type %d\n", constant->u.State[0]); + } + break; + default: - debug_printf("r300: Implementation error: Unhandled constant type %i\n", - constant->Type); - return zero; + debug_printf("r300: Implementation error: " + "Unhandled constant type %d\n", constant->Type); } + + /* This should either be (0, 0, 0, 1), which should be a relatively safe + * RGBA or STRQ value, or it could be one of the RC_CONSTANT_STATE + * state factors. */ + return vec; } /* Convert a normal single-precision float into the 7.16 format @@ -173,18 +201,15 @@ static uint32_t pack_float24(float f) } void r300_emit_fragment_program_code(struct r300_context* r300, - struct rX00_fragment_program_code* generic_code, - struct r300_constant_buffer* externals) + struct rX00_fragment_program_code* generic_code) { struct r300_fragment_program_code * code = &generic_code->code.r300; - struct rc_constant_list * constants = &generic_code->constants; int i; CS_LOCALS(r300); BEGIN_CS(15 + code->alu.length * 4 + - (code->tex.length ? (1 + code->tex.length) : 0) + - (constants->Count ? (1 + constants->Count * 4) : 0)); + (code->tex.length ? (1 + code->tex.length) : 0)); OUT_CS_REG(R300_US_CONFIG, code->config); OUT_CS_REG(R300_US_PIXSIZE, code->pixsize); @@ -216,32 +241,41 @@ void r300_emit_fragment_program_code(struct r300_context* r300, OUT_CS(code->tex.inst[i]); } - if (constants->Count) { - OUT_CS_REG_SEQ(R300_PFS_PARAM_0_X, constants->Count * 4); - for(i = 0; i < constants->Count; ++i) { - const float * data = get_shader_constant(r300, &constants->Constants[i], externals); - OUT_CS(pack_float24(data[0])); - OUT_CS(pack_float24(data[1])); - OUT_CS(pack_float24(data[2])); - OUT_CS(pack_float24(data[3])); - } - } + END_CS; +} + +void r300_emit_fs_constant_buffer(struct r300_context* r300, + struct rc_constant_list* constants) +{ + int i; + CS_LOCALS(r300); + + if (constants->Count == 0) + return; + BEGIN_CS(constants->Count * 4 + 1); + OUT_CS_REG_SEQ(R300_PFS_PARAM_0_X, constants->Count * 4); + for(i = 0; i < constants->Count; ++i) { + const float * data = get_shader_constant(r300, + &constants->Constants[i], + &r300->shader_constants[PIPE_SHADER_FRAGMENT]); + OUT_CS(pack_float24(data[0])); + OUT_CS(pack_float24(data[1])); + OUT_CS(pack_float24(data[2])); + OUT_CS(pack_float24(data[3])); + } END_CS; } void r500_emit_fragment_program_code(struct r300_context* r300, - struct rX00_fragment_program_code* generic_code, - struct r300_constant_buffer* externals) + struct rX00_fragment_program_code* generic_code) { struct r500_fragment_program_code * code = &generic_code->code.r500; - struct rc_constant_list * constants = &generic_code->constants; int i; CS_LOCALS(r300); BEGIN_CS(13 + - ((code->inst_end + 1) * 6) + - (constants->Count ? (3 + (constants->Count * 4)) : 0)); + ((code->inst_end + 1) * 6)); OUT_CS_REG(R500_US_CONFIG, 0); OUT_CS_REG(R500_US_PIXSIZE, code->max_temp_idx); OUT_CS_REG(R500_US_CODE_RANGE, @@ -261,18 +295,30 @@ void r500_emit_fragment_program_code(struct r300_context* r300, OUT_CS(code->inst[i].inst5); } - if (constants->Count) { - OUT_CS_REG(R500_GA_US_VECTOR_INDEX, R500_GA_US_VECTOR_INDEX_TYPE_CONST); - OUT_CS_ONE_REG(R500_GA_US_VECTOR_DATA, constants->Count * 4); - for (i = 0; i < constants->Count; i++) { - const float * data = get_shader_constant(r300, &constants->Constants[i], externals); - OUT_CS_32F(data[0]); - OUT_CS_32F(data[1]); - OUT_CS_32F(data[2]); - OUT_CS_32F(data[3]); - } - } + END_CS; +} + +void r500_emit_fs_constant_buffer(struct r300_context* r300, + struct rc_constant_list* constants) +{ + int i; + CS_LOCALS(r300); + + if (constants->Count == 0) + return; + BEGIN_CS(constants->Count * 4 + 3); + OUT_CS_REG(R500_GA_US_VECTOR_INDEX, R500_GA_US_VECTOR_INDEX_TYPE_CONST); + OUT_CS_ONE_REG(R500_GA_US_VECTOR_DATA, constants->Count * 4); + for (i = 0; i < constants->Count; i++) { + const float * data = get_shader_constant(r300, + &constants->Constants[i], + &r300->shader_constants[PIPE_SHADER_FRAGMENT]); + OUT_CS_32F(data[0]); + OUT_CS_32F(data[1]); + OUT_CS_32F(data[2]); + OUT_CS_32F(data[3]); + } END_CS; } @@ -280,48 +326,50 @@ void r300_emit_fb_state(struct r300_context* r300, struct pipe_framebuffer_state* fb) { struct r300_texture* tex; - unsigned pixpitch; + struct pipe_surface* surf; int i; CS_LOCALS(r300); BEGIN_CS((10 * fb->nr_cbufs) + (fb->zsbuf ? 10 : 0) + 4); + OUT_CS_REG(R300_RB3D_DSTCACHE_CTLSTAT, + R300_RB3D_DSTCACHE_CTLSTAT_DC_FREE_FREE_3D_TAGS | + R300_RB3D_DSTCACHE_CTLSTAT_DC_FLUSH_FLUSH_DIRTY_3D); + OUT_CS_REG(R300_ZB_ZCACHE_CTLSTAT, + R300_ZB_ZCACHE_CTLSTAT_ZC_FLUSH_FLUSH_AND_FREE | + R300_ZB_ZCACHE_CTLSTAT_ZC_FREE_FREE); + for (i = 0; i < fb->nr_cbufs; i++) { - tex = (struct r300_texture*)fb->cbufs[i]->texture; + surf = fb->cbufs[i]; + tex = (struct r300_texture*)surf->texture; assert(tex && tex->buffer && "cbuf is marked, but NULL!"); - pixpitch = r300_texture_get_stride(tex, 0) / tex->tex.block.size; OUT_CS_REG_SEQ(R300_RB3D_COLOROFFSET0 + (4 * i), 1); - OUT_CS_RELOC(tex->buffer, 0, 0, RADEON_GEM_DOMAIN_VRAM, 0); + OUT_CS_RELOC(tex->buffer, surf->offset, 0, RADEON_GEM_DOMAIN_VRAM, 0); OUT_CS_REG_SEQ(R300_RB3D_COLORPITCH0 + (4 * i), 1); - OUT_CS_RELOC(tex->buffer, pixpitch | + OUT_CS_RELOC(tex->buffer, tex->pitch[surf->level] | r300_translate_colorformat(tex->tex.format), 0, RADEON_GEM_DOMAIN_VRAM, 0); OUT_CS_REG(R300_US_OUT_FMT_0 + (4 * i), - r300_translate_out_fmt(fb->cbufs[i]->format)); + r300_translate_out_fmt(surf->format)); } if (fb->zsbuf) { - tex = (struct r300_texture*)fb->zsbuf->texture; + surf = fb->zsbuf; + tex = (struct r300_texture*)surf->texture; assert(tex && tex->buffer && "zsbuf is marked, but NULL!"); - pixpitch = r300_texture_get_stride(tex, 0) / tex->tex.block.size; OUT_CS_REG_SEQ(R300_ZB_DEPTHOFFSET, 1); - OUT_CS_RELOC(tex->buffer, 0, 0, RADEON_GEM_DOMAIN_VRAM, 0); + OUT_CS_RELOC(tex->buffer, surf->offset, 0, RADEON_GEM_DOMAIN_VRAM, 0); OUT_CS_REG(R300_ZB_FORMAT, r300_translate_zsformat(tex->tex.format)); OUT_CS_REG_SEQ(R300_ZB_DEPTHPITCH, 1); - OUT_CS_RELOC(tex->buffer, pixpitch, 0, RADEON_GEM_DOMAIN_VRAM, 0); + OUT_CS_RELOC(tex->buffer, tex->pitch[surf->level], 0, + RADEON_GEM_DOMAIN_VRAM, 0); } - OUT_CS_REG(R300_RB3D_DSTCACHE_CTLSTAT, - R300_RB3D_DSTCACHE_CTLSTAT_DC_FREE_FREE_3D_TAGS | - R300_RB3D_DSTCACHE_CTLSTAT_DC_FLUSH_FLUSH_DIRTY_3D); - OUT_CS_REG(R300_ZB_ZCACHE_CTLSTAT, - R300_ZB_ZCACHE_CTLSTAT_ZC_FLUSH_FLUSH_AND_FREE | - R300_ZB_ZCACHE_CTLSTAT_ZC_FREE_FREE); END_CS; } @@ -457,7 +505,7 @@ void r300_emit_rs_state(struct r300_context* r300, struct r300_rs_state* rs) { CS_LOCALS(r300); - BEGIN_CS(20); + BEGIN_CS(22); OUT_CS_REG(R300_VAP_CNTL_STATUS, rs->vap_control_status); OUT_CS_REG(R300_GA_POINT_SIZE, rs->point_size); OUT_CS_REG_SEQ(R300_GA_POINT_MINMAX, 2); @@ -473,6 +521,7 @@ void r300_emit_rs_state(struct r300_context* r300, struct r300_rs_state* rs) OUT_CS_REG(R300_GA_LINE_STIPPLE_CONFIG, rs->line_stipple_config); OUT_CS_REG(R300_GA_LINE_STIPPLE_VALUE, rs->line_stipple_value); OUT_CS_REG(R300_GA_COLOR_CONTROL, rs->color_control); + OUT_CS_REG(R300_GA_POLY_MODE, rs->polygon_mode); END_CS; } @@ -483,6 +532,8 @@ void r300_emit_rs_block_state(struct r300_context* r300, struct r300_screen* r300screen = r300_screen(r300->context.screen); CS_LOCALS(r300); + DBG(r300, DBG_DRAW, "r300: RS emit:\n"); + BEGIN_CS(21); if (r300screen->caps->is_r500) { OUT_CS_REG_SEQ(R500_RS_IP_0, 8); @@ -491,7 +542,7 @@ void r300_emit_rs_block_state(struct r300_context* r300, } for (i = 0; i < 8; i++) { OUT_CS(rs->ip[i]); - /* debug_printf("ip %d: 0x%08x\n", i, rs->ip[i]); */ + DBG(r300, DBG_DRAW, " : ip %d: 0x%08x\n", i, rs->ip[i]); } OUT_CS_REG_SEQ(R300_RS_COUNT, 2); @@ -505,11 +556,11 @@ void r300_emit_rs_block_state(struct r300_context* r300, } for (i = 0; i < 8; i++) { OUT_CS(rs->inst[i]); - /* debug_printf("inst %d: 0x%08x\n", i, rs->inst[i]); */ + DBG(r300, DBG_DRAW, " : inst %d: 0x%08x\n", i, rs->inst[i]); } - /* debug_printf("count: 0x%08x inst_count: 0x%08x\n", rs->count, - * rs->inst_count); */ + DBG(r300, DBG_DRAW, " : count: 0x%08x inst_count: 0x%08x\n", + rs->count, rs->inst_count); END_CS; } @@ -531,15 +582,31 @@ void r300_emit_texture(struct r300_context* r300, struct r300_texture* tex, unsigned offset) { + uint32_t filter0 = sampler->filter0; + uint32_t format0 = tex->state.format0; + unsigned min_level, max_level; CS_LOCALS(r300); + /* to emulate 1D textures through 2D ones correctly */ + if (tex->tex.target == PIPE_TEXTURE_1D) { + filter0 &= ~R300_TX_WRAP_T_MASK; + filter0 |= R300_TX_WRAP_T(R300_TX_CLAMP_TO_EDGE); + } + + /* determine min/max levels */ + /* the MAX_MIP level is the largest (finest) one */ + max_level = MIN2(sampler->max_lod, tex->tex.last_level); + min_level = MIN2(sampler->min_lod, max_level); + format0 |= R300_TX_NUM_LEVELS(max_level); + filter0 |= R300_TX_MAX_MIP_LEVEL(min_level); + BEGIN_CS(16); - OUT_CS_REG(R300_TX_FILTER0_0 + (offset * 4), sampler->filter0 | + OUT_CS_REG(R300_TX_FILTER0_0 + (offset * 4), filter0 | (offset << 28)); OUT_CS_REG(R300_TX_FILTER1_0 + (offset * 4), sampler->filter1); OUT_CS_REG(R300_TX_BORDER_COLOR_0 + (offset * 4), sampler->border_color); - OUT_CS_REG(R300_TX_FORMAT0_0 + (offset * 4), tex->state.format0); + OUT_CS_REG(R300_TX_FORMAT0_0 + (offset * 4), format0); OUT_CS_REG(R300_TX_FORMAT1_0 + (offset * 4), tex->state.format1); OUT_CS_REG(R300_TX_FORMAT2_0 + (offset * 4), tex->state.format2); OUT_CS_REG_SEQ(R300_TX_OFFSET_0 + (offset * 4), 1); @@ -548,7 +615,52 @@ void r300_emit_texture(struct r300_context* r300, END_CS; } -void r300_emit_vertex_buffer(struct r300_context* r300) +/* XXX I can't read this and that's not good */ +void r300_emit_aos(struct r300_context* r300, unsigned offset) +{ + struct pipe_vertex_buffer *vbuf = r300->vertex_buffer; + struct pipe_vertex_element *velem = r300->vertex_element; + CS_LOCALS(r300); + int i; + unsigned aos_count = r300->vertex_element_count; + + unsigned packet_size = (aos_count * 3 + 1) / 2; + BEGIN_CS(2 + packet_size + aos_count * 2); + OUT_CS_PKT3(R300_PACKET3_3D_LOAD_VBPNTR, packet_size); + OUT_CS(aos_count); + for (i = 0; i < aos_count - 1; i += 2) { + int buf_num1 = velem[i].vertex_buffer_index; + int buf_num2 = velem[i+1].vertex_buffer_index; + assert(vbuf[buf_num1].stride % 4 == 0 && pf_get_size(velem[i].src_format) % 4 == 0); + assert(vbuf[buf_num2].stride % 4 == 0 && pf_get_size(velem[i+1].src_format) % 4 == 0); + OUT_CS((pf_get_size(velem[i].src_format) >> 2) | (vbuf[buf_num1].stride << 6) | + (pf_get_size(velem[i+1].src_format) << 14) | (vbuf[buf_num2].stride << 22)); + OUT_CS(vbuf[buf_num1].buffer_offset + velem[i].src_offset + + offset * vbuf[buf_num1].stride); + OUT_CS(vbuf[buf_num2].buffer_offset + velem[i+1].src_offset + + offset * vbuf[buf_num2].stride); + } + if (aos_count & 1) { + int buf_num = velem[i].vertex_buffer_index; + assert(vbuf[buf_num].stride % 4 == 0 && pf_get_size(velem[i].src_format) % 4 == 0); + OUT_CS((pf_get_size(velem[i].src_format) >> 2) | (vbuf[buf_num].stride << 6)); + OUT_CS(vbuf[buf_num].buffer_offset + velem[i].src_offset + + offset * vbuf[buf_num].stride); + } + + /* XXX bare CS reloc */ + for (i = 0; i < aos_count; i++) { + cs_winsys->write_cs_reloc(cs_winsys, + vbuf[velem[i].vertex_buffer_index].buffer, + RADEON_GEM_DOMAIN_GTT, + 0, + 0); + cs_count -= 2; + } + END_CS; +} +#if 0 +void r300_emit_draw_packet(struct r300_context* r300) { CS_LOCALS(r300); @@ -571,12 +683,15 @@ void r300_emit_vertex_buffer(struct r300_context* r300) OUT_CS_RELOC(r300->vbo, 0, RADEON_GEM_DOMAIN_GTT, 0, 0); END_CS; } +#endif void r300_emit_vertex_format_state(struct r300_context* r300) { int i; CS_LOCALS(r300); + DBG(r300, DBG_DRAW, "r300: VAP/PSC emit:\n"); + BEGIN_CS(26); OUT_CS_REG(R300_VAP_VTX_SIZE, r300->vertex_info->vinfo.size); @@ -586,33 +701,42 @@ void r300_emit_vertex_format_state(struct r300_context* r300) OUT_CS_REG_SEQ(R300_VAP_OUTPUT_VTX_FMT_0, 2); OUT_CS(r300->vertex_info->vinfo.hwfmt[2]); OUT_CS(r300->vertex_info->vinfo.hwfmt[3]); - /* for (i = 0; i < 4; i++) { - * debug_printf("hwfmt%d: 0x%08x\n", i, - * r300->vertex_info->vinfo.hwfmt[i]); - * } */ + for (i = 0; i < 4; i++) { + DBG(r300, DBG_DRAW, " : hwfmt%d: 0x%08x\n", i, + r300->vertex_info->vinfo.hwfmt[i]); + } OUT_CS_REG_SEQ(R300_VAP_PROG_STREAM_CNTL_0, 8); for (i = 0; i < 8; i++) { OUT_CS(r300->vertex_info->vap_prog_stream_cntl[i]); - /* debug_printf("prog_stream_cntl%d: 0x%08x\n", i, - * r300->vertex_info->vap_prog_stream_cntl[i]); */ + DBG(r300, DBG_DRAW, " : prog_stream_cntl%d: 0x%08x\n", i, + r300->vertex_info->vap_prog_stream_cntl[i]); } OUT_CS_REG_SEQ(R300_VAP_PROG_STREAM_CNTL_EXT_0, 8); for (i = 0; i < 8; i++) { OUT_CS(r300->vertex_info->vap_prog_stream_cntl_ext[i]); - /* debug_printf("prog_stream_cntl_ext%d: 0x%08x\n", i, - * r300->vertex_info->vap_prog_stream_cntl_ext[i]); */ + DBG(r300, DBG_DRAW, " : prog_stream_cntl_ext%d: 0x%08x\n", i, + r300->vertex_info->vap_prog_stream_cntl_ext[i]); } END_CS; } + void r300_emit_vertex_program_code(struct r300_context* r300, - struct r300_vertex_program_code* code, - struct r300_constant_buffer* constants) + struct r300_vertex_program_code* code) { int i; struct r300_screen* r300screen = r300_screen(r300->context.screen); unsigned instruction_count = code->length / 4; + + int vtx_mem_size = r300screen->caps->is_r500 ? 128 : 72; + int input_count = MAX2(util_bitcount(code->InputsRead), 1); + int output_count = MAX2(util_bitcount(code->OutputsWritten), 1); + int temp_count = MAX2(code->num_temporaries, 1); + int pvs_num_slots = MIN3(vtx_mem_size / input_count, + vtx_mem_size / output_count, 10); + int pvs_num_controllers = MIN2(vtx_mem_size / temp_count, 6); + CS_LOCALS(r300); if (!r300screen->caps->has_tcl) { @@ -621,17 +745,11 @@ void r300_emit_vertex_program_code(struct r300_context* r300, return; } - if (code->constants.Count) { - BEGIN_CS(14 + code->length + (code->constants.Count * 4)); - } else { - BEGIN_CS(11 + code->length); - } - + BEGIN_CS(9 + code->length); /* R300_VAP_PVS_CODE_CNTL_0 * R300_VAP_PVS_CONST_CNTL * R300_VAP_PVS_CODE_CNTL_1 - * See the r5xx docs for instructions on how to use these. - * XXX these could be optimized to select better values... */ + * See the r5xx docs for instructions on how to use these. */ OUT_CS_REG_SEQ(R300_VAP_PVS_CODE_CNTL_0, 3); OUT_CS(R300_PVS_FIRST_INST(0) | R300_PVS_XYZW_VALID_INST(instruction_count - 1) | @@ -644,32 +762,51 @@ void r300_emit_vertex_program_code(struct r300_context* r300, for (i = 0; i < code->length; i++) OUT_CS(code->body.d[i]); - if (code->constants.Count) { - OUT_CS_REG(R300_VAP_PVS_VECTOR_INDX_REG, - (r300screen->caps->is_r500 ? - R500_PVS_CONST_START : R300_PVS_CONST_START)); - OUT_CS_ONE_REG(R300_VAP_PVS_UPLOAD_DATA, code->constants.Count * 4); - for (i = 0; i < code->constants.Count; i++) { - const float * data = get_shader_constant(r300, &code->constants.Constants[i], constants); - OUT_CS_32F(data[0]); - OUT_CS_32F(data[1]); - OUT_CS_32F(data[2]); - OUT_CS_32F(data[3]); - } - } - - OUT_CS_REG(R300_VAP_CNTL, R300_PVS_NUM_SLOTS(10) | - R300_PVS_NUM_CNTLRS(5) | + OUT_CS_REG(R300_VAP_CNTL, R300_PVS_NUM_SLOTS(pvs_num_slots) | + R300_PVS_NUM_CNTLRS(pvs_num_controllers) | R300_PVS_NUM_FPUS(r300screen->caps->num_vert_fpus) | - R300_PVS_VF_MAX_VTX_NUM(12)); - OUT_CS_REG(R300_VAP_PVS_STATE_FLUSH_REG, 0x0); + R300_PVS_VF_MAX_VTX_NUM(12) | + (r300screen->caps->is_r500 ? R500_TCL_STATE_OPTIMIZATION : 0)); END_CS; } void r300_emit_vertex_shader(struct r300_context* r300, struct r300_vertex_shader* vs) { - r300_emit_vertex_program_code(r300, &vs->code, &r300->shader_constants[PIPE_SHADER_VERTEX]); + r300_emit_vertex_program_code(r300, &vs->code); +} + +void r300_emit_vs_constant_buffer(struct r300_context* r300, + struct rc_constant_list* constants) +{ + int i; + struct r300_screen* r300screen = r300_screen(r300->context.screen); + CS_LOCALS(r300); + + if (!r300screen->caps->has_tcl) { + debug_printf("r300: Implementation error: emit_vertex_shader called," + " but has_tcl is FALSE!\n"); + return; + } + + if (constants->Count == 0) + return; + + BEGIN_CS(constants->Count * 4 + 3); + OUT_CS_REG(R300_VAP_PVS_VECTOR_INDX_REG, + (r300screen->caps->is_r500 ? + R500_PVS_CONST_START : R300_PVS_CONST_START)); + OUT_CS_ONE_REG(R300_VAP_PVS_UPLOAD_DATA, constants->Count * 4); + for (i = 0; i < constants->Count; i++) { + const float * data = get_shader_constant(r300, + &constants->Constants[i], + &r300->shader_constants[PIPE_SHADER_VERTEX]); + OUT_CS_32F(data[0]); + OUT_CS_32F(data[1]); + OUT_CS_32F(data[2]); + OUT_CS_32F(data[3]); + } + END_CS; } void r300_emit_viewport_state(struct r300_context* r300, @@ -694,13 +831,31 @@ void r300_emit_viewport_state(struct r300_context* r300, END_CS; } +void r300_emit_texture_count(struct r300_context* r300) +{ + CS_LOCALS(r300); + + BEGIN_CS(2); + OUT_CS_REG(R300_TX_ENABLE, (1 << r300->texture_count) - 1); + END_CS; + +} + void r300_flush_textures(struct r300_context* r300) { CS_LOCALS(r300); - BEGIN_CS(4); + BEGIN_CS(2); OUT_CS_REG(R300_TX_INVALTAGS, 0); - OUT_CS_REG(R300_TX_ENABLE, (1 << r300->texture_count) - 1); + END_CS; +} + +static void r300_flush_pvs(struct r300_context* r300) +{ + CS_LOCALS(r300); + + BEGIN_CS(2); + OUT_CS_REG(R300_VAP_PVS_STATE_FLUSH_REG, 0x0); END_CS; } @@ -716,8 +871,6 @@ void r300_emit_dirty_state(struct r300_context* r300) return; } - r300_update_derived_state(r300); - /* Clean out BOs. */ r300->winsys->reset_bos(r300->winsys); @@ -747,7 +900,7 @@ validate: for (i = 0; i < r300->texture_count; i++) { tex = r300->textures[i]; if (!tex) - continue; + continue; if (!r300->winsys->add_buffer(r300->winsys, tex->buffer, RADEON_GEM_DOMAIN_GTT | RADEON_GEM_DOMAIN_VRAM, 0)) { r300->context.flush(&r300->context, 0, NULL); @@ -768,7 +921,7 @@ validate: goto validate; } } else { - debug_printf("No VBO while emitting dirty state!\n"); + // debug_printf("No VBO while emitting dirty state!\n"); } if (!r300->winsys->validate(r300->winsys)) { r300->context.flush(&r300->context, 0, NULL); @@ -808,13 +961,22 @@ validate: if (r300->dirty_state & R300_NEW_FRAGMENT_SHADER) { if (r300screen->caps->is_r500) { - r500_emit_fragment_program_code(r300, &r300->fs->code, &r300->shader_constants[PIPE_SHADER_FRAGMENT]); + r500_emit_fragment_program_code(r300, &r300->fs->code); } else { - r300_emit_fragment_program_code(r300, &r300->fs->code, &r300->shader_constants[PIPE_SHADER_FRAGMENT]); + r300_emit_fragment_program_code(r300, &r300->fs->code); } r300->dirty_state &= ~R300_NEW_FRAGMENT_SHADER; } + if (r300->dirty_state & R300_NEW_FRAGMENT_SHADER_CONSTANTS) { + if (r300screen->caps->is_r500) { + r500_emit_fs_constant_buffer(r300, &r300->fs->code.constants); + } else { + r300_emit_fs_constant_buffer(r300, &r300->fs->code.constants); + } + r300->dirty_state &= ~R300_NEW_FRAGMENT_SHADER_CONSTANTS; + } + if (r300->dirty_state & R300_NEW_FRAMEBUFFERS) { r300_emit_fb_state(r300, &r300->framebuffer_state); r300->dirty_state &= ~R300_NEW_FRAMEBUFFERS; @@ -838,6 +1000,8 @@ validate: /* Samplers and textures are tracked separately but emitted together. */ if (r300->dirty_state & (R300_ANY_NEW_SAMPLERS | R300_ANY_NEW_TEXTURES)) { + r300_emit_texture_count(r300); + for (i = 0; i < MIN2(r300->sampler_count, r300->texture_count); i++) { if (r300->dirty_state & ((R300_NEW_SAMPLER << i) | (R300_NEW_TEXTURE << i))) { @@ -868,17 +1032,26 @@ validate: r300->dirty_state &= ~R300_NEW_VERTEX_FORMAT; } + if (r300->dirty_state & (R300_NEW_VERTEX_SHADER | R300_NEW_VERTEX_SHADER_CONSTANTS)) { + r300_flush_pvs(r300); + } + if (r300->dirty_state & R300_NEW_VERTEX_SHADER) { r300_emit_vertex_shader(r300, r300->vs); r300->dirty_state &= ~R300_NEW_VERTEX_SHADER; } + if (r300->dirty_state & R300_NEW_VERTEX_SHADER_CONSTANTS) { + r300_emit_vs_constant_buffer(r300, &r300->vs->code.constants); + r300->dirty_state &= ~R300_NEW_VERTEX_SHADER_CONSTANTS; + } + /* XXX assert(r300->dirty_state == 0); */ /* Finally, emit the VBO. */ - r300_emit_vertex_buffer(r300); + //r300_emit_vertex_buffer(r300); r300->dirty_hw++; } |