diff options
Diffstat (limited to 'src/gallium/drivers/r300/r300_state_derived.c')
-rw-r--r-- | src/gallium/drivers/r300/r300_state_derived.c | 339 |
1 files changed, 132 insertions, 207 deletions
diff --git a/src/gallium/drivers/r300/r300_state_derived.c b/src/gallium/drivers/r300/r300_state_derived.c index 7166694edf..6fb780cb29 100644 --- a/src/gallium/drivers/r300/r300_state_derived.c +++ b/src/gallium/drivers/r300/r300_state_derived.c @@ -47,8 +47,8 @@ struct r300_shader_derived_value { unsigned r300_shader_key_hash(void* key) { struct r300_shader_key* shader_key = (struct r300_shader_key*)key; - unsigned vs = (unsigned)shader_key->vs; - unsigned fs = (unsigned)shader_key->fs; + unsigned vs = (intptr_t)shader_key->vs; + unsigned fs = (intptr_t)shader_key->fs; return (vs << 16) | (fs & 0xffff); } @@ -61,88 +61,46 @@ int r300_shader_key_compare(void* key1, void* key2) { (shader_key1->fs == shader_key2->fs); } -/* Set up the vs_tab and routes. */ -static void r300_vs_tab_routes(struct r300_context* r300, - struct r300_vertex_info* vformat) +/* Set up the vs_output_tab and routes. */ +static void r300_vs_output_tab_routes(struct r300_context* r300, + int* vs_output_tab) { - struct r300_screen* r300screen = r300_screen(r300->context.screen); - struct vertex_info* vinfo = &vformat->vinfo; - int* tab = vformat->vs_tab; + struct vertex_info* vinfo = &r300->vertex_info->vinfo; boolean pos = FALSE, psize = FALSE, fog = FALSE; int i, texs = 0, cols = 0; - struct tgsi_shader_info* info; - - if (r300screen->caps->has_tcl) { - /* Use vertex shader to determine required routes. */ - info = &r300->vs->info; - } else { - /* Use fragment shader to determine required routes. */ - info = &r300->fs->info; - } + struct tgsi_shader_info* info = &r300->fs->info; - assert(info->num_inputs <= 16); + /* XXX One day we should figure out how to handle a different number of + * VS outputs and FS inputs, as well as a different number of vertex streams + * and VS inputs. It's definitely one of the sources of hardlocks. */ - if (!r300screen->caps->has_tcl || !r300->rs_state->enable_vte) - { - for (i = 0; i < info->num_inputs; i++) { - switch (r300->vs->code.inputs[i]) { - case TGSI_SEMANTIC_POSITION: - pos = TRUE; - tab[i] = 0; - break; - case TGSI_SEMANTIC_COLOR: - tab[i] = 2 + cols; - cols++; - break; - case TGSI_SEMANTIC_PSIZE: - assert(psize == FALSE); - psize = TRUE; - tab[i] = 15; - break; - case TGSI_SEMANTIC_FOG: - assert(fog == FALSE); - fog = TRUE; - /* Fall through */ - case TGSI_SEMANTIC_GENERIC: - tab[i] = 6 + texs; - texs++; - break; - default: - debug_printf("r300: Unknown vertex input %d\n", - info->input_semantic_name[i]); - break; - } - } - } - else - { - /* Just copy vert attribs over as-is. */ - for (i = 0; i < info->num_inputs; i++) { - tab[i] = i; - } - - for (i = 0; i < info->num_outputs; i++) { - switch (info->output_semantic_name[i]) { - case TGSI_SEMANTIC_POSITION: - pos = TRUE; - break; - case TGSI_SEMANTIC_COLOR: - cols++; - break; - case TGSI_SEMANTIC_PSIZE: - psize = TRUE; - break; - case TGSI_SEMANTIC_FOG: - fog = TRUE; - /* Fall through */ - case TGSI_SEMANTIC_GENERIC: - texs++; - break; - default: - debug_printf("r300: Unknown vertex output %d\n", - info->output_semantic_name[i]); - break; - } + for (i = 0; i < info->num_inputs; i++) { + switch (info->input_semantic_name[i]) { + case TGSI_SEMANTIC_POSITION: + pos = TRUE; + vs_output_tab[i] = 0; + break; + case TGSI_SEMANTIC_COLOR: + vs_output_tab[i] = 2 + cols; + cols++; + break; + case TGSI_SEMANTIC_PSIZE: + assert(psize == FALSE); + psize = TRUE; + vs_output_tab[i] = 15; + break; + case TGSI_SEMANTIC_FOG: + assert(fog == FALSE); + fog = TRUE; + /* Fall through */ + case TGSI_SEMANTIC_GENERIC: + vs_output_tab[i] = 6 + texs; + texs++; + break; + default: + debug_printf("r300: Unknown vertex input %d\n", + info->input_semantic_name[i]); + break; } } @@ -161,14 +119,13 @@ static void r300_vs_tab_routes(struct r300_context* r300, /* We need to add vertex position attribute only for SW TCL case, * for HW TCL case it could be generated by vertex shader */ - if (!pos && !r300screen->caps->has_tcl) { - debug_printf("r300: Forcing vertex position attribute emit...\n"); + if (!pos) { /* Make room for the position attribute - * at the beginning of the tab. */ + * at the beginning of the vs_output_tab. */ for (i = 15; i > 0; i--) { - tab[i] = tab[i-1]; + vs_output_tab[i] = vs_output_tab[i-1]; } - tab[0] = 0; + vs_output_tab[0] = 0; } /* Position. */ @@ -227,43 +184,78 @@ static void r300_vs_tab_routes(struct r300_context* r300, } /* Update the PSC tables. */ -static void r300_vertex_psc(struct r300_context* r300, - struct r300_vertex_info* vformat) +static void r300_vertex_psc(struct r300_context* r300) { - struct r300_screen* r300screen = r300_screen(r300->context.screen); - struct vertex_info* vinfo = &vformat->vinfo; - int* tab = vformat->vs_tab; + struct r300_vertex_info *vformat = r300->vertex_info; uint16_t type, swizzle; enum pipe_format format; - unsigned i, attrib_count; + unsigned i; /* Vertex shaders have no semantics on their inputs, - * so PSC should just route stuff based on their info, + * so PSC should just route stuff based on the vertex elements, * and not on attrib information. */ - if (r300screen->caps->has_tcl) { - attrib_count = r300->vs->info.num_inputs; - DBG(r300, DBG_DRAW, "r300: routing %d attribs in psc for vs\n", - attrib_count); - } else { - attrib_count = vinfo->num_attribs; - DBG(r300, DBG_DRAW, "r300: attrib count: %d\n", attrib_count); - for (i = 0; i < attrib_count; i++) { - DBG(r300, DBG_DRAW, "r300: attrib: offset %d, interp %d, size %d," - " tab %d\n", vinfo->attrib[i].src_index, - vinfo->attrib[i].interp_mode, vinfo->attrib[i].emit, - tab[i]); + DBG(r300, DBG_DRAW, "r300: vs expects %d attribs, routing %d elements" + " in psc\n", + r300->vs->info.num_inputs, + r300->vertex_element_count); + + for (i = 0; i < r300->vertex_element_count; i++) { + format = r300->vertex_element[i].src_format; + + type = r300_translate_vertex_data_type(format) | + (i << R300_DST_VEC_LOC_SHIFT); + swizzle = r300_translate_vertex_data_swizzle(format); + + if (i % 2) { + vformat->vap_prog_stream_cntl[i >> 1] |= type << 16; + vformat->vap_prog_stream_cntl_ext[i >> 1] |= swizzle << 16; + } else { + vformat->vap_prog_stream_cntl[i >> 1] |= type; + vformat->vap_prog_stream_cntl_ext[i >> 1] |= swizzle; } } + + assert(i <= 15); + + /* Set the last vector in the PSC. */ + if (i) { + i -= 1; + } + vformat->vap_prog_stream_cntl[i >> 1] |= + (R300_LAST_VEC << (i & 1 ? 16 : 0)); +} + +/* Update the PSC tables for SW TCL, using Draw. */ +static void r300_swtcl_vertex_psc(struct r300_context* r300, + int* vs_output_tab) +{ + struct r300_vertex_info *vformat = r300->vertex_info; + struct vertex_info* vinfo = &vformat->vinfo; + uint16_t type, swizzle; + enum pipe_format format; + unsigned i, attrib_count; + + /* For each Draw attribute, route it to the fragment shader according + * to the vs_output_tab. */ + attrib_count = vinfo->num_attribs; + DBG(r300, DBG_DRAW, "r300: attrib count: %d\n", attrib_count); + for (i = 0; i < attrib_count; i++) { + DBG(r300, DBG_DRAW, "r300: attrib: offset %d, interp %d, size %d," + " vs_output_tab %d\n", vinfo->attrib[i].src_index, + vinfo->attrib[i].interp_mode, vinfo->attrib[i].emit, + vs_output_tab[i]); + } + for (i = 0; i < attrib_count; i++) { /* Make sure we have a proper destination for our attribute. */ - assert(tab[i] != -1); + assert(vs_output_tab[i] != -1); format = draw_translate_vinfo_format(vinfo->attrib[i].emit); /* Obtain the type of data in this attribute. */ type = r300_translate_vertex_data_type(format) | - tab[i] << R300_DST_VEC_LOC_SHIFT; + vs_output_tab[i] << R300_DST_VEC_LOC_SHIFT; /* Obtain the swizzle for this attribute. Note that the default * swizzle in the hardware is not XYZW! */ @@ -272,12 +264,10 @@ static void r300_vertex_psc(struct r300_context* r300, /* Add the attribute to the PSC table. */ if (i & 1) { vformat->vap_prog_stream_cntl[i >> 1] |= type << 16; - vformat->vap_prog_stream_cntl_ext[i >> 1] |= swizzle << 16; } else { - vformat->vap_prog_stream_cntl[i >> 1] |= type << 0; - - vformat->vap_prog_stream_cntl_ext[i >> 1] |= swizzle << 0; + vformat->vap_prog_stream_cntl[i >> 1] |= type; + vformat->vap_prog_stream_cntl_ext[i >> 1] |= swizzle; } } @@ -289,74 +279,12 @@ static void r300_vertex_psc(struct r300_context* r300, (R300_LAST_VEC << (i & 1 ? 16 : 0)); } -/* Set up the mappings from GB to US, for RS block. */ -static void r300_update_fs_tab(struct r300_context* r300, - struct r300_vertex_info* vformat) -{ - struct tgsi_shader_info* info = &r300->fs->info; - int i, cols = 0, texs = 0, cols_emitted = 0; - int* tab = vformat->fs_tab; - - for (i = 0; i < 16; i++) { - tab[i] = -1; - } - - assert(info->num_inputs <= 16); - for (i = 0; i < info->num_inputs; i++) { - switch (info->input_semantic_name[i]) { - case TGSI_SEMANTIC_COLOR: - tab[i] = INTERP_LINEAR; - cols++; - break; - case TGSI_SEMANTIC_POSITION: - case TGSI_SEMANTIC_PSIZE: - debug_printf("r300: Implementation error: Can't use " - "pos attribs in fragshader yet!\n"); - /* Pass through for now */ - case TGSI_SEMANTIC_FOG: - case TGSI_SEMANTIC_GENERIC: - tab[i] = INTERP_PERSPECTIVE; - break; - default: - debug_printf("r300: Unknown vertex input %d\n", - info->input_semantic_name[i]); - break; - } - } - - /* Now that we know where everything is... */ - DBG(r300, DBG_DRAW, "r300: fp input count: %d\n", info->num_inputs); - for (i = 0; i < info->num_inputs; i++) { - switch (tab[i]) { - case INTERP_LINEAR: - DBG(r300, DBG_DRAW, "r300: attrib: " - "stack offset %d, color, tab %d\n", - i, cols_emitted); - tab[i] = cols_emitted; - cols_emitted++; - break; - case INTERP_PERSPECTIVE: - DBG(r300, DBG_DRAW, "r300: attrib: " - "stack offset %d, texcoord, tab %d\n", - i, cols + texs); - tab[i] = cols + texs; - texs++; - break; - case -1: - debug_printf("r300: Implementation error: Bad fp interp!\n"); - default: - break; - } - } - -} - /* Set up the RS block. This is the part of the chipset that actually does * the rasterization of vertices into fragments. This is also the part of the * chipset that locks up if any part of it is even slightly wrong. */ -static void r300_update_rs_block(struct r300_context* r300, - struct r300_rs_block* rs) +static void r300_update_rs_block(struct r300_context* r300) { + struct r300_rs_block* rs = r300->rs_block; struct tgsi_shader_info* info = &r300->fs->info; int col_count = 0, fp_offset = 0, i, tex_count = 0; int rs_tex_comp = 0; @@ -390,17 +318,18 @@ static void r300_update_rs_block(struct r300_context* r300, col_count++; } + for (i = 0; i < col_count; i++) { + rs->inst[i] |= R500_RS_INST_COL_ID(i) | + R500_RS_INST_COL_CN_WRITE | R500_RS_INST_COL_ADDR(fp_offset); + fp_offset++; + } + for (i = 0; i < tex_count; i++) { rs->inst[i] |= R500_RS_INST_TEX_ID(i) | R500_RS_INST_TEX_CN_WRITE | R500_RS_INST_TEX_ADDR(fp_offset); fp_offset++; } - for (i = 0; i < col_count; i++) { - rs->inst[i] |= R500_RS_INST_COL_ID(i) | - R500_RS_INST_COL_CN_WRITE | R500_RS_INST_COL_ADDR(fp_offset); - fp_offset++; - } } else { for (i = 0; i < info->num_inputs; i++) { switch (info->input_semantic_name[i]) { @@ -425,8 +354,10 @@ static void r300_update_rs_block(struct r300_context* r300, } } + /* Rasterize at least one color, or bad things happen. */ if (col_count == 0) { rs->ip[0] |= R300_RS_COL_FMT(R300_RS_COL_FMT_0001); + col_count++; } if (tex_count == 0) { @@ -437,9 +368,10 @@ static void r300_update_rs_block(struct r300_context* r300, R300_RS_SEL_Q(R300_RS_SEL_K1); } - /* Rasterize at least one color, or bad things happen. */ - if ((col_count == 0) && (tex_count == 0)) { - col_count++; + for (i = 0; i < col_count; i++) { + rs->inst[i] |= R300_RS_INST_COL_ID(i) | + R300_RS_INST_COL_CN_WRITE | R300_RS_INST_COL_ADDR(fp_offset); + fp_offset++; } for (i = 0; i < tex_count; i++) { @@ -447,28 +379,22 @@ static void r300_update_rs_block(struct r300_context* r300, R300_RS_INST_TEX_CN_WRITE | R300_RS_INST_TEX_ADDR(fp_offset); fp_offset++; } - - for (i = 0; i < col_count; i++) { - rs->inst[i] |= R300_RS_INST_COL_ID(i) | - R300_RS_INST_COL_CN_WRITE | R300_RS_INST_COL_ADDR(fp_offset); - fp_offset++; - } } rs->count = (rs_tex_comp) | (col_count << R300_IC_COUNT_SHIFT) | R300_HIRES_EN; - rs->inst_count = MAX2(MAX2(col_count - 1, tex_count - 1), 0); + rs->inst_count = MAX3(col_count - 1, tex_count - 1, 0); } /* Update the vertex format. */ static void r300_update_derived_shader_state(struct r300_context* r300) { struct r300_screen* r300screen = r300_screen(r300->context.screen); - struct r300_vertex_info* vformat; - struct r300_rs_block* rs_block; + int vs_output_tab[16]; int i; + /* struct r300_shader_key* key; struct r300_shader_derived_value* value; @@ -495,27 +421,26 @@ static void r300_update_derived_shader_state(struct r300_context* r300) (void*)key, (void*)value); } */ - /* XXX This will be refactored ASAP. */ - vformat = CALLOC_STRUCT(r300_vertex_info); - rs_block = CALLOC_STRUCT(r300_rs_block); + /* Reset structures */ + memset(r300->rs_block, 0, sizeof(struct r300_rs_block)); + memset(r300->vertex_info, 0, sizeof(struct r300_vertex_info)); for (i = 0; i < 16; i++) { - vformat->vs_tab[i] = -1; - vformat->fs_tab[i] = -1; + vs_output_tab[i] = -1; } - r300_vs_tab_routes(r300, vformat); - r300_vertex_psc(r300, vformat); - r300_update_fs_tab(r300, vformat); + /* Update states */ + r300_vs_output_tab_routes(r300, vs_output_tab); - r300_update_rs_block(r300, rs_block); + if (r300screen->caps->has_tcl) { + r300_vertex_psc(r300); + } else { + r300_swtcl_vertex_psc(r300, vs_output_tab); + } - FREE(r300->vertex_info); - FREE(r300->rs_block); + r300_update_rs_block(r300); - r300->vertex_info = vformat; - r300->rs_block = rs_block; - r300->dirty_state |= (R300_NEW_VERTEX_FORMAT | R300_NEW_RS_BLOCK); + r300->dirty_state |= R300_NEW_RS_BLOCK; } static void r300_update_ztop(struct r300_context* r300) @@ -553,9 +478,9 @@ static void r300_update_ztop(struct r300_context* r300) void r300_update_derived_state(struct r300_context* r300) { - /* XXX */ - if (TRUE || r300->dirty_state & - (R300_NEW_FRAGMENT_SHADER | R300_NEW_VERTEX_SHADER)) { + if (r300->dirty_state & + (R300_NEW_FRAGMENT_SHADER | R300_NEW_VERTEX_SHADER | + R300_NEW_VERTEX_FORMAT)) { r300_update_derived_shader_state(r300); } |