diff options
Diffstat (limited to 'src/gallium/drivers/r300/r300_state_derived.c')
-rw-r--r-- | src/gallium/drivers/r300/r300_state_derived.c | 241 |
1 files changed, 164 insertions, 77 deletions
diff --git a/src/gallium/drivers/r300/r300_state_derived.c b/src/gallium/drivers/r300/r300_state_derived.c index c01e61a9b1..42aee7231e 100644 --- a/src/gallium/drivers/r300/r300_state_derived.c +++ b/src/gallium/drivers/r300/r300_state_derived.c @@ -20,14 +20,47 @@ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE * USE OR OTHER DEALINGS IN THE SOFTWARE. */ -#include "r300_state_derived.h" +#include "draw/draw_context.h" + +#include "util/u_math.h" +#include "util/u_memory.h" +#include "r300_context.h" #include "r300_fs.h" +#include "r300_screen.h" +#include "r300_state_derived.h" +#include "r300_state_inlines.h" #include "r300_vs.h" /* r300_state_derived: Various bits of state which are dependent upon * currently bound CSO data. */ +struct r300_shader_key { + struct r300_vertex_shader* vs; + struct r300_fragment_shader* fs; +}; + +struct r300_shader_derived_value { + struct r300_vertex_format* vformat; + struct r300_rs_block* rs_block; +}; + +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; + + return (vs << 16) | (fs & 0xffff); +} + +int r300_shader_key_compare(void* key1, void* key2) { + struct r300_shader_key* shader_key1 = (struct r300_shader_key*)key1; + struct r300_shader_key* shader_key2 = (struct r300_shader_key*)key2; + + return (shader_key1->vs == shader_key2->vs) && + (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_format* vformat) @@ -52,7 +85,7 @@ static void r300_vs_tab_routes(struct r300_context* r300, if (!r300screen->caps->has_tcl || !r300->rs_state->enable_vte) { for (i = 0; i < info->num_inputs; i++) { - switch (info->input_semantic_name[i]) { + switch (r300->vs->code.inputs[i]) { case TGSI_SEMANTIC_POSITION: pos = TRUE; tab[i] = 0; @@ -62,10 +95,12 @@ static void r300_vs_tab_routes(struct r300_context* r300, 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: @@ -124,7 +159,9 @@ static void r300_vs_tab_routes(struct r300_context* r300, vinfo->hwfmt[0] = 0x5555; /* XXX this is classic Mesa bonghits */ - if (!pos) { + /* 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"); /* Make room for the position attribute * at the beginning of the tab. */ @@ -163,7 +200,7 @@ static void r300_vs_tab_routes(struct r300_context* r300, vinfo->hwfmt[3] |= (4 << (3 * i)); } - for (i; i < texs; i++) { + for (; i < texs; i++) { draw_emit_vertex_attr(vinfo, EMIT_4F, INTERP_PERSPECTIVE, draw_find_vs_output(r300->draw, TGSI_SEMANTIC_GENERIC, i)); vinfo->hwfmt[1] |= (R300_INPUT_CNTL_TC0 << i); @@ -188,20 +225,20 @@ static void r300_vertex_psc(struct r300_context* r300, struct vertex_info* vinfo = &vformat->vinfo; int* tab = vformat->vs_tab; uint32_t temp; - int i, attrib_count; + unsigned i, attrib_count; /* Vertex shaders have no semantics on their inputs, * so PSC should just route stuff based on their info, * and not on attrib information. */ if (r300screen->caps->has_tcl) { attrib_count = r300->vs->info.num_inputs; - debug_printf("r300: routing %d attribs in psc for vs\n", + DBG(r300, DBG_DRAW, "r300: routing %d attribs in psc for vs\n", attrib_count); } else { attrib_count = vinfo->num_attribs; - debug_printf("r300: attrib count: %d\n", attrib_count); + DBG(r300, DBG_DRAW, "r300: attrib count: %d\n", attrib_count); for (i = 0; i < attrib_count; i++) { - debug_printf("r300: attrib: offset %d, interp %d, size %d," + 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]); @@ -213,10 +250,8 @@ static void r300_vertex_psc(struct r300_context* r300, assert(tab[i] != -1); /* Add the attribute to the PSC table. */ - temp = r300screen->caps->has_tcl ? - R300_DATA_TYPE_FLOAT_4 : - translate_vertex_data_type(vinfo->attrib[i].emit); - temp |= tab[i] << R300_DST_VEC_LOC_SHIFT; + temp = translate_draw_vertex_data_type(vinfo->attrib[i].emit) | + tab[i] << R300_DST_VEC_LOC_SHIFT; if (i & 1) { vformat->vap_prog_stream_cntl[i >> 1] &= 0x0000ffff; @@ -234,39 +269,17 @@ static void r300_vertex_psc(struct r300_context* r300, } /* Set the last vector in the PSC. */ - i--; + if (i) { + i -= 1; + } vformat->vap_prog_stream_cntl[i >> 1] |= (R300_LAST_VEC << (i & 1 ? 16 : 0)); } -/* Update the vertex format. */ -static void r300_update_vertex_format(struct r300_context* r300) -{ - struct r300_vertex_format vformat; - int i; - - memset(&vformat, 0, sizeof(struct r300_vertex_format)); - for (i = 0; i < 16; i++) { - vformat.vs_tab[i] = -1; - vformat.fs_tab[i] = -1; - } - - r300_vs_tab_routes(r300, &vformat); - - r300_vertex_psc(r300, &vformat); - - if (memcmp(&r300->vertex_info, &vformat, - sizeof(struct r300_vertex_format))) { - memcpy(&r300->vertex_info, &vformat, - sizeof(struct r300_vertex_format)); - r300->dirty_state |= R300_NEW_VERTEX_FORMAT; - } -} - /* Set up the mappings from GB to US, for RS block. */ -static void r300_update_fs_tab(struct r300_context* r300) +static void r300_update_fs_tab(struct r300_context* r300, + struct r300_vertex_format* vformat) { - struct r300_vertex_format* vformat = &r300->vertex_info; struct tgsi_shader_info* info = &r300->fs->info; int i, cols = 0, texs = 0, cols_emitted = 0; int* tab = vformat->fs_tab; @@ -299,18 +312,18 @@ static void r300_update_fs_tab(struct r300_context* r300) } /* Now that we know where everything is... */ - debug_printf("r300: fp input count: %d\n", info->num_inputs); + 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: - debug_printf("r300: attrib: " + 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: - debug_printf("r300: attrib: " + DBG(r300, DBG_DRAW, "r300: attrib: " "stack offset %d, texcoord, tab %d\n", i, cols + texs); tab[i] = cols + texs; @@ -328,53 +341,39 @@ static void r300_update_fs_tab(struct r300_context* r300) /* 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) +static void r300_update_rs_block(struct r300_context* r300, + struct r300_rs_block* rs) { - struct r300_rs_block* rs = r300->rs_block; struct tgsi_shader_info* info = &r300->fs->info; - int* tab = r300->vertex_info.fs_tab; - int col_count = 0, fp_offset = 0, i, memory_pos, tex_count = 0; - - memset(rs, 0, sizeof(struct r300_rs_block)); + int col_count = 0, fp_offset = 0, i, tex_count = 0; + int rs_tex_comp = 0; if (r300_screen(r300->context.screen)->caps->is_r500) { for (i = 0; i < info->num_inputs; i++) { - assert(tab[i] != -1); - memory_pos = tab[i] * 4; switch (info->input_semantic_name[i]) { case TGSI_SEMANTIC_COLOR: rs->ip[col_count] |= - R500_RS_COL_PTR(memory_pos) | + R500_RS_COL_PTR(col_count) | R500_RS_COL_FMT(R300_RS_COL_FMT_RGBA); col_count++; break; case TGSI_SEMANTIC_GENERIC: rs->ip[tex_count] |= - R500_RS_SEL_S(memory_pos) | - R500_RS_SEL_T(memory_pos + 1) | - R500_RS_SEL_R(memory_pos + 2) | - R500_RS_SEL_Q(memory_pos + 3); + R500_RS_SEL_S(rs_tex_comp) | + R500_RS_SEL_T(rs_tex_comp + 1) | + R500_RS_SEL_R(rs_tex_comp + 2) | + R500_RS_SEL_Q(rs_tex_comp + 3); tex_count++; + rs_tex_comp += 4; break; default: break; } } - if (col_count == 0) { - rs->ip[0] |= R500_RS_COL_FMT(R300_RS_COL_FMT_0001); - } - - if (tex_count == 0) { - rs->ip[0] |= - R500_RS_SEL_S(R500_RS_IP_PTR_K0) | - R500_RS_SEL_T(R500_RS_IP_PTR_K0) | - R500_RS_SEL_R(R500_RS_IP_PTR_K0) | - R500_RS_SEL_Q(R500_RS_IP_PTR_K1); - } - /* Rasterize at least one color, or bad things happen. */ if ((col_count == 0) && (tex_count == 0)) { + rs->ip[0] |= R500_RS_COL_FMT(R300_RS_COL_FMT_0001); col_count++; } @@ -391,23 +390,22 @@ static void r300_update_rs_block(struct r300_context* r300) } } else { for (i = 0; i < info->num_inputs; i++) { - assert(tab[i] != -1); - memory_pos = tab[i] * 4; switch (info->input_semantic_name[i]) { case TGSI_SEMANTIC_COLOR: rs->ip[col_count] |= - R300_RS_COL_PTR(memory_pos) | + R300_RS_COL_PTR(col_count) | R300_RS_COL_FMT(R300_RS_COL_FMT_RGBA); col_count++; break; case TGSI_SEMANTIC_GENERIC: rs->ip[tex_count] |= - R300_RS_TEX_PTR(memory_pos) | + R300_RS_TEX_PTR(rs_tex_comp) | R300_RS_SEL_S(R300_RS_SEL_C0) | R300_RS_SEL_T(R300_RS_SEL_C1) | R300_RS_SEL_R(R300_RS_SEL_C2) | R300_RS_SEL_Q(R300_RS_SEL_C3); tex_count++; + rs_tex_comp+=4; break; default: break; @@ -444,21 +442,110 @@ static void r300_update_rs_block(struct r300_context* r300) } } - rs->count = (tex_count * 4) | (col_count << R300_IC_COUNT_SHIFT) | + 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); } +/* Update the vertex format. */ +static void r300_update_derived_shader_state(struct r300_context* r300) +{ + struct r300_shader_key* key; + struct r300_vertex_format* vformat; + struct r300_rs_block* rs_block; + struct r300_shader_derived_value* value; + int i; + + /* + key = CALLOC_STRUCT(r300_shader_key); + key->vs = r300->vs; + key->fs = r300->fs; + + value = (struct r300_shader_derived_value*) + util_hash_table_get(r300->shader_hash_table, (void*)key); + if (value) { + //vformat = value->vformat; + rs_block = value->rs_block; + + FREE(key); + } else { + rs_block = CALLOC_STRUCT(r300_rs_block); + value = CALLOC_STRUCT(r300_shader_derived_value); + + r300_update_rs_block(r300, rs_block); + + //value->vformat = vformat; + value->rs_block = rs_block; + util_hash_table_set(r300->shader_hash_table, + (void*)key, (void*)value); + } */ + + /* XXX This will be refactored ASAP. */ + vformat = CALLOC_STRUCT(r300_vertex_format); + rs_block = CALLOC_STRUCT(r300_rs_block); + + for (i = 0; i < 16; i++) { + vformat->vs_tab[i] = -1; + vformat->fs_tab[i] = -1; + } + + r300_vs_tab_routes(r300, vformat); + r300_vertex_psc(r300, vformat); + r300_update_fs_tab(r300, vformat); + + r300_update_rs_block(r300, rs_block); + + FREE(r300->vertex_info); + FREE(r300->rs_block); + + r300->vertex_info = vformat; + r300->rs_block = rs_block; + r300->dirty_state |= (R300_NEW_VERTEX_FORMAT | R300_NEW_RS_BLOCK); +} + +static void r300_update_ztop(struct r300_context* r300) +{ + r300->ztop_state.z_buffer_top = R300_ZTOP_ENABLE; + + /* This is important enough that I felt it warranted a comment. + * + * According to the docs, these are the conditions where ZTOP must be + * disabled: + * 1) Alpha testing enabled + * 2) Texture kill instructions in fragment shader + * 3) Chroma key culling enabled + * 4) W-buffering enabled + * + * The docs claim that for the first three cases, if no ZS writes happen, + * then ZTOP can be used. + * + * Additionally, the following conditions require disabled ZTOP: + * ~) Depth writes in fragment shader + * ~) Outstanding occlusion queries + * + * ~C. + */ + if (r300->dsa_state->alpha_function) { + r300->ztop_state.z_buffer_top = R300_ZTOP_DISABLE; + } else if (r300->fs->info.uses_kill) { + r300->ztop_state.z_buffer_top = R300_ZTOP_DISABLE; + } else if (r300_fragment_shader_writes_depth(r300->fs)) { + r300->ztop_state.z_buffer_top = R300_ZTOP_DISABLE; + } else if (r300->query_current) { + r300->ztop_state.z_buffer_top = R300_ZTOP_DISABLE; + } +} + void r300_update_derived_state(struct r300_context* r300) { if (r300->dirty_state & - (R300_NEW_FRAGMENT_SHADER | R300_NEW_VERTEX_SHADER)) { - r300_update_vertex_format(r300); + (R300_NEW_FRAGMENT_SHADER | R300_NEW_VERTEX_SHADER)) { + r300_update_derived_shader_state(r300); } - if (r300->dirty_state & R300_NEW_VERTEX_FORMAT) { - r300_update_fs_tab(r300); - r300_update_rs_block(r300); + if (r300->dirty_state & + (R300_NEW_DSA | R300_NEW_FRAGMENT_SHADER | R300_NEW_QUERY)) { + r300_update_ztop(r300); } } |