src/gallium/drivers/r300/r300_state_derived.c

   1 /*
   2  * Copyright 2008 Corbin Simpson <MostAwesomeDude@gmail.com>
   3  *
   4  * Permission is hereby granted, free of charge, to any person obtaining a
   5  * copy of this software and associated documentation files (the "Software"),
   6  * to deal in the Software without restriction, including without limitation
   7  * on the rights to use, copy, modify, merge, publish, distribute, sub
   8  * license, and/or sell copies of the Software, and to permit persons to whom
   9  * the Software is furnished to do so, subject to the following conditions:
  10  *
  11  * The above copyright notice and this permission notice (including the next
  12  * paragraph) shall be included in all copies or substantial portions of the
  13  * Software.
  14  *
  15  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  16  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  17  * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
  18  * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
  19  * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
  20  * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
  21  * USE OR OTHER DEALINGS IN THE SOFTWARE. */
  22
  23 #include "r300_state_derived.h"
  24
  25 #include "r300_fs.h"
  26 #include "r300_state_inlines.h"
  27 #include "r300_vs.h"
  28
  29 /* r300_state_derived: Various bits of state which are dependent upon
  30  * currently bound CSO data. */
  31
  32 struct r300_shader_key {
  33     struct r300_vertex_shader* vs;
  34     struct r300_fragment_shader* fs;
  35 };
  36
  37 struct r300_shader_derived_value {
  38     struct r300_vertex_format* vformat;
  39     struct r300_rs_block* rs_block;
  40 };
  41
  42 unsigned r300_shader_key_hash(void* key) {
  43     struct r300_shader_key* shader_key = (struct r300_shader_key*)key;
  44     unsigned vs = (unsigned)shader_key->vs;
  45     unsigned fs = (unsigned)shader_key->fs;
  46
  47     return (vs << 16) | (fs & 0xffff);
  48 }
  49
  50 int r300_shader_key_compare(void* key1, void* key2) {
  51     struct r300_shader_key* shader_key1 = (struct r300_shader_key*)key1;
  52     struct r300_shader_key* shader_key2 = (struct r300_shader_key*)key2;
  53
  54     return (shader_key1->vs == shader_key2->vs) &&
  55         (shader_key1->fs == shader_key2->fs);
  56 }
  57
  58 /* Set up the vs_tab and routes. */
  59 static void r300_vs_tab_routes(struct r300_context* r300,
  60                                struct r300_vertex_format* vformat)
  61 {
  62     struct r300_screen* r300screen = r300_screen(r300->context.screen);
  63     struct vertex_info* vinfo = &vformat->vinfo;
  64     int* tab = vformat->vs_tab;
  65     boolean pos = FALSE, psize = FALSE, fog = FALSE;
  66     int i, texs = 0, cols = 0;
  67     struct tgsi_shader_info* info;
  68
  69     if (r300screen->caps->has_tcl) {
  70         /* Use vertex shader to determine required routes. */
  71         info = &r300->vs->info;
  72     } else {
  73         /* Use fragment shader to determine required routes. */
  74         info = &r300->fs->info;
  75     }
  76
  77     assert(info->num_inputs <= 16);
  78
  79     if (!r300screen->caps->has_tcl || !r300->rs_state->enable_vte)
  80     {
  81         for (i = 0; i < info->num_inputs; i++) {
  82             switch (r300->vs->code.inputs[i]) {
  83                 case TGSI_SEMANTIC_POSITION:
  84                     pos = TRUE;
  85                     tab[i] = 0;
  86                     break;
  87                 case TGSI_SEMANTIC_COLOR:
  88                     tab[i] = 2 + cols;
  89                     cols++;
  90                     break;
  91                 case TGSI_SEMANTIC_PSIZE:
  92                     assert(psize == FALSE);
  93                     psize = TRUE;
  94                     tab[i] = 15;
  95                     break;
  96                 case TGSI_SEMANTIC_FOG:
  97                     assert(fog == FALSE);
  98                     fog = TRUE;
  99                     /* Fall through */
 100                 case TGSI_SEMANTIC_GENERIC:
 101                     tab[i] = 6 + texs;
 102                     texs++;
 103                     break;
 104                 default:
 105                     debug_printf("r300: Unknown vertex input %d\n",
 106                         info->input_semantic_name[i]);
 107                     break;
 108             }
 109         }
 110     }
 111     else
 112     {
 113         /* Just copy vert attribs over as-is. */
 114         for (i = 0; i < info->num_inputs; i++) {
 115             tab[i] = i;
 116         }
 117
 118         for (i = 0; i < info->num_outputs; i++) {
 119             switch (info->output_semantic_name[i]) {
 120                 case TGSI_SEMANTIC_POSITION:
 121                     pos = TRUE;
 122                     break;
 123                 case TGSI_SEMANTIC_COLOR:
 124                     cols++;
 125                     break;
 126                 case TGSI_SEMANTIC_PSIZE:
 127                     psize = TRUE;
 128                     break;
 129                 case TGSI_SEMANTIC_FOG:
 130                     fog = TRUE;
 131                     /* Fall through */
 132                 case TGSI_SEMANTIC_GENERIC:
 133                     texs++;
 134                     break;
 135                 default:
 136                     debug_printf("r300: Unknown vertex output %d\n",
 137                         info->output_semantic_name[i]);
 138                     break;
 139             }
 140         }
 141     }
 142
 143     /* XXX magic */
 144     assert(texs <= 8);
 145
 146     /* Do the actual vertex_info setup.
 147      *
 148      * vertex_info has four uints of hardware-specific data in it.
 149      * vinfo.hwfmt[0] is R300_VAP_VTX_STATE_CNTL
 150      * vinfo.hwfmt[1] is R300_VAP_VSM_VTX_ASSM
 151      * vinfo.hwfmt[2] is R300_VAP_OUTPUT_VTX_FMT_0
 152      * vinfo.hwfmt[3] is R300_VAP_OUTPUT_VTX_FMT_1 */
 153
 154     vinfo->hwfmt[0] = 0x5555; /* XXX this is classic Mesa bonghits */
 155
 156     /* We need to add vertex position attribute only for SW TCL case,
 157      * for HW TCL case it could be generated by vertex shader */
 158     if (!pos && !r300screen->caps->has_tcl) {
 159         debug_printf("r300: Forcing vertex position attribute emit...\n");
 160         /* Make room for the position attribute
 161          * at the beginning of the tab. */
 162         for (i = 15; i > 0; i--) {
 163             tab[i] = tab[i-1];
 164         }
 165         tab[0] = 0;
 166     }
 167     draw_emit_vertex_attr(vinfo, EMIT_4F, INTERP_PERSPECTIVE,
 168         draw_find_vs_output(r300->draw, TGSI_SEMANTIC_POSITION, 0));
 169     vinfo->hwfmt[1] |= R300_INPUT_CNTL_POS;
 170     vinfo->hwfmt[2] |= R300_VAP_OUTPUT_VTX_FMT_0__POS_PRESENT;
 171
 172     if (psize) {
 173         draw_emit_vertex_attr(vinfo, EMIT_1F_PSIZE, INTERP_POS,
 174             draw_find_vs_output(r300->draw, TGSI_SEMANTIC_PSIZE, 0));
 175         vinfo->hwfmt[2] |= R300_VAP_OUTPUT_VTX_FMT_0__PT_SIZE_PRESENT;
 176     }
 177
 178     for (i = 0; i < cols; i++) {
 179         draw_emit_vertex_attr(vinfo, EMIT_4F, INTERP_LINEAR,
 180             draw_find_vs_output(r300->draw, TGSI_SEMANTIC_COLOR, i));
 181         vinfo->hwfmt[1] |= R300_INPUT_CNTL_COLOR;
 182         vinfo->hwfmt[2] |= (R300_VAP_OUTPUT_VTX_FMT_0__COLOR_0_PRESENT << i);
 183     }
 184
 185     /* Init i right here, increment it if fog is enabled.
 186      * This gets around a double-increment problem. */
 187     i = 0;
 188
 189     if (fog) {
 190         i++;
 191         draw_emit_vertex_attr(vinfo, EMIT_4F, INTERP_PERSPECTIVE,
 192             draw_find_vs_output(r300->draw, TGSI_SEMANTIC_FOG, 0));
 193         vinfo->hwfmt[1] |= (R300_INPUT_CNTL_TC0 << i);
 194         vinfo->hwfmt[3] |= (4 << (3 * i));
 195     }
 196
 197     for (; i < texs; i++) {
 198         draw_emit_vertex_attr(vinfo, EMIT_4F, INTERP_PERSPECTIVE,
 199             draw_find_vs_output(r300->draw, TGSI_SEMANTIC_GENERIC, i));
 200         vinfo->hwfmt[1] |= (R300_INPUT_CNTL_TC0 << i);
 201         vinfo->hwfmt[3] |= (4 << (3 * i));
 202     }
 203
 204     /* Handle the case where the vertex shader will be generating some of
 205      * the attribs based on its inputs. */
 206     if (r300screen->caps->has_tcl &&
 207             info->num_inputs < info->num_outputs) {
 208         vinfo->num_attribs = info->num_inputs;
 209     }
 210
 211     draw_compute_vertex_size(vinfo);
 212 }
 213
 214 /* Update the PSC tables. */
 215 static void r300_vertex_psc(struct r300_context* r300,
 216                             struct r300_vertex_format* vformat)
 217 {
 218     struct r300_screen* r300screen = r300_screen(r300->context.screen);
 219     struct vertex_info* vinfo = &vformat->vinfo;
 220     int* tab = vformat->vs_tab;
 221     uint32_t temp;
 222     unsigned i, attrib_count;
 223
 224     /* Vertex shaders have no semantics on their inputs,
 225      * so PSC should just route stuff based on their info,
 226      * and not on attrib information. */
 227     if (r300screen->caps->has_tcl) {
 228         attrib_count = r300->vs->info.num_inputs;
 229         DBG(r300, DBG_DRAW, "r300: routing %d attribs in psc for vs\n",
 230                 attrib_count);
 231     } else {
 232         attrib_count = vinfo->num_attribs;
 233         DBG(r300, DBG_DRAW, "r300: attrib count: %d\n", attrib_count);
 234         for (i = 0; i < attrib_count; i++) {
 235             DBG(r300, DBG_DRAW, "r300: attrib: offset %d, interp %d, size %d,"
 236                    " tab %d\n", vinfo->attrib[i].src_index,
 237                    vinfo->attrib[i].interp_mode, vinfo->attrib[i].emit,
 238                    tab[i]);
 239         }
 240     }
 241
 242     for (i = 0; i < attrib_count; i++) {
 243         /* Make sure we have a proper destination for our attribute */
 244         assert(tab[i] != -1);
 245
 246         /* Add the attribute to the PSC table. */
 247         temp = translate_vertex_data_type(vinfo->attrib[i].emit) |
 248             tab[i] << R300_DST_VEC_LOC_SHIFT;
 249
 250         if (i & 1) {
 251             vformat->vap_prog_stream_cntl[i >> 1] &= 0x0000ffff;
 252             vformat->vap_prog_stream_cntl[i >> 1] |= temp << 16;
 253
 254             vformat->vap_prog_stream_cntl_ext[i >> 1] |=
 255                 (R300_VAP_SWIZZLE_XYZW << 16);
 256         } else {
 257             vformat->vap_prog_stream_cntl[i >> 1] &= 0xffff0000;
 258             vformat->vap_prog_stream_cntl[i >> 1] |= temp <<  0;
 259
 260             vformat->vap_prog_stream_cntl_ext[i >> 1] |=
 261                 (R300_VAP_SWIZZLE_XYZW <<  0);
 262         }
 263     }
 264
 265     /* Set the last vector in the PSC. */
 266     if (i) {
 267         i -= 1;
 268     }
 269     vformat->vap_prog_stream_cntl[i >> 1] |=
 270         (R300_LAST_VEC << (i & 1 ? 16 : 0));
 271 }
 272
 273 /* Set up the mappings from GB to US, for RS block. */
 274 static void r300_update_fs_tab(struct r300_context* r300,
 275                                struct r300_vertex_format* vformat)
 276 {
 277     struct tgsi_shader_info* info = &r300->fs->info;
 278     int i, cols = 0, texs = 0, cols_emitted = 0;
 279     int* tab = vformat->fs_tab;
 280
 281     for (i = 0; i < 16; i++) {
 282         tab[i] = -1;
 283     }
 284
 285     assert(info->num_inputs <= 16);
 286     for (i = 0; i < info->num_inputs; i++) {
 287         switch (info->input_semantic_name[i]) {
 288             case TGSI_SEMANTIC_COLOR:
 289                 tab[i] = INTERP_LINEAR;
 290                 cols++;
 291                 break;
 292             case TGSI_SEMANTIC_POSITION:
 293             case TGSI_SEMANTIC_PSIZE:
 294                 debug_printf("r300: Implementation error: Can't use "
 295                         "pos attribs in fragshader yet!\n");
 296                 /* Pass through for now */
 297             case TGSI_SEMANTIC_FOG:
 298             case TGSI_SEMANTIC_GENERIC:
 299                 tab[i] = INTERP_PERSPECTIVE;
 300                 break;
 301             default:
 302                 debug_printf("r300: Unknown vertex input %d\n",
 303                     info->input_semantic_name[i]);
 304                 break;
 305         }
 306     }
 307
 308     /* Now that we know where everything is... */
 309     DBG(r300, DBG_DRAW, "r300: fp input count: %d\n", info->num_inputs);
 310     for (i = 0; i < info->num_inputs; i++) {
 311         switch (tab[i]) {
 312             case INTERP_LINEAR:
 313                 DBG(r300, DBG_DRAW, "r300: attrib: "
 314                         "stack offset %d, color,    tab %d\n",
 315                         i, cols_emitted);
 316                 tab[i] = cols_emitted;
 317                 cols_emitted++;
 318                 break;
 319             case INTERP_PERSPECTIVE:
 320                 DBG(r300, DBG_DRAW, "r300: attrib: "
 321                         "stack offset %d, texcoord, tab %d\n",
 322                         i, cols + texs);
 323                 tab[i] = cols + texs;
 324                 texs++;
 325                 break;
 326             case -1:
 327                 debug_printf("r300: Implementation error: Bad fp interp!\n");
 328             default:
 329                 break;
 330         }
 331     }
 332
 333 }
 334
 335 /* Set up the RS block. This is the part of the chipset that actually does
 336  * the rasterization of vertices into fragments. This is also the part of the
 337  * chipset that locks up if any part of it is even slightly wrong. */
 338 static void r300_update_rs_block(struct r300_context* r300,
 339                                  struct r300_rs_block* rs)
 340 {
 341     struct tgsi_shader_info* info = &r300->fs->info;
 342     int col_count = 0, fp_offset = 0, i, tex_count = 0;
 343     int rs_tex_comp = 0;
 344
 345     if (r300_screen(r300->context.screen)->caps->is_r500) {
 346         for (i = 0; i < info->num_inputs; i++) {
 347             switch (info->input_semantic_name[i]) {
 348                 case TGSI_SEMANTIC_COLOR:
 349                     rs->ip[col_count] |=
 350                         R500_RS_COL_PTR(col_count) |
 351                         R500_RS_COL_FMT(R300_RS_COL_FMT_RGBA);
 352                     col_count++;
 353                     break;
 354                 case TGSI_SEMANTIC_GENERIC:
 355                     rs->ip[tex_count] |=
 356                         R500_RS_SEL_S(rs_tex_comp) |
 357                         R500_RS_SEL_T(rs_tex_comp + 1) |
 358                         R500_RS_SEL_R(rs_tex_comp + 2) |
 359                         R500_RS_SEL_Q(rs_tex_comp + 3);
 360                     tex_count++;
 361                     rs_tex_comp += 4;
 362                     break;
 363                 default:
 364                     break;
 365             }
 366         }
 367
 368         /* Rasterize at least one color, or bad things happen. */
 369         if ((col_count == 0) && (tex_count == 0)) {
 370             rs->ip[0] |= R500_RS_COL_FMT(R300_RS_COL_FMT_0001);
 371             col_count++;
 372         }
 373
 374         for (i = 0; i < tex_count; i++) {
 375             rs->inst[i] |= R500_RS_INST_TEX_ID(i) |
 376                 R500_RS_INST_TEX_CN_WRITE | R500_RS_INST_TEX_ADDR(fp_offset);
 377             fp_offset++;
 378         }
 379
 380         for (i = 0; i < col_count; i++) {
 381             rs->inst[i] |= R500_RS_INST_COL_ID(i) |
 382                 R500_RS_INST_COL_CN_WRITE | R500_RS_INST_COL_ADDR(fp_offset);
 383             fp_offset++;
 384         }
 385     } else {
 386         for (i = 0; i < info->num_inputs; i++) {
 387             switch (info->input_semantic_name[i]) {
 388                 case TGSI_SEMANTIC_COLOR:
 389                     rs->ip[col_count] |=
 390                         R300_RS_COL_PTR(col_count) |
 391                         R300_RS_COL_FMT(R300_RS_COL_FMT_RGBA);
 392                     col_count++;
 393                     break;
 394                 case TGSI_SEMANTIC_GENERIC:
 395                     rs->ip[tex_count] |=
 396                         R300_RS_TEX_PTR(rs_tex_comp) |
 397                         R300_RS_SEL_S(R300_RS_SEL_C0) |
 398                         R300_RS_SEL_T(R300_RS_SEL_C1) |
 399                         R300_RS_SEL_R(R300_RS_SEL_C2) |
 400                         R300_RS_SEL_Q(R300_RS_SEL_C3);
 401                     tex_count++;
 402                     rs_tex_comp+=4;
 403                     break;
 404                 default:
 405                     break;
 406             }
 407         }
 408
 409         if (col_count == 0) {
 410             rs->ip[0] |= R300_RS_COL_FMT(R300_RS_COL_FMT_0001);
 411         }
 412
 413         if (tex_count == 0) {
 414             rs->ip[0] |=
 415                 R300_RS_SEL_S(R300_RS_SEL_K0) |
 416                 R300_RS_SEL_T(R300_RS_SEL_K0) |
 417                 R300_RS_SEL_R(R300_RS_SEL_K0) |
 418                 R300_RS_SEL_Q(R300_RS_SEL_K1);
 419         }
 420
 421         /* Rasterize at least one color, or bad things happen. */
 422         if ((col_count == 0) && (tex_count == 0)) {
 423             col_count++;
 424         }
 425
 426         for (i = 0; i < tex_count; i++) {
 427             rs->inst[i] |= R300_RS_INST_TEX_ID(i) |
 428                 R300_RS_INST_TEX_CN_WRITE | R300_RS_INST_TEX_ADDR(fp_offset);
 429             fp_offset++;
 430         }
 431
 432         for (i = 0; i < col_count; i++) {
 433             rs->inst[i] |= R300_RS_INST_COL_ID(i) |
 434                 R300_RS_INST_COL_CN_WRITE | R300_RS_INST_COL_ADDR(fp_offset);
 435             fp_offset++;
 436         }
 437     }
 438
 439     rs->count = (rs_tex_comp) | (col_count << R300_IC_COUNT_SHIFT) |
 440         R300_HIRES_EN;
 441
 442     rs->inst_count = MAX2(MAX2(col_count - 1, tex_count - 1), 0);
 443 }
 444
 445 /* Update the vertex format. */
 446 static void r300_update_derived_shader_state(struct r300_context* r300)
 447 {
 448     struct r300_shader_key* key;
 449     struct r300_vertex_format* vformat;
 450     struct r300_rs_block* rs_block;
 451     struct r300_shader_derived_value* value;
 452     int i;
 453
 454     key = CALLOC_STRUCT(r300_shader_key);
 455     key->vs = r300->vs;
 456     key->fs = r300->fs;
 457
 458     value = (struct r300_shader_derived_value*)
 459         util_hash_table_get(r300->shader_hash_table, (void*)key);
 460     if (value) {
 461         //vformat = value->vformat;
 462         rs_block = value->rs_block;
 463
 464         FREE(key);
 465     } else {
 466         rs_block = CALLOC_STRUCT(r300_rs_block);
 467         value = CALLOC_STRUCT(r300_shader_derived_value);
 468
 469         r300_update_rs_block(r300, rs_block);
 470
 471         //value->vformat = vformat;
 472         value->rs_block = rs_block;
 473         util_hash_table_set(r300->shader_hash_table,
 474             (void*)key, (void*)value);
 475     }
 476
 477     /* XXX This will be refactored ASAP. */
 478     vformat = CALLOC_STRUCT(r300_vertex_format);
 479
 480     for (i = 0; i < 16; i++) {
 481         vformat->vs_tab[i] = -1;
 482         vformat->fs_tab[i] = -1;
 483     }
 484
 485     r300_vs_tab_routes(r300, vformat);
 486     r300_vertex_psc(r300, vformat);
 487     r300_update_fs_tab(r300, vformat);
 488
 489     FREE(r300->vertex_info);
 490
 491     r300->vertex_info = vformat;
 492     r300->rs_block = rs_block;
 493     r300->dirty_state |= (R300_NEW_VERTEX_FORMAT | R300_NEW_RS_BLOCK);
 494 }
 495
 496 static void r300_update_ztop(struct r300_context* r300)
 497 {
 498     r300->ztop_state.z_buffer_top = R300_ZTOP_ENABLE;
 499
 500     /* This is important enough that I felt it warranted a comment.
 501      *
 502      * According to the docs, these are the conditions where ZTOP must be
 503      * disabled:
 504      * 1) Alpha testing enabled
 505      * 2) Texture kill instructions in fragment shader
 506      * 3) Chroma key culling enabled
 507      * 4) W-buffering enabled
 508      *
 509      * The docs claim that for the first three cases, if no ZS writes happen,
 510      * then ZTOP can be used.
 511      *
 512      * Additionally, the following conditions require disabled ZTOP:
 513      * ~) Depth writes in fragment shader
 514      * ~) Outstanding occlusion queries
 515      *
 516      * ~C.
 517      */
 518     if (r300->dsa_state->alpha_function) {
 519         r300->ztop_state.z_buffer_top = R300_ZTOP_DISABLE;
 520     } else if (r300->fs->info.uses_kill) {
 521         r300->ztop_state.z_buffer_top = R300_ZTOP_DISABLE;
 522     } else if (r300_fragment_shader_writes_depth(r300->fs)) {
 523         r300->ztop_state.z_buffer_top = R300_ZTOP_DISABLE;
 524     } else if (r300->query_current) {
 525         r300->ztop_state.z_buffer_top = R300_ZTOP_DISABLE;
 526     }
 527 }
 528
 529 void r300_update_derived_state(struct r300_context* r300)
 530 {
 531     if (r300->dirty_state &
 532         (R300_NEW_FRAGMENT_SHADER | R300_NEW_VERTEX_SHADER)) {
 533         r300_update_derived_shader_state(r300);
 534     }
 535
 536     if (r300->dirty_state &
 537             (R300_NEW_DSA | R300_NEW_FRAGMENT_SHADER | R300_NEW_QUERY)) {
 538         r300_update_ztop(r300);
 539     }
 540 }