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 "draw/draw_context.h"
  24
  25 #include "util/u_math.h"
  26 #include "util/u_memory.h"
  27
  28 #include "r300_context.h"
  29 #include "r300_fs.h"
  30 #include "r300_screen.h"
  31 #include "r300_state_derived.h"
  32 #include "r300_state_inlines.h"
  33 #include "r300_vs.h"
  34
  35 /* r300_state_derived: Various bits of state which are dependent upon
  36  * currently bound CSO data. */
  37
  38 struct r300_shader_key {
  39     struct r300_vertex_shader* vs;
  40     struct r300_fragment_shader* fs;
  41 };
  42
  43 struct r300_shader_derived_value {
  44     struct r300_vertex_format* vformat;
  45     struct r300_rs_block* rs_block;
  46 };
  47
  48 unsigned r300_shader_key_hash(void* key) {
  49     struct r300_shader_key* shader_key = (struct r300_shader_key*)key;
  50     unsigned vs = (unsigned)shader_key->vs;
  51     unsigned fs = (unsigned)shader_key->fs;
  52
  53     return (vs << 16) | (fs & 0xffff);
  54 }
  55
  56 int r300_shader_key_compare(void* key1, void* key2) {
  57     struct r300_shader_key* shader_key1 = (struct r300_shader_key*)key1;
  58     struct r300_shader_key* shader_key2 = (struct r300_shader_key*)key2;
  59
  60     return (shader_key1->vs == shader_key2->vs) &&
  61         (shader_key1->fs == shader_key2->fs);
  62 }
  63
  64 /* Set up the vs_tab and routes. */
  65 static void r300_vs_tab_routes(struct r300_context* r300,
  66                                struct r300_vertex_info* vformat)
  67 {
  68     struct r300_screen* r300screen = r300_screen(r300->context.screen);
  69     struct vertex_info* vinfo = &vformat->vinfo;
  70     int* tab = vformat->vs_tab;
  71     boolean pos = FALSE, psize = FALSE, fog = FALSE;
  72     int i, texs = 0, cols = 0;
  73     struct tgsi_shader_info* info;
  74
  75     if (r300screen->caps->has_tcl) {
  76         /* Use vertex shader to determine required routes. */
  77         info = &r300->vs->info;
  78     } else {
  79         /* Use fragment shader to determine required routes. */
  80         info = &r300->fs->info;
  81     }
  82
  83     assert(info->num_inputs <= 16);
  84
  85     if (!r300screen->caps->has_tcl || !r300->rs_state->enable_vte)
  86     {
  87         for (i = 0; i < info->num_inputs; i++) {
  88             switch (r300->vs->code.inputs[i]) {
  89                 case TGSI_SEMANTIC_POSITION:
  90                     pos = TRUE;
  91                     tab[i] = 0;
  92                     break;
  93                 case TGSI_SEMANTIC_COLOR:
  94                     tab[i] = 2 + cols;
  95                     cols++;
  96                     break;
  97                 case TGSI_SEMANTIC_PSIZE:
  98                     assert(psize == FALSE);
  99                     psize = TRUE;
 100                     tab[i] = 15;
 101                     break;
 102                 case TGSI_SEMANTIC_FOG:
 103                     assert(fog == FALSE);
 104                     fog = TRUE;
 105                     /* Fall through */
 106                 case TGSI_SEMANTIC_GENERIC:
 107                     tab[i] = 6 + texs;
 108                     texs++;
 109                     break;
 110                 default:
 111                     debug_printf("r300: Unknown vertex input %d\n",
 112                         info->input_semantic_name[i]);
 113                     break;
 114             }
 115         }
 116     }
 117     else
 118     {
 119         /* Just copy vert attribs over as-is. */
 120         for (i = 0; i < info->num_inputs; i++) {
 121             tab[i] = i;
 122         }
 123
 124         for (i = 0; i < info->num_outputs; i++) {
 125             switch (info->output_semantic_name[i]) {
 126                 case TGSI_SEMANTIC_POSITION:
 127                     pos = TRUE;
 128                     break;
 129                 case TGSI_SEMANTIC_COLOR:
 130                     cols++;
 131                     break;
 132                 case TGSI_SEMANTIC_PSIZE:
 133                     psize = TRUE;
 134                     break;
 135                 case TGSI_SEMANTIC_FOG:
 136                     fog = TRUE;
 137                     /* Fall through */
 138                 case TGSI_SEMANTIC_GENERIC:
 139                     texs++;
 140                     break;
 141                 default:
 142                     debug_printf("r300: Unknown vertex output %d\n",
 143                         info->output_semantic_name[i]);
 144                     break;
 145             }
 146         }
 147     }
 148
 149     /* XXX magic */
 150     assert(texs <= 8);
 151
 152     /* Do the actual vertex_info setup.
 153      *
 154      * vertex_info has four uints of hardware-specific data in it.
 155      * vinfo.hwfmt[0] is R300_VAP_VTX_STATE_CNTL
 156      * vinfo.hwfmt[1] is R300_VAP_VSM_VTX_ASSM
 157      * vinfo.hwfmt[2] is R300_VAP_OUTPUT_VTX_FMT_0
 158      * vinfo.hwfmt[3] is R300_VAP_OUTPUT_VTX_FMT_1 */
 159
 160     vinfo->hwfmt[0] = 0x5555; /* XXX this is classic Mesa bonghits */
 161
 162     /* We need to add vertex position attribute only for SW TCL case,
 163      * for HW TCL case it could be generated by vertex shader */
 164     if (!pos && !r300screen->caps->has_tcl) {
 165         debug_printf("r300: Forcing vertex position attribute emit...\n");
 166         /* Make room for the position attribute
 167          * at the beginning of the tab. */
 168         for (i = 15; i > 0; i--) {
 169             tab[i] = tab[i-1];
 170         }
 171         tab[0] = 0;
 172     }
 173
 174     /* Position. */
 175     if (r300->draw) {
 176         draw_emit_vertex_attr(vinfo, EMIT_4F, INTERP_PERSPECTIVE,
 177             draw_find_vs_output(r300->draw, TGSI_SEMANTIC_POSITION, 0));
 178     }
 179     vinfo->hwfmt[1] |= R300_INPUT_CNTL_POS;
 180     vinfo->hwfmt[2] |= R300_VAP_OUTPUT_VTX_FMT_0__POS_PRESENT;
 181
 182     /* Point size. */
 183     if (psize) {
 184         if (r300->draw) {
 185             draw_emit_vertex_attr(vinfo, EMIT_1F_PSIZE, INTERP_POS,
 186                 draw_find_vs_output(r300->draw, TGSI_SEMANTIC_PSIZE, 0));
 187         }
 188         vinfo->hwfmt[2] |= R300_VAP_OUTPUT_VTX_FMT_0__PT_SIZE_PRESENT;
 189     }
 190
 191     /* Colors. */
 192     for (i = 0; i < cols; i++) {
 193         if (r300->draw) {
 194             draw_emit_vertex_attr(vinfo, EMIT_4F, INTERP_LINEAR,
 195                 draw_find_vs_output(r300->draw, TGSI_SEMANTIC_COLOR, i));
 196         }
 197         vinfo->hwfmt[1] |= R300_INPUT_CNTL_COLOR;
 198         vinfo->hwfmt[2] |= (R300_VAP_OUTPUT_VTX_FMT_0__COLOR_0_PRESENT << i);
 199     }
 200
 201     /* Init i right here, increment it if fog is enabled.
 202      * This gets around a double-increment problem. */
 203     i = 0;
 204
 205     /* Fog. This is a special-cased texcoord. */
 206     if (fog) {
 207         i++;
 208         if (r300->draw) {
 209             draw_emit_vertex_attr(vinfo, EMIT_4F, INTERP_PERSPECTIVE,
 210                 draw_find_vs_output(r300->draw, TGSI_SEMANTIC_FOG, 0));
 211         }
 212         vinfo->hwfmt[1] |= (R300_INPUT_CNTL_TC0 << i);
 213         vinfo->hwfmt[3] |= (4 << (3 * i));
 214     }
 215
 216     /* Texcoords. */
 217     for (; i < texs; i++) {
 218         if (r300->draw) {
 219             draw_emit_vertex_attr(vinfo, EMIT_4F, INTERP_PERSPECTIVE,
 220                 draw_find_vs_output(r300->draw, TGSI_SEMANTIC_GENERIC, i));
 221         }
 222         vinfo->hwfmt[1] |= (R300_INPUT_CNTL_TC0 << i);
 223         vinfo->hwfmt[3] |= (4 << (3 * i));
 224     }
 225
 226     draw_compute_vertex_size(vinfo);
 227 }
 228
 229 /* Update the PSC tables. */
 230 static void r300_vertex_psc(struct r300_context* r300,
 231                             struct r300_vertex_info* vformat)
 232 {
 233     struct r300_screen* r300screen = r300_screen(r300->context.screen);
 234     struct vertex_info* vinfo = &vformat->vinfo;
 235     int* tab = vformat->vs_tab;
 236     uint16_t type, swizzle;
 237     enum pipe_format format;
 238     unsigned i, attrib_count;
 239
 240     /* Vertex shaders have no semantics on their inputs,
 241      * so PSC should just route stuff based on their info,
 242      * and not on attrib information. */
 243     if (r300screen->caps->has_tcl) {
 244         attrib_count = r300->vs->info.num_inputs;
 245         DBG(r300, DBG_DRAW, "r300: routing %d attribs in psc for vs\n",
 246                 attrib_count);
 247     } else {
 248         attrib_count = vinfo->num_attribs;
 249         DBG(r300, DBG_DRAW, "r300: attrib count: %d\n", attrib_count);
 250         for (i = 0; i < attrib_count; i++) {
 251             DBG(r300, DBG_DRAW, "r300: attrib: offset %d, interp %d, size %d,"
 252                    " tab %d\n", vinfo->attrib[i].src_index,
 253                    vinfo->attrib[i].interp_mode, vinfo->attrib[i].emit,
 254                    tab[i]);
 255         }
 256     }
 257
 258     for (i = 0; i < attrib_count; i++) {
 259         /* Make sure we have a proper destination for our attribute. */
 260         assert(tab[i] != -1);
 261
 262         format = draw_translate_vinfo_format(vinfo->attrib[i].emit);
 263
 264         /* Obtain the type of data in this attribute. */
 265         type = r300_translate_vertex_data_type(format) |
 266             tab[i] << R300_DST_VEC_LOC_SHIFT;
 267
 268         /* Obtain the swizzle for this attribute. Note that the default
 269          * swizzle in the hardware is not XYZW! */
 270         swizzle = r300_translate_vertex_data_swizzle(format);
 271
 272         /* Add the attribute to the PSC table. */
 273         if (i & 1) {
 274             vformat->vap_prog_stream_cntl[i >> 1] |= type << 16;
 275
 276             vformat->vap_prog_stream_cntl_ext[i >> 1] |= swizzle << 16;
 277         } else {
 278             vformat->vap_prog_stream_cntl[i >> 1] |= type <<  0;
 279
 280             vformat->vap_prog_stream_cntl_ext[i >> 1] |= swizzle << 0;
 281         }
 282     }
 283
 284     /* Set the last vector in the PSC. */
 285     if (i) {
 286         i -= 1;
 287     }
 288     vformat->vap_prog_stream_cntl[i >> 1] |=
 289         (R300_LAST_VEC << (i & 1 ? 16 : 0));
 290 }
 291
 292 /* Set up the mappings from GB to US, for RS block. */
 293 static void r300_update_fs_tab(struct r300_context* r300,
 294                                struct r300_vertex_info* vformat)
 295 {
 296     struct tgsi_shader_info* info = &r300->fs->info;
 297     int i, cols = 0, texs = 0, cols_emitted = 0;
 298     int* tab = vformat->fs_tab;
 299
 300     for (i = 0; i < 16; i++) {
 301         tab[i] = -1;
 302     }
 303
 304     assert(info->num_inputs <= 16);
 305     for (i = 0; i < info->num_inputs; i++) {
 306         switch (info->input_semantic_name[i]) {
 307             case TGSI_SEMANTIC_COLOR:
 308                 tab[i] = INTERP_LINEAR;
 309                 cols++;
 310                 break;
 311             case TGSI_SEMANTIC_POSITION:
 312             case TGSI_SEMANTIC_PSIZE:
 313                 debug_printf("r300: Implementation error: Can't use "
 314                         "pos attribs in fragshader yet!\n");
 315                 /* Pass through for now */
 316             case TGSI_SEMANTIC_FOG:
 317             case TGSI_SEMANTIC_GENERIC:
 318                 tab[i] = INTERP_PERSPECTIVE;
 319                 break;
 320             default:
 321                 debug_printf("r300: Unknown vertex input %d\n",
 322                     info->input_semantic_name[i]);
 323                 break;
 324         }
 325     }
 326
 327     /* Now that we know where everything is... */
 328     DBG(r300, DBG_DRAW, "r300: fp input count: %d\n", info->num_inputs);
 329     for (i = 0; i < info->num_inputs; i++) {
 330         switch (tab[i]) {
 331             case INTERP_LINEAR:
 332                 DBG(r300, DBG_DRAW, "r300: attrib: "
 333                         "stack offset %d, color,    tab %d\n",
 334                         i, cols_emitted);
 335                 tab[i] = cols_emitted;
 336                 cols_emitted++;
 337                 break;
 338             case INTERP_PERSPECTIVE:
 339                 DBG(r300, DBG_DRAW, "r300: attrib: "
 340                         "stack offset %d, texcoord, tab %d\n",
 341                         i, cols + texs);
 342                 tab[i] = cols + texs;
 343                 texs++;
 344                 break;
 345             case -1:
 346                 debug_printf("r300: Implementation error: Bad fp interp!\n");
 347             default:
 348                 break;
 349         }
 350     }
 351
 352 }
 353
 354 /* Set up the RS block. This is the part of the chipset that actually does
 355  * the rasterization of vertices into fragments. This is also the part of the
 356  * chipset that locks up if any part of it is even slightly wrong. */
 357 static void r300_update_rs_block(struct r300_context* r300,
 358                                  struct r300_rs_block* rs)
 359 {
 360     struct tgsi_shader_info* info = &r300->fs->info;
 361     int col_count = 0, fp_offset = 0, i, tex_count = 0;
 362     int rs_tex_comp = 0;
 363
 364     if (r300_screen(r300->context.screen)->caps->is_r500) {
 365         for (i = 0; i < info->num_inputs; i++) {
 366             switch (info->input_semantic_name[i]) {
 367                 case TGSI_SEMANTIC_COLOR:
 368                     rs->ip[col_count] |=
 369                         R500_RS_COL_PTR(col_count) |
 370                         R500_RS_COL_FMT(R300_RS_COL_FMT_RGBA);
 371                     col_count++;
 372                     break;
 373                 case TGSI_SEMANTIC_GENERIC:
 374                     rs->ip[tex_count] |=
 375                         R500_RS_SEL_S(rs_tex_comp) |
 376                         R500_RS_SEL_T(rs_tex_comp + 1) |
 377                         R500_RS_SEL_R(rs_tex_comp + 2) |
 378                         R500_RS_SEL_Q(rs_tex_comp + 3);
 379                     tex_count++;
 380                     rs_tex_comp += 4;
 381                     break;
 382                 default:
 383                     break;
 384             }
 385         }
 386
 387         /* Rasterize at least one color, or bad things happen. */
 388         if ((col_count == 0) && (tex_count == 0)) {
 389             rs->ip[0] |= R500_RS_COL_FMT(R300_RS_COL_FMT_0001);
 390             col_count++;
 391         }
 392
 393         for (i = 0; i < tex_count; i++) {
 394             rs->inst[i] |= R500_RS_INST_TEX_ID(i) |
 395                 R500_RS_INST_TEX_CN_WRITE | R500_RS_INST_TEX_ADDR(fp_offset);
 396             fp_offset++;
 397         }
 398
 399         for (i = 0; i < col_count; i++) {
 400             rs->inst[i] |= R500_RS_INST_COL_ID(i) |
 401                 R500_RS_INST_COL_CN_WRITE | R500_RS_INST_COL_ADDR(fp_offset);
 402             fp_offset++;
 403         }
 404     } else {
 405         for (i = 0; i < info->num_inputs; i++) {
 406             switch (info->input_semantic_name[i]) {
 407                 case TGSI_SEMANTIC_COLOR:
 408                     rs->ip[col_count] |=
 409                         R300_RS_COL_PTR(col_count) |
 410                         R300_RS_COL_FMT(R300_RS_COL_FMT_RGBA);
 411                     col_count++;
 412                     break;
 413                 case TGSI_SEMANTIC_GENERIC:
 414                     rs->ip[tex_count] |=
 415                         R300_RS_TEX_PTR(rs_tex_comp) |
 416                         R300_RS_SEL_S(R300_RS_SEL_C0) |
 417                         R300_RS_SEL_T(R300_RS_SEL_C1) |
 418                         R300_RS_SEL_R(R300_RS_SEL_C2) |
 419                         R300_RS_SEL_Q(R300_RS_SEL_C3);
 420                     tex_count++;
 421                     rs_tex_comp+=4;
 422                     break;
 423                 default:
 424                     break;
 425             }
 426         }
 427
 428         if (col_count == 0) {
 429             rs->ip[0] |= R300_RS_COL_FMT(R300_RS_COL_FMT_0001);
 430         }
 431
 432         if (tex_count == 0) {
 433             rs->ip[0] |=
 434                 R300_RS_SEL_S(R300_RS_SEL_K0) |
 435                 R300_RS_SEL_T(R300_RS_SEL_K0) |
 436                 R300_RS_SEL_R(R300_RS_SEL_K0) |
 437                 R300_RS_SEL_Q(R300_RS_SEL_K1);
 438         }
 439
 440         /* Rasterize at least one color, or bad things happen. */
 441         if ((col_count == 0) && (tex_count == 0)) {
 442             col_count++;
 443         }
 444
 445         for (i = 0; i < tex_count; i++) {
 446             rs->inst[i] |= R300_RS_INST_TEX_ID(i) |
 447                 R300_RS_INST_TEX_CN_WRITE | R300_RS_INST_TEX_ADDR(fp_offset);
 448             fp_offset++;
 449         }
 450
 451         for (i = 0; i < col_count; i++) {
 452             rs->inst[i] |= R300_RS_INST_COL_ID(i) |
 453                 R300_RS_INST_COL_CN_WRITE | R300_RS_INST_COL_ADDR(fp_offset);
 454             fp_offset++;
 455         }
 456     }
 457
 458     rs->count = (rs_tex_comp) | (col_count << R300_IC_COUNT_SHIFT) |
 459         R300_HIRES_EN;
 460
 461     rs->inst_count = MAX2(MAX2(col_count - 1, tex_count - 1), 0);
 462 }
 463
 464 /* Update the vertex format. */
 465 static void r300_update_derived_shader_state(struct r300_context* r300)
 466 {
 467     struct r300_screen* r300screen = r300_screen(r300->context.screen);
 468     struct r300_vertex_info* vformat;
 469     struct r300_rs_block* rs_block;
 470     int i;
 471
 472     /*
 473     struct r300_shader_key* key;
 474     struct r300_shader_derived_value* value;
 475     key = CALLOC_STRUCT(r300_shader_key);
 476     key->vs = r300->vs;
 477     key->fs = r300->fs;
 478
 479     value = (struct r300_shader_derived_value*)
 480         util_hash_table_get(r300->shader_hash_table, (void*)key);
 481     if (value) {
 482         //vformat = value->vformat;
 483         rs_block = value->rs_block;
 484
 485         FREE(key);
 486     } else {
 487         rs_block = CALLOC_STRUCT(r300_rs_block);
 488         value = CALLOC_STRUCT(r300_shader_derived_value);
 489
 490         r300_update_rs_block(r300, rs_block);
 491
 492         //value->vformat = vformat;
 493         value->rs_block = rs_block;
 494         util_hash_table_set(r300->shader_hash_table,
 495             (void*)key, (void*)value);
 496     } */
 497
 498     /* XXX This will be refactored ASAP. */
 499     vformat = CALLOC_STRUCT(r300_vertex_info);
 500     rs_block = CALLOC_STRUCT(r300_rs_block);
 501
 502     for (i = 0; i < 16; i++) {
 503         vformat->vs_tab[i] = -1;
 504         vformat->fs_tab[i] = -1;
 505     }
 506
 507     r300_vs_tab_routes(r300, vformat);
 508     r300_vertex_psc(r300, vformat);
 509     r300_update_fs_tab(r300, vformat);
 510
 511     r300_update_rs_block(r300, rs_block);
 512
 513     FREE(r300->vertex_info);
 514     FREE(r300->rs_block);
 515
 516     r300->vertex_info = vformat;
 517     r300->rs_block = rs_block;
 518     r300->dirty_state |= (R300_NEW_VERTEX_FORMAT | R300_NEW_RS_BLOCK);
 519 }
 520
 521 static void r300_update_ztop(struct r300_context* r300)
 522 {
 523     r300->ztop_state.z_buffer_top = R300_ZTOP_ENABLE;
 524
 525     /* This is important enough that I felt it warranted a comment.
 526      *
 527      * According to the docs, these are the conditions where ZTOP must be
 528      * disabled:
 529      * 1) Alpha testing enabled
 530      * 2) Texture kill instructions in fragment shader
 531      * 3) Chroma key culling enabled
 532      * 4) W-buffering enabled
 533      *
 534      * The docs claim that for the first three cases, if no ZS writes happen,
 535      * then ZTOP can be used.
 536      *
 537      * Additionally, the following conditions require disabled ZTOP:
 538      * ~) Depth writes in fragment shader
 539      * ~) Outstanding occlusion queries
 540      *
 541      * ~C.
 542      */
 543     if (r300->dsa_state->alpha_function) {
 544         r300->ztop_state.z_buffer_top = R300_ZTOP_DISABLE;
 545     } else if (r300->fs->info.uses_kill) {
 546         r300->ztop_state.z_buffer_top = R300_ZTOP_DISABLE;
 547     } else if (r300_fragment_shader_writes_depth(r300->fs)) {
 548         r300->ztop_state.z_buffer_top = R300_ZTOP_DISABLE;
 549     } else if (r300->query_current) {
 550         r300->ztop_state.z_buffer_top = R300_ZTOP_DISABLE;
 551     }
 552 }
 553
 554 void r300_update_derived_state(struct r300_context* r300)
 555 {
 556     /* XXX */
 557     if (TRUE || r300->dirty_state &
 558         (R300_NEW_FRAGMENT_SHADER | R300_NEW_VERTEX_SHADER)) {
 559         r300_update_derived_shader_state(r300);
 560     }
 561
 562     if (r300->dirty_state &
 563             (R300_NEW_DSA | R300_NEW_FRAGMENT_SHADER | R300_NEW_QUERY)) {
 564         r300_update_ztop(r300);
 565     }
 566 }