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_vs.h"
  27
  28 /* r300_state_derived: Various bits of state which are dependent upon
  29  * currently bound CSO data. */
  30
  31 /* Set up the vs_tab and routes. */
  32 static void r300_vs_tab_routes(struct r300_context* r300,
  33                                struct r300_vertex_format* vformat)
  34 {
  35     struct r300_screen* r300screen = r300_screen(r300->context.screen);
  36     struct vertex_info* vinfo = &vformat->vinfo;
  37     int* tab = vformat->vs_tab;
  38     boolean pos = FALSE, psize = FALSE, fog = FALSE;
  39     int i, texs = 0, cols = 0;
  40     struct tgsi_shader_info* info;
  41
  42     if (r300screen->caps->has_tcl) {
  43         /* Use vertex shader to determine required routes. */
  44         info = &r300->vs->info;
  45     } else {
  46         /* Use fragment shader to determine required routes. */
  47         info = &r300->fs->info;
  48     }
  49
  50     assert(info->num_inputs <= 16);
  51
  52     if (!r300screen->caps->has_tcl || !r300->rs_state->enable_vte)
  53     {
  54         for (i = 0; i < info->num_inputs; i++) {
  55             switch (info->input_semantic_name[i]) {
  56                 case TGSI_SEMANTIC_POSITION:
  57                     pos = TRUE;
  58                     tab[i] = 0;
  59                     break;
  60                 case TGSI_SEMANTIC_COLOR:
  61                     tab[i] = 2 + cols;
  62                     cols++;
  63                     break;
  64                 case TGSI_SEMANTIC_PSIZE:
  65                     psize = TRUE;
  66                     tab[i] = 15;
  67                     break;
  68                 case TGSI_SEMANTIC_FOG:
  69                     fog = TRUE;
  70                     /* Fall through */
  71                 case TGSI_SEMANTIC_GENERIC:
  72                     tab[i] = 6 + texs;
  73                     texs++;
  74                     break;
  75                 default:
  76                     debug_printf("r300: Unknown vertex input %d\n",
  77                         info->input_semantic_name[i]);
  78                     break;
  79             }
  80         }
  81     }
  82     else
  83     {
  84         /* Just copy vert attribs over as-is. */
  85         for (i = 0; i < info->num_inputs; i++) {
  86             tab[i] = i;
  87         }
  88
  89         for (i = 0; i < info->num_outputs; i++) {
  90             switch (info->output_semantic_name[i]) {
  91                 case TGSI_SEMANTIC_POSITION:
  92                     pos = TRUE;
  93                     break;
  94                 case TGSI_SEMANTIC_COLOR:
  95                     cols++;
  96                     break;
  97                 case TGSI_SEMANTIC_PSIZE:
  98                     psize = TRUE;
  99                     break;
 100                 case TGSI_SEMANTIC_FOG:
 101                     fog = TRUE;
 102                     /* Fall through */
 103                 case TGSI_SEMANTIC_GENERIC:
 104                     texs++;
 105                     break;
 106                 default:
 107                     debug_printf("r300: Unknown vertex output %d\n",
 108                         info->output_semantic_name[i]);
 109                     break;
 110             }
 111         }
 112     }
 113
 114     /* XXX magic */
 115     assert(texs <= 8);
 116
 117     /* Do the actual vertex_info setup.
 118      *
 119      * vertex_info has four uints of hardware-specific data in it.
 120      * vinfo.hwfmt[0] is R300_VAP_VTX_STATE_CNTL
 121      * vinfo.hwfmt[1] is R300_VAP_VSM_VTX_ASSM
 122      * vinfo.hwfmt[2] is R300_VAP_OUTPUT_VTX_FMT_0
 123      * vinfo.hwfmt[3] is R300_VAP_OUTPUT_VTX_FMT_1 */
 124
 125     vinfo->hwfmt[0] = 0x5555; /* XXX this is classic Mesa bonghits */
 126
 127     if (!pos) {
 128         debug_printf("r300: Forcing vertex position attribute emit...\n");
 129         /* Make room for the position attribute
 130          * at the beginning of the tab. */
 131         for (i = 15; i > 0; i--) {
 132             tab[i] = tab[i-1];
 133         }
 134         tab[0] = 0;
 135     }
 136     draw_emit_vertex_attr(vinfo, EMIT_4F, INTERP_PERSPECTIVE,
 137         draw_find_vs_output(r300->draw, TGSI_SEMANTIC_POSITION, 0));
 138     vinfo->hwfmt[1] |= R300_INPUT_CNTL_POS;
 139     vinfo->hwfmt[2] |= R300_VAP_OUTPUT_VTX_FMT_0__POS_PRESENT;
 140
 141     if (psize) {
 142         draw_emit_vertex_attr(vinfo, EMIT_1F_PSIZE, INTERP_POS,
 143             draw_find_vs_output(r300->draw, TGSI_SEMANTIC_PSIZE, 0));
 144         vinfo->hwfmt[2] |= R300_VAP_OUTPUT_VTX_FMT_0__PT_SIZE_PRESENT;
 145     }
 146
 147     for (i = 0; i < cols; i++) {
 148         draw_emit_vertex_attr(vinfo, EMIT_4F, INTERP_LINEAR,
 149             draw_find_vs_output(r300->draw, TGSI_SEMANTIC_COLOR, i));
 150         vinfo->hwfmt[1] |= R300_INPUT_CNTL_COLOR;
 151         vinfo->hwfmt[2] |= (R300_VAP_OUTPUT_VTX_FMT_0__COLOR_0_PRESENT << i);
 152     }
 153
 154     /* Init i right here, increment it if fog is enabled.
 155      * This gets around a double-increment problem. */
 156     i = 0;
 157
 158     if (fog) {
 159         i++;
 160         draw_emit_vertex_attr(vinfo, EMIT_4F, INTERP_PERSPECTIVE,
 161             draw_find_vs_output(r300->draw, TGSI_SEMANTIC_FOG, 0));
 162         vinfo->hwfmt[1] |= (R300_INPUT_CNTL_TC0 << i);
 163         vinfo->hwfmt[3] |= (4 << (3 * i));
 164     }
 165
 166     for (i; i < texs; i++) {
 167         draw_emit_vertex_attr(vinfo, EMIT_4F, INTERP_PERSPECTIVE,
 168             draw_find_vs_output(r300->draw, TGSI_SEMANTIC_GENERIC, i));
 169         vinfo->hwfmt[1] |= (R300_INPUT_CNTL_TC0 << i);
 170         vinfo->hwfmt[3] |= (4 << (3 * i));
 171     }
 172
 173     /* Handle the case where the vertex shader will be generating some of
 174      * the attribs based on its inputs. */
 175     if (r300screen->caps->has_tcl &&
 176             info->num_inputs < info->num_outputs) {
 177         vinfo->num_attribs = info->num_inputs;
 178     }
 179
 180     draw_compute_vertex_size(vinfo);
 181 }
 182
 183 /* Update the PSC tables. */
 184 static void r300_vertex_psc(struct r300_context* r300,
 185                             struct r300_vertex_format* vformat)
 186 {
 187     struct r300_screen* r300screen = r300_screen(r300->context.screen);
 188     struct vertex_info* vinfo = &vformat->vinfo;
 189     int* tab = vformat->vs_tab;
 190     uint32_t temp;
 191     int i, attrib_count;
 192
 193     /* Vertex shaders have no semantics on their inputs,
 194      * so PSC should just route stuff based on their info,
 195      * and not on attrib information. */
 196     if (r300screen->caps->has_tcl) {
 197         attrib_count = r300->vs->info.num_inputs;
 198         debug_printf("r300: routing %d attribs in psc for vs\n",
 199                 attrib_count);
 200     } else {
 201         attrib_count = vinfo->num_attribs;
 202         debug_printf("r300: attrib count: %d\n", attrib_count);
 203         for (i = 0; i < attrib_count; i++) {
 204             debug_printf("r300: attrib: offset %d, interp %d, size %d,"
 205                    " tab %d\n", vinfo->attrib[i].src_index,
 206                    vinfo->attrib[i].interp_mode, vinfo->attrib[i].emit,
 207                    tab[i]);
 208         }
 209     }
 210
 211     for (i = 0; i < attrib_count; i++) {
 212         /* Make sure we have a proper destination for our attribute */
 213         assert(tab[i] != -1);
 214
 215         /* Add the attribute to the PSC table. */
 216         temp = r300screen->caps->has_tcl ?
 217             R300_DATA_TYPE_FLOAT_4 :
 218             translate_vertex_data_type(vinfo->attrib[i].emit);
 219         temp |= tab[i] << R300_DST_VEC_LOC_SHIFT;
 220
 221         if (i & 1) {
 222             vformat->vap_prog_stream_cntl[i >> 1] &= 0x0000ffff;
 223             vformat->vap_prog_stream_cntl[i >> 1] |= temp << 16;
 224
 225             vformat->vap_prog_stream_cntl_ext[i >> 1] |=
 226                 (R300_VAP_SWIZZLE_XYZW << 16);
 227         } else {
 228             vformat->vap_prog_stream_cntl[i >> 1] &= 0xffff0000;
 229             vformat->vap_prog_stream_cntl[i >> 1] |= temp <<  0;
 230
 231             vformat->vap_prog_stream_cntl_ext[i >> 1] |=
 232                 (R300_VAP_SWIZZLE_XYZW <<  0);
 233         }
 234     }
 235
 236     /* Set the last vector in the PSC. */
 237     i--;
 238     vformat->vap_prog_stream_cntl[i >> 1] |=
 239         (R300_LAST_VEC << (i & 1 ? 16 : 0));
 240 }
 241
 242 /* Update the vertex format. */
 243 static void r300_update_vertex_format(struct r300_context* r300)
 244 {
 245     struct r300_vertex_format vformat;
 246     int i;
 247
 248     memset(&vformat, 0, sizeof(struct r300_vertex_format));
 249     for (i = 0; i < 16; i++) {
 250         vformat.vs_tab[i] = -1;
 251         vformat.fs_tab[i] = -1;
 252     }
 253
 254     r300_vs_tab_routes(r300, &vformat);
 255
 256     r300_vertex_psc(r300, &vformat);
 257
 258     if (memcmp(&r300->vertex_info, &vformat,
 259                 sizeof(struct r300_vertex_format))) {
 260         memcpy(&r300->vertex_info, &vformat,
 261                 sizeof(struct r300_vertex_format));
 262         r300->dirty_state |= R300_NEW_VERTEX_FORMAT;
 263     }
 264 }
 265
 266 /* Set up the mappings from GB to US, for RS block. */
 267 static void r300_update_fs_tab(struct r300_context* r300)
 268 {
 269     struct r300_vertex_format* vformat = &r300->vertex_info;
 270     struct tgsi_shader_info* info = &r300->fs->info;
 271     int i, cols = 0, texs = 0, cols_emitted = 0;
 272     int* tab = vformat->fs_tab;
 273
 274     for (i = 0; i < 16; i++) {
 275         tab[i] = -1;
 276     }
 277
 278     assert(info->num_inputs <= 16);
 279     for (i = 0; i < info->num_inputs; i++) {
 280         switch (info->input_semantic_name[i]) {
 281             case TGSI_SEMANTIC_COLOR:
 282                 tab[i] = INTERP_LINEAR;
 283                 cols++;
 284                 break;
 285             case TGSI_SEMANTIC_POSITION:
 286             case TGSI_SEMANTIC_PSIZE:
 287                 debug_printf("r300: Implementation error: Can't use "
 288                         "pos attribs in fragshader yet!\n");
 289                 /* Pass through for now */
 290             case TGSI_SEMANTIC_FOG:
 291             case TGSI_SEMANTIC_GENERIC:
 292                 tab[i] = INTERP_PERSPECTIVE;
 293                 break;
 294             default:
 295                 debug_printf("r300: Unknown vertex input %d\n",
 296                     info->input_semantic_name[i]);
 297                 break;
 298         }
 299     }
 300
 301     /* Now that we know where everything is... */
 302     debug_printf("r300: fp input count: %d\n", info->num_inputs);
 303     for (i = 0; i < info->num_inputs; i++) {
 304         switch (tab[i]) {
 305             case INTERP_LINEAR:
 306                 debug_printf("r300: attrib: "
 307                         "stack offset %d, color,    tab %d\n",
 308                         i, cols_emitted);
 309                 tab[i] = cols_emitted;
 310                 cols_emitted++;
 311                 break;
 312             case INTERP_PERSPECTIVE:
 313                 debug_printf("r300: attrib: "
 314                         "stack offset %d, texcoord, tab %d\n",
 315                         i, cols + texs);
 316                 tab[i] = cols + texs;
 317                 texs++;
 318                 break;
 319             case -1:
 320                 debug_printf("r300: Implementation error: Bad fp interp!\n");
 321             default:
 322                 break;
 323         }
 324     }
 325
 326 }
 327
 328 /* Set up the RS block. This is the part of the chipset that actually does
 329  * the rasterization of vertices into fragments. This is also the part of the
 330  * chipset that locks up if any part of it is even slightly wrong. */
 331 static void r300_update_rs_block(struct r300_context* r300)
 332 {
 333     struct r300_rs_block* rs = r300->rs_block;
 334     struct tgsi_shader_info* info = &r300->fs->info;
 335     int* tab = r300->vertex_info.fs_tab;
 336     int col_count = 0, fp_offset = 0, i, memory_pos, tex_count = 0;
 337
 338     memset(rs, 0, sizeof(struct r300_rs_block));
 339
 340     if (r300_screen(r300->context.screen)->caps->is_r500) {
 341         for (i = 0; i < info->num_inputs; i++) {
 342             assert(tab[i] != -1);
 343             memory_pos = tab[i] * 4;
 344             switch (info->input_semantic_name[i]) {
 345                 case TGSI_SEMANTIC_COLOR:
 346                     rs->ip[col_count] |=
 347                         R500_RS_COL_PTR(memory_pos) |
 348                         R500_RS_COL_FMT(R300_RS_COL_FMT_RGBA);
 349                     col_count++;
 350                     break;
 351                 case TGSI_SEMANTIC_GENERIC:
 352                     rs->ip[tex_count] |=
 353                         R500_RS_SEL_S(memory_pos) |
 354                         R500_RS_SEL_T(memory_pos + 1) |
 355                         R500_RS_SEL_R(memory_pos + 2) |
 356                         R500_RS_SEL_Q(memory_pos + 3);
 357                     tex_count++;
 358                     break;
 359                 default:
 360                     break;
 361             }
 362         }
 363
 364         if (col_count == 0) {
 365             rs->ip[0] |= R500_RS_COL_FMT(R300_RS_COL_FMT_0001);
 366         }
 367
 368         if (tex_count == 0) {
 369             rs->ip[0] |=
 370                 R500_RS_SEL_S(R500_RS_IP_PTR_K0) |
 371                 R500_RS_SEL_T(R500_RS_IP_PTR_K0) |
 372                 R500_RS_SEL_R(R500_RS_IP_PTR_K0) |
 373                 R500_RS_SEL_Q(R500_RS_IP_PTR_K1);
 374         }
 375
 376         /* Rasterize at least one color, or bad things happen. */
 377         if ((col_count == 0) && (tex_count == 0)) {
 378             col_count++;
 379         }
 380
 381         for (i = 0; i < tex_count; i++) {
 382             rs->inst[i] |= R500_RS_INST_TEX_ID(i) |
 383                 R500_RS_INST_TEX_CN_WRITE | R500_RS_INST_TEX_ADDR(fp_offset);
 384             fp_offset++;
 385         }
 386
 387         for (i = 0; i < col_count; i++) {
 388             rs->inst[i] |= R500_RS_INST_COL_ID(i) |
 389                 R500_RS_INST_COL_CN_WRITE | R500_RS_INST_COL_ADDR(fp_offset);
 390             fp_offset++;
 391         }
 392     } else {
 393         for (i = 0; i < info->num_inputs; i++) {
 394             assert(tab[i] != -1);
 395             memory_pos = tab[i] * 4;
 396             switch (info->input_semantic_name[i]) {
 397                 case TGSI_SEMANTIC_COLOR:
 398                     rs->ip[col_count] |=
 399                         R300_RS_COL_PTR(memory_pos) |
 400                         R300_RS_COL_FMT(R300_RS_COL_FMT_RGBA);
 401                     col_count++;
 402                     break;
 403                 case TGSI_SEMANTIC_GENERIC:
 404                     rs->ip[tex_count] |=
 405                         R300_RS_TEX_PTR(memory_pos) |
 406                         R300_RS_SEL_S(R300_RS_SEL_C0) |
 407                         R300_RS_SEL_T(R300_RS_SEL_C1) |
 408                         R300_RS_SEL_R(R300_RS_SEL_C2) |
 409                         R300_RS_SEL_Q(R300_RS_SEL_C3);
 410                     tex_count++;
 411                     break;
 412                 default:
 413                     break;
 414             }
 415         }
 416
 417         if (col_count == 0) {
 418             rs->ip[0] |= R300_RS_COL_FMT(R300_RS_COL_FMT_0001);
 419         }
 420
 421         if (tex_count == 0) {
 422             rs->ip[0] |=
 423                 R300_RS_SEL_S(R300_RS_SEL_K0) |
 424                 R300_RS_SEL_T(R300_RS_SEL_K0) |
 425                 R300_RS_SEL_R(R300_RS_SEL_K0) |
 426                 R300_RS_SEL_Q(R300_RS_SEL_K1);
 427         }
 428
 429         /* Rasterize at least one color, or bad things happen. */
 430         if ((col_count == 0) && (tex_count == 0)) {
 431             col_count++;
 432         }
 433
 434         for (i = 0; i < tex_count; i++) {
 435             rs->inst[i] |= R300_RS_INST_TEX_ID(i) |
 436                 R300_RS_INST_TEX_CN_WRITE | R300_RS_INST_TEX_ADDR(fp_offset);
 437             fp_offset++;
 438         }
 439
 440         for (i = 0; i < col_count; i++) {
 441             rs->inst[i] |= R300_RS_INST_COL_ID(i) |
 442                 R300_RS_INST_COL_CN_WRITE | R300_RS_INST_COL_ADDR(fp_offset);
 443             fp_offset++;
 444         }
 445     }
 446
 447     rs->count = (tex_count * 4) | (col_count << R300_IC_COUNT_SHIFT) |
 448         R300_HIRES_EN;
 449
 450     rs->inst_count = MAX2(MAX2(col_count - 1, tex_count - 1), 0);
 451 }
 452
 453 void r300_update_derived_state(struct r300_context* r300)
 454 {
 455     if (r300->dirty_state &
 456             (R300_NEW_FRAGMENT_SHADER | R300_NEW_VERTEX_SHADER)) {
 457         r300_update_vertex_format(r300);
 458     }
 459
 460     if (r300->dirty_state & R300_NEW_VERTEX_FORMAT) {
 461         r300_update_fs_tab(r300);
 462         r300_update_rs_block(r300);
 463     }
 464 }