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 /* Set up the vs_tab and routes. */
  33 static void r300_vs_tab_routes(struct r300_context* r300,
  34                                struct r300_vertex_format* vformat)
  35 {
  36     struct r300_screen* r300screen = r300_screen(r300->context.screen);
  37     struct vertex_info* vinfo = &vformat->vinfo;
  38     int* tab = vformat->vs_tab;
  39     boolean pos = FALSE, psize = FALSE, fog = FALSE;
  40     int i, texs = 0, cols = 0;
  41     struct tgsi_shader_info* info;
  42
  43     if (r300screen->caps->has_tcl) {
  44         /* Use vertex shader to determine required routes. */
  45         info = &r300->vs->info;
  46     } else {
  47         /* Use fragment shader to determine required routes. */
  48         info = &r300->fs->info;
  49     }
  50
  51     assert(info->num_inputs <= 16);
  52
  53     if (!r300screen->caps->has_tcl || !r300->rs_state->enable_vte)
  54     {
  55         for (i = 0; i < info->num_inputs; i++) {
  56             switch (info->input_semantic_name[i]) {
  57                 case TGSI_SEMANTIC_POSITION:
  58                     pos = TRUE;
  59                     tab[i] = 0;
  60                     break;
  61                 case TGSI_SEMANTIC_COLOR:
  62                     tab[i] = 2 + cols;
  63                     cols++;
  64                     break;
  65                 case TGSI_SEMANTIC_PSIZE:
  66                     psize = TRUE;
  67                     tab[i] = 15;
  68                     break;
  69                 case TGSI_SEMANTIC_FOG:
  70                     fog = TRUE;
  71                     /* Fall through */
  72                 case TGSI_SEMANTIC_GENERIC:
  73                     tab[i] = 6 + texs;
  74                     texs++;
  75                     break;
  76                 default:
  77                     debug_printf("r300: Unknown vertex input %d\n",
  78                         info->input_semantic_name[i]);
  79                     break;
  80             }
  81         }
  82     }
  83     else
  84     {
  85         /* Just copy vert attribs over as-is. */
  86         for (i = 0; i < info->num_inputs; i++) {
  87             tab[i] = i;
  88         }
  89
  90         for (i = 0; i < info->num_outputs; i++) {
  91             switch (info->output_semantic_name[i]) {
  92                 case TGSI_SEMANTIC_POSITION:
  93                     pos = TRUE;
  94                     break;
  95                 case TGSI_SEMANTIC_COLOR:
  96                     cols++;
  97                     break;
  98                 case TGSI_SEMANTIC_PSIZE:
  99                     psize = TRUE;
 100                     break;
 101                 case TGSI_SEMANTIC_FOG:
 102                     fog = TRUE;
 103                     /* Fall through */
 104                 case TGSI_SEMANTIC_GENERIC:
 105                     texs++;
 106                     break;
 107                 default:
 108                     debug_printf("r300: Unknown vertex output %d\n",
 109                         info->output_semantic_name[i]);
 110                     break;
 111             }
 112         }
 113     }
 114
 115     /* XXX magic */
 116     assert(texs <= 8);
 117
 118     /* Do the actual vertex_info setup.
 119      *
 120      * vertex_info has four uints of hardware-specific data in it.
 121      * vinfo.hwfmt[0] is R300_VAP_VTX_STATE_CNTL
 122      * vinfo.hwfmt[1] is R300_VAP_VSM_VTX_ASSM
 123      * vinfo.hwfmt[2] is R300_VAP_OUTPUT_VTX_FMT_0
 124      * vinfo.hwfmt[3] is R300_VAP_OUTPUT_VTX_FMT_1 */
 125
 126     vinfo->hwfmt[0] = 0x5555; /* XXX this is classic Mesa bonghits */
 127
 128     if (!pos) {
 129         debug_printf("r300: Forcing vertex position attribute emit...\n");
 130         /* Make room for the position attribute
 131          * at the beginning of the tab. */
 132         for (i = 15; i > 0; i--) {
 133             tab[i] = tab[i-1];
 134         }
 135         tab[0] = 0;
 136     }
 137     draw_emit_vertex_attr(vinfo, EMIT_4F, INTERP_PERSPECTIVE,
 138         draw_find_vs_output(r300->draw, TGSI_SEMANTIC_POSITION, 0));
 139     vinfo->hwfmt[1] |= R300_INPUT_CNTL_POS;
 140     vinfo->hwfmt[2] |= R300_VAP_OUTPUT_VTX_FMT_0__POS_PRESENT;
 141
 142     if (psize) {
 143         draw_emit_vertex_attr(vinfo, EMIT_1F_PSIZE, INTERP_POS,
 144             draw_find_vs_output(r300->draw, TGSI_SEMANTIC_PSIZE, 0));
 145         vinfo->hwfmt[2] |= R300_VAP_OUTPUT_VTX_FMT_0__PT_SIZE_PRESENT;
 146     }
 147
 148     for (i = 0; i < cols; i++) {
 149         draw_emit_vertex_attr(vinfo, EMIT_4F, INTERP_LINEAR,
 150             draw_find_vs_output(r300->draw, TGSI_SEMANTIC_COLOR, i));
 151         vinfo->hwfmt[1] |= R300_INPUT_CNTL_COLOR;
 152         vinfo->hwfmt[2] |= (R300_VAP_OUTPUT_VTX_FMT_0__COLOR_0_PRESENT << i);
 153     }
 154
 155     /* Init i right here, increment it if fog is enabled.
 156      * This gets around a double-increment problem. */
 157     i = 0;
 158
 159     if (fog) {
 160         i++;
 161         draw_emit_vertex_attr(vinfo, EMIT_4F, INTERP_PERSPECTIVE,
 162             draw_find_vs_output(r300->draw, TGSI_SEMANTIC_FOG, 0));
 163         vinfo->hwfmt[1] |= (R300_INPUT_CNTL_TC0 << i);
 164         vinfo->hwfmt[3] |= (4 << (3 * i));
 165     }
 166
 167     for (i; i < texs; i++) {
 168         draw_emit_vertex_attr(vinfo, EMIT_4F, INTERP_PERSPECTIVE,
 169             draw_find_vs_output(r300->draw, TGSI_SEMANTIC_GENERIC, i));
 170         vinfo->hwfmt[1] |= (R300_INPUT_CNTL_TC0 << i);
 171         vinfo->hwfmt[3] |= (4 << (3 * i));
 172     }
 173
 174     /* Handle the case where the vertex shader will be generating some of
 175      * the attribs based on its inputs. */
 176     if (r300screen->caps->has_tcl &&
 177             info->num_inputs < info->num_outputs) {
 178         vinfo->num_attribs = info->num_inputs;
 179     }
 180
 181     draw_compute_vertex_size(vinfo);
 182 }
 183
 184 /* Update the PSC tables. */
 185 static void r300_vertex_psc(struct r300_context* r300,
 186                             struct r300_vertex_format* vformat)
 187 {
 188     struct r300_screen* r300screen = r300_screen(r300->context.screen);
 189     struct vertex_info* vinfo = &vformat->vinfo;
 190     int* tab = vformat->vs_tab;
 191     uint32_t temp;
 192     int i, attrib_count;
 193
 194     /* Vertex shaders have no semantics on their inputs,
 195      * so PSC should just route stuff based on their info,
 196      * and not on attrib information. */
 197     if (r300screen->caps->has_tcl) {
 198         attrib_count = r300->vs->info.num_inputs;
 199         DBG(r300, DBG_DRAW, "r300: routing %d attribs in psc for vs\n",
 200                 attrib_count);
 201     } else {
 202         attrib_count = vinfo->num_attribs;
 203         DBG(r300, DBG_DRAW, "r300: attrib count: %d\n", attrib_count);
 204         for (i = 0; i < attrib_count; i++) {
 205             DBG(r300, DBG_DRAW, "r300: attrib: offset %d, interp %d, size %d,"
 206                    " tab %d\n", vinfo->attrib[i].src_index,
 207                    vinfo->attrib[i].interp_mode, vinfo->attrib[i].emit,
 208                    tab[i]);
 209         }
 210     }
 211
 212     for (i = 0; i < attrib_count; i++) {
 213         /* Make sure we have a proper destination for our attribute */
 214         assert(tab[i] != -1);
 215
 216         /* Add the attribute to the PSC table. */
 217         temp = r300screen->caps->has_tcl ?
 218             R300_DATA_TYPE_FLOAT_4 :
 219             translate_vertex_data_type(vinfo->attrib[i].emit);
 220         temp |= tab[i] << R300_DST_VEC_LOC_SHIFT;
 221
 222         if (i & 1) {
 223             vformat->vap_prog_stream_cntl[i >> 1] &= 0x0000ffff;
 224             vformat->vap_prog_stream_cntl[i >> 1] |= temp << 16;
 225
 226             vformat->vap_prog_stream_cntl_ext[i >> 1] |=
 227                 (R300_VAP_SWIZZLE_XYZW << 16);
 228         } else {
 229             vformat->vap_prog_stream_cntl[i >> 1] &= 0xffff0000;
 230             vformat->vap_prog_stream_cntl[i >> 1] |= temp <<  0;
 231
 232             vformat->vap_prog_stream_cntl_ext[i >> 1] |=
 233                 (R300_VAP_SWIZZLE_XYZW <<  0);
 234         }
 235     }
 236
 237     /* Set the last vector in the PSC. */
 238     i--;
 239     vformat->vap_prog_stream_cntl[i >> 1] |=
 240         (R300_LAST_VEC << (i & 1 ? 16 : 0));
 241 }
 242
 243 /* Update the vertex format. */
 244 static void r300_update_vertex_format(struct r300_context* r300)
 245 {
 246     struct r300_vertex_format vformat;
 247     int i;
 248
 249     memset(&vformat, 0, sizeof(struct r300_vertex_format));
 250     for (i = 0; i < 16; i++) {
 251         vformat.vs_tab[i] = -1;
 252         vformat.fs_tab[i] = -1;
 253     }
 254
 255     r300_vs_tab_routes(r300, &vformat);
 256
 257     r300_vertex_psc(r300, &vformat);
 258
 259     if (memcmp(&r300->vertex_info, &vformat,
 260                 sizeof(struct r300_vertex_format))) {
 261         memcpy(&r300->vertex_info, &vformat,
 262                 sizeof(struct r300_vertex_format));
 263         r300->dirty_state |= R300_NEW_VERTEX_FORMAT;
 264     }
 265 }
 266
 267 /* Set up the mappings from GB to US, for RS block. */
 268 static void r300_update_fs_tab(struct r300_context* r300)
 269 {
 270     struct r300_vertex_format* vformat = &r300->vertex_info;
 271     struct tgsi_shader_info* info = &r300->fs->info;
 272     int i, cols = 0, texs = 0, cols_emitted = 0;
 273     int* tab = vformat->fs_tab;
 274
 275     for (i = 0; i < 16; i++) {
 276         tab[i] = -1;
 277     }
 278
 279     assert(info->num_inputs <= 16);
 280     for (i = 0; i < info->num_inputs; i++) {
 281         switch (info->input_semantic_name[i]) {
 282             case TGSI_SEMANTIC_COLOR:
 283                 tab[i] = INTERP_LINEAR;
 284                 cols++;
 285                 break;
 286             case TGSI_SEMANTIC_POSITION:
 287             case TGSI_SEMANTIC_PSIZE:
 288                 debug_printf("r300: Implementation error: Can't use "
 289                         "pos attribs in fragshader yet!\n");
 290                 /* Pass through for now */
 291             case TGSI_SEMANTIC_FOG:
 292             case TGSI_SEMANTIC_GENERIC:
 293                 tab[i] = INTERP_PERSPECTIVE;
 294                 break;
 295             default:
 296                 debug_printf("r300: Unknown vertex input %d\n",
 297                     info->input_semantic_name[i]);
 298                 break;
 299         }
 300     }
 301
 302     /* Now that we know where everything is... */
 303     DBG(r300, DBG_DRAW, "r300: fp input count: %d\n", info->num_inputs);
 304     for (i = 0; i < info->num_inputs; i++) {
 305         switch (tab[i]) {
 306             case INTERP_LINEAR:
 307                 DBG(r300, DBG_DRAW, "r300: attrib: "
 308                         "stack offset %d, color,    tab %d\n",
 309                         i, cols_emitted);
 310                 tab[i] = cols_emitted;
 311                 cols_emitted++;
 312                 break;
 313             case INTERP_PERSPECTIVE:
 314                 DBG(r300, DBG_DRAW, "r300: attrib: "
 315                         "stack offset %d, texcoord, tab %d\n",
 316                         i, cols + texs);
 317                 tab[i] = cols + texs;
 318                 texs++;
 319                 break;
 320             case -1:
 321                 debug_printf("r300: Implementation error: Bad fp interp!\n");
 322             default:
 323                 break;
 324         }
 325     }
 326
 327 }
 328
 329 /* Set up the RS block. This is the part of the chipset that actually does
 330  * the rasterization of vertices into fragments. This is also the part of the
 331  * chipset that locks up if any part of it is even slightly wrong. */
 332 static void r300_update_rs_block(struct r300_context* r300)
 333 {
 334     struct r300_rs_block* rs = r300->rs_block;
 335     struct tgsi_shader_info* info = &r300->fs->info;
 336     int* tab = r300->vertex_info.fs_tab;
 337     int col_count = 0, fp_offset = 0, i, memory_pos, tex_count = 0;
 338
 339     memset(rs, 0, sizeof(struct r300_rs_block));
 340
 341     if (r300_screen(r300->context.screen)->caps->is_r500) {
 342         for (i = 0; i < info->num_inputs; i++) {
 343             assert(tab[i] != -1);
 344             memory_pos = tab[i] * 4;
 345             switch (info->input_semantic_name[i]) {
 346                 case TGSI_SEMANTIC_COLOR:
 347                     rs->ip[col_count] |=
 348                         R500_RS_COL_PTR(memory_pos) |
 349                         R500_RS_COL_FMT(R300_RS_COL_FMT_RGBA);
 350                     col_count++;
 351                     break;
 352                 case TGSI_SEMANTIC_GENERIC:
 353                     rs->ip[tex_count] |=
 354                         R500_RS_SEL_S(memory_pos) |
 355                         R500_RS_SEL_T(memory_pos + 1) |
 356                         R500_RS_SEL_R(memory_pos + 2) |
 357                         R500_RS_SEL_Q(memory_pos + 3);
 358                     tex_count++;
 359                     break;
 360                 default:
 361                     break;
 362             }
 363         }
 364
 365         if (col_count == 0) {
 366             rs->ip[0] |= R500_RS_COL_FMT(R300_RS_COL_FMT_0001);
 367         }
 368
 369         if (tex_count == 0) {
 370             rs->ip[0] |=
 371                 R500_RS_SEL_S(R500_RS_IP_PTR_K0) |
 372                 R500_RS_SEL_T(R500_RS_IP_PTR_K0) |
 373                 R500_RS_SEL_R(R500_RS_IP_PTR_K0) |
 374                 R500_RS_SEL_Q(R500_RS_IP_PTR_K1);
 375         }
 376
 377         /* Rasterize at least one color, or bad things happen. */
 378         if ((col_count == 0) && (tex_count == 0)) {
 379             col_count++;
 380         }
 381
 382         for (i = 0; i < tex_count; i++) {
 383             rs->inst[i] |= R500_RS_INST_TEX_ID(i) |
 384                 R500_RS_INST_TEX_CN_WRITE | R500_RS_INST_TEX_ADDR(fp_offset);
 385             fp_offset++;
 386         }
 387
 388         for (i = 0; i < col_count; i++) {
 389             rs->inst[i] |= R500_RS_INST_COL_ID(i) |
 390                 R500_RS_INST_COL_CN_WRITE | R500_RS_INST_COL_ADDR(fp_offset);
 391             fp_offset++;
 392         }
 393     } else {
 394         for (i = 0; i < info->num_inputs; i++) {
 395             assert(tab[i] != -1);
 396             memory_pos = tab[i] * 4;
 397             switch (info->input_semantic_name[i]) {
 398                 case TGSI_SEMANTIC_COLOR:
 399                     rs->ip[col_count] |=
 400                         R300_RS_COL_PTR(memory_pos) |
 401                         R300_RS_COL_FMT(R300_RS_COL_FMT_RGBA);
 402                     col_count++;
 403                     break;
 404                 case TGSI_SEMANTIC_GENERIC:
 405                     rs->ip[tex_count] |=
 406                         R300_RS_TEX_PTR(memory_pos) |
 407                         R300_RS_SEL_S(R300_RS_SEL_C0) |
 408                         R300_RS_SEL_T(R300_RS_SEL_C1) |
 409                         R300_RS_SEL_R(R300_RS_SEL_C2) |
 410                         R300_RS_SEL_Q(R300_RS_SEL_C3);
 411                     tex_count++;
 412                     break;
 413                 default:
 414                     break;
 415             }
 416         }
 417
 418         if (col_count == 0) {
 419             rs->ip[0] |= R300_RS_COL_FMT(R300_RS_COL_FMT_0001);
 420         }
 421
 422         if (tex_count == 0) {
 423             rs->ip[0] |=
 424                 R300_RS_SEL_S(R300_RS_SEL_K0) |
 425                 R300_RS_SEL_T(R300_RS_SEL_K0) |
 426                 R300_RS_SEL_R(R300_RS_SEL_K0) |
 427                 R300_RS_SEL_Q(R300_RS_SEL_K1);
 428         }
 429
 430         /* Rasterize at least one color, or bad things happen. */
 431         if ((col_count == 0) && (tex_count == 0)) {
 432             col_count++;
 433         }
 434
 435         for (i = 0; i < tex_count; i++) {
 436             rs->inst[i] |= R300_RS_INST_TEX_ID(i) |
 437                 R300_RS_INST_TEX_CN_WRITE | R300_RS_INST_TEX_ADDR(fp_offset);
 438             fp_offset++;
 439         }
 440
 441         for (i = 0; i < col_count; i++) {
 442             rs->inst[i] |= R300_RS_INST_COL_ID(i) |
 443                 R300_RS_INST_COL_CN_WRITE | R300_RS_INST_COL_ADDR(fp_offset);
 444             fp_offset++;
 445         }
 446     }
 447
 448     rs->count = (tex_count * 4) | (col_count << R300_IC_COUNT_SHIFT) |
 449         R300_HIRES_EN;
 450
 451     rs->inst_count = MAX2(MAX2(col_count - 1, tex_count - 1), 0);
 452 }
 453
 454 void r300_update_derived_state(struct r300_context* r300)
 455 {
 456     if (r300->dirty_state &
 457             (R300_NEW_FRAGMENT_SHADER | R300_NEW_VERTEX_SHADER)) {
 458         r300_update_vertex_format(r300);
 459     }
 460
 461     if (r300->dirty_state & R300_NEW_VERTEX_FORMAT) {
 462         r300_update_fs_tab(r300);
 463         r300_update_rs_block(r300);
 464     }
 465 }