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