src/gallium/drivers/r300/r300_vs.c

   1 /*
   2  * Copyright 2009 Corbin Simpson <MostAwesomeDude@gmail.com>
   3  * Copyright 2009 Marek Olšák <maraeo@gmail.com>
   4  *
   5  * Permission is hereby granted, free of charge, to any person obtaining a
   6  * copy of this software and associated documentation files (the "Software"),
   7  * to deal in the Software without restriction, including without limitation
   8  * on the rights to use, copy, modify, merge, publish, distribute, sub
   9  * license, and/or sell copies of the Software, and to permit persons to whom
  10  * the Software is furnished to do so, subject to the following conditions:
  11  *
  12  * The above copyright notice and this permission notice (including the next
  13  * paragraph) shall be included in all copies or substantial portions of the
  14  * Software.
  15  *
  16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  18  * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
  19  * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
  20  * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
  21  * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
  22  * USE OR OTHER DEALINGS IN THE SOFTWARE. */
  23
  24 #include "r300_vs.h"
  25 #include "r300_fs.h"
  26
  27 #include "r300_context.h"
  28 #include "r300_screen.h"
  29 #include "r300_tgsi_to_rc.h"
  30 #include "r300_reg.h"
  31
  32 #include "tgsi/tgsi_dump.h"
  33 #include "tgsi/tgsi_parse.h"
  34
  35 #include "radeon_compiler.h"
  36
  37 #include "util/u_math.h"
  38
  39 /* Convert info about VS output semantics into r300_shader_semantics. */
  40 static void r300_shader_read_vs_outputs(
  41     struct tgsi_shader_info* info,
  42     struct r300_shader_semantics* vs_outputs)
  43 {
  44     int i;
  45     unsigned index;
  46
  47     r300_shader_semantics_reset(vs_outputs);
  48
  49     for (i = 0; i < info->num_outputs; i++) {
  50         index = info->output_semantic_index[i];
  51
  52         switch (info->output_semantic_name[i]) {
  53             case TGSI_SEMANTIC_POSITION:
  54                 assert(index == 0);
  55                 vs_outputs->pos = i;
  56                 break;
  57
  58             case TGSI_SEMANTIC_PSIZE:
  59                 assert(index == 0);
  60                 vs_outputs->psize = i;
  61                 break;
  62
  63             case TGSI_SEMANTIC_COLOR:
  64                 assert(index <= ATTR_COLOR_COUNT);
  65                 vs_outputs->color[index] = i;
  66                 break;
  67
  68             case TGSI_SEMANTIC_BCOLOR:
  69                 assert(index <= ATTR_COLOR_COUNT);
  70                 vs_outputs->bcolor[index] = i;
  71                 break;
  72
  73             case TGSI_SEMANTIC_GENERIC:
  74                 assert(index <= ATTR_GENERIC_COUNT);
  75                 vs_outputs->generic[index] = i;
  76                 break;
  77
  78             case TGSI_SEMANTIC_FOG:
  79                 assert(index == 0);
  80                 vs_outputs->fog = i;
  81                 break;
  82
  83             case TGSI_SEMANTIC_EDGEFLAG:
  84                 assert(index == 0);
  85                 fprintf(stderr, "r300 VP: cannot handle edgeflag output\n");
  86                 assert(0);
  87                 break;
  88             default:
  89                 assert(0);
  90         }
  91     }
  92 }
  93
  94 static void r300_shader_vap_output_fmt(struct r300_vertex_shader* vs)
  95 {
  96     struct r300_shader_semantics* vs_outputs = &vs->outputs;
  97     uint32_t* hwfmt = vs->hwfmt;
  98     int i, gen_count;
  99     boolean any_bcolor_used = vs_outputs->bcolor[0] != ATTR_UNUSED ||
 100                               vs_outputs->bcolor[1] != ATTR_UNUSED;
 101
 102     /* Do the actual vertex_info setup.
 103      *
 104      * vertex_info has four uints of hardware-specific data in it.
 105      * vinfo.hwfmt[0] is R300_VAP_VTX_STATE_CNTL
 106      * vinfo.hwfmt[1] is R300_VAP_VSM_VTX_ASSM
 107      * vinfo.hwfmt[2] is R300_VAP_OUTPUT_VTX_FMT_0
 108      * vinfo.hwfmt[3] is R300_VAP_OUTPUT_VTX_FMT_1 */
 109
 110     hwfmt[0] = 0x5555; /* XXX this is classic Mesa bonghits */
 111
 112     /* Position. */
 113     if (vs_outputs->pos != ATTR_UNUSED) {
 114         hwfmt[1] |= R300_INPUT_CNTL_POS;
 115         hwfmt[2] |= R300_VAP_OUTPUT_VTX_FMT_0__POS_PRESENT;
 116     } else {
 117         assert(0);
 118     }
 119
 120     /* Point size. */
 121     if (vs_outputs->psize != ATTR_UNUSED) {
 122         hwfmt[2] |= R300_VAP_OUTPUT_VTX_FMT_0__PT_SIZE_PRESENT;
 123     }
 124
 125     /* Colors. */
 126     for (i = 0; i < ATTR_COLOR_COUNT; i++) {
 127         if (vs_outputs->color[i] != ATTR_UNUSED || any_bcolor_used) {
 128             hwfmt[1] |= R300_INPUT_CNTL_COLOR;
 129             hwfmt[2] |= R300_VAP_OUTPUT_VTX_FMT_0__COLOR_0_PRESENT << i;
 130         }
 131     }
 132
 133     /* Back-face colors. */
 134     if (any_bcolor_used) {
 135         for (i = 0; i < ATTR_COLOR_COUNT; i++) {
 136             hwfmt[1] |= R300_INPUT_CNTL_COLOR;
 137             hwfmt[2] |= R300_VAP_OUTPUT_VTX_FMT_0__COLOR_0_PRESENT << (2+i);
 138         }
 139     }
 140
 141     /* Texture coordinates. */
 142     gen_count = 0;
 143     for (i = 0; i < ATTR_GENERIC_COUNT; i++) {
 144         if (vs_outputs->generic[i] != ATTR_UNUSED) {
 145             hwfmt[1] |= (R300_INPUT_CNTL_TC0 << gen_count);
 146             hwfmt[3] |= (4 << (3 * gen_count));
 147             gen_count++;
 148         }
 149     }
 150
 151     /* Fog coordinates. */
 152     if (vs_outputs->fog != ATTR_UNUSED) {
 153         hwfmt[1] |= (R300_INPUT_CNTL_TC0 << gen_count);
 154         hwfmt[3] |= (4 << (3 * gen_count));
 155         gen_count++;
 156     }
 157
 158     /* XXX magic */
 159     assert(gen_count <= 8);
 160
 161     /* WPOS. */
 162     vs->wpos_tex_output = gen_count;
 163 }
 164
 165 /* Sets up stream mapping to equivalent VS outputs if TCL is bypassed
 166  * or isn't present. */
 167 static void r300_stream_locations_notcl(
 168     struct r300_shader_semantics* vs_outputs,
 169     int* stream_loc)
 170 {
 171     int i, tabi = 0, gen_count;
 172     boolean any_bcolor_used = vs_outputs->bcolor[0] != ATTR_UNUSED ||
 173                               vs_outputs->bcolor[1] != ATTR_UNUSED;
 174
 175     /* Position. */
 176     stream_loc[tabi++] = 0;
 177
 178     /* Point size. */
 179     if (vs_outputs->psize != ATTR_UNUSED) {
 180         stream_loc[tabi++] = 1;
 181     }
 182
 183     /* Colors. */
 184     for (i = 0; i < ATTR_COLOR_COUNT; i++) {
 185         if (vs_outputs->color[i] != ATTR_UNUSED || any_bcolor_used) {
 186             stream_loc[tabi++] = 2 + i;
 187         }
 188     }
 189
 190     /* Back-face colors. */
 191     if (any_bcolor_used) {
 192         for (i = 0; i < ATTR_COLOR_COUNT; i++) {
 193             stream_loc[tabi++] = 4 + i;
 194         }
 195     }
 196
 197     /* Texture coordinates. */
 198     gen_count = 0;
 199     for (i = 0; i < ATTR_GENERIC_COUNT; i++) {
 200         if (vs_outputs->generic[i] != ATTR_UNUSED) {
 201             assert(tabi < 16);
 202             stream_loc[tabi++] = 6 + gen_count;
 203             gen_count++;
 204         }
 205     }
 206
 207     /* Fog coordinates. */
 208     if (vs_outputs->fog != ATTR_UNUSED) {
 209         assert(tabi < 16);
 210         stream_loc[tabi++] = 6 + gen_count;
 211         gen_count++;
 212     }
 213
 214     /* WPOS. */
 215     if (vs_outputs->wpos != ATTR_UNUSED) {
 216         assert(tabi < 16);
 217         stream_loc[tabi++] = 6 + gen_count;
 218         gen_count++;
 219     }
 220
 221     for (; tabi < 16;) {
 222         stream_loc[tabi++] = -1;
 223     }
 224 }
 225
 226 static void set_vertex_inputs_outputs(struct r300_vertex_program_compiler * c)
 227 {
 228     struct r300_vertex_shader * vs = c->UserData;
 229     struct r300_shader_semantics* outputs = &vs->outputs;
 230     struct tgsi_shader_info* info = &vs->info;
 231     int i, reg = 0;
 232     boolean any_bcolor_used = outputs->bcolor[0] != ATTR_UNUSED ||
 233                               outputs->bcolor[1] != ATTR_UNUSED;
 234
 235     /* Fill in the input mapping */
 236     for (i = 0; i < info->num_inputs; i++)
 237         c->code->inputs[i] = i;
 238
 239     /* Position. */
 240     if (outputs->pos != ATTR_UNUSED) {
 241         c->code->outputs[outputs->pos] = reg++;
 242     } else {
 243         assert(0);
 244     }
 245
 246     /* Point size. */
 247     if (outputs->psize != ATTR_UNUSED) {
 248         c->code->outputs[outputs->psize] = reg++;
 249     }
 250
 251     /* If we're writing back facing colors we need to send
 252      * four colors to make front/back face colors selection work.
 253      * If the vertex program doesn't write all 4 colors, lets
 254      * pretend it does by skipping output index reg so the colors
 255      * get written into appropriate output vectors.
 256      */
 257
 258     /* Colors. */
 259     for (i = 0; i < ATTR_COLOR_COUNT; i++) {
 260         if (outputs->color[i] != ATTR_UNUSED) {
 261             c->code->outputs[outputs->color[i]] = reg++;
 262         } else if (any_bcolor_used) {
 263             reg++;
 264         }
 265     }
 266
 267     /* Back-face colors. */
 268     for (i = 0; i < ATTR_COLOR_COUNT; i++) {
 269         if (outputs->bcolor[i] != ATTR_UNUSED) {
 270             c->code->outputs[outputs->bcolor[i]] = reg++;
 271         } else if (any_bcolor_used) {
 272             reg++;
 273         }
 274     }
 275
 276     /* Texture coordinates. */
 277     for (i = 0; i < ATTR_GENERIC_COUNT; i++) {
 278         if (outputs->generic[i] != ATTR_UNUSED) {
 279             c->code->outputs[outputs->generic[i]] = reg++;
 280         }
 281     }
 282
 283     /* Fog coordinates. */
 284     if (outputs->fog != ATTR_UNUSED) {
 285         c->code->outputs[outputs->fog] = reg++;
 286     }
 287
 288     /* WPOS. */
 289     if (outputs->wpos != ATTR_UNUSED) {
 290         c->code->outputs[outputs->wpos] = reg++;
 291     }
 292 }
 293
 294 static void r300_insert_wpos(struct r300_vertex_program_compiler* c,
 295                              struct r300_shader_semantics* outputs)
 296 {
 297     int i, lastOutput = 0;
 298
 299     /* Find the max output index. */
 300     lastOutput = MAX2(lastOutput, outputs->psize);
 301     for (i = 0; i < ATTR_COLOR_COUNT; i++) {
 302         lastOutput = MAX2(lastOutput, outputs->color[i]);
 303         lastOutput = MAX2(lastOutput, outputs->bcolor[i]);
 304     }
 305     for (i = 0; i < ATTR_GENERIC_COUNT; i++) {
 306         lastOutput = MAX2(lastOutput, outputs->generic[i]);
 307     }
 308     lastOutput = MAX2(lastOutput, outputs->fog);
 309
 310     /* Set WPOS after the last output. */
 311     lastOutput++;
 312     rc_copy_output(&c->Base, 0, lastOutput); /* out[lastOutput] = out[0]; */
 313     outputs->wpos = lastOutput;
 314 }
 315
 316 void r300_translate_vertex_shader(struct r300_context* r300,
 317                                   struct r300_vertex_shader* vs)
 318 {
 319     struct r300_vertex_program_compiler compiler;
 320     struct tgsi_to_rc ttr;
 321
 322     /* Initialize. */
 323     r300_shader_read_vs_outputs(&vs->info, &vs->outputs);
 324
 325     /* Setup the compiler */
 326     rc_init(&compiler.Base);
 327
 328     compiler.Base.Debug = DBG_ON(r300, DBG_VP);
 329     compiler.code = &vs->code;
 330     compiler.UserData = vs;
 331
 332     if (compiler.Base.Debug) {
 333         debug_printf("r300: Initial vertex program\n");
 334         tgsi_dump(vs->state.tokens, 0);
 335     }
 336
 337     /* Translate TGSI to our internal representation */
 338     ttr.compiler = &compiler.Base;
 339     ttr.info = &vs->info;
 340
 341     r300_tgsi_to_rc(&ttr, vs->state.tokens);
 342
 343     compiler.RequiredOutputs = ~(~0 << (vs->info.num_outputs+1));
 344     compiler.SetHwInputOutput = &set_vertex_inputs_outputs;
 345
 346     /* Insert the WPOS output. */
 347     r300_insert_wpos(&compiler, &vs->outputs);
 348
 349     r300_shader_vap_output_fmt(vs);
 350     r300_stream_locations_notcl(&vs->outputs, vs->stream_loc_notcl);
 351
 352     /* Invoke the compiler */
 353     r3xx_compile_vertex_program(&compiler);
 354     if (compiler.Base.Error) {
 355         /* XXX We should fallback using Draw. */
 356         fprintf(stderr, "r300 VP: Compiler error\n");
 357         abort();
 358     }
 359
 360     /* And, finally... */
 361     rc_destroy(&compiler.Base);
 362     vs->translated = TRUE;
 363 }
 364
 365 boolean r300_vertex_shader_setup_wpos(struct r300_context* r300)
 366 {
 367     struct r300_vertex_shader* vs = r300->vs;
 368     int tex_output = r300->vs->wpos_tex_output;
 369     uint32_t tex_fmt = R300_INPUT_CNTL_TC0 << tex_output;
 370     uint32_t* hwfmt = vs->hwfmt;
 371
 372     if (r300->fs->inputs.wpos != ATTR_UNUSED) {
 373         /* Enable WPOS in VAP. */
 374         if (!(hwfmt[1] & tex_fmt)) {
 375             hwfmt[1] |= tex_fmt;
 376             hwfmt[3] |= (4 << (3 * tex_output));
 377
 378             assert(tex_output < 8);
 379             return TRUE;
 380         }
 381     } else {
 382         /* Disable WPOS in VAP. */
 383         if (hwfmt[1] & tex_fmt) {
 384             hwfmt[1] &= ~tex_fmt;
 385             hwfmt[3] &= ~(4 << (3 * tex_output));
 386             return TRUE;
 387         }
 388     }
 389     return FALSE;
 390 }