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     /* WPOS is a straight copy of POSITION and it's always emitted. */
  94     vs_outputs->wpos = i;
  95 }
  96
  97 /* This function sets up:
  98  * - VAP mapping, which maps VS registers to output semantics and
  99  *   at the same time it indicates which attributes are enabled and should
 100  *   be rasterized.
 101  * - Stream mapping to VS outputs if TCL is not present. */
 102 static void r300_init_vs_output_mapping(struct r300_vertex_shader* vs)
 103 {
 104     struct r300_shader_semantics* vs_outputs = &vs->outputs;
 105     struct r300_vap_output_state *vap_out = &vs->vap_out;
 106     int *stream_loc = vs->stream_loc_notcl;
 107     int i, gen_count, tabi = 0;
 108     boolean any_bcolor_used = vs_outputs->bcolor[0] != ATTR_UNUSED ||
 109                               vs_outputs->bcolor[1] != ATTR_UNUSED;
 110
 111     vap_out->vap_vtx_state_cntl = 0x5555; /* XXX this is classic Mesa bonghits */
 112
 113     /* Position. */
 114     if (vs_outputs->pos != ATTR_UNUSED) {
 115         vap_out->vap_vsm_vtx_assm |= R300_INPUT_CNTL_POS;
 116         vap_out->vap_out_vtx_fmt[0] |= R300_VAP_OUTPUT_VTX_FMT_0__POS_PRESENT;
 117
 118         stream_loc[tabi++] = 0;
 119     } else {
 120         assert(0);
 121     }
 122
 123     /* Point size. */
 124     if (vs_outputs->psize != ATTR_UNUSED) {
 125         vap_out->vap_out_vtx_fmt[0] |= R300_VAP_OUTPUT_VTX_FMT_0__PT_SIZE_PRESENT;
 126
 127         stream_loc[tabi++] = 1;
 128     }
 129
 130     /* Colors. */
 131     for (i = 0; i < ATTR_COLOR_COUNT; i++) {
 132         if (vs_outputs->color[i] != ATTR_UNUSED || any_bcolor_used ||
 133             vs_outputs->color[1] != ATTR_UNUSED) {
 134             vap_out->vap_vsm_vtx_assm |= R300_INPUT_CNTL_COLOR;
 135             vap_out->vap_out_vtx_fmt[0] |= R300_VAP_OUTPUT_VTX_FMT_0__COLOR_0_PRESENT << i;
 136
 137             stream_loc[tabi++] = 2 + i;
 138         }
 139     }
 140
 141     /* Back-face colors. */
 142     if (any_bcolor_used) {
 143         for (i = 0; i < ATTR_COLOR_COUNT; i++) {
 144             vap_out->vap_vsm_vtx_assm |= R300_INPUT_CNTL_COLOR;
 145             vap_out->vap_out_vtx_fmt[0] |= R300_VAP_OUTPUT_VTX_FMT_0__COLOR_0_PRESENT << (2+i);
 146
 147             stream_loc[tabi++] = 4 + i;
 148         }
 149     }
 150
 151     /* Texture coordinates. */
 152     gen_count = 0;
 153     for (i = 0; i < ATTR_GENERIC_COUNT; i++) {
 154         if (vs_outputs->generic[i] != ATTR_UNUSED) {
 155             vap_out->vap_vsm_vtx_assm |= (R300_INPUT_CNTL_TC0 << gen_count);
 156             vap_out->vap_out_vtx_fmt[1] |= (4 << (3 * gen_count));
 157
 158             assert(tabi < 16);
 159             stream_loc[tabi++] = 6 + gen_count;
 160             gen_count++;
 161         }
 162     }
 163
 164     /* Fog coordinates. */
 165     if (vs_outputs->fog != ATTR_UNUSED) {
 166         vap_out->vap_vsm_vtx_assm |= (R300_INPUT_CNTL_TC0 << gen_count);
 167         vap_out->vap_out_vtx_fmt[1] |= (4 << (3 * gen_count));
 168
 169         assert(tabi < 16);
 170         stream_loc[tabi++] = 6 + gen_count;
 171         gen_count++;
 172     }
 173
 174     /* XXX magic */
 175     assert(gen_count <= 8);
 176
 177     /* WPOS. */
 178     vs->wpos_tex_output = gen_count;
 179
 180     assert(tabi < 16);
 181     stream_loc[tabi++] = 6 + gen_count;
 182
 183     for (; tabi < 16;) {
 184         stream_loc[tabi++] = -1;
 185     }
 186 }
 187
 188 static void set_vertex_inputs_outputs(struct r300_vertex_program_compiler * c)
 189 {
 190     struct r300_vertex_shader * vs = c->UserData;
 191     struct r300_shader_semantics* outputs = &vs->outputs;
 192     struct tgsi_shader_info* info = &vs->info;
 193     int i, reg = 0;
 194     boolean any_bcolor_used = outputs->bcolor[0] != ATTR_UNUSED ||
 195                               outputs->bcolor[1] != ATTR_UNUSED;
 196
 197     /* Fill in the input mapping */
 198     for (i = 0; i < info->num_inputs; i++)
 199         c->code->inputs[i] = i;
 200
 201     /* Position. */
 202     if (outputs->pos != ATTR_UNUSED) {
 203         c->code->outputs[outputs->pos] = reg++;
 204     } else {
 205         assert(0);
 206     }
 207
 208     /* Point size. */
 209     if (outputs->psize != ATTR_UNUSED) {
 210         c->code->outputs[outputs->psize] = reg++;
 211     }
 212
 213     /* If we're writing back facing colors we need to send
 214      * four colors to make front/back face colors selection work.
 215      * If the vertex program doesn't write all 4 colors, lets
 216      * pretend it does by skipping output index reg so the colors
 217      * get written into appropriate output vectors.
 218      */
 219
 220     /* Colors. */
 221     for (i = 0; i < ATTR_COLOR_COUNT; i++) {
 222         if (outputs->color[i] != ATTR_UNUSED) {
 223             c->code->outputs[outputs->color[i]] = reg++;
 224         } else if (any_bcolor_used ||
 225                    outputs->color[1] != ATTR_UNUSED) {
 226             reg++;
 227         }
 228     }
 229
 230     /* Back-face colors. */
 231     for (i = 0; i < ATTR_COLOR_COUNT; i++) {
 232         if (outputs->bcolor[i] != ATTR_UNUSED) {
 233             c->code->outputs[outputs->bcolor[i]] = reg++;
 234         } else if (any_bcolor_used) {
 235             reg++;
 236         }
 237     }
 238
 239     /* Texture coordinates. */
 240     for (i = 0; i < ATTR_GENERIC_COUNT; i++) {
 241         if (outputs->generic[i] != ATTR_UNUSED) {
 242             c->code->outputs[outputs->generic[i]] = reg++;
 243         }
 244     }
 245
 246     /* Fog coordinates. */
 247     if (outputs->fog != ATTR_UNUSED) {
 248         c->code->outputs[outputs->fog] = reg++;
 249     }
 250
 251     /* WPOS. */
 252     if (outputs->wpos != ATTR_UNUSED) {
 253         c->code->outputs[outputs->wpos] = reg++;
 254     }
 255 }
 256
 257 void r300_vertex_shader_common_init(struct r300_vertex_shader *vs,
 258                                     const struct pipe_shader_state *shader)
 259 {
 260     /* Copy state directly into shader. */
 261     vs->state = *shader;
 262     vs->state.tokens = tgsi_dup_tokens(shader->tokens);
 263     tgsi_scan_shader(shader->tokens, &vs->info);
 264
 265     r300_shader_read_vs_outputs(&vs->info, &vs->outputs);
 266     r300_init_vs_output_mapping(vs);
 267 }
 268
 269 void r300_translate_vertex_shader(struct r300_context* r300,
 270                                   struct r300_vertex_shader* vs)
 271 {
 272     struct r300_vertex_program_compiler compiler;
 273     struct tgsi_to_rc ttr;
 274
 275     /* Setup the compiler */
 276     rc_init(&compiler.Base);
 277
 278     compiler.Base.Debug = DBG_ON(r300, DBG_VP);
 279     compiler.code = &vs->code;
 280     compiler.UserData = vs;
 281
 282     if (compiler.Base.Debug) {
 283         debug_printf("r300: Initial vertex program\n");
 284         tgsi_dump(vs->state.tokens, 0);
 285     }
 286
 287     /* Translate TGSI to our internal representation */
 288     ttr.compiler = &compiler.Base;
 289     ttr.info = &vs->info;
 290     ttr.use_half_swizzles = FALSE;
 291
 292     r300_tgsi_to_rc(&ttr, vs->state.tokens);
 293
 294     compiler.RequiredOutputs = ~(~0 << (vs->info.num_outputs+1));
 295     compiler.SetHwInputOutput = &set_vertex_inputs_outputs;
 296
 297     /* Insert the WPOS output. */
 298     rc_copy_output(&compiler.Base, 0, vs->outputs.wpos);
 299
 300     /* Invoke the compiler */
 301     r3xx_compile_vertex_program(&compiler);
 302     if (compiler.Base.Error) {
 303         /* XXX We should fallback using Draw. */
 304         fprintf(stderr, "r300 VP: Compiler error\n");
 305         abort();
 306     }
 307
 308     /* And, finally... */
 309     rc_destroy(&compiler.Base);
 310 }
 311
 312 boolean r300_vertex_shader_setup_wpos(struct r300_context* r300)
 313 {
 314     struct r300_vertex_shader* vs = r300->vs_state.state;
 315     struct r300_vap_output_state *vap_out = &vs->vap_out;
 316     int tex_output = vs->wpos_tex_output;
 317     uint32_t tex_fmt = R300_INPUT_CNTL_TC0 << tex_output;
 318
 319     if (r300->fs->inputs.wpos != ATTR_UNUSED) {
 320         /* Enable WPOS in VAP. */
 321         if (!(vap_out->vap_vsm_vtx_assm & tex_fmt)) {
 322             vap_out->vap_vsm_vtx_assm |= tex_fmt;
 323             vap_out->vap_out_vtx_fmt[1] |= (4 << (3 * tex_output));
 324
 325             assert(tex_output < 8);
 326             return TRUE;
 327         }
 328     } else {
 329         /* Disable WPOS in VAP. */
 330         if (vap_out->vap_vsm_vtx_assm & tex_fmt) {
 331             vap_out->vap_vsm_vtx_assm &= ~tex_fmt;
 332             vap_out->vap_out_vtx_fmt[1] &= ~(4 << (3 * tex_output));
 333             return TRUE;
 334         }
 335     }
 336     return FALSE;
 337 }