src/gallium/drivers/r300/r300_vs_draw.c

   1 /**************************************************************************
   2  *
   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
   7  * "Software"), to deal in the Software without restriction, including
   8  * without limitation the rights to use, copy, modify, merge, publish,
   9  * distribute, sub license, and/or sell copies of the Software, and to
  10  * permit persons to whom the Software is furnished to do so, subject to
  11  * the following conditions:
  12  *
  13  * The above copyright notice and this permission notice (including the
  14  * next paragraph) shall be included in all copies or substantial portions
  15  * of the Software.
  16  *
  17  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
  18  * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  19  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
  20  * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
  21  * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
  22  * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
  23  * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  24  *
  25  **************************************************************************/
  26
  27 /* This file contains the vertex shader tranformations for SW TCL needed
  28  * to overcome the limitations of the r300 rasterizer.
  29  *
  30  * Transformations:
  31  * 1) If the secondary color output is present, the primary color must be
  32  *    inserted before it.
  33  * 2) If any back-face color output is present, there must be all 4 color
  34  *    outputs and missing ones must be inserted.
  35  * 3) Insert a trailing texcoord output containing a copy of POS, for WPOS.
  36  *
  37  * I know this code is cumbersome, but I don't know of any nicer way
  38  * of transforming TGSI shaders. ~ M.
  39  */
  40
  41 #include "r300_vs.h"
  42
  43 #include <stdio.h>
  44
  45 #include "tgsi/tgsi_transform.h"
  46 #include "tgsi/tgsi_dump.h"
  47
  48 #include "draw/draw_context.h"
  49
  50 struct vs_transform_context {
  51     struct tgsi_transform_context base;
  52
  53     boolean color_used[2];
  54     boolean bcolor_used[2];
  55     boolean temp_used[128];
  56
  57     /* Index of the pos output, typically 0. */
  58     unsigned pos_output;
  59     /* Index of the pos temp where all writes of pos are redirected to. */
  60     unsigned pos_temp;
  61     /* The index of the last generic output, after which we insert a new
  62      * output for WPOS. */
  63     int last_generic;
  64
  65     unsigned num_outputs;
  66     /* Used to shift output decl. indices when inserting new ones. */
  67     unsigned decl_shift;
  68     /* Used to remap writes to output decls if their indices changed. */
  69     unsigned out_remap[32];
  70
  71     /* First instruction processed? */
  72     boolean first_instruction;
  73     /* End instruction processed? */
  74     boolean end_instruction;
  75 };
  76
  77 static void emit_temp(struct tgsi_transform_context *ctx, unsigned reg)
  78 {
  79     struct tgsi_full_declaration decl;
  80
  81     decl = tgsi_default_full_declaration();
  82     decl.Declaration.File = TGSI_FILE_TEMPORARY;
  83     decl.Range.First = decl.Range.Last = reg;
  84     ctx->emit_declaration(ctx, &decl);
  85 }
  86
  87 static void emit_output(struct tgsi_transform_context *ctx,
  88                         unsigned name, unsigned index, unsigned interp,
  89                         unsigned reg)
  90 {
  91     struct vs_transform_context *vsctx = (struct vs_transform_context *)ctx;
  92     struct tgsi_full_declaration decl;
  93
  94     decl = tgsi_default_full_declaration();
  95     decl.Declaration.File = TGSI_FILE_OUTPUT;
  96     decl.Declaration.Interpolate = interp;
  97     decl.Declaration.Semantic = TRUE;
  98     decl.Semantic.Name = name;
  99     decl.Semantic.Index = index;
 100     decl.Range.First = decl.Range.Last = reg;
 101     ctx->emit_declaration(ctx, &decl);
 102     ++vsctx->num_outputs;
 103 }
 104
 105 static void insert_output(struct tgsi_transform_context *ctx,
 106                           struct tgsi_full_declaration *before,
 107                           unsigned name, unsigned index, unsigned interp)
 108 {
 109     struct vs_transform_context *vsctx = (struct vs_transform_context *)ctx;
 110     unsigned i;
 111
 112     /* Make a place for the new output. */
 113     for (i = before->Range.First; i < Elements(vsctx->out_remap); i++) {
 114         ++vsctx->out_remap[i];
 115     }
 116
 117     /* Insert the new output. */
 118     emit_output(ctx, name, index, interp, before->Range.First);
 119
 120     ++vsctx->decl_shift;
 121 }
 122
 123 static void insert_trailing_bcolor(struct tgsi_transform_context *ctx,
 124                                    struct tgsi_full_declaration *before)
 125 {
 126     struct vs_transform_context *vsctx = (struct vs_transform_context *)ctx;
 127
 128     /* If BCOLOR0 is used, make sure BCOLOR1 is present too. Otherwise
 129      * the rasterizer doesn't do the color selection correctly. */
 130     if (vsctx->bcolor_used[0] && !vsctx->bcolor_used[1]) {
 131         if (before) {
 132             insert_output(ctx, before, TGSI_SEMANTIC_BCOLOR, 1,
 133                           TGSI_INTERPOLATE_LINEAR);
 134         } else {
 135             emit_output(ctx, TGSI_SEMANTIC_BCOLOR, 1,
 136                         TGSI_INTERPOLATE_LINEAR, vsctx->num_outputs);
 137         }
 138         vsctx->bcolor_used[1] = TRUE;
 139     }
 140 }
 141
 142 static void transform_decl(struct tgsi_transform_context *ctx,
 143                            struct tgsi_full_declaration *decl)
 144 {
 145     struct vs_transform_context *vsctx = (struct vs_transform_context *)ctx;
 146     unsigned i;
 147
 148     if (decl->Declaration.File == TGSI_FILE_OUTPUT) {
 149         switch (decl->Semantic.Name) {
 150             case TGSI_SEMANTIC_POSITION:
 151                 vsctx->pos_output = decl->Range.First;
 152                 break;
 153
 154             case TGSI_SEMANTIC_COLOR:
 155                 assert(decl->Semantic.Index < 2);
 156                 vsctx->color_used[decl->Semantic.Index] = TRUE;
 157
 158                 /* We must rasterize the first color if the second one is
 159                  * used, otherwise the rasterizer doesn't do the color
 160                  * selection correctly. Declare it, but don't write to it. */
 161                 if (decl->Semantic.Index == 1 && !vsctx->color_used[0]) {
 162                     insert_output(ctx, decl, TGSI_SEMANTIC_COLOR, 0,
 163                                   TGSI_INTERPOLATE_LINEAR);
 164                     vsctx->color_used[0] = TRUE;
 165                 }
 166                 break;
 167
 168             case TGSI_SEMANTIC_BCOLOR:
 169                 assert(decl->Semantic.Index < 2);
 170                 vsctx->bcolor_used[decl->Semantic.Index] = TRUE;
 171
 172                 /* We must rasterize all 4 colors if back-face colors are
 173                  * used, otherwise the rasterizer doesn't do the color
 174                  * selection correctly. Declare it, but don't write to it. */
 175                 if (!vsctx->color_used[0]) {
 176                     insert_output(ctx, decl, TGSI_SEMANTIC_COLOR, 0,
 177                                   TGSI_INTERPOLATE_LINEAR);
 178                     vsctx->color_used[0] = TRUE;
 179                 }
 180                 if (!vsctx->color_used[1]) {
 181                     insert_output(ctx, decl, TGSI_SEMANTIC_COLOR, 1,
 182                                   TGSI_INTERPOLATE_LINEAR);
 183                     vsctx->color_used[1] = TRUE;
 184                 }
 185                 if (decl->Semantic.Index == 1 && !vsctx->bcolor_used[0]) {
 186                     insert_output(ctx, decl, TGSI_SEMANTIC_BCOLOR, 0,
 187                                   TGSI_INTERPOLATE_LINEAR);
 188                     vsctx->color_used[2] = TRUE;
 189                 }
 190                 /* One more case is handled in insert_trailing_bcolor. */
 191                 break;
 192
 193             case TGSI_SEMANTIC_GENERIC:
 194                 vsctx->last_generic = MAX2(vsctx->last_generic, decl->Semantic.Index);
 195                 break;
 196         }
 197
 198         if (decl->Semantic.Name != TGSI_SEMANTIC_BCOLOR) {
 199             /* Insert it as soon as possible. */
 200             insert_trailing_bcolor(ctx, decl);
 201         }
 202
 203         /* Since we're inserting new outputs in between, the following outputs
 204          * should be moved to the right so that they don't overlap with
 205          * the newly added ones. */
 206         decl->Range.First += vsctx->decl_shift;
 207         decl->Range.Last += vsctx->decl_shift;
 208
 209         ++vsctx->num_outputs;
 210     } else if (decl->Declaration.File == TGSI_FILE_TEMPORARY) {
 211         for (i = decl->Range.First; i <= decl->Range.Last; i++) {
 212            vsctx->temp_used[i] = TRUE;
 213         }
 214     }
 215
 216     ctx->emit_declaration(ctx, decl);
 217 }
 218
 219 static void transform_inst(struct tgsi_transform_context *ctx,
 220                            struct tgsi_full_instruction *inst)
 221 {
 222     struct vs_transform_context *vsctx = (struct vs_transform_context *) ctx;
 223     struct tgsi_full_instruction new_inst;
 224     unsigned i;
 225
 226     if (!vsctx->first_instruction) {
 227         vsctx->first_instruction = TRUE;
 228
 229         /* The trailing BCOLOR should be inserted before the code
 230          * if it hasn't already been done so. */
 231         insert_trailing_bcolor(ctx, NULL);
 232
 233         /* Insert the generic output for WPOS. */
 234         emit_output(ctx, TGSI_SEMANTIC_GENERIC, vsctx->last_generic + 1,
 235                     TGSI_INTERPOLATE_PERSPECTIVE, vsctx->num_outputs);
 236
 237         /* Find a free temp for POSITION. */
 238         for (i = 0; i < Elements(vsctx->temp_used); i++) {
 239             if (!vsctx->temp_used[i]) {
 240                 emit_temp(ctx, i);
 241                 vsctx->pos_temp = i;
 242                 break;
 243             }
 244         }
 245     }
 246
 247     if (inst->Instruction.Opcode == TGSI_OPCODE_END) {
 248         /* MOV OUT[pos_output], TEMP[pos_temp]; */
 249         new_inst = tgsi_default_full_instruction();
 250         new_inst.Instruction.Opcode = TGSI_OPCODE_MOV;
 251         new_inst.Instruction.NumDstRegs = 1;
 252         new_inst.Dst[0].Register.File = TGSI_FILE_OUTPUT;
 253         new_inst.Dst[0].Register.Index = vsctx->pos_output;
 254         new_inst.Dst[0].Register.WriteMask = TGSI_WRITEMASK_XYZW;
 255         new_inst.Instruction.NumSrcRegs = 1;
 256         new_inst.Src[0].Register.File = TGSI_FILE_TEMPORARY;
 257         new_inst.Src[0].Register.Index = vsctx->pos_temp;
 258         ctx->emit_instruction(ctx, &new_inst);
 259
 260         /* MOV OUT[n-1], TEMP[pos_temp]; */
 261         new_inst = tgsi_default_full_instruction();
 262         new_inst.Instruction.Opcode = TGSI_OPCODE_MOV;
 263         new_inst.Instruction.NumDstRegs = 1;
 264         new_inst.Dst[0].Register.File = TGSI_FILE_OUTPUT;
 265         new_inst.Dst[0].Register.Index = vsctx->num_outputs - 1;
 266         new_inst.Dst[0].Register.WriteMask = TGSI_WRITEMASK_XYZW;
 267         new_inst.Instruction.NumSrcRegs = 1;
 268         new_inst.Src[0].Register.File = TGSI_FILE_TEMPORARY;
 269         new_inst.Src[0].Register.Index = vsctx->pos_temp;
 270         ctx->emit_instruction(ctx, &new_inst);
 271
 272         vsctx->end_instruction = TRUE;
 273     } else {
 274         /* Not an END instruction. */
 275         /* Fix writes to outputs. */
 276         for (i = 0; i < inst->Instruction.NumDstRegs; i++) {
 277             struct tgsi_full_dst_register *dst = &inst->Dst[i];
 278             if (dst->Register.File == TGSI_FILE_OUTPUT) {
 279                 if (dst->Register.Index == vsctx->pos_output) {
 280                     /* Replace writes to OUT[pos_output] with TEMP[pos_temp]. */
 281                     dst->Register.File = TGSI_FILE_TEMPORARY;
 282                     dst->Register.Index = vsctx->pos_temp;
 283                 } else {
 284                     /* Not a position, good...
 285                      * Since we were changing the indices of output decls,
 286                      * we must redirect writes into them too. */
 287                     dst->Register.Index = vsctx->out_remap[dst->Register.Index];
 288                 }
 289             }
 290         }
 291
 292         /* Inserting 2 instructions before the END opcode moves all following
 293          * labels by 2. Subroutines are always after the END opcode so
 294          * they're always moved. */
 295         if (inst->Instruction.Opcode == TGSI_OPCODE_CAL) {
 296             inst->Label.Label += 2;
 297         }
 298         /* The labels of the following opcodes are moved only after
 299          * the END opcode. */
 300         if (vsctx->end_instruction &&
 301             (inst->Instruction.Opcode == TGSI_OPCODE_IF ||
 302              inst->Instruction.Opcode == TGSI_OPCODE_ELSE ||
 303              inst->Instruction.Opcode == TGSI_OPCODE_BGNLOOP ||
 304              inst->Instruction.Opcode == TGSI_OPCODE_ENDLOOP)) {
 305             inst->Label.Label += 2;
 306         }
 307     }
 308
 309     ctx->emit_instruction(ctx, inst);
 310 }
 311
 312 void r300_draw_init_vertex_shader(struct draw_context *draw,
 313                                   struct r300_vertex_shader *vs)
 314 {
 315     struct pipe_shader_state new_vs;
 316     struct vs_transform_context transform;
 317     const uint newLen = tgsi_num_tokens(vs->state.tokens) + 100 /* XXX */;
 318     unsigned i;
 319
 320     new_vs.tokens = tgsi_alloc_tokens(newLen);
 321     if (new_vs.tokens == NULL)
 322         return;
 323
 324     memset(&transform, 0, sizeof(transform));
 325     for (i = 0; i < Elements(transform.out_remap); i++) {
 326         transform.out_remap[i] = i;
 327     }
 328     transform.last_generic = -1;
 329     transform.base.transform_instruction = transform_inst;
 330     transform.base.transform_declaration = transform_decl;
 331
 332     tgsi_transform_shader(vs->state.tokens,
 333                           (struct tgsi_token*)new_vs.tokens,
 334                           newLen, &transform.base);
 335
 336 #if 0
 337     printf("----------------------------------------------\norig shader:\n");
 338     tgsi_dump(vs->state.tokens, 0);
 339     printf("----------------------------------------------\nnew shader:\n");
 340     tgsi_dump(new_vs.tokens, 0);
 341     printf("----------------------------------------------\n");
 342 #endif
 343
 344     /* Free old tokens. */
 345     FREE((void*)vs->state.tokens);
 346
 347     vs->draw_vs = draw_create_vertex_shader(draw, &new_vs);
 348
 349     /* Instead of duplicating and freeing the tokens, copy the pointer directly. */
 350     vs->state.tokens = new_vs.tokens;
 351
 352     /* Init the VS output table for the rasterizer. */
 353     r300_init_vs_outputs(vs);
 354
 355     /* Make the last generic be WPOS. */
 356     vs->outputs.wpos = vs->outputs.generic[transform.last_generic + 1];
 357     vs->outputs.generic[transform.last_generic + 1] = ATTR_UNUSED;
 358 }