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  *    present too.
  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
  56     /* Index of the pos output, typically 0. */
  57     unsigned pos_output;
  58     /* Index of the pos temp where all writes of pos are redirected to. */
  59     unsigned pos_temp;
  60     /* The index of the last generic output, after which we insert a new
  61      * output for WPOS. */
  62     int last_generic;
  63
  64     unsigned num_outputs;
  65     /* Used to shift output decl. indices when inserting new ones. */
  66     unsigned decl_shift;
  67     /* Used to remap writes to output decls if their indices changed. */
  68     unsigned out_remap[32];
  69
  70     /* First instruction processed? */
  71     boolean first_instruction;
  72     /* End instruction processed? */
  73     boolean end_instruction;
  74
  75     boolean temp_used[1024];
  76 };
  77
  78 static void emit_temp(struct tgsi_transform_context *ctx, unsigned reg)
  79 {
  80     struct tgsi_full_declaration decl;
  81
  82     decl = tgsi_default_full_declaration();
  83     decl.Declaration.File = TGSI_FILE_TEMPORARY;
  84     decl.Range.First = decl.Range.Last = reg;
  85     ctx->emit_declaration(ctx, &decl);
  86 }
  87
  88 static void emit_output(struct tgsi_transform_context *ctx,
  89                         unsigned name, unsigned index, unsigned interp,
  90                         unsigned reg)
  91 {
  92     struct vs_transform_context *vsctx = (struct vs_transform_context *)ctx;
  93     struct tgsi_full_declaration decl;
  94
  95     decl = tgsi_default_full_declaration();
  96     decl.Declaration.File = TGSI_FILE_OUTPUT;
  97     decl.Declaration.Interpolate = interp;
  98     decl.Declaration.Semantic = TRUE;
  99     decl.Semantic.Name = name;
 100     decl.Semantic.Index = index;
 101     decl.Range.First = decl.Range.Last = reg;
 102     ctx->emit_declaration(ctx, &decl);
 103     ++vsctx->num_outputs;
 104 }
 105
 106 static void insert_output_before(struct tgsi_transform_context *ctx,
 107                                  struct tgsi_full_declaration *before,
 108                                  unsigned name, unsigned index, unsigned interp)
 109 {
 110     struct vs_transform_context *vsctx = (struct vs_transform_context *)ctx;
 111     unsigned i;
 112
 113     /* Make a place for the new output. */
 114     for (i = before->Range.First; i < Elements(vsctx->out_remap); i++) {
 115         ++vsctx->out_remap[i];
 116     }
 117
 118     /* Insert the new output. */
 119     emit_output(ctx, name, index, interp,
 120                 before->Range.First + vsctx->decl_shift);
 121
 122     ++vsctx->decl_shift;
 123 }
 124
 125 static void insert_output_after(struct tgsi_transform_context *ctx,
 126                                 struct tgsi_full_declaration *after,
 127                                 unsigned name, unsigned index, unsigned interp)
 128 {
 129     struct vs_transform_context *vsctx = (struct vs_transform_context *)ctx;
 130     unsigned i;
 131
 132     /* Make a place for the new output. */
 133     for (i = after->Range.First+1; i < Elements(vsctx->out_remap); i++) {
 134         ++vsctx->out_remap[i];
 135     }
 136
 137     /* Insert the new output. */
 138     emit_output(ctx, name, index, interp,
 139                 after->Range.First + 1);
 140
 141     ++vsctx->decl_shift;
 142 }
 143
 144 static void transform_decl(struct tgsi_transform_context *ctx,
 145                            struct tgsi_full_declaration *decl)
 146 {
 147     struct vs_transform_context *vsctx = (struct vs_transform_context *)ctx;
 148     unsigned i;
 149
 150     if (decl->Declaration.File == TGSI_FILE_OUTPUT) {
 151         switch (decl->Semantic.Name) {
 152             case TGSI_SEMANTIC_POSITION:
 153                 vsctx->pos_output = decl->Range.First;
 154                 break;
 155
 156             case TGSI_SEMANTIC_COLOR:
 157                 assert(decl->Semantic.Index < 2);
 158
 159                 /* We must rasterize the first color if the second one is
 160                  * used, otherwise the rasterizer doesn't do the color
 161                  * selection correctly. Declare it, but don't write to it. */
 162                 if (decl->Semantic.Index == 1 && !vsctx->color_used[0]) {
 163                     insert_output_before(ctx, decl, TGSI_SEMANTIC_COLOR, 0,
 164                                          TGSI_INTERPOLATE_LINEAR);
 165                     vsctx->color_used[0] = TRUE;
 166                 }
 167                 break;
 168
 169             case TGSI_SEMANTIC_BCOLOR:
 170                 assert(decl->Semantic.Index < 2);
 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_before(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_before(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_before(ctx, decl, TGSI_SEMANTIC_BCOLOR, 0,
 187                                          TGSI_INTERPOLATE_LINEAR);
 188                     vsctx->bcolor_used[0] = TRUE;
 189                 }
 190                 break;
 191
 192             case TGSI_SEMANTIC_GENERIC:
 193                 vsctx->last_generic = MAX2(vsctx->last_generic, decl->Semantic.Index);
 194                 break;
 195         }
 196
 197         /* Since we're inserting new outputs in between, the following outputs
 198          * should be moved to the right so that they don't overlap with
 199          * the newly added ones. */
 200         decl->Range.First += vsctx->decl_shift;
 201         decl->Range.Last += vsctx->decl_shift;
 202
 203         ++vsctx->num_outputs;
 204     } else if (decl->Declaration.File == TGSI_FILE_TEMPORARY) {
 205         for (i = decl->Range.First; i <= decl->Range.Last; i++) {
 206            vsctx->temp_used[i] = TRUE;
 207         }
 208     }
 209
 210     ctx->emit_declaration(ctx, decl);
 211
 212     /* Insert BCOLOR1 if needed. */
 213     if (decl->Declaration.File == TGSI_FILE_OUTPUT &&
 214         decl->Semantic.Name == TGSI_SEMANTIC_BCOLOR &&
 215         !vsctx->bcolor_used[1]) {
 216         insert_output_after(ctx, decl, TGSI_SEMANTIC_BCOLOR, 1,
 217                             TGSI_INTERPOLATE_LINEAR);
 218     }
 219 }
 220
 221 static void transform_inst(struct tgsi_transform_context *ctx,
 222                            struct tgsi_full_instruction *inst)
 223 {
 224     struct vs_transform_context *vsctx = (struct vs_transform_context *) ctx;
 225     struct tgsi_full_instruction new_inst;
 226     unsigned i;
 227
 228     if (!vsctx->first_instruction) {
 229         vsctx->first_instruction = TRUE;
 230
 231         /* Insert the generic output for WPOS. */
 232         emit_output(ctx, TGSI_SEMANTIC_GENERIC, vsctx->last_generic + 1,
 233                     TGSI_INTERPOLATE_PERSPECTIVE, vsctx->num_outputs);
 234
 235         /* Find a free temp for POSITION. */
 236         for (i = 0; i < Elements(vsctx->temp_used); i++) {
 237             if (!vsctx->temp_used[i]) {
 238                 emit_temp(ctx, i);
 239                 vsctx->pos_temp = i;
 240                 break;
 241             }
 242         }
 243     }
 244
 245     if (inst->Instruction.Opcode == TGSI_OPCODE_END) {
 246         /* MOV OUT[pos_output], TEMP[pos_temp]; */
 247         new_inst = tgsi_default_full_instruction();
 248         new_inst.Instruction.Opcode = TGSI_OPCODE_MOV;
 249         new_inst.Instruction.NumDstRegs = 1;
 250         new_inst.Dst[0].Register.File = TGSI_FILE_OUTPUT;
 251         new_inst.Dst[0].Register.Index = vsctx->pos_output;
 252         new_inst.Dst[0].Register.WriteMask = TGSI_WRITEMASK_XYZW;
 253         new_inst.Instruction.NumSrcRegs = 1;
 254         new_inst.Src[0].Register.File = TGSI_FILE_TEMPORARY;
 255         new_inst.Src[0].Register.Index = vsctx->pos_temp;
 256         ctx->emit_instruction(ctx, &new_inst);
 257
 258         /* MOV OUT[n-1], TEMP[pos_temp]; */
 259         new_inst = tgsi_default_full_instruction();
 260         new_inst.Instruction.Opcode = TGSI_OPCODE_MOV;
 261         new_inst.Instruction.NumDstRegs = 1;
 262         new_inst.Dst[0].Register.File = TGSI_FILE_OUTPUT;
 263         new_inst.Dst[0].Register.Index = vsctx->num_outputs - 1;
 264         new_inst.Dst[0].Register.WriteMask = TGSI_WRITEMASK_XYZW;
 265         new_inst.Instruction.NumSrcRegs = 1;
 266         new_inst.Src[0].Register.File = TGSI_FILE_TEMPORARY;
 267         new_inst.Src[0].Register.Index = vsctx->pos_temp;
 268         ctx->emit_instruction(ctx, &new_inst);
 269
 270         vsctx->end_instruction = TRUE;
 271     } else {
 272         /* Not an END instruction. */
 273         /* Fix writes to outputs. */
 274         for (i = 0; i < inst->Instruction.NumDstRegs; i++) {
 275             struct tgsi_full_dst_register *dst = &inst->Dst[i];
 276             if (dst->Register.File == TGSI_FILE_OUTPUT) {
 277                 if (dst->Register.Index == vsctx->pos_output) {
 278                     /* Replace writes to OUT[pos_output] with TEMP[pos_temp]. */
 279                     dst->Register.File = TGSI_FILE_TEMPORARY;
 280                     dst->Register.Index = vsctx->pos_temp;
 281                 } else {
 282                     /* Not a position, good...
 283                      * Since we were changing the indices of output decls,
 284                      * we must redirect writes into them too. */
 285                     dst->Register.Index = vsctx->out_remap[dst->Register.Index];
 286                 }
 287             }
 288         }
 289
 290         /* Inserting 2 instructions before the END opcode moves all following
 291          * labels by 2. Subroutines are always after the END opcode so
 292          * they're always moved. */
 293         if (inst->Instruction.Opcode == TGSI_OPCODE_CAL) {
 294             inst->Label.Label += 2;
 295         }
 296         /* The labels of the following opcodes are moved only after
 297          * the END opcode. */
 298         if (vsctx->end_instruction &&
 299             (inst->Instruction.Opcode == TGSI_OPCODE_IF ||
 300              inst->Instruction.Opcode == TGSI_OPCODE_ELSE ||
 301              inst->Instruction.Opcode == TGSI_OPCODE_BGNLOOP ||
 302              inst->Instruction.Opcode == TGSI_OPCODE_ENDLOOP)) {
 303             inst->Label.Label += 2;
 304         }
 305     }
 306
 307     ctx->emit_instruction(ctx, inst);
 308 }
 309
 310 void r300_draw_init_vertex_shader(struct r300_context *r300,
 311                                   struct r300_vertex_shader *vs)
 312 {
 313     struct draw_context *draw = r300->draw;
 314     struct pipe_shader_state new_vs;
 315     struct tgsi_shader_info info;
 316     struct vs_transform_context transform;
 317     const uint newLen = tgsi_num_tokens(vs->state.tokens) + 100 /* XXX */;
 318     unsigned i;
 319
 320     tgsi_scan_shader(vs->state.tokens, &info);
 321
 322     new_vs.tokens = tgsi_alloc_tokens(newLen);
 323     if (new_vs.tokens == NULL)
 324         return;
 325
 326     memset(&transform, 0, sizeof(transform));
 327     for (i = 0; i < Elements(transform.out_remap); i++) {
 328         transform.out_remap[i] = i;
 329     }
 330     transform.last_generic = -1;
 331     transform.base.transform_instruction = transform_inst;
 332     transform.base.transform_declaration = transform_decl;
 333
 334     for (i = 0; i < info.num_outputs; i++) {
 335         unsigned index = info.output_semantic_index[i];
 336
 337         switch (info.output_semantic_name[i]) {
 338             case TGSI_SEMANTIC_COLOR:
 339                 assert(index < 2);
 340                 transform.color_used[index] = TRUE;
 341                 break;
 342
 343             case TGSI_SEMANTIC_BCOLOR:
 344                 assert(index < 2);
 345                 transform.bcolor_used[index] = TRUE;
 346                 break;
 347         }
 348     }
 349
 350     tgsi_transform_shader(vs->state.tokens,
 351                           (struct tgsi_token*)new_vs.tokens,
 352                           newLen, &transform.base);
 353
 354 #if 0
 355     printf("----------------------------------------------\norig shader:\n");
 356     tgsi_dump(vs->state.tokens, 0);
 357     printf("----------------------------------------------\nnew shader:\n");
 358     tgsi_dump(new_vs.tokens, 0);
 359     printf("----------------------------------------------\n");
 360 #endif
 361
 362     /* Free old tokens. */
 363     FREE((void*)vs->state.tokens);
 364
 365     vs->draw_vs = draw_create_vertex_shader(draw, &new_vs);
 366
 367     /* Instead of duplicating and freeing the tokens, copy the pointer directly. */
 368     vs->state.tokens = new_vs.tokens;
 369
 370     /* Init the VS output table for the rasterizer. */
 371     r300_init_vs_outputs(r300, vs);
 372
 373     /* Make the last generic be WPOS. */
 374     vs->outputs.wpos = vs->outputs.generic[transform.last_generic + 1];
 375     vs->outputs.generic[transform.last_generic + 1] = ATTR_UNUSED;
 376 }