src/gallium/drivers/r300/r300_tgsi_to_rc.c

   1 /*
   2  * Copyright 2009 Nicolai Hähnle <nhaehnle@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 COPYRIGHT HOLDER(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 <stdio.h>
  24 #include "r300_tgsi_to_rc.h"
  25
  26 #include "compiler/radeon_compiler.h"
  27
  28 #include "tgsi/tgsi_info.h"
  29 #include "tgsi/tgsi_parse.h"
  30 #include "tgsi/tgsi_scan.h"
  31 #include "tgsi/tgsi_util.h"
  32
  33 static unsigned translate_opcode(unsigned opcode)
  34 {
  35     switch(opcode) {
  36         case TGSI_OPCODE_ARL: return RC_OPCODE_ARL;
  37         case TGSI_OPCODE_MOV: return RC_OPCODE_MOV;
  38         case TGSI_OPCODE_LIT: return RC_OPCODE_LIT;
  39         case TGSI_OPCODE_RCP: return RC_OPCODE_RCP;
  40         case TGSI_OPCODE_RSQ: return RC_OPCODE_RSQ;
  41         case TGSI_OPCODE_EXP: return RC_OPCODE_EXP;
  42         case TGSI_OPCODE_LOG: return RC_OPCODE_LOG;
  43         case TGSI_OPCODE_MUL: return RC_OPCODE_MUL;
  44         case TGSI_OPCODE_ADD: return RC_OPCODE_ADD;
  45         case TGSI_OPCODE_DP3: return RC_OPCODE_DP3;
  46         case TGSI_OPCODE_DP4: return RC_OPCODE_DP4;
  47         case TGSI_OPCODE_DST: return RC_OPCODE_DST;
  48         case TGSI_OPCODE_MIN: return RC_OPCODE_MIN;
  49         case TGSI_OPCODE_MAX: return RC_OPCODE_MAX;
  50         case TGSI_OPCODE_SLT: return RC_OPCODE_SLT;
  51         case TGSI_OPCODE_SGE: return RC_OPCODE_SGE;
  52         case TGSI_OPCODE_MAD: return RC_OPCODE_MAD;
  53         case TGSI_OPCODE_LRP: return RC_OPCODE_LRP;
  54         case TGSI_OPCODE_FRC: return RC_OPCODE_FRC;
  55         case TGSI_OPCODE_FLR: return RC_OPCODE_FLR;
  56         case TGSI_OPCODE_ROUND: return RC_OPCODE_ROUND;
  57         case TGSI_OPCODE_EX2: return RC_OPCODE_EX2;
  58         case TGSI_OPCODE_LG2: return RC_OPCODE_LG2;
  59         case TGSI_OPCODE_POW: return RC_OPCODE_POW;
  60         case TGSI_OPCODE_COS: return RC_OPCODE_COS;
  61         case TGSI_OPCODE_DDX: return RC_OPCODE_DDX;
  62         case TGSI_OPCODE_DDY: return RC_OPCODE_DDY;
  63         case TGSI_OPCODE_KILL: return RC_OPCODE_KILP;
  64      /* case TGSI_OPCODE_PK2H: return RC_OPCODE_PK2H; */
  65      /* case TGSI_OPCODE_PK2US: return RC_OPCODE_PK2US; */
  66      /* case TGSI_OPCODE_PK4B: return RC_OPCODE_PK4B; */
  67      /* case TGSI_OPCODE_PK4UB: return RC_OPCODE_PK4UB; */
  68         case TGSI_OPCODE_SEQ: return RC_OPCODE_SEQ;
  69         case TGSI_OPCODE_SGT: return RC_OPCODE_SGT;
  70         case TGSI_OPCODE_SIN: return RC_OPCODE_SIN;
  71         case TGSI_OPCODE_SLE: return RC_OPCODE_SLE;
  72         case TGSI_OPCODE_SNE: return RC_OPCODE_SNE;
  73         case TGSI_OPCODE_TEX: return RC_OPCODE_TEX;
  74         case TGSI_OPCODE_TXD: return RC_OPCODE_TXD;
  75         case TGSI_OPCODE_TXP: return RC_OPCODE_TXP;
  76      /* case TGSI_OPCODE_UP2H: return RC_OPCODE_UP2H; */
  77      /* case TGSI_OPCODE_UP2US: return RC_OPCODE_UP2US; */
  78      /* case TGSI_OPCODE_UP4B: return RC_OPCODE_UP4B; */
  79      /* case TGSI_OPCODE_UP4UB: return RC_OPCODE_UP4UB; */
  80         case TGSI_OPCODE_ARR: return RC_OPCODE_ARR;
  81      /* case TGSI_OPCODE_CAL: return RC_OPCODE_CAL; */
  82      /* case TGSI_OPCODE_RET: return RC_OPCODE_RET; */
  83         case TGSI_OPCODE_SSG: return RC_OPCODE_SSG;
  84         case TGSI_OPCODE_CMP: return RC_OPCODE_CMP;
  85         case TGSI_OPCODE_TXB: return RC_OPCODE_TXB;
  86      /* case TGSI_OPCODE_DIV: return RC_OPCODE_DIV; */
  87         case TGSI_OPCODE_DP2: return RC_OPCODE_DP2;
  88         case TGSI_OPCODE_TXL: return RC_OPCODE_TXL;
  89         case TGSI_OPCODE_BRK: return RC_OPCODE_BRK;
  90         case TGSI_OPCODE_IF: return RC_OPCODE_IF;
  91         case TGSI_OPCODE_BGNLOOP: return RC_OPCODE_BGNLOOP;
  92         case TGSI_OPCODE_ELSE: return RC_OPCODE_ELSE;
  93         case TGSI_OPCODE_ENDIF: return RC_OPCODE_ENDIF;
  94         case TGSI_OPCODE_ENDLOOP: return RC_OPCODE_ENDLOOP;
  95      /* case TGSI_OPCODE_PUSHA: return RC_OPCODE_PUSHA; */
  96      /* case TGSI_OPCODE_POPA: return RC_OPCODE_POPA; */
  97         case TGSI_OPCODE_CEIL: return RC_OPCODE_CEIL;
  98      /* case TGSI_OPCODE_I2F: return RC_OPCODE_I2F; */
  99      /* case TGSI_OPCODE_NOT: return RC_OPCODE_NOT; */
 100         case TGSI_OPCODE_TRUNC: return RC_OPCODE_TRUNC;
 101      /* case TGSI_OPCODE_SHL: return RC_OPCODE_SHL; */
 102      /* case TGSI_OPCODE_ISHR: return RC_OPCODE_SHR; */
 103      /* case TGSI_OPCODE_AND: return RC_OPCODE_AND; */
 104      /* case TGSI_OPCODE_OR: return RC_OPCODE_OR; */
 105      /* case TGSI_OPCODE_MOD: return RC_OPCODE_MOD; */
 106      /* case TGSI_OPCODE_XOR: return RC_OPCODE_XOR; */
 107      /* case TGSI_OPCODE_TXF: return RC_OPCODE_TXF; */
 108      /* case TGSI_OPCODE_TXQ: return RC_OPCODE_TXQ; */
 109         case TGSI_OPCODE_CONT: return RC_OPCODE_CONT;
 110      /* case TGSI_OPCODE_EMIT: return RC_OPCODE_EMIT; */
 111      /* case TGSI_OPCODE_ENDPRIM: return RC_OPCODE_ENDPRIM; */
 112      /* case TGSI_OPCODE_BGNLOOP2: return RC_OPCODE_BGNLOOP2; */
 113      /* case TGSI_OPCODE_BGNSUB: return RC_OPCODE_BGNSUB; */
 114      /* case TGSI_OPCODE_ENDLOOP2: return RC_OPCODE_ENDLOOP2; */
 115      /* case TGSI_OPCODE_ENDSUB: return RC_OPCODE_ENDSUB; */
 116         case TGSI_OPCODE_NOP: return RC_OPCODE_NOP;
 117         case TGSI_OPCODE_KILL_IF: return RC_OPCODE_KIL;
 118     }
 119
 120     fprintf(stderr, "r300: Unknown TGSI/RC opcode: %s\n", tgsi_get_opcode_name(opcode));
 121     return RC_OPCODE_ILLEGAL_OPCODE;
 122 }
 123
 124 static unsigned translate_saturate(unsigned saturate)
 125 {
 126     return saturate ? RC_SATURATE_ZERO_ONE : RC_SATURATE_NONE;
 127 }
 128
 129 static unsigned translate_register_file(unsigned file)
 130 {
 131     switch(file) {
 132         case TGSI_FILE_CONSTANT: return RC_FILE_CONSTANT;
 133         case TGSI_FILE_IMMEDIATE: return RC_FILE_CONSTANT;
 134         case TGSI_FILE_INPUT: return RC_FILE_INPUT;
 135         case TGSI_FILE_OUTPUT: return RC_FILE_OUTPUT;
 136         default:
 137             fprintf(stderr, "Unhandled register file: %i\n", file);
 138             /* fall-through */
 139         case TGSI_FILE_TEMPORARY: return RC_FILE_TEMPORARY;
 140         case TGSI_FILE_ADDRESS: return RC_FILE_ADDRESS;
 141     }
 142 }
 143
 144 static int translate_register_index(
 145     struct tgsi_to_rc * ttr,
 146     unsigned file,
 147     int index)
 148 {
 149     if (file == TGSI_FILE_IMMEDIATE)
 150         return ttr->immediate_offset + index;
 151
 152     return index;
 153 }
 154
 155 static void transform_dstreg(
 156     struct tgsi_to_rc * ttr,
 157     struct rc_dst_register * dst,
 158     struct tgsi_full_dst_register * src)
 159 {
 160     dst->File = translate_register_file(src->Register.File);
 161     dst->Index = translate_register_index(ttr, src->Register.File, src->Register.Index);
 162     dst->WriteMask = src->Register.WriteMask;
 163
 164     if (src->Register.Indirect) {
 165         ttr->error = TRUE;
 166         fprintf(stderr, "r300: Relative addressing of destination operands "
 167                 "is unsupported.\n");
 168     }
 169 }
 170
 171 static void transform_srcreg(
 172     struct tgsi_to_rc * ttr,
 173     struct rc_src_register * dst,
 174     struct tgsi_full_src_register * src)
 175 {
 176     unsigned i, j;
 177
 178     dst->File = translate_register_file(src->Register.File);
 179     dst->Index = translate_register_index(ttr, src->Register.File, src->Register.Index);
 180     dst->RelAddr = src->Register.Indirect;
 181     dst->Swizzle = tgsi_util_get_full_src_register_swizzle(src, 0);
 182     dst->Swizzle |= tgsi_util_get_full_src_register_swizzle(src, 1) << 3;
 183     dst->Swizzle |= tgsi_util_get_full_src_register_swizzle(src, 2) << 6;
 184     dst->Swizzle |= tgsi_util_get_full_src_register_swizzle(src, 3) << 9;
 185     dst->Abs = src->Register.Absolute;
 186     dst->Negate = src->Register.Negate ? RC_MASK_XYZW : 0;
 187
 188     if (src->Register.File == TGSI_FILE_IMMEDIATE) {
 189         for (i = 0; i < ttr->imms_to_swizzle_count; i++) {
 190             if (ttr->imms_to_swizzle[i].index == src->Register.Index) {
 191                 dst->File = RC_FILE_TEMPORARY;
 192                 dst->Index = 0;
 193                 dst->Swizzle = 0;
 194                 for (j = 0; j < 4; j++) {
 195                     dst->Swizzle |= GET_SWZ(ttr->imms_to_swizzle[i].swizzle,
 196                         tgsi_util_get_full_src_register_swizzle(src, j)) << (j * 3);
 197                 }
 198                 break;
 199             }
 200         }
 201     }
 202 }
 203
 204 static void transform_texture(struct rc_instruction * dst, struct tgsi_instruction_texture src,
 205                               uint32_t *shadowSamplers)
 206 {
 207     switch(src.Texture) {
 208         case TGSI_TEXTURE_1D:
 209             dst->U.I.TexSrcTarget = RC_TEXTURE_1D;
 210             break;
 211         case TGSI_TEXTURE_2D:
 212             dst->U.I.TexSrcTarget = RC_TEXTURE_2D;
 213             break;
 214         case TGSI_TEXTURE_3D:
 215             dst->U.I.TexSrcTarget = RC_TEXTURE_3D;
 216             break;
 217         case TGSI_TEXTURE_CUBE:
 218             dst->U.I.TexSrcTarget = RC_TEXTURE_CUBE;
 219             break;
 220         case TGSI_TEXTURE_RECT:
 221             dst->U.I.TexSrcTarget = RC_TEXTURE_RECT;
 222             break;
 223         case TGSI_TEXTURE_SHADOW1D:
 224             dst->U.I.TexSrcTarget = RC_TEXTURE_1D;
 225             dst->U.I.TexShadow = 1;
 226             *shadowSamplers |= 1 << dst->U.I.TexSrcUnit;
 227             break;
 228         case TGSI_TEXTURE_SHADOW2D:
 229             dst->U.I.TexSrcTarget = RC_TEXTURE_2D;
 230             dst->U.I.TexShadow = 1;
 231             *shadowSamplers |= 1 << dst->U.I.TexSrcUnit;
 232             break;
 233         case TGSI_TEXTURE_SHADOWRECT:
 234             dst->U.I.TexSrcTarget = RC_TEXTURE_RECT;
 235             dst->U.I.TexShadow = 1;
 236             *shadowSamplers |= 1 << dst->U.I.TexSrcUnit;
 237             break;
 238     }
 239     dst->U.I.TexSwizzle = RC_SWIZZLE_XYZW;
 240 }
 241
 242 static void transform_instruction(struct tgsi_to_rc * ttr, struct tgsi_full_instruction * src)
 243 {
 244     struct rc_instruction * dst;
 245     int i;
 246
 247     dst = rc_insert_new_instruction(ttr->compiler, ttr->compiler->Program.Instructions.Prev);
 248     dst->U.I.Opcode = translate_opcode(src->Instruction.Opcode);
 249     dst->U.I.SaturateMode = translate_saturate(src->Instruction.Saturate);
 250
 251     if (src->Instruction.NumDstRegs)
 252         transform_dstreg(ttr, &dst->U.I.DstReg, &src->Dst[0]);
 253
 254     for(i = 0; i < src->Instruction.NumSrcRegs; ++i) {
 255         if (src->Src[i].Register.File == TGSI_FILE_SAMPLER)
 256             dst->U.I.TexSrcUnit = src->Src[i].Register.Index;
 257         else
 258             transform_srcreg(ttr, &dst->U.I.SrcReg[i], &src->Src[i]);
 259     }
 260
 261     /* Texturing. */
 262     if (src->Instruction.Texture)
 263         transform_texture(dst, src->Texture,
 264                           &ttr->compiler->Program.ShadowSamplers);
 265 }
 266
 267 static void handle_immediate(struct tgsi_to_rc * ttr,
 268                              struct tgsi_full_immediate * imm,
 269                              unsigned index)
 270 {
 271     struct rc_constant constant;
 272     unsigned swizzle = 0;
 273     boolean can_swizzle = TRUE;
 274     unsigned i;
 275
 276     for (i = 0; i < 4; i++) {
 277         if (imm->u[i].Float == 0.0f) {
 278             swizzle |= RC_SWIZZLE_ZERO << (i * 3);
 279         } else if (imm->u[i].Float == 0.5f && ttr->use_half_swizzles) {
 280             swizzle |= RC_SWIZZLE_HALF << (i * 3);
 281         } else if (imm->u[i].Float == 1.0f) {
 282             swizzle |= RC_SWIZZLE_ONE << (i * 3);
 283         } else {
 284             can_swizzle = FALSE;
 285             break;
 286         }
 287     }
 288
 289     if (can_swizzle) {
 290         ttr->imms_to_swizzle[ttr->imms_to_swizzle_count].index = index;
 291         ttr->imms_to_swizzle[ttr->imms_to_swizzle_count].swizzle = swizzle;
 292         ttr->imms_to_swizzle_count++;
 293     } else {
 294         constant.Type = RC_CONSTANT_IMMEDIATE;
 295         constant.Size = 4;
 296         for(i = 0; i < 4; ++i)
 297             constant.u.Immediate[i] = imm->u[i].Float;
 298         rc_constants_add(&ttr->compiler->Program.Constants, &constant);
 299     }
 300 }
 301
 302 void r300_tgsi_to_rc(struct tgsi_to_rc * ttr,
 303                      const struct tgsi_token * tokens)
 304 {
 305     struct tgsi_full_instruction *inst;
 306     struct tgsi_parse_context parser;
 307     unsigned imm_index = 0;
 308     int i;
 309
 310     ttr->error = FALSE;
 311
 312     /* Allocate constants placeholders.
 313      *
 314      * Note: What if declared constants are not contiguous? */
 315     for(i = 0; i <= ttr->info->file_max[TGSI_FILE_CONSTANT]; ++i) {
 316         struct rc_constant constant;
 317         memset(&constant, 0, sizeof(constant));
 318         constant.Type = RC_CONSTANT_EXTERNAL;
 319         constant.Size = 4;
 320         constant.u.External = i;
 321         rc_constants_add(&ttr->compiler->Program.Constants, &constant);
 322     }
 323
 324     ttr->immediate_offset = ttr->compiler->Program.Constants.Count;
 325
 326     ttr->imms_to_swizzle = malloc(ttr->info->immediate_count * sizeof(struct swizzled_imms));
 327     ttr->imms_to_swizzle_count = 0;
 328
 329     tgsi_parse_init(&parser, tokens);
 330
 331     while (!tgsi_parse_end_of_tokens(&parser)) {
 332         tgsi_parse_token(&parser);
 333
 334         switch (parser.FullToken.Token.Type) {
 335             case TGSI_TOKEN_TYPE_DECLARATION:
 336                 break;
 337             case TGSI_TOKEN_TYPE_IMMEDIATE:
 338                 handle_immediate(ttr, &parser.FullToken.FullImmediate, imm_index);
 339                 imm_index++;
 340                 break;
 341             case TGSI_TOKEN_TYPE_INSTRUCTION:
 342                 inst = &parser.FullToken.FullInstruction;
 343                 if (inst->Instruction.Opcode == TGSI_OPCODE_END) {
 344                     break;
 345                 }
 346
 347                 transform_instruction(ttr, inst);
 348                 break;
 349         }
 350     }
 351
 352     tgsi_parse_free(&parser);
 353
 354     free(ttr->imms_to_swizzle);
 355
 356     rc_calculate_inputs_outputs(ttr->compiler);
 357 }