src/gallium/auxiliary/gallivm/lp_bld_tgsi.c

   1 /**************************************************************************
   2  *
   3  * Copyright 2011-2012 Advanced Micro Devices, Inc.
   4  * Copyright 2010 VMware, Inc.
   5  * Copyright 2009 VMware, Inc.
   6  * Copyright 2007-2008 VMware, Inc.
   7  * All Rights Reserved.
   8  *
   9  * Permission is hereby granted, free of charge, to any person obtaining a
  10  * copy of this software and associated documentation files (the
  11  * "Software"), to deal in the Software without restriction, including
  12  * without limitation the rights to use, copy, modify, merge, publish,
  13  * distribute, sub license, and/or sell copies of the Software, and to
  14  * permit persons to whom the Software is furnished to do so, subject to
  15  * the following conditions:
  16  *
  17  * The above copyright notice and this permission notice (including the
  18  * next paragraph) shall be included in all copies or substantial portions
  19  * of the Software.
  20  *
  21  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
  22  * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  23  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
  24  * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
  25  * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
  26  * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
  27  * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  28  *
  29  **************************************************************************/
  30
  31 #include "gallivm/lp_bld_tgsi.h"
  32
  33 #include "gallivm/lp_bld_arit.h"
  34 #include "gallivm/lp_bld_gather.h"
  35 #include "gallivm/lp_bld_init.h"
  36 #include "gallivm/lp_bld_intr.h"
  37 #include "tgsi/tgsi_info.h"
  38 #include "tgsi/tgsi_parse.h"
  39 #include "tgsi/tgsi_util.h"
  40 #include "util/u_memory.h"
  41
  42 /* The user is responsible for freeing list->instructions */
  43 unsigned lp_bld_tgsi_list_init(struct lp_build_tgsi_context * bld_base)
  44 {
  45    bld_base->instructions = (struct tgsi_full_instruction *)
  46          MALLOC( LP_MAX_INSTRUCTIONS * sizeof(struct tgsi_full_instruction) );
  47    if (!bld_base->instructions) {
  48       return 0;
  49    }
  50    bld_base->max_instructions = LP_MAX_INSTRUCTIONS;
  51    return 1;
  52 }
  53
  54
  55 unsigned lp_bld_tgsi_add_instruction(
  56    struct lp_build_tgsi_context * bld_base,
  57    const struct tgsi_full_instruction *inst_to_add)
  58 {
  59
  60    if (bld_base->num_instructions == bld_base->max_instructions) {
  61       struct tgsi_full_instruction *instructions;
  62       instructions = REALLOC(bld_base->instructions, bld_base->max_instructions
  63                                       * sizeof(struct tgsi_full_instruction),
  64                                       (bld_base->max_instructions + LP_MAX_INSTRUCTIONS)
  65                                       * sizeof(struct tgsi_full_instruction));
  66       if (!instructions) {
  67          return 0;
  68       }
  69       bld_base->instructions = instructions;
  70       bld_base->max_instructions += LP_MAX_INSTRUCTIONS;
  71    }
  72    memcpy(bld_base->instructions + bld_base->num_instructions, inst_to_add,
  73           sizeof(bld_base->instructions[0]));
  74
  75    bld_base->num_instructions++;
  76
  77    return 1;
  78 }
  79
  80
  81 /**
  82  * This function assumes that all the args in emit_data have been set.
  83  */
  84 static void
  85 lp_build_action_set_dst_type(
  86    struct lp_build_emit_data * emit_data,
  87    struct lp_build_tgsi_context *bld_base,
  88    unsigned tgsi_opcode)
  89 {
  90    if (emit_data->arg_count == 0) {
  91       emit_data->dst_type = LLVMVoidTypeInContext(bld_base->base.gallivm->context);
  92    } else {
  93       /* XXX: Not all opcodes have the same src and dst types. */
  94       emit_data->dst_type = LLVMTypeOf(emit_data->args[0]);
  95    }
  96 }
  97
  98 void
  99 lp_build_tgsi_intrinsic(
 100  const struct lp_build_tgsi_action * action,
 101  struct lp_build_tgsi_context * bld_base,
 102  struct lp_build_emit_data * emit_data)
 103 {
 104    struct lp_build_context * base = &bld_base->base;
 105    emit_data->output[emit_data->chan] = lp_build_intrinsic(
 106                base->gallivm->builder, action->intr_name,
 107                emit_data->dst_type, emit_data->args, emit_data->arg_count, 0);
 108 }
 109
 110 LLVMValueRef
 111 lp_build_emit_llvm(
 112    struct lp_build_tgsi_context *bld_base,
 113    unsigned tgsi_opcode,
 114    struct lp_build_emit_data * emit_data)
 115 {
 116    struct lp_build_tgsi_action * action = &bld_base->op_actions[tgsi_opcode];
 117    /* XXX: Assert that this is a componentwise or replicate instruction */
 118
 119    lp_build_action_set_dst_type(emit_data, bld_base, tgsi_opcode);
 120    emit_data->chan = 0;
 121    assert(action->emit);
 122    action->emit(action, bld_base, emit_data);
 123    return emit_data->output[0];
 124 }
 125
 126 LLVMValueRef
 127 lp_build_emit_llvm_unary(
 128    struct lp_build_tgsi_context *bld_base,
 129    unsigned tgsi_opcode,
 130    LLVMValueRef arg0)
 131 {
 132    struct lp_build_emit_data emit_data;
 133    emit_data.arg_count = 1;
 134    emit_data.args[0] = arg0;
 135    return lp_build_emit_llvm(bld_base, tgsi_opcode, &emit_data);
 136 }
 137
 138 LLVMValueRef
 139 lp_build_emit_llvm_binary(
 140    struct lp_build_tgsi_context *bld_base,
 141    unsigned tgsi_opcode,
 142    LLVMValueRef arg0,
 143    LLVMValueRef arg1)
 144 {
 145    struct lp_build_emit_data emit_data;
 146    emit_data.arg_count = 2;
 147    emit_data.args[0] = arg0;
 148    emit_data.args[1] = arg1;
 149    return lp_build_emit_llvm(bld_base, tgsi_opcode, &emit_data);
 150 }
 151
 152 LLVMValueRef
 153 lp_build_emit_llvm_ternary(
 154    struct lp_build_tgsi_context *bld_base,
 155    unsigned tgsi_opcode,
 156    LLVMValueRef arg0,
 157    LLVMValueRef arg1,
 158    LLVMValueRef arg2)
 159 {
 160    struct lp_build_emit_data emit_data;
 161    emit_data.arg_count = 3;
 162    emit_data.args[0] = arg0;
 163    emit_data.args[1] = arg1;
 164    emit_data.args[2] = arg2;
 165    return lp_build_emit_llvm(bld_base, tgsi_opcode, &emit_data);
 166 }
 167
 168 /**
 169  * The default fetch implementation.
 170  */
 171 void lp_build_fetch_args(
 172    struct lp_build_tgsi_context * bld_base,
 173    struct lp_build_emit_data * emit_data)
 174 {
 175    unsigned src;
 176    for (src = 0; src < emit_data->info->num_src; src++) {
 177       emit_data->args[src] = lp_build_emit_fetch(bld_base, emit_data->inst, src,
 178                                                  emit_data->src_chan);
 179    }
 180    emit_data->arg_count = emit_data->info->num_src;
 181    lp_build_action_set_dst_type(emit_data, bld_base,
 182                 emit_data->inst->Instruction.Opcode);
 183 }
 184
 185 /**
 186  * with doubles src and dst channels aren't 1:1.
 187  * check the src/dst types for the opcode,
 188  * 1. if neither is double then src == dst;
 189  * 2. if dest is double
 190  *     - don't store to y or w
 191  *     - if src is double then src == dst.
 192  *     - else for f2d, d.xy = s.x
 193  *     - else for f2d, d.zw = s.y
 194  * 3. if dst is single, src is double
 195  *    - map dst x,z to src xy;
 196  *    - map dst y,w to src zw;
 197  */
 198 static int get_src_chan_idx(unsigned opcode,
 199                             int dst_chan_index)
 200 {
 201    enum tgsi_opcode_type dtype = tgsi_opcode_infer_dst_type(opcode);
 202    enum tgsi_opcode_type stype = tgsi_opcode_infer_src_type(opcode);
 203
 204    if (dtype != TGSI_TYPE_DOUBLE && stype != TGSI_TYPE_DOUBLE)
 205       return dst_chan_index;
 206    if (dtype == TGSI_TYPE_DOUBLE) {
 207       if (dst_chan_index == 1 || dst_chan_index == 3)
 208          return -1;
 209       if (stype == TGSI_TYPE_DOUBLE)
 210          return dst_chan_index;
 211       if (dst_chan_index == 0)
 212          return 0;
 213       if (dst_chan_index == 2)
 214          return 1;
 215    } else {
 216       if (dst_chan_index == 0 || dst_chan_index == 2)
 217          return 0;
 218       if (dst_chan_index == 1 || dst_chan_index == 3)
 219          return 2;
 220    }
 221    return -1;
 222 }
 223
 224 /* XXX: COMMENT
 225  * It should be assumed that this function ignores writemasks
 226  */
 227 boolean
 228 lp_build_tgsi_inst_llvm(
 229    struct lp_build_tgsi_context * bld_base,
 230    const struct tgsi_full_instruction * inst)
 231 {
 232    unsigned tgsi_opcode = inst->Instruction.Opcode;
 233    const struct tgsi_opcode_info * info = tgsi_get_opcode_info(tgsi_opcode);
 234    const struct lp_build_tgsi_action * action =
 235                                          &bld_base->op_actions[tgsi_opcode];
 236    struct lp_build_emit_data emit_data;
 237    unsigned chan_index;
 238    LLVMValueRef val;
 239    bld_base->pc++;
 240
 241    if (bld_base->emit_debug) {
 242       bld_base->emit_debug(bld_base, inst, info);
 243    }
 244
 245    /* Ignore deprecated instructions */
 246    switch (inst->Instruction.Opcode) {
 247
 248    case TGSI_OPCODE_UP2H:
 249    case TGSI_OPCODE_UP2US:
 250    case TGSI_OPCODE_UP4B:
 251    case TGSI_OPCODE_UP4UB:
 252    case TGSI_OPCODE_PUSHA:
 253    case TGSI_OPCODE_POPA:
 254    case TGSI_OPCODE_SAD:
 255       /* deprecated? */
 256       assert(0);
 257       return FALSE;
 258       break;
 259    }
 260
 261    /* Check if the opcode has been implemented */
 262    if (!action->emit) {
 263       return FALSE;
 264    }
 265
 266    memset(&emit_data, 0, sizeof(emit_data));
 267
 268    assert(info->num_dst <= 1);
 269    if (info->num_dst) {
 270       TGSI_FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
 271          emit_data.output[chan_index] = bld_base->base.undef;
 272       }
 273    }
 274
 275    emit_data.inst = inst;
 276    emit_data.info = info;
 277
 278    /* Emit the instructions */
 279    if (info->output_mode == TGSI_OUTPUT_COMPONENTWISE && bld_base->soa) {
 280       TGSI_FOR_EACH_DST0_ENABLED_CHANNEL(inst, chan_index) {
 281          int src_index = get_src_chan_idx(inst->Instruction.Opcode, chan_index);
 282          /* ignore channels 1/3 in double dst */
 283          if (src_index == -1)
 284             continue;
 285          emit_data.chan = chan_index;
 286          emit_data.src_chan = src_index;
 287          if (!action->fetch_args) {
 288             lp_build_fetch_args(bld_base, &emit_data);
 289          } else {
 290              action->fetch_args(bld_base, &emit_data);
 291          }
 292          action->emit(action, bld_base, &emit_data);
 293       }
 294    } else {
 295       emit_data.chan = LP_CHAN_ALL;
 296       if (action->fetch_args) {
 297          action->fetch_args(bld_base, &emit_data);
 298       }
 299       /* Make sure the output value is stored in emit_data.output[0], unless
 300        * the opcode is channel dependent */
 301       if (info->output_mode != TGSI_OUTPUT_CHAN_DEPENDENT) {
 302          emit_data.chan = 0;
 303       }
 304       action->emit(action, bld_base, &emit_data);
 305
 306       /* Replicate the output values */
 307       if (info->output_mode == TGSI_OUTPUT_REPLICATE && bld_base->soa) {
 308          val = emit_data.output[0];
 309          memset(emit_data.output, 0, sizeof(emit_data.output));
 310          TGSI_FOR_EACH_DST0_ENABLED_CHANNEL(inst, chan_index) {
 311             emit_data.output[chan_index] = val;
 312          }
 313       }
 314    }
 315
 316    if (info->num_dst > 0) {
 317       bld_base->emit_store(bld_base, inst, info, emit_data.output);
 318    }
 319    return TRUE;
 320 }
 321
 322
 323 LLVMValueRef
 324 lp_build_emit_fetch(
 325    struct lp_build_tgsi_context *bld_base,
 326    const struct tgsi_full_instruction *inst,
 327    unsigned src_op,
 328    const unsigned chan_index)
 329 {
 330    const struct tgsi_full_src_register *reg = &inst->Src[src_op];
 331    unsigned swizzle;
 332    LLVMValueRef res;
 333    enum tgsi_opcode_type stype = tgsi_opcode_infer_src_type(inst->Instruction.Opcode);
 334
 335    if (chan_index == LP_CHAN_ALL) {
 336       swizzle = ~0;
 337    } else {
 338       swizzle = tgsi_util_get_full_src_register_swizzle(reg, chan_index);
 339       if (swizzle > 3) {
 340          assert(0 && "invalid swizzle in emit_fetch()");
 341          return bld_base->base.undef;
 342       }
 343    }
 344
 345    assert(reg->Register.Index <= bld_base->info->file_max[reg->Register.File]);
 346
 347    if (bld_base->emit_fetch_funcs[reg->Register.File]) {
 348       res = bld_base->emit_fetch_funcs[reg->Register.File](bld_base, reg, stype,
 349                                                            swizzle);
 350    } else {
 351       assert(0 && "invalid src register in emit_fetch()");
 352       return bld_base->base.undef;
 353    }
 354
 355    if (reg->Register.Absolute) {
 356       switch (stype) {
 357       case TGSI_TYPE_FLOAT:
 358       case TGSI_TYPE_DOUBLE:
 359       case TGSI_TYPE_UNTYPED:
 360           /* modifiers on movs assume data is float */
 361          res = lp_build_emit_llvm_unary(bld_base, TGSI_OPCODE_ABS, res);
 362          break;
 363       case TGSI_TYPE_UNSIGNED:
 364       case TGSI_TYPE_SIGNED:
 365       case TGSI_TYPE_VOID:
 366       default:
 367          /* abs modifier is only legal on floating point types */
 368          assert(0);
 369          break;
 370       }
 371    }
 372
 373    if (reg->Register.Negate) {
 374       switch (stype) {
 375       case TGSI_TYPE_FLOAT:
 376       case TGSI_TYPE_UNTYPED:
 377          /* modifiers on movs assume data is float */
 378          res = lp_build_negate( &bld_base->base, res );
 379          break;
 380       case TGSI_TYPE_DOUBLE:
 381          /* no double build context */
 382          assert(0);
 383          break;
 384       case TGSI_TYPE_SIGNED:
 385       case TGSI_TYPE_UNSIGNED:
 386          res = lp_build_negate( &bld_base->int_bld, res );
 387          break;
 388       case TGSI_TYPE_VOID:
 389       default:
 390          assert(0);
 391          break;
 392       }
 393    }
 394
 395    /*
 396     * Swizzle the argument
 397     */
 398
 399    if (swizzle == ~0) {
 400       res = bld_base->emit_swizzle(bld_base, res,
 401                      reg->Register.SwizzleX,
 402                      reg->Register.SwizzleY,
 403                      reg->Register.SwizzleZ,
 404                      reg->Register.SwizzleW);
 405    }
 406
 407    return res;
 408
 409 }
 410
 411
 412 LLVMValueRef
 413 lp_build_emit_fetch_texoffset(
 414    struct lp_build_tgsi_context *bld_base,
 415    const struct tgsi_full_instruction *inst,
 416    unsigned tex_off_op,
 417    const unsigned chan_index)
 418 {
 419    const struct tgsi_texture_offset *off = &inst->TexOffsets[tex_off_op];
 420    struct tgsi_full_src_register reg;
 421    unsigned swizzle;
 422    LLVMValueRef res;
 423    enum tgsi_opcode_type stype = TGSI_TYPE_SIGNED;
 424
 425    /* convert offset "register" to ordinary register so can use normal emit funcs */
 426    memset(&reg, 0, sizeof(reg));
 427    reg.Register.File = off->File;
 428    reg.Register.Index = off->Index;
 429    reg.Register.SwizzleX = off->SwizzleX;
 430    reg.Register.SwizzleY = off->SwizzleY;
 431    reg.Register.SwizzleZ = off->SwizzleZ;
 432
 433    if (chan_index == LP_CHAN_ALL) {
 434       swizzle = ~0;
 435    } else {
 436       assert(chan_index < TGSI_SWIZZLE_W);
 437       swizzle = tgsi_util_get_src_register_swizzle(&reg.Register, chan_index);
 438    }
 439
 440    assert(off->Index <= bld_base->info->file_max[off->File]);
 441
 442    if (bld_base->emit_fetch_funcs[off->File]) {
 443       res = bld_base->emit_fetch_funcs[off->File](bld_base, &reg, stype,
 444                                                            swizzle);
 445    } else {
 446       assert(0 && "invalid src register in emit_fetch_texoffset()");
 447       return bld_base->base.undef;
 448    }
 449
 450    /*
 451     * Swizzle the argument
 452     */
 453
 454    if (swizzle == ~0) {
 455       res = bld_base->emit_swizzle(bld_base, res,
 456                                    off->SwizzleX,
 457                                    off->SwizzleY,
 458                                    off->SwizzleZ,
 459                                    /* there's no 4th channel */
 460                                    off->SwizzleX);
 461    }
 462
 463    return res;
 464
 465 }
 466
 467
 468 boolean
 469 lp_build_tgsi_llvm(
 470    struct lp_build_tgsi_context * bld_base,
 471    const struct tgsi_token *tokens)
 472 {
 473    struct tgsi_parse_context parse;
 474
 475    if (bld_base->emit_prologue) {
 476       bld_base->emit_prologue(bld_base);
 477    }
 478
 479    if (!lp_bld_tgsi_list_init(bld_base)) {
 480       return FALSE;
 481    }
 482
 483    tgsi_parse_init( &parse, tokens );
 484
 485    while( !tgsi_parse_end_of_tokens( &parse ) ) {
 486       tgsi_parse_token( &parse );
 487
 488       switch( parse.FullToken.Token.Type ) {
 489       case TGSI_TOKEN_TYPE_DECLARATION:
 490          /* Inputs already interpolated */
 491          bld_base->emit_declaration(bld_base, &parse.FullToken.FullDeclaration);
 492          break;
 493
 494       case TGSI_TOKEN_TYPE_INSTRUCTION:
 495          lp_bld_tgsi_add_instruction(bld_base, &parse.FullToken.FullInstruction);
 496          break;
 497
 498       case TGSI_TOKEN_TYPE_IMMEDIATE:
 499          bld_base->emit_immediate(bld_base, &parse.FullToken.FullImmediate);
 500          break;
 501
 502       case TGSI_TOKEN_TYPE_PROPERTY:
 503          break;
 504
 505       default:
 506          assert( 0 );
 507       }
 508    }
 509
 510    while (bld_base->pc != -1) {
 511       const struct tgsi_full_instruction *instr =
 512          bld_base->instructions + bld_base->pc;
 513       const struct tgsi_opcode_info *opcode_info =
 514          tgsi_get_opcode_info(instr->Instruction.Opcode);
 515       if (!lp_build_tgsi_inst_llvm(bld_base, instr)) {
 516          _debug_printf("warning: failed to translate tgsi opcode %s to LLVM\n",
 517                        opcode_info->mnemonic);
 518          return FALSE;
 519       }
 520    }
 521
 522    tgsi_parse_free(&parse);
 523
 524    FREE(bld_base->instructions);
 525
 526    if (bld_base->emit_epilogue) {
 527       bld_base->emit_epilogue(bld_base);
 528    }
 529
 530    return TRUE;
 531 }