src/gallium/drivers/radeonsi/si_shader_tgsi_alu.c

   1 /*
   2  * Copyright 2016 Advanced Micro Devices, Inc.
   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 AUTHOR(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
  24 #include "si_shader_internal.h"
  25 #include "gallivm/lp_bld_const.h"
  26 #include "gallivm/lp_bld_intr.h"
  27 #include "gallivm/lp_bld_gather.h"
  28 #include "tgsi/tgsi_parse.h"
  29 #include "amd/common/ac_llvm_build.h"
  30
  31 static void kill_if_fetch_args(struct lp_build_tgsi_context *bld_base,
  32                                struct lp_build_emit_data *emit_data)
  33 {
  34         const struct tgsi_full_instruction *inst = emit_data->inst;
  35         struct gallivm_state *gallivm = bld_base->base.gallivm;
  36         LLVMBuilderRef builder = gallivm->builder;
  37         unsigned i;
  38         LLVMValueRef conds[TGSI_NUM_CHANNELS];
  39
  40         for (i = 0; i < TGSI_NUM_CHANNELS; i++) {
  41                 LLVMValueRef value = lp_build_emit_fetch(bld_base, inst, 0, i);
  42                 conds[i] = LLVMBuildFCmp(builder, LLVMRealOLT, value,
  43                                         bld_base->base.zero, "");
  44         }
  45
  46         /* Or the conditions together */
  47         for (i = TGSI_NUM_CHANNELS - 1; i > 0; i--) {
  48                 conds[i - 1] = LLVMBuildOr(builder, conds[i], conds[i - 1], "");
  49         }
  50
  51         emit_data->dst_type = LLVMVoidTypeInContext(gallivm->context);
  52         emit_data->arg_count = 1;
  53         emit_data->args[0] = LLVMBuildSelect(builder, conds[0],
  54                                         lp_build_const_float(gallivm, -1.0f),
  55                                         bld_base->base.zero, "");
  56 }
  57
  58 static void kil_emit(const struct lp_build_tgsi_action *action,
  59                      struct lp_build_tgsi_context *bld_base,
  60                      struct lp_build_emit_data *emit_data)
  61 {
  62         struct si_shader_context *ctx = si_shader_context(bld_base);
  63         LLVMBuilderRef builder = ctx->gallivm.builder;
  64
  65         if (ctx->postponed_kill) {
  66                 if (emit_data->inst->Instruction.Opcode == TGSI_OPCODE_KILL_IF) {
  67                         LLVMValueRef val;
  68
  69                         /* Take the minimum kill value. This is the same as OR
  70                          * between 2 kill values. If the value is negative,
  71                          * the pixel will be killed.
  72                          */
  73                         val = LLVMBuildLoad(builder, ctx->postponed_kill, "");
  74                         val = lp_build_emit_llvm_binary(bld_base, TGSI_OPCODE_MIN,
  75                                                         val, emit_data->args[0]);
  76                         LLVMBuildStore(builder, val, ctx->postponed_kill);
  77                 } else {
  78                         LLVMBuildStore(builder,
  79                                        LLVMConstReal(ctx->f32, -1),
  80                                        ctx->postponed_kill);
  81                 }
  82                 return;
  83         }
  84
  85         if (emit_data->inst->Instruction.Opcode == TGSI_OPCODE_KILL_IF)
  86                 ac_build_kill(&ctx->ac, emit_data->args[0]);
  87         else
  88                 ac_build_kill(&ctx->ac, NULL);
  89 }
  90
  91 static void emit_icmp(const struct lp_build_tgsi_action *action,
  92                       struct lp_build_tgsi_context *bld_base,
  93                       struct lp_build_emit_data *emit_data)
  94 {
  95         unsigned pred;
  96         LLVMBuilderRef builder = bld_base->base.gallivm->builder;
  97         LLVMContextRef context = bld_base->base.gallivm->context;
  98
  99         switch (emit_data->inst->Instruction.Opcode) {
 100         case TGSI_OPCODE_USEQ:
 101         case TGSI_OPCODE_U64SEQ: pred = LLVMIntEQ; break;
 102         case TGSI_OPCODE_USNE:
 103         case TGSI_OPCODE_U64SNE: pred = LLVMIntNE; break;
 104         case TGSI_OPCODE_USGE:
 105         case TGSI_OPCODE_U64SGE: pred = LLVMIntUGE; break;
 106         case TGSI_OPCODE_USLT:
 107         case TGSI_OPCODE_U64SLT: pred = LLVMIntULT; break;
 108         case TGSI_OPCODE_ISGE:
 109         case TGSI_OPCODE_I64SGE: pred = LLVMIntSGE; break;
 110         case TGSI_OPCODE_ISLT:
 111         case TGSI_OPCODE_I64SLT: pred = LLVMIntSLT; break;
 112         default:
 113                 assert(!"unknown instruction");
 114                 pred = 0;
 115                 break;
 116         }
 117
 118         LLVMValueRef v = LLVMBuildICmp(builder, pred,
 119                         emit_data->args[0], emit_data->args[1],"");
 120
 121         v = LLVMBuildSExtOrBitCast(builder, v,
 122                         LLVMInt32TypeInContext(context), "");
 123
 124         emit_data->output[emit_data->chan] = v;
 125 }
 126
 127 static void emit_ucmp(const struct lp_build_tgsi_action *action,
 128                       struct lp_build_tgsi_context *bld_base,
 129                       struct lp_build_emit_data *emit_data)
 130 {
 131         LLVMBuilderRef builder = bld_base->base.gallivm->builder;
 132
 133         LLVMValueRef arg0 = LLVMBuildBitCast(builder, emit_data->args[0],
 134                                              bld_base->uint_bld.elem_type, "");
 135
 136         LLVMValueRef v = LLVMBuildICmp(builder, LLVMIntNE, arg0,
 137                                        bld_base->uint_bld.zero, "");
 138
 139         emit_data->output[emit_data->chan] =
 140                 LLVMBuildSelect(builder, v, emit_data->args[1], emit_data->args[2], "");
 141 }
 142
 143 static void emit_cmp(const struct lp_build_tgsi_action *action,
 144                      struct lp_build_tgsi_context *bld_base,
 145                      struct lp_build_emit_data *emit_data)
 146 {
 147         LLVMBuilderRef builder = bld_base->base.gallivm->builder;
 148         LLVMValueRef cond, *args = emit_data->args;
 149
 150         cond = LLVMBuildFCmp(builder, LLVMRealOLT, args[0],
 151                              bld_base->base.zero, "");
 152
 153         emit_data->output[emit_data->chan] =
 154                 LLVMBuildSelect(builder, cond, args[1], args[2], "");
 155 }
 156
 157 static void emit_set_cond(const struct lp_build_tgsi_action *action,
 158                           struct lp_build_tgsi_context *bld_base,
 159                           struct lp_build_emit_data *emit_data)
 160 {
 161         LLVMBuilderRef builder = bld_base->base.gallivm->builder;
 162         LLVMRealPredicate pred;
 163         LLVMValueRef cond;
 164
 165         /* Use ordered for everything but NE (which is usual for
 166          * float comparisons)
 167          */
 168         switch (emit_data->inst->Instruction.Opcode) {
 169         case TGSI_OPCODE_SGE: pred = LLVMRealOGE; break;
 170         case TGSI_OPCODE_SEQ: pred = LLVMRealOEQ; break;
 171         case TGSI_OPCODE_SLE: pred = LLVMRealOLE; break;
 172         case TGSI_OPCODE_SLT: pred = LLVMRealOLT; break;
 173         case TGSI_OPCODE_SNE: pred = LLVMRealUNE; break;
 174         case TGSI_OPCODE_SGT: pred = LLVMRealOGT; break;
 175         default: assert(!"unknown instruction"); pred = 0; break;
 176         }
 177
 178         cond = LLVMBuildFCmp(builder,
 179                 pred, emit_data->args[0], emit_data->args[1], "");
 180
 181         emit_data->output[emit_data->chan] = LLVMBuildSelect(builder,
 182                 cond, bld_base->base.one, bld_base->base.zero, "");
 183 }
 184
 185 static void emit_fcmp(const struct lp_build_tgsi_action *action,
 186                       struct lp_build_tgsi_context *bld_base,
 187                       struct lp_build_emit_data *emit_data)
 188 {
 189         LLVMBuilderRef builder = bld_base->base.gallivm->builder;
 190         LLVMContextRef context = bld_base->base.gallivm->context;
 191         LLVMRealPredicate pred;
 192
 193         /* Use ordered for everything but NE (which is usual for
 194          * float comparisons)
 195          */
 196         switch (emit_data->inst->Instruction.Opcode) {
 197         case TGSI_OPCODE_FSEQ: pred = LLVMRealOEQ; break;
 198         case TGSI_OPCODE_FSGE: pred = LLVMRealOGE; break;
 199         case TGSI_OPCODE_FSLT: pred = LLVMRealOLT; break;
 200         case TGSI_OPCODE_FSNE: pred = LLVMRealUNE; break;
 201         default: assert(!"unknown instruction"); pred = 0; break;
 202         }
 203
 204         LLVMValueRef v = LLVMBuildFCmp(builder, pred,
 205                         emit_data->args[0], emit_data->args[1],"");
 206
 207         v = LLVMBuildSExtOrBitCast(builder, v,
 208                         LLVMInt32TypeInContext(context), "");
 209
 210         emit_data->output[emit_data->chan] = v;
 211 }
 212
 213 static void emit_dcmp(const struct lp_build_tgsi_action *action,
 214                       struct lp_build_tgsi_context *bld_base,
 215                       struct lp_build_emit_data *emit_data)
 216 {
 217         LLVMBuilderRef builder = bld_base->base.gallivm->builder;
 218         LLVMContextRef context = bld_base->base.gallivm->context;
 219         LLVMRealPredicate pred;
 220
 221         /* Use ordered for everything but NE (which is usual for
 222          * float comparisons)
 223          */
 224         switch (emit_data->inst->Instruction.Opcode) {
 225         case TGSI_OPCODE_DSEQ: pred = LLVMRealOEQ; break;
 226         case TGSI_OPCODE_DSGE: pred = LLVMRealOGE; break;
 227         case TGSI_OPCODE_DSLT: pred = LLVMRealOLT; break;
 228         case TGSI_OPCODE_DSNE: pred = LLVMRealUNE; break;
 229         default: assert(!"unknown instruction"); pred = 0; break;
 230         }
 231
 232         LLVMValueRef v = LLVMBuildFCmp(builder, pred,
 233                         emit_data->args[0], emit_data->args[1],"");
 234
 235         v = LLVMBuildSExtOrBitCast(builder, v,
 236                         LLVMInt32TypeInContext(context), "");
 237
 238         emit_data->output[emit_data->chan] = v;
 239 }
 240
 241 static void emit_not(const struct lp_build_tgsi_action *action,
 242                      struct lp_build_tgsi_context *bld_base,
 243                      struct lp_build_emit_data *emit_data)
 244 {
 245         LLVMBuilderRef builder = bld_base->base.gallivm->builder;
 246         LLVMValueRef v = bitcast(bld_base, TGSI_TYPE_UNSIGNED,
 247                         emit_data->args[0]);
 248         emit_data->output[emit_data->chan] = LLVMBuildNot(builder, v, "");
 249 }
 250
 251 static void emit_arl(const struct lp_build_tgsi_action *action,
 252                      struct lp_build_tgsi_context *bld_base,
 253                      struct lp_build_emit_data *emit_data)
 254 {
 255         LLVMBuilderRef builder = bld_base->base.gallivm->builder;
 256         LLVMValueRef floor_index =  lp_build_emit_llvm_unary(bld_base, TGSI_OPCODE_FLR, emit_data->args[0]);
 257         emit_data->output[emit_data->chan] = LLVMBuildFPToSI(builder,
 258                         floor_index, bld_base->base.int_elem_type , "");
 259 }
 260
 261 static void emit_and(const struct lp_build_tgsi_action *action,
 262                      struct lp_build_tgsi_context *bld_base,
 263                      struct lp_build_emit_data *emit_data)
 264 {
 265         LLVMBuilderRef builder = bld_base->base.gallivm->builder;
 266         emit_data->output[emit_data->chan] = LLVMBuildAnd(builder,
 267                         emit_data->args[0], emit_data->args[1], "");
 268 }
 269
 270 static void emit_or(const struct lp_build_tgsi_action *action,
 271                     struct lp_build_tgsi_context *bld_base,
 272                     struct lp_build_emit_data *emit_data)
 273 {
 274         LLVMBuilderRef builder = bld_base->base.gallivm->builder;
 275         emit_data->output[emit_data->chan] = LLVMBuildOr(builder,
 276                         emit_data->args[0], emit_data->args[1], "");
 277 }
 278
 279 static void emit_uadd(const struct lp_build_tgsi_action *action,
 280                       struct lp_build_tgsi_context *bld_base,
 281                       struct lp_build_emit_data *emit_data)
 282 {
 283         LLVMBuilderRef builder = bld_base->base.gallivm->builder;
 284         emit_data->output[emit_data->chan] = LLVMBuildAdd(builder,
 285                         emit_data->args[0], emit_data->args[1], "");
 286 }
 287
 288 static void emit_udiv(const struct lp_build_tgsi_action *action,
 289                       struct lp_build_tgsi_context *bld_base,
 290                       struct lp_build_emit_data *emit_data)
 291 {
 292         LLVMBuilderRef builder = bld_base->base.gallivm->builder;
 293         emit_data->output[emit_data->chan] = LLVMBuildUDiv(builder,
 294                         emit_data->args[0], emit_data->args[1], "");
 295 }
 296
 297 static void emit_idiv(const struct lp_build_tgsi_action *action,
 298                       struct lp_build_tgsi_context *bld_base,
 299                       struct lp_build_emit_data *emit_data)
 300 {
 301         LLVMBuilderRef builder = bld_base->base.gallivm->builder;
 302         emit_data->output[emit_data->chan] = LLVMBuildSDiv(builder,
 303                         emit_data->args[0], emit_data->args[1], "");
 304 }
 305
 306 static void emit_mod(const struct lp_build_tgsi_action *action,
 307                      struct lp_build_tgsi_context *bld_base,
 308                      struct lp_build_emit_data *emit_data)
 309 {
 310         LLVMBuilderRef builder = bld_base->base.gallivm->builder;
 311         emit_data->output[emit_data->chan] = LLVMBuildSRem(builder,
 312                         emit_data->args[0], emit_data->args[1], "");
 313 }
 314
 315 static void emit_umod(const struct lp_build_tgsi_action *action,
 316                       struct lp_build_tgsi_context *bld_base,
 317                       struct lp_build_emit_data *emit_data)
 318 {
 319         LLVMBuilderRef builder = bld_base->base.gallivm->builder;
 320         emit_data->output[emit_data->chan] = LLVMBuildURem(builder,
 321                         emit_data->args[0], emit_data->args[1], "");
 322 }
 323
 324 static void emit_shl(const struct lp_build_tgsi_action *action,
 325                      struct lp_build_tgsi_context *bld_base,
 326                      struct lp_build_emit_data *emit_data)
 327 {
 328         LLVMBuilderRef builder = bld_base->base.gallivm->builder;
 329         emit_data->output[emit_data->chan] = LLVMBuildShl(builder,
 330                         emit_data->args[0], emit_data->args[1], "");
 331 }
 332
 333 static void emit_ushr(const struct lp_build_tgsi_action *action,
 334                       struct lp_build_tgsi_context *bld_base,
 335                       struct lp_build_emit_data *emit_data)
 336 {
 337         LLVMBuilderRef builder = bld_base->base.gallivm->builder;
 338         emit_data->output[emit_data->chan] = LLVMBuildLShr(builder,
 339                         emit_data->args[0], emit_data->args[1], "");
 340 }
 341 static void emit_ishr(const struct lp_build_tgsi_action *action,
 342                       struct lp_build_tgsi_context *bld_base,
 343                       struct lp_build_emit_data *emit_data)
 344 {
 345         LLVMBuilderRef builder = bld_base->base.gallivm->builder;
 346         emit_data->output[emit_data->chan] = LLVMBuildAShr(builder,
 347                         emit_data->args[0], emit_data->args[1], "");
 348 }
 349
 350 static void emit_xor(const struct lp_build_tgsi_action *action,
 351                      struct lp_build_tgsi_context *bld_base,
 352                      struct lp_build_emit_data *emit_data)
 353 {
 354         LLVMBuilderRef builder = bld_base->base.gallivm->builder;
 355         emit_data->output[emit_data->chan] = LLVMBuildXor(builder,
 356                         emit_data->args[0], emit_data->args[1], "");
 357 }
 358
 359 static void emit_ssg(const struct lp_build_tgsi_action *action,
 360                      struct lp_build_tgsi_context *bld_base,
 361                      struct lp_build_emit_data *emit_data)
 362 {
 363         LLVMBuilderRef builder = bld_base->base.gallivm->builder;
 364
 365         LLVMValueRef cmp, val;
 366
 367         if (emit_data->inst->Instruction.Opcode == TGSI_OPCODE_I64SSG) {
 368                 cmp = LLVMBuildICmp(builder, LLVMIntSGT, emit_data->args[0], bld_base->int64_bld.zero, "");
 369                 val = LLVMBuildSelect(builder, cmp, bld_base->int64_bld.one, emit_data->args[0], "");
 370                 cmp = LLVMBuildICmp(builder, LLVMIntSGE, val, bld_base->int64_bld.zero, "");
 371                 val = LLVMBuildSelect(builder, cmp, val, LLVMConstInt(bld_base->int64_bld.elem_type, -1, true), "");
 372         } else if (emit_data->inst->Instruction.Opcode == TGSI_OPCODE_ISSG) {
 373                 cmp = LLVMBuildICmp(builder, LLVMIntSGT, emit_data->args[0], bld_base->int_bld.zero, "");
 374                 val = LLVMBuildSelect(builder, cmp, bld_base->int_bld.one, emit_data->args[0], "");
 375                 cmp = LLVMBuildICmp(builder, LLVMIntSGE, val, bld_base->int_bld.zero, "");
 376                 val = LLVMBuildSelect(builder, cmp, val, LLVMConstInt(bld_base->int_bld.elem_type, -1, true), "");
 377         } else if (emit_data->inst->Instruction.Opcode == TGSI_OPCODE_DSSG) {
 378                 cmp = LLVMBuildFCmp(builder, LLVMRealOGT, emit_data->args[0], bld_base->dbl_bld.zero, "");
 379                 val = LLVMBuildSelect(builder, cmp, bld_base->dbl_bld.one, emit_data->args[0], "");
 380                 cmp = LLVMBuildFCmp(builder, LLVMRealOGE, val, bld_base->dbl_bld.zero, "");
 381                 val = LLVMBuildSelect(builder, cmp, val, LLVMConstReal(bld_base->dbl_bld.elem_type, -1), "");
 382         } else { // float SSG
 383                 cmp = LLVMBuildFCmp(builder, LLVMRealOGT, emit_data->args[0], bld_base->base.zero, "");
 384                 val = LLVMBuildSelect(builder, cmp, bld_base->base.one, emit_data->args[0], "");
 385                 cmp = LLVMBuildFCmp(builder, LLVMRealOGE, val, bld_base->base.zero, "");
 386                 val = LLVMBuildSelect(builder, cmp, val, LLVMConstReal(bld_base->base.elem_type, -1), "");
 387         }
 388
 389         emit_data->output[emit_data->chan] = val;
 390 }
 391
 392 static void emit_ineg(const struct lp_build_tgsi_action *action,
 393                       struct lp_build_tgsi_context *bld_base,
 394                       struct lp_build_emit_data *emit_data)
 395 {
 396         LLVMBuilderRef builder = bld_base->base.gallivm->builder;
 397         emit_data->output[emit_data->chan] = LLVMBuildNeg(builder,
 398                         emit_data->args[0], "");
 399 }
 400
 401 static void emit_dneg(const struct lp_build_tgsi_action *action,
 402                       struct lp_build_tgsi_context *bld_base,
 403                       struct lp_build_emit_data *emit_data)
 404 {
 405         LLVMBuilderRef builder = bld_base->base.gallivm->builder;
 406         emit_data->output[emit_data->chan] = LLVMBuildFNeg(builder,
 407                         emit_data->args[0], "");
 408 }
 409
 410 static void emit_frac(const struct lp_build_tgsi_action *action,
 411                       struct lp_build_tgsi_context *bld_base,
 412                       struct lp_build_emit_data *emit_data)
 413 {
 414         LLVMBuilderRef builder = bld_base->base.gallivm->builder;
 415         char *intr;
 416
 417         if (emit_data->info->opcode == TGSI_OPCODE_FRC)
 418                 intr = "llvm.floor.f32";
 419         else if (emit_data->info->opcode == TGSI_OPCODE_DFRAC)
 420                 intr = "llvm.floor.f64";
 421         else {
 422                 assert(0);
 423                 return;
 424         }
 425
 426         LLVMValueRef floor = lp_build_intrinsic(builder, intr, emit_data->dst_type,
 427                                                 &emit_data->args[0], 1,
 428                                                 LP_FUNC_ATTR_READNONE);
 429         emit_data->output[emit_data->chan] = LLVMBuildFSub(builder,
 430                         emit_data->args[0], floor, "");
 431 }
 432
 433 static void emit_f2i(const struct lp_build_tgsi_action *action,
 434                      struct lp_build_tgsi_context *bld_base,
 435                      struct lp_build_emit_data *emit_data)
 436 {
 437         LLVMBuilderRef builder = bld_base->base.gallivm->builder;
 438         emit_data->output[emit_data->chan] = LLVMBuildFPToSI(builder,
 439                         emit_data->args[0], bld_base->int_bld.elem_type, "");
 440 }
 441
 442 static void emit_f2u(const struct lp_build_tgsi_action *action,
 443                      struct lp_build_tgsi_context *bld_base,
 444                      struct lp_build_emit_data *emit_data)
 445 {
 446         LLVMBuilderRef builder = bld_base->base.gallivm->builder;
 447         emit_data->output[emit_data->chan] = LLVMBuildFPToUI(builder,
 448                         emit_data->args[0], bld_base->uint_bld.elem_type, "");
 449 }
 450
 451 static void emit_i2f(const struct lp_build_tgsi_action *action,
 452                      struct lp_build_tgsi_context *bld_base,
 453                      struct lp_build_emit_data *emit_data)
 454 {
 455         LLVMBuilderRef builder = bld_base->base.gallivm->builder;
 456         emit_data->output[emit_data->chan] = LLVMBuildSIToFP(builder,
 457                         emit_data->args[0], bld_base->base.elem_type, "");
 458 }
 459
 460 static void emit_u2f(const struct lp_build_tgsi_action *action,
 461                      struct lp_build_tgsi_context *bld_base,
 462                      struct lp_build_emit_data *emit_data)
 463 {
 464         LLVMBuilderRef builder = bld_base->base.gallivm->builder;
 465         emit_data->output[emit_data->chan] = LLVMBuildUIToFP(builder,
 466                         emit_data->args[0], bld_base->base.elem_type, "");
 467 }
 468
 469 static void
 470 build_tgsi_intrinsic_nomem(const struct lp_build_tgsi_action *action,
 471                            struct lp_build_tgsi_context *bld_base,
 472                            struct lp_build_emit_data *emit_data)
 473 {
 474         struct lp_build_context *base = &bld_base->base;
 475         emit_data->output[emit_data->chan] =
 476                 lp_build_intrinsic(base->gallivm->builder, action->intr_name,
 477                                    emit_data->dst_type, emit_data->args,
 478                                    emit_data->arg_count, LP_FUNC_ATTR_READNONE);
 479 }
 480
 481 static void emit_bfi(const struct lp_build_tgsi_action *action,
 482                      struct lp_build_tgsi_context *bld_base,
 483                      struct lp_build_emit_data *emit_data)
 484 {
 485         struct gallivm_state *gallivm = bld_base->base.gallivm;
 486         LLVMBuilderRef builder = gallivm->builder;
 487         LLVMValueRef bfi_args[3];
 488         LLVMValueRef bfi_sm5;
 489         LLVMValueRef cond;
 490
 491         // Calculate the bitmask: (((1 << src3) - 1) << src2
 492         bfi_args[0] = LLVMBuildShl(builder,
 493                                    LLVMBuildSub(builder,
 494                                                 LLVMBuildShl(builder,
 495                                                              bld_base->int_bld.one,
 496                                                              emit_data->args[3], ""),
 497                                                 bld_base->int_bld.one, ""),
 498                                    emit_data->args[2], "");
 499
 500         bfi_args[1] = LLVMBuildShl(builder, emit_data->args[1],
 501                                    emit_data->args[2], "");
 502
 503         bfi_args[2] = emit_data->args[0];
 504
 505         /* Calculate:
 506          *   (arg0 & arg1) | (~arg0 & arg2) = arg2 ^ (arg0 & (arg1 ^ arg2)
 507          * Use the right-hand side, which the LLVM backend can convert to V_BFI.
 508          */
 509         bfi_sm5 =
 510                 LLVMBuildXor(builder, bfi_args[2],
 511                         LLVMBuildAnd(builder, bfi_args[0],
 512                                 LLVMBuildXor(builder, bfi_args[1], bfi_args[2],
 513                                              ""), ""), "");
 514
 515         /* Since shifts of >= 32 bits are undefined in LLVM IR, the backend
 516          * uses the convenient V_BFI lowering for the above, which follows SM5
 517          * and disagrees with GLSL semantics when bits (src3) is 32.
 518          */
 519         cond = LLVMBuildICmp(builder, LLVMIntUGE, emit_data->args[3],
 520                              lp_build_const_int32(gallivm, 32), "");
 521         emit_data->output[emit_data->chan] =
 522                 LLVMBuildSelect(builder, cond, emit_data->args[1], bfi_sm5, "");
 523 }
 524
 525 static void emit_bfe(const struct lp_build_tgsi_action *action,
 526                      struct lp_build_tgsi_context *bld_base,
 527                      struct lp_build_emit_data *emit_data)
 528 {
 529         struct si_shader_context *ctx = si_shader_context(bld_base);
 530         struct gallivm_state *gallivm = &ctx->gallivm;
 531         LLVMBuilderRef builder = gallivm->builder;
 532         LLVMValueRef bfe_sm5;
 533         LLVMValueRef cond;
 534
 535         bfe_sm5 = ac_build_bfe(&ctx->ac, emit_data->args[0],
 536                                emit_data->args[1], emit_data->args[2],
 537                                emit_data->info->opcode == TGSI_OPCODE_IBFE);
 538
 539         /* Correct for GLSL semantics. */
 540         cond = LLVMBuildICmp(builder, LLVMIntUGE, emit_data->args[2],
 541                              LLVMConstInt(ctx->i32, 32, 0), "");
 542         emit_data->output[emit_data->chan] =
 543                 LLVMBuildSelect(builder, cond, emit_data->args[0], bfe_sm5, "");
 544 }
 545
 546 /* this is ffs in C */
 547 static void emit_lsb(const struct lp_build_tgsi_action *action,
 548                      struct lp_build_tgsi_context *bld_base,
 549                      struct lp_build_emit_data *emit_data)
 550 {
 551         struct gallivm_state *gallivm = bld_base->base.gallivm;
 552         LLVMBuilderRef builder = gallivm->builder;
 553         LLVMValueRef args[2] = {
 554                 emit_data->args[0],
 555
 556                 /* The value of 1 means that ffs(x=0) = undef, so LLVM won't
 557                  * add special code to check for x=0. The reason is that
 558                  * the LLVM behavior for x=0 is different from what we
 559                  * need here. However, LLVM also assumes that ffs(x) is
 560                  * in [0, 31], but GLSL expects that ffs(0) = -1, so
 561                  * a conditional assignment to handle 0 is still required.
 562                  */
 563                 LLVMConstInt(LLVMInt1TypeInContext(gallivm->context), 1, 0)
 564         };
 565
 566         LLVMValueRef lsb =
 567                 lp_build_intrinsic(gallivm->builder, "llvm.cttz.i32",
 568                                 emit_data->dst_type, args, ARRAY_SIZE(args),
 569                                 LP_FUNC_ATTR_READNONE);
 570
 571         /* TODO: We need an intrinsic to skip this conditional. */
 572         /* Check for zero: */
 573         emit_data->output[emit_data->chan] =
 574                 LLVMBuildSelect(builder,
 575                                 LLVMBuildICmp(builder, LLVMIntEQ, args[0],
 576                                               bld_base->uint_bld.zero, ""),
 577                                 lp_build_const_int32(gallivm, -1), lsb, "");
 578 }
 579
 580 /* Find the last bit set. */
 581 static void emit_umsb(const struct lp_build_tgsi_action *action,
 582                       struct lp_build_tgsi_context *bld_base,
 583                       struct lp_build_emit_data *emit_data)
 584 {
 585         struct si_shader_context *ctx = si_shader_context(bld_base);
 586
 587         emit_data->output[emit_data->chan] =
 588                 ac_build_umsb(&ctx->ac, emit_data->args[0], emit_data->dst_type);
 589 }
 590
 591 /* Find the last bit opposite of the sign bit. */
 592 static void emit_imsb(const struct lp_build_tgsi_action *action,
 593                       struct lp_build_tgsi_context *bld_base,
 594                       struct lp_build_emit_data *emit_data)
 595 {
 596         struct si_shader_context *ctx = si_shader_context(bld_base);
 597         emit_data->output[emit_data->chan] =
 598                 ac_build_imsb(&ctx->ac, emit_data->args[0],
 599                               emit_data->dst_type);
 600 }
 601
 602 static void emit_iabs(const struct lp_build_tgsi_action *action,
 603                       struct lp_build_tgsi_context *bld_base,
 604                       struct lp_build_emit_data *emit_data)
 605 {
 606         LLVMBuilderRef builder = bld_base->base.gallivm->builder;
 607
 608         emit_data->output[emit_data->chan] =
 609                 lp_build_emit_llvm_binary(bld_base, TGSI_OPCODE_IMAX,
 610                                           emit_data->args[0],
 611                                           LLVMBuildNeg(builder,
 612                                                        emit_data->args[0], ""));
 613 }
 614
 615 static void emit_minmax_int(const struct lp_build_tgsi_action *action,
 616                             struct lp_build_tgsi_context *bld_base,
 617                             struct lp_build_emit_data *emit_data)
 618 {
 619         LLVMBuilderRef builder = bld_base->base.gallivm->builder;
 620         LLVMIntPredicate op;
 621
 622         switch (emit_data->info->opcode) {
 623         default:
 624                 assert(0);
 625         case TGSI_OPCODE_IMAX:
 626         case TGSI_OPCODE_I64MAX:
 627                 op = LLVMIntSGT;
 628                 break;
 629         case TGSI_OPCODE_IMIN:
 630         case TGSI_OPCODE_I64MIN:
 631                 op = LLVMIntSLT;
 632                 break;
 633         case TGSI_OPCODE_UMAX:
 634         case TGSI_OPCODE_U64MAX:
 635                 op = LLVMIntUGT;
 636                 break;
 637         case TGSI_OPCODE_UMIN:
 638         case TGSI_OPCODE_U64MIN:
 639                 op = LLVMIntULT;
 640                 break;
 641         }
 642
 643         emit_data->output[emit_data->chan] =
 644                 LLVMBuildSelect(builder,
 645                                 LLVMBuildICmp(builder, op, emit_data->args[0],
 646                                               emit_data->args[1], ""),
 647                                 emit_data->args[0],
 648                                 emit_data->args[1], "");
 649 }
 650
 651 static void pk2h_fetch_args(struct lp_build_tgsi_context *bld_base,
 652                             struct lp_build_emit_data *emit_data)
 653 {
 654         emit_data->args[0] = lp_build_emit_fetch(bld_base, emit_data->inst,
 655                                                  0, TGSI_CHAN_X);
 656         emit_data->args[1] = lp_build_emit_fetch(bld_base, emit_data->inst,
 657                                                  0, TGSI_CHAN_Y);
 658 }
 659
 660 static void emit_pk2h(const struct lp_build_tgsi_action *action,
 661                       struct lp_build_tgsi_context *bld_base,
 662                       struct lp_build_emit_data *emit_data)
 663 {
 664         /* From the GLSL 4.50 spec:
 665          *   "The rounding mode cannot be set and is undefined."
 666          *
 667          * v_cvt_pkrtz_f16 rounds to zero, but it's fastest.
 668          */
 669         emit_data->output[emit_data->chan] =
 670                 ac_build_cvt_pkrtz_f16(&si_shader_context(bld_base)->ac,
 671                                        emit_data->args);
 672 }
 673
 674 static void up2h_fetch_args(struct lp_build_tgsi_context *bld_base,
 675                             struct lp_build_emit_data *emit_data)
 676 {
 677         emit_data->args[0] = lp_build_emit_fetch(bld_base, emit_data->inst,
 678                                                  0, TGSI_CHAN_X);
 679 }
 680
 681 static void emit_up2h(const struct lp_build_tgsi_action *action,
 682                       struct lp_build_tgsi_context *bld_base,
 683                       struct lp_build_emit_data *emit_data)
 684 {
 685         LLVMBuilderRef builder = bld_base->base.gallivm->builder;
 686         LLVMContextRef context = bld_base->base.gallivm->context;
 687         struct lp_build_context *uint_bld = &bld_base->uint_bld;
 688         LLVMTypeRef fp16, i16;
 689         LLVMValueRef const16, input, val;
 690         unsigned i;
 691
 692         fp16 = LLVMHalfTypeInContext(context);
 693         i16 = LLVMInt16TypeInContext(context);
 694         const16 = lp_build_const_int32(uint_bld->gallivm, 16);
 695         input = emit_data->args[0];
 696
 697         for (i = 0; i < 2; i++) {
 698                 val = i == 1 ? LLVMBuildLShr(builder, input, const16, "") : input;
 699                 val = LLVMBuildTrunc(builder, val, i16, "");
 700                 val = LLVMBuildBitCast(builder, val, fp16, "");
 701                 emit_data->output[i] =
 702                         LLVMBuildFPExt(builder, val, bld_base->base.elem_type, "");
 703         }
 704 }
 705
 706 static void emit_fdiv(const struct lp_build_tgsi_action *action,
 707                       struct lp_build_tgsi_context *bld_base,
 708                       struct lp_build_emit_data *emit_data)
 709 {
 710         struct si_shader_context *ctx = si_shader_context(bld_base);
 711
 712         emit_data->output[emit_data->chan] =
 713                 LLVMBuildFDiv(ctx->gallivm.builder,
 714                               emit_data->args[0], emit_data->args[1], "");
 715
 716         /* Use v_rcp_f32 instead of precise division. */
 717         if (!LLVMIsConstant(emit_data->output[emit_data->chan]))
 718                 LLVMSetMetadata(emit_data->output[emit_data->chan],
 719                                 ctx->fpmath_md_kind, ctx->fpmath_md_2p5_ulp);
 720 }
 721
 722 /* 1/sqrt is translated to rsq for f32 if fp32 denormals are not enabled in
 723  * the target machine. f64 needs global unsafe math flags to get rsq. */
 724 static void emit_rsq(const struct lp_build_tgsi_action *action,
 725                      struct lp_build_tgsi_context *bld_base,
 726                      struct lp_build_emit_data *emit_data)
 727 {
 728         LLVMValueRef sqrt =
 729                 lp_build_emit_llvm_unary(bld_base, TGSI_OPCODE_SQRT,
 730                                          emit_data->args[0]);
 731
 732         emit_data->output[emit_data->chan] =
 733                 lp_build_emit_llvm_binary(bld_base, TGSI_OPCODE_DIV,
 734                                           bld_base->base.one, sqrt);
 735 }
 736
 737 void si_shader_context_init_alu(struct lp_build_tgsi_context *bld_base)
 738 {
 739         lp_set_default_actions(bld_base);
 740
 741         bld_base->op_actions[TGSI_OPCODE_AND].emit = emit_and;
 742         bld_base->op_actions[TGSI_OPCODE_ARL].emit = emit_arl;
 743         bld_base->op_actions[TGSI_OPCODE_BFI].emit = emit_bfi;
 744         bld_base->op_actions[TGSI_OPCODE_BREV].emit = build_tgsi_intrinsic_nomem;
 745         bld_base->op_actions[TGSI_OPCODE_BREV].intr_name = "llvm.bitreverse.i32";
 746         bld_base->op_actions[TGSI_OPCODE_CEIL].emit = build_tgsi_intrinsic_nomem;
 747         bld_base->op_actions[TGSI_OPCODE_CEIL].intr_name = "llvm.ceil.f32";
 748         bld_base->op_actions[TGSI_OPCODE_CMP].emit = emit_cmp;
 749         bld_base->op_actions[TGSI_OPCODE_COS].emit = build_tgsi_intrinsic_nomem;
 750         bld_base->op_actions[TGSI_OPCODE_COS].intr_name = "llvm.cos.f32";
 751         bld_base->op_actions[TGSI_OPCODE_DABS].emit = build_tgsi_intrinsic_nomem;
 752         bld_base->op_actions[TGSI_OPCODE_DABS].intr_name = "llvm.fabs.f64";
 753         bld_base->op_actions[TGSI_OPCODE_DCEIL].emit = build_tgsi_intrinsic_nomem;
 754         bld_base->op_actions[TGSI_OPCODE_DCEIL].intr_name = "llvm.ceil.f64";
 755         bld_base->op_actions[TGSI_OPCODE_DFLR].emit = build_tgsi_intrinsic_nomem;
 756         bld_base->op_actions[TGSI_OPCODE_DFLR].intr_name = "llvm.floor.f64";
 757         bld_base->op_actions[TGSI_OPCODE_DFMA].emit = build_tgsi_intrinsic_nomem;
 758         bld_base->op_actions[TGSI_OPCODE_DFMA].intr_name = "llvm.fma.f64";
 759         bld_base->op_actions[TGSI_OPCODE_DFRAC].emit = emit_frac;
 760         bld_base->op_actions[TGSI_OPCODE_DIV].emit = emit_fdiv;
 761         bld_base->op_actions[TGSI_OPCODE_DNEG].emit = emit_dneg;
 762         bld_base->op_actions[TGSI_OPCODE_DROUND].emit = build_tgsi_intrinsic_nomem;
 763         bld_base->op_actions[TGSI_OPCODE_DROUND].intr_name = "llvm.rint.f64";
 764         bld_base->op_actions[TGSI_OPCODE_DSEQ].emit = emit_dcmp;
 765         bld_base->op_actions[TGSI_OPCODE_DSGE].emit = emit_dcmp;
 766         bld_base->op_actions[TGSI_OPCODE_DSLT].emit = emit_dcmp;
 767         bld_base->op_actions[TGSI_OPCODE_DSNE].emit = emit_dcmp;
 768         bld_base->op_actions[TGSI_OPCODE_DSSG].emit = emit_ssg;
 769         bld_base->op_actions[TGSI_OPCODE_DRSQ].emit = build_tgsi_intrinsic_nomem;
 770         bld_base->op_actions[TGSI_OPCODE_DRSQ].intr_name = "llvm.amdgcn.rsq.f64";
 771         bld_base->op_actions[TGSI_OPCODE_DSQRT].emit = build_tgsi_intrinsic_nomem;
 772         bld_base->op_actions[TGSI_OPCODE_DSQRT].intr_name = "llvm.sqrt.f64";
 773         bld_base->op_actions[TGSI_OPCODE_DTRUNC].emit = build_tgsi_intrinsic_nomem;
 774         bld_base->op_actions[TGSI_OPCODE_DTRUNC].intr_name = "llvm.trunc.f64";
 775         bld_base->op_actions[TGSI_OPCODE_EX2].emit = build_tgsi_intrinsic_nomem;
 776         bld_base->op_actions[TGSI_OPCODE_EX2].intr_name = "llvm.exp2.f32";
 777         bld_base->op_actions[TGSI_OPCODE_FLR].emit = build_tgsi_intrinsic_nomem;
 778         bld_base->op_actions[TGSI_OPCODE_FLR].intr_name = "llvm.floor.f32";
 779         bld_base->op_actions[TGSI_OPCODE_FMA].emit =
 780                 bld_base->op_actions[TGSI_OPCODE_MAD].emit;
 781         bld_base->op_actions[TGSI_OPCODE_FRC].emit = emit_frac;
 782         bld_base->op_actions[TGSI_OPCODE_F2I].emit = emit_f2i;
 783         bld_base->op_actions[TGSI_OPCODE_F2U].emit = emit_f2u;
 784         bld_base->op_actions[TGSI_OPCODE_FSEQ].emit = emit_fcmp;
 785         bld_base->op_actions[TGSI_OPCODE_FSGE].emit = emit_fcmp;
 786         bld_base->op_actions[TGSI_OPCODE_FSLT].emit = emit_fcmp;
 787         bld_base->op_actions[TGSI_OPCODE_FSNE].emit = emit_fcmp;
 788         bld_base->op_actions[TGSI_OPCODE_IABS].emit = emit_iabs;
 789         bld_base->op_actions[TGSI_OPCODE_IBFE].emit = emit_bfe;
 790         bld_base->op_actions[TGSI_OPCODE_IDIV].emit = emit_idiv;
 791         bld_base->op_actions[TGSI_OPCODE_IMAX].emit = emit_minmax_int;
 792         bld_base->op_actions[TGSI_OPCODE_IMIN].emit = emit_minmax_int;
 793         bld_base->op_actions[TGSI_OPCODE_IMSB].emit = emit_imsb;
 794         bld_base->op_actions[TGSI_OPCODE_INEG].emit = emit_ineg;
 795         bld_base->op_actions[TGSI_OPCODE_ISHR].emit = emit_ishr;
 796         bld_base->op_actions[TGSI_OPCODE_ISGE].emit = emit_icmp;
 797         bld_base->op_actions[TGSI_OPCODE_ISLT].emit = emit_icmp;
 798         bld_base->op_actions[TGSI_OPCODE_ISSG].emit = emit_ssg;
 799         bld_base->op_actions[TGSI_OPCODE_I2F].emit = emit_i2f;
 800         bld_base->op_actions[TGSI_OPCODE_KILL_IF].fetch_args = kill_if_fetch_args;
 801         bld_base->op_actions[TGSI_OPCODE_KILL_IF].emit = kil_emit;
 802         bld_base->op_actions[TGSI_OPCODE_KILL].emit = kil_emit;
 803         bld_base->op_actions[TGSI_OPCODE_LSB].emit = emit_lsb;
 804         bld_base->op_actions[TGSI_OPCODE_LG2].emit = build_tgsi_intrinsic_nomem;
 805         bld_base->op_actions[TGSI_OPCODE_LG2].intr_name = "llvm.log2.f32";
 806         bld_base->op_actions[TGSI_OPCODE_MAX].emit = build_tgsi_intrinsic_nomem;
 807         bld_base->op_actions[TGSI_OPCODE_MAX].intr_name = "llvm.maxnum.f32";
 808         bld_base->op_actions[TGSI_OPCODE_MIN].emit = build_tgsi_intrinsic_nomem;
 809         bld_base->op_actions[TGSI_OPCODE_MIN].intr_name = "llvm.minnum.f32";
 810         bld_base->op_actions[TGSI_OPCODE_MOD].emit = emit_mod;
 811         bld_base->op_actions[TGSI_OPCODE_UMSB].emit = emit_umsb;
 812         bld_base->op_actions[TGSI_OPCODE_NOT].emit = emit_not;
 813         bld_base->op_actions[TGSI_OPCODE_OR].emit = emit_or;
 814         bld_base->op_actions[TGSI_OPCODE_PK2H].fetch_args = pk2h_fetch_args;
 815         bld_base->op_actions[TGSI_OPCODE_PK2H].emit = emit_pk2h;
 816         bld_base->op_actions[TGSI_OPCODE_POPC].emit = build_tgsi_intrinsic_nomem;
 817         bld_base->op_actions[TGSI_OPCODE_POPC].intr_name = "llvm.ctpop.i32";
 818         bld_base->op_actions[TGSI_OPCODE_POW].emit = build_tgsi_intrinsic_nomem;
 819         bld_base->op_actions[TGSI_OPCODE_POW].intr_name = "llvm.pow.f32";
 820         bld_base->op_actions[TGSI_OPCODE_ROUND].emit = build_tgsi_intrinsic_nomem;
 821         bld_base->op_actions[TGSI_OPCODE_ROUND].intr_name = "llvm.rint.f32";
 822         bld_base->op_actions[TGSI_OPCODE_RSQ].emit = emit_rsq;
 823         bld_base->op_actions[TGSI_OPCODE_SGE].emit = emit_set_cond;
 824         bld_base->op_actions[TGSI_OPCODE_SEQ].emit = emit_set_cond;
 825         bld_base->op_actions[TGSI_OPCODE_SHL].emit = emit_shl;
 826         bld_base->op_actions[TGSI_OPCODE_SLE].emit = emit_set_cond;
 827         bld_base->op_actions[TGSI_OPCODE_SLT].emit = emit_set_cond;
 828         bld_base->op_actions[TGSI_OPCODE_SNE].emit = emit_set_cond;
 829         bld_base->op_actions[TGSI_OPCODE_SGT].emit = emit_set_cond;
 830         bld_base->op_actions[TGSI_OPCODE_SIN].emit = build_tgsi_intrinsic_nomem;
 831         bld_base->op_actions[TGSI_OPCODE_SIN].intr_name = "llvm.sin.f32";
 832         bld_base->op_actions[TGSI_OPCODE_SQRT].emit = build_tgsi_intrinsic_nomem;
 833         bld_base->op_actions[TGSI_OPCODE_SQRT].intr_name = "llvm.sqrt.f32";
 834         bld_base->op_actions[TGSI_OPCODE_SSG].emit = emit_ssg;
 835         bld_base->op_actions[TGSI_OPCODE_TRUNC].emit = build_tgsi_intrinsic_nomem;
 836         bld_base->op_actions[TGSI_OPCODE_TRUNC].intr_name = "llvm.trunc.f32";
 837         bld_base->op_actions[TGSI_OPCODE_UADD].emit = emit_uadd;
 838         bld_base->op_actions[TGSI_OPCODE_UBFE].emit = emit_bfe;
 839         bld_base->op_actions[TGSI_OPCODE_UDIV].emit = emit_udiv;
 840         bld_base->op_actions[TGSI_OPCODE_UMAX].emit = emit_minmax_int;
 841         bld_base->op_actions[TGSI_OPCODE_UMIN].emit = emit_minmax_int;
 842         bld_base->op_actions[TGSI_OPCODE_UMOD].emit = emit_umod;
 843         bld_base->op_actions[TGSI_OPCODE_USEQ].emit = emit_icmp;
 844         bld_base->op_actions[TGSI_OPCODE_USGE].emit = emit_icmp;
 845         bld_base->op_actions[TGSI_OPCODE_USHR].emit = emit_ushr;
 846         bld_base->op_actions[TGSI_OPCODE_USLT].emit = emit_icmp;
 847         bld_base->op_actions[TGSI_OPCODE_USNE].emit = emit_icmp;
 848         bld_base->op_actions[TGSI_OPCODE_U2F].emit = emit_u2f;
 849         bld_base->op_actions[TGSI_OPCODE_XOR].emit = emit_xor;
 850         bld_base->op_actions[TGSI_OPCODE_UCMP].emit = emit_ucmp;
 851         bld_base->op_actions[TGSI_OPCODE_UP2H].fetch_args = up2h_fetch_args;
 852         bld_base->op_actions[TGSI_OPCODE_UP2H].emit = emit_up2h;
 853
 854         bld_base->op_actions[TGSI_OPCODE_I64MAX].emit = emit_minmax_int;
 855         bld_base->op_actions[TGSI_OPCODE_I64MIN].emit = emit_minmax_int;
 856         bld_base->op_actions[TGSI_OPCODE_U64MAX].emit = emit_minmax_int;
 857         bld_base->op_actions[TGSI_OPCODE_U64MIN].emit = emit_minmax_int;
 858         bld_base->op_actions[TGSI_OPCODE_I64ABS].emit = emit_iabs;
 859         bld_base->op_actions[TGSI_OPCODE_I64SSG].emit = emit_ssg;
 860         bld_base->op_actions[TGSI_OPCODE_I64NEG].emit = emit_ineg;
 861
 862         bld_base->op_actions[TGSI_OPCODE_U64SEQ].emit = emit_icmp;
 863         bld_base->op_actions[TGSI_OPCODE_U64SNE].emit = emit_icmp;
 864         bld_base->op_actions[TGSI_OPCODE_U64SGE].emit = emit_icmp;
 865         bld_base->op_actions[TGSI_OPCODE_U64SLT].emit = emit_icmp;
 866         bld_base->op_actions[TGSI_OPCODE_I64SGE].emit = emit_icmp;
 867         bld_base->op_actions[TGSI_OPCODE_I64SLT].emit = emit_icmp;
 868
 869         bld_base->op_actions[TGSI_OPCODE_U64ADD].emit = emit_uadd;
 870         bld_base->op_actions[TGSI_OPCODE_U64SHL].emit = emit_shl;
 871         bld_base->op_actions[TGSI_OPCODE_U64SHR].emit = emit_ushr;
 872         bld_base->op_actions[TGSI_OPCODE_I64SHR].emit = emit_ishr;
 873
 874         bld_base->op_actions[TGSI_OPCODE_U64MOD].emit = emit_umod;
 875         bld_base->op_actions[TGSI_OPCODE_I64MOD].emit = emit_mod;
 876         bld_base->op_actions[TGSI_OPCODE_U64DIV].emit = emit_udiv;
 877         bld_base->op_actions[TGSI_OPCODE_I64DIV].emit = emit_idiv;
 878 }