lp_bld_logic.c revision 1aede69d3a8d288af11c2ef620b51e71c2ce89b2
11aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca/**************************************************************************
21aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca *
31aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca * Copyright 2009 VMware, Inc.
41aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca * All Rights Reserved.
51aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca *
61aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca * Permission is hereby granted, free of charge, to any person obtaining a
71aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca * copy of this software and associated documentation files (the
81aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca * "Software"), to deal in the Software without restriction, including
91aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca * without limitation the rights to use, copy, modify, merge, publish,
101aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca * distribute, sub license, and/or sell copies of the Software, and to
111aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca * permit persons to whom the Software is furnished to do so, subject to
121aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca * the following conditions:
131aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca *
141aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca * The above copyright notice and this permission notice (including the
151aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca * next paragraph) shall be included in all copies or substantial portions
161aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca * of the Software.
171aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca *
181aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
191aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
201aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
211aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
221aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
231aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
241aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
251aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca *
261aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca **************************************************************************/
271aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca
281aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca
291aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca#include "pipe/p_defines.h"
301aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca#include "lp_bld_type.h"
311aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca#include "lp_bld_intr.h"
321aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca#include "lp_bld_logic.h"
331aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca
341aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca
351aede69d3a8d288af11c2ef620b51e71c2ce89b2José FonsecaLLVMValueRef
361aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonsecalp_build_cmp(struct lp_build_context *bld,
371aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca             unsigned func,
381aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca             LLVMValueRef a,
391aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca             LLVMValueRef b)
401aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca{
411aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca   const union lp_type type = bld->type;
421aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca   LLVMTypeRef vec_type = lp_build_vec_type(type);
431aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca   LLVMTypeRef int_vec_type = lp_build_int_vec_type(type);
441aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca   LLVMValueRef zeros = LLVMConstNull(int_vec_type);
451aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca   LLVMValueRef ones = LLVMConstAllOnes(int_vec_type);
461aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca   LLVMValueRef cond;
471aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca
481aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca   if(func == PIPE_FUNC_NEVER)
491aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca      return zeros;
501aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca   if(func == PIPE_FUNC_ALWAYS)
511aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca      return ones;
521aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca
531aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca   /* TODO: optimize the constant case */
541aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca
551aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca   /* XXX: It is not clear if we should use the ordered or unordered operators */
561aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca
571aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca#if defined(PIPE_ARCH_X86) || defined(PIPE_ARCH_X86_64)
581aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca   if(type.width * type.length == 128) {
591aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca      if(type.floating) {
601aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca         LLVMValueRef args[3];
611aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca         unsigned cc;
621aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca         boolean swap;
631aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca         LLVMValueRef res;
641aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca
651aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca         swap = FALSE;
661aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca         switch(func) {
671aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca         case PIPE_FUNC_EQUAL:
681aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca            cc = 0;
691aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca            break;
701aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca         case PIPE_FUNC_NOTEQUAL:
711aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca            cc = 4;
721aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca            break;
731aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca         case PIPE_FUNC_LESS:
741aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca            cc = 1;
751aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca            break;
761aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca         case PIPE_FUNC_LEQUAL:
771aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca            cc = 2;
781aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca            break;
791aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca         case PIPE_FUNC_GREATER:
801aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca            cc = 1;
811aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca            swap = TRUE;
821aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca            break;
831aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca         case PIPE_FUNC_GEQUAL:
841aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca            cc = 2;
851aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca            swap = TRUE;
861aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca            break;
871aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca         default:
881aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca            assert(0);
891aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca            return bld->undef;
901aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca         }
911aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca
921aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca         if(swap) {
931aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca            args[0] = b;
941aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca            args[1] = a;
951aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca         }
961aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca         else {
971aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca            args[0] = a;
981aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca            args[1] = b;
991aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca         }
1001aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca
1011aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca         args[2] = LLVMConstInt(LLVMInt8Type(), cc, 0);
1021aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca         res = lp_build_intrinsic(bld->builder,
1031aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca                                  "llvm.x86.sse.cmp.ps",
1041aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca                                  vec_type,
1051aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca                                  args, 3);
1061aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca         res = LLVMBuildBitCast(bld->builder, res, int_vec_type, "");
1071aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca         return res;
1081aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca      }
1091aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca   }
1101aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca#endif
1111aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca
1121aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca   if(type.floating) {
1131aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca      LLVMRealPredicate op;
1141aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca      switch(func) {
1151aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca      case PIPE_FUNC_NEVER:
1161aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca         op = LLVMRealPredicateFalse;
1171aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca         break;
1181aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca      case PIPE_FUNC_ALWAYS:
1191aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca         op = LLVMRealPredicateTrue;
1201aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca         break;
1211aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca      case PIPE_FUNC_EQUAL:
1221aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca         op = LLVMRealUEQ;
1231aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca         break;
1241aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca      case PIPE_FUNC_NOTEQUAL:
1251aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca         op = LLVMRealUNE;
1261aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca         break;
1271aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca      case PIPE_FUNC_LESS:
1281aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca         op = LLVMRealULT;
1291aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca         break;
1301aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca      case PIPE_FUNC_LEQUAL:
1311aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca         op = LLVMRealULE;
1321aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca         break;
1331aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca      case PIPE_FUNC_GREATER:
1341aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca         op = LLVMRealUGT;
1351aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca         break;
1361aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca      case PIPE_FUNC_GEQUAL:
1371aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca         op = LLVMRealUGE;
1381aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca         break;
1391aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca      default:
1401aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca         assert(0);
1411aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca         return bld->undef;
1421aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca      }
1431aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca      cond = LLVMBuildFCmp(bld->builder, op, a, b, "");
1441aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca   }
1451aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca   else {
1461aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca      LLVMIntPredicate op;
1471aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca      switch(func) {
1481aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca      case PIPE_FUNC_EQUAL:
1491aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca         op = LLVMIntEQ;
1501aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca         break;
1511aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca      case PIPE_FUNC_NOTEQUAL:
1521aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca         op = LLVMIntNE;
1531aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca         break;
1541aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca      case PIPE_FUNC_LESS:
1551aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca         op = type.sign ? LLVMIntSLT : LLVMIntULT;
1561aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca         break;
1571aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca      case PIPE_FUNC_LEQUAL:
1581aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca         op = type.sign ? LLVMIntSLE : LLVMIntULE;
1591aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca         break;
1601aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca      case PIPE_FUNC_GREATER:
1611aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca         op = type.sign ? LLVMIntSGT : LLVMIntUGT;
1621aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca         break;
1631aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca      case PIPE_FUNC_GEQUAL:
1641aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca         op = type.sign ? LLVMIntSGE : LLVMIntUGE;
1651aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca         break;
1661aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca      default:
1671aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca         assert(0);
1681aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca         return bld->undef;
1691aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca      }
1701aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca      cond = LLVMBuildICmp(bld->builder, op, a, b, "");
1711aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca   }
1721aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca
1731aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca   return LLVMBuildSelect(bld->builder, cond, ones, zeros, "");
1741aede69d3a8d288af11c2ef620b51e71c2ce89b2José Fonseca}
175