[PATCH] D119855: [SLP]Fix vectorization of the alternate cmp instruction with swapped predicates.

[Alexey Bataev via Phabricator via llvm-commits]

ABataev created this revision.
ABataev added reviewers: RKSimon, anton-afanasyev, dtemirbulatov.
Herald added a subscriber: hiraditya.
ABataev requested review of this revision.
Herald added a project: LLVM.

If the alternate cmp instruction is a swapped predicate of the main cmp
instruction, need to generate alternate instruction, not the one with
the swapped predicate. Also, the lane with the alternate opcode should
be selected only, if the corresponding operands are not compatible.

Correctness confirmed:
https://alive2.llvm.org/ce/z/94BG66


Repository:
  rG LLVM Github Monorepo

https://reviews.llvm.org/D119855

Files:
  llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
  llvm/test/Transforms/SLPVectorizer/X86/alternate-cmp-swapped-pred.ll


Index: llvm/test/Transforms/SLPVectorizer/X86/alternate-cmp-swapped-pred.ll
===================================================================
--- llvm/test/Transforms/SLPVectorizer/X86/alternate-cmp-swapped-pred.ll
+++ llvm/test/Transforms/SLPVectorizer/X86/alternate-cmp-swapped-pred.ll
@@ -9,10 +9,11 @@
 ; CHECK-NEXT:    [[TMP1:%.*]] = insertelement <8 x i16> <i16 0, i16 0, i16 0, i16 poison, i16 0, i16 0, i16 poison, i16 poison>, i16 [[CALL37:%.*]], i32 3
 ; CHECK-NEXT:    [[SHUFFLE:%.*]] = shufflevector <8 x i16> [[TMP1]], <8 x i16> poison, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 3, i32 4, i32 3, i32 5>
 ; CHECK-NEXT:    [[TMP2:%.*]] = icmp slt <8 x i16> [[TMP0]], [[SHUFFLE]]
-; CHECK-NEXT:    [[TMP3:%.*]] = shufflevector <8 x i1> [[TMP2]], <8 x i1> [[TMP2]], <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 12, i32 5, i32 6, i32 7>
-; CHECK-NEXT:    [[TMP4:%.*]] = zext <8 x i1> [[TMP3]] to <8 x i16>
-; CHECK-NEXT:    [[TMP5:%.*]] = call i16 @llvm.vector.reduce.add.v8i16(<8 x i16> [[TMP4]])
-; CHECK-NEXT:    [[OP_EXTRA:%.*]] = add i16 [[TMP5]], 0
+; CHECK-NEXT:    [[TMP3:%.*]] = icmp sgt <8 x i16> [[TMP0]], [[SHUFFLE]]
+; CHECK-NEXT:    [[TMP4:%.*]] = shufflevector <8 x i1> [[TMP2]], <8 x i1> [[TMP3]], <8 x i32> <i32 0, i32 1, i32 2, i32 11, i32 4, i32 5, i32 14, i32 7>
+; CHECK-NEXT:    [[TMP5:%.*]] = zext <8 x i1> [[TMP4]] to <8 x i16>
+; CHECK-NEXT:    [[TMP6:%.*]] = call i16 @llvm.vector.reduce.add.v8i16(<8 x i16> [[TMP5]])
+; CHECK-NEXT:    [[OP_EXTRA:%.*]] = add i16 [[TMP6]], 0
 ; CHECK-NEXT:    ret i16 [[OP_EXTRA]]
 ;
 entry:
Index: llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
===================================================================
--- llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
+++ llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
@@ -5571,14 +5571,9 @@
                   CmpInst::getSwappedPredicate(AltP0);
               CmpInst::Predicate CurrentPred = CI->getPredicate();
               if (P0 == AltP0Swapped)
-                return (P0 == CurrentPred &&
-                        !areCompatibleCmpOps(
-                            CI0->getOperand(0), CI0->getOperand(1),
-                            CI->getOperand(0), CI->getOperand(1))) ||
-                       (AltP0 == CurrentPred &&
-                        !areCompatibleCmpOps(
-                            CI0->getOperand(0), CI0->getOperand(1),
-                            CI->getOperand(1), CI->getOperand(0)));
+                return !areCompatibleCmpOps(
+                    CI0->getOperand(0), CI0->getOperand(1), CI->getOperand(0),
+                    CI->getOperand(1));
               return AltP0 == CurrentPred || AltP0Swapped == CurrentPred;
             }
             return I->getOpcode() == E->getAltOpcode();
@@ -7081,10 +7076,6 @@
         V0 = Builder.CreateCmp(CI0->getPredicate(), LHS, RHS);
         auto *AltCI = cast<CmpInst>(E->getAltOp());
         CmpInst::Predicate AltPred = AltCI->getPredicate();
-        unsigned AltIdx =
-            std::distance(E->Scalars.begin(), find(E->Scalars, AltCI));
-        if (AltCI->getOperand(0) != E->getOperand(0)[AltIdx])
-          AltPred = CmpInst::getSwappedPredicate(AltPred);
         V1 = Builder.CreateCmp(AltPred, LHS, RHS);
       } else {
         V0 = Builder.CreateCast(
@@ -7121,14 +7112,9 @@
                   CmpInst::getSwappedPredicate(AltP0);
               CmpInst::Predicate CurrentPred = CI->getPredicate();
               if (P0 == AltP0Swapped)
-                return (P0 == CurrentPred &&
-                        !areCompatibleCmpOps(
-                            CI0->getOperand(0), CI0->getOperand(1),
-                            CI->getOperand(0), CI->getOperand(1))) ||
-                       (AltP0 == CurrentPred &&
-                        !areCompatibleCmpOps(
-                            CI0->getOperand(0), CI0->getOperand(1),
-                            CI->getOperand(1), CI->getOperand(0)));
+                return !areCompatibleCmpOps(
+                    CI0->getOperand(0), CI0->getOperand(1), CI->getOperand(0),
+                    CI->getOperand(1));
               return AltP0 == CurrentPred || AltP0Swapped == CurrentPred;
             }
             return I->getOpcode() == E->getAltOpcode();


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D119855.408906.patch
Type: text/x-patch
Size: 4247 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20220215/a4787bfc/attachment.bin>