[MergeFuncs] Don't merge shufflevectors with different masks

[nikic]

Summary

When the shufflevector mask operand was converted into special instruction data, the FunctionComparator was not updated to account for this. As such, MergeFuncs will happily merge shufflevectors with different masks.

This fixes https://bugs.llvm.org/show_bug.cgi?id=45773.

[efriedma]

LGTM

I thought I caught all the places that needed new code to compare masks, but I missed this; thanks.

llvm/lib/Transforms/Utils/FunctionComparator.cpp

if (int Res = cmpNumbers(RMWI->isVolatile(),
                              cast<AtomicRMWInst>(R)->isVolatile()))
       return Res;
     if (int Res = cmpOrderings(RMWI->getOrdering(),
                                cast<AtomicRMWInst>(R)->getOrdering()))
       return Res;
     return cmpNumbers(RMWI->getSyncScopeID(),
                       cast<AtomicRMWInst>(R)->getSyncScopeID());
   }
+  if (const ShuffleVectorInst *SVI = dyn_cast<ShuffleVectorInst>(L)) {
+    ArrayRef<int> LMask = SVI->getShuffleMask();
+    ArrayRef<int> RMask = cast<ShuffleVectorInst>(R)->getShuffleMask();
+    if (int Res = cmpNumbers(LMask.size(), RMask.size()))
+      return Res;
+    for (size_t i = 0, e = LMask.size(); i != e; ++i) {
+      if (int Res = cmpNumbers(LMask[i], RMask[i]))
+        return Res;
+    }
+  }
   if (const PHINode *PNL = dyn_cast<PHINode>(L)) {
     const PHINode *PNR = cast<PHINode>(R);
     // Ensure that in addition to the incoming values being identical
     // (checked by the caller of this function), the incoming blocks
     // are also identical.
     for (unsigned i = 0, e = PNL->getNumIncomingValues(); i != e; ++i) {
       if (int Res =
               cmpValues(PNL->getIncomingBlock(i), PNR->getIncomingBlock(i)))

llvm/test/Transforms/MergeFunc/shufflevector.ll

+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt -S -mergefunc < %s | FileCheck %s
+
+define internal <2 x i32> @test1(<2 x i32> %v1, <2 x i32> %v2) {
+; CHECK-LABEL: @test1(
+; CHECK-NEXT:    [[X:%.*]] = shufflevector <2 x i32> [[V1:%.*]], <2 x i32> [[V2:%.*]], <2 x i32> <i32 0, i32 1>
+; CHECK-NEXT:    ret <2 x i32> [[X]]
+;
+  %x = shufflevector <2 x i32> %v1, <2 x i32> %v2, <2 x i32> <i32 0, i32 1>
+  ret <2 x i32> %x
+}
+
+; Same mask as test1.
+define internal <2 x i32> @test2(<2 x i32> %v1, <2 x i32> %v2) {
+  %x = shufflevector <2 x i32> %v1, <2 x i32> %v2, <2 x i32> <i32 0, i32 1>
+  ret <2 x i32> %x
+}
+
+; Different mask than test1, don't merge.
+define internal <2 x i32> @test3(<2 x i32> %v1, <2 x i32> %v2) {
+; CHECK-LABEL: @test3(
+; CHECK-NEXT:    [[X:%.*]] = shufflevector <2 x i32> [[V1:%.*]], <2 x i32> [[V2:%.*]], <2 x i32> <i32 1, i32 0>
+; CHECK-NEXT:    ret <2 x i32> [[X]]
+;
+  %x = shufflevector <2 x i32> %v1, <2 x i32> %v2, <2 x i32> <i32 1, i32 0>
+  ret <2 x i32> %x
+}
+
+define void @caller(<2 x i32> %v1, <2 x i32> %v2) {
+; CHECK-LABEL: @caller(
+; CHECK-NEXT:    [[TMP1:%.*]] = call <2 x i32> @test1(<2 x i32> [[V1:%.*]], <2 x i32> [[V2:%.*]])
+; CHECK-NEXT:    [[TMP2:%.*]] = call <2 x i32> @test1(<2 x i32> [[V1]], <2 x i32> [[V2]])
+; CHECK-NEXT:    [[TMP3:%.*]] = call <2 x i32> @test3(<2 x i32> [[V1]], <2 x i32> [[V2]])
+; CHECK-NEXT:    ret void
+;
+  call <2 x i32> @test1(<2 x i32> %v1, <2 x i32> %v2)
+  call <2 x i32> @test2(<2 x i32> %v1, <2 x i32> %v2)
+  call <2 x i32> @test3(<2 x i32> %v1, <2 x i32> %v2)
+  ret void
+}