[SLP]Do not account external uses in EH block and in non-returning blocks

No need to account the cost of the external uses in EH and non-returning
basic blocks.

Reviewers: RKSimon

Reviewed By: RKSimon

Pull Request: llvm#112045

Author: alexey-bataev

llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp

@@ -11977,11 +11977,13 @@ InstructionCost BoUpSLP::getTreeCost(ArrayRef<Value *> VectorizedVals) {
     if (EphValues.count(EU.User))
       continue;
 
-    // Used in unreachable blocks or in landing pads (rarely executed).
+    // Used in unreachable blocks or in EH pads (rarely executed) or is
+    // terminated with unreachable instruction.
     if (BasicBlock *UserParent =
             EU.User ? cast<Instruction>(EU.User)->getParent() : nullptr;
         UserParent &&
-        (!DT->isReachableFromEntry(UserParent) || UserParent->isLandingPad()))
+        (!DT->isReachableFromEntry(UserParent) || UserParent->isEHPad() ||
+         isa_and_present<UnreachableInst>(UserParent->getTerminator())))
       continue;
 
     // We only add extract cost once for the same scalar.

llvm/test/Transforms/SLPVectorizer/X86/pr27163.ll

@@ -9,7 +9,7 @@ define void @test1(ptr %p) personality ptr @__CxxFrameHandler3 {
 ; CHECK-LABEL: @test1(
 ; CHECK-NEXT:  invoke.cont:
 ; CHECK-NEXT:    [[TMP0:%.*]] = load <2 x i64>, ptr [[P:%.*]], align 8
-; CHECK-NEXT:    [[LOAD1:%.*]] = load i64, ptr [[P]], align 8
+; CHECK-NEXT:    [[LOAD1:%.*]] = extractelement <2 x i64> [[TMP0]], i32 0
 ; CHECK-NEXT:    store <2 x i64> [[TMP0]], ptr [[P]], align 8
 ; CHECK-NEXT:    invoke void @throw()
 ; CHECK-NEXT:            to label [[UNREACHABLE:%.*]] unwind label [[CATCH_DISPATCH:%.*]]

llvm/test/Transforms/SLPVectorizer/X86/reorder_repeated_ops.ll

@@ -11,41 +11,23 @@ define void @hoge() {
 ; CHECK-NEXT:    ret void
 ; CHECK:       bb2:
 ; CHECK-NEXT:    [[T:%.*]] = select i1 undef, i16 undef, i16 15
-; CHECK-NEXT:    [[T3:%.*]] = sext i16 undef to i32
-; CHECK-NEXT:    [[T4:%.*]] = sext i16 [[T]] to i32
-; CHECK-NEXT:    [[T5:%.*]] = sub nsw i32 undef, [[T4]]
-; CHECK-NEXT:    [[T6:%.*]] = sub i32 [[T5]], undef
-; CHECK-NEXT:    [[T7:%.*]] = sub nsw i32 63, [[T3]]
-; CHECK-NEXT:    [[T8:%.*]] = sub i32 [[T7]], undef
-; CHECK-NEXT:    [[T9:%.*]] = add i32 [[T8]], undef
-; CHECK-NEXT:    [[T10:%.*]] = add nsw i32 [[T6]], 15
-; CHECK-NEXT:    [[T11:%.*]] = icmp sgt i32 [[T9]], [[T10]]
-; CHECK-NEXT:    [[T12:%.*]] = select i1 [[T11]], i32 [[T9]], i32 [[T10]]
-; CHECK-NEXT:    [[T13:%.*]] = add nsw i32 [[T6]], 31
-; CHECK-NEXT:    [[T14:%.*]] = icmp sgt i32 [[T12]], [[T13]]
-; CHECK-NEXT:    [[T15:%.*]] = select i1 [[T14]], i32 [[T12]], i32 [[T13]]
-; CHECK-NEXT:    [[T16:%.*]] = add nsw i32 [[T6]], 47
-; CHECK-NEXT:    [[T17:%.*]] = icmp sgt i32 [[T15]], [[T16]]
-; CHECK-NEXT:    [[T18:%.*]] = select i1 [[T17]], i32 [[T15]], i32 [[T16]]
+; CHECK-NEXT:    [[TMP0:%.*]] = insertelement <2 x i16> <i16 poison, i16 undef>, i16 [[T]], i32 0
+; CHECK-NEXT:    [[TMP1:%.*]] = sext <2 x i16> [[TMP0]] to <2 x i32>
+; CHECK-NEXT:    [[TMP2:%.*]] = sub nsw <2 x i32> <i32 undef, i32 63>, [[TMP1]]
+; CHECK-NEXT:    [[TMP3:%.*]] = sub <2 x i32> [[TMP2]], undef
+; CHECK-NEXT:    [[TMP4:%.*]] = shufflevector <2 x i32> [[TMP3]], <2 x i32> poison, <4 x i32> <i32 1, i32 0, i32 0, i32 0>
+; CHECK-NEXT:    [[TMP5:%.*]] = add <4 x i32> [[TMP4]], <i32 undef, i32 15, i32 31, i32 47>
+; CHECK-NEXT:    [[T18:%.*]] = call i32 @llvm.vector.reduce.smax.v4i32(<4 x i32> [[TMP5]])
 ; CHECK-NEXT:    [[T19:%.*]] = select i1 undef, i32 [[T18]], i32 undef
 ; CHECK-NEXT:    [[T20:%.*]] = icmp sgt i32 [[T19]], 63
-; CHECK-NEXT:    [[T21:%.*]] = sub nsw i32 undef, [[T3]]
-; CHECK-NEXT:    [[T22:%.*]] = sub i32 [[T21]], undef
-; CHECK-NEXT:    [[T23:%.*]] = sub nsw i32 undef, [[T4]]
-; CHECK-NEXT:    [[T24:%.*]] = sub i32 [[T23]], undef
-; CHECK-NEXT:    [[T25:%.*]] = add nsw i32 [[T24]], -49
-; CHECK-NEXT:    [[T30:%.*]] = add nsw i32 [[T22]], -33
-; CHECK-NEXT:    [[T35:%.*]] = add nsw i32 [[T24]], -33
-; CHECK-NEXT:    [[T40:%.*]] = add nsw i32 [[T22]], -17
+; CHECK-NEXT:    [[TMP7:%.*]] = sub nsw <2 x i32> undef, [[TMP1]]
+; CHECK-NEXT:    [[TMP8:%.*]] = sub <2 x i32> [[TMP7]], undef
+; CHECK-NEXT:    [[TMP9:%.*]] = shufflevector <2 x i32> [[TMP8]], <2 x i32> poison, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
+; CHECK-NEXT:    [[TMP10:%.*]] = add nsw <4 x i32> [[TMP9]], <i32 -49, i32 -33, i32 -33, i32 -17>
+; CHECK-NEXT:    [[T25:%.*]] = call i32 @llvm.vector.reduce.smin.v4i32(<4 x i32> [[TMP10]])
 ; CHECK-NEXT:    [[OP_RDX:%.*]] = icmp slt i32 undef, [[T25]]
 ; CHECK-NEXT:    [[OP_RDX1:%.*]] = select i1 [[OP_RDX]], i32 undef, i32 [[T25]]
-; CHECK-NEXT:    [[OP_RDX2:%.*]] = icmp slt i32 [[T30]], [[T35]]
-; CHECK-NEXT:    [[OP_RDX3:%.*]] = select i1 [[OP_RDX2]], i32 [[T30]], i32 [[T35]]
-; CHECK-NEXT:    [[OP_RDX4:%.*]] = icmp slt i32 [[OP_RDX1]], [[OP_RDX3]]
-; CHECK-NEXT:    [[OP_RDX5:%.*]] = select i1 [[OP_RDX4]], i32 [[OP_RDX1]], i32 [[OP_RDX3]]
-; CHECK-NEXT:    [[OP_RDX6:%.*]] = icmp slt i32 [[OP_RDX5]], [[T40]]
-; CHECK-NEXT:    [[OP_RDX7:%.*]] = select i1 [[OP_RDX6]], i32 [[OP_RDX5]], i32 [[T40]]
-; CHECK-NEXT:    [[T45:%.*]] = icmp sgt i32 undef, [[OP_RDX7]]
+; CHECK-NEXT:    [[T45:%.*]] = icmp sgt i32 undef, [[OP_RDX1]]
 ; CHECK-NEXT:    unreachable
 ;
 bb: