[InstSimplify] Fold insertelement vec, poison, idx into vec

[aqjune]

Summary

This is a simple patch that adds folding from insertelement vec, poison, idx into vec.

Alive2 proof: https://alive2.llvm.org/ce/z/2y2vbC

[nikic]

llvm/lib/Analysis/InstructionSimplify.cpp

4407

You can also switch this to use isGuaranteedNotToBePoison() to be more precise.

[aqjune]

Address comment

[aqjune]

llvm/lib/Analysis/InstructionSimplify.cpp

4407

Made D94053 to make the function distinguish constant vector with undef elements but not poison. :)

[nikic]

It looks like there are some test failures.

[aqjune]

Oh yes, I'll look into them

[aqjune]

Resolve failures
A few of those was regressions. It was because ConstantVector::get(Arr) was unable to understand Poison array. I updated those.

If the diffs look okay, I'll split this into two commits (one with the ConstantVector and another one with the InstSimplify) when landing.

[nikic]

LGTM. Nice catch on the ConstantVector issue.

[aqjune]

llvm/test/Transforms/SLPVectorizer/X86/alternate-int.ll

428

I think something is happening here, I imagine some transformation is passing undef to SimplifyInsertElementInst instead of poison (because sdiv _, undef is simplified to poison) I will look into this later.

llvm/lib/Analysis/InstructionSimplify.cpp

if (isa<FixedVectorType>(Vec->getType()) &&
         CI->uge(cast<FixedVectorType>(Vec->getType())->getNumElements()))
       return PoisonValue::get(Vec->getType());
   }
 
   // If index is undef, it might be out of bounds (see above case)
   if (Q.isUndefValue(Idx))
     return PoisonValue::get(Vec->getType());
 
-  // If the scalar is undef, and there is no risk of propagating poison from the
-  // vector value, simplify to the vector value.
-  if (Q.isUndefValue(Val) &&
-      isGuaranteedNotToBeUndefOrPoison(Vec))
+  // If the scalar is poison, or it is undef and there is no risk of
+  // propagating poison from the vector value, simplify to the vector value.
+  if (isa<PoisonValue>(Val) ||
+      (Q.isUndefValue(Val) && isGuaranteedNotToBePoison(Vec)))
     return Vec;
 
   // If we are extracting a value from a vector, then inserting it into the same
   // place, that's the input vector:
   // insertelt Vec, (extractelt Vec, Idx), Idx --> Vec
   if (match(Val, m_ExtractElt(m_Specific(Vec), m_Specific(Idx))))
     return Vec;
 

llvm/test/Transforms/InstCombine/extractelement-inseltpoison.ll

 ; does not consider the possibility of invalid insert/extract operands.
 
 define <4 x double> @invalid_extractelement(<2 x double> %a, <4 x double> %b, double* %p) {
 ; ANY-LABEL: @invalid_extractelement(
 ; ANY-NEXT:    [[TMP1:%.*]] = shufflevector <2 x double> [[A:%.*]], <2 x double> undef, <4 x i32> <i32 0, i32 undef, i32 undef, i32 undef>
 ; ANY-NEXT:    [[T4:%.*]] = shufflevector <4 x double> [[B:%.*]], <4 x double> [[TMP1]], <4 x i32> <i32 undef, i32 1, i32 4, i32 3>
 ; ANY-NEXT:    [[E:%.*]] = extractelement <4 x double> [[B]], i32 1
 ; ANY-NEXT:    store double [[E]], double* [[P:%.*]], align 8
-; ANY-NEXT:    [[R:%.*]] = insertelement <4 x double> [[T4]], double poison, i64 0
-; ANY-NEXT:    ret <4 x double> [[R]]
+; ANY-NEXT:    ret <4 x double> [[T4]]
 ;
   %t3 = extractelement <2 x double> %a, i32 0
   %t4 = insertelement <4 x double> %b, double %t3, i32 2
   %e = extractelement <4 x double> %t4, i32 1
   store double %e, double* %p
   %e1 = extractelement <2 x double> %a, i32 4 ; invalid index
   %r = insertelement <4 x double> %t4, double %e1, i64 0
   ret <4 x double> %r
 }

llvm/test/Transforms/InstCombine/extractelement.ll

 ; does not consider the possibility of invalid insert/extract operands.
 
 define <4 x double> @invalid_extractelement(<2 x double> %a, <4 x double> %b, double* %p) {
 ; ANY-LABEL: @invalid_extractelement(
 ; ANY-NEXT:    [[TMP1:%.*]] = shufflevector <2 x double> [[A:%.*]], <2 x double> undef, <4 x i32> <i32 0, i32 undef, i32 undef, i32 undef>
 ; ANY-NEXT:    [[T4:%.*]] = shufflevector <4 x double> [[B:%.*]], <4 x double> [[TMP1]], <4 x i32> <i32 undef, i32 1, i32 4, i32 3>
 ; ANY-NEXT:    [[E:%.*]] = extractelement <4 x double> [[B]], i32 1
 ; ANY-NEXT:    store double [[E]], double* [[P:%.*]], align 8
-; ANY-NEXT:    [[R:%.*]] = insertelement <4 x double> [[T4]], double poison, i64 0
-; ANY-NEXT:    ret <4 x double> [[R]]
+; ANY-NEXT:    ret <4 x double> [[T4]]
 ;
   %t3 = extractelement <2 x double> %a, i32 0
   %t4 = insertelement <4 x double> %b, double %t3, i32 2
   %e = extractelement <4 x double> %t4, i32 1
   store double %e, double* %p
   %e1 = extractelement <2 x double> %a, i32 4 ; invalid index
   %r = insertelement <4 x double> %t4, double %e1, i64 0
   ret <4 x double> %r
 }

llvm/test/Transforms/InstSimplify/insertelement.ll

 ; CHECK-NEXT:    ret <4 x i32> poison
 ;
   %I = insertelement <4 x i32> %A, i32 5, i64 poison
   ret <4 x i32> %I
 }
 
 define <4 x i32> @elem_poison(<4 x i32> %A) {
 ; CHECK-LABEL: @elem_poison(
-; CHECK-NEXT:    [[B:%.*]] = insertelement <4 x i32> [[A:%.*]], i32 poison, i32 1
-; CHECK-NEXT:    ret <4 x i32> [[B]]
+; CHECK-NEXT:    ret <4 x i32> [[A:%.*]]
 ;
   %B = insertelement <4 x i32> %A, i32 poison, i32 1
   ret <4 x i32> %B
 }
 
 ; The undef may be replacing a poison value, so it is not safe to just return 'A'.
 
 define <4 x i32> @PR1286(<4 x i32> %A) {

llvm/test/Transforms/SLPVectorizer/X86/alternate-int.ll

 ; CHECK-NEXT:    [[A6:%.*]] = extractelement <8 x i32> [[A]], i32 6
 ; CHECK-NEXT:    [[A7:%.*]] = extractelement <8 x i32> [[A]], i32 7
 ; CHECK-NEXT:    [[AB1:%.*]] = sdiv i32 [[A1]], 4
 ; CHECK-NEXT:    [[AB2:%.*]] = sdiv i32 [[A2]], 8
 ; CHECK-NEXT:    [[AB3:%.*]] = sdiv i32 [[A3]], 16
 ; CHECK-NEXT:    [[AB5:%.*]] = sdiv i32 [[A5]], 4
 ; CHECK-NEXT:    [[AB6:%.*]] = sdiv i32 [[A6]], 8
 ; CHECK-NEXT:    [[AB7:%.*]] = sdiv i32 [[A7]], 16
-; CHECK-NEXT:    [[TMP1:%.*]] = insertelement <8 x i32> poison, i32 [[AB1]], i32 1
-; CHECK-NEXT:    [[TMP2:%.*]] = insertelement <8 x i32> [[TMP1]], i32 [[AB2]], i32 2
-; CHECK-NEXT:    [[R4:%.*]] = insertelement <8 x i32> [[TMP2]], i32 [[AB3]], i32 3
-; CHECK-NEXT:    [[R5:%.*]] = insertelement <8 x i32> [[R4]], i32 [[AB5]], i32 5
+; CHECK-NEXT:    [[R1:%.*]] = insertelement <8 x i32> <i32 poison, i32 poison, i32 poison, i32 poison, i32 undef, i32 poison, i32 poison, i32 poison>, i32 [[AB1]], i32 1
+; CHECK-NEXT:    [[R2:%.*]] = insertelement <8 x i32> [[R1]], i32 [[AB2]], i32 2
+; CHECK-NEXT:    [[R3:%.*]] = insertelement <8 x i32> [[R2]], i32 [[AB3]], i32 3
+; CHECK-NEXT:    [[R5:%.*]] = insertelement <8 x i32> [[R3]], i32 [[AB5]], i32 5
 ; CHECK-NEXT:    [[R6:%.*]] = insertelement <8 x i32> [[R5]], i32 [[AB6]], i32 6
 ; CHECK-NEXT:    [[R7:%.*]] = insertelement <8 x i32> [[R6]], i32 [[AB7]], i32 7
 ; CHECK-NEXT:    ret <8 x i32> [[R7]]
 ;
   %a0 = extractelement <8 x i32> %a, i32 0
   %a1 = extractelement <8 x i32> %a, i32 1
   %a2 = extractelement <8 x i32> %a, i32 2
   %a3 = extractelement <8 x i32> %a, i32 3