[llvm] r316430 - [ConstantFolding] Avoid assert when folding ptrtoint of vectorized GEP

[Bjorn Pettersson via llvm-commits]

Author: bjope
Date: Tue Oct 24 05:08:11 2017
New Revision: 316430

URL: http://llvm.org/viewvc/llvm-project?rev=316430&view=rev
Log:
[ConstantFolding] Avoid assert when folding ptrtoint of vectorized GEP

Summary:
Got asserts in llvm::CastInst::getCastOpcode saying:
`DestBits == SrcBits && "Illegal cast to vector (wrong type or size)"' failed.

Problem seemed to be that llvm::ConstantFoldCastInstruction did
not handle ptrtoint cast of a getelementptr returning a vector
correctly. I assume such situations are quite rare, since the
GEP needs to be considered as a constant value (base pointer
being null).
The solution used here is to simply avoid the constant fold
of ptrtoint when the value is a vector. It is not supported,
and by bailing out we do not fail on assertions later on.

Reviewers: craig.topper, majnemer, davide, filcab, efriedma

Reviewed By: efriedma

Subscribers: efriedma, filcab, llvm-commits

Differential Revision: https://reviews.llvm.org/D38546

Added:
    llvm/trunk/test/Analysis/ConstantFolding/cast-vector.ll
Modified:
    llvm/trunk/lib/IR/ConstantFold.cpp

Modified: llvm/trunk/lib/IR/ConstantFold.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/IR/ConstantFold.cpp?rev=316430&r1=316429&r2=316430&view=diff
==============================================================================
--- llvm/trunk/lib/IR/ConstantFold.cpp (original)
+++ llvm/trunk/lib/IR/ConstantFold.cpp Tue Oct 24 05:08:11 2017
@@ -629,6 +629,15 @@ Constant *llvm::ConstantFoldCastInstruct
     if (ConstantExpr *CE = dyn_cast<ConstantExpr>(V))
       if (CE->getOpcode() == Instruction::GetElementPtr &&
           CE->getOperand(0)->isNullValue()) {
+        // FIXME: Looks like getFoldedSizeOf(), getFoldedOffsetOf() and
+        // getFoldedAlignOf() don't handle the case when DestTy is a vector of
+        // pointers yet. We end up in asserts in CastInst::getCastOpcode (see
+        // test/Analysis/ConstantFolding/cast-vector.ll). I've only seen this
+        // happen in one "real" C-code test case, so it does not seem to be an
+        // important optimization to handle vectors here. For now, simply bail
+        // out.
+        if (DestTy->isVectorTy())
+          return nullptr;
         GEPOperator *GEPO = cast<GEPOperator>(CE);
         Type *Ty = GEPO->getSourceElementType();
         if (CE->getNumOperands() == 2) {

Added: llvm/trunk/test/Analysis/ConstantFolding/cast-vector.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Analysis/ConstantFolding/cast-vector.ll?rev=316430&view=auto
==============================================================================
--- llvm/trunk/test/Analysis/ConstantFolding/cast-vector.ll (added)
+++ llvm/trunk/test/Analysis/ConstantFolding/cast-vector.ll Tue Oct 24 05:08:11 2017
@@ -0,0 +1,32 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt < %s -instsimplify -S | FileCheck %s
+
+; Test constant fold of constant expression GEP used by ptrtoint (the
+; "offsetof-like expression" case).
+; This used to hit an assert due to not supporting vectors in
+; llvm::ConstantFoldCastInstruction when handling ptrtoint.
+define <2 x i16> @test1() {
+; CHECK-LABEL: @test1(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    ret <2 x i16> ptrtoint (<2 x i32*> getelementptr ([10 x i32], [10 x i32]* null, <2 x i64> zeroinitializer, <2 x i64> <i64 5, i64 7>) to <2 x i16>)
+;
+entry:
+  %gep = getelementptr inbounds [10 x i32], [10 x i32]* null, i16 0, <2 x i16> <i16 5, i16 7>
+  %vec = ptrtoint <2 x i32*> %gep to <2 x i16>
+  ret <2 x i16> %vec
+}
+
+; Test constant fold of constant expression GEP used by ptrtoint (the
+; "sizeof-like expression" case).
+; This used to hit an assert due to not supporting vectors in
+; llvm::ConstantFoldCastInstruction when handling ptrtoint.
+define <2 x i16> @test2() {
+; CHECK-LABEL: @test2(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    ret <2 x i16> ptrtoint (<2 x i32*> getelementptr (i32, i32* null, <2 x i64> <i64 5, i64 7>) to <2 x i16>)
+;
+entry:
+  %gep = getelementptr i32, i32* null, <2 x i16> <i16 5, i16 7>
+  %vec = ptrtoint <2 x i32*> %gep to <2 x i16>
+  ret <2 x i16> %vec
+}