[PATCH] D35956: Don't try to fold gep when the idx is a vector

[Davide Italiano via Phabricator via llvm-commits]

davide created this revision.

The code in `ConstantFoldGetElementPtr()` assumes integer, and therefore it crashes trying to get the integer bidwith of a vector type (in this case `<4 x i32>`. I just changed the code to prevent the folding in case of vectors and I didn't bother to generalize as this doesn't seem to me something that really happens in practice, but I'm willing to change the patch if you think it's worth it.
This  is hard to trigger from `-instsimplify`  or `-instcombine` only as the second instruction is dead, so the test uses loop-unroll.


https://reviews.llvm.org/D35956

Files:
  lib/IR/ConstantFold.cpp
  test/Transforms/InstSimplify/pr33957.ll


Index: test/Transforms/InstSimplify/pr33957.ll
===================================================================
--- /dev/null
+++ test/Transforms/InstSimplify/pr33957.ll
@@ -0,0 +1,29 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt -loop-unroll -S %s | FileCheck %s
+
+%struct.bar = type { i32 }
+
+ at global = external constant [78 x %struct.bar], align 4
+
+define void @patatino(i32 %x) {
+; CHECK-LABEL: @patatino(
+; CHECK-NEXT:  bb:
+; CHECK-NEXT:    br i1 true, label [[BB1_PREHEADER:%.*]], label [[BB3:%.*]]
+; CHECK:       bb1.preheader:
+; CHECK-NEXT:    br label [[BB1:%.*]]
+; CHECK:       bb1:
+; CHECK-NEXT:    br label [[BB3]]
+; CHECK:       bb3:
+; CHECK-NEXT:    ret void
+;
+bb:
+  br i1 true, label %bb1, label %bb3
+
+bb1:
+  %tmp = getelementptr inbounds [78 x %struct.bar], [78 x %struct.bar]* @global, i32 0, <4 x i32> undef
+  %tmp2 = getelementptr inbounds %struct.bar, <4 x %struct.bar*> %tmp, i32 1
+  br i1 true, label %bb3, label %bb1
+
+bb3:
+  ret void
+}
Index: lib/IR/ConstantFold.cpp
===================================================================
--- lib/IR/ConstantFold.cpp
+++ lib/IR/ConstantFold.cpp
@@ -2103,9 +2103,12 @@
         PerformFold = true;
       else if (LastI.isSequential())
         if (ConstantInt *CI = dyn_cast<ConstantInt>(Idx0))
-          PerformFold =
-              !LastI.isBoundedSequential() ||
-              isIndexInRangeOfArrayType(LastI.getSequentialNumElements(), CI);
+          PerformFold = (!LastI.isBoundedSequential() ||
+                         isIndexInRangeOfArrayType(
+                             LastI.getSequentialNumElements(), CI)) &&
+                        !CE->getOperand(CE->getNumOperands() - 1)
+                             ->getType()
+                             ->isVectorTy();
 
       if (PerformFold) {
         SmallVector<Value*, 16> NewIndices;


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D35956.108527.patch
Type: text/x-patch
Size: 1899 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20170727/47602410/attachment.bin>