[llvm-bugs] [Bug 27236] New: Instcombine tries to create invalid IR

[via llvm-bugs]

https://llvm.org/bugs/show_bug.cgi?id=27236

            Bug ID: 27236
           Summary: Instcombine tries to create invalid IR
           Product: libraries
           Version: trunk
          Hardware: PC
                OS: All
            Status: NEW
          Severity: normal
          Priority: P
         Component: Scalar Optimizations
          Assignee: unassignedbugs at nondot.org
          Reporter: anton at korobeynikov.info
                CC: llvm-bugs at lists.llvm.org
    Classification: Unclassified

Created attachment 16174
  --> https://llvm.org/bugs/attachment.cgi?id=16174&action=edit
Testcase

Consider the attached IR.

opt -instcombine -debug yields:

IC: Replacing   %.sroa.4.0 = select i1 %4, float %3, float 0.000000e+00
    with   %2 = select i1 %1, i32 1, i32 %0
Assertion failed: (New->getType() == getType() && "replaceAllUses of value with
new value of different type!"), function replaceAllUsesWith, file
/Users/asl/Projects/llvm/2commit/llvm/lib/IR/Value.cpp, line 375.

Which is certainly incorrect. I believe the problem was exposed by the recent
changes in min/max optimizations and min(min) / max(max) folding.

The "big hammer" fix is to have something like this:

Index: lib/Transforms/InstCombine/InstCombineSelect.cpp
===================================================================
--- lib/Transforms/InstCombine/InstCombineSelect.cpp    (revision 265487)
+++ lib/Transforms/InstCombine/InstCombineSelect.cpp    (working copy)
@@ -642,6 +642,9 @@
                                         Value *A, Value *B,
                                         Instruction &Outer,
                                         SelectPatternFlavor SPF2, Value *C) {
+  if (Outer.getType() != Inner->getType())
+    return nullptr;
+
   if (C == A || C == B) {
     // MAX(MAX(A, B), B) -> MAX(A, B)
     // MIN(MIN(a, b), a) -> MIN(a, b)

However, I believe we need to make sure we never call stuff this way

-- 
You are receiving this mail because:
You are on the CC list for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-bugs/attachments/20160406/1ab8a188/attachment.html>