[ValueTracking] matchSelectPattern needs to be more careful around FP

matchSelectPattern attempts to see through casts which mask min/max
patterns from being more obvious.  Under certain circumstances, it would
misidentify a sequence of instructions as a min/max because it assumed
that folding casts would preserve the result.  This is not the case for
floating point <-> integer casts.

This fixes PR27575.

llvm-svn: 268086

1 files changed, 26 insertions, 0 deletions

diff --git a/llvm/test/Transforms/InstCombine/minmax-fp.ll b/llvm/test/Transforms/InstCombine/minmax-fp.ll
index b90afe3405f..b6eb1bb6834 100644
--- a/llvm/test/Transforms/InstCombine/minmax-fp.ll
+++ b/llvm/test/Transforms/InstCombine/minmax-fp.ll
@@ -154,3 +154,29 @@ define i8 @t15(float %a) {
   %3 = select i1 %1, i8 %2, i8 0
   ret i8 %3
 }
+
+; CHECK-LABEL: @t16
+; CHECK:  %[[cmp:.*]] = icmp sgt i32 %x, 0
+; CHECK:  %[[cst:.*]] = sitofp i32 %x to double
+; CHECK:  %[[sel:.*]] = select i1 %[[cmp]], double %[[cst]], double 5.000000e-01
+; CHECK:  ret double %[[sel]]
+define double @t16(i32 %x) {
+entry:
+  %cmp = icmp sgt i32 %x, 0
+  %cst = sitofp i32 %x to double
+  %sel = select i1 %cmp, double %cst, double 5.000000e-01
+  ret double %sel
+}
+
+; CHECK-LABEL: @t17
+; CHECK:  %[[cmp:.*]] = icmp sgt i32 %x, 2
+; CHECK:  %[[sel:.*]] = select i1 %[[cmp]], i32 %x, i32 2
+; CHECK:  %[[cst:.*]] = sitofp i32 %[[sel]] to double
+; CHECK:  ret double %[[cst]]
+define double @t17(i32 %x) {
+entry:
+  %cmp = icmp sgt i32 %x, 2
+  %cst = sitofp i32 %x to double
+  %sel = select i1 %cmp, double %cst, double 2.0
+  ret double %sel
+}