Fix PR4748: don't fold gep(bitcast(x)) into bitcast(gep) when x

is itself a bitcast.  Since we have gep(bitcast(bitcast(y))) in this
case, just wait for the two bitcasts to get zapped.  This prevents
instcombine from confusing some aliasing stuff, and allows it to
directly eliminate the load in the testcase.

llvm-svn: 80508

1 files changed, 21 insertions, 0 deletions

diff --git a/llvm/test/Transforms/InstCombine/getelementptr-2.ll b/llvm/test/Transforms/InstCombine/getelementptr-2.ll
new file mode 100644
index 00000000000..4cffbcd4bcd
--- /dev/null
+++ b/llvm/test/Transforms/InstCombine/getelementptr-2.ll
@@ -0,0 +1,21 @@
+; RUN: llvm-as < %s | opt -instcombine | llvm-dis | not grep load
+; PR4748
+target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:64:64-v128:128:128-a0:0:64-f80:128:128"
+target triple = "i386-apple-darwin9"
+	%struct.B = type { double }
+	%struct.A = type { %struct.B, i32, i32 }
+
+define i32 @_Z4funcv(%struct.A* %a) {
+entry:
+  %g3 = getelementptr %struct.A* %a, i32 0, i32 1
+  store i32 10, i32* %g3, align 4
+
+  %g4 = getelementptr %struct.A* %a, i32 0, i32 0
+  
+  %new_a = bitcast %struct.B* %g4 to %struct.A*
+
+  %g5 = getelementptr %struct.A* %new_a, i32 0, i32 1	
+  %a_a = load i32* %g5, align 4	
+  ret i32 %a_a
+}
+