|  | ; RUN: opt < %s -basicaa -memcpyopt -S | FileCheck %s | 
|  | ; PR10067 | 
|  | ; Make sure the call+copy isn't optimized in such a way that | 
|  | ; %ret ends up with the wrong value. | 
|  |  | 
|  | 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-n8:16:32" | 
|  | target triple = "i386-apple-darwin10" | 
|  |  | 
|  | %struct1 = type { i32, i32 } | 
|  | %struct2 = type { %struct1, i8* } | 
|  |  | 
|  | declare void @bar(%struct1* nocapture sret %agg.result) nounwind | 
|  |  | 
|  | define i32 @foo() nounwind { | 
|  | %x = alloca %struct1, align 8 | 
|  | %y = alloca %struct2, align 8 | 
|  | call void @bar(%struct1* sret %x) nounwind | 
|  | ; CHECK: call void @bar(%struct1* sret %x) | 
|  |  | 
|  | %gepn1 = getelementptr inbounds %struct2, %struct2* %y, i32 0, i32 0, i32 0 | 
|  | store i32 0, i32* %gepn1, align 8 | 
|  | %gepn2 = getelementptr inbounds %struct2, %struct2* %y, i32 0, i32 0, i32 1 | 
|  | store i32 0, i32* %gepn2, align 4 | 
|  |  | 
|  | %bit1 = bitcast %struct1* %x to i64* | 
|  | %bit2 = bitcast %struct2* %y to i64* | 
|  | %load = load i64, i64* %bit1, align 8 | 
|  | store i64 %load, i64* %bit2, align 8 | 
|  |  | 
|  | ; CHECK: %load = load i64, i64* %bit1, align 8 | 
|  | ; CHECK: store i64 %load, i64* %bit2, align 8 | 
|  |  | 
|  | %gep1 = getelementptr %struct2, %struct2* %y, i32 0, i32 0, i32 0 | 
|  | %ret = load i32, i32* %gep1 | 
|  | ret i32 %ret | 
|  | } |