1 ; RUN: opt < %s -basicaa -memcpyopt -dse -S | FileCheck %s
3 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"
4 target triple = "i686-apple-darwin9"
6 define void @test1({ x86_fp80, x86_fp80 }* sret %agg.result, x86_fp80 %z.0, x86_fp80 %z.1) nounwind {
8 %tmp2 = alloca { x86_fp80, x86_fp80 } ; <{ x86_fp80, x86_fp80 }*> [#uses=1]
9 %memtmp = alloca { x86_fp80, x86_fp80 }, align 16 ; <{ x86_fp80, x86_fp80 }*> [#uses=2]
10 %tmp5 = fsub x86_fp80 0xK80000000000000000000, %z.1 ; <x86_fp80> [#uses=1]
11 call void @ccoshl( { x86_fp80, x86_fp80 }* sret %memtmp, x86_fp80 %tmp5, x86_fp80 %z.0 ) nounwind
12 %tmp219 = bitcast { x86_fp80, x86_fp80 }* %tmp2 to i8* ; <i8*> [#uses=2]
13 %memtmp20 = bitcast { x86_fp80, x86_fp80 }* %memtmp to i8* ; <i8*> [#uses=1]
14 call void @llvm.memcpy.i32( i8* %tmp219, i8* %memtmp20, i32 32, i32 16 )
15 %agg.result21 = bitcast { x86_fp80, x86_fp80 }* %agg.result to i8* ; <i8*> [#uses=1]
16 call void @llvm.memcpy.i32( i8* %agg.result21, i8* %tmp219, i32 32, i32 16 )
18 ; Check that one of the memcpy's are removed.
19 ;; FIXME: PR 8643 We should be able to eliminate the last memcpy here.
22 ; CHECK: call void @ccoshl
23 ; CHECK: call void @llvm.memcpy
24 ; CHECK-NOT: llvm.memcpy
29 declare void @ccoshl({ x86_fp80, x86_fp80 }* sret , x86_fp80, x86_fp80) nounwind
31 declare void @llvm.memcpy.i32(i8*, i8*, i32, i32) nounwind
34 ; The intermediate alloca and one of the memcpy's should be eliminated, the
35 ; other should be related with a memmove.
36 define void @test2(i8* %P, i8* %Q) nounwind {
37 %memtmp = alloca { x86_fp80, x86_fp80 }, align 16
38 %R = bitcast { x86_fp80, x86_fp80 }* %memtmp to i8*
39 call void @llvm.memcpy.i32( i8* %R, i8* %P, i32 32, i32 16 )
40 call void @llvm.memcpy.i32( i8* %Q, i8* %R, i32 32, i32 16 )
44 ; CHECK-NEXT: call void @llvm.memmove{{.*}}(i8* %Q, i8* %P
45 ; CHECK-NEXT: ret void
51 @x = external global { x86_fp80, x86_fp80 }
53 define void @test3({ x86_fp80, x86_fp80 }* noalias sret %agg.result) nounwind {
54 %x.0 = alloca { x86_fp80, x86_fp80 }
55 %x.01 = bitcast { x86_fp80, x86_fp80 }* %x.0 to i8*
56 call void @llvm.memcpy.i32( i8* %x.01, i8* bitcast ({ x86_fp80, x86_fp80 }* @x to i8*), i32 32, i32 16 )
57 %agg.result2 = bitcast { x86_fp80, x86_fp80 }* %agg.result to i8*
58 call void @llvm.memcpy.i32( i8* %agg.result2, i8* %x.01, i32 32, i32 16 )
61 ; CHECK-NEXT: %agg.result2 = bitcast
62 ; CHECK-NEXT: call void @llvm.memcpy
63 ; CHECK-NEXT: ret void
68 define void @test4(i8 *%P) {
69 %A = alloca {i32, i32}
70 %a = bitcast {i32, i32}* %A to i8*
71 call void @llvm.memcpy.p0i8.p0i8.i64(i8* %a, i8* %P, i64 8, i32 4, i1 false)
72 call void @test4a(i8* byval align 1 %a)
75 ; CHECK-NEXT: call void @test4a(
78 declare void @test4a(i8* byval align 1)
79 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* nocapture, i8* nocapture, i64, i32, i1) nounwind
81 %struct.S = type { i128, [4 x i8]}
83 @sS = external global %struct.S, align 16
85 declare void @test5a(%struct.S* byval align 16) nounwind ssp
87 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* nocapture, i8* nocapture, i64, i32, i1) nounwind
89 ; rdar://8713376 - This memcpy can't be eliminated.
90 define i32 @test5(i32 %x) nounwind ssp {
92 %y = alloca %struct.S, align 16
93 %tmp = bitcast %struct.S* %y to i8*
94 call void @llvm.memcpy.p0i8.p0i8.i64(i8* %tmp, i8* bitcast (%struct.S* @sS to i8*), i64 32, i32 16, i1 false)
95 %a = getelementptr %struct.S* %y, i64 0, i32 1, i64 0
97 call void @test5a(%struct.S* byval align 16 %y)
101 ; CHECK: call void @test5a(%struct.S* byval align 16 %y)
104 ;; Noop memcpy should be zapped.
105 define void @test6(i8 *%P) {
106 call void @llvm.memcpy.p0i8.p0i8.i64(i8* %P, i8* %P, i64 8, i32 4, i1 false)
109 ; CHECK-NEXT: ret void