1 ; RUN: opt < %s -sroa -S | FileCheck %s
2 target datalayout = "E-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-n8:16:32:64"
4 declare void @llvm.memcpy.p0i8.p0i8.i32(i8*, i8*, i32, i32, i1)
6 define void @test1({ i8, i8 }* %a, { i8, i8 }* %b) {
8 ; CHECK: %[[gep_a0:.*]] = getelementptr inbounds { i8, i8 }* %a, i64 0, i32 0
9 ; CHECK: %[[a0:.*]] = load i8* %[[gep_a0]], align 16
10 ; CHECK: %[[gep_a1:.*]] = getelementptr inbounds { i8, i8 }* %a, i64 0, i32 1
11 ; CHECK: %[[a1:.*]] = load i8* %[[gep_a1]], align 1
12 ; CHECK: %[[gep_b0:.*]] = getelementptr inbounds { i8, i8 }* %b, i64 0, i32 0
13 ; CHECK: store i8 %[[a0]], i8* %[[gep_b0]], align 16
14 ; CHECK: %[[gep_b1:.*]] = getelementptr inbounds { i8, i8 }* %b, i64 0, i32 1
15 ; CHECK: store i8 %[[a1]], i8* %[[gep_b1]], align 1
19 %alloca = alloca { i8, i8 }, align 16
20 %gep_a = getelementptr { i8, i8 }* %a, i32 0, i32 0
21 %gep_alloca = getelementptr { i8, i8 }* %alloca, i32 0, i32 0
22 %gep_b = getelementptr { i8, i8 }* %b, i32 0, i32 0
24 store i8 420, i8* %gep_alloca, align 16
26 call void @llvm.memcpy.p0i8.p0i8.i32(i8* %gep_alloca, i8* %gep_a, i32 2, i32 16, i1 false)
27 call void @llvm.memcpy.p0i8.p0i8.i32(i8* %gep_b, i8* %gep_alloca, i32 2, i32 16, i1 false)
31 define void @test2() {
34 ; CHECK: load i8* %{{.*}}
35 ; CHECK: store i8 42, i8* %{{.*}}
39 %a = alloca { i8, i8, i8, i8 }, align 2
40 %gep1 = getelementptr { i8, i8, i8, i8 }* %a, i32 0, i32 1
41 %cast1 = bitcast i8* %gep1 to i16*
42 store volatile i16 0, i16* %cast1
43 %gep2 = getelementptr { i8, i8, i8, i8 }* %a, i32 0, i32 2
44 %result = load i8* %gep2
45 store i8 42, i8* %gep2
49 define void @PR13920(<2 x i64>* %a, i16* %b) {
50 ; Test that alignments on memcpy intrinsics get propagated to loads and stores.
52 ; CHECK: load <2 x i64>* %a, align 2
53 ; CHECK: store <2 x i64> {{.*}}, <2 x i64>* {{.*}}, align 2
57 %aa = alloca <2 x i64>, align 16
58 %aptr = bitcast <2 x i64>* %a to i8*
59 %aaptr = bitcast <2 x i64>* %aa to i8*
60 call void @llvm.memcpy.p0i8.p0i8.i32(i8* %aaptr, i8* %aptr, i32 16, i32 2, i1 false)
61 %bptr = bitcast i16* %b to i8*
62 call void @llvm.memcpy.p0i8.p0i8.i32(i8* %bptr, i8* %aaptr, i32 16, i32 2, i1 false)
66 define void @test3(i8* %x) {
67 ; Test that when we promote an alloca to a type with lower ABI alignment, we
68 ; provide the needed explicit alignment that code using the alloca may be
69 ; expecting. However, also check that any offset within an alloca can in turn
70 ; reduce the alignment.
72 ; CHECK: alloca [22 x i8], align 8
73 ; CHECK: alloca [18 x i8], align 2
77 %a = alloca { i8*, i8*, i8* }
78 %b = alloca { i8*, i8*, i8* }
79 %a_raw = bitcast { i8*, i8*, i8* }* %a to i8*
80 call void @llvm.memcpy.p0i8.p0i8.i32(i8* %a_raw, i8* %x, i32 22, i32 8, i1 false)
81 %b_raw = bitcast { i8*, i8*, i8* }* %b to i8*
82 %b_gep = getelementptr i8* %b_raw, i32 6
83 call void @llvm.memcpy.p0i8.p0i8.i32(i8* %b_gep, i8* %x, i32 18, i32 2, i1 false)
87 %struct.S = type { i8, { i64 } }
89 define void @test4() {
90 ; This test case triggered very strange alignment behavior with memcpy due to
91 ; strange splitting. Reported by Duncan.
95 %D.2113 = alloca %struct.S
96 %Op = alloca %struct.S
97 %D.2114 = alloca %struct.S
98 %gep1 = getelementptr inbounds %struct.S* %Op, i32 0, i32 0
99 store i8 0, i8* %gep1, align 8
100 %gep2 = getelementptr inbounds %struct.S* %Op, i32 0, i32 1, i32 0
101 %cast = bitcast i64* %gep2 to double*
102 store double 0.000000e+00, double* %cast, align 8
103 store i64 0, i64* %gep2, align 8
104 %dst1 = bitcast %struct.S* %D.2114 to i8*
105 %src1 = bitcast %struct.S* %Op to i8*
106 call void @llvm.memcpy.p0i8.p0i8.i32(i8* %dst1, i8* %src1, i32 16, i32 8, i1 false)
107 %dst2 = bitcast %struct.S* %D.2113 to i8*
108 %src2 = bitcast %struct.S* %D.2114 to i8*
109 call void @llvm.memcpy.p0i8.p0i8.i32(i8* %dst2, i8* %src2, i32 16, i32 8, i1 false)
110 ; We get 3 memcpy calls with various reasons to shrink their alignment to 1.
111 ; CHECK: @llvm.memcpy.p0i8.p0i8.i32(i8* %{{.*}}, i8* %{{.*}}, i32 3, i32 1, i1 false)
112 ; CHECK: @llvm.memcpy.p0i8.p0i8.i32(i8* %{{.*}}, i8* %{{.*}}, i32 8, i32 1, i1 false)
113 ; CHECK: @llvm.memcpy.p0i8.p0i8.i32(i8* %{{.*}}, i8* %{{.*}}, i32 11, i32 1, i1 false)
118 define void @test5() {
119 ; Test that we preserve underaligned loads and stores when splitting.
121 ; CHECK: alloca [9 x i8]
122 ; CHECK: alloca [9 x i8]
123 ; CHECK: store volatile double 0.0{{.*}}, double* %{{.*}}, align 1
124 ; CHECK: load i16* %{{.*}}, align 1
125 ; CHECK: load double* %{{.*}}, align 1
126 ; CHECK: store volatile double %{{.*}}, double* %{{.*}}, align 1
127 ; CHECK: load i16* %{{.*}}, align 1
131 %a = alloca [18 x i8]
132 %raw1 = getelementptr inbounds [18 x i8]* %a, i32 0, i32 0
133 %ptr1 = bitcast i8* %raw1 to double*
134 store volatile double 0.0, double* %ptr1, align 1
135 %weird_gep1 = getelementptr inbounds [18 x i8]* %a, i32 0, i32 7
136 %weird_cast1 = bitcast i8* %weird_gep1 to i16*
137 %weird_load1 = load i16* %weird_cast1, align 1
139 %raw2 = getelementptr inbounds [18 x i8]* %a, i32 0, i32 9
140 %ptr2 = bitcast i8* %raw2 to double*
141 %d1 = load double* %ptr1, align 1
142 store volatile double %d1, double* %ptr2, align 1
143 %weird_gep2 = getelementptr inbounds [18 x i8]* %a, i32 0, i32 16
144 %weird_cast2 = bitcast i8* %weird_gep2 to i16*
145 %weird_load2 = load i16* %weird_cast2, align 1
150 define void @test6() {
151 ; Test that we promote alignment when the underlying alloca switches to one
152 ; that innately provides it.
154 ; CHECK: alloca double
155 ; CHECK: alloca double
160 %a = alloca [16 x i8]
161 %raw1 = getelementptr inbounds [16 x i8]* %a, i32 0, i32 0
162 %ptr1 = bitcast i8* %raw1 to double*
163 store volatile double 0.0, double* %ptr1, align 1
165 %raw2 = getelementptr inbounds [16 x i8]* %a, i32 0, i32 8
166 %ptr2 = bitcast i8* %raw2 to double*
167 %val = load double* %ptr1, align 1
168 store volatile double %val, double* %ptr2, align 1