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" 3 4 declare void @llvm.memcpy.p0i8.p0i8.i32(i8*, i8*, i32, i32, i1) 5 6 define void @test1({ i8, i8 }* %a, { i8, i8 }* %b) { 7 ; CHECK-LABEL: @test1( 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 16 ; CHECK: ret void 17 18 entry: 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 23 24 store i8 420, i8* %gep_alloca, align 16 25 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) 28 ret void 29 } 30 31 define void @test2() { 32 ; CHECK-LABEL: @test2( 33 ; CHECK: alloca i16 34 ; CHECK: load i8* %{{.*}} 35 ; CHECK: store i8 42, i8* %{{.*}} 36 ; CHECK: ret void 37 38 entry: 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 46 ret void 47 } 48 49 define void @PR13920(<2 x i64>* %a, i16* %b) { 50 ; Test that alignments on memcpy intrinsics get propagated to loads and stores. 51 ; CHECK-LABEL: @PR13920( 52 ; CHECK: load <2 x i64>* %a, align 2 53 ; CHECK: store <2 x i64> {{.*}}, <2 x i64>* {{.*}}, align 2 54 ; CHECK: ret void 55 56 entry: 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) 63 ret void 64 } 65 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. 71 ; CHECK-LABEL: @test3( 72 ; CHECK: alloca [22 x i8], align 8 73 ; CHECK: alloca [18 x i8], align 2 74 ; CHECK: ret void 75 76 entry: 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) 84 ret void 85 } 86 87 define void @test5() { 88 ; Test that we preserve underaligned loads and stores when splitting. 89 ; CHECK-LABEL: @test5( 90 ; CHECK: alloca [9 x i8] 91 ; CHECK: alloca [9 x i8] 92 ; CHECK: store volatile double 0.0{{.*}}, double* %{{.*}}, align 1 93 ; CHECK: load i16* %{{.*}}, align 1 94 ; CHECK: load double* %{{.*}}, align 1 95 ; CHECK: store volatile double %{{.*}}, double* %{{.*}}, align 1 96 ; CHECK: load i16* %{{.*}}, align 1 97 ; CHECK: ret void 98 99 entry: 100 %a = alloca [18 x i8] 101 %raw1 = getelementptr inbounds [18 x i8]* %a, i32 0, i32 0 102 %ptr1 = bitcast i8* %raw1 to double* 103 store volatile double 0.0, double* %ptr1, align 1 104 %weird_gep1 = getelementptr inbounds [18 x i8]* %a, i32 0, i32 7 105 %weird_cast1 = bitcast i8* %weird_gep1 to i16* 106 %weird_load1 = load i16* %weird_cast1, align 1 107 108 %raw2 = getelementptr inbounds [18 x i8]* %a, i32 0, i32 9 109 %ptr2 = bitcast i8* %raw2 to double* 110 %d1 = load double* %ptr1, align 1 111 store volatile double %d1, double* %ptr2, align 1 112 %weird_gep2 = getelementptr inbounds [18 x i8]* %a, i32 0, i32 16 113 %weird_cast2 = bitcast i8* %weird_gep2 to i16* 114 %weird_load2 = load i16* %weird_cast2, align 1 115 116 ret void 117 } 118 119 define void @test6() { 120 ; Test that we promote alignment when the underlying alloca switches to one 121 ; that innately provides it. 122 ; CHECK-LABEL: @test6( 123 ; CHECK: alloca double 124 ; CHECK: alloca double 125 ; CHECK-NOT: align 126 ; CHECK: ret void 127 128 entry: 129 %a = alloca [16 x i8] 130 %raw1 = getelementptr inbounds [16 x i8]* %a, i32 0, i32 0 131 %ptr1 = bitcast i8* %raw1 to double* 132 store volatile double 0.0, double* %ptr1, align 1 133 134 %raw2 = getelementptr inbounds [16 x i8]* %a, i32 0, i32 8 135 %ptr2 = bitcast i8* %raw2 to double* 136 %val = load double* %ptr1, align 1 137 store volatile double %val, double* %ptr2, align 1 138 139 ret void 140 } 141 142 define void @test7(i8* %out) { 143 ; Test that we properly compute the destination alignment when rewriting 144 ; memcpys as direct loads or stores. 145 ; CHECK-LABEL: @test7( 146 ; CHECK-NOT: alloca 147 148 entry: 149 %a = alloca [16 x i8] 150 %raw1 = getelementptr inbounds [16 x i8]* %a, i32 0, i32 0 151 %ptr1 = bitcast i8* %raw1 to double* 152 %raw2 = getelementptr inbounds [16 x i8]* %a, i32 0, i32 8 153 %ptr2 = bitcast i8* %raw2 to double* 154 155 call void @llvm.memcpy.p0i8.p0i8.i32(i8* %raw1, i8* %out, i32 16, i32 0, i1 false) 156 ; CHECK: %[[val2:.*]] = load double* %{{.*}}, align 1 157 ; CHECK: %[[val1:.*]] = load double* %{{.*}}, align 1 158 159 %val1 = load double* %ptr2, align 1 160 %val2 = load double* %ptr1, align 1 161 162 store double %val1, double* %ptr1, align 1 163 store double %val2, double* %ptr2, align 1 164 165 call void @llvm.memcpy.p0i8.p0i8.i32(i8* %out, i8* %raw1, i32 16, i32 0, i1 false) 166 ; CHECK: store double %[[val1]], double* %{{.*}}, align 1 167 ; CHECK: store double %[[val2]], double* %{{.*}}, align 1 168 169 ret void 170 ; CHECK: ret void 171 } 172
