1; RUN: opt < %s -sroa -S | FileCheck %s 2target 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 4declare void @llvm.memcpy.p0i8.p0i8.i32(i8*, i8*, i32, i32, i1) 5 6define void @test1({ i8, i8 }* %a, { i8, i8 }* %b) { 7; CHECK-LABEL: @test1( 8; CHECK: %[[gep_a0:.*]] = getelementptr inbounds { i8, i8 }, { i8, i8 }* %a, i64 0, i32 0 9; CHECK: %[[a0:.*]] = load i8, i8* %[[gep_a0]], align 16 10; CHECK: %[[gep_a1:.*]] = getelementptr inbounds { i8, i8 }, { i8, i8 }* %a, i64 0, i32 1 11; CHECK: %[[a1:.*]] = load i8, i8* %[[gep_a1]], align 1 12; CHECK: %[[gep_b0:.*]] = getelementptr inbounds { i8, i8 }, { 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 }, { i8, i8 }* %b, i64 0, i32 1 15; CHECK: store i8 %[[a1]], i8* %[[gep_b1]], align 1 16; CHECK: ret void 17 18entry: 19 %alloca = alloca { i8, i8 }, align 16 20 %gep_a = getelementptr { i8, i8 }, { i8, i8 }* %a, i32 0, i32 0 21 %gep_alloca = getelementptr { i8, i8 }, { i8, i8 }* %alloca, i32 0, i32 0 22 %gep_b = getelementptr { i8, i8 }, { 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 31define void @test2() { 32; CHECK-LABEL: @test2( 33; CHECK: alloca i16 34; CHECK: load i8, i8* %{{.*}} 35; CHECK: store i8 42, i8* %{{.*}} 36; CHECK: ret void 37 38entry: 39 %a = alloca { i8, i8, i8, i8 }, align 2 40 %gep1 = getelementptr { i8, i8, i8, i8 }, { 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 }, { i8, i8, i8, i8 }* %a, i32 0, i32 2 44 %result = load i8, i8* %gep2 45 store i8 42, i8* %gep2 46 ret void 47} 48 49define 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>, <2 x i64>* %a, align 2 53; CHECK: store <2 x i64> {{.*}}, <2 x i64>* {{.*}}, align 2 54; CHECK: ret void 55 56entry: 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 66define 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 76entry: 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, 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 87define void @test5() { 88; Test that we preserve underaligned loads and stores when splitting. The use 89; of volatile in this test case is just to force the loads and stores to not be 90; split or promoted out of existence. 91; 92; CHECK-LABEL: @test5( 93; CHECK: alloca [9 x i8] 94; CHECK: alloca [9 x i8] 95; CHECK: store volatile double 0.0{{.*}}, double* %{{.*}}, align 1 96; CHECK: load volatile i16, i16* %{{.*}}, align 1 97; CHECK: load double, double* %{{.*}}, align 1 98; CHECK: store volatile double %{{.*}}, double* %{{.*}}, align 1 99; CHECK: load volatile i16, i16* %{{.*}}, align 1 100; CHECK: ret void 101 102entry: 103 %a = alloca [18 x i8] 104 %raw1 = getelementptr inbounds [18 x i8], [18 x i8]* %a, i32 0, i32 0 105 %ptr1 = bitcast i8* %raw1 to double* 106 store volatile double 0.0, double* %ptr1, align 1 107 %weird_gep1 = getelementptr inbounds [18 x i8], [18 x i8]* %a, i32 0, i32 7 108 %weird_cast1 = bitcast i8* %weird_gep1 to i16* 109 %weird_load1 = load volatile i16, i16* %weird_cast1, align 1 110 111 %raw2 = getelementptr inbounds [18 x i8], [18 x i8]* %a, i32 0, i32 9 112 %ptr2 = bitcast i8* %raw2 to double* 113 %d1 = load double, double* %ptr1, align 1 114 store volatile double %d1, double* %ptr2, align 1 115 %weird_gep2 = getelementptr inbounds [18 x i8], [18 x i8]* %a, i32 0, i32 16 116 %weird_cast2 = bitcast i8* %weird_gep2 to i16* 117 %weird_load2 = load volatile i16, i16* %weird_cast2, align 1 118 119 ret void 120} 121 122define void @test6() { 123; Test that we promote alignment when the underlying alloca switches to one 124; that innately provides it. 125; CHECK-LABEL: @test6( 126; CHECK: alloca double 127; CHECK: alloca double 128; CHECK-NOT: align 129; CHECK: ret void 130 131entry: 132 %a = alloca [16 x i8] 133 %raw1 = getelementptr inbounds [16 x i8], [16 x i8]* %a, i32 0, i32 0 134 %ptr1 = bitcast i8* %raw1 to double* 135 store volatile double 0.0, double* %ptr1, align 1 136 137 %raw2 = getelementptr inbounds [16 x i8], [16 x i8]* %a, i32 0, i32 8 138 %ptr2 = bitcast i8* %raw2 to double* 139 %val = load double, double* %ptr1, align 1 140 store volatile double %val, double* %ptr2, align 1 141 142 ret void 143} 144 145define void @test7(i8* %out) { 146; Test that we properly compute the destination alignment when rewriting 147; memcpys as direct loads or stores. 148; CHECK-LABEL: @test7( 149; CHECK-NOT: alloca 150 151entry: 152 %a = alloca [16 x i8] 153 %raw1 = getelementptr inbounds [16 x i8], [16 x i8]* %a, i32 0, i32 0 154 %ptr1 = bitcast i8* %raw1 to double* 155 %raw2 = getelementptr inbounds [16 x i8], [16 x i8]* %a, i32 0, i32 8 156 %ptr2 = bitcast i8* %raw2 to double* 157 158 call void @llvm.memcpy.p0i8.p0i8.i32(i8* %raw1, i8* %out, i32 16, i32 0, i1 false) 159; CHECK: %[[val2:.*]] = load double, double* %{{.*}}, align 1 160; CHECK: %[[val1:.*]] = load double, double* %{{.*}}, align 1 161 162 %val1 = load double, double* %ptr2, align 1 163 %val2 = load double, double* %ptr1, align 1 164 165 store double %val1, double* %ptr1, align 1 166 store double %val2, double* %ptr2, align 1 167 168 call void @llvm.memcpy.p0i8.p0i8.i32(i8* %out, i8* %raw1, i32 16, i32 0, i1 false) 169; CHECK: store double %[[val1]], double* %{{.*}}, align 1 170; CHECK: store double %[[val2]], double* %{{.*}}, align 1 171 172 ret void 173; CHECK: ret void 174} 175