1; RUN: opt < %s -sroa -S | FileCheck %s 2target datalayout = "e-p:64:64:64-p1:16:16:16-p3:32:32:32-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* nocapture, i8* nocapture readonly, i32, i1) 5declare void @llvm.memcpy.p1i8.p0i8.i32(i8 addrspace(1)* nocapture, i8* nocapture readonly, i32, i1) 6declare void @llvm.memcpy.p0i8.p1i8.i32(i8* nocapture, i8 addrspace(1)* nocapture readonly, i32, i1) 7declare void @llvm.memcpy.p1i8.p1i8.i32(i8 addrspace(1)* nocapture, i8 addrspace(1)* nocapture readonly, i32, i1) 8 9 10; Make sure an illegal bitcast isn't introduced 11define void @test_address_space_1_1(<2 x i64> addrspace(1)* %a, i16 addrspace(1)* %b) { 12; CHECK-LABEL: @test_address_space_1_1( 13; CHECK: load <2 x i64>, <2 x i64> addrspace(1)* %a, align 2 14; CHECK: store <2 x i64> {{.*}}, <2 x i64> addrspace(1)* {{.*}}, align 2 15; CHECK: ret void 16 %aa = alloca <2 x i64>, align 16 17 %aptr = bitcast <2 x i64> addrspace(1)* %a to i8 addrspace(1)* 18 %aaptr = bitcast <2 x i64>* %aa to i8* 19 call void @llvm.memcpy.p0i8.p1i8.i32(i8* align 2 %aaptr, i8 addrspace(1)* align 2 %aptr, i32 16, i1 false) 20 %bptr = bitcast i16 addrspace(1)* %b to i8 addrspace(1)* 21 call void @llvm.memcpy.p1i8.p0i8.i32(i8 addrspace(1)* align 2 %bptr, i8* align 2 %aaptr, i32 16, i1 false) 22 ret void 23} 24 25define void @test_address_space_1_0(<2 x i64> addrspace(1)* %a, i16* %b) { 26; CHECK-LABEL: @test_address_space_1_0( 27; CHECK: load <2 x i64>, <2 x i64> addrspace(1)* %a, align 2 28; CHECK: store <2 x i64> {{.*}}, <2 x i64>* {{.*}}, align 2 29; CHECK: ret void 30 %aa = alloca <2 x i64>, align 16 31 %aptr = bitcast <2 x i64> addrspace(1)* %a to i8 addrspace(1)* 32 %aaptr = bitcast <2 x i64>* %aa to i8* 33 call void @llvm.memcpy.p0i8.p1i8.i32(i8* align 2 %aaptr, i8 addrspace(1)* align 2 %aptr, i32 16, i1 false) 34 %bptr = bitcast i16* %b to i8* 35 call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 2 %bptr, i8* align 2 %aaptr, i32 16, i1 false) 36 ret void 37} 38 39define void @test_address_space_0_1(<2 x i64>* %a, i16 addrspace(1)* %b) { 40; CHECK-LABEL: @test_address_space_0_1( 41; CHECK: load <2 x i64>, <2 x i64>* %a, align 2 42; CHECK: store <2 x i64> {{.*}}, <2 x i64> addrspace(1)* {{.*}}, align 2 43; CHECK: ret void 44 %aa = alloca <2 x i64>, align 16 45 %aptr = bitcast <2 x i64>* %a to i8* 46 %aaptr = bitcast <2 x i64>* %aa to i8* 47 call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 2 %aaptr, i8* align 2 %aptr, i32 16, i1 false) 48 %bptr = bitcast i16 addrspace(1)* %b to i8 addrspace(1)* 49 call void @llvm.memcpy.p1i8.p0i8.i32(i8 addrspace(1)* align 2 %bptr, i8* align 2 %aaptr, i32 16, i1 false) 50 ret void 51} 52 53%struct.struct_test_27.0.13 = type { i32, float, i64, i8, [4 x i32] } 54 55; Function Attrs: nounwind 56define void @copy_struct([5 x i64] %in.coerce) { 57; CHECK-LABEL: @copy_struct( 58; CHECK-NOT: memcpy 59for.end: 60 %in = alloca %struct.struct_test_27.0.13, align 8 61 %0 = bitcast %struct.struct_test_27.0.13* %in to [5 x i64]* 62 store [5 x i64] %in.coerce, [5 x i64]* %0, align 8 63 %scevgep9 = getelementptr %struct.struct_test_27.0.13, %struct.struct_test_27.0.13* %in, i32 0, i32 4, i32 0 64 %scevgep910 = bitcast i32* %scevgep9 to i8* 65 call void @llvm.memcpy.p1i8.p0i8.i32(i8 addrspace(1)* align 4 undef, i8* align 4 %scevgep910, i32 16, i1 false) 66 ret void 67} 68 69%union.anon = type { i32* } 70 71@g = common global i32 0, align 4 72@l = common addrspace(3) global i32 0, align 4 73 74; If pointers from different address spaces have different sizes, make sure an 75; illegal bitcast isn't introduced 76define void @pr27557() { 77; CHECK-LABEL: @pr27557( 78; CHECK: %[[CAST:.*]] = bitcast i32** {{.*}} to i32 addrspace(3)** 79; CHECK: store i32 addrspace(3)* @l, i32 addrspace(3)** %[[CAST]] 80 %1 = alloca %union.anon, align 8 81 %2 = bitcast %union.anon* %1 to i32** 82 store i32* @g, i32** %2, align 8 83 %3 = bitcast %union.anon* %1 to i32 addrspace(3)** 84 store i32 addrspace(3)* @l, i32 addrspace(3)** %3, align 8 85 ret void 86} 87 88@l2 = common addrspace(2) global i32 0, align 4 89 90; If pointers from different address spaces have the same size, that pointer 91; should be promoted through the pair of `ptrtoint`/`inttoptr`. 92define i32* @pr27557.alt() { 93; CHECK-LABEL: @pr27557.alt( 94; CHECK: ret i32* inttoptr (i64 ptrtoint (i32 addrspace(2)* @l2 to i64) to i32*) 95 %1 = alloca %union.anon, align 8 96 %2 = bitcast %union.anon* %1 to i32 addrspace(2)** 97 store i32 addrspace(2)* @l2, i32 addrspace(2)** %2, align 8 98 %3 = bitcast %union.anon* %1 to i32** 99 %4 = load i32*, i32** %3, align 8 100 ret i32* %4 101} 102 103; Make sure pre-splitting doesn't try to introduce an illegal bitcast 104define float @presplit(i64 addrspace(1)* %p) { 105entry: 106; CHECK-LABEL: @presplit( 107; CHECK: %[[CAST:.*]] = bitcast i64 addrspace(1)* {{.*}} to i32 addrspace(1)* 108; CHECK: load i32, i32 addrspace(1)* %[[CAST]] 109 %b = alloca i64 110 %b.cast = bitcast i64* %b to [2 x float]* 111 %b.gep1 = getelementptr [2 x float], [2 x float]* %b.cast, i32 0, i32 0 112 %b.gep2 = getelementptr [2 x float], [2 x float]* %b.cast, i32 0, i32 1 113 %l = load i64, i64 addrspace(1)* %p 114 store i64 %l, i64* %b 115 %f1 = load float, float* %b.gep1 116 %f2 = load float, float* %b.gep2 117 %ret = fadd float %f1, %f2 118 ret float %ret 119} 120 121; Test load from and store to non-zero address space. 122define void @test_load_store_diff_addr_space([2 x float] addrspace(1)* %complex1, [2 x float] addrspace(1)* %complex2) { 123; CHECK-LABEL: @test_load_store_diff_addr_space 124; CHECK-NOT: alloca 125; CHECK: load i32, i32 addrspace(1)* 126; CHECK: load i32, i32 addrspace(1)* 127; CHECK: store i32 %{{.*}}, i32 addrspace(1)* 128; CHECK: store i32 %{{.*}}, i32 addrspace(1)* 129 %a = alloca i64 130 %a.cast = bitcast i64* %a to [2 x float]* 131 %a.gep1 = getelementptr [2 x float], [2 x float]* %a.cast, i32 0, i32 0 132 %a.gep2 = getelementptr [2 x float], [2 x float]* %a.cast, i32 0, i32 1 133 %complex1.gep = getelementptr [2 x float], [2 x float] addrspace(1)* %complex1, i32 0, i32 0 134 %p1 = bitcast float addrspace(1)* %complex1.gep to i64 addrspace(1)* 135 %v1 = load i64, i64 addrspace(1)* %p1 136 store i64 %v1, i64* %a 137 %f1 = load float, float* %a.gep1 138 %f2 = load float, float* %a.gep2 139 %sum = fadd float %f1, %f2 140 store float %sum, float* %a.gep1 141 store float %sum, float* %a.gep2 142 %v2 = load i64, i64* %a 143 %complex2.gep = getelementptr [2 x float], [2 x float] addrspace(1)* %complex2, i32 0, i32 0 144 %p2 = bitcast float addrspace(1)* %complex2.gep to i64 addrspace(1)* 145 store i64 %v2, i64 addrspace(1)* %p2 146 ret void 147} 148