1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: -p --check-globals 2; RUN: opt -S -mtriple=amdgcn-- -amdgpu-lower-module-lds --amdgpu-lower-module-lds-strategy=module < %s | FileCheck %s 3; RUN: opt -S -mtriple=amdgcn-- -passes=amdgpu-lower-module-lds --amdgpu-lower-module-lds-strategy=module < %s | FileCheck %s 4 5@lds.1 = internal unnamed_addr addrspace(3) global [2 x i8] poison, align 1 6 7; CHECK: %llvm.amdgcn.kernel.k0.lds.t = type { [2 x i8] } 8; CHECK: %llvm.amdgcn.kernel.k1.lds.t = type { [2 x i8] } 9; CHECK: %llvm.amdgcn.kernel.k2.lds.t = type { i32 } 10; CHECK: %llvm.amdgcn.kernel.k3.lds.t = type { [32 x i8] } 11; CHECK: %llvm.amdgcn.kernel.k4.lds.t = type { [2 x i8] } 12; CHECK: %llvm.amdgcn.kernel.k5.lds.t = type { [505 x i32] } 13; CHECK: %llvm.amdgcn.kernel.k6.lds.t = type { [4 x i32] } 14 15; Use constant from different kernels 16;. 17; CHECK: @llvm.amdgcn.kernel.k0.lds = internal addrspace(3) global %llvm.amdgcn.kernel.k0.lds.t poison, align 2 18; CHECK: @llvm.amdgcn.kernel.k1.lds = internal addrspace(3) global %llvm.amdgcn.kernel.k1.lds.t poison, align 2 19; CHECK: @llvm.amdgcn.kernel.k2.lds = internal addrspace(3) global %llvm.amdgcn.kernel.k2.lds.t poison, align 4 20; CHECK: @llvm.amdgcn.kernel.k3.lds = internal addrspace(3) global %llvm.amdgcn.kernel.k3.lds.t poison, align 16 21; CHECK: @llvm.amdgcn.kernel.k4.lds = internal addrspace(3) global %llvm.amdgcn.kernel.k4.lds.t poison, align 2 22; CHECK: @llvm.amdgcn.kernel.k5.lds = internal addrspace(3) global %llvm.amdgcn.kernel.k5.lds.t poison, align 16 23; CHECK: @llvm.amdgcn.kernel.k6.lds = internal addrspace(3) global %llvm.amdgcn.kernel.k6.lds.t poison, align 16 24;. 25define amdgpu_kernel void @k0(i64 %x) { 26; CHECK-LABEL: @k0( 27; CHECK-NEXT: %1 = addrspacecast ptr addrspace(3) @llvm.amdgcn.kernel.k0.lds to ptr 28; CHECK-NEXT: %ptr = getelementptr inbounds i8, ptr %1, i64 %x 29; CHECK-NEXT: store i8 1, ptr %ptr, align 1 30; CHECK-NEXT: ret void 31; 32 %ptr = getelementptr inbounds i8, ptr addrspacecast (ptr addrspace(3) @lds.1 to ptr), i64 %x 33 store i8 1, ptr addrspace(0) %ptr, align 1 34 ret void 35} 36 37define amdgpu_kernel void @k1(i64 %x) { 38; CHECK-LABEL: @k1( 39; CHECK-NEXT: %1 = addrspacecast ptr addrspace(3) @llvm.amdgcn.kernel.k1.lds to ptr 40; CHECK-NEXT: %ptr = getelementptr inbounds i8, ptr %1, i64 %x 41; CHECK-NEXT: store i8 1, ptr %ptr, align 1 42; CHECK-NEXT: ret void 43; 44 %ptr = getelementptr inbounds i8, ptr addrspacecast (ptr addrspace(3) @lds.1 to ptr), i64 %x 45 store i8 1, ptr addrspace(0) %ptr, align 1 46 ret void 47} 48 49@lds.2 = internal unnamed_addr addrspace(3) global i32 poison, align 4 50 51; Use constant twice from the same kernel 52define amdgpu_kernel void @k2(i64 %x) { 53; CHECK-LABEL: @k2( 54; CHECK-NEXT: store i8 1, ptr addrspace(3) @llvm.amdgcn.kernel.k2.lds, align 4 55; CHECK-NEXT: store i8 2, ptr addrspace(3) @llvm.amdgcn.kernel.k2.lds, align 4 56; CHECK-NEXT: ret void 57; 58 store i8 1, ptr addrspace(3) @lds.2, align 4 59 store i8 2, ptr addrspace(3) @lds.2, align 4 60 ret void 61} 62 63@lds.3 = internal unnamed_addr addrspace(3) global [32 x i8] poison, align 1 64 65; Use constant twice from the same kernel but a different other constant. 66define amdgpu_kernel void @k3(i64 %x) { 67; CHECK-LABEL: @k3( 68; CHECK-NEXT: %1 = getelementptr inbounds [32 x i8], ptr addrspace(3) @llvm.amdgcn.kernel.k3.lds, i32 0, i32 16 69; CHECK-NEXT: %ptr1 = addrspacecast ptr addrspace(3) %1 to ptr 70; CHECK-NEXT: store i64 1, ptr %ptr1, align 1 71; CHECK-NEXT: %2 = getelementptr inbounds [32 x i8], ptr addrspace(3) @llvm.amdgcn.kernel.k3.lds, i32 0, i32 24 72; CHECK-NEXT: %ptr2 = addrspacecast ptr addrspace(3) %2 to ptr 73; CHECK-NEXT: store i64 2, ptr %ptr2, align 8 74; CHECK-NEXT: ret void 75; 76 %ptr1 = addrspacecast ptr addrspace(3) getelementptr inbounds ([32 x i8], ptr addrspace(3) @lds.3, i32 0, i32 16) to ptr 77 store i64 1, ptr %ptr1, align 1 78 %ptr2 = addrspacecast ptr addrspace(3) getelementptr inbounds ([32 x i8], ptr addrspace(3) @lds.3, i32 0, i32 24) to ptr 79 store i64 2, ptr %ptr2, align 1 80 ret void 81} 82 83; @lds.1 is used from constant expressions in different kernels. 84define amdgpu_kernel void @k4(i64 %x) { 85; CHECK-LABEL: @k4( 86; CHECK-NEXT: %1 = addrspacecast ptr addrspace(3) @llvm.amdgcn.kernel.k4.lds to ptr 87; CHECK-NEXT: %ptr = getelementptr inbounds i8, ptr %1, i64 %x 88; CHECK-NEXT: store i8 1, ptr %ptr, align 1 89; CHECK-NEXT: ret void 90; 91 %ptr = getelementptr inbounds i8, ptr addrspacecast (ptr addrspace(3) @lds.1 to ptr), i64 %x 92 store i8 1, ptr addrspace(0) %ptr, align 1 93 ret void 94} 95 96@lds.4 = internal unnamed_addr addrspace(3) global [505 x i32] poison, align 4 97 98; Multiple constexpr use in a same instruction. 99define amdgpu_kernel void @k5() { 100; CHECK-LABEL: @k5( 101; CHECK-NEXT: %1 = addrspacecast ptr addrspace(3) @llvm.amdgcn.kernel.k5.lds to ptr 102; CHECK-NEXT: %2 = addrspacecast ptr addrspace(3) @llvm.amdgcn.kernel.k5.lds to ptr 103; CHECK-NEXT: call void poison(ptr %1, ptr %2) 104; 105 call void poison(ptr addrspacecast (ptr addrspace(3) @lds.4 to ptr), ptr addrspacecast (ptr addrspace(3) @lds.4 to ptr)) 106 ret void 107} 108 109@lds.5 = internal addrspace(3) global [4 x i32] poison, align 4 110 111; Both the *value* and *pointer* operands of store instruction are constant expressions, and 112; both of these constant expression paths use same lds - @lds.5. Hence both of these constant 113; expression operands of store should be replaced by equivalent instruction sequences. 114define amdgpu_kernel void @k6() { 115; CHECK-LABEL: @k6( 116 117; CHECK-NEXT: %1 = getelementptr inbounds [4 x i32], ptr addrspace(3) @llvm.amdgcn.kernel.k6.lds, i32 0, i32 2 118; CHECK-NEXT: %2 = ptrtoint ptr addrspace(3) %1 to i32 119; CHECK-NEXT: %3 = getelementptr inbounds [4 x i32], ptr addrspace(3) @llvm.amdgcn.kernel.k6.lds, i32 0, i32 2 120; CHECK-NEXT: store i32 %2, ptr addrspace(3) %3, align 8 121; CHECK-NEXT: ret void 122; 123 store i32 ptrtoint (ptr addrspace(3) getelementptr inbounds ([4 x i32], ptr addrspace(3) @lds.5, i32 0, i32 2) to i32), ptr addrspace(3) getelementptr inbounds ([4 x i32], ptr addrspace(3) @lds.5, i32 0, i32 2) 124 ret void 125} 126