168 lines · plain
1; RUN: opt -S -mtriple=amdgcn-- -amdgpu-lower-module-lds < %s | FileCheck %s2; RUN: opt -S -mtriple=amdgcn-- -passes=amdgpu-lower-module-lds < %s | FileCheck %s3 4; Properly aligned, same size as alignment.5; CHECK: %llvm.amdgcn.kernel.k0.lds.t = type { [16 x i8], [8 x i8], [4 x i8], [2 x i8], [1 x i8] }6 7; Different properly aligned values, but same size of 1.8; CHECK: %llvm.amdgcn.kernel.k1.lds.t = type { [1 x i8], [1 x i8], [1 x i8], [1 x i8], [1 x i8], [3 x i8], [1 x i8] }9 10; All are under-aligned, requires to fix each on different alignment boundary.11; CHECK: %llvm.amdgcn.kernel.k2.lds.t = type { [9 x i8], [1 x i8], [2 x i8], [3 x i8], [1 x i8], [5 x i8] }12 13; All LDS are underaligned, requires to allocate on 8 byte boundary14; CHECK: %llvm.amdgcn.kernel.k3.lds.t = type { [7 x i8], [1 x i8], [7 x i8], [1 x i8], [6 x i8], [2 x i8], [5 x i8] }15 16; All LDS are underaligned, requires to allocate on 16 byte boundary17; CHECK: %llvm.amdgcn.kernel.k4.lds.t = type { [12 x i8], [4 x i8], [11 x i8], [5 x i8], [10 x i8], [6 x i8], [9 x i8] }18 19; All LDS are properly aligned on 16 byte boundary, but they are of different size.20; CHECK: %llvm.amdgcn.kernel.k5.lds.t = type { [20 x i8], [12 x i8], [19 x i8], [13 x i8], [18 x i8], [14 x i8], [17 x i8] }21 22; CHECK: @llvm.amdgcn.kernel.k0.lds = internal addrspace(3) global %llvm.amdgcn.kernel.k0.lds.t poison, align 1623; CHECK: @llvm.amdgcn.kernel.k1.lds = internal addrspace(3) global %llvm.amdgcn.kernel.k1.lds.t poison, align 1624; CHECK: @llvm.amdgcn.kernel.k2.lds = internal addrspace(3) global %llvm.amdgcn.kernel.k2.lds.t poison, align 1625; CHECK: @llvm.amdgcn.kernel.k3.lds = internal addrspace(3) global %llvm.amdgcn.kernel.k3.lds.t poison, align 826; CHECK: @llvm.amdgcn.kernel.k4.lds = internal addrspace(3) global %llvm.amdgcn.kernel.k4.lds.t poison, align 1627; CHECK: @llvm.amdgcn.kernel.k5.lds = internal addrspace(3) global %llvm.amdgcn.kernel.k5.lds.t poison, align 1628 29 30; Properly aligned, same size as alignment.31; CHECK-NOT: @k0.lds.size.1.align.132; CHECK-NOT: @k0.lds.size.2.align.233; CHECK-NOT: @k0.lds.size.4.align.434; CHECK-NOT: @k0.lds.size.8.align.835; CHECK-NOT: @k0.lds.size.16.align.1636@k0.lds.size.1.align.1 = internal unnamed_addr addrspace(3) global [1 x i8] poison, align 137@k0.lds.size.2.align.2 = internal unnamed_addr addrspace(3) global [2 x i8] poison, align 238@k0.lds.size.4.align.4 = internal unnamed_addr addrspace(3) global [4 x i8] poison, align 439@k0.lds.size.8.align.8 = internal unnamed_addr addrspace(3) global [8 x i8] poison, align 840@k0.lds.size.16.align.16 = internal unnamed_addr addrspace(3) global [16 x i8] poison, align 1641 42define amdgpu_kernel void @k0() {43 store i8 1, ptr addrspace(3) @k0.lds.size.1.align.1, align 144 45 store i8 2, ptr addrspace(3) @k0.lds.size.2.align.2, align 246 47 store i8 3, ptr addrspace(3) @k0.lds.size.4.align.4, align 448 49 store i8 4, ptr addrspace(3) @k0.lds.size.8.align.8, align 850 51 store i8 5, ptr addrspace(3) @k0.lds.size.16.align.16, align 1652 53 ret void54}55 56; Different properly aligned values, but same size of 1.57; CHECK-NOT: @k1.lds.size.1.align.158; CHECK-NOT: @k1.lds.size.1.align.259; CHECK-NOT: @k1.lds.size.1.align.460; CHECK-NOT: @k1.lds.size.1.align.861; CHECK-NOT: @k1.lds.size.1.align.1662@k1.lds.size.1.align.1 = internal unnamed_addr addrspace(3) global [1 x i8] poison, align 163@k1.lds.size.1.align.2 = internal unnamed_addr addrspace(3) global [1 x i8] poison, align 264@k1.lds.size.1.align.4 = internal unnamed_addr addrspace(3) global [1 x i8] poison, align 465@k1.lds.size.1.align.8 = internal unnamed_addr addrspace(3) global [1 x i8] poison, align 866@k1.lds.size.1.align.16 = internal unnamed_addr addrspace(3) global [1 x i8] poison, align 1667 68define amdgpu_kernel void @k1() {69 store i8 1, ptr addrspace(3) @k1.lds.size.1.align.1, align 170 71 store i8 2, ptr addrspace(3) @k1.lds.size.1.align.2, align 272 73 store i8 3, ptr addrspace(3) @k1.lds.size.1.align.4, align 474 75 store i8 4, ptr addrspace(3) @k1.lds.size.1.align.8, align 876 77 store i8 5, ptr addrspace(3) @k1.lds.size.1.align.16, align 1678 79 ret void80}81 82; All are under-aligned, requires to fix each on different alignment boundary.83; CHECK-NOT: @k2.lds.size.2.align.184; CHECK-NOT: @k2.lds.size.3.align.285; CHECK-NOT: @k2.lds.size.5.align.486; CHECK-NOT: @k2.lds.size.9.align.887@k2.lds.size.2.align.1 = internal unnamed_addr addrspace(3) global [2 x i8] poison, align 188@k2.lds.size.3.align.2 = internal unnamed_addr addrspace(3) global [3 x i8] poison, align 289@k2.lds.size.5.align.4 = internal unnamed_addr addrspace(3) global [5 x i8] poison, align 490@k2.lds.size.9.align.8 = internal unnamed_addr addrspace(3) global [9 x i8] poison, align 891 92define amdgpu_kernel void @k2() {93 store i8 1, ptr addrspace(3) @k2.lds.size.2.align.1, align 194 95 store i8 2, ptr addrspace(3) @k2.lds.size.3.align.2, align 296 97 store i8 3, ptr addrspace(3) @k2.lds.size.5.align.4, align 498 99 store i8 4, ptr addrspace(3) @k2.lds.size.9.align.8, align 8100 101 ret void102}103 104; All LDS are underaligned, requires to allocate on 8 byte boundary105; CHECK-NOT: @k3.lds.size.5.align.2106; CHECK-NOT: @k3.lds.size.6.align.2107; CHECK-NOT: @k3.lds.size.7.align.2108; CHECK-NOT: @k3.lds.size.7.align.4109@k3.lds.size.5.align.2 = internal unnamed_addr addrspace(3) global [5 x i8] poison, align 2110@k3.lds.size.6.align.2 = internal unnamed_addr addrspace(3) global [6 x i8] poison, align 2111@k3.lds.size.7.align.2 = internal unnamed_addr addrspace(3) global [7 x i8] poison, align 2112@k3.lds.size.7.align.4 = internal unnamed_addr addrspace(3) global [7 x i8] poison, align 4113 114define amdgpu_kernel void @k3() {115 store i8 1, ptr addrspace(3) @k3.lds.size.5.align.2, align 2116 117 store i8 2, ptr addrspace(3) @k3.lds.size.6.align.2, align 2118 119 store i8 3, ptr addrspace(3) @k3.lds.size.7.align.2, align 2120 121 store i8 4, ptr addrspace(3) @k3.lds.size.7.align.4, align 4122 123 ret void124}125 126; All LDS are underaligned, requires to allocate on 16 byte boundary127; CHECK-NOT: @k4.lds.size.9.align.1128; CHECK-NOT: @k4.lds.size.10.align.2129; CHECK-NOT: @k4.lds.size.11.align.4130; CHECK-NOT: @k4.lds.size.12.align.8131@k4.lds.size.9.align.1 = internal unnamed_addr addrspace(3) global [9 x i8] poison, align 1132@k4.lds.size.10.align.2 = internal unnamed_addr addrspace(3) global [10 x i8] poison, align 2133@k4.lds.size.11.align.4 = internal unnamed_addr addrspace(3) global [11 x i8] poison, align 4134@k4.lds.size.12.align.8 = internal unnamed_addr addrspace(3) global [12 x i8] poison, align 8135 136define amdgpu_kernel void @k4() {137 store i8 1, ptr addrspace(3) @k4.lds.size.9.align.1, align 1138 139 store i8 2, ptr addrspace(3) @k4.lds.size.10.align.2, align 2140 141 store i8 3, ptr addrspace(3) @k4.lds.size.11.align.4, align 4142 143 store i8 4, ptr addrspace(3) @k4.lds.size.12.align.8, align 8144 145 ret void146}147 148; CHECK-NOT: @k5.lds.size.17.align.16149; CHECK-NOT: @k5.lds.size.18.align.16150; CHECK-NOT: @k5.lds.size.19.align.16151; CHECK-NOT: @k5.lds.size.20.align.16152@k5.lds.size.17.align.16 = internal unnamed_addr addrspace(3) global [17 x i8] poison, align 16153@k5.lds.size.18.align.16 = internal unnamed_addr addrspace(3) global [18 x i8] poison, align 16154@k5.lds.size.19.align.16 = internal unnamed_addr addrspace(3) global [19 x i8] poison, align 16155@k5.lds.size.20.align.16 = internal unnamed_addr addrspace(3) global [20 x i8] poison, align 16156 157define amdgpu_kernel void @k5() {158 store i8 1, ptr addrspace(3) @k5.lds.size.17.align.16, align 16159 160 store i8 2, ptr addrspace(3) @k5.lds.size.18.align.16, align 16161 162 store i8 3, ptr addrspace(3) @k5.lds.size.19.align.16, align 16163 164 store i8 4, ptr addrspace(3) @k5.lds.size.20.align.16, align 16165 166 ret void167}168