1; RUN: opt -S -mtriple=amdgcn-- -amdgpu-lower-module-lds --amdgpu-lower-module-lds-strategy=module < %s | FileCheck %s 2; RUN: opt -S -mtriple=amdgcn-- -passes=amdgpu-lower-module-lds --amdgpu-lower-module-lds-strategy=module < %s | FileCheck %s 3 4; Padding to meet alignment, so references to @var1 replaced with gep ptr, 0, 2 5; No i64 as addrspace(3) types with initializers are ignored. Likewise no addrspace(4). 6; CHECK: %llvm.amdgcn.module.lds.t = type { float, [4 x i8], i32 } 7 8; Variable removed by pass 9; CHECK-NOT: @var0 10 11@var0 = addrspace(3) global float poison, align 8 12@var1 = addrspace(3) global i32 poison, align 8 13 14; The invalid use by the global is left unchanged 15; CHECK: @var1 = addrspace(3) global i32 poison, align 8 16; CHECK: @ptr = addrspace(1) global ptr addrspace(3) @var1, align 4 17@ptr = addrspace(1) global ptr addrspace(3) @var1, align 4 18 19; A variable that is unchanged by pass 20; CHECK: @with_init = addrspace(3) global i64 0 21@with_init = addrspace(3) global i64 0 22 23; Instance of new type, aligned to max of element alignment 24; CHECK: @llvm.amdgcn.module.lds = internal addrspace(3) global %llvm.amdgcn.module.lds.t poison, align 8 25 26; Use in func rewritten to access struct at address zero 27; CHECK-LABEL: @func() 28; CHECK: %dec = atomicrmw fsub ptr addrspace(3) @llvm.amdgcn.module.lds, float 1.0 29; CHECK: %val0 = load i32, ptr addrspace(3) getelementptr inbounds (%llvm.amdgcn.module.lds.t, ptr addrspace(3) @llvm.amdgcn.module.lds, i32 0, i32 2), align 8 30; CHECK: %val1 = add i32 %val0, 4 31; CHECK: store i32 %val1, ptr addrspace(3) getelementptr inbounds (%llvm.amdgcn.module.lds.t, ptr addrspace(3) @llvm.amdgcn.module.lds, i32 0, i32 2), align 8 32; CHECK: %unused0 = atomicrmw add ptr addrspace(3) @with_init, i64 1 monotonic 33define void @func() { 34 %dec = atomicrmw fsub ptr addrspace(3) @var0, float 1.0 monotonic 35 %val0 = load i32, ptr addrspace(3) @var1, align 4 36 %val1 = add i32 %val0, 4 37 store i32 %val1, ptr addrspace(3) @var1, align 4 38 %unused0 = atomicrmw add ptr addrspace(3) @with_init, i64 1 monotonic 39 ret void 40} 41 42; This kernel calls a function that uses LDS so needs the block 43; CHECK-LABEL: @kern_call() #0 44; CHECK: call void @llvm.donothing() [ "ExplicitUse"(ptr addrspace(3) @llvm.amdgcn.module.lds) ] 45; CHECK: call void @func() 46; CHECK: %dec = atomicrmw fsub ptr addrspace(3) @llvm.amdgcn.module.lds, float 2.000000e+00 monotonic, align 8 47define amdgpu_kernel void @kern_call() { 48 call void @func() 49 %dec = atomicrmw fsub ptr addrspace(3) @var0, float 2.0 monotonic 50 ret void 51} 52 53; This kernel does alloc the LDS block as it makes no calls 54; CHECK-LABEL: @kern_empty() 55; CHECK-NOT: call void @llvm.donothing() 56define spir_kernel void @kern_empty() { 57 ret void 58} 59 60; Make sure we don't crash trying to insert code into a kernel 61; declaration. 62declare amdgpu_kernel void @kernel_declaration() 63 64; CHECK: attributes #0 = { "amdgpu-lds-size"="12" } 65