92 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 62; RUN: opt -S -mtriple=amdgcn-- -amdgpu-lower-module-lds --amdgpu-lower-module-lds-strategy=module < %s | FileCheck %s3; RUN: opt -S -mtriple=amdgcn-- -passes=amdgpu-lower-module-lds --amdgpu-lower-module-lds-strategy=module < %s | FileCheck %s4 5; CHECK: %llvm.amdgcn.module.lds.t = type { float, float }6; CHECK: %llvm.amdgcn.kernel.timestwo.lds.t = type { float, float }7 8@a_func = addrspace(3) global float poison, align 49 10@kern = addrspace(3) global float poison, align 411 12; @a_func is only used from a non-kernel function so is rewritten13; @b_both is used from a non-kernel function so is rewritten14; sorted both < func, so @b_both at null and @a_func at 415@b_both = addrspace(3) global float poison, align 416 17 18define i32 @get_func() local_unnamed_addr #0 {19; CHECK-LABEL: define i32 @get_func() local_unnamed_addr {20; CHECK-NEXT: [[ENTRY:.*:]]21; CHECK-NEXT: [[TMP0:%.*]] = addrspacecast ptr addrspace(3) @llvm.amdgcn.module.lds to ptr22; CHECK-NEXT: [[TMP1:%.*]] = ptrtoint ptr [[TMP0]] to i6423; CHECK-NEXT: [[TMP2:%.*]] = add i64 [[TMP1]], [[TMP1]]24; CHECK-NEXT: [[TMP3:%.*]] = inttoptr i64 [[TMP2]] to ptr25; CHECK-NEXT: [[TMP4:%.*]] = load i32, ptr [[TMP3]], align 426; CHECK-NEXT: ret i32 [[TMP4]]27;28entry:29 %0 = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr addrspacecast (ptr addrspace(3) @a_func to ptr) to i64), i64 ptrtoint (ptr addrspacecast (ptr addrspace(3) @a_func to ptr) to i64)) to ptr), align 430 ret i32 %031}32 33define void @set_func(i32 %x) {34; CHECK-LABEL: define void @set_func(35; CHECK-SAME: i32 [[X:%.*]]) {36; CHECK-NEXT: [[ENTRY:.*:]]37; CHECK-NEXT: [[TMP0:%.*]] = addrspacecast ptr addrspace(3) getelementptr inbounds ([[LLVM_AMDGCN_MODULE_LDS_T:%.*]], ptr addrspace(3) @llvm.amdgcn.module.lds, i32 0, i32 1) to ptr38; CHECK-NEXT: [[TMP1:%.*]] = ptrtoint ptr [[TMP0]] to i6439; CHECK-NEXT: [[TMP2:%.*]] = add i64 [[TMP1]], [[TMP1]]40; CHECK-NEXT: [[TMP3:%.*]] = inttoptr i64 [[TMP2]] to ptr41; CHECK-NEXT: store i32 [[X]], ptr [[TMP3]], align 442; CHECK-NEXT: ret void43;44entry:45 store i32 %x, ptr inttoptr (i64 add (i64 ptrtoint (ptr addrspacecast (ptr addrspace(3) @b_both to ptr) to i64), i64 ptrtoint (ptr addrspacecast (ptr addrspace(3) @b_both to ptr) to i64)) to ptr), align 446 ret void47}48 49 50define amdgpu_kernel void @timestwo() {51; CHECK-LABEL: define amdgpu_kernel void @timestwo(52; CHECK-SAME: ) #[[ATTR0:[0-9]+]] {53; CHECK-NEXT: [[TMP1:%.*]] = addrspacecast ptr addrspace(3) @llvm.amdgcn.kernel.timestwo.lds to ptr54; CHECK-NEXT: [[TMP2:%.*]] = ptrtoint ptr [[TMP1]] to i6455; CHECK-NEXT: [[TMP3:%.*]] = addrspacecast ptr addrspace(3) getelementptr inbounds ([[LLVM_AMDGCN_KERNEL_TIMESTWO_LDS_T:%.*]], ptr addrspace(3) @llvm.amdgcn.kernel.timestwo.lds, i32 0, i32 1) to ptr56; CHECK-NEXT: [[TMP4:%.*]] = ptrtoint ptr [[TMP3]] to i6457; CHECK-NEXT: [[TMP5:%.*]] = add i64 [[TMP2]], [[TMP4]]58; CHECK-NEXT: [[TMP6:%.*]] = inttoptr i64 [[TMP5]] to ptr59; CHECK-NEXT: [[LD:%.*]] = load i32, ptr [[TMP6]], align 460; CHECK-NEXT: [[MUL:%.*]] = mul i32 [[LD]], 261; CHECK-NEXT: [[TMP7:%.*]] = addrspacecast ptr addrspace(3) getelementptr inbounds ([[LLVM_AMDGCN_KERNEL_TIMESTWO_LDS_T]], ptr addrspace(3) @llvm.amdgcn.kernel.timestwo.lds, i32 0, i32 1) to ptr62; CHECK-NEXT: [[TMP8:%.*]] = ptrtoint ptr [[TMP7]] to i6463; CHECK-NEXT: [[TMP9:%.*]] = addrspacecast ptr addrspace(3) @llvm.amdgcn.kernel.timestwo.lds to ptr64; CHECK-NEXT: [[TMP10:%.*]] = ptrtoint ptr [[TMP9]] to i6465; CHECK-NEXT: [[TMP11:%.*]] = add i64 [[TMP8]], [[TMP10]]66; CHECK-NEXT: [[TMP12:%.*]] = inttoptr i64 [[TMP11]] to ptr67; CHECK-NEXT: store i32 [[MUL]], ptr [[TMP12]], align 468; CHECK-NEXT: ret void69;70 %ld = load i32, ptr inttoptr (i64 add (i64 ptrtoint (ptr addrspacecast (ptr addrspace(3) @b_both to ptr) to i64), i64 ptrtoint (ptr addrspacecast (ptr addrspace(3) @kern to ptr) to i64)) to ptr), align 471 %mul = mul i32 %ld, 272 store i32 %mul, ptr inttoptr (i64 add (i64 ptrtoint (ptr addrspacecast (ptr addrspace(3) @kern to ptr) to i64), i64 ptrtoint (ptr addrspacecast (ptr addrspace(3) @b_both to ptr) to i64)) to ptr), align 473 ret void74}75 76define amdgpu_kernel void @through_functions() {77; CHECK-LABEL: define amdgpu_kernel void @through_functions(78; CHECK-SAME: ) #[[ATTR0]] {79; CHECK-NEXT: call void @llvm.donothing() [ "ExplicitUse"(ptr addrspace(3) @llvm.amdgcn.module.lds) ]80; CHECK-NEXT: [[LD:%.*]] = call i32 @get_func()81; CHECK-NEXT: [[MUL:%.*]] = mul i32 [[LD]], 482; CHECK-NEXT: call void @set_func(i32 [[MUL]])83; CHECK-NEXT: ret void84;85 %ld = call i32 @get_func()86 %mul = mul i32 %ld, 487 call void @set_func(i32 %mul)88 ret void89}90 91; CHECK: attributes #0 = { "amdgpu-lds-size"="8" }92