1; RUN: opt -S -mtriple=amdgcn-- -amdgpu-replace-lds-use-with-pointer -amdgpu-enable-lds-replace-with-pointer=true < %s | FileCheck %s 2 3; DESCRIPTION: 4; 5; There are three lds globals defined here, and these three lds are used respectively within 6; three non-kernel functions. There are three kernels, which *indirectly* call two of the 7; non-kernel functions. Hence pointer replacement should take place for all three lds, and 8; pointer initialization within kernel should selectively happen depending on which lds is 9; reachable from the kernel. 10; 11 12; Original LDS should exist. 13; CHECK: @lds_used_within_function_1 = internal addrspace(3) global [4 x i32] undef, align 4 14; CHECK: @lds_used_within_function_2 = internal addrspace(3) global [4 x i32] undef, align 4 15; CHECK: @lds_used_within_function_3 = internal addrspace(3) global [4 x i32] undef, align 4 16@lds_used_within_function_1 = internal addrspace(3) global [4 x i32] undef, align 4 17@lds_used_within_function_2 = internal addrspace(3) global [4 x i32] undef, align 4 18@lds_used_within_function_3 = internal addrspace(3) global [4 x i32] undef, align 4 19 20; Function pointers should exist. 21; CHECK: @ptr_to_func1 = internal local_unnamed_addr externally_initialized global void (float)* @function_1, align 8 22; CHECK: @ptr_to_func2 = internal local_unnamed_addr externally_initialized global void (i16)* @function_2, align 8 23; CHECK: @ptr_to_func3 = internal local_unnamed_addr externally_initialized global void (i8)* @function_3, align 8 24@ptr_to_func1 = internal local_unnamed_addr externally_initialized global void (float)* @function_1, align 8 25@ptr_to_func2 = internal local_unnamed_addr externally_initialized global void (i16)* @function_2, align 8 26@ptr_to_func3 = internal local_unnamed_addr externally_initialized global void (i8)* @function_3, align 8 27 28; Pointers should be created. 29; CHECK: @lds_used_within_function_1.ptr = internal unnamed_addr addrspace(3) global i16 undef, align 2 30; CHECK: @lds_used_within_function_2.ptr = internal unnamed_addr addrspace(3) global i16 undef, align 2 31; CHECK: @lds_used_within_function_3.ptr = internal unnamed_addr addrspace(3) global i16 undef, align 2 32 33; Pointer replacement code should be added. 34define internal void @function_3(i8 %c) { 35; CHECK-LABEL: entry: 36; CHECK: %0 = load i16, i16 addrspace(3)* @lds_used_within_function_3.ptr, align 2 37; CHECK: %1 = getelementptr i8, i8 addrspace(3)* null, i16 %0 38; CHECK: %2 = bitcast i8 addrspace(3)* %1 to [4 x i32] addrspace(3)* 39; CHECK: %gep = getelementptr inbounds [4 x i32], [4 x i32] addrspace(3)* %2, i32 0, i32 0 40; CHECK: ret void 41entry: 42 %gep = getelementptr inbounds [4 x i32], [4 x i32] addrspace(3)* @lds_used_within_function_3, i32 0, i32 0 43 ret void 44} 45 46; Pointer replacement code should be added. 47define internal void @function_2(i16 %i) { 48; CHECK-LABEL: entry: 49; CHECK: %0 = load i16, i16 addrspace(3)* @lds_used_within_function_2.ptr, align 2 50; CHECK: %1 = getelementptr i8, i8 addrspace(3)* null, i16 %0 51; CHECK: %2 = bitcast i8 addrspace(3)* %1 to [4 x i32] addrspace(3)* 52; CHECK: %gep = getelementptr inbounds [4 x i32], [4 x i32] addrspace(3)* %2, i32 0, i32 0 53; CHECK: ret void 54entry: 55 %gep = getelementptr inbounds [4 x i32], [4 x i32] addrspace(3)* @lds_used_within_function_2, i32 0, i32 0 56 ret void 57} 58 59; Pointer replacement code should be added. 60define internal void @function_1(float %f) { 61; CHECK-LABEL: entry: 62; CHECK: %0 = load i16, i16 addrspace(3)* @lds_used_within_function_1.ptr, align 2 63; CHECK: %1 = getelementptr i8, i8 addrspace(3)* null, i16 %0 64; CHECK: %2 = bitcast i8 addrspace(3)* %1 to [4 x i32] addrspace(3)* 65; CHECK: %gep = getelementptr inbounds [4 x i32], [4 x i32] addrspace(3)* %2, i32 0, i32 0 66; CHECK: ret void 67entry: 68 %gep = getelementptr inbounds [4 x i32], [4 x i32] addrspace(3)* @lds_used_within_function_1, i32 0, i32 0 69 ret void 70} 71 72; Pointer initialization code shoud be added 73define protected amdgpu_kernel void @kernel_calls_function_3_and_1() { 74; CHECK-LABEL: entry: 75; CHECK: %0 = call i32 @llvm.amdgcn.mbcnt.lo(i32 -1, i32 0) 76; CHECK: %1 = icmp eq i32 %0, 0 77; CHECK: br i1 %1, label %2, label %3 78; 79; CHECK-LABEL: 2: 80; CHECK: store i16 ptrtoint ([4 x i32] addrspace(3)* @lds_used_within_function_3 to i16), i16 addrspace(3)* @lds_used_within_function_3.ptr, align 2 81; CHECK: store i16 ptrtoint ([4 x i32] addrspace(3)* @lds_used_within_function_1 to i16), i16 addrspace(3)* @lds_used_within_function_1.ptr, align 2 82; CHECK: br label %3 83; 84; CHECK-LABEL: 3: 85; CHECK: call void @llvm.amdgcn.wave.barrier() 86; CHECK: %fptr3 = load void (i8)*, void (i8)** @ptr_to_func3, align 8 87; CHECK: %fptr1 = load void (float)*, void (float)** @ptr_to_func1, align 8 88; CHECK: call void %fptr3(i8 1) 89; CHECK: call void %fptr1(float 2.000000e+00) 90; CHECK: ret void 91entry: 92 %fptr3 = load void (i8)*, void (i8)** @ptr_to_func3, align 8 93 %fptr1 = load void (float)*, void (float)** @ptr_to_func1, align 8 94 call void %fptr3(i8 1) 95 call void %fptr1(float 2.0) 96 ret void 97} 98 99; Pointer initialization code shoud be added 100define protected amdgpu_kernel void @kernel_calls_function_2_and_3() { 101; CHECK-LABEL: entry: 102; CHECK: %0 = call i32 @llvm.amdgcn.mbcnt.lo(i32 -1, i32 0) 103; CHECK: %1 = icmp eq i32 %0, 0 104; CHECK: br i1 %1, label %2, label %3 105; 106; CHECK-LABEL: 2: 107; CHECK: store i16 ptrtoint ([4 x i32] addrspace(3)* @lds_used_within_function_3 to i16), i16 addrspace(3)* @lds_used_within_function_3.ptr, align 2 108; CHECK: store i16 ptrtoint ([4 x i32] addrspace(3)* @lds_used_within_function_2 to i16), i16 addrspace(3)* @lds_used_within_function_2.ptr, align 2 109; CHECK: br label %3 110; 111; CHECK-LABEL: 3: 112; CHECK: call void @llvm.amdgcn.wave.barrier() 113; CHECK: %fptr2 = load void (i16)*, void (i16)** @ptr_to_func2, align 8 114; CHECK: %fptr3 = load void (i8)*, void (i8)** @ptr_to_func3, align 8 115; CHECK: call void %fptr2(i16 3) 116; CHECK: call void %fptr3(i8 4) 117; CHECK: ret void 118entry: 119 %fptr2 = load void (i16)*, void (i16)** @ptr_to_func2, align 8 120 %fptr3 = load void (i8)*, void (i8)** @ptr_to_func3, align 8 121 call void %fptr2(i16 3) 122 call void %fptr3(i8 4) 123 ret void 124} 125 126; Pointer initialization code shoud be added 127define protected amdgpu_kernel void @kernel_calls_function_1_and_2() { 128; CHECK-LABEL: entry: 129; CHECK: %0 = call i32 @llvm.amdgcn.mbcnt.lo(i32 -1, i32 0) 130; CHECK: %1 = icmp eq i32 %0, 0 131; CHECK: br i1 %1, label %2, label %3 132; 133; CHECK-LABEL: 2: 134; CHECK: store i16 ptrtoint ([4 x i32] addrspace(3)* @lds_used_within_function_2 to i16), i16 addrspace(3)* @lds_used_within_function_2.ptr, align 2 135; CHECK: store i16 ptrtoint ([4 x i32] addrspace(3)* @lds_used_within_function_1 to i16), i16 addrspace(3)* @lds_used_within_function_1.ptr, align 2 136; CHECK: br label %3 137; 138; CHECK-LABEL: 3: 139; CHECK: call void @llvm.amdgcn.wave.barrier() 140; CHECK: %fptr1 = load void (float)*, void (float)** @ptr_to_func1, align 8 141; CHECK: %fptr2 = load void (i16)*, void (i16)** @ptr_to_func2, align 8 142; CHECK: call void %fptr1(float 5.000000e+00) 143; CHECK: call void %fptr2(i16 6) 144; CHECK: ret void 145entry: 146 %fptr1 = load void (float)*, void (float)** @ptr_to_func1, align 8 147 %fptr2 = load void (i16)*, void (i16)** @ptr_to_func2, align 8 148 call void %fptr1(float 5.0) 149 call void %fptr2(i16 6) 150 ret void 151} 152