xref: /llvm-project/llvm/test/CodeGen/AMDGPU/lower-module-lds-single-var-ambiguous.ll (revision 6fed2c90d3b9fabe8742ae1308632c06dd4b76ea)
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: opt -S -mtriple=amdgcn-- -passes=amdgpu-lower-module-lds < %s --amdgpu-lower-module-lds-strategy=module | FileCheck -check-prefixes=CHECK,M_OR_HY %s
3; RUN: opt -S -mtriple=amdgcn-- -passes=amdgpu-lower-module-lds < %s --amdgpu-lower-module-lds-strategy=table | FileCheck -check-prefixes=CHECK,TABLE %s
4; RUN: not --crash opt -S -mtriple=amdgcn-- -passes=amdgpu-lower-module-lds < %s --amdgpu-lower-module-lds-strategy=kernel 2>&1 | FileCheck -check-prefixes=KERNEL %s
5; RUN: opt -S -mtriple=amdgcn-- -passes=amdgpu-lower-module-lds < %s --amdgpu-lower-module-lds-strategy=hybrid | FileCheck -check-prefixes=CHECK,M_OR_HY %s
6
7;; Two kernels access the same variable, specialisation gives them each their own copy of it
8
9@kernel.lds = addrspace(3) global i8 undef
10define amdgpu_kernel void @k0() {
11; CHECK-LABEL: @k0(
12; CHECK-NEXT:    [[LD:%.*]] = load i8, ptr addrspace(3) @llvm.amdgcn.kernel.k0.lds, align 1
13; CHECK-NEXT:    [[MUL:%.*]] = mul i8 [[LD]], 2
14; CHECK-NEXT:    store i8 [[MUL]], ptr addrspace(3) @llvm.amdgcn.kernel.k0.lds, align 1
15; CHECK-NEXT:    ret void
16;
17  %ld = load i8, ptr addrspace(3) @kernel.lds
18  %mul = mul i8 %ld, 2
19  store i8 %mul, ptr  addrspace(3) @kernel.lds
20  ret void
21}
22
23define amdgpu_kernel void @k1() {
24; CHECK-LABEL: @k1(
25; CHECK-NEXT:    [[LD:%.*]] = load i8, ptr addrspace(3) @llvm.amdgcn.kernel.k1.lds, align 1
26; CHECK-NEXT:    [[MUL:%.*]] = mul i8 [[LD]], 3
27; CHECK-NEXT:    store i8 [[MUL]], ptr addrspace(3) @llvm.amdgcn.kernel.k1.lds, align 1
28; CHECK-NEXT:    ret void
29;
30  %ld = load i8, ptr addrspace(3) @kernel.lds
31  %mul = mul i8 %ld, 3
32  store i8 %mul, ptr  addrspace(3) @kernel.lds
33  ret void
34}
35
36;; Function accesses variable, reachable from two kernels, can't use kernel lowering for either
37;; Hybrid can put it in module lds without cost as the first variable is free
38
39; KERNEL: LLVM ERROR: cannot lower LDS 'function.lds' to kernel access as it is reachable from multiple kernels
40
41@function.lds = addrspace(3) global i16 undef
42define void @f0() {
43; M_OR_HY-LABEL: @f0(
44; M_OR_HY-NEXT:    [[LD:%.*]] = load i16, ptr addrspace(3) @llvm.amdgcn.module.lds, align 2
45; M_OR_HY-NEXT:    [[MUL:%.*]] = mul i16 [[LD]], 4
46; M_OR_HY-NEXT:    store i16 [[MUL]], ptr addrspace(3) @llvm.amdgcn.module.lds, align 2
47; M_OR_HY-NEXT:    ret void
48;
49; TABLE-LABEL: @f0(
50; TABLE-NEXT:    [[TMP1:%.*]] = call i32 @llvm.amdgcn.lds.kernel.id()
51; TABLE-NEXT:    [[FUNCTION_LDS2:%.*]] = getelementptr inbounds [2 x [1 x i32]], ptr addrspace(4) @llvm.amdgcn.lds.offset.table, i32 0, i32 [[TMP1]], i32 0
52; TABLE-NEXT:    [[TMP2:%.*]] = load i32, ptr addrspace(4) [[FUNCTION_LDS2]], align 4
53; TABLE-NEXT:    [[FUNCTION_LDS3:%.*]] = inttoptr i32 [[TMP2]] to ptr addrspace(3)
54; TABLE-NEXT:    [[LD:%.*]] = load i16, ptr addrspace(3) [[FUNCTION_LDS3]], align 2
55; TABLE-NEXT:    [[MUL:%.*]] = mul i16 [[LD]], 4
56; TABLE-NEXT:    [[FUNCTION_LDS:%.*]] = getelementptr inbounds [2 x [1 x i32]], ptr addrspace(4) @llvm.amdgcn.lds.offset.table, i32 0, i32 [[TMP1]], i32 0
57; TABLE-NEXT:    [[TMP3:%.*]] = load i32, ptr addrspace(4) [[FUNCTION_LDS]], align 4
58; TABLE-NEXT:    [[FUNCTION_LDS1:%.*]] = inttoptr i32 [[TMP3]] to ptr addrspace(3)
59; TABLE-NEXT:    store i16 [[MUL]], ptr addrspace(3) [[FUNCTION_LDS1]], align 2
60; TABLE-NEXT:    ret void
61;
62  %ld = load i16, ptr addrspace(3) @function.lds
63  %mul = mul i16 %ld, 4
64  store i16 %mul, ptr  addrspace(3) @function.lds
65  ret void
66}
67
68
69define amdgpu_kernel void @k0_f0() {
70; M_OR_HY-LABEL: @k0_f0(
71; M_OR_HY-NEXT:    call void @llvm.donothing() [ "ExplicitUse"(ptr addrspace(3) @llvm.amdgcn.module.lds) ]
72; M_OR_HY-NEXT:    call void @f0()
73; M_OR_HY-NEXT:    ret void
74;
75; TABLE-LABEL: @k0_f0(
76; TABLE-NEXT:    call void @llvm.donothing() [ "ExplicitUse"(ptr addrspace(3) @llvm.amdgcn.kernel.k0_f0.lds) ]
77; TABLE-NEXT:    call void @f0()
78; TABLE-NEXT:    ret void
79;
80  call void @f0()
81  ret void
82}
83
84define amdgpu_kernel void @k1_f0() {
85; M_OR_HY-LABEL: @k1_f0(
86; M_OR_HY-NEXT:    call void @llvm.donothing() [ "ExplicitUse"(ptr addrspace(3) @llvm.amdgcn.module.lds) ]
87; M_OR_HY-NEXT:    call void @f0()
88; M_OR_HY-NEXT:    ret void
89;
90; TABLE-LABEL: @k1_f0(
91; TABLE-NEXT:    call void @llvm.donothing() [ "ExplicitUse"(ptr addrspace(3) @llvm.amdgcn.kernel.k1_f0.lds) ]
92; TABLE-NEXT:    call void @f0()
93; TABLE-NEXT:    ret void
94;
95  call void @f0()
96  ret void
97}
98