xref: /llvm-project/llvm/test/Transforms/LowerMatrixIntrinsics/store-align-volatile.ll (revision 256d5ad3e8db42e125e4954fadeba2b37421d91f) 
1; RUN: opt -passes='lower-matrix-intrinsics' -S < %s | FileCheck %s
2
3define void @strided_store_volatile(<6 x i32> %in, ptr %out) {
4; CHECK-LABEL: @strided_store_volatile(
5; CHECK-NEXT:    [[SPLIT:%.*]] = shufflevector <6 x i32> [[IN:%.*]], <6 x i32> poison, <3 x i32> <i32 0, i32 1, i32 2>
6; CHECK-NEXT:    [[SPLIT1:%.*]] = shufflevector <6 x i32> [[IN]], <6 x i32> poison, <3 x i32> <i32 3, i32 4, i32 5>
7; CHECK-NEXT:    store volatile <3 x i32> [[SPLIT]], ptr [[OUT:%.*]], align 4
8; CHECK-NEXT:    [[VEC_GEP:%.*]] = getelementptr i32, ptr [[OUT]], i64 5
9; CHECK-NEXT:    store volatile <3 x i32> [[SPLIT1]], ptr [[VEC_GEP]], align 4
10; CHECK-NEXT:    ret void
11;
12  call void @llvm.matrix.column.major.store(<6 x i32> %in, ptr %out, i64 5, i1 true, i32 3, i32 2)
13  ret void
14}
15
16declare void @llvm.matrix.column.major.store(<6 x i32>, ptr, i64, i1, i32, i32)
17
18
19define void @multiply_store_volatile(<4 x i32> %in, ptr %out) {
20; CHECK-LABEL: @multiply_store_volatile(
21; CHECK:    store volatile <2 x i32> {{.*}}, ptr %out, align 4
22; CHECK-NEXT:    [[VEC_GEP:%.*]] = getelementptr i32, ptr %out, i64 2
23; CHECK-NEXT:    store volatile <2 x i32> {{.*}}, ptr [[VEC_GEP]], align 4
24; CHECK-NEXT:    ret void
25;
26  %res = call <4 x i32> @llvm.matrix.multiply(<4 x i32> %in, <4 x i32> %in, i32 2, i32 2, i32 2)
27  store volatile <4 x i32> %res, ptr %out, align 4
28  ret void
29}
30
31declare <4 x i32> @llvm.matrix.multiply(<4 x i32>, <4 x i32>, i32, i32, i32)
32
33define void @strided_store_align32(<6 x i32> %in, i64 %stride, ptr %out) {
34; CHECK-LABEL: @strided_store_align32(
35; CHECK-NEXT:    [[SPLIT:%.*]] = shufflevector <6 x i32> [[IN:%.*]], <6 x i32> poison, <3 x i32> <i32 0, i32 1, i32 2>
36; CHECK-NEXT:    [[SPLIT1:%.*]] = shufflevector <6 x i32> [[IN]], <6 x i32> poison, <3 x i32> <i32 3, i32 4, i32 5>
37; CHECK-NEXT:    [[VEC_START:%.*]] = mul i64 0, [[STRIDE:%.*]]
38; CHECK-NEXT:    [[VEC_GEP:%.*]] = getelementptr i32, ptr [[OUT:%.*]], i64 [[VEC_START]]
39; CHECK-NEXT:    store volatile <3 x i32> [[SPLIT]], ptr [[VEC_GEP]], align 32
40; CHECK-NEXT:    [[VEC_START2:%.*]] = mul i64 1, [[STRIDE]]
41; CHECK-NEXT:    [[VEC_GEP3:%.*]] = getelementptr i32, ptr [[OUT]], i64 [[VEC_START2]]
42; CHECK-NEXT:    store volatile <3 x i32> [[SPLIT1]], ptr [[VEC_GEP3]], align 4
43; CHECK-NEXT:    ret void
44;
45  call void @llvm.matrix.column.major.store(<6 x i32> %in, ptr align 32 %out, i64 %stride, i1 true, i32 3, i32 2)
46  ret void
47}
48
49define void @strided_store_align2(<6 x i32> %in, i64 %stride, ptr %out) {
50; CHECK-LABEL: @strided_store_align2(
51; CHECK-NEXT:    [[SPLIT:%.*]] = shufflevector <6 x i32> [[IN:%.*]], <6 x i32> poison, <3 x i32> <i32 0, i32 1, i32 2>
52; CHECK-NEXT:    [[SPLIT1:%.*]] = shufflevector <6 x i32> [[IN]], <6 x i32> poison, <3 x i32> <i32 3, i32 4, i32 5>
53; CHECK-NEXT:    [[VEC_START:%.*]] = mul i64 0, [[STRIDE:%.*]]
54; CHECK-NEXT:    [[VEC_GEP:%.*]] = getelementptr i32, ptr [[OUT:%.*]], i64 [[VEC_START]]
55; CHECK-NEXT:    store volatile <3 x i32> [[SPLIT]], ptr [[VEC_GEP]], align 2
56; CHECK-NEXT:    [[VEC_START2:%.*]] = mul i64 1, [[STRIDE]]
57; CHECK-NEXT:    [[VEC_GEP3:%.*]] = getelementptr i32, ptr [[OUT]], i64 [[VEC_START2]]
58; CHECK-NEXT:    store volatile <3 x i32> [[SPLIT1]], ptr [[VEC_GEP3]], align 2
59; CHECK-NEXT:    ret void
60;
61  call void @llvm.matrix.column.major.store(<6 x i32> %in, ptr align 2 %out, i64 %stride, i1 true, i32 3, i32 2)
62  ret void
63}
64
65define void @multiply_store_align16_stride8(<4 x i32> %in, ptr %out) {
66; CHECK-LABEL: @multiply_store_align16_stride8(
67; CHECK:    store <2 x i32> {{.*}}, ptr %out, align 16
68; CHECK-NEXT:    [[VEC_GEP:%.*]] = getelementptr i32, ptr %out, i64 2
69; CHECK-NEXT:    store <2 x i32> {{.*}}, ptr [[VEC_GEP]], align 8
70; CHECK-NEXT:    ret void
71;
72  %res = call <4 x i32> @llvm.matrix.multiply(<4 x i32> %in, <4 x i32> %in, i32 2, i32 2, i32 2)
73  store <4 x i32> %res, ptr %out, align 16
74  ret void
75}
76
77define void @strided_store_align8_stride12(<6 x i32> %in, ptr %out) {
78; CHECK-LABEL: @strided_store_align8_stride12(
79; CHECK-NEXT:    [[SPLIT:%.*]] = shufflevector <6 x i32> [[IN:%.*]], <6 x i32> poison, <2 x i32> <i32 0, i32 1>
80; CHECK-NEXT:    [[SPLIT1:%.*]] = shufflevector <6 x i32> [[IN]], <6 x i32> poison, <2 x i32> <i32 2, i32 3>
81; CHECK-NEXT:    [[SPLIT2:%.*]] = shufflevector <6 x i32> [[IN]], <6 x i32> poison, <2 x i32> <i32 4, i32 5>
82; CHECK-NEXT:    store <2 x i32> [[SPLIT]], ptr [[OUT:%.*]], align 8
83; CHECK-NEXT:    [[VEC_GEP:%.*]] = getelementptr i32, ptr [[OUT]], i64 3
84; CHECK-NEXT:    store <2 x i32> [[SPLIT1]], ptr [[VEC_GEP]], align 4
85; CHECK-NEXT:    [[VEC_GEP4:%.*]] = getelementptr i32, ptr [[OUT]], i64 6
86; CHECK-NEXT:    store <2 x i32> [[SPLIT2]], ptr [[VEC_GEP4]], align 8
87; CHECK-NEXT:    ret void
88;
89  call void @llvm.matrix.column.major.store(<6 x i32> %in, ptr align 8 %out, i64 3, i1 false, i32 2, i32 3)
90  ret void
91}
92