1 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2 ; RUN: opt -passes='lower-matrix-intrinsics' -S < %s | FileCheck %s
4 define <9 x double> @strided_load_3x3(ptr %in, i64 %stride) {
5 ; CHECK-LABEL: @strided_load_3x3(
7 ; CHECK-NEXT: [[VEC_START:%.*]] = mul i64 0, [[STRIDE:%.*]]
8 ; CHECK-NEXT: [[VEC_GEP:%.*]] = getelementptr double, ptr [[IN:%.*]], i64 [[VEC_START]]
9 ; CHECK-NEXT: [[COL_LOAD:%.*]] = load <3 x double>, ptr [[VEC_GEP]], align 8
10 ; CHECK-NEXT: [[VEC_START1:%.*]] = mul i64 1, [[STRIDE]]
11 ; CHECK-NEXT: [[VEC_GEP2:%.*]] = getelementptr double, ptr [[IN]], i64 [[VEC_START1]]
12 ; CHECK-NEXT: [[COL_LOAD4:%.*]] = load <3 x double>, ptr [[VEC_GEP2]], align 8
13 ; CHECK-NEXT: [[VEC_START5:%.*]] = mul i64 2, [[STRIDE]]
14 ; CHECK-NEXT: [[VEC_GEP6:%.*]] = getelementptr double, ptr [[IN]], i64 [[VEC_START5]]
15 ; CHECK-NEXT: [[COL_LOAD8:%.*]] = load <3 x double>, ptr [[VEC_GEP6]], align 8
16 ; CHECK-NEXT: [[TMP0:%.*]] = shufflevector <3 x double> [[COL_LOAD]], <3 x double> [[COL_LOAD4]], <6 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5>
17 ; CHECK-NEXT: [[TMP1:%.*]] = shufflevector <3 x double> [[COL_LOAD8]], <3 x double> poison, <6 x i32> <i32 0, i32 1, i32 2, i32 poison, i32 poison, i32 poison>
18 ; CHECK-NEXT: [[TMP2:%.*]] = shufflevector <6 x double> [[TMP0]], <6 x double> [[TMP1]], <9 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8>
19 ; CHECK-NEXT: ret <9 x double> [[TMP2]]
22 %load = call <9 x double> @llvm.matrix.column.major.load.v9f64.i64(ptr %in, i64 %stride, i1 false, i32 3, i32 3)
23 ret <9 x double> %load
26 declare <9 x double> @llvm.matrix.column.major.load.v9f64.i64(ptr, i64, i1, i32, i32)
28 define <9 x double> @strided_load_9x1(ptr %in, i64 %stride) {
29 ; CHECK-LABEL: @strided_load_9x1(
31 ; CHECK-NEXT: [[VEC_START:%.*]] = mul i64 0, [[STRIDE:%.*]]
32 ; CHECK-NEXT: [[VEC_GEP:%.*]] = getelementptr double, ptr [[IN:%.*]], i64 [[VEC_START]]
33 ; CHECK-NEXT: [[COL_LOAD:%.*]] = load <9 x double>, ptr [[VEC_GEP]], align 8
34 ; CHECK-NEXT: ret <9 x double> [[COL_LOAD]]
37 %load = call <9 x double> @llvm.matrix.column.major.load.v9f64.i64(ptr %in, i64 %stride, i1 false, i32 9, i32 1)
38 ret <9 x double> %load
41 declare <8 x double> @llvm.matrix.column.major.load.v8f64.i64(ptr, i64, i1, i32, i32)
43 define <8 x double> @strided_load_4x2(ptr %in, i64 %stride) {
44 ; CHECK-LABEL: @strided_load_4x2(
46 ; CHECK-NEXT: [[VEC_START:%.*]] = mul i64 0, [[STRIDE:%.*]]
47 ; CHECK-NEXT: [[VEC_GEP:%.*]] = getelementptr double, ptr [[IN:%.*]], i64 [[VEC_START]]
48 ; CHECK-NEXT: [[COL_LOAD:%.*]] = load <4 x double>, ptr [[VEC_GEP]], align 8
49 ; CHECK-NEXT: [[VEC_START1:%.*]] = mul i64 1, [[STRIDE]]
50 ; CHECK-NEXT: [[VEC_GEP2:%.*]] = getelementptr double, ptr [[IN]], i64 [[VEC_START1]]
51 ; CHECK-NEXT: [[COL_LOAD4:%.*]] = load <4 x double>, ptr [[VEC_GEP2]], align 8
52 ; CHECK-NEXT: [[TMP0:%.*]] = shufflevector <4 x double> [[COL_LOAD]], <4 x double> [[COL_LOAD4]], <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
53 ; CHECK-NEXT: ret <8 x double> [[TMP0]]
56 %load = call <8 x double> @llvm.matrix.column.major.load.v8f64.i64(ptr %in, i64 %stride, i1 false, i32 4, i32 2)
57 ret <8 x double> %load
60 declare <8 x double> @llvm.matrix.column.major.load.v8f64.i32(ptr, i32, i1, i32, i32)
62 define <8 x double> @strided_load_4x2_stride_i32(ptr %in, i32 %stride) {
63 ; CHECK-LABEL: @strided_load_4x2_stride_i32(
65 ; CHECK-NEXT: [[VEC_START:%.*]] = mul i32 0, [[STRIDE:%.*]]
66 ; CHECK-NEXT: [[VEC_GEP:%.*]] = getelementptr double, ptr [[IN:%.*]], i32 [[VEC_START]]
67 ; CHECK-NEXT: [[COL_LOAD:%.*]] = load <4 x double>, ptr [[VEC_GEP]], align 8
68 ; CHECK-NEXT: [[VEC_START1:%.*]] = mul i32 1, [[STRIDE]]
69 ; CHECK-NEXT: [[VEC_GEP2:%.*]] = getelementptr double, ptr [[IN]], i32 [[VEC_START1]]
70 ; CHECK-NEXT: [[COL_LOAD4:%.*]] = load <4 x double>, ptr [[VEC_GEP2]], align 8
71 ; CHECK-NEXT: [[TMP0:%.*]] = shufflevector <4 x double> [[COL_LOAD]], <4 x double> [[COL_LOAD4]], <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
72 ; CHECK-NEXT: ret <8 x double> [[TMP0]]
75 %load = call <8 x double> @llvm.matrix.column.major.load.v8f64.i32(ptr %in, i32 %stride, i1 false, i32 4, i32 2)
76 ret <8 x double> %load