| blob | 8ad9e2e77860a036fff97f208bb82d921438a551 |
1 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
3 ; RUN: opt -passes=lower-matrix-intrinsics -matrix-default-layout=row-major -S < %s | FileCheck --check-prefix=RM %s
7 define <4 x i32> @multiply_2x2(<4 x i32> %a, <4 x i32> %b) {
8 ; RM-LABEL: @multiply_2x2(
9 ; RM-NEXT: entry:
10 ; RM-NEXT: [[SPLIT:%.*]] = shufflevector <4 x i32> [[A:%.*]], <4 x i32> poison, <2 x i32> <i32 0, i32 1>
11 ; RM-NEXT: [[SPLIT1:%.*]] = shufflevector <4 x i32> [[A]], <4 x i32> poison, <2 x i32> <i32 2, i32 3>
12 ; RM-NEXT: [[SPLIT2:%.*]] = shufflevector <4 x i32> [[B:%.*]], <4 x i32> poison, <2 x i32> <i32 0, i32 1>
13 ; RM-NEXT: [[SPLIT3:%.*]] = shufflevector <4 x i32> [[B]], <4 x i32> poison, <2 x i32> <i32 2, i32 3>
14 ; RM-NEXT: [[BLOCK:%.*]] = shufflevector <2 x i32> [[SPLIT2]], <2 x i32> poison, <1 x i32> zeroinitializer
15 ; RM-NEXT: [[TMP0:%.*]] = extractelement <2 x i32> [[SPLIT]], i64 0
16 ; RM-NEXT: [[SPLAT_SPLATINSERT:%.*]] = insertelement <1 x i32> poison, i32 [[TMP0]], i64 0
17 ; RM-NEXT: [[SPLAT_SPLAT:%.*]] = shufflevector <1 x i32> [[SPLAT_SPLATINSERT]], <1 x i32> poison, <1 x i32> zeroinitializer
18 ; RM-NEXT: [[TMP1:%.*]] = mul <1 x i32> [[SPLAT_SPLAT]], [[BLOCK]]
19 ; RM-NEXT: [[BLOCK4:%.*]] = shufflevector <2 x i32> [[SPLIT3]], <2 x i32> poison, <1 x i32> zeroinitializer
20 ; RM-NEXT: [[TMP2:%.*]] = extractelement <2 x i32> [[SPLIT]], i64 1
21 ; RM-NEXT: [[SPLAT_SPLATINSERT5:%.*]] = insertelement <1 x i32> poison, i32 [[TMP2]], i64 0
22 ; RM-NEXT: [[SPLAT_SPLAT6:%.*]] = shufflevector <1 x i32> [[SPLAT_SPLATINSERT5]], <1 x i32> poison, <1 x i32> zeroinitializer
23 ; RM-NEXT: [[TMP3:%.*]] = mul <1 x i32> [[SPLAT_SPLAT6]], [[BLOCK4]]
24 ; RM-NEXT: [[TMP4:%.*]] = add <1 x i32> [[TMP1]], [[TMP3]]
25 ; RM-NEXT: [[TMP5:%.*]] = shufflevector <1 x i32> [[TMP4]], <1 x i32> poison, <2 x i32> <i32 0, i32 poison>
26 ; RM-NEXT: [[TMP6:%.*]] = shufflevector <2 x i32> poison, <2 x i32> [[TMP5]], <2 x i32> <i32 2, i32 1>
27 ; RM-NEXT: [[BLOCK7:%.*]] = shufflevector <2 x i32> [[SPLIT2]], <2 x i32> poison, <1 x i32> <i32 1>
28 ; RM-NEXT: [[TMP7:%.*]] = extractelement <2 x i32> [[SPLIT]], i64 0
29 ; RM-NEXT: [[SPLAT_SPLATINSERT8:%.*]] = insertelement <1 x i32> poison, i32 [[TMP7]], i64 0
30 ; RM-NEXT: [[SPLAT_SPLAT9:%.*]] = shufflevector <1 x i32> [[SPLAT_SPLATINSERT8]], <1 x i32> poison, <1 x i32> zeroinitializer
31 ; RM-NEXT: [[TMP8:%.*]] = mul <1 x i32> [[SPLAT_SPLAT9]], [[BLOCK7]]
32 ; RM-NEXT: [[BLOCK10:%.*]] = shufflevector <2 x i32> [[SPLIT3]], <2 x i32> poison, <1 x i32> <i32 1>
33 ; RM-NEXT: [[TMP9:%.*]] = extractelement <2 x i32> [[SPLIT]], i64 1
34 ; RM-NEXT: [[SPLAT_SPLATINSERT11:%.*]] = insertelement <1 x i32> poison, i32 [[TMP9]], i64 0
35 ; RM-NEXT: [[SPLAT_SPLAT12:%.*]] = shufflevector <1 x i32> [[SPLAT_SPLATINSERT11]], <1 x i32> poison, <1 x i32> zeroinitializer
36 ; RM-NEXT: [[TMP10:%.*]] = mul <1 x i32> [[SPLAT_SPLAT12]], [[BLOCK10]]
37 ; RM-NEXT: [[TMP11:%.*]] = add <1 x i32> [[TMP8]], [[TMP10]]
38 ; RM-NEXT: [[TMP12:%.*]] = shufflevector <1 x i32> [[TMP11]], <1 x i32> poison, <2 x i32> <i32 0, i32 poison>
39 ; RM-NEXT: [[TMP13:%.*]] = shufflevector <2 x i32> [[TMP6]], <2 x i32> [[TMP12]], <2 x i32> <i32 0, i32 2>
40 ; RM-NEXT: [[BLOCK13:%.*]] = shufflevector <2 x i32> [[SPLIT2]], <2 x i32> poison, <1 x i32> zeroinitializer
41 ; RM-NEXT: [[TMP14:%.*]] = extractelement <2 x i32> [[SPLIT1]], i64 0
42 ; RM-NEXT: [[SPLAT_SPLATINSERT14:%.*]] = insertelement <1 x i32> poison, i32 [[TMP14]], i64 0
43 ; RM-NEXT: [[SPLAT_SPLAT15:%.*]] = shufflevector <1 x i32> [[SPLAT_SPLATINSERT14]], <1 x i32> poison, <1 x i32> zeroinitializer
44 ; RM-NEXT: [[TMP15:%.*]] = mul <1 x i32> [[SPLAT_SPLAT15]], [[BLOCK13]]
45 ; RM-NEXT: [[BLOCK16:%.*]] = shufflevector <2 x i32> [[SPLIT3]], <2 x i32> poison, <1 x i32> zeroinitializer
46 ; RM-NEXT: [[TMP16:%.*]] = extractelement <2 x i32> [[SPLIT1]], i64 1
47 ; RM-NEXT: [[SPLAT_SPLATINSERT17:%.*]] = insertelement <1 x i32> poison, i32 [[TMP16]], i64 0
48 ; RM-NEXT: [[SPLAT_SPLAT18:%.*]] = shufflevector <1 x i32> [[SPLAT_SPLATINSERT17]], <1 x i32> poison, <1 x i32> zeroinitializer
49 ; RM-NEXT: [[TMP17:%.*]] = mul <1 x i32> [[SPLAT_SPLAT18]], [[BLOCK16]]
50 ; RM-NEXT: [[TMP18:%.*]] = add <1 x i32> [[TMP15]], [[TMP17]]
51 ; RM-NEXT: [[TMP19:%.*]] = shufflevector <1 x i32> [[TMP18]], <1 x i32> poison, <2 x i32> <i32 0, i32 poison>
52 ; RM-NEXT: [[TMP20:%.*]] = shufflevector <2 x i32> poison, <2 x i32> [[TMP19]], <2 x i32> <i32 2, i32 1>
53 ; RM-NEXT: [[BLOCK19:%.*]] = shufflevector <2 x i32> [[SPLIT2]], <2 x i32> poison, <1 x i32> <i32 1>
54 ; RM-NEXT: [[TMP21:%.*]] = extractelement <2 x i32> [[SPLIT1]], i64 0
55 ; RM-NEXT: [[SPLAT_SPLATINSERT20:%.*]] = insertelement <1 x i32> poison, i32 [[TMP21]], i64 0
56 ; RM-NEXT: [[SPLAT_SPLAT21:%.*]] = shufflevector <1 x i32> [[SPLAT_SPLATINSERT20]], <1 x i32> poison, <1 x i32> zeroinitializer
57 ; RM-NEXT: [[TMP22:%.*]] = mul <1 x i32> [[SPLAT_SPLAT21]], [[BLOCK19]]
58 ; RM-NEXT: [[BLOCK22:%.*]] = shufflevector <2 x i32> [[SPLIT3]], <2 x i32> poison, <1 x i32> <i32 1>
59 ; RM-NEXT: [[TMP23:%.*]] = extractelement <2 x i32> [[SPLIT1]], i64 1
60 ; RM-NEXT: [[SPLAT_SPLATINSERT23:%.*]] = insertelement <1 x i32> poison, i32 [[TMP23]], i64 0
61 ; RM-NEXT: [[SPLAT_SPLAT24:%.*]] = shufflevector <1 x i32> [[SPLAT_SPLATINSERT23]], <1 x i32> poison, <1 x i32> zeroinitializer
62 ; RM-NEXT: [[TMP24:%.*]] = mul <1 x i32> [[SPLAT_SPLAT24]], [[BLOCK22]]
63 ; RM-NEXT: [[TMP25:%.*]] = add <1 x i32> [[TMP22]], [[TMP24]]
64 ; RM-NEXT: [[TMP26:%.*]] = shufflevector <1 x i32> [[TMP25]], <1 x i32> poison, <2 x i32> <i32 0, i32 poison>
65 ; RM-NEXT: [[TMP27:%.*]] = shufflevector <2 x i32> [[TMP20]], <2 x i32> [[TMP26]], <2 x i32> <i32 0, i32 2>
66 ; RM-NEXT: [[TMP28:%.*]] = shufflevector <2 x i32> [[TMP13]], <2 x i32> [[TMP27]], <4 x i32> <i32 0, i32 1, i32 2, i32 3>
67 ; RM-NEXT: ret <4 x i32> [[TMP28]]
68 ;
69 entry:
70 %c = call <4 x i32> @llvm.matrix.multiply.v4i32.v4i32.v4i32(<4 x i32> %a, <4 x i32> %b, i32 2, i32 2, i32 2)
71 ret <4 x i32> %c
72 }
74 declare <4 x i32> @llvm.matrix.multiply.v4i32.v4i32.v4i32(<4 x i32>, <4 x i32>, i32, i32, i32)
76 define <4 x i32> @multiply_1x2(<2 x i32> %a, <2 x i32> %b) {
77 ; RM-LABEL: @multiply_1x2(
78 ; RM-NEXT: entry:
79 ; RM-NEXT: [[SPLIT:%.*]] = shufflevector <2 x i32> [[A:%.*]], <2 x i32> poison, <1 x i32> zeroinitializer
80 ; RM-NEXT: [[SPLIT1:%.*]] = shufflevector <2 x i32> [[A]], <2 x i32> poison, <1 x i32> <i32 1>
81 ; RM-NEXT: [[SPLIT2:%.*]] = shufflevector <2 x i32> [[B:%.*]], <2 x i32> poison, <2 x i32> <i32 0, i32 1>
82 ; RM-NEXT: [[BLOCK:%.*]] = shufflevector <2 x i32> [[SPLIT2]], <2 x i32> poison, <1 x i32> zeroinitializer
83 ; RM-NEXT: [[TMP0:%.*]] = extractelement <1 x i32> [[SPLIT]], i64 0
84 ; RM-NEXT: [[SPLAT_SPLATINSERT:%.*]] = insertelement <1 x i32> poison, i32 [[TMP0]], i64 0
85 ; RM-NEXT: [[SPLAT_SPLAT:%.*]] = shufflevector <1 x i32> [[SPLAT_SPLATINSERT]], <1 x i32> poison, <1 x i32> zeroinitializer
86 ; RM-NEXT: [[TMP1:%.*]] = mul <1 x i32> [[SPLAT_SPLAT]], [[BLOCK]]
87 ; RM-NEXT: [[TMP2:%.*]] = shufflevector <1 x i32> [[TMP1]], <1 x i32> poison, <2 x i32> <i32 0, i32 poison>
88 ; RM-NEXT: [[TMP3:%.*]] = shufflevector <2 x i32> poison, <2 x i32> [[TMP2]], <2 x i32> <i32 2, i32 1>
89 ; RM-NEXT: [[BLOCK3:%.*]] = shufflevector <2 x i32> [[SPLIT2]], <2 x i32> poison, <1 x i32> <i32 1>
90 ; RM-NEXT: [[TMP4:%.*]] = extractelement <1 x i32> [[SPLIT]], i64 0
91 ; RM-NEXT: [[SPLAT_SPLATINSERT4:%.*]] = insertelement <1 x i32> poison, i32 [[TMP4]], i64 0
92 ; RM-NEXT: [[SPLAT_SPLAT5:%.*]] = shufflevector <1 x i32> [[SPLAT_SPLATINSERT4]], <1 x i32> poison, <1 x i32> zeroinitializer
93 ; RM-NEXT: [[TMP5:%.*]] = mul <1 x i32> [[SPLAT_SPLAT5]], [[BLOCK3]]
94 ; RM-NEXT: [[TMP6:%.*]] = shufflevector <1 x i32> [[TMP5]], <1 x i32> poison, <2 x i32> <i32 0, i32 poison>
95 ; RM-NEXT: [[TMP7:%.*]] = shufflevector <2 x i32> [[TMP3]], <2 x i32> [[TMP6]], <2 x i32> <i32 0, i32 2>
96 ; RM-NEXT: [[BLOCK6:%.*]] = shufflevector <2 x i32> [[SPLIT2]], <2 x i32> poison, <1 x i32> zeroinitializer
97 ; RM-NEXT: [[TMP8:%.*]] = extractelement <1 x i32> [[SPLIT1]], i64 0
98 ; RM-NEXT: [[SPLAT_SPLATINSERT7:%.*]] = insertelement <1 x i32> poison, i32 [[TMP8]], i64 0
99 ; RM-NEXT: [[SPLAT_SPLAT8:%.*]] = shufflevector <1 x i32> [[SPLAT_SPLATINSERT7]], <1 x i32> poison, <1 x i32> zeroinitializer
100 ; RM-NEXT: [[TMP9:%.*]] = mul <1 x i32> [[SPLAT_SPLAT8]], [[BLOCK6]]
101 ; RM-NEXT: [[TMP10:%.*]] = shufflevector <1 x i32> [[TMP9]], <1 x i32> poison, <2 x i32> <i32 0, i32 poison>
102 ; RM-NEXT: [[TMP11:%.*]] = shufflevector <2 x i32> poison, <2 x i32> [[TMP10]], <2 x i32> <i32 2, i32 1>
103 ; RM-NEXT: [[BLOCK9:%.*]] = shufflevector <2 x i32> [[SPLIT2]], <2 x i32> poison, <1 x i32> <i32 1>
104 ; RM-NEXT: [[TMP12:%.*]] = extractelement <1 x i32> [[SPLIT1]], i64 0
105 ; RM-NEXT: [[SPLAT_SPLATINSERT10:%.*]] = insertelement <1 x i32> poison, i32 [[TMP12]], i64 0
106 ; RM-NEXT: [[SPLAT_SPLAT11:%.*]] = shufflevector <1 x i32> [[SPLAT_SPLATINSERT10]], <1 x i32> poison, <1 x i32> zeroinitializer
107 ; RM-NEXT: [[TMP13:%.*]] = mul <1 x i32> [[SPLAT_SPLAT11]], [[BLOCK9]]
108 ; RM-NEXT: [[TMP14:%.*]] = shufflevector <1 x i32> [[TMP13]], <1 x i32> poison, <2 x i32> <i32 0, i32 poison>
109 ; RM-NEXT: [[TMP15:%.*]] = shufflevector <2 x i32> [[TMP11]], <2 x i32> [[TMP14]], <2 x i32> <i32 0, i32 2>
110 ; RM-NEXT: [[TMP16:%.*]] = shufflevector <2 x i32> [[TMP7]], <2 x i32> [[TMP15]], <4 x i32> <i32 0, i32 1, i32 2, i32 3>
111 ; RM-NEXT: ret <4 x i32> [[TMP16]]
112 ;
113 entry:
114 %c = call <4 x i32> @llvm.matrix.multiply.v4i32.v2i32.v2i32(<2 x i32> %a, <2 x i32> %b, i32 2, i32 1, i32 2)
115 ret <4 x i32> %c
116 }
118 declare <4 x i32> @llvm.matrix.multiply.v4i32.v2i32.v2i32(<2 x i32>, <2 x i32>, i32, i32, i32)
120 define <9 x i32> @multiply_2x3(<6 x i32> %a, <6 x i32> %b) {
121 ; RM-LABEL: @multiply_2x3(
122 ; RM-NEXT: entry:
123 ; RM-NEXT: [[SPLIT:%.*]] = shufflevector <6 x i32> [[A:%.*]], <6 x i32> poison, <2 x i32> <i32 0, i32 1>
124 ; RM-NEXT: [[SPLIT1:%.*]] = shufflevector <6 x i32> [[A]], <6 x i32> poison, <2 x i32> <i32 2, i32 3>
125 ; RM-NEXT: [[SPLIT2:%.*]] = shufflevector <6 x i32> [[A]], <6 x i32> poison, <2 x i32> <i32 4, i32 5>
126 ; RM-NEXT: [[SPLIT3:%.*]] = shufflevector <6 x i32> [[B:%.*]], <6 x i32> poison, <3 x i32> <i32 0, i32 1, i32 2>
127 ; RM-NEXT: [[SPLIT4:%.*]] = shufflevector <6 x i32> [[B]], <6 x i32> poison, <3 x i32> <i32 3, i32 4, i32 5>
128 ; RM-NEXT: [[BLOCK:%.*]] = shufflevector <3 x i32> [[SPLIT3]], <3 x i32> poison, <1 x i32> zeroinitializer
129 ; RM-NEXT: [[TMP0:%.*]] = extractelement <2 x i32> [[SPLIT]], i64 0
130 ; RM-NEXT: [[SPLAT_SPLATINSERT:%.*]] = insertelement <1 x i32> poison, i32 [[TMP0]], i64 0
131 ; RM-NEXT: [[SPLAT_SPLAT:%.*]] = shufflevector <1 x i32> [[SPLAT_SPLATINSERT]], <1 x i32> poison, <1 x i32> zeroinitializer
132 ; RM-NEXT: [[TMP1:%.*]] = mul <1 x i32> [[SPLAT_SPLAT]], [[BLOCK]]
133 ; RM-NEXT: [[BLOCK5:%.*]] = shufflevector <3 x i32> [[SPLIT4]], <3 x i32> poison, <1 x i32> zeroinitializer
134 ; RM-NEXT: [[TMP2:%.*]] = extractelement <2 x i32> [[SPLIT]], i64 1
135 ; RM-NEXT: [[SPLAT_SPLATINSERT6:%.*]] = insertelement <1 x i32> poison, i32 [[TMP2]], i64 0
136 ; RM-NEXT: [[SPLAT_SPLAT7:%.*]] = shufflevector <1 x i32> [[SPLAT_SPLATINSERT6]], <1 x i32> poison, <1 x i32> zeroinitializer
137 ; RM-NEXT: [[TMP3:%.*]] = mul <1 x i32> [[SPLAT_SPLAT7]], [[BLOCK5]]
138 ; RM-NEXT: [[TMP4:%.*]] = add <1 x i32> [[TMP1]], [[TMP3]]
139 ; RM-NEXT: [[TMP5:%.*]] = shufflevector <1 x i32> [[TMP4]], <1 x i32> poison, <3 x i32> <i32 0, i32 poison, i32 poison>
140 ; RM-NEXT: [[TMP6:%.*]] = shufflevector <3 x i32> poison, <3 x i32> [[TMP5]], <3 x i32> <i32 3, i32 1, i32 2>
141 ; RM-NEXT: [[BLOCK8:%.*]] = shufflevector <3 x i32> [[SPLIT3]], <3 x i32> poison, <1 x i32> <i32 1>
142 ; RM-NEXT: [[TMP7:%.*]] = extractelement <2 x i32> [[SPLIT]], i64 0
143 ; RM-NEXT: [[SPLAT_SPLATINSERT9:%.*]] = insertelement <1 x i32> poison, i32 [[TMP7]], i64 0
144 ; RM-NEXT: [[SPLAT_SPLAT10:%.*]] = shufflevector <1 x i32> [[SPLAT_SPLATINSERT9]], <1 x i32> poison, <1 x i32> zeroinitializer
145 ; RM-NEXT: [[TMP8:%.*]] = mul <1 x i32> [[SPLAT_SPLAT10]], [[BLOCK8]]
146 ; RM-NEXT: [[BLOCK11:%.*]] = shufflevector <3 x i32> [[SPLIT4]], <3 x i32> poison, <1 x i32> <i32 1>
147 ; RM-NEXT: [[TMP9:%.*]] = extractelement <2 x i32> [[SPLIT]], i64 1
148 ; RM-NEXT: [[SPLAT_SPLATINSERT12:%.*]] = insertelement <1 x i32> poison, i32 [[TMP9]], i64 0
149 ; RM-NEXT: [[SPLAT_SPLAT13:%.*]] = shufflevector <1 x i32> [[SPLAT_SPLATINSERT12]], <1 x i32> poison, <1 x i32> zeroinitializer
150 ; RM-NEXT: [[TMP10:%.*]] = mul <1 x i32> [[SPLAT_SPLAT13]], [[BLOCK11]]
151 ; RM-NEXT: [[TMP11:%.*]] = add <1 x i32> [[TMP8]], [[TMP10]]
152 ; RM-NEXT: [[TMP12:%.*]] = shufflevector <1 x i32> [[TMP11]], <1 x i32> poison, <3 x i32> <i32 0, i32 poison, i32 poison>
153 ; RM-NEXT: [[TMP13:%.*]] = shufflevector <3 x i32> [[TMP6]], <3 x i32> [[TMP12]], <3 x i32> <i32 0, i32 3, i32 2>
154 ; RM-NEXT: [[BLOCK14:%.*]] = shufflevector <3 x i32> [[SPLIT3]], <3 x i32> poison, <1 x i32> <i32 2>
155 ; RM-NEXT: [[TMP14:%.*]] = extractelement <2 x i32> [[SPLIT]], i64 0
156 ; RM-NEXT: [[SPLAT_SPLATINSERT15:%.*]] = insertelement <1 x i32> poison, i32 [[TMP14]], i64 0
157 ; RM-NEXT: [[SPLAT_SPLAT16:%.*]] = shufflevector <1 x i32> [[SPLAT_SPLATINSERT15]], <1 x i32> poison, <1 x i32> zeroinitializer
158 ; RM-NEXT: [[TMP15:%.*]] = mul <1 x i32> [[SPLAT_SPLAT16]], [[BLOCK14]]
159 ; RM-NEXT: [[BLOCK17:%.*]] = shufflevector <3 x i32> [[SPLIT4]], <3 x i32> poison, <1 x i32> <i32 2>
160 ; RM-NEXT: [[TMP16:%.*]] = extractelement <2 x i32> [[SPLIT]], i64 1
161 ; RM-NEXT: [[SPLAT_SPLATINSERT18:%.*]] = insertelement <1 x i32> poison, i32 [[TMP16]], i64 0
162 ; RM-NEXT: [[SPLAT_SPLAT19:%.*]] = shufflevector <1 x i32> [[SPLAT_SPLATINSERT18]], <1 x i32> poison, <1 x i32> zeroinitializer
163 ; RM-NEXT: [[TMP17:%.*]] = mul <1 x i32> [[SPLAT_SPLAT19]], [[BLOCK17]]
164 ; RM-NEXT: [[TMP18:%.*]] = add <1 x i32> [[TMP15]], [[TMP17]]
165 ; RM-NEXT: [[TMP19:%.*]] = shufflevector <1 x i32> [[TMP18]], <1 x i32> poison, <3 x i32> <i32 0, i32 poison, i32 poison>
166 ; RM-NEXT: [[TMP20:%.*]] = shufflevector <3 x i32> [[TMP13]], <3 x i32> [[TMP19]], <3 x i32> <i32 0, i32 1, i32 3>
167 ; RM-NEXT: [[BLOCK20:%.*]] = shufflevector <3 x i32> [[SPLIT3]], <3 x i32> poison, <1 x i32> zeroinitializer
168 ; RM-NEXT: [[TMP21:%.*]] = extractelement <2 x i32> [[SPLIT1]], i64 0
169 ; RM-NEXT: [[SPLAT_SPLATINSERT21:%.*]] = insertelement <1 x i32> poison, i32 [[TMP21]], i64 0
170 ; RM-NEXT: [[SPLAT_SPLAT22:%.*]] = shufflevector <1 x i32> [[SPLAT_SPLATINSERT21]], <1 x i32> poison, <1 x i32> zeroinitializer
171 ; RM-NEXT: [[TMP22:%.*]] = mul <1 x i32> [[SPLAT_SPLAT22]], [[BLOCK20]]
172 ; RM-NEXT: [[BLOCK23:%.*]] = shufflevector <3 x i32> [[SPLIT4]], <3 x i32> poison, <1 x i32> zeroinitializer
173 ; RM-NEXT: [[TMP23:%.*]] = extractelement <2 x i32> [[SPLIT1]], i64 1
174 ; RM-NEXT: [[SPLAT_SPLATINSERT24:%.*]] = insertelement <1 x i32> poison, i32 [[TMP23]], i64 0
175 ; RM-NEXT: [[SPLAT_SPLAT25:%.*]] = shufflevector <1 x i32> [[SPLAT_SPLATINSERT24]], <1 x i32> poison, <1 x i32> zeroinitializer
176 ; RM-NEXT: [[TMP24:%.*]] = mul <1 x i32> [[SPLAT_SPLAT25]], [[BLOCK23]]
177 ; RM-NEXT: [[TMP25:%.*]] = add <1 x i32> [[TMP22]], [[TMP24]]
178 ; RM-NEXT: [[TMP26:%.*]] = shufflevector <1 x i32> [[TMP25]], <1 x i32> poison, <3 x i32> <i32 0, i32 poison, i32 poison>
179 ; RM-NEXT: [[TMP27:%.*]] = shufflevector <3 x i32> poison, <3 x i32> [[TMP26]], <3 x i32> <i32 3, i32 1, i32 2>
180 ; RM-NEXT: [[BLOCK26:%.*]] = shufflevector <3 x i32> [[SPLIT3]], <3 x i32> poison, <1 x i32> <i32 1>
181 ; RM-NEXT: [[TMP28:%.*]] = extractelement <2 x i32> [[SPLIT1]], i64 0
182 ; RM-NEXT: [[SPLAT_SPLATINSERT27:%.*]] = insertelement <1 x i32> poison, i32 [[TMP28]], i64 0
183 ; RM-NEXT: [[SPLAT_SPLAT28:%.*]] = shufflevector <1 x i32> [[SPLAT_SPLATINSERT27]], <1 x i32> poison, <1 x i32> zeroinitializer
184 ; RM-NEXT: [[TMP29:%.*]] = mul <1 x i32> [[SPLAT_SPLAT28]], [[BLOCK26]]
185 ; RM-NEXT: [[BLOCK29:%.*]] = shufflevector <3 x i32> [[SPLIT4]], <3 x i32> poison, <1 x i32> <i32 1>
186 ; RM-NEXT: [[TMP30:%.*]] = extractelement <2 x i32> [[SPLIT1]], i64 1
187 ; RM-NEXT: [[SPLAT_SPLATINSERT30:%.*]] = insertelement <1 x i32> poison, i32 [[TMP30]], i64 0
188 ; RM-NEXT: [[SPLAT_SPLAT31:%.*]] = shufflevector <1 x i32> [[SPLAT_SPLATINSERT30]], <1 x i32> poison, <1 x i32> zeroinitializer
189 ; RM-NEXT: [[TMP31:%.*]] = mul <1 x i32> [[SPLAT_SPLAT31]], [[BLOCK29]]
190 ; RM-NEXT: [[TMP32:%.*]] = add <1 x i32> [[TMP29]], [[TMP31]]
191 ; RM-NEXT: [[TMP33:%.*]] = shufflevector <1 x i32> [[TMP32]], <1 x i32> poison, <3 x i32> <i32 0, i32 poison, i32 poison>
192 ; RM-NEXT: [[TMP34:%.*]] = shufflevector <3 x i32> [[TMP27]], <3 x i32> [[TMP33]], <3 x i32> <i32 0, i32 3, i32 2>
193 ; RM-NEXT: [[BLOCK32:%.*]] = shufflevector <3 x i32> [[SPLIT3]], <3 x i32> poison, <1 x i32> <i32 2>
194 ; RM-NEXT: [[TMP35:%.*]] = extractelement <2 x i32> [[SPLIT1]], i64 0
195 ; RM-NEXT: [[SPLAT_SPLATINSERT33:%.*]] = insertelement <1 x i32> poison, i32 [[TMP35]], i64 0
196 ; RM-NEXT: [[SPLAT_SPLAT34:%.*]] = shufflevector <1 x i32> [[SPLAT_SPLATINSERT33]], <1 x i32> poison, <1 x i32> zeroinitializer
197 ; RM-NEXT: [[TMP36:%.*]] = mul <1 x i32> [[SPLAT_SPLAT34]], [[BLOCK32]]
198 ; RM-NEXT: [[BLOCK35:%.*]] = shufflevector <3 x i32> [[SPLIT4]], <3 x i32> poison, <1 x i32> <i32 2>
199 ; RM-NEXT: [[TMP37:%.*]] = extractelement <2 x i32> [[SPLIT1]], i64 1
200 ; RM-NEXT: [[SPLAT_SPLATINSERT36:%.*]] = insertelement <1 x i32> poison, i32 [[TMP37]], i64 0
201 ; RM-NEXT: [[SPLAT_SPLAT37:%.*]] = shufflevector <1 x i32> [[SPLAT_SPLATINSERT36]], <1 x i32> poison, <1 x i32> zeroinitializer
202 ; RM-NEXT: [[TMP38:%.*]] = mul <1 x i32> [[SPLAT_SPLAT37]], [[BLOCK35]]
203 ; RM-NEXT: [[TMP39:%.*]] = add <1 x i32> [[TMP36]], [[TMP38]]
204 ; RM-NEXT: [[TMP40:%.*]] = shufflevector <1 x i32> [[TMP39]], <1 x i32> poison, <3 x i32> <i32 0, i32 poison, i32 poison>
205 ; RM-NEXT: [[TMP41:%.*]] = shufflevector <3 x i32> [[TMP34]], <3 x i32> [[TMP40]], <3 x i32> <i32 0, i32 1, i32 3>
206 ; RM-NEXT: [[BLOCK38:%.*]] = shufflevector <3 x i32> [[SPLIT3]], <3 x i32> poison, <1 x i32> zeroinitializer
207 ; RM-NEXT: [[TMP42:%.*]] = extractelement <2 x i32> [[SPLIT2]], i64 0
208 ; RM-NEXT: [[SPLAT_SPLATINSERT39:%.*]] = insertelement <1 x i32> poison, i32 [[TMP42]], i64 0
209 ; RM-NEXT: [[SPLAT_SPLAT40:%.*]] = shufflevector <1 x i32> [[SPLAT_SPLATINSERT39]], <1 x i32> poison, <1 x i32> zeroinitializer
210 ; RM-NEXT: [[TMP43:%.*]] = mul <1 x i32> [[SPLAT_SPLAT40]], [[BLOCK38]]
211 ; RM-NEXT: [[BLOCK41:%.*]] = shufflevector <3 x i32> [[SPLIT4]], <3 x i32> poison, <1 x i32> zeroinitializer
212 ; RM-NEXT: [[TMP44:%.*]] = extractelement <2 x i32> [[SPLIT2]], i64 1
213 ; RM-NEXT: [[SPLAT_SPLATINSERT42:%.*]] = insertelement <1 x i32> poison, i32 [[TMP44]], i64 0
214 ; RM-NEXT: [[SPLAT_SPLAT43:%.*]] = shufflevector <1 x i32> [[SPLAT_SPLATINSERT42]], <1 x i32> poison, <1 x i32> zeroinitializer
215 ; RM-NEXT: [[TMP45:%.*]] = mul <1 x i32> [[SPLAT_SPLAT43]], [[BLOCK41]]
216 ; RM-NEXT: [[TMP46:%.*]] = add <1 x i32> [[TMP43]], [[TMP45]]
217 ; RM-NEXT: [[TMP47:%.*]] = shufflevector <1 x i32> [[TMP46]], <1 x i32> poison, <3 x i32> <i32 0, i32 poison, i32 poison>
218 ; RM-NEXT: [[TMP48:%.*]] = shufflevector <3 x i32> poison, <3 x i32> [[TMP47]], <3 x i32> <i32 3, i32 1, i32 2>
219 ; RM-NEXT: [[BLOCK44:%.*]] = shufflevector <3 x i32> [[SPLIT3]], <3 x i32> poison, <1 x i32> <i32 1>
220 ; RM-NEXT: [[TMP49:%.*]] = extractelement <2 x i32> [[SPLIT2]], i64 0
221 ; RM-NEXT: [[SPLAT_SPLATINSERT45:%.*]] = insertelement <1 x i32> poison, i32 [[TMP49]], i64 0
222 ; RM-NEXT: [[SPLAT_SPLAT46:%.*]] = shufflevector <1 x i32> [[SPLAT_SPLATINSERT45]], <1 x i32> poison, <1 x i32> zeroinitializer
223 ; RM-NEXT: [[TMP50:%.*]] = mul <1 x i32> [[SPLAT_SPLAT46]], [[BLOCK44]]
224 ; RM-NEXT: [[BLOCK47:%.*]] = shufflevector <3 x i32> [[SPLIT4]], <3 x i32> poison, <1 x i32> <i32 1>
225 ; RM-NEXT: [[TMP51:%.*]] = extractelement <2 x i32> [[SPLIT2]], i64 1
226 ; RM-NEXT: [[SPLAT_SPLATINSERT48:%.*]] = insertelement <1 x i32> poison, i32 [[TMP51]], i64 0
227 ; RM-NEXT: [[SPLAT_SPLAT49:%.*]] = shufflevector <1 x i32> [[SPLAT_SPLATINSERT48]], <1 x i32> poison, <1 x i32> zeroinitializer
228 ; RM-NEXT: [[TMP52:%.*]] = mul <1 x i32> [[SPLAT_SPLAT49]], [[BLOCK47]]
229 ; RM-NEXT: [[TMP53:%.*]] = add <1 x i32> [[TMP50]], [[TMP52]]
230 ; RM-NEXT: [[TMP54:%.*]] = shufflevector <1 x i32> [[TMP53]], <1 x i32> poison, <3 x i32> <i32 0, i32 poison, i32 poison>
231 ; RM-NEXT: [[TMP55:%.*]] = shufflevector <3 x i32> [[TMP48]], <3 x i32> [[TMP54]], <3 x i32> <i32 0, i32 3, i32 2>
232 ; RM-NEXT: [[BLOCK50:%.*]] = shufflevector <3 x i32> [[SPLIT3]], <3 x i32> poison, <1 x i32> <i32 2>
233 ; RM-NEXT: [[TMP56:%.*]] = extractelement <2 x i32> [[SPLIT2]], i64 0
234 ; RM-NEXT: [[SPLAT_SPLATINSERT51:%.*]] = insertelement <1 x i32> poison, i32 [[TMP56]], i64 0
235 ; RM-NEXT: [[SPLAT_SPLAT52:%.*]] = shufflevector <1 x i32> [[SPLAT_SPLATINSERT51]], <1 x i32> poison, <1 x i32> zeroinitializer
236 ; RM-NEXT: [[TMP57:%.*]] = mul <1 x i32> [[SPLAT_SPLAT52]], [[BLOCK50]]
237 ; RM-NEXT: [[BLOCK53:%.*]] = shufflevector <3 x i32> [[SPLIT4]], <3 x i32> poison, <1 x i32> <i32 2>
238 ; RM-NEXT: [[TMP58:%.*]] = extractelement <2 x i32> [[SPLIT2]], i64 1
239 ; RM-NEXT: [[SPLAT_SPLATINSERT54:%.*]] = insertelement <1 x i32> poison, i32 [[TMP58]], i64 0
240 ; RM-NEXT: [[SPLAT_SPLAT55:%.*]] = shufflevector <1 x i32> [[SPLAT_SPLATINSERT54]], <1 x i32> poison, <1 x i32> zeroinitializer
241 ; RM-NEXT: [[TMP59:%.*]] = mul <1 x i32> [[SPLAT_SPLAT55]], [[BLOCK53]]
242 ; RM-NEXT: [[TMP60:%.*]] = add <1 x i32> [[TMP57]], [[TMP59]]
243 ; RM-NEXT: [[TMP61:%.*]] = shufflevector <1 x i32> [[TMP60]], <1 x i32> poison, <3 x i32> <i32 0, i32 poison, i32 poison>
244 ; RM-NEXT: [[TMP62:%.*]] = shufflevector <3 x i32> [[TMP55]], <3 x i32> [[TMP61]], <3 x i32> <i32 0, i32 1, i32 3>
245 ; RM-NEXT: [[TMP63:%.*]] = shufflevector <3 x i32> [[TMP20]], <3 x i32> [[TMP41]], <6 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5>
246 ; RM-NEXT: [[TMP64:%.*]] = shufflevector <3 x i32> [[TMP62]], <3 x i32> poison, <6 x i32> <i32 0, i32 1, i32 2, i32 poison, i32 poison, i32 poison>
247 ; RM-NEXT: [[TMP65:%.*]] = shufflevector <6 x i32> [[TMP63]], <6 x i32> [[TMP64]], <9 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8>
248 ; RM-NEXT: ret <9 x i32> [[TMP65]]
249 ;
250 entry:
251 %c = call <9 x i32> @llvm.matrix.multiply.v9i32.v6i32.v6i32(<6 x i32> %a, <6 x i32> %b, i32 3, i32 2, i32 3)
252 ret <9 x i32> %c
253 }
255 declare <9 x i32> @llvm.matrix.multiply.v9i32.v6i32.v6i32(<6 x i32>, <6 x i32>, i32, i32, i32)