xref: /llvm-project/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fma.mir (revision 373c343a77a7afaa07179db1754a52b620dfaf2e) 
1# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
2# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=tahiti -run-pass=legalizer %s -o - | FileCheck -check-prefix=SI  %s
3# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=fiji -run-pass=legalizer %s -o - | FileCheck -check-prefix=VI %s
4# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 -run-pass=legalizer %s -o - | FileCheck -check-prefix=GFX9  %s
5# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1010 -run-pass=legalizer %s -o - | FileCheck -check-prefix=GFX9  %s
6# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -run-pass=legalizer %s -o - | FileCheck -check-prefix=GFX9  %s
7
8---
9name: test_fma_s32
10body: |
11  bb.0:
12    liveins: $vgpr0, $vgpr1, $vgpr2
13
14    ; SI-LABEL: name: test_fma_s32
15    ; SI: liveins: $vgpr0, $vgpr1, $vgpr2
16    ; SI-NEXT: {{  $}}
17    ; SI-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
18    ; SI-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
19    ; SI-NEXT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
20    ; SI-NEXT: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[COPY]], [[COPY1]], [[COPY2]]
21    ; SI-NEXT: $vgpr0 = COPY [[FMA]](s32)
22    ;
23    ; VI-LABEL: name: test_fma_s32
24    ; VI: liveins: $vgpr0, $vgpr1, $vgpr2
25    ; VI-NEXT: {{  $}}
26    ; VI-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
27    ; VI-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
28    ; VI-NEXT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
29    ; VI-NEXT: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[COPY]], [[COPY1]], [[COPY2]]
30    ; VI-NEXT: $vgpr0 = COPY [[FMA]](s32)
31    ;
32    ; GFX9-LABEL: name: test_fma_s32
33    ; GFX9: liveins: $vgpr0, $vgpr1, $vgpr2
34    ; GFX9-NEXT: {{  $}}
35    ; GFX9-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
36    ; GFX9-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
37    ; GFX9-NEXT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
38    ; GFX9-NEXT: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[COPY]], [[COPY1]], [[COPY2]]
39    ; GFX9-NEXT: $vgpr0 = COPY [[FMA]](s32)
40    %0:_(s32) = COPY $vgpr0
41    %1:_(s32) = COPY $vgpr1
42    %2:_(s32) = COPY $vgpr2
43    %3:_(s32) = G_FMA %0, %1, %2
44    $vgpr0 = COPY %3
45...
46---
47name: test_fma_s64
48body: |
49  bb.0:
50    liveins: $vgpr0_vgpr1, $vgpr2_vgpr3,  $vgpr4_vgpr5
51
52    ; SI-LABEL: name: test_fma_s64
53    ; SI: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $vgpr4_vgpr5
54    ; SI-NEXT: {{  $}}
55    ; SI-NEXT: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1
56    ; SI-NEXT: [[COPY1:%[0-9]+]]:_(s64) = COPY $vgpr2_vgpr3
57    ; SI-NEXT: [[COPY2:%[0-9]+]]:_(s64) = COPY $vgpr4_vgpr5
58    ; SI-NEXT: [[FMA:%[0-9]+]]:_(s64) = G_FMA [[COPY]], [[COPY1]], [[COPY2]]
59    ; SI-NEXT: $vgpr0_vgpr1 = COPY [[FMA]](s64)
60    ;
61    ; VI-LABEL: name: test_fma_s64
62    ; VI: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $vgpr4_vgpr5
63    ; VI-NEXT: {{  $}}
64    ; VI-NEXT: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1
65    ; VI-NEXT: [[COPY1:%[0-9]+]]:_(s64) = COPY $vgpr2_vgpr3
66    ; VI-NEXT: [[COPY2:%[0-9]+]]:_(s64) = COPY $vgpr4_vgpr5
67    ; VI-NEXT: [[FMA:%[0-9]+]]:_(s64) = G_FMA [[COPY]], [[COPY1]], [[COPY2]]
68    ; VI-NEXT: $vgpr0_vgpr1 = COPY [[FMA]](s64)
69    ;
70    ; GFX9-LABEL: name: test_fma_s64
71    ; GFX9: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $vgpr4_vgpr5
72    ; GFX9-NEXT: {{  $}}
73    ; GFX9-NEXT: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1
74    ; GFX9-NEXT: [[COPY1:%[0-9]+]]:_(s64) = COPY $vgpr2_vgpr3
75    ; GFX9-NEXT: [[COPY2:%[0-9]+]]:_(s64) = COPY $vgpr4_vgpr5
76    ; GFX9-NEXT: [[FMA:%[0-9]+]]:_(s64) = G_FMA [[COPY]], [[COPY1]], [[COPY2]]
77    ; GFX9-NEXT: $vgpr0_vgpr1 = COPY [[FMA]](s64)
78    %0:_(s64) = COPY $vgpr0_vgpr1
79    %1:_(s64) = COPY $vgpr2_vgpr3
80    %2:_(s64) = COPY $vgpr4_vgpr5
81    %3:_(s64) = G_FMA %0, %1, %2
82    $vgpr0_vgpr1 = COPY %3
83...
84
85---
86name: test_fma_s16
87body: |
88  bb.0:
89    liveins: $vgpr0, $vgpr1, $vgpr2
90
91    ; SI-LABEL: name: test_fma_s16
92    ; SI: liveins: $vgpr0, $vgpr1, $vgpr2
93    ; SI-NEXT: {{  $}}
94    ; SI-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
95    ; SI-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
96    ; SI-NEXT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr1
97    ; SI-NEXT: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
98    ; SI-NEXT: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32)
99    ; SI-NEXT: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[COPY2]](s32)
100    ; SI-NEXT: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC]](s16)
101    ; SI-NEXT: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC1]](s16)
102    ; SI-NEXT: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC2]](s16)
103    ; SI-NEXT: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[FPEXT]], [[FPEXT1]], [[FPEXT2]]
104    ; SI-NEXT: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMA]](s32)
105    ; SI-NEXT: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FPTRUNC]](s16)
106    ; SI-NEXT: $vgpr0 = COPY [[ANYEXT]](s32)
107    ;
108    ; VI-LABEL: name: test_fma_s16
109    ; VI: liveins: $vgpr0, $vgpr1, $vgpr2
110    ; VI-NEXT: {{  $}}
111    ; VI-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
112    ; VI-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
113    ; VI-NEXT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr1
114    ; VI-NEXT: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
115    ; VI-NEXT: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32)
116    ; VI-NEXT: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[COPY2]](s32)
117    ; VI-NEXT: [[FMA:%[0-9]+]]:_(s16) = G_FMA [[TRUNC]], [[TRUNC1]], [[TRUNC2]]
118    ; VI-NEXT: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FMA]](s16)
119    ; VI-NEXT: $vgpr0 = COPY [[ANYEXT]](s32)
120    ;
121    ; GFX9-LABEL: name: test_fma_s16
122    ; GFX9: liveins: $vgpr0, $vgpr1, $vgpr2
123    ; GFX9-NEXT: {{  $}}
124    ; GFX9-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
125    ; GFX9-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
126    ; GFX9-NEXT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr1
127    ; GFX9-NEXT: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
128    ; GFX9-NEXT: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32)
129    ; GFX9-NEXT: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[COPY2]](s32)
130    ; GFX9-NEXT: [[FMA:%[0-9]+]]:_(s16) = G_FMA [[TRUNC]], [[TRUNC1]], [[TRUNC2]]
131    ; GFX9-NEXT: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FMA]](s16)
132    ; GFX9-NEXT: $vgpr0 = COPY [[ANYEXT]](s32)
133    %0:_(s32) = COPY $vgpr0
134    %1:_(s32) = COPY $vgpr1
135    %2:_(s32) = COPY $vgpr1
136    %3:_(s16) = G_TRUNC %0
137    %4:_(s16) = G_TRUNC %1
138    %5:_(s16) = G_TRUNC %2
139
140    %6:_(s16) = G_FMA %3, %4, %5
141    %7:_(s32) = G_ANYEXT %6
142    $vgpr0 = COPY %7
143...
144
145---
146name: test_fma_v2s32
147body: |
148  bb.0:
149    liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $vgpr4_vgpr5
150
151    ; SI-LABEL: name: test_fma_v2s32
152    ; SI: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $vgpr4_vgpr5
153    ; SI-NEXT: {{  $}}
154    ; SI-NEXT: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
155    ; SI-NEXT: [[COPY1:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr2_vgpr3
156    ; SI-NEXT: [[COPY2:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr4_vgpr5
157    ; SI-NEXT: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<2 x s32>)
158    ; SI-NEXT: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<2 x s32>)
159    ; SI-NEXT: [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY2]](<2 x s32>)
160    ; SI-NEXT: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[UV]], [[UV2]], [[UV4]]
161    ; SI-NEXT: [[FMA1:%[0-9]+]]:_(s32) = G_FMA [[UV1]], [[UV3]], [[UV5]]
162    ; SI-NEXT: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[FMA]](s32), [[FMA1]](s32)
163    ; SI-NEXT: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
164    ;
165    ; VI-LABEL: name: test_fma_v2s32
166    ; VI: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $vgpr4_vgpr5
167    ; VI-NEXT: {{  $}}
168    ; VI-NEXT: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
169    ; VI-NEXT: [[COPY1:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr2_vgpr3
170    ; VI-NEXT: [[COPY2:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr4_vgpr5
171    ; VI-NEXT: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<2 x s32>)
172    ; VI-NEXT: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<2 x s32>)
173    ; VI-NEXT: [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY2]](<2 x s32>)
174    ; VI-NEXT: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[UV]], [[UV2]], [[UV4]]
175    ; VI-NEXT: [[FMA1:%[0-9]+]]:_(s32) = G_FMA [[UV1]], [[UV3]], [[UV5]]
176    ; VI-NEXT: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[FMA]](s32), [[FMA1]](s32)
177    ; VI-NEXT: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
178    ;
179    ; GFX9-LABEL: name: test_fma_v2s32
180    ; GFX9: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $vgpr4_vgpr5
181    ; GFX9-NEXT: {{  $}}
182    ; GFX9-NEXT: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
183    ; GFX9-NEXT: [[COPY1:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr2_vgpr3
184    ; GFX9-NEXT: [[COPY2:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr4_vgpr5
185    ; GFX9-NEXT: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<2 x s32>)
186    ; GFX9-NEXT: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<2 x s32>)
187    ; GFX9-NEXT: [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY2]](<2 x s32>)
188    ; GFX9-NEXT: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[UV]], [[UV2]], [[UV4]]
189    ; GFX9-NEXT: [[FMA1:%[0-9]+]]:_(s32) = G_FMA [[UV1]], [[UV3]], [[UV5]]
190    ; GFX9-NEXT: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[FMA]](s32), [[FMA1]](s32)
191    ; GFX9-NEXT: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
192    %0:_(<2 x s32>) = COPY $vgpr0_vgpr1
193    %1:_(<2 x s32>) = COPY $vgpr2_vgpr3
194    %2:_(<2 x s32>) = COPY $vgpr4_vgpr5
195    %3:_(<2 x s32>) = G_FMA %0, %1, %2
196    $vgpr0_vgpr1 = COPY %3
197...
198
199---
200name: test_fma_v3s32
201body: |
202  bb.0:
203    liveins: $vgpr0_vgpr1_vgpr2, $vgpr3_vgpr4_vgpr5, $vgpr6_vgpr7_vgpr8
204
205    ; SI-LABEL: name: test_fma_v3s32
206    ; SI: liveins: $vgpr0_vgpr1_vgpr2, $vgpr3_vgpr4_vgpr5, $vgpr6_vgpr7_vgpr8
207    ; SI-NEXT: {{  $}}
208    ; SI-NEXT: [[COPY:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr0_vgpr1_vgpr2
209    ; SI-NEXT: [[COPY1:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr3_vgpr4_vgpr5
210    ; SI-NEXT: [[COPY2:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr6_vgpr7_vgpr8
211    ; SI-NEXT: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<3 x s32>)
212    ; SI-NEXT: [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<3 x s32>)
213    ; SI-NEXT: [[UV6:%[0-9]+]]:_(s32), [[UV7:%[0-9]+]]:_(s32), [[UV8:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY2]](<3 x s32>)
214    ; SI-NEXT: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[UV]], [[UV3]], [[UV6]]
215    ; SI-NEXT: [[FMA1:%[0-9]+]]:_(s32) = G_FMA [[UV1]], [[UV4]], [[UV7]]
216    ; SI-NEXT: [[FMA2:%[0-9]+]]:_(s32) = G_FMA [[UV2]], [[UV5]], [[UV8]]
217    ; SI-NEXT: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[FMA]](s32), [[FMA1]](s32), [[FMA2]](s32)
218    ; SI-NEXT: $vgpr0_vgpr1_vgpr2 = COPY [[BUILD_VECTOR]](<3 x s32>)
219    ;
220    ; VI-LABEL: name: test_fma_v3s32
221    ; VI: liveins: $vgpr0_vgpr1_vgpr2, $vgpr3_vgpr4_vgpr5, $vgpr6_vgpr7_vgpr8
222    ; VI-NEXT: {{  $}}
223    ; VI-NEXT: [[COPY:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr0_vgpr1_vgpr2
224    ; VI-NEXT: [[COPY1:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr3_vgpr4_vgpr5
225    ; VI-NEXT: [[COPY2:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr6_vgpr7_vgpr8
226    ; VI-NEXT: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<3 x s32>)
227    ; VI-NEXT: [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<3 x s32>)
228    ; VI-NEXT: [[UV6:%[0-9]+]]:_(s32), [[UV7:%[0-9]+]]:_(s32), [[UV8:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY2]](<3 x s32>)
229    ; VI-NEXT: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[UV]], [[UV3]], [[UV6]]
230    ; VI-NEXT: [[FMA1:%[0-9]+]]:_(s32) = G_FMA [[UV1]], [[UV4]], [[UV7]]
231    ; VI-NEXT: [[FMA2:%[0-9]+]]:_(s32) = G_FMA [[UV2]], [[UV5]], [[UV8]]
232    ; VI-NEXT: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[FMA]](s32), [[FMA1]](s32), [[FMA2]](s32)
233    ; VI-NEXT: $vgpr0_vgpr1_vgpr2 = COPY [[BUILD_VECTOR]](<3 x s32>)
234    ;
235    ; GFX9-LABEL: name: test_fma_v3s32
236    ; GFX9: liveins: $vgpr0_vgpr1_vgpr2, $vgpr3_vgpr4_vgpr5, $vgpr6_vgpr7_vgpr8
237    ; GFX9-NEXT: {{  $}}
238    ; GFX9-NEXT: [[COPY:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr0_vgpr1_vgpr2
239    ; GFX9-NEXT: [[COPY1:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr3_vgpr4_vgpr5
240    ; GFX9-NEXT: [[COPY2:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr6_vgpr7_vgpr8
241    ; GFX9-NEXT: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<3 x s32>)
242    ; GFX9-NEXT: [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<3 x s32>)
243    ; GFX9-NEXT: [[UV6:%[0-9]+]]:_(s32), [[UV7:%[0-9]+]]:_(s32), [[UV8:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY2]](<3 x s32>)
244    ; GFX9-NEXT: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[UV]], [[UV3]], [[UV6]]
245    ; GFX9-NEXT: [[FMA1:%[0-9]+]]:_(s32) = G_FMA [[UV1]], [[UV4]], [[UV7]]
246    ; GFX9-NEXT: [[FMA2:%[0-9]+]]:_(s32) = G_FMA [[UV2]], [[UV5]], [[UV8]]
247    ; GFX9-NEXT: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[FMA]](s32), [[FMA1]](s32), [[FMA2]](s32)
248    ; GFX9-NEXT: $vgpr0_vgpr1_vgpr2 = COPY [[BUILD_VECTOR]](<3 x s32>)
249    %0:_(<3 x s32>) = COPY $vgpr0_vgpr1_vgpr2
250    %1:_(<3 x s32>) = COPY $vgpr3_vgpr4_vgpr5
251    %2:_(<3 x s32>) = COPY $vgpr6_vgpr7_vgpr8
252    %3:_(<3 x s32>) = G_FMA %0, %1, %2
253    $vgpr0_vgpr1_vgpr2 = COPY %3
254...
255
256---
257name: test_fma_v4s32
258body: |
259  bb.0:
260    liveins: $vgpr0_vgpr1_vgpr2_vgpr3, $vgpr4_vgpr5_vgpr6_vgpr7, $vgpr8_vgpr9_vgpr10_vgpr11
261
262    ; SI-LABEL: name: test_fma_v4s32
263    ; SI: liveins: $vgpr0_vgpr1_vgpr2_vgpr3, $vgpr4_vgpr5_vgpr6_vgpr7, $vgpr8_vgpr9_vgpr10_vgpr11
264    ; SI-NEXT: {{  $}}
265    ; SI-NEXT: [[COPY:%[0-9]+]]:_(<4 x s32>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
266    ; SI-NEXT: [[COPY1:%[0-9]+]]:_(<4 x s32>) = COPY $vgpr4_vgpr5_vgpr6_vgpr7
267    ; SI-NEXT: [[COPY2:%[0-9]+]]:_(<4 x s32>) = COPY $vgpr8_vgpr9_vgpr10_vgpr11
268    ; SI-NEXT: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<4 x s32>)
269    ; SI-NEXT: [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32), [[UV6:%[0-9]+]]:_(s32), [[UV7:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<4 x s32>)
270    ; SI-NEXT: [[UV8:%[0-9]+]]:_(s32), [[UV9:%[0-9]+]]:_(s32), [[UV10:%[0-9]+]]:_(s32), [[UV11:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY2]](<4 x s32>)
271    ; SI-NEXT: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[UV]], [[UV4]], [[UV8]]
272    ; SI-NEXT: [[FMA1:%[0-9]+]]:_(s32) = G_FMA [[UV1]], [[UV5]], [[UV9]]
273    ; SI-NEXT: [[FMA2:%[0-9]+]]:_(s32) = G_FMA [[UV2]], [[UV6]], [[UV10]]
274    ; SI-NEXT: [[FMA3:%[0-9]+]]:_(s32) = G_FMA [[UV3]], [[UV7]], [[UV11]]
275    ; SI-NEXT: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[FMA]](s32), [[FMA1]](s32), [[FMA2]](s32), [[FMA3]](s32)
276    ; SI-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<4 x s32>)
277    ;
278    ; VI-LABEL: name: test_fma_v4s32
279    ; VI: liveins: $vgpr0_vgpr1_vgpr2_vgpr3, $vgpr4_vgpr5_vgpr6_vgpr7, $vgpr8_vgpr9_vgpr10_vgpr11
280    ; VI-NEXT: {{  $}}
281    ; VI-NEXT: [[COPY:%[0-9]+]]:_(<4 x s32>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
282    ; VI-NEXT: [[COPY1:%[0-9]+]]:_(<4 x s32>) = COPY $vgpr4_vgpr5_vgpr6_vgpr7
283    ; VI-NEXT: [[COPY2:%[0-9]+]]:_(<4 x s32>) = COPY $vgpr8_vgpr9_vgpr10_vgpr11
284    ; VI-NEXT: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<4 x s32>)
285    ; VI-NEXT: [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32), [[UV6:%[0-9]+]]:_(s32), [[UV7:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<4 x s32>)
286    ; VI-NEXT: [[UV8:%[0-9]+]]:_(s32), [[UV9:%[0-9]+]]:_(s32), [[UV10:%[0-9]+]]:_(s32), [[UV11:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY2]](<4 x s32>)
287    ; VI-NEXT: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[UV]], [[UV4]], [[UV8]]
288    ; VI-NEXT: [[FMA1:%[0-9]+]]:_(s32) = G_FMA [[UV1]], [[UV5]], [[UV9]]
289    ; VI-NEXT: [[FMA2:%[0-9]+]]:_(s32) = G_FMA [[UV2]], [[UV6]], [[UV10]]
290    ; VI-NEXT: [[FMA3:%[0-9]+]]:_(s32) = G_FMA [[UV3]], [[UV7]], [[UV11]]
291    ; VI-NEXT: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[FMA]](s32), [[FMA1]](s32), [[FMA2]](s32), [[FMA3]](s32)
292    ; VI-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<4 x s32>)
293    ;
294    ; GFX9-LABEL: name: test_fma_v4s32
295    ; GFX9: liveins: $vgpr0_vgpr1_vgpr2_vgpr3, $vgpr4_vgpr5_vgpr6_vgpr7, $vgpr8_vgpr9_vgpr10_vgpr11
296    ; GFX9-NEXT: {{  $}}
297    ; GFX9-NEXT: [[COPY:%[0-9]+]]:_(<4 x s32>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
298    ; GFX9-NEXT: [[COPY1:%[0-9]+]]:_(<4 x s32>) = COPY $vgpr4_vgpr5_vgpr6_vgpr7
299    ; GFX9-NEXT: [[COPY2:%[0-9]+]]:_(<4 x s32>) = COPY $vgpr8_vgpr9_vgpr10_vgpr11
300    ; GFX9-NEXT: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<4 x s32>)
301    ; GFX9-NEXT: [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32), [[UV6:%[0-9]+]]:_(s32), [[UV7:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<4 x s32>)
302    ; GFX9-NEXT: [[UV8:%[0-9]+]]:_(s32), [[UV9:%[0-9]+]]:_(s32), [[UV10:%[0-9]+]]:_(s32), [[UV11:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY2]](<4 x s32>)
303    ; GFX9-NEXT: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[UV]], [[UV4]], [[UV8]]
304    ; GFX9-NEXT: [[FMA1:%[0-9]+]]:_(s32) = G_FMA [[UV1]], [[UV5]], [[UV9]]
305    ; GFX9-NEXT: [[FMA2:%[0-9]+]]:_(s32) = G_FMA [[UV2]], [[UV6]], [[UV10]]
306    ; GFX9-NEXT: [[FMA3:%[0-9]+]]:_(s32) = G_FMA [[UV3]], [[UV7]], [[UV11]]
307    ; GFX9-NEXT: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[FMA]](s32), [[FMA1]](s32), [[FMA2]](s32), [[FMA3]](s32)
308    ; GFX9-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<4 x s32>)
309    %0:_(<4 x s32>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
310    %1:_(<4 x s32>) = COPY $vgpr4_vgpr5_vgpr6_vgpr7
311    %2:_(<4 x s32>) = COPY $vgpr8_vgpr9_vgpr10_vgpr11
312    %3:_(<4 x s32>) = G_FMA %0, %1, %2
313    $vgpr0_vgpr1_vgpr2_vgpr3 = COPY %3
314...
315
316---
317name: test_fma_v2s64
318body: |
319  bb.0:
320    liveins: $vgpr0_vgpr1_vgpr2_vgpr3, $vgpr4_vgpr5_vgpr6_vgpr7, $vgpr8_vgpr9_vgpr10_vgpr11
321
322    ; SI-LABEL: name: test_fma_v2s64
323    ; SI: liveins: $vgpr0_vgpr1_vgpr2_vgpr3, $vgpr4_vgpr5_vgpr6_vgpr7, $vgpr8_vgpr9_vgpr10_vgpr11
324    ; SI-NEXT: {{  $}}
325    ; SI-NEXT: [[COPY:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
326    ; SI-NEXT: [[COPY1:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr4_vgpr5_vgpr6_vgpr7
327    ; SI-NEXT: [[COPY2:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr8_vgpr9_vgpr10_vgpr11
328    ; SI-NEXT: [[UV:%[0-9]+]]:_(s64), [[UV1:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY]](<2 x s64>)
329    ; SI-NEXT: [[UV2:%[0-9]+]]:_(s64), [[UV3:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY1]](<2 x s64>)
330    ; SI-NEXT: [[UV4:%[0-9]+]]:_(s64), [[UV5:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY2]](<2 x s64>)
331    ; SI-NEXT: [[FMA:%[0-9]+]]:_(s64) = G_FMA [[UV]], [[UV2]], [[UV4]]
332    ; SI-NEXT: [[FMA1:%[0-9]+]]:_(s64) = G_FMA [[UV1]], [[UV3]], [[UV5]]
333    ; SI-NEXT: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s64>) = G_BUILD_VECTOR [[FMA]](s64), [[FMA1]](s64)
334    ; SI-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<2 x s64>)
335    ;
336    ; VI-LABEL: name: test_fma_v2s64
337    ; VI: liveins: $vgpr0_vgpr1_vgpr2_vgpr3, $vgpr4_vgpr5_vgpr6_vgpr7, $vgpr8_vgpr9_vgpr10_vgpr11
338    ; VI-NEXT: {{  $}}
339    ; VI-NEXT: [[COPY:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
340    ; VI-NEXT: [[COPY1:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr4_vgpr5_vgpr6_vgpr7
341    ; VI-NEXT: [[COPY2:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr8_vgpr9_vgpr10_vgpr11
342    ; VI-NEXT: [[UV:%[0-9]+]]:_(s64), [[UV1:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY]](<2 x s64>)
343    ; VI-NEXT: [[UV2:%[0-9]+]]:_(s64), [[UV3:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY1]](<2 x s64>)
344    ; VI-NEXT: [[UV4:%[0-9]+]]:_(s64), [[UV5:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY2]](<2 x s64>)
345    ; VI-NEXT: [[FMA:%[0-9]+]]:_(s64) = G_FMA [[UV]], [[UV2]], [[UV4]]
346    ; VI-NEXT: [[FMA1:%[0-9]+]]:_(s64) = G_FMA [[UV1]], [[UV3]], [[UV5]]
347    ; VI-NEXT: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s64>) = G_BUILD_VECTOR [[FMA]](s64), [[FMA1]](s64)
348    ; VI-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<2 x s64>)
349    ;
350    ; GFX9-LABEL: name: test_fma_v2s64
351    ; GFX9: liveins: $vgpr0_vgpr1_vgpr2_vgpr3, $vgpr4_vgpr5_vgpr6_vgpr7, $vgpr8_vgpr9_vgpr10_vgpr11
352    ; GFX9-NEXT: {{  $}}
353    ; GFX9-NEXT: [[COPY:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
354    ; GFX9-NEXT: [[COPY1:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr4_vgpr5_vgpr6_vgpr7
355    ; GFX9-NEXT: [[COPY2:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr8_vgpr9_vgpr10_vgpr11
356    ; GFX9-NEXT: [[UV:%[0-9]+]]:_(s64), [[UV1:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY]](<2 x s64>)
357    ; GFX9-NEXT: [[UV2:%[0-9]+]]:_(s64), [[UV3:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY1]](<2 x s64>)
358    ; GFX9-NEXT: [[UV4:%[0-9]+]]:_(s64), [[UV5:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY2]](<2 x s64>)
359    ; GFX9-NEXT: [[FMA:%[0-9]+]]:_(s64) = G_FMA [[UV]], [[UV2]], [[UV4]]
360    ; GFX9-NEXT: [[FMA1:%[0-9]+]]:_(s64) = G_FMA [[UV1]], [[UV3]], [[UV5]]
361    ; GFX9-NEXT: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s64>) = G_BUILD_VECTOR [[FMA]](s64), [[FMA1]](s64)
362    ; GFX9-NEXT: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<2 x s64>)
363    %0:_(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
364    %1:_(<2 x s64>) = COPY $vgpr4_vgpr5_vgpr6_vgpr7
365    %2:_(<2 x s64>) = COPY $vgpr8_vgpr9_vgpr10_vgpr11
366    %3:_(<2 x s64>) = G_FMA %0, %1, %2
367    $vgpr0_vgpr1_vgpr2_vgpr3 = COPY %3
368...
369
370---
371name: test_fma_v2s16
372body: |
373  bb.0:
374    liveins: $vgpr0, $vgpr1, $vgpr2
375
376    ; SI-LABEL: name: test_fma_v2s16
377    ; SI: liveins: $vgpr0, $vgpr1, $vgpr2
378    ; SI-NEXT: {{  $}}
379    ; SI-NEXT: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
380    ; SI-NEXT: [[COPY1:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr1
381    ; SI-NEXT: [[COPY2:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr2
382    ; SI-NEXT: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY]](<2 x s16>)
383    ; SI-NEXT: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
384    ; SI-NEXT: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
385    ; SI-NEXT: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
386    ; SI-NEXT: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
387    ; SI-NEXT: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[COPY1]](<2 x s16>)
388    ; SI-NEXT: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST1]](s32)
389    ; SI-NEXT: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
390    ; SI-NEXT: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR1]](s32)
391    ; SI-NEXT: [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[COPY2]](<2 x s16>)
392    ; SI-NEXT: [[TRUNC4:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST2]](s32)
393    ; SI-NEXT: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST2]], [[C]](s32)
394    ; SI-NEXT: [[TRUNC5:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR2]](s32)
395    ; SI-NEXT: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC]](s16)
396    ; SI-NEXT: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC2]](s16)
397    ; SI-NEXT: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC4]](s16)
398    ; SI-NEXT: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[FPEXT]], [[FPEXT1]], [[FPEXT2]]
399    ; SI-NEXT: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMA]](s32)
400    ; SI-NEXT: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC1]](s16)
401    ; SI-NEXT: [[FPEXT4:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC3]](s16)
402    ; SI-NEXT: [[FPEXT5:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC5]](s16)
403    ; SI-NEXT: [[FMA1:%[0-9]+]]:_(s32) = G_FMA [[FPEXT3]], [[FPEXT4]], [[FPEXT5]]
404    ; SI-NEXT: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMA1]](s32)
405    ; SI-NEXT: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[FPTRUNC]](s16)
406    ; SI-NEXT: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[FPTRUNC1]](s16)
407    ; SI-NEXT: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[ZEXT1]], [[C]](s32)
408    ; SI-NEXT: [[OR:%[0-9]+]]:_(s32) = G_OR [[ZEXT]], [[SHL]]
409    ; SI-NEXT: [[BITCAST3:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR]](s32)
410    ; SI-NEXT: $vgpr0 = COPY [[BITCAST3]](<2 x s16>)
411    ;
412    ; VI-LABEL: name: test_fma_v2s16
413    ; VI: liveins: $vgpr0, $vgpr1, $vgpr2
414    ; VI-NEXT: {{  $}}
415    ; VI-NEXT: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
416    ; VI-NEXT: [[COPY1:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr1
417    ; VI-NEXT: [[COPY2:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr2
418    ; VI-NEXT: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY]](<2 x s16>)
419    ; VI-NEXT: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
420    ; VI-NEXT: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
421    ; VI-NEXT: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
422    ; VI-NEXT: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
423    ; VI-NEXT: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[COPY1]](<2 x s16>)
424    ; VI-NEXT: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST1]](s32)
425    ; VI-NEXT: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
426    ; VI-NEXT: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR1]](s32)
427    ; VI-NEXT: [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[COPY2]](<2 x s16>)
428    ; VI-NEXT: [[TRUNC4:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST2]](s32)
429    ; VI-NEXT: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST2]], [[C]](s32)
430    ; VI-NEXT: [[TRUNC5:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR2]](s32)
431    ; VI-NEXT: [[FMA:%[0-9]+]]:_(s16) = G_FMA [[TRUNC]], [[TRUNC2]], [[TRUNC4]]
432    ; VI-NEXT: [[FMA1:%[0-9]+]]:_(s16) = G_FMA [[TRUNC1]], [[TRUNC3]], [[TRUNC5]]
433    ; VI-NEXT: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[FMA]](s16)
434    ; VI-NEXT: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[FMA1]](s16)
435    ; VI-NEXT: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[ZEXT1]], [[C]](s32)
436    ; VI-NEXT: [[OR:%[0-9]+]]:_(s32) = G_OR [[ZEXT]], [[SHL]]
437    ; VI-NEXT: [[BITCAST3:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR]](s32)
438    ; VI-NEXT: $vgpr0 = COPY [[BITCAST3]](<2 x s16>)
439    ;
440    ; GFX9-LABEL: name: test_fma_v2s16
441    ; GFX9: liveins: $vgpr0, $vgpr1, $vgpr2
442    ; GFX9-NEXT: {{  $}}
443    ; GFX9-NEXT: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
444    ; GFX9-NEXT: [[COPY1:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr1
445    ; GFX9-NEXT: [[COPY2:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr2
446    ; GFX9-NEXT: [[FMA:%[0-9]+]]:_(<2 x s16>) = G_FMA [[COPY]], [[COPY1]], [[COPY2]]
447    ; GFX9-NEXT: $vgpr0 = COPY [[FMA]](<2 x s16>)
448    %0:_(<2 x s16>) = COPY $vgpr0
449    %1:_(<2 x s16>) = COPY $vgpr1
450    %2:_(<2 x s16>) = COPY $vgpr2
451    %3:_(<2 x s16>) = G_FMA %0, %1, %2
452    $vgpr0 = COPY %3
453...
454
455---
456name: test_fma_v3s16
457body: |
458  bb.0:
459    liveins: $vgpr0_vgpr1_vgpr2, $vgpr3_vgpr4_vgpr5, $vgpr6_vgpr7_vgpr8
460
461    ; SI-LABEL: name: test_fma_v3s16
462    ; SI: liveins: $vgpr0_vgpr1_vgpr2, $vgpr3_vgpr4_vgpr5, $vgpr6_vgpr7_vgpr8
463    ; SI-NEXT: {{  $}}
464    ; SI-NEXT: [[COPY:%[0-9]+]]:_(<6 x s16>) = COPY $vgpr0_vgpr1_vgpr2
465    ; SI-NEXT: [[COPY1:%[0-9]+]]:_(<6 x s16>) = COPY $vgpr3_vgpr4_vgpr5
466    ; SI-NEXT: [[COPY2:%[0-9]+]]:_(<6 x s16>) = COPY $vgpr6_vgpr7_vgpr8
467    ; SI-NEXT: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>), [[UV2:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<6 x s16>)
468    ; SI-NEXT: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV]](<2 x s16>)
469    ; SI-NEXT: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
470    ; SI-NEXT: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
471    ; SI-NEXT: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
472    ; SI-NEXT: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
473    ; SI-NEXT: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV1]](<2 x s16>)
474    ; SI-NEXT: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST1]](s32)
475    ; SI-NEXT: [[UV3:%[0-9]+]]:_(<2 x s16>), [[UV4:%[0-9]+]]:_(<2 x s16>), [[UV5:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY1]](<6 x s16>)
476    ; SI-NEXT: [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[UV3]](<2 x s16>)
477    ; SI-NEXT: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST2]](s32)
478    ; SI-NEXT: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST2]], [[C]](s32)
479    ; SI-NEXT: [[TRUNC4:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR1]](s32)
480    ; SI-NEXT: [[BITCAST3:%[0-9]+]]:_(s32) = G_BITCAST [[UV4]](<2 x s16>)
481    ; SI-NEXT: [[TRUNC5:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST3]](s32)
482    ; SI-NEXT: [[UV6:%[0-9]+]]:_(<2 x s16>), [[UV7:%[0-9]+]]:_(<2 x s16>), [[UV8:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY2]](<6 x s16>)
483    ; SI-NEXT: [[BITCAST4:%[0-9]+]]:_(s32) = G_BITCAST [[UV6]](<2 x s16>)
484    ; SI-NEXT: [[TRUNC6:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST4]](s32)
485    ; SI-NEXT: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST4]], [[C]](s32)
486    ; SI-NEXT: [[TRUNC7:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR2]](s32)
487    ; SI-NEXT: [[BITCAST5:%[0-9]+]]:_(s32) = G_BITCAST [[UV7]](<2 x s16>)
488    ; SI-NEXT: [[TRUNC8:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST5]](s32)
489    ; SI-NEXT: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC]](s16)
490    ; SI-NEXT: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC3]](s16)
491    ; SI-NEXT: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC6]](s16)
492    ; SI-NEXT: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[FPEXT]], [[FPEXT1]], [[FPEXT2]]
493    ; SI-NEXT: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMA]](s32)
494    ; SI-NEXT: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC1]](s16)
495    ; SI-NEXT: [[FPEXT4:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC4]](s16)
496    ; SI-NEXT: [[FPEXT5:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC7]](s16)
497    ; SI-NEXT: [[FMA1:%[0-9]+]]:_(s32) = G_FMA [[FPEXT3]], [[FPEXT4]], [[FPEXT5]]
498    ; SI-NEXT: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMA1]](s32)
499    ; SI-NEXT: [[FPEXT6:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC2]](s16)
500    ; SI-NEXT: [[FPEXT7:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC5]](s16)
501    ; SI-NEXT: [[FPEXT8:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC8]](s16)
502    ; SI-NEXT: [[FMA2:%[0-9]+]]:_(s32) = G_FMA [[FPEXT6]], [[FPEXT7]], [[FPEXT8]]
503    ; SI-NEXT: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMA2]](s32)
504    ; SI-NEXT: [[DEF:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
505    ; SI-NEXT: [[UV9:%[0-9]+]]:_(<2 x s16>), [[UV10:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[DEF]](<4 x s16>)
506    ; SI-NEXT: [[BITCAST6:%[0-9]+]]:_(s32) = G_BITCAST [[UV9]](<2 x s16>)
507    ; SI-NEXT: [[LSHR3:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST6]], [[C]](s32)
508    ; SI-NEXT: [[BITCAST7:%[0-9]+]]:_(s32) = G_BITCAST [[UV10]](<2 x s16>)
509    ; SI-NEXT: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[FPTRUNC]](s16)
510    ; SI-NEXT: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[FPTRUNC1]](s16)
511    ; SI-NEXT: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[ZEXT1]], [[C]](s32)
512    ; SI-NEXT: [[OR:%[0-9]+]]:_(s32) = G_OR [[ZEXT]], [[SHL]]
513    ; SI-NEXT: [[BITCAST8:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR]](s32)
514    ; SI-NEXT: [[ZEXT2:%[0-9]+]]:_(s32) = G_ZEXT [[FPTRUNC2]](s16)
515    ; SI-NEXT: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
516    ; SI-NEXT: [[AND:%[0-9]+]]:_(s32) = G_AND [[BITCAST6]], [[C1]]
517    ; SI-NEXT: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND]], [[C]](s32)
518    ; SI-NEXT: [[OR1:%[0-9]+]]:_(s32) = G_OR [[ZEXT2]], [[SHL1]]
519    ; SI-NEXT: [[BITCAST9:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR1]](s32)
520    ; SI-NEXT: [[AND1:%[0-9]+]]:_(s32) = G_AND [[BITCAST7]], [[C1]]
521    ; SI-NEXT: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C]](s32)
522    ; SI-NEXT: [[OR2:%[0-9]+]]:_(s32) = G_OR [[LSHR3]], [[SHL2]]
523    ; SI-NEXT: [[BITCAST10:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR2]](s32)
524    ; SI-NEXT: [[CONCAT_VECTORS:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[BITCAST8]](<2 x s16>), [[BITCAST9]](<2 x s16>), [[BITCAST10]](<2 x s16>)
525    ; SI-NEXT: $vgpr0_vgpr1_vgpr2 = COPY [[CONCAT_VECTORS]](<6 x s16>)
526    ;
527    ; VI-LABEL: name: test_fma_v3s16
528    ; VI: liveins: $vgpr0_vgpr1_vgpr2, $vgpr3_vgpr4_vgpr5, $vgpr6_vgpr7_vgpr8
529    ; VI-NEXT: {{  $}}
530    ; VI-NEXT: [[COPY:%[0-9]+]]:_(<6 x s16>) = COPY $vgpr0_vgpr1_vgpr2
531    ; VI-NEXT: [[COPY1:%[0-9]+]]:_(<6 x s16>) = COPY $vgpr3_vgpr4_vgpr5
532    ; VI-NEXT: [[COPY2:%[0-9]+]]:_(<6 x s16>) = COPY $vgpr6_vgpr7_vgpr8
533    ; VI-NEXT: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>), [[UV2:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<6 x s16>)
534    ; VI-NEXT: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV]](<2 x s16>)
535    ; VI-NEXT: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
536    ; VI-NEXT: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
537    ; VI-NEXT: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
538    ; VI-NEXT: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
539    ; VI-NEXT: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV1]](<2 x s16>)
540    ; VI-NEXT: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST1]](s32)
541    ; VI-NEXT: [[UV3:%[0-9]+]]:_(<2 x s16>), [[UV4:%[0-9]+]]:_(<2 x s16>), [[UV5:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY1]](<6 x s16>)
542    ; VI-NEXT: [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[UV3]](<2 x s16>)
543    ; VI-NEXT: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST2]](s32)
544    ; VI-NEXT: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST2]], [[C]](s32)
545    ; VI-NEXT: [[TRUNC4:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR1]](s32)
546    ; VI-NEXT: [[BITCAST3:%[0-9]+]]:_(s32) = G_BITCAST [[UV4]](<2 x s16>)
547    ; VI-NEXT: [[TRUNC5:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST3]](s32)
548    ; VI-NEXT: [[UV6:%[0-9]+]]:_(<2 x s16>), [[UV7:%[0-9]+]]:_(<2 x s16>), [[UV8:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY2]](<6 x s16>)
549    ; VI-NEXT: [[BITCAST4:%[0-9]+]]:_(s32) = G_BITCAST [[UV6]](<2 x s16>)
550    ; VI-NEXT: [[TRUNC6:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST4]](s32)
551    ; VI-NEXT: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST4]], [[C]](s32)
552    ; VI-NEXT: [[TRUNC7:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR2]](s32)
553    ; VI-NEXT: [[BITCAST5:%[0-9]+]]:_(s32) = G_BITCAST [[UV7]](<2 x s16>)
554    ; VI-NEXT: [[TRUNC8:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST5]](s32)
555    ; VI-NEXT: [[FMA:%[0-9]+]]:_(s16) = G_FMA [[TRUNC]], [[TRUNC3]], [[TRUNC6]]
556    ; VI-NEXT: [[FMA1:%[0-9]+]]:_(s16) = G_FMA [[TRUNC1]], [[TRUNC4]], [[TRUNC7]]
557    ; VI-NEXT: [[FMA2:%[0-9]+]]:_(s16) = G_FMA [[TRUNC2]], [[TRUNC5]], [[TRUNC8]]
558    ; VI-NEXT: [[DEF:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
559    ; VI-NEXT: [[UV9:%[0-9]+]]:_(<2 x s16>), [[UV10:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[DEF]](<4 x s16>)
560    ; VI-NEXT: [[BITCAST6:%[0-9]+]]:_(s32) = G_BITCAST [[UV9]](<2 x s16>)
561    ; VI-NEXT: [[LSHR3:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST6]], [[C]](s32)
562    ; VI-NEXT: [[BITCAST7:%[0-9]+]]:_(s32) = G_BITCAST [[UV10]](<2 x s16>)
563    ; VI-NEXT: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[FMA]](s16)
564    ; VI-NEXT: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[FMA1]](s16)
565    ; VI-NEXT: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[ZEXT1]], [[C]](s32)
566    ; VI-NEXT: [[OR:%[0-9]+]]:_(s32) = G_OR [[ZEXT]], [[SHL]]
567    ; VI-NEXT: [[BITCAST8:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR]](s32)
568    ; VI-NEXT: [[ZEXT2:%[0-9]+]]:_(s32) = G_ZEXT [[FMA2]](s16)
569    ; VI-NEXT: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
570    ; VI-NEXT: [[AND:%[0-9]+]]:_(s32) = G_AND [[BITCAST6]], [[C1]]
571    ; VI-NEXT: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND]], [[C]](s32)
572    ; VI-NEXT: [[OR1:%[0-9]+]]:_(s32) = G_OR [[ZEXT2]], [[SHL1]]
573    ; VI-NEXT: [[BITCAST9:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR1]](s32)
574    ; VI-NEXT: [[AND1:%[0-9]+]]:_(s32) = G_AND [[BITCAST7]], [[C1]]
575    ; VI-NEXT: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C]](s32)
576    ; VI-NEXT: [[OR2:%[0-9]+]]:_(s32) = G_OR [[LSHR3]], [[SHL2]]
577    ; VI-NEXT: [[BITCAST10:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR2]](s32)
578    ; VI-NEXT: [[CONCAT_VECTORS:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[BITCAST8]](<2 x s16>), [[BITCAST9]](<2 x s16>), [[BITCAST10]](<2 x s16>)
579    ; VI-NEXT: $vgpr0_vgpr1_vgpr2 = COPY [[CONCAT_VECTORS]](<6 x s16>)
580    ;
581    ; GFX9-LABEL: name: test_fma_v3s16
582    ; GFX9: liveins: $vgpr0_vgpr1_vgpr2, $vgpr3_vgpr4_vgpr5, $vgpr6_vgpr7_vgpr8
583    ; GFX9-NEXT: {{  $}}
584    ; GFX9-NEXT: [[COPY:%[0-9]+]]:_(<6 x s16>) = COPY $vgpr0_vgpr1_vgpr2
585    ; GFX9-NEXT: [[COPY1:%[0-9]+]]:_(<6 x s16>) = COPY $vgpr3_vgpr4_vgpr5
586    ; GFX9-NEXT: [[COPY2:%[0-9]+]]:_(<6 x s16>) = COPY $vgpr6_vgpr7_vgpr8
587    ; GFX9-NEXT: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>), [[UV2:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<6 x s16>)
588    ; GFX9-NEXT: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV1]](<2 x s16>)
589    ; GFX9-NEXT: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
590    ; GFX9-NEXT: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
591    ; GFX9-NEXT: [[DEF:%[0-9]+]]:_(s16) = G_IMPLICIT_DEF
592    ; GFX9-NEXT: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR [[TRUNC]](s16), [[DEF]](s16)
593    ; GFX9-NEXT: [[UV3:%[0-9]+]]:_(<2 x s16>), [[UV4:%[0-9]+]]:_(<2 x s16>), [[UV5:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY1]](<6 x s16>)
594    ; GFX9-NEXT: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV4]](<2 x s16>)
595    ; GFX9-NEXT: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST1]](s32)
596    ; GFX9-NEXT: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR [[TRUNC1]](s16), [[DEF]](s16)
597    ; GFX9-NEXT: [[UV6:%[0-9]+]]:_(<2 x s16>), [[UV7:%[0-9]+]]:_(<2 x s16>), [[UV8:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY2]](<6 x s16>)
598    ; GFX9-NEXT: [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[UV7]](<2 x s16>)
599    ; GFX9-NEXT: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST2]](s32)
600    ; GFX9-NEXT: [[BUILD_VECTOR2:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR [[TRUNC2]](s16), [[DEF]](s16)
601    ; GFX9-NEXT: [[FMA:%[0-9]+]]:_(<2 x s16>) = G_FMA [[UV]], [[UV3]], [[UV6]]
602    ; GFX9-NEXT: [[FMA1:%[0-9]+]]:_(<2 x s16>) = G_FMA [[BUILD_VECTOR]], [[BUILD_VECTOR1]], [[BUILD_VECTOR2]]
603    ; GFX9-NEXT: [[BITCAST3:%[0-9]+]]:_(s32) = G_BITCAST [[FMA]](<2 x s16>)
604    ; GFX9-NEXT: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST3]](s32)
605    ; GFX9-NEXT: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST3]], [[C]](s32)
606    ; GFX9-NEXT: [[TRUNC4:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
607    ; GFX9-NEXT: [[BITCAST4:%[0-9]+]]:_(s32) = G_BITCAST [[FMA1]](<2 x s16>)
608    ; GFX9-NEXT: [[TRUNC5:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST4]](s32)
609    ; GFX9-NEXT: [[DEF1:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
610    ; GFX9-NEXT: [[UV9:%[0-9]+]]:_(<2 x s16>), [[UV10:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[DEF1]](<4 x s16>)
611    ; GFX9-NEXT: [[BITCAST5:%[0-9]+]]:_(s32) = G_BITCAST [[UV9]](<2 x s16>)
612    ; GFX9-NEXT: [[TRUNC6:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST5]](s32)
613    ; GFX9-NEXT: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST5]], [[C]](s32)
614    ; GFX9-NEXT: [[TRUNC7:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR1]](s32)
615    ; GFX9-NEXT: [[BITCAST6:%[0-9]+]]:_(s32) = G_BITCAST [[UV10]](<2 x s16>)
616    ; GFX9-NEXT: [[TRUNC8:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST6]](s32)
617    ; GFX9-NEXT: [[BUILD_VECTOR3:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR [[TRUNC3]](s16), [[TRUNC4]](s16)
618    ; GFX9-NEXT: [[BUILD_VECTOR4:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR [[TRUNC5]](s16), [[TRUNC6]](s16)
619    ; GFX9-NEXT: [[BUILD_VECTOR5:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR [[TRUNC7]](s16), [[TRUNC8]](s16)
620    ; GFX9-NEXT: [[CONCAT_VECTORS:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR3]](<2 x s16>), [[BUILD_VECTOR4]](<2 x s16>), [[BUILD_VECTOR5]](<2 x s16>)
621    ; GFX9-NEXT: $vgpr0_vgpr1_vgpr2 = COPY [[CONCAT_VECTORS]](<6 x s16>)
622    %0:_(<6 x s16>) = COPY $vgpr0_vgpr1_vgpr2
623    %1:_(<6 x s16>) = COPY $vgpr3_vgpr4_vgpr5
624    %2:_(<6 x s16>) = COPY $vgpr6_vgpr7_vgpr8
625    %3:_(<3 x s16>), %4:_(<3 x s16>) = G_UNMERGE_VALUES %0
626    %5:_(<3 x s16>), %6:_(<3 x s16>) = G_UNMERGE_VALUES %1
627    %7:_(<3 x s16>), %8:_(<3 x s16>) = G_UNMERGE_VALUES %2
628    %9:_(<3 x s16>) = G_FMA %3, %5, %7
629    %10:_(<3 x s16>) = G_IMPLICIT_DEF
630    %11:_(<6 x s16>) = G_CONCAT_VECTORS %9, %10
631    $vgpr0_vgpr1_vgpr2 = COPY %11
632
633...
634
635---
636name: test_fma_v4s16
637body: |
638  bb.0:
639    liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $vgpr4_vgpr5
640
641    ; SI-LABEL: name: test_fma_v4s16
642    ; SI: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $vgpr4_vgpr5
643    ; SI-NEXT: {{  $}}
644    ; SI-NEXT: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
645    ; SI-NEXT: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3
646    ; SI-NEXT: [[COPY2:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr4_vgpr5
647    ; SI-NEXT: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
648    ; SI-NEXT: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV]](<2 x s16>)
649    ; SI-NEXT: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
650    ; SI-NEXT: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
651    ; SI-NEXT: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
652    ; SI-NEXT: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
653    ; SI-NEXT: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV1]](<2 x s16>)
654    ; SI-NEXT: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST1]](s32)
655    ; SI-NEXT: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
656    ; SI-NEXT: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR1]](s32)
657    ; SI-NEXT: [[UV2:%[0-9]+]]:_(<2 x s16>), [[UV3:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>)
658    ; SI-NEXT: [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[UV2]](<2 x s16>)
659    ; SI-NEXT: [[TRUNC4:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST2]](s32)
660    ; SI-NEXT: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST2]], [[C]](s32)
661    ; SI-NEXT: [[TRUNC5:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR2]](s32)
662    ; SI-NEXT: [[BITCAST3:%[0-9]+]]:_(s32) = G_BITCAST [[UV3]](<2 x s16>)
663    ; SI-NEXT: [[TRUNC6:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST3]](s32)
664    ; SI-NEXT: [[LSHR3:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST3]], [[C]](s32)
665    ; SI-NEXT: [[TRUNC7:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR3]](s32)
666    ; SI-NEXT: [[UV4:%[0-9]+]]:_(<2 x s16>), [[UV5:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY2]](<4 x s16>)
667    ; SI-NEXT: [[BITCAST4:%[0-9]+]]:_(s32) = G_BITCAST [[UV4]](<2 x s16>)
668    ; SI-NEXT: [[TRUNC8:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST4]](s32)
669    ; SI-NEXT: [[LSHR4:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST4]], [[C]](s32)
670    ; SI-NEXT: [[TRUNC9:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR4]](s32)
671    ; SI-NEXT: [[BITCAST5:%[0-9]+]]:_(s32) = G_BITCAST [[UV5]](<2 x s16>)
672    ; SI-NEXT: [[TRUNC10:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST5]](s32)
673    ; SI-NEXT: [[LSHR5:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST5]], [[C]](s32)
674    ; SI-NEXT: [[TRUNC11:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR5]](s32)
675    ; SI-NEXT: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC]](s16)
676    ; SI-NEXT: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC4]](s16)
677    ; SI-NEXT: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC8]](s16)
678    ; SI-NEXT: [[FMA:%[0-9]+]]:_(s32) = G_FMA [[FPEXT]], [[FPEXT1]], [[FPEXT2]]
679    ; SI-NEXT: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMA]](s32)
680    ; SI-NEXT: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC1]](s16)
681    ; SI-NEXT: [[FPEXT4:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC5]](s16)
682    ; SI-NEXT: [[FPEXT5:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC9]](s16)
683    ; SI-NEXT: [[FMA1:%[0-9]+]]:_(s32) = G_FMA [[FPEXT3]], [[FPEXT4]], [[FPEXT5]]
684    ; SI-NEXT: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMA1]](s32)
685    ; SI-NEXT: [[FPEXT6:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC2]](s16)
686    ; SI-NEXT: [[FPEXT7:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC6]](s16)
687    ; SI-NEXT: [[FPEXT8:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC10]](s16)
688    ; SI-NEXT: [[FMA2:%[0-9]+]]:_(s32) = G_FMA [[FPEXT6]], [[FPEXT7]], [[FPEXT8]]
689    ; SI-NEXT: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMA2]](s32)
690    ; SI-NEXT: [[FPEXT9:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC3]](s16)
691    ; SI-NEXT: [[FPEXT10:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC7]](s16)
692    ; SI-NEXT: [[FPEXT11:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC11]](s16)
693    ; SI-NEXT: [[FMA3:%[0-9]+]]:_(s32) = G_FMA [[FPEXT9]], [[FPEXT10]], [[FPEXT11]]
694    ; SI-NEXT: [[FPTRUNC3:%[0-9]+]]:_(s16) = G_FPTRUNC [[FMA3]](s32)
695    ; SI-NEXT: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[FPTRUNC]](s16)
696    ; SI-NEXT: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[FPTRUNC1]](s16)
697    ; SI-NEXT: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[ZEXT1]], [[C]](s32)
698    ; SI-NEXT: [[OR:%[0-9]+]]:_(s32) = G_OR [[ZEXT]], [[SHL]]
699    ; SI-NEXT: [[BITCAST6:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR]](s32)
700    ; SI-NEXT: [[ZEXT2:%[0-9]+]]:_(s32) = G_ZEXT [[FPTRUNC2]](s16)
701    ; SI-NEXT: [[ZEXT3:%[0-9]+]]:_(s32) = G_ZEXT [[FPTRUNC3]](s16)
702    ; SI-NEXT: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[ZEXT3]], [[C]](s32)
703    ; SI-NEXT: [[OR1:%[0-9]+]]:_(s32) = G_OR [[ZEXT2]], [[SHL1]]
704    ; SI-NEXT: [[BITCAST7:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR1]](s32)
705    ; SI-NEXT: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BITCAST6]](<2 x s16>), [[BITCAST7]](<2 x s16>)
706    ; SI-NEXT: $vgpr0_vgpr1 = COPY [[CONCAT_VECTORS]](<4 x s16>)
707    ;
708    ; VI-LABEL: name: test_fma_v4s16
709    ; VI: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $vgpr4_vgpr5
710    ; VI-NEXT: {{  $}}
711    ; VI-NEXT: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
712    ; VI-NEXT: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3
713    ; VI-NEXT: [[COPY2:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr4_vgpr5
714    ; VI-NEXT: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
715    ; VI-NEXT: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV]](<2 x s16>)
716    ; VI-NEXT: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
717    ; VI-NEXT: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
718    ; VI-NEXT: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
719    ; VI-NEXT: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
720    ; VI-NEXT: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV1]](<2 x s16>)
721    ; VI-NEXT: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST1]](s32)
722    ; VI-NEXT: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
723    ; VI-NEXT: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR1]](s32)
724    ; VI-NEXT: [[UV2:%[0-9]+]]:_(<2 x s16>), [[UV3:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>)
725    ; VI-NEXT: [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[UV2]](<2 x s16>)
726    ; VI-NEXT: [[TRUNC4:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST2]](s32)
727    ; VI-NEXT: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST2]], [[C]](s32)
728    ; VI-NEXT: [[TRUNC5:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR2]](s32)
729    ; VI-NEXT: [[BITCAST3:%[0-9]+]]:_(s32) = G_BITCAST [[UV3]](<2 x s16>)
730    ; VI-NEXT: [[TRUNC6:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST3]](s32)
731    ; VI-NEXT: [[LSHR3:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST3]], [[C]](s32)
732    ; VI-NEXT: [[TRUNC7:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR3]](s32)
733    ; VI-NEXT: [[UV4:%[0-9]+]]:_(<2 x s16>), [[UV5:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY2]](<4 x s16>)
734    ; VI-NEXT: [[BITCAST4:%[0-9]+]]:_(s32) = G_BITCAST [[UV4]](<2 x s16>)
735    ; VI-NEXT: [[TRUNC8:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST4]](s32)
736    ; VI-NEXT: [[LSHR4:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST4]], [[C]](s32)
737    ; VI-NEXT: [[TRUNC9:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR4]](s32)
738    ; VI-NEXT: [[BITCAST5:%[0-9]+]]:_(s32) = G_BITCAST [[UV5]](<2 x s16>)
739    ; VI-NEXT: [[TRUNC10:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST5]](s32)
740    ; VI-NEXT: [[LSHR5:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST5]], [[C]](s32)
741    ; VI-NEXT: [[TRUNC11:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR5]](s32)
742    ; VI-NEXT: [[FMA:%[0-9]+]]:_(s16) = G_FMA [[TRUNC]], [[TRUNC4]], [[TRUNC8]]
743    ; VI-NEXT: [[FMA1:%[0-9]+]]:_(s16) = G_FMA [[TRUNC1]], [[TRUNC5]], [[TRUNC9]]
744    ; VI-NEXT: [[FMA2:%[0-9]+]]:_(s16) = G_FMA [[TRUNC2]], [[TRUNC6]], [[TRUNC10]]
745    ; VI-NEXT: [[FMA3:%[0-9]+]]:_(s16) = G_FMA [[TRUNC3]], [[TRUNC7]], [[TRUNC11]]
746    ; VI-NEXT: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[FMA]](s16)
747    ; VI-NEXT: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[FMA1]](s16)
748    ; VI-NEXT: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[ZEXT1]], [[C]](s32)
749    ; VI-NEXT: [[OR:%[0-9]+]]:_(s32) = G_OR [[ZEXT]], [[SHL]]
750    ; VI-NEXT: [[BITCAST6:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR]](s32)
751    ; VI-NEXT: [[ZEXT2:%[0-9]+]]:_(s32) = G_ZEXT [[FMA2]](s16)
752    ; VI-NEXT: [[ZEXT3:%[0-9]+]]:_(s32) = G_ZEXT [[FMA3]](s16)
753    ; VI-NEXT: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[ZEXT3]], [[C]](s32)
754    ; VI-NEXT: [[OR1:%[0-9]+]]:_(s32) = G_OR [[ZEXT2]], [[SHL1]]
755    ; VI-NEXT: [[BITCAST7:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR1]](s32)
756    ; VI-NEXT: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BITCAST6]](<2 x s16>), [[BITCAST7]](<2 x s16>)
757    ; VI-NEXT: $vgpr0_vgpr1 = COPY [[CONCAT_VECTORS]](<4 x s16>)
758    ;
759    ; GFX9-LABEL: name: test_fma_v4s16
760    ; GFX9: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $vgpr4_vgpr5
761    ; GFX9-NEXT: {{  $}}
762    ; GFX9-NEXT: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
763    ; GFX9-NEXT: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3
764    ; GFX9-NEXT: [[COPY2:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr4_vgpr5
765    ; GFX9-NEXT: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
766    ; GFX9-NEXT: [[UV2:%[0-9]+]]:_(<2 x s16>), [[UV3:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>)
767    ; GFX9-NEXT: [[UV4:%[0-9]+]]:_(<2 x s16>), [[UV5:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY2]](<4 x s16>)
768    ; GFX9-NEXT: [[FMA:%[0-9]+]]:_(<2 x s16>) = G_FMA [[UV]], [[UV2]], [[UV4]]
769    ; GFX9-NEXT: [[FMA1:%[0-9]+]]:_(<2 x s16>) = G_FMA [[UV1]], [[UV3]], [[UV5]]
770    ; GFX9-NEXT: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[FMA]](<2 x s16>), [[FMA1]](<2 x s16>)
771    ; GFX9-NEXT: $vgpr0_vgpr1 = COPY [[CONCAT_VECTORS]](<4 x s16>)
772    %0:_(<4 x s16>) = COPY $vgpr0_vgpr1
773    %1:_(<4 x s16>) = COPY $vgpr2_vgpr3
774    %2:_(<4 x s16>) = COPY $vgpr4_vgpr5
775    %3:_(<4 x s16>) = G_FMA %0, %1, %2
776    $vgpr0_vgpr1 = COPY %3
777...
778