xref: /llvm-project/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-smulh.mir (revision 373c343a77a7afaa07179db1754a52b620dfaf2e) 
1# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
2# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=fiji -O0 -run-pass=legalizer %s -o - | FileCheck %s --check-prefix=GFX8
3# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx906 -O0 -run-pass=legalizer %s -o - | FileCheck %s --check-prefix=GFX9
4
5---
6name: test_smulh_s32
7body: |
8  bb.0:
9    liveins: $vgpr0, $vgpr1
10
11    ; GFX8-LABEL: name: test_smulh_s32
12    ; GFX8: liveins: $vgpr0, $vgpr1
13    ; GFX8-NEXT: {{  $}}
14    ; GFX8-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
15    ; GFX8-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
16    ; GFX8-NEXT: [[SMULH:%[0-9]+]]:_(s32) = G_SMULH [[COPY]], [[COPY1]]
17    ; GFX8-NEXT: $vgpr0 = COPY [[SMULH]](s32)
18    ;
19    ; GFX9-LABEL: name: test_smulh_s32
20    ; GFX9: liveins: $vgpr0, $vgpr1
21    ; GFX9-NEXT: {{  $}}
22    ; GFX9-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
23    ; GFX9-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
24    ; GFX9-NEXT: [[SMULH:%[0-9]+]]:_(s32) = G_SMULH [[COPY]], [[COPY1]]
25    ; GFX9-NEXT: $vgpr0 = COPY [[SMULH]](s32)
26    %0:_(s32) = COPY $vgpr0
27    %1:_(s32) = COPY $vgpr1
28    %2:_(s32) = G_SMULH %0, %1
29    $vgpr0 = COPY %2
30...
31
32---
33name: test_smulh_v2s32
34body: |
35  bb.0:
36    liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
37
38    ; GFX8-LABEL: name: test_smulh_v2s32
39    ; GFX8: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
40    ; GFX8-NEXT: {{  $}}
41    ; GFX8-NEXT: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
42    ; GFX8-NEXT: [[COPY1:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr2_vgpr3
43    ; GFX8-NEXT: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<2 x s32>)
44    ; GFX8-NEXT: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<2 x s32>)
45    ; GFX8-NEXT: [[SMULH:%[0-9]+]]:_(s32) = G_SMULH [[UV]], [[UV2]]
46    ; GFX8-NEXT: [[SMULH1:%[0-9]+]]:_(s32) = G_SMULH [[UV1]], [[UV3]]
47    ; GFX8-NEXT: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[SMULH]](s32), [[SMULH1]](s32)
48    ; GFX8-NEXT: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
49    ;
50    ; GFX9-LABEL: name: test_smulh_v2s32
51    ; GFX9: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
52    ; GFX9-NEXT: {{  $}}
53    ; GFX9-NEXT: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
54    ; GFX9-NEXT: [[COPY1:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr2_vgpr3
55    ; GFX9-NEXT: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<2 x s32>)
56    ; GFX9-NEXT: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<2 x s32>)
57    ; GFX9-NEXT: [[SMULH:%[0-9]+]]:_(s32) = G_SMULH [[UV]], [[UV2]]
58    ; GFX9-NEXT: [[SMULH1:%[0-9]+]]:_(s32) = G_SMULH [[UV1]], [[UV3]]
59    ; GFX9-NEXT: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[SMULH]](s32), [[SMULH1]](s32)
60    ; GFX9-NEXT: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
61    %0:_(<2 x s32>) = COPY $vgpr0_vgpr1
62    %1:_(<2 x s32>) = COPY $vgpr2_vgpr3
63    %2:_(<2 x s32>) = G_SMULH %0, %1
64    $vgpr0_vgpr1 = COPY %2
65...
66
67---
68name: test_smulh_s16
69body: |
70  bb.0:
71    liveins: $vgpr0, $vgpr1
72
73    ; GFX8-LABEL: name: test_smulh_s16
74    ; GFX8: liveins: $vgpr0, $vgpr1
75    ; GFX8-NEXT: {{  $}}
76    ; GFX8-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
77    ; GFX8-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
78    ; GFX8-NEXT: [[SEXT_INREG:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY]], 16
79    ; GFX8-NEXT: [[SEXT_INREG1:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY1]], 16
80    ; GFX8-NEXT: [[MUL:%[0-9]+]]:_(s32) = G_MUL [[SEXT_INREG]], [[SEXT_INREG1]]
81    ; GFX8-NEXT: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
82    ; GFX8-NEXT: [[ASHR:%[0-9]+]]:_(s32) = G_ASHR [[MUL]], [[C]](s32)
83    ; GFX8-NEXT: [[SEXT_INREG2:%[0-9]+]]:_(s32) = G_SEXT_INREG [[ASHR]], 16
84    ; GFX8-NEXT: $vgpr0 = COPY [[SEXT_INREG2]](s32)
85    ;
86    ; GFX9-LABEL: name: test_smulh_s16
87    ; GFX9: liveins: $vgpr0, $vgpr1
88    ; GFX9-NEXT: {{  $}}
89    ; GFX9-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
90    ; GFX9-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
91    ; GFX9-NEXT: [[SEXT_INREG:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY]], 16
92    ; GFX9-NEXT: [[SEXT_INREG1:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY1]], 16
93    ; GFX9-NEXT: [[MUL:%[0-9]+]]:_(s32) = G_MUL [[SEXT_INREG]], [[SEXT_INREG1]]
94    ; GFX9-NEXT: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
95    ; GFX9-NEXT: [[ASHR:%[0-9]+]]:_(s32) = G_ASHR [[MUL]], [[C]](s32)
96    ; GFX9-NEXT: [[SEXT_INREG2:%[0-9]+]]:_(s32) = G_SEXT_INREG [[ASHR]], 16
97    ; GFX9-NEXT: $vgpr0 = COPY [[SEXT_INREG2]](s32)
98    %0:_(s32) = COPY $vgpr0
99    %1:_(s32) = COPY $vgpr1
100    %2:_(s16) = G_TRUNC %0
101    %3:_(s16) = G_TRUNC %1
102    %4:_(s16) = G_SMULH %2, %3
103    %5:_(s32) = G_SEXT %4
104    $vgpr0 = COPY %5
105...
106
107---
108name: test_smulh_s8
109body: |
110  bb.0:
111    liveins: $vgpr0, $vgpr1
112
113    ; GFX8-LABEL: name: test_smulh_s8
114    ; GFX8: liveins: $vgpr0, $vgpr1
115    ; GFX8-NEXT: {{  $}}
116    ; GFX8-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
117    ; GFX8-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
118    ; GFX8-NEXT: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
119    ; GFX8-NEXT: [[C:%[0-9]+]]:_(s16) = G_CONSTANT i16 8
120    ; GFX8-NEXT: [[SHL:%[0-9]+]]:_(s16) = G_SHL [[TRUNC]], [[C]](s16)
121    ; GFX8-NEXT: [[ASHR:%[0-9]+]]:_(s16) = G_ASHR [[SHL]], [[C]](s16)
122    ; GFX8-NEXT: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32)
123    ; GFX8-NEXT: [[SHL1:%[0-9]+]]:_(s16) = G_SHL [[TRUNC1]], [[C]](s16)
124    ; GFX8-NEXT: [[ASHR1:%[0-9]+]]:_(s16) = G_ASHR [[SHL1]], [[C]](s16)
125    ; GFX8-NEXT: [[MUL:%[0-9]+]]:_(s16) = G_MUL [[ASHR]], [[ASHR1]]
126    ; GFX8-NEXT: [[ASHR2:%[0-9]+]]:_(s16) = G_ASHR [[MUL]], [[C]](s16)
127    ; GFX8-NEXT: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[ASHR2]](s16)
128    ; GFX8-NEXT: [[SEXT_INREG:%[0-9]+]]:_(s32) = G_SEXT_INREG [[ANYEXT]], 8
129    ; GFX8-NEXT: $vgpr0 = COPY [[SEXT_INREG]](s32)
130    ;
131    ; GFX9-LABEL: name: test_smulh_s8
132    ; GFX9: liveins: $vgpr0, $vgpr1
133    ; GFX9-NEXT: {{  $}}
134    ; GFX9-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
135    ; GFX9-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
136    ; GFX9-NEXT: [[SEXT_INREG:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY]], 8
137    ; GFX9-NEXT: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[SEXT_INREG]](s32)
138    ; GFX9-NEXT: [[SEXT_INREG1:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY1]], 8
139    ; GFX9-NEXT: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[SEXT_INREG1]](s32)
140    ; GFX9-NEXT: [[MUL:%[0-9]+]]:_(s16) = G_MUL [[TRUNC]], [[TRUNC1]]
141    ; GFX9-NEXT: [[C:%[0-9]+]]:_(s16) = G_CONSTANT i16 8
142    ; GFX9-NEXT: [[ASHR:%[0-9]+]]:_(s16) = G_ASHR [[MUL]], [[C]](s16)
143    ; GFX9-NEXT: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[ASHR]](s16)
144    ; GFX9-NEXT: [[SEXT_INREG2:%[0-9]+]]:_(s32) = G_SEXT_INREG [[ANYEXT]], 8
145    ; GFX9-NEXT: $vgpr0 = COPY [[SEXT_INREG2]](s32)
146    %0:_(s32) = COPY $vgpr0
147    %1:_(s32) = COPY $vgpr1
148    %2:_(s8) = G_TRUNC %0
149    %3:_(s8) = G_TRUNC %1
150    %4:_(s8) = G_SMULH %2, %3
151    %5:_(s32) = G_SEXT %4
152    $vgpr0 = COPY %5
153...
154
155---
156name: test_smulh_v2s16
157body: |
158  bb.0:
159    liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
160    ; GFX8-LABEL: name: test_smulh_v2s16
161    ; GFX8: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
162    ; GFX8-NEXT: {{  $}}
163    ; GFX8-NEXT: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
164    ; GFX8-NEXT: [[COPY1:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr2_vgpr3
165    ; GFX8-NEXT: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<2 x s32>)
166    ; GFX8-NEXT: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<2 x s32>)
167    ; GFX8-NEXT: [[SEXT_INREG:%[0-9]+]]:_(s32) = G_SEXT_INREG [[UV]], 16
168    ; GFX8-NEXT: [[SEXT_INREG1:%[0-9]+]]:_(s32) = G_SEXT_INREG [[UV2]], 16
169    ; GFX8-NEXT: [[MUL:%[0-9]+]]:_(s32) = G_MUL [[SEXT_INREG]], [[SEXT_INREG1]]
170    ; GFX8-NEXT: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
171    ; GFX8-NEXT: [[ASHR:%[0-9]+]]:_(s32) = G_ASHR [[MUL]], [[C]](s32)
172    ; GFX8-NEXT: [[SEXT_INREG2:%[0-9]+]]:_(s32) = G_SEXT_INREG [[UV1]], 16
173    ; GFX8-NEXT: [[SEXT_INREG3:%[0-9]+]]:_(s32) = G_SEXT_INREG [[UV3]], 16
174    ; GFX8-NEXT: [[MUL1:%[0-9]+]]:_(s32) = G_MUL [[SEXT_INREG2]], [[SEXT_INREG3]]
175    ; GFX8-NEXT: [[ASHR1:%[0-9]+]]:_(s32) = G_ASHR [[MUL1]], [[C]](s32)
176    ; GFX8-NEXT: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
177    ; GFX8-NEXT: [[AND:%[0-9]+]]:_(s32) = G_AND [[ASHR]], [[C1]]
178    ; GFX8-NEXT: [[AND1:%[0-9]+]]:_(s32) = G_AND [[ASHR1]], [[C1]]
179    ; GFX8-NEXT: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C]](s32)
180    ; GFX8-NEXT: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
181    ; GFX8-NEXT: [[BITCAST:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR]](s32)
182    ; GFX8-NEXT: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[BITCAST]](<2 x s16>)
183    ; GFX8-NEXT: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
184    ; GFX8-NEXT: [[SEXT_INREG4:%[0-9]+]]:_(s32) = G_SEXT_INREG [[BITCAST1]], 16
185    ; GFX8-NEXT: [[SEXT_INREG5:%[0-9]+]]:_(s32) = G_SEXT_INREG [[LSHR]], 16
186    ; GFX8-NEXT: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[SEXT_INREG4]](s32), [[SEXT_INREG5]](s32)
187    ; GFX8-NEXT: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
188    ;
189    ; GFX9-LABEL: name: test_smulh_v2s16
190    ; GFX9: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
191    ; GFX9-NEXT: {{  $}}
192    ; GFX9-NEXT: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
193    ; GFX9-NEXT: [[COPY1:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr2_vgpr3
194    ; GFX9-NEXT: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<2 x s32>)
195    ; GFX9-NEXT: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<2 x s32>)
196    ; GFX9-NEXT: [[SEXT_INREG:%[0-9]+]]:_(s32) = G_SEXT_INREG [[UV]], 16
197    ; GFX9-NEXT: [[SEXT_INREG1:%[0-9]+]]:_(s32) = G_SEXT_INREG [[UV2]], 16
198    ; GFX9-NEXT: [[MUL:%[0-9]+]]:_(s32) = G_MUL [[SEXT_INREG]], [[SEXT_INREG1]]
199    ; GFX9-NEXT: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
200    ; GFX9-NEXT: [[ASHR:%[0-9]+]]:_(s32) = G_ASHR [[MUL]], [[C]](s32)
201    ; GFX9-NEXT: [[SEXT_INREG2:%[0-9]+]]:_(s32) = G_SEXT_INREG [[UV1]], 16
202    ; GFX9-NEXT: [[SEXT_INREG3:%[0-9]+]]:_(s32) = G_SEXT_INREG [[UV3]], 16
203    ; GFX9-NEXT: [[MUL1:%[0-9]+]]:_(s32) = G_MUL [[SEXT_INREG2]], [[SEXT_INREG3]]
204    ; GFX9-NEXT: [[ASHR1:%[0-9]+]]:_(s32) = G_ASHR [[MUL1]], [[C]](s32)
205    ; GFX9-NEXT: [[SEXT_INREG4:%[0-9]+]]:_(s32) = G_SEXT_INREG [[ASHR]], 16
206    ; GFX9-NEXT: [[SEXT_INREG5:%[0-9]+]]:_(s32) = G_SEXT_INREG [[ASHR1]], 16
207    ; GFX9-NEXT: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[SEXT_INREG4]](s32), [[SEXT_INREG5]](s32)
208    ; GFX9-NEXT: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
209    %0:_(<2 x s32>) = COPY $vgpr0_vgpr1
210    %1:_(<2 x s32>) = COPY $vgpr2_vgpr3
211    %2:_(<2 x s16>) = G_TRUNC %0
212    %3:_(<2 x s16>) = G_TRUNC %1
213    %4:_(<2 x s16>) = G_SMULH %2, %3
214    %5:_(<2 x s32>) = G_SEXT %4
215    $vgpr0_vgpr1 = COPY %5
216...
217
218---
219name: test_smulh_v2s8
220body: |
221  bb.0:
222    liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3
223    ; GFX8-LABEL: name: test_smulh_v2s8
224    ; GFX8: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3
225    ; GFX8-NEXT: {{  $}}
226    ; GFX8-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
227    ; GFX8-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
228    ; GFX8-NEXT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
229    ; GFX8-NEXT: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
230    ; GFX8-NEXT: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
231    ; GFX8-NEXT: [[C:%[0-9]+]]:_(s16) = G_CONSTANT i16 8
232    ; GFX8-NEXT: [[SHL:%[0-9]+]]:_(s16) = G_SHL [[TRUNC]], [[C]](s16)
233    ; GFX8-NEXT: [[ASHR:%[0-9]+]]:_(s16) = G_ASHR [[SHL]], [[C]](s16)
234    ; GFX8-NEXT: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY2]](s32)
235    ; GFX8-NEXT: [[SHL1:%[0-9]+]]:_(s16) = G_SHL [[TRUNC1]], [[C]](s16)
236    ; GFX8-NEXT: [[ASHR1:%[0-9]+]]:_(s16) = G_ASHR [[SHL1]], [[C]](s16)
237    ; GFX8-NEXT: [[MUL:%[0-9]+]]:_(s16) = G_MUL [[ASHR]], [[ASHR1]]
238    ; GFX8-NEXT: [[ASHR2:%[0-9]+]]:_(s16) = G_ASHR [[MUL]], [[C]](s16)
239    ; GFX8-NEXT: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32)
240    ; GFX8-NEXT: [[SHL2:%[0-9]+]]:_(s16) = G_SHL [[TRUNC2]], [[C]](s16)
241    ; GFX8-NEXT: [[ASHR3:%[0-9]+]]:_(s16) = G_ASHR [[SHL2]], [[C]](s16)
242    ; GFX8-NEXT: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[COPY3]](s32)
243    ; GFX8-NEXT: [[SHL3:%[0-9]+]]:_(s16) = G_SHL [[TRUNC3]], [[C]](s16)
244    ; GFX8-NEXT: [[ASHR4:%[0-9]+]]:_(s16) = G_ASHR [[SHL3]], [[C]](s16)
245    ; GFX8-NEXT: [[MUL1:%[0-9]+]]:_(s16) = G_MUL [[ASHR3]], [[ASHR4]]
246    ; GFX8-NEXT: [[ASHR5:%[0-9]+]]:_(s16) = G_ASHR [[MUL1]], [[C]](s16)
247    ; GFX8-NEXT: [[C1:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
248    ; GFX8-NEXT: [[AND:%[0-9]+]]:_(s16) = G_AND [[ASHR2]], [[C1]]
249    ; GFX8-NEXT: [[AND1:%[0-9]+]]:_(s16) = G_AND [[ASHR5]], [[C1]]
250    ; GFX8-NEXT: [[SHL4:%[0-9]+]]:_(s16) = G_SHL [[AND1]], [[C]](s16)
251    ; GFX8-NEXT: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[SHL4]]
252    ; GFX8-NEXT: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[OR]](s16)
253    ; GFX8-NEXT: $vgpr0 = COPY [[ANYEXT]](s32)
254    ;
255    ; GFX9-LABEL: name: test_smulh_v2s8
256    ; GFX9: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3
257    ; GFX9-NEXT: {{  $}}
258    ; GFX9-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
259    ; GFX9-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
260    ; GFX9-NEXT: [[COPY2:%[0-9]+]]:_(s32) = COPY $vgpr2
261    ; GFX9-NEXT: [[COPY3:%[0-9]+]]:_(s32) = COPY $vgpr3
262    ; GFX9-NEXT: [[SEXT_INREG:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY]], 8
263    ; GFX9-NEXT: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[SEXT_INREG]](s32)
264    ; GFX9-NEXT: [[SEXT_INREG1:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY1]], 8
265    ; GFX9-NEXT: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[SEXT_INREG1]](s32)
266    ; GFX9-NEXT: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR [[TRUNC]](s16), [[TRUNC1]](s16)
267    ; GFX9-NEXT: [[SEXT_INREG2:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY2]], 8
268    ; GFX9-NEXT: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[SEXT_INREG2]](s32)
269    ; GFX9-NEXT: [[SEXT_INREG3:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY3]], 8
270    ; GFX9-NEXT: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[SEXT_INREG3]](s32)
271    ; GFX9-NEXT: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR [[TRUNC2]](s16), [[TRUNC3]](s16)
272    ; GFX9-NEXT: [[MUL:%[0-9]+]]:_(<2 x s16>) = G_MUL [[BUILD_VECTOR]], [[BUILD_VECTOR1]]
273    ; GFX9-NEXT: [[C:%[0-9]+]]:_(s16) = G_CONSTANT i16 8
274    ; GFX9-NEXT: [[BUILD_VECTOR2:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR [[C]](s16), [[C]](s16)
275    ; GFX9-NEXT: [[ASHR:%[0-9]+]]:_(<2 x s16>) = G_ASHR [[MUL]], [[BUILD_VECTOR2]](<2 x s16>)
276    ; GFX9-NEXT: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[ASHR]](<2 x s16>)
277    ; GFX9-NEXT: [[TRUNC4:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
278    ; GFX9-NEXT: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
279    ; GFX9-NEXT: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C1]](s32)
280    ; GFX9-NEXT: [[TRUNC5:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
281    ; GFX9-NEXT: [[C2:%[0-9]+]]:_(s16) = G_CONSTANT i16 255
282    ; GFX9-NEXT: [[AND:%[0-9]+]]:_(s16) = G_AND [[TRUNC4]], [[C2]]
283    ; GFX9-NEXT: [[AND1:%[0-9]+]]:_(s16) = G_AND [[TRUNC5]], [[C2]]
284    ; GFX9-NEXT: [[SHL:%[0-9]+]]:_(s16) = G_SHL [[AND1]], [[C]](s16)
285    ; GFX9-NEXT: [[OR:%[0-9]+]]:_(s16) = G_OR [[AND]], [[SHL]]
286    ; GFX9-NEXT: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[OR]](s16)
287    ; GFX9-NEXT: $vgpr0 = COPY [[ANYEXT]](s32)
288    %0:_(s32) = COPY $vgpr0
289    %1:_(s32) = COPY $vgpr1
290    %2:_(s32) = COPY $vgpr2
291    %3:_(s32) = COPY $vgpr3
292    %5:_(s8) = G_TRUNC %0
293    %6:_(s8) = G_TRUNC %1
294    %7:_(s8) = G_TRUNC %2
295    %8:_(s8) = G_TRUNC %3
296    %11:_(<2 x s8>) = G_BUILD_VECTOR %5, %6
297    %12:_(<2 x s8>) = G_BUILD_VECTOR %7, %8
298    %13:_(<2 x s8>) = G_SMULH %11, %12
299    %14:_(s8), %15:_(s8) = G_UNMERGE_VALUES %13
300    %17:_(s16) = G_MERGE_VALUES %14, %15
301    %18:_(s32) = G_ANYEXT %17
302    $vgpr0 = COPY %18
303
304...
305
306---
307name: test_smulh_v4s8
308body: |
309  bb.0:
310    liveins: $vgpr0, $vgpr1
311    ; GFX8-LABEL: name: test_smulh_v4s8
312    ; GFX8: liveins: $vgpr0, $vgpr1
313    ; GFX8-NEXT: {{  $}}
314    ; GFX8-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
315    ; GFX8-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
316    ; GFX8-NEXT: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
317    ; GFX8-NEXT: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C]](s32)
318    ; GFX8-NEXT: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
319    ; GFX8-NEXT: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C1]](s32)
320    ; GFX8-NEXT: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
321    ; GFX8-NEXT: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C2]](s32)
322    ; GFX8-NEXT: [[LSHR3:%[0-9]+]]:_(s32) = G_LSHR [[COPY1]], [[C]](s32)
323    ; GFX8-NEXT: [[LSHR4:%[0-9]+]]:_(s32) = G_LSHR [[COPY1]], [[C1]](s32)
324    ; GFX8-NEXT: [[LSHR5:%[0-9]+]]:_(s32) = G_LSHR [[COPY1]], [[C2]](s32)
325    ; GFX8-NEXT: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
326    ; GFX8-NEXT: [[C3:%[0-9]+]]:_(s16) = G_CONSTANT i16 8
327    ; GFX8-NEXT: [[SHL:%[0-9]+]]:_(s16) = G_SHL [[TRUNC]], [[C3]](s16)
328    ; GFX8-NEXT: [[ASHR:%[0-9]+]]:_(s16) = G_ASHR [[SHL]], [[C3]](s16)
329    ; GFX8-NEXT: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32)
330    ; GFX8-NEXT: [[SHL1:%[0-9]+]]:_(s16) = G_SHL [[TRUNC1]], [[C3]](s16)
331    ; GFX8-NEXT: [[ASHR1:%[0-9]+]]:_(s16) = G_ASHR [[SHL1]], [[C3]](s16)
332    ; GFX8-NEXT: [[MUL:%[0-9]+]]:_(s16) = G_MUL [[ASHR]], [[ASHR1]]
333    ; GFX8-NEXT: [[ASHR2:%[0-9]+]]:_(s16) = G_ASHR [[MUL]], [[C3]](s16)
334    ; GFX8-NEXT: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
335    ; GFX8-NEXT: [[SHL2:%[0-9]+]]:_(s16) = G_SHL [[TRUNC2]], [[C3]](s16)
336    ; GFX8-NEXT: [[ASHR3:%[0-9]+]]:_(s16) = G_ASHR [[SHL2]], [[C3]](s16)
337    ; GFX8-NEXT: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR3]](s32)
338    ; GFX8-NEXT: [[SHL3:%[0-9]+]]:_(s16) = G_SHL [[TRUNC3]], [[C3]](s16)
339    ; GFX8-NEXT: [[ASHR4:%[0-9]+]]:_(s16) = G_ASHR [[SHL3]], [[C3]](s16)
340    ; GFX8-NEXT: [[MUL1:%[0-9]+]]:_(s16) = G_MUL [[ASHR3]], [[ASHR4]]
341    ; GFX8-NEXT: [[ASHR5:%[0-9]+]]:_(s16) = G_ASHR [[MUL1]], [[C3]](s16)
342    ; GFX8-NEXT: [[TRUNC4:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR1]](s32)
343    ; GFX8-NEXT: [[SHL4:%[0-9]+]]:_(s16) = G_SHL [[TRUNC4]], [[C3]](s16)
344    ; GFX8-NEXT: [[ASHR6:%[0-9]+]]:_(s16) = G_ASHR [[SHL4]], [[C3]](s16)
345    ; GFX8-NEXT: [[TRUNC5:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR4]](s32)
346    ; GFX8-NEXT: [[SHL5:%[0-9]+]]:_(s16) = G_SHL [[TRUNC5]], [[C3]](s16)
347    ; GFX8-NEXT: [[ASHR7:%[0-9]+]]:_(s16) = G_ASHR [[SHL5]], [[C3]](s16)
348    ; GFX8-NEXT: [[MUL2:%[0-9]+]]:_(s16) = G_MUL [[ASHR6]], [[ASHR7]]
349    ; GFX8-NEXT: [[ASHR8:%[0-9]+]]:_(s16) = G_ASHR [[MUL2]], [[C3]](s16)
350    ; GFX8-NEXT: [[TRUNC6:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR2]](s32)
351    ; GFX8-NEXT: [[SHL6:%[0-9]+]]:_(s16) = G_SHL [[TRUNC6]], [[C3]](s16)
352    ; GFX8-NEXT: [[ASHR9:%[0-9]+]]:_(s16) = G_ASHR [[SHL6]], [[C3]](s16)
353    ; GFX8-NEXT: [[TRUNC7:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR5]](s32)
354    ; GFX8-NEXT: [[SHL7:%[0-9]+]]:_(s16) = G_SHL [[TRUNC7]], [[C3]](s16)
355    ; GFX8-NEXT: [[ASHR10:%[0-9]+]]:_(s16) = G_ASHR [[SHL7]], [[C3]](s16)
356    ; GFX8-NEXT: [[MUL3:%[0-9]+]]:_(s16) = G_MUL [[ASHR9]], [[ASHR10]]
357    ; GFX8-NEXT: [[ASHR11:%[0-9]+]]:_(s16) = G_ASHR [[MUL3]], [[C3]](s16)
358    ; GFX8-NEXT: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[ASHR2]](s16)
359    ; GFX8-NEXT: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
360    ; GFX8-NEXT: [[AND:%[0-9]+]]:_(s32) = G_AND [[ANYEXT]], [[C4]]
361    ; GFX8-NEXT: [[ANYEXT1:%[0-9]+]]:_(s32) = G_ANYEXT [[ASHR5]](s16)
362    ; GFX8-NEXT: [[AND1:%[0-9]+]]:_(s32) = G_AND [[ANYEXT1]], [[C4]]
363    ; GFX8-NEXT: [[SHL8:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C]](s32)
364    ; GFX8-NEXT: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL8]]
365    ; GFX8-NEXT: [[ANYEXT2:%[0-9]+]]:_(s32) = G_ANYEXT [[ASHR8]](s16)
366    ; GFX8-NEXT: [[AND2:%[0-9]+]]:_(s32) = G_AND [[ANYEXT2]], [[C4]]
367    ; GFX8-NEXT: [[SHL9:%[0-9]+]]:_(s32) = G_SHL [[AND2]], [[C1]](s32)
368    ; GFX8-NEXT: [[OR1:%[0-9]+]]:_(s32) = G_OR [[OR]], [[SHL9]]
369    ; GFX8-NEXT: [[ANYEXT3:%[0-9]+]]:_(s32) = G_ANYEXT [[ASHR11]](s16)
370    ; GFX8-NEXT: [[AND3:%[0-9]+]]:_(s32) = G_AND [[ANYEXT3]], [[C4]]
371    ; GFX8-NEXT: [[SHL10:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C2]](s32)
372    ; GFX8-NEXT: [[OR2:%[0-9]+]]:_(s32) = G_OR [[OR1]], [[SHL10]]
373    ; GFX8-NEXT: $vgpr0 = COPY [[OR2]](s32)
374    ;
375    ; GFX9-LABEL: name: test_smulh_v4s8
376    ; GFX9: liveins: $vgpr0, $vgpr1
377    ; GFX9-NEXT: {{  $}}
378    ; GFX9-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
379    ; GFX9-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
380    ; GFX9-NEXT: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
381    ; GFX9-NEXT: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C]](s32)
382    ; GFX9-NEXT: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
383    ; GFX9-NEXT: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C1]](s32)
384    ; GFX9-NEXT: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 24
385    ; GFX9-NEXT: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[COPY]], [[C2]](s32)
386    ; GFX9-NEXT: [[LSHR3:%[0-9]+]]:_(s32) = G_LSHR [[COPY1]], [[C]](s32)
387    ; GFX9-NEXT: [[LSHR4:%[0-9]+]]:_(s32) = G_LSHR [[COPY1]], [[C1]](s32)
388    ; GFX9-NEXT: [[LSHR5:%[0-9]+]]:_(s32) = G_LSHR [[COPY1]], [[C2]](s32)
389    ; GFX9-NEXT: [[SEXT_INREG:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY]], 8
390    ; GFX9-NEXT: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[SEXT_INREG]](s32)
391    ; GFX9-NEXT: [[SEXT_INREG1:%[0-9]+]]:_(s32) = G_SEXT_INREG [[LSHR]], 8
392    ; GFX9-NEXT: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[SEXT_INREG1]](s32)
393    ; GFX9-NEXT: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR [[TRUNC]](s16), [[TRUNC1]](s16)
394    ; GFX9-NEXT: [[SEXT_INREG2:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY1]], 8
395    ; GFX9-NEXT: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[SEXT_INREG2]](s32)
396    ; GFX9-NEXT: [[SEXT_INREG3:%[0-9]+]]:_(s32) = G_SEXT_INREG [[LSHR3]], 8
397    ; GFX9-NEXT: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[SEXT_INREG3]](s32)
398    ; GFX9-NEXT: [[BUILD_VECTOR1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR [[TRUNC2]](s16), [[TRUNC3]](s16)
399    ; GFX9-NEXT: [[MUL:%[0-9]+]]:_(<2 x s16>) = G_MUL [[BUILD_VECTOR]], [[BUILD_VECTOR1]]
400    ; GFX9-NEXT: [[C3:%[0-9]+]]:_(s16) = G_CONSTANT i16 8
401    ; GFX9-NEXT: [[BUILD_VECTOR2:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR [[C3]](s16), [[C3]](s16)
402    ; GFX9-NEXT: [[ASHR:%[0-9]+]]:_(<2 x s16>) = G_ASHR [[MUL]], [[BUILD_VECTOR2]](<2 x s16>)
403    ; GFX9-NEXT: [[SEXT_INREG4:%[0-9]+]]:_(s32) = G_SEXT_INREG [[LSHR1]], 8
404    ; GFX9-NEXT: [[TRUNC4:%[0-9]+]]:_(s16) = G_TRUNC [[SEXT_INREG4]](s32)
405    ; GFX9-NEXT: [[SEXT_INREG5:%[0-9]+]]:_(s32) = G_SEXT_INREG [[LSHR2]], 8
406    ; GFX9-NEXT: [[TRUNC5:%[0-9]+]]:_(s16) = G_TRUNC [[SEXT_INREG5]](s32)
407    ; GFX9-NEXT: [[BUILD_VECTOR3:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR [[TRUNC4]](s16), [[TRUNC5]](s16)
408    ; GFX9-NEXT: [[SEXT_INREG6:%[0-9]+]]:_(s32) = G_SEXT_INREG [[LSHR4]], 8
409    ; GFX9-NEXT: [[TRUNC6:%[0-9]+]]:_(s16) = G_TRUNC [[SEXT_INREG6]](s32)
410    ; GFX9-NEXT: [[SEXT_INREG7:%[0-9]+]]:_(s32) = G_SEXT_INREG [[LSHR5]], 8
411    ; GFX9-NEXT: [[TRUNC7:%[0-9]+]]:_(s16) = G_TRUNC [[SEXT_INREG7]](s32)
412    ; GFX9-NEXT: [[BUILD_VECTOR4:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR [[TRUNC6]](s16), [[TRUNC7]](s16)
413    ; GFX9-NEXT: [[MUL1:%[0-9]+]]:_(<2 x s16>) = G_MUL [[BUILD_VECTOR3]], [[BUILD_VECTOR4]]
414    ; GFX9-NEXT: [[BUILD_VECTOR5:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR [[C3]](s16), [[C3]](s16)
415    ; GFX9-NEXT: [[ASHR1:%[0-9]+]]:_(<2 x s16>) = G_ASHR [[MUL1]], [[BUILD_VECTOR5]](<2 x s16>)
416    ; GFX9-NEXT: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[ASHR]](<2 x s16>)
417    ; GFX9-NEXT: [[LSHR6:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C1]](s32)
418    ; GFX9-NEXT: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[ASHR1]](<2 x s16>)
419    ; GFX9-NEXT: [[LSHR7:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C1]](s32)
420    ; GFX9-NEXT: [[C4:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
421    ; GFX9-NEXT: [[AND:%[0-9]+]]:_(s32) = G_AND [[BITCAST]], [[C4]]
422    ; GFX9-NEXT: [[AND1:%[0-9]+]]:_(s32) = G_AND [[LSHR6]], [[C4]]
423    ; GFX9-NEXT: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C]](s32)
424    ; GFX9-NEXT: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
425    ; GFX9-NEXT: [[AND2:%[0-9]+]]:_(s32) = G_AND [[BITCAST1]], [[C4]]
426    ; GFX9-NEXT: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND2]], [[C1]](s32)
427    ; GFX9-NEXT: [[OR1:%[0-9]+]]:_(s32) = G_OR [[OR]], [[SHL1]]
428    ; GFX9-NEXT: [[AND3:%[0-9]+]]:_(s32) = G_AND [[LSHR7]], [[C4]]
429    ; GFX9-NEXT: [[SHL2:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C2]](s32)
430    ; GFX9-NEXT: [[OR2:%[0-9]+]]:_(s32) = G_OR [[OR1]], [[SHL2]]
431    ; GFX9-NEXT: $vgpr0 = COPY [[OR2]](s32)
432    %0:_(s32) = COPY $vgpr0
433    %1:_(s32) = COPY $vgpr1
434    %2:_(s8), %3:_(s8), %4:_(s8), %5:_(s8) = G_UNMERGE_VALUES %0
435    %6:_(s8), %7:_(s8), %8:_(s8), %9:_(s8) = G_UNMERGE_VALUES %1
436    %10:_(<4 x s8>) = G_BUILD_VECTOR %2:_(s8), %3:_(s8), %4:_(s8), %5:_(s8)
437    %11:_(<4 x s8>) = G_BUILD_VECTOR %6:_(s8), %7:_(s8), %8:_(s8), %9:_(s8)
438    %12:_(<4 x s8>) = G_SMULH %10:_, %11:_
439    %13:_(s8), %14:_(s8), %15:_(s8), %16:_(s8) = G_UNMERGE_VALUES %12:_(<4 x s8>)
440    %17:_(s32) = G_MERGE_VALUES %13, %14, %15, %16
441    $vgpr0 = COPY %17
442...
443