xref: /llvm-project/llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-store-global.s96.mir (revision 9e9907f1cfa424366fba58d9520f9305b537cec9)
1# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
2# RUN: llc -mtriple=amdgcn -mcpu=hawaii -run-pass=instruction-select -verify-machineinstrs  -global-isel-abort=0 -o - %s | FileCheck -check-prefix=GFX7 %s
3# RUN: llc -mtriple=amdgcn -mcpu=hawaii -mattr=+flat-for-global -run-pass=instruction-select -verify-machineinstrs  -global-isel-abort=0 -o - %s | FileCheck -check-prefix=GFX7-FLAT %s
4# RUN: llc -mtriple=amdgcn -mcpu=fiji -run-pass=instruction-select -verify-machineinstrs -global-isel-abort=0 -o - %s | FileCheck -check-prefix=GFX8 %s
5# RUN: llc -mtriple=amdgcn -mcpu=gfx900 -run-pass=instruction-select -verify-machineinstrs -global-isel-abort=0 -o - %s | FileCheck -check-prefix=GFX9 %s
6# RUN: llc -mtriple=amdgcn -mcpu=gfx1010 -run-pass=instruction-select -verify-machineinstrs -global-isel-abort=0 -o - %s | FileCheck -check-prefix=GFX10 %s
7# RUN: llc -mtriple=amdgcn -mcpu=gfx1100 -run-pass=instruction-select -verify-machineinstrs -global-isel-abort=0 -o - %s | FileCheck -check-prefix=GFX10 %s
8# RUN: llc -mtriple=amdgcn -mcpu=gfx1200 -run-pass=instruction-select -verify-machineinstrs -global-isel-abort=0 -o - %s | FileCheck -check-prefix=GFX10 %s
9
10# 96-bit load/store is only legal for SI, so split the test out.
11
12---
13
14name: store_global_v3s32
15legalized:       true
16tracksRegLiveness: true
17regBankSelected: true
18
19body: |
20  bb.0:
21    liveins:  $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4
22
23    ; GFX7-LABEL: name: store_global_v3s32
24    ; GFX7: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4
25    ; GFX7-NEXT: {{  $}}
26    ; GFX7-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
27    ; GFX7-NEXT: [[COPY1:%[0-9]+]]:vreg_96 = COPY $vgpr2_vgpr3_vgpr4
28    ; GFX7-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 0
29    ; GFX7-NEXT: [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 61440
30    ; GFX7-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[S_MOV_B32_]], %subreg.sub0, [[S_MOV_B32_1]], %subreg.sub1
31    ; GFX7-NEXT: [[S_MOV_B64_:%[0-9]+]]:sreg_64 = S_MOV_B64 0
32    ; GFX7-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[S_MOV_B64_]], %subreg.sub0_sub1, [[REG_SEQUENCE]], %subreg.sub2_sub3
33    ; GFX7-NEXT: BUFFER_STORE_DWORDX3_ADDR64 [[COPY1]], [[COPY]], [[REG_SEQUENCE1]], 0, 0, 0, 0, implicit $exec :: (store (<3 x s32>), align 16, addrspace 1)
34    ; GFX7-FLAT-LABEL: name: store_global_v3s32
35    ; GFX7-FLAT: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4
36    ; GFX7-FLAT-NEXT: {{  $}}
37    ; GFX7-FLAT-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
38    ; GFX7-FLAT-NEXT: [[COPY1:%[0-9]+]]:vreg_96 = COPY $vgpr2_vgpr3_vgpr4
39    ; GFX7-FLAT-NEXT: FLAT_STORE_DWORDX3 [[COPY]], [[COPY1]], 0, 0, implicit $exec, implicit $flat_scr :: (store (<3 x s32>), align 16, addrspace 1)
40    ; GFX8-LABEL: name: store_global_v3s32
41    ; GFX8: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4
42    ; GFX8-NEXT: {{  $}}
43    ; GFX8-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
44    ; GFX8-NEXT: [[COPY1:%[0-9]+]]:vreg_96 = COPY $vgpr2_vgpr3_vgpr4
45    ; GFX8-NEXT: FLAT_STORE_DWORDX3 [[COPY]], [[COPY1]], 0, 0, implicit $exec, implicit $flat_scr :: (store (<3 x s32>), align 16, addrspace 1)
46    ; GFX9-LABEL: name: store_global_v3s32
47    ; GFX9: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4
48    ; GFX9-NEXT: {{  $}}
49    ; GFX9-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
50    ; GFX9-NEXT: [[COPY1:%[0-9]+]]:vreg_96 = COPY $vgpr2_vgpr3_vgpr4
51    ; GFX9-NEXT: GLOBAL_STORE_DWORDX3 [[COPY]], [[COPY1]], 0, 0, implicit $exec :: (store (<3 x s32>), align 16, addrspace 1)
52    ; GFX10-LABEL: name: store_global_v3s32
53    ; GFX10: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4
54    ; GFX10-NEXT: {{  $}}
55    ; GFX10-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
56    ; GFX10-NEXT: [[COPY1:%[0-9]+]]:vreg_96 = COPY $vgpr2_vgpr3_vgpr4
57    ; GFX10-NEXT: GLOBAL_STORE_DWORDX3 [[COPY]], [[COPY1]], 0, 0, implicit $exec :: (store (<3 x s32>), align 16, addrspace 1)
58    %0:vgpr(p1) = COPY $vgpr0_vgpr1
59    %1:vgpr(<3 x s32>) = COPY $vgpr2_vgpr3_vgpr4
60    G_STORE %1, %0 :: (store (<3 x s32>), align 16, addrspace 1)
61
62...
63
64---
65
66name: store_global_s96
67legalized:       true
68tracksRegLiveness: true
69regBankSelected: true
70
71body: |
72  bb.0:
73    liveins:  $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4
74
75    ; GFX7-LABEL: name: store_global_s96
76    ; GFX7: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4
77    ; GFX7-NEXT: {{  $}}
78    ; GFX7-NEXT: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
79    ; GFX7-NEXT: [[COPY1:%[0-9]+]]:vgpr(s96) = COPY $vgpr2_vgpr3_vgpr4
80    ; GFX7-NEXT: G_STORE [[COPY1]](s96), [[COPY]](p1) :: (store (s96), align 16, addrspace 1)
81    ; GFX7-FLAT-LABEL: name: store_global_s96
82    ; GFX7-FLAT: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4
83    ; GFX7-FLAT-NEXT: {{  $}}
84    ; GFX7-FLAT-NEXT: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
85    ; GFX7-FLAT-NEXT: [[COPY1:%[0-9]+]]:vgpr(s96) = COPY $vgpr2_vgpr3_vgpr4
86    ; GFX7-FLAT-NEXT: G_STORE [[COPY1]](s96), [[COPY]](p1) :: (store (s96), align 16, addrspace 1)
87    ; GFX8-LABEL: name: store_global_s96
88    ; GFX8: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4
89    ; GFX8-NEXT: {{  $}}
90    ; GFX8-NEXT: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
91    ; GFX8-NEXT: [[COPY1:%[0-9]+]]:vgpr(s96) = COPY $vgpr2_vgpr3_vgpr4
92    ; GFX8-NEXT: G_STORE [[COPY1]](s96), [[COPY]](p1) :: (store (s96), align 16, addrspace 1)
93    ; GFX9-LABEL: name: store_global_s96
94    ; GFX9: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4
95    ; GFX9-NEXT: {{  $}}
96    ; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
97    ; GFX9-NEXT: [[COPY1:%[0-9]+]]:vgpr(s96) = COPY $vgpr2_vgpr3_vgpr4
98    ; GFX9-NEXT: G_STORE [[COPY1]](s96), [[COPY]](p1) :: (store (s96), align 16, addrspace 1)
99    ; GFX10-LABEL: name: store_global_s96
100    ; GFX10: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4
101    ; GFX10-NEXT: {{  $}}
102    ; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
103    ; GFX10-NEXT: [[COPY1:%[0-9]+]]:vgpr(s96) = COPY $vgpr2_vgpr3_vgpr4
104    ; GFX10-NEXT: G_STORE [[COPY1]](s96), [[COPY]](p1) :: (store (s96), align 16, addrspace 1)
105    %0:vgpr(p1) = COPY $vgpr0_vgpr1
106    %1:vgpr(s96) = COPY $vgpr2_vgpr3_vgpr4
107    G_STORE %1, %0 :: (store (s96), align 16, addrspace 1)
108
109...
110
111---
112
113name: store_global_v6s16
114legalized:       true
115tracksRegLiveness: true
116regBankSelected: true
117
118body: |
119  bb.0:
120    liveins:  $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4
121
122    ; GFX7-LABEL: name: store_global_v6s16
123    ; GFX7: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4
124    ; GFX7-NEXT: {{  $}}
125    ; GFX7-NEXT: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
126    ; GFX7-NEXT: [[COPY1:%[0-9]+]]:vgpr(<6 x s16>) = COPY $vgpr2_vgpr3_vgpr4
127    ; GFX7-NEXT: G_STORE [[COPY1]](<6 x s16>), [[COPY]](p1) :: (store (<6 x s16>), align 16, addrspace 1)
128    ; GFX7-FLAT-LABEL: name: store_global_v6s16
129    ; GFX7-FLAT: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4
130    ; GFX7-FLAT-NEXT: {{  $}}
131    ; GFX7-FLAT-NEXT: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
132    ; GFX7-FLAT-NEXT: [[COPY1:%[0-9]+]]:vgpr(<6 x s16>) = COPY $vgpr2_vgpr3_vgpr4
133    ; GFX7-FLAT-NEXT: G_STORE [[COPY1]](<6 x s16>), [[COPY]](p1) :: (store (<6 x s16>), align 16, addrspace 1)
134    ; GFX8-LABEL: name: store_global_v6s16
135    ; GFX8: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4
136    ; GFX8-NEXT: {{  $}}
137    ; GFX8-NEXT: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
138    ; GFX8-NEXT: [[COPY1:%[0-9]+]]:vgpr(<6 x s16>) = COPY $vgpr2_vgpr3_vgpr4
139    ; GFX8-NEXT: G_STORE [[COPY1]](<6 x s16>), [[COPY]](p1) :: (store (<6 x s16>), align 16, addrspace 1)
140    ; GFX9-LABEL: name: store_global_v6s16
141    ; GFX9: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4
142    ; GFX9-NEXT: {{  $}}
143    ; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
144    ; GFX9-NEXT: [[COPY1:%[0-9]+]]:vgpr(<6 x s16>) = COPY $vgpr2_vgpr3_vgpr4
145    ; GFX9-NEXT: G_STORE [[COPY1]](<6 x s16>), [[COPY]](p1) :: (store (<6 x s16>), align 16, addrspace 1)
146    ; GFX10-LABEL: name: store_global_v6s16
147    ; GFX10: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4
148    ; GFX10-NEXT: {{  $}}
149    ; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
150    ; GFX10-NEXT: [[COPY1:%[0-9]+]]:vgpr(<6 x s16>) = COPY $vgpr2_vgpr3_vgpr4
151    ; GFX10-NEXT: G_STORE [[COPY1]](<6 x s16>), [[COPY]](p1) :: (store (<6 x s16>), align 16, addrspace 1)
152    %0:vgpr(p1) = COPY $vgpr0_vgpr1
153    %1:vgpr(<6 x s16>) = COPY $vgpr2_vgpr3_vgpr4
154    G_STORE %1, %0 :: (store (<6 x s16>), align 16, addrspace 1)
155
156...
157