xref: /llvm-project/llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-load-global.s96.mir (revision ed25d1a82e51595b6fdfc3d187352ed823f80491)
1# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
2# RUN: llc -amdgpu-global-isel-new-legality -mtriple=amdgcn -mcpu=hawaii -run-pass=instruction-select -verify-machineinstrs  -global-isel-abort=0 -o - %s | FileCheck -check-prefix=GFX7 %s
3# RUN: llc -amdgpu-global-isel-new-legality -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 -amdgpu-global-isel-new-legality -mtriple=amdgcn -mcpu=fiji -run-pass=instruction-select -verify-machineinstrs -global-isel-abort=0 -o - %s | FileCheck -check-prefix=GFX8 %s
5# RUN: llc -amdgpu-global-isel-new-legality -mtriple=amdgcn -mcpu=gfx900 -run-pass=instruction-select -verify-machineinstrs -global-isel-abort=0 -o - %s | FileCheck -check-prefix=GFX9 %s
6# RUN: llc -amdgpu-global-isel-new-legality -mtriple=amdgcn -mcpu=gfx1010 -run-pass=instruction-select -verify-machineinstrs -global-isel-abort=0 -o - %s | FileCheck -check-prefix=GFX10 %s
7# RUN: llc -amdgpu-global-isel-new-legality -mtriple=amdgcn -mcpu=gfx1100 -run-pass=instruction-select -verify-machineinstrs -global-isel-abort=0 -o - %s | FileCheck -check-prefix=GFX10 %s
8# RUN: llc -amdgpu-global-isel-new-legality -mtriple=amdgcn -mcpu=gfx1200 -run-pass=instruction-select -verify-machineinstrs -global-isel-abort=0 -o - %s | FileCheck -check-prefix=GFX10 %s
9
10---
11
12name: load_global_v3s32
13legalized:       true
14regBankSelected: true
15tracksRegLiveness: true
16
17body: |
18  bb.0:
19    liveins:  $vgpr0_vgpr1
20
21    ; GFX7-LABEL: name: load_global_v3s32
22    ; GFX7: liveins: $vgpr0_vgpr1
23    ; GFX7-NEXT: {{  $}}
24    ; GFX7-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
25    ; GFX7-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 0
26    ; GFX7-NEXT: [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 61440
27    ; GFX7-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[S_MOV_B32_]], %subreg.sub0, [[S_MOV_B32_1]], %subreg.sub1
28    ; GFX7-NEXT: [[S_MOV_B64_:%[0-9]+]]:sreg_64 = S_MOV_B64 0
29    ; GFX7-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[S_MOV_B64_]], %subreg.sub0_sub1, [[REG_SEQUENCE]], %subreg.sub2_sub3
30    ; GFX7-NEXT: [[BUFFER_LOAD_DWORDX3_ADDR64_:%[0-9]+]]:vreg_96 = BUFFER_LOAD_DWORDX3_ADDR64 [[COPY]], [[REG_SEQUENCE1]], 0, 0, 0, 0, implicit $exec :: (load (<3 x s32>), align 4, addrspace 1)
31    ; GFX7-NEXT: $vgpr0_vgpr1_vgpr2 = COPY [[BUFFER_LOAD_DWORDX3_ADDR64_]]
32    ;
33    ; GFX7-FLAT-LABEL: name: load_global_v3s32
34    ; GFX7-FLAT: liveins: $vgpr0_vgpr1
35    ; GFX7-FLAT-NEXT: {{  $}}
36    ; GFX7-FLAT-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
37    ; GFX7-FLAT-NEXT: [[FLAT_LOAD_DWORDX3_:%[0-9]+]]:vreg_96 = FLAT_LOAD_DWORDX3 [[COPY]], 0, 0, implicit $exec, implicit $flat_scr :: (load (<3 x s32>), align 4, addrspace 1)
38    ; GFX7-FLAT-NEXT: $vgpr0_vgpr1_vgpr2 = COPY [[FLAT_LOAD_DWORDX3_]]
39    ;
40    ; GFX8-LABEL: name: load_global_v3s32
41    ; GFX8: liveins: $vgpr0_vgpr1
42    ; GFX8-NEXT: {{  $}}
43    ; GFX8-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
44    ; GFX8-NEXT: [[FLAT_LOAD_DWORDX3_:%[0-9]+]]:vreg_96 = FLAT_LOAD_DWORDX3 [[COPY]], 0, 0, implicit $exec, implicit $flat_scr :: (load (<3 x s32>), align 4, addrspace 1)
45    ; GFX8-NEXT: $vgpr0_vgpr1_vgpr2 = COPY [[FLAT_LOAD_DWORDX3_]]
46    ;
47    ; GFX9-LABEL: name: load_global_v3s32
48    ; GFX9: liveins: $vgpr0_vgpr1
49    ; GFX9-NEXT: {{  $}}
50    ; GFX9-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
51    ; GFX9-NEXT: [[GLOBAL_LOAD_DWORDX3_:%[0-9]+]]:vreg_96 = GLOBAL_LOAD_DWORDX3 [[COPY]], 0, 0, implicit $exec :: (load (<3 x s32>), align 4, addrspace 1)
52    ; GFX9-NEXT: $vgpr0_vgpr1_vgpr2 = COPY [[GLOBAL_LOAD_DWORDX3_]]
53    ;
54    ; GFX10-LABEL: name: load_global_v3s32
55    ; GFX10: liveins: $vgpr0_vgpr1
56    ; GFX10-NEXT: {{  $}}
57    ; GFX10-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
58    ; GFX10-NEXT: [[GLOBAL_LOAD_DWORDX3_:%[0-9]+]]:vreg_96 = GLOBAL_LOAD_DWORDX3 [[COPY]], 0, 0, implicit $exec :: (load (<3 x s32>), align 4, addrspace 1)
59    ; GFX10-NEXT: $vgpr0_vgpr1_vgpr2 = COPY [[GLOBAL_LOAD_DWORDX3_]]
60    %0:vgpr(p1) = COPY $vgpr0_vgpr1
61    %1:vgpr(<3 x s32>) = G_LOAD %0 :: (load (<3 x s32>), align 4, addrspace 1)
62    $vgpr0_vgpr1_vgpr2 = COPY %1
63
64...
65
66---
67
68name: load_global_s96
69legalized:       true
70regBankSelected: true
71tracksRegLiveness: true
72
73body: |
74  bb.0:
75    liveins:  $vgpr0_vgpr1
76
77    ; GFX7-LABEL: name: load_global_s96
78    ; GFX7: liveins: $vgpr0_vgpr1
79    ; GFX7-NEXT: {{  $}}
80    ; GFX7-NEXT: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
81    ; GFX7-NEXT: [[LOAD:%[0-9]+]]:vreg_96(s96) = G_LOAD [[COPY]](p1) :: (load (s96), align 4, addrspace 1)
82    ; GFX7-NEXT: $vgpr0_vgpr1_vgpr2 = COPY [[LOAD]](s96)
83    ;
84    ; GFX7-FLAT-LABEL: name: load_global_s96
85    ; GFX7-FLAT: liveins: $vgpr0_vgpr1
86    ; GFX7-FLAT-NEXT: {{  $}}
87    ; GFX7-FLAT-NEXT: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
88    ; GFX7-FLAT-NEXT: [[LOAD:%[0-9]+]]:vreg_96(s96) = G_LOAD [[COPY]](p1) :: (load (s96), align 4, addrspace 1)
89    ; GFX7-FLAT-NEXT: $vgpr0_vgpr1_vgpr2 = COPY [[LOAD]](s96)
90    ;
91    ; GFX8-LABEL: name: load_global_s96
92    ; GFX8: liveins: $vgpr0_vgpr1
93    ; GFX8-NEXT: {{  $}}
94    ; GFX8-NEXT: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
95    ; GFX8-NEXT: [[LOAD:%[0-9]+]]:vreg_96(s96) = G_LOAD [[COPY]](p1) :: (load (s96), align 4, addrspace 1)
96    ; GFX8-NEXT: $vgpr0_vgpr1_vgpr2 = COPY [[LOAD]](s96)
97    ;
98    ; GFX9-LABEL: name: load_global_s96
99    ; GFX9: liveins: $vgpr0_vgpr1
100    ; GFX9-NEXT: {{  $}}
101    ; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
102    ; GFX9-NEXT: [[LOAD:%[0-9]+]]:vreg_96(s96) = G_LOAD [[COPY]](p1) :: (load (s96), align 4, addrspace 1)
103    ; GFX9-NEXT: $vgpr0_vgpr1_vgpr2 = COPY [[LOAD]](s96)
104    ;
105    ; GFX10-LABEL: name: load_global_s96
106    ; GFX10: liveins: $vgpr0_vgpr1
107    ; GFX10-NEXT: {{  $}}
108    ; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
109    ; GFX10-NEXT: [[LOAD:%[0-9]+]]:vreg_96(s96) = G_LOAD [[COPY]](p1) :: (load (s96), align 4, addrspace 1)
110    ; GFX10-NEXT: $vgpr0_vgpr1_vgpr2 = COPY [[LOAD]](s96)
111    %0:vgpr(p1) = COPY $vgpr0_vgpr1
112    %1:vgpr(s96) = G_LOAD %0 :: (load (s96), align 4, addrspace 1)
113    $vgpr0_vgpr1_vgpr2 = COPY %1
114
115...
116
117---
118
119name: load_global_v6s16
120legalized:       true
121regBankSelected: true
122tracksRegLiveness: true
123
124body: |
125  bb.0:
126    liveins:  $vgpr0_vgpr1
127
128    ; GFX7-LABEL: name: load_global_v6s16
129    ; GFX7: liveins: $vgpr0_vgpr1
130    ; GFX7-NEXT: {{  $}}
131    ; GFX7-NEXT: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
132    ; GFX7-NEXT: [[LOAD:%[0-9]+]]:vreg_96(<6 x s16>) = G_LOAD [[COPY]](p1) :: (load (<6 x s16>), align 4, addrspace 1)
133    ; GFX7-NEXT: $vgpr0_vgpr1_vgpr2 = COPY [[LOAD]](<6 x s16>)
134    ;
135    ; GFX7-FLAT-LABEL: name: load_global_v6s16
136    ; GFX7-FLAT: liveins: $vgpr0_vgpr1
137    ; GFX7-FLAT-NEXT: {{  $}}
138    ; GFX7-FLAT-NEXT: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
139    ; GFX7-FLAT-NEXT: [[LOAD:%[0-9]+]]:vreg_96(<6 x s16>) = G_LOAD [[COPY]](p1) :: (load (<6 x s16>), align 4, addrspace 1)
140    ; GFX7-FLAT-NEXT: $vgpr0_vgpr1_vgpr2 = COPY [[LOAD]](<6 x s16>)
141    ;
142    ; GFX8-LABEL: name: load_global_v6s16
143    ; GFX8: liveins: $vgpr0_vgpr1
144    ; GFX8-NEXT: {{  $}}
145    ; GFX8-NEXT: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
146    ; GFX8-NEXT: [[LOAD:%[0-9]+]]:vreg_96(<6 x s16>) = G_LOAD [[COPY]](p1) :: (load (<6 x s16>), align 4, addrspace 1)
147    ; GFX8-NEXT: $vgpr0_vgpr1_vgpr2 = COPY [[LOAD]](<6 x s16>)
148    ;
149    ; GFX9-LABEL: name: load_global_v6s16
150    ; GFX9: liveins: $vgpr0_vgpr1
151    ; GFX9-NEXT: {{  $}}
152    ; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
153    ; GFX9-NEXT: [[LOAD:%[0-9]+]]:vreg_96(<6 x s16>) = G_LOAD [[COPY]](p1) :: (load (<6 x s16>), align 4, addrspace 1)
154    ; GFX9-NEXT: $vgpr0_vgpr1_vgpr2 = COPY [[LOAD]](<6 x s16>)
155    ;
156    ; GFX10-LABEL: name: load_global_v6s16
157    ; GFX10: liveins: $vgpr0_vgpr1
158    ; GFX10-NEXT: {{  $}}
159    ; GFX10-NEXT: [[COPY:%[0-9]+]]:vgpr(p1) = COPY $vgpr0_vgpr1
160    ; GFX10-NEXT: [[LOAD:%[0-9]+]]:vreg_96(<6 x s16>) = G_LOAD [[COPY]](p1) :: (load (<6 x s16>), align 4, addrspace 1)
161    ; GFX10-NEXT: $vgpr0_vgpr1_vgpr2 = COPY [[LOAD]](<6 x s16>)
162    %0:vgpr(p1) = COPY $vgpr0_vgpr1
163    %1:vgpr(<6 x s16>) = G_LOAD %0 :: (load (<6 x s16>), align 4, addrspace 1)
164    $vgpr0_vgpr1_vgpr2 = COPY %1
165
166...
167