1# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
2# RUN: llc -march=amdgcn -mcpu=gfx90a -start-before=machine-scheduler -stop-after=postmisched %s -o - 2>&1 | FileCheck -check-prefix=DEFAULT %s
3# RUN: llc -march=amdgcn -mcpu=gfx90a -start-before=machine-scheduler -stop-after=postmisched %s -o - -amdgpu-igrouplp=1 2>&1 | FileCheck -check-prefix=PIPELINE %s
4
5---
6name: no_pipeline
7tracksRegLiveness: true
8body:             |
9  bb.0:
10    liveins: $sgpr0, $vgpr10_vgpr11
11    ; DEFAULT-LABEL: name: no_pipeline
12    ; DEFAULT: liveins: $sgpr0, $vgpr10_vgpr11
13    ; DEFAULT-NEXT: {{  $}}
14    ; DEFAULT-NEXT: $vgpr1 = V_MOV_B32_e32 1, implicit $exec
15    ; DEFAULT-NEXT: $vgpr0 = V_MOV_B32_e32 1, implicit $exec
16    ; DEFAULT-NEXT: $vgpr1 = V_ADD_F16_e32 killed $vgpr1, $vgpr0, implicit $mode, implicit $exec
17    ; DEFAULT-NEXT: GLOBAL_STORE_DWORD killed $vgpr10_vgpr11, $vgpr1, 0, 0, implicit $exec
18    ; DEFAULT-NEXT: $vgpr2 = V_MOV_B32_e32 1, implicit $exec
19    ; DEFAULT-NEXT: $vgpr3 = DS_READ_U16_gfx9 killed $vgpr2, 0, 0, implicit $exec
20    ; DEFAULT-NEXT: $vgpr5 = V_XOR_B32_e32 $vgpr1, killed $vgpr0, implicit $exec
21    ; DEFAULT-NEXT: $vgpr6 = V_MUL_LO_U32_e64 killed $vgpr1, killed $sgpr0, implicit $exec
22    ; DEFAULT-NEXT: $vgpr8 = V_MOV_B32_e32 0, implicit $exec
23    ; DEFAULT-NEXT: $vgpr9 = V_MOV_B32_e32 9, implicit $exec
24    ; PIPELINE-LABEL: name: no_pipeline
25    ; PIPELINE: liveins: $sgpr0, $vgpr10_vgpr11
26    ; PIPELINE-NEXT: {{  $}}
27    ; PIPELINE-NEXT: $vgpr1 = V_MOV_B32_e32 1, implicit $exec
28    ; PIPELINE-NEXT: $vgpr0 = V_MOV_B32_e32 1, implicit $exec
29    ; PIPELINE-NEXT: $vgpr1 = V_ADD_F16_e32 killed $vgpr1, $vgpr0, implicit $mode, implicit $exec
30    ; PIPELINE-NEXT: GLOBAL_STORE_DWORD killed $vgpr10_vgpr11, $vgpr1, 0, 0, implicit $exec
31    ; PIPELINE-NEXT: $vgpr2 = V_MOV_B32_e32 1, implicit $exec
32    ; PIPELINE-NEXT: $vgpr3 = DS_READ_U16_gfx9 killed $vgpr2, 0, 0, implicit $exec
33    ; PIPELINE-NEXT: $vgpr5 = V_XOR_B32_e32 $vgpr1, killed $vgpr0, implicit $exec
34    ; PIPELINE-NEXT: $vgpr6 = V_MUL_LO_U32_e64 killed $vgpr1, killed $sgpr0, implicit $exec
35    ; PIPELINE-NEXT: $vgpr8 = V_MOV_B32_e32 0, implicit $exec
36    ; PIPELINE-NEXT: $vgpr9 = V_MOV_B32_e32 9, implicit $exec
37    $vgpr1 = V_MOV_B32_e32 1, implicit $exec
38    $vgpr0 = V_MOV_B32_e32 1, implicit $exec
39    $vgpr8 = V_MOV_B32_e32 0, implicit $exec
40    $vgpr9 = V_MOV_B32_e32 9, implicit $exec
41    $vgpr1 = V_ADD_F16_e32 $vgpr1, $vgpr0, implicit $mode, implicit $exec
42    GLOBAL_STORE_DWORD $vgpr10_vgpr11, $vgpr1, 0, 0, implicit $exec
43    $vgpr2 = V_MOV_B32_e32 1, implicit $exec
44    $vgpr3 = DS_READ_U16_gfx9 $vgpr2, 0, 0, implicit $exec
45    $vgpr5 = V_XOR_B32_e32 $vgpr1, $vgpr0, implicit $exec
46    $vgpr6 = V_MUL_LO_U32_e64 $vgpr1, $sgpr0, implicit $exec
47...
48
49
50---
51name: full_pipe
52tracksRegLiveness: true
53body:             |
54  bb.0:
55    liveins:  $agpr0_agpr1_agpr2_agpr3, $agpr4_agpr5_agpr6_agpr7,  $agpr8_agpr9_agpr10_agpr11, $agpr12_agpr13_agpr14_agpr15, $agpr16_agpr17_agpr18_agpr19, $sgpr0, $vgpr10_vgpr11
56    ; DEFAULT-LABEL: name: full_pipe
57    ; DEFAULT: liveins: $sgpr0, $agpr0_agpr1_agpr2_agpr3, $agpr4_agpr5_agpr6_agpr7, $agpr8_agpr9_agpr10_agpr11, $agpr12_agpr13_agpr14_agpr15, $agpr16_agpr17_agpr18_agpr19, $vgpr10_vgpr11
58    ; DEFAULT-NEXT: {{  $}}
59    ; DEFAULT-NEXT: $vgpr0 = V_MOV_B32_e32 0, implicit $exec
60    ; DEFAULT-NEXT: $vgpr1 = V_MOV_B32_e32 1, implicit $exec
61    ; DEFAULT-NEXT: $vgpr2 = V_MOV_B32_e32 2, implicit $exec
62    ; DEFAULT-NEXT: $vgpr3 = V_MOV_B32_e32 3, implicit $exec
63    ; DEFAULT-NEXT: $vgpr6 = GLOBAL_LOAD_USHORT $vgpr0_vgpr1, 0, 0, implicit $exec
64    ; DEFAULT-NEXT: $vgpr7 = GLOBAL_LOAD_USHORT $vgpr2_vgpr3, 0, 0, implicit $exec
65    ; DEFAULT-NEXT: $vgpr4 = V_MOV_B32_e32 4, implicit $exec
66    ; DEFAULT-NEXT: $vgpr5 = V_MOV_B32_e32 5, implicit $exec
67    ; DEFAULT-NEXT: $vgpr8 = GLOBAL_LOAD_USHORT $vgpr4_vgpr5, 0, 0, implicit $exec
68    ; DEFAULT-NEXT: $vgpr1 = V_ADD_F16_e32 killed $vgpr1, $vgpr0, implicit $mode, implicit $exec
69    ; DEFAULT-NEXT: $vgpr26 = V_MOV_B32_e32 1, implicit $exec
70    ; DEFAULT-NEXT: $vgpr27 = V_MOV_B32_e32 1, implicit $exec
71    ; DEFAULT-NEXT: $agpr4_agpr5_agpr6_agpr7 = V_MFMA_F32_4X4X1F32_e64 $vgpr3, $vgpr4, killed $agpr4_agpr5_agpr6_agpr7, 0, 0, 0, implicit $mode, implicit $exec
72    ; DEFAULT-NEXT: $vgpr23 = V_XOR_B32_e32 $vgpr1, $vgpr0, implicit $exec
73    ; DEFAULT-NEXT: $vgpr9 = V_MOV_B32_e32 1, implicit $exec
74    ; DEFAULT-NEXT: $vgpr24 = V_MOV_B32_e32 1, implicit $exec
75    ; DEFAULT-NEXT: $agpr0_agpr1_agpr2_agpr3 = V_MFMA_F32_4X4X1F32_e64 $vgpr1, $vgpr0, killed $agpr0_agpr1_agpr2_agpr3, 0, 0, 0, implicit $mode, implicit $exec
76    ; DEFAULT-NEXT: $vgpr22 = V_XOR_B32_e32 $vgpr1, $vgpr0, implicit $exec
77    ; DEFAULT-NEXT: $agpr8_agpr9_agpr10_agpr11 = V_MFMA_F32_4X4X1F32_e64 $vgpr3, killed $vgpr4, killed $agpr8_agpr9_agpr10_agpr11, 0, 0, 0, implicit $mode, implicit $exec
78    ; DEFAULT-NEXT: $vgpr21 = V_MUL_LO_U32_e64 $vgpr1, killed $sgpr0, implicit $exec
79    ; DEFAULT-NEXT: $agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_4X4X1F32_e64 $vgpr1, $vgpr0, killed $agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
80    ; DEFAULT-NEXT: $vgpr30 = V_MOV_B32_e32 30, implicit $exec
81    ; DEFAULT-NEXT: $vgpr17 = V_MOV_B32_e32 1, implicit $exec
82    ; DEFAULT-NEXT: $vgpr18 = V_MOV_B32_e32 1, implicit $exec
83    ; DEFAULT-NEXT: BUNDLE implicit-def $vgpr10, implicit-def $vgpr10_lo16, implicit-def $vgpr10_hi16, implicit-def $vgpr11, implicit-def $vgpr11_lo16, implicit-def $vgpr11_hi16, implicit-def $vgpr12, implicit-def $vgpr12_lo16, implicit-def $vgpr12_hi16, implicit-def $vgpr15, implicit-def $vgpr15_lo16, implicit-def $vgpr15_hi16, implicit-def $vgpr16, implicit-def $vgpr16_lo16, implicit-def $vgpr16_hi16, implicit $vgpr7, implicit $exec {
84    ; DEFAULT-NEXT:   $vgpr10 = DS_READ_U16_gfx9 $vgpr7, 0, 512, implicit $exec
85    ; DEFAULT-NEXT:   $vgpr11 = DS_READ_U16_gfx9 $vgpr7, 0, 2048, implicit $exec
86    ; DEFAULT-NEXT:   $vgpr12 = DS_READ_U16_gfx9 $vgpr7, 0, 1024, implicit $exec
87    ; DEFAULT-NEXT:   $vgpr15 = DS_READ_U16_gfx9 $vgpr7, 0, 4096, implicit $exec
88    ; DEFAULT-NEXT:   $vgpr16 = DS_READ_U16_gfx9 $vgpr7, 0, 2048, implicit $exec
89    ; DEFAULT-NEXT: }
90    ; DEFAULT-NEXT: DS_WRITE_B32 $vgpr3, killed $vgpr1, 0, 16, implicit $m0, implicit $exec
91    ; DEFAULT-NEXT: BUNDLE implicit-def $vgpr19, implicit-def $vgpr19_lo16, implicit-def $vgpr19_hi16, implicit-def $vgpr20, implicit-def $vgpr20_lo16, implicit-def $vgpr20_hi16, implicit killed $vgpr26_vgpr27, implicit $exec {
92    ; DEFAULT-NEXT:   $vgpr19 = GLOBAL_LOAD_USHORT $vgpr26_vgpr27, 0, 0, implicit $exec
93    ; DEFAULT-NEXT:   $vgpr20 = GLOBAL_LOAD_USHORT killed $vgpr26_vgpr27, 0, 0, implicit $exec
94    ; DEFAULT-NEXT: }
95    ; DEFAULT-NEXT: $agpr4_agpr5_agpr6_agpr7 = V_MFMA_F32_4X4X1F32_e64 killed $vgpr5, killed $vgpr6, killed $agpr4_agpr5_agpr6_agpr7, 0, 0, 0, implicit $mode, implicit $exec
96    ; DEFAULT-NEXT: DS_WRITE_B32 killed $vgpr0, killed $vgpr7, 0, 16, implicit $m0, implicit $exec
97    ; DEFAULT-NEXT: $agpr16_agpr17_agpr18_agpr19 = V_MFMA_F32_4X4X1F32_e64 killed $vgpr10, killed $vgpr11, killed $agpr16_agpr17_agpr18_agpr19, 0, 0, 0, implicit $mode, implicit $exec
98    ; DEFAULT-NEXT: DS_WRITE_B32 killed $vgpr23, killed $vgpr3, 0, 16, implicit $m0, implicit $exec
99    ; DEFAULT-NEXT: DS_WRITE_B32 killed $vgpr9, killed $vgpr24, 0, 16, implicit $m0, implicit $exec
100    ; PIPELINE-LABEL: name: full_pipe
101    ; PIPELINE: liveins: $sgpr0, $agpr0_agpr1_agpr2_agpr3, $agpr4_agpr5_agpr6_agpr7, $agpr8_agpr9_agpr10_agpr11, $agpr12_agpr13_agpr14_agpr15, $agpr16_agpr17_agpr18_agpr19, $vgpr10_vgpr11
102    ; PIPELINE-NEXT: {{  $}}
103    ; PIPELINE-NEXT: $vgpr0 = V_MOV_B32_e32 0, implicit $exec
104    ; PIPELINE-NEXT: $vgpr1 = V_MOV_B32_e32 1, implicit $exec
105    ; PIPELINE-NEXT: $vgpr2 = V_MOV_B32_e32 2, implicit $exec
106    ; PIPELINE-NEXT: $vgpr3 = V_MOV_B32_e32 3, implicit $exec
107    ; PIPELINE-NEXT: $vgpr6 = GLOBAL_LOAD_USHORT $vgpr0_vgpr1, 0, 0, implicit $exec
108    ; PIPELINE-NEXT: $vgpr7 = GLOBAL_LOAD_USHORT $vgpr2_vgpr3, 0, 0, implicit $exec
109    ; PIPELINE-NEXT: $vgpr4 = V_MOV_B32_e32 4, implicit $exec
110    ; PIPELINE-NEXT: $vgpr5 = V_MOV_B32_e32 5, implicit $exec
111    ; PIPELINE-NEXT: $vgpr8 = GLOBAL_LOAD_USHORT $vgpr4_vgpr5, 0, 0, implicit $exec
112    ; PIPELINE-NEXT: $vgpr1 = V_ADD_F16_e32 killed $vgpr1, $vgpr0, implicit $mode, implicit $exec
113    ; PIPELINE-NEXT: $vgpr26 = V_MOV_B32_e32 1, implicit $exec
114    ; PIPELINE-NEXT: $vgpr27 = V_MOV_B32_e32 1, implicit $exec
115    ; PIPELINE-NEXT: $vgpr9 = V_MOV_B32_e32 1, implicit $exec
116    ; PIPELINE-NEXT: $vgpr24 = V_MOV_B32_e32 1, implicit $exec
117    ; PIPELINE-NEXT: $vgpr23 = V_XOR_B32_e32 $vgpr1, $vgpr0, implicit $exec
118    ; PIPELINE-NEXT: $vgpr22 = V_XOR_B32_e32 $vgpr1, $vgpr0, implicit $exec
119    ; PIPELINE-NEXT: $vgpr21 = V_MUL_LO_U32_e64 $vgpr1, killed $sgpr0, implicit $exec
120    ; PIPELINE-NEXT: $vgpr30 = V_MOV_B32_e32 30, implicit $exec
121    ; PIPELINE-NEXT: $vgpr17 = V_MOV_B32_e32 1, implicit $exec
122    ; PIPELINE-NEXT: $vgpr18 = V_MOV_B32_e32 1, implicit $exec
123    ; PIPELINE-NEXT: BUNDLE implicit-def $vgpr10, implicit-def $vgpr10_lo16, implicit-def $vgpr10_hi16, implicit-def $vgpr11, implicit-def $vgpr11_lo16, implicit-def $vgpr11_hi16, implicit-def $vgpr12, implicit-def $vgpr12_lo16, implicit-def $vgpr12_hi16, implicit-def $vgpr15, implicit-def $vgpr15_lo16, implicit-def $vgpr15_hi16, implicit-def $vgpr16, implicit-def $vgpr16_lo16, implicit-def $vgpr16_hi16, implicit $vgpr7, implicit $exec {
124    ; PIPELINE-NEXT:   $vgpr10 = DS_READ_U16_gfx9 $vgpr7, 0, 512, implicit $exec
125    ; PIPELINE-NEXT:   $vgpr11 = DS_READ_U16_gfx9 $vgpr7, 0, 2048, implicit $exec
126    ; PIPELINE-NEXT:   $vgpr12 = DS_READ_U16_gfx9 $vgpr7, 0, 1024, implicit $exec
127    ; PIPELINE-NEXT:   $vgpr15 = DS_READ_U16_gfx9 $vgpr7, 0, 4096, implicit $exec
128    ; PIPELINE-NEXT:   $vgpr16 = DS_READ_U16_gfx9 $vgpr7, 0, 2048, implicit $exec
129    ; PIPELINE-NEXT: }
130    ; PIPELINE-NEXT: DS_WRITE_B32 $vgpr3, $vgpr1, 0, 16, implicit $m0, implicit $exec
131    ; PIPELINE-NEXT: BUNDLE implicit-def $vgpr19, implicit-def $vgpr19_lo16, implicit-def $vgpr19_hi16, implicit-def $vgpr20, implicit-def $vgpr20_lo16, implicit-def $vgpr20_hi16, implicit killed $vgpr26_vgpr27, implicit $exec {
132    ; PIPELINE-NEXT:   $vgpr19 = GLOBAL_LOAD_USHORT $vgpr26_vgpr27, 0, 0, implicit $exec
133    ; PIPELINE-NEXT:   $vgpr20 = GLOBAL_LOAD_USHORT killed $vgpr26_vgpr27, 0, 0, implicit $exec
134    ; PIPELINE-NEXT: }
135    ; PIPELINE-NEXT: $agpr4_agpr5_agpr6_agpr7 = V_MFMA_F32_4X4X1F32_e64 $vgpr3, $vgpr4, killed $agpr4_agpr5_agpr6_agpr7, 0, 0, 0, implicit $mode, implicit $exec
136    ; PIPELINE-NEXT: $agpr0_agpr1_agpr2_agpr3 = V_MFMA_F32_4X4X1F32_e64 $vgpr1, $vgpr0, killed $agpr0_agpr1_agpr2_agpr3, 0, 0, 0, implicit $mode, implicit $exec
137    ; PIPELINE-NEXT: $agpr8_agpr9_agpr10_agpr11 = V_MFMA_F32_4X4X1F32_e64 $vgpr3, killed $vgpr4, killed $agpr8_agpr9_agpr10_agpr11, 0, 0, 0, implicit $mode, implicit $exec
138    ; PIPELINE-NEXT: $agpr4_agpr5_agpr6_agpr7 = V_MFMA_F32_4X4X1F32_e64 killed $vgpr5, killed $vgpr6, killed $agpr4_agpr5_agpr6_agpr7, 0, 0, 0, implicit $mode, implicit $exec
139    ; PIPELINE-NEXT: $agpr16_agpr17_agpr18_agpr19 = V_MFMA_F32_4X4X1F32_e64 killed $vgpr10, killed $vgpr11, killed $agpr16_agpr17_agpr18_agpr19, 0, 0, 0, implicit $mode, implicit $exec
140    ; PIPELINE-NEXT: $agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_4X4X1F32_e64 killed $vgpr1, $vgpr0, killed $agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
141    ; PIPELINE-NEXT: BUNDLE implicit killed $vgpr0, implicit killed $vgpr7, implicit $m0, implicit $exec, implicit killed $vgpr23, implicit killed $vgpr3 {
142    ; PIPELINE-NEXT:   DS_WRITE_B32 killed $vgpr0, killed $vgpr7, 0, 16, implicit $m0, implicit $exec
143    ; PIPELINE-NEXT:   DS_WRITE_B32 killed $vgpr23, killed $vgpr3, 0, 16, implicit $m0, implicit $exec
144    ; PIPELINE-NEXT: }
145    ; PIPELINE-NEXT: DS_WRITE_B32 killed $vgpr9, killed $vgpr24, 0, 16, implicit $m0, implicit $exec
146    $vgpr0 = V_MOV_B32_e32 0, implicit $exec
147    $vgpr1 = V_MOV_B32_e32 1, implicit $exec
148    $vgpr2 = V_MOV_B32_e32 2, implicit $exec
149    $vgpr3 = V_MOV_B32_e32 3, implicit $exec
150    $vgpr4 = V_MOV_B32_e32 4, implicit $exec
151    $vgpr5 = V_MOV_B32_e32 5, implicit $exec
152    $vgpr30 = V_MOV_B32_e32 30, implicit $exec
153    $vgpr6 = GLOBAL_LOAD_USHORT $vgpr0_vgpr1, 0, 0, implicit $exec
154    $vgpr7 = GLOBAL_LOAD_USHORT $vgpr2_vgpr3, 0, 0, implicit $exec
155    $vgpr8 = GLOBAL_LOAD_USHORT $vgpr4_vgpr5, 0, 0, implicit $exec
156    $vgpr9 = V_MOV_B32_e32 1, implicit $exec
157    $vgpr1 = V_ADD_F16_e32 $vgpr1, $vgpr0, implicit $mode, implicit $exec
158    $agpr0_agpr1_agpr2_agpr3 = V_MFMA_F32_4X4X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3, 0, 0, 0, implicit $mode, implicit $exec
159    $agpr4_agpr5_agpr6_agpr7 = V_MFMA_F32_4X4X1F32_e64 $vgpr3, $vgpr4, $agpr4_agpr5_agpr6_agpr7, 0, 0, 0, implicit $mode, implicit $exec
160    $vgpr23 = V_XOR_B32_e32 $vgpr1, $vgpr0, implicit $exec
161    $vgpr24 = V_MOV_B32_e32 1, implicit $exec
162    $agpr8_agpr9_agpr10_agpr11 = V_MFMA_F32_4X4X1F32_e64 $vgpr3, $vgpr4, $agpr8_agpr9_agpr10_agpr11, 0, 0, 0, implicit $mode, implicit $exec
163    $vgpr10 = DS_READ_U16_gfx9 $vgpr7, 0, 512, implicit $exec
164    $vgpr11 = DS_READ_U16_gfx9 $vgpr7, 0, 2048, implicit $exec
165    $vgpr26 = V_MOV_B32_e32 1, implicit $exec
166    $vgpr27 = V_MOV_B32_e32 1, implicit $exec
167    $vgpr12 = DS_READ_U16_gfx9 $vgpr7, 0, 1024, implicit $exec
168    $agpr4_agpr5_agpr6_agpr7 = V_MFMA_F32_4X4X1F32_e64 $vgpr5, $vgpr6, $agpr4_agpr5_agpr6_agpr7, 0, 0, 0, implicit $mode, implicit $exec
169    $vgpr22 = V_XOR_B32_e32 $vgpr1, $vgpr0, implicit $exec
170    $vgpr21 = V_MUL_LO_U32_e64 $vgpr1, $sgpr0, implicit $exec
171    $vgpr15 = DS_READ_U16_gfx9 $vgpr7, 0, 4096, implicit $exec
172    $vgpr16 = DS_READ_U16_gfx9 $vgpr7, 0, 2048, implicit $exec
173    DS_WRITE_B32 $vgpr3, $vgpr1, 0, 16, implicit $m0, implicit $exec
174    $vgpr19 = GLOBAL_LOAD_USHORT $vgpr26_vgpr27, 0, 0, implicit $exec
175    $vgpr17 = V_MOV_B32_e32 1, implicit $exec
176    $vgpr18 = V_MOV_B32_e32 1, implicit $exec
177    $vgpr20 = GLOBAL_LOAD_USHORT $vgpr26_vgpr27, 0, 0, implicit $exec
178    DS_WRITE_B32 $vgpr0, $vgpr7, 0, 16, implicit $m0, implicit $exec
179    $agpr16_agpr17_agpr18_agpr19 = V_MFMA_F32_4X4X1F32_e64 $vgpr10, $vgpr11, $agpr16_agpr17_agpr18_agpr19, 0, 0, 0, implicit $mode, implicit $exec
180    DS_WRITE_B32 $vgpr23, $vgpr3, 0, 16, implicit $m0, implicit $exec
181    $agpr12_agpr13_agpr14_agpr15 = V_MFMA_F32_4X4X1F32_e64 $vgpr1, $vgpr0, $agpr12_agpr13_agpr14_agpr15, 0, 0, 0, implicit $mode, implicit $exec
182    DS_WRITE_B32 $vgpr9, $vgpr24, 0, 16, implicit $m0, implicit $exec
183...
184
185---
186name: block_ends_in_bundle
187tracksRegLiveness: true
188body:             |
189  bb.0:
190    liveins: $vgpr0, $vgpr1, $vgpr7, $agpr0_agpr1_agpr2_agpr3
191    ; DEFAULT-LABEL: name: block_ends_in_bundle
192    ; DEFAULT: liveins: $vgpr0, $vgpr1, $vgpr7, $agpr0_agpr1_agpr2_agpr3
193    ; DEFAULT-NEXT: {{  $}}
194    ; DEFAULT-NEXT: BUNDLE implicit-def $vgpr10, implicit-def $vgpr10_lo16, implicit-def $vgpr10_hi16, implicit-def $vgpr11, implicit-def $vgpr11_lo16, implicit-def $vgpr11_hi16, implicit-def $vgpr12, implicit-def $vgpr12_lo16, implicit-def $vgpr12_hi16, implicit-def $vgpr15, implicit-def $vgpr15_lo16, implicit-def $vgpr15_hi16, implicit-def $vgpr16, implicit-def $vgpr16_lo16, implicit-def $vgpr16_hi16, implicit killed $vgpr7, implicit $exec {
195    ; DEFAULT-NEXT:   $vgpr10 = DS_READ_U16_gfx9 $vgpr7, 0, 512, implicit $exec
196    ; DEFAULT-NEXT:   $vgpr11 = DS_READ_U16_gfx9 $vgpr7, 0, 2048, implicit $exec
197    ; DEFAULT-NEXT:   $vgpr12 = DS_READ_U16_gfx9 $vgpr7, 0, 1024, implicit $exec
198    ; DEFAULT-NEXT:   $vgpr15 = DS_READ_U16_gfx9 $vgpr7, 0, 4096, implicit $exec
199    ; DEFAULT-NEXT:   $vgpr16 = DS_READ_U16_gfx9 killed $vgpr7, 0, 2048, implicit $exec
200    ; DEFAULT-NEXT: }
201    ; DEFAULT-NEXT: $agpr0_agpr1_agpr2_agpr3 = V_MFMA_F32_4X4X1F32_e64 killed $vgpr1, killed $vgpr0, killed $agpr0_agpr1_agpr2_agpr3, 0, 0, 0, implicit $mode, implicit $exec
202    ; PIPELINE-LABEL: name: block_ends_in_bundle
203    ; PIPELINE: liveins: $vgpr0, $vgpr1, $vgpr7, $agpr0_agpr1_agpr2_agpr3
204    ; PIPELINE-NEXT: {{  $}}
205    ; PIPELINE-NEXT: BUNDLE implicit-def $vgpr10, implicit-def $vgpr10_lo16, implicit-def $vgpr10_hi16, implicit-def $vgpr11, implicit-def $vgpr11_lo16, implicit-def $vgpr11_hi16, implicit-def $vgpr12, implicit-def $vgpr12_lo16, implicit-def $vgpr12_hi16, implicit-def $vgpr15, implicit-def $vgpr15_lo16, implicit-def $vgpr15_hi16, implicit-def $vgpr16, implicit-def $vgpr16_lo16, implicit-def $vgpr16_hi16, implicit killed $vgpr7, implicit $exec {
206    ; PIPELINE-NEXT:   $vgpr10 = DS_READ_U16_gfx9 $vgpr7, 0, 512, implicit $exec
207    ; PIPELINE-NEXT:   $vgpr11 = DS_READ_U16_gfx9 $vgpr7, 0, 2048, implicit $exec
208    ; PIPELINE-NEXT:   $vgpr12 = DS_READ_U16_gfx9 $vgpr7, 0, 1024, implicit $exec
209    ; PIPELINE-NEXT:   $vgpr15 = DS_READ_U16_gfx9 $vgpr7, 0, 4096, implicit $exec
210    ; PIPELINE-NEXT:   $vgpr16 = DS_READ_U16_gfx9 killed $vgpr7, 0, 2048, implicit $exec
211    ; PIPELINE-NEXT: }
212    ; PIPELINE-NEXT: $agpr0_agpr1_agpr2_agpr3 = V_MFMA_F32_4X4X1F32_e64 killed $vgpr1, killed $vgpr0, killed $agpr0_agpr1_agpr2_agpr3, 0, 0, 0, implicit $mode, implicit $exec
213      $agpr0_agpr1_agpr2_agpr3 = V_MFMA_F32_4X4X1F32_e64 $vgpr1, $vgpr0, $agpr0_agpr1_agpr2_agpr3, 0, 0, 0, implicit $mode, implicit $exec
214      BUNDLE implicit-def $vgpr10, implicit-def $vgpr10_lo16, implicit-def $vgpr10_hi16, implicit-def $vgpr11, implicit-def $vgpr11_lo16, implicit-def $vgpr11_hi16, implicit-def $vgpr12, implicit-def $vgpr12_lo16, implicit-def $vgpr12_hi16, implicit-def $vgpr15, implicit-def $vgpr15_lo16, implicit-def $vgpr15_hi16, implicit-def $vgpr16, implicit-def $vgpr16_lo16, implicit-def $vgpr16_hi16, implicit $vgpr7, implicit $exec {
215        $vgpr10 = DS_READ_U16_gfx9 $vgpr7, 0, 512, implicit $exec
216        $vgpr11 = DS_READ_U16_gfx9 $vgpr7, 0, 2048, implicit $exec
217        $vgpr12 = DS_READ_U16_gfx9 $vgpr7, 0, 1024, implicit $exec
218        $vgpr15 = DS_READ_U16_gfx9 $vgpr7, 0, 4096, implicit $exec
219        $vgpr16 = DS_READ_U16_gfx9 $vgpr7, 0, 2048, implicit $exec
220    }
221...
222