Skip to content

Commit 6030f75

Browse files
AMDGPU: precommit test for bug in RenameIndependentSubregs
1 parent d1a69e4 commit 6030f75

File tree

1 file changed

+207
-0
lines changed

1 file changed

+207
-0
lines changed
Lines changed: 207 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,207 @@
1+
# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 4
2+
# RUN: llc -mtriple=amdgcn--amdpal -mcpu=gfx1010 -start-before=rename-independent-subregs -mattr=+wavefrontsize64 -stop-before=amdgpu-mark-last-scratch-load %s -o - | FileCheck -check-prefix=REG_ALLOC %s
3+
# RUN: llc -mtriple=amdgcn--amdpal -mcpu=gfx1010 -start-before=rename-independent-subregs -mattr=+wavefrontsize64 -stop-after=machine-cp %s -o - | FileCheck -check-prefix=DEAD_INST_DEL %s
4+
5+
---
6+
name: _amdgpu_cs_main
7+
tracksRegLiveness: true
8+
body: |
9+
; REG_ALLOC-LABEL: name: _amdgpu_cs_main
10+
; REG_ALLOC: bb.0:
11+
; REG_ALLOC-NEXT: successors: %bb.2(0x40000000), %bb.1(0x40000000)
12+
; REG_ALLOC-NEXT: liveins: $sgpr0, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr10, $sgpr2_sgpr3, $sgpr4_sgpr5, $sgpr8_sgpr9_sgpr10_sgpr11
13+
; REG_ALLOC-NEXT: {{ $}}
14+
; REG_ALLOC-NEXT: renamable $vgpr5_vgpr6_vgpr7_vgpr8 = BUFFER_LOAD_DWORDX4_OFFEN renamable $vgpr3, renamable $sgpr8_sgpr9_sgpr10_sgpr11, 0, 0, 0, 0, implicit $exec :: (dereferenceable load (<4 x s32>), align 1, addrspace 8)
15+
; REG_ALLOC-NEXT: renamable $vgpr15_vgpr16_vgpr17_vgpr18 = BUFFER_LOAD_DWORDX4_OFFEN renamable $vgpr2, renamable $sgpr8_sgpr9_sgpr10_sgpr11, 0, 0, 0, 0, implicit $exec :: (dereferenceable load (<4 x s32>), align 1, addrspace 8)
16+
; REG_ALLOC-NEXT: renamable $vgpr11_vgpr12_vgpr13_vgpr14 = BUFFER_LOAD_DWORDX4_OFFEN renamable $vgpr0, renamable $sgpr8_sgpr9_sgpr10_sgpr11, 0, 0, 0, 0, implicit $exec :: (dereferenceable load (<4 x s32>), align 1, addrspace 8)
17+
; REG_ALLOC-NEXT: KILL killed renamable $vgpr2
18+
; REG_ALLOC-NEXT: KILL killed renamable $vgpr0
19+
; REG_ALLOC-NEXT: KILL killed renamable $vgpr3
20+
; REG_ALLOC-NEXT: renamable $sgpr12 = V_READFIRSTLANE_B32 killed $vgpr5, implicit $exec
21+
; REG_ALLOC-NEXT: renamable $vgpr4_vgpr5_vgpr6_vgpr7 = BUFFER_LOAD_DWORDX4_OFFEN killed renamable $vgpr4, killed renamable $sgpr8_sgpr9_sgpr10_sgpr11, 0, 0, 0, 0, implicit $exec :: (dereferenceable load (<4 x s32>), align 1, addrspace 8)
22+
; REG_ALLOC-NEXT: renamable $sgpr13 = V_READFIRSTLANE_B32 killed $vgpr15, implicit $exec
23+
; REG_ALLOC-NEXT: renamable $sgpr6_sgpr7 = V_CMP_NE_U32_e64 killed $vgpr1, 0, implicit $exec
24+
; REG_ALLOC-NEXT: S_CMP_EQ_U64 killed renamable $sgpr12_sgpr13, killed renamable $sgpr2_sgpr3, implicit-def $scc
25+
; REG_ALLOC-NEXT: renamable $sgpr1 = S_CSELECT_B32 1, 0, implicit $scc
26+
; REG_ALLOC-NEXT: renamable $vgpr0 = IMPLICIT_DEF
27+
; REG_ALLOC-NEXT: $exec = S_MOV_B64_term renamable $sgpr6_sgpr7
28+
; REG_ALLOC-NEXT: S_CBRANCH_EXECZ %bb.1, implicit $exec
29+
; REG_ALLOC-NEXT: S_BRANCH %bb.2
30+
; REG_ALLOC-NEXT: {{ $}}
31+
; REG_ALLOC-NEXT: bb.1:
32+
; REG_ALLOC-NEXT: successors: %bb.3(0x40000000), %bb.5(0x40000000)
33+
; REG_ALLOC-NEXT: liveins: $sgpr0, $vgpr0, $vgpr10, $sgpr4_sgpr5, $sgpr6_sgpr7, $vgpr11_vgpr12_vgpr13_vgpr14:0x0000000000000003, $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7:0x0000000000000300
34+
; REG_ALLOC-NEXT: {{ $}}
35+
; REG_ALLOC-NEXT: renamable $sgpr2_sgpr3 = S_OR_SAVEEXEC_B64 killed renamable $sgpr6_sgpr7, implicit-def $exec, implicit-def $scc, implicit $exec
36+
; REG_ALLOC-NEXT: $exec = S_XOR_B64_term $exec, renamable $sgpr2_sgpr3, implicit-def $scc
37+
; REG_ALLOC-NEXT: S_CBRANCH_EXECZ %bb.5, implicit $exec
38+
; REG_ALLOC-NEXT: S_BRANCH %bb.3
39+
; REG_ALLOC-NEXT: {{ $}}
40+
; REG_ALLOC-NEXT: bb.2:
41+
; REG_ALLOC-NEXT: successors: %bb.1(0x80000000)
42+
; REG_ALLOC-NEXT: liveins: $sgpr0, $sgpr1, $vgpr10, $sgpr4_sgpr5, $sgpr6_sgpr7
43+
; REG_ALLOC-NEXT: {{ $}}
44+
; REG_ALLOC-NEXT: renamable $sgpr1 = S_OR_B32 killed renamable $sgpr1, 2, implicit-def dead $scc
45+
; REG_ALLOC-NEXT: renamable $vgpr0 = COPY killed renamable $sgpr1
46+
; REG_ALLOC-NEXT: renamable $vgpr11_vgpr12 = IMPLICIT_DEF
47+
; REG_ALLOC-NEXT: renamable $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7 = IMPLICIT_DEF
48+
; REG_ALLOC-NEXT: S_BRANCH %bb.1
49+
; REG_ALLOC-NEXT: {{ $}}
50+
; REG_ALLOC-NEXT: bb.3:
51+
; REG_ALLOC-NEXT: successors: %bb.5(0x80000000)
52+
; REG_ALLOC-NEXT: liveins: $sgpr0, $vgpr10, $sgpr2_sgpr3, $sgpr4_sgpr5, $vgpr11_vgpr12_vgpr13_vgpr14:0x0000000000000003, $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7:0x0000000000000300
53+
; REG_ALLOC-NEXT: {{ $}}
54+
; REG_ALLOC-NEXT: renamable $sgpr1 = V_READFIRSTLANE_B32 killed $vgpr11, implicit $exec
55+
; REG_ALLOC-NEXT: renamable $sgpr6 = V_READFIRSTLANE_B32 killed $vgpr4, implicit $exec
56+
; REG_ALLOC-NEXT: S_CMP_EQ_U32 killed renamable $sgpr6, killed renamable $sgpr1, implicit-def $scc
57+
; REG_ALLOC-NEXT: renamable $sgpr1 = S_CSELECT_B32 1, 0, implicit $scc
58+
; REG_ALLOC-NEXT: renamable $vgpr0 = COPY killed renamable $sgpr1
59+
; REG_ALLOC-NEXT: S_BRANCH %bb.5
60+
; REG_ALLOC-NEXT: {{ $}}
61+
; REG_ALLOC-NEXT: bb.4:
62+
; REG_ALLOC-NEXT: liveins: $sgpr0, $vgpr0, $vgpr10, $sgpr4_sgpr5
63+
; REG_ALLOC-NEXT: {{ $}}
64+
; REG_ALLOC-NEXT: renamable $sgpr4_sgpr5_sgpr6_sgpr7 = S_LOAD_DWORDX4_IMM killed renamable $sgpr4_sgpr5, 0, 0 :: (invariant load (<4 x s32>), addrspace 4)
65+
; REG_ALLOC-NEXT: renamable $vgpr1 = V_ADD_U32_e64 killed $sgpr0, killed $vgpr10, 0, implicit $exec
66+
; REG_ALLOC-NEXT: BUFFER_STORE_DWORD_OFFEN_exact killed renamable $vgpr0, killed renamable $vgpr1, killed renamable $sgpr4_sgpr5_sgpr6_sgpr7, 0, 0, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 8)
67+
; REG_ALLOC-NEXT: S_ENDPGM 0
68+
; REG_ALLOC-NEXT: {{ $}}
69+
; REG_ALLOC-NEXT: bb.5:
70+
; REG_ALLOC-NEXT: successors: %bb.4(0x80000000)
71+
; REG_ALLOC-NEXT: liveins: $sgpr0, $vgpr0, $vgpr10, $sgpr2_sgpr3, $sgpr4_sgpr5
72+
; REG_ALLOC-NEXT: {{ $}}
73+
; REG_ALLOC-NEXT: $exec = S_OR_B64 $exec, killed renamable $sgpr2_sgpr3, implicit-def $scc
74+
; REG_ALLOC-NEXT: S_BRANCH %bb.4
75+
;
76+
; DEAD_INST_DEL-LABEL: name: _amdgpu_cs_main
77+
; DEAD_INST_DEL: bb.0:
78+
; DEAD_INST_DEL-NEXT: successors: %bb.2(0x40000000), %bb.1(0x40000000)
79+
; DEAD_INST_DEL-NEXT: liveins: $sgpr0, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr10, $sgpr2_sgpr3, $sgpr4_sgpr5, $sgpr8_sgpr9_sgpr10_sgpr11
80+
; DEAD_INST_DEL-NEXT: {{ $}}
81+
; DEAD_INST_DEL-NEXT: renamable $vgpr5_vgpr6_vgpr7_vgpr8 = BUFFER_LOAD_DWORDX4_OFFEN renamable $vgpr3, renamable $sgpr8_sgpr9_sgpr10_sgpr11, 0, 0, 0, 0, implicit $exec :: (dereferenceable load (<4 x s32>), align 1, addrspace 8)
82+
; DEAD_INST_DEL-NEXT: renamable $vgpr15_vgpr16_vgpr17_vgpr18 = BUFFER_LOAD_DWORDX4_OFFEN renamable $vgpr2, renamable $sgpr8_sgpr9_sgpr10_sgpr11, 0, 0, 0, 0, implicit $exec :: (dereferenceable load (<4 x s32>), align 1, addrspace 8)
83+
; DEAD_INST_DEL-NEXT: renamable $vgpr11_vgpr12_vgpr13_vgpr14 = BUFFER_LOAD_DWORDX4_OFFEN renamable $vgpr0, renamable $sgpr8_sgpr9_sgpr10_sgpr11, 0, 0, 0, 0, implicit $exec :: (dereferenceable load (<4 x s32>), align 1, addrspace 8)
84+
; DEAD_INST_DEL-NEXT: KILL killed renamable $vgpr2
85+
; DEAD_INST_DEL-NEXT: KILL killed renamable $vgpr0
86+
; DEAD_INST_DEL-NEXT: KILL killed renamable $vgpr3
87+
; DEAD_INST_DEL-NEXT: renamable $sgpr12 = V_READFIRSTLANE_B32 killed $vgpr5, implicit $exec
88+
; DEAD_INST_DEL-NEXT: renamable $vgpr4_vgpr5_vgpr6_vgpr7 = BUFFER_LOAD_DWORDX4_OFFEN killed renamable $vgpr4, killed renamable $sgpr8_sgpr9_sgpr10_sgpr11, 0, 0, 0, 0, implicit $exec :: (dereferenceable load (<4 x s32>), align 1, addrspace 8)
89+
; DEAD_INST_DEL-NEXT: renamable $sgpr13 = V_READFIRSTLANE_B32 killed $vgpr15, implicit $exec
90+
; DEAD_INST_DEL-NEXT: renamable $sgpr6_sgpr7 = V_CMP_NE_U32_e64 killed $vgpr1, 0, implicit $exec
91+
; DEAD_INST_DEL-NEXT: S_CMP_EQ_U64 killed renamable $sgpr12_sgpr13, killed renamable $sgpr2_sgpr3, implicit-def $scc
92+
; DEAD_INST_DEL-NEXT: renamable $sgpr1 = S_CSELECT_B32 1, 0, implicit $scc
93+
; DEAD_INST_DEL-NEXT: renamable $vgpr0 = IMPLICIT_DEF
94+
; DEAD_INST_DEL-NEXT: $exec = S_MOV_B64_term renamable $sgpr6_sgpr7
95+
; DEAD_INST_DEL-NEXT: S_CBRANCH_EXECZ %bb.1, implicit $exec
96+
; DEAD_INST_DEL-NEXT: S_BRANCH %bb.2
97+
; DEAD_INST_DEL-NEXT: {{ $}}
98+
; DEAD_INST_DEL-NEXT: bb.1:
99+
; DEAD_INST_DEL-NEXT: successors: %bb.3(0x40000000), %bb.5(0x40000000)
100+
; DEAD_INST_DEL-NEXT: liveins: $sgpr0, $vgpr0, $vgpr10, $sgpr4_sgpr5, $sgpr6_sgpr7, $vgpr11_vgpr12_vgpr13_vgpr14:0x0000000000000003, $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7:0x0000000000000300
101+
; DEAD_INST_DEL-NEXT: {{ $}}
102+
; DEAD_INST_DEL-NEXT: renamable $sgpr2_sgpr3 = S_OR_SAVEEXEC_B64 killed renamable $sgpr6_sgpr7, implicit-def $exec, implicit-def $scc, implicit $exec
103+
; DEAD_INST_DEL-NEXT: $exec = S_XOR_B64_term $exec, renamable $sgpr2_sgpr3, implicit-def $scc
104+
; DEAD_INST_DEL-NEXT: S_CBRANCH_EXECZ %bb.5, implicit $exec
105+
; DEAD_INST_DEL-NEXT: S_BRANCH %bb.3
106+
; DEAD_INST_DEL-NEXT: {{ $}}
107+
; DEAD_INST_DEL-NEXT: bb.2:
108+
; DEAD_INST_DEL-NEXT: successors: %bb.1(0x80000000)
109+
; DEAD_INST_DEL-NEXT: liveins: $sgpr0, $sgpr1, $vgpr10, $sgpr4_sgpr5, $sgpr6_sgpr7
110+
; DEAD_INST_DEL-NEXT: {{ $}}
111+
; DEAD_INST_DEL-NEXT: renamable $sgpr1 = S_OR_B32 killed renamable $sgpr1, 2, implicit-def dead $scc
112+
; DEAD_INST_DEL-NEXT: renamable $vgpr11_vgpr12 = IMPLICIT_DEF
113+
; DEAD_INST_DEL-NEXT: renamable $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7 = IMPLICIT_DEF
114+
; DEAD_INST_DEL-NEXT: S_BRANCH %bb.1
115+
; DEAD_INST_DEL-NEXT: {{ $}}
116+
; DEAD_INST_DEL-NEXT: bb.3:
117+
; DEAD_INST_DEL-NEXT: successors: %bb.5(0x80000000)
118+
; DEAD_INST_DEL-NEXT: liveins: $sgpr0, $vgpr10, $sgpr2_sgpr3, $sgpr4_sgpr5, $vgpr11_vgpr12_vgpr13_vgpr14:0x0000000000000003, $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7:0x0000000000000300
119+
; DEAD_INST_DEL-NEXT: {{ $}}
120+
; DEAD_INST_DEL-NEXT: renamable $sgpr1 = V_READFIRSTLANE_B32 killed $vgpr11, implicit $exec
121+
; DEAD_INST_DEL-NEXT: renamable $sgpr6 = V_READFIRSTLANE_B32 killed $vgpr4, implicit $exec
122+
; DEAD_INST_DEL-NEXT: S_CMP_EQ_U32 killed renamable $sgpr6, killed renamable $sgpr1, implicit-def $scc
123+
; DEAD_INST_DEL-NEXT: renamable $sgpr1 = S_CSELECT_B32 1, 0, implicit $scc
124+
; DEAD_INST_DEL-NEXT: renamable $vgpr0 = COPY killed renamable $sgpr1
125+
; DEAD_INST_DEL-NEXT: S_BRANCH %bb.5
126+
; DEAD_INST_DEL-NEXT: {{ $}}
127+
; DEAD_INST_DEL-NEXT: bb.4:
128+
; DEAD_INST_DEL-NEXT: liveins: $sgpr0, $vgpr0, $vgpr10, $sgpr4_sgpr5
129+
; DEAD_INST_DEL-NEXT: {{ $}}
130+
; DEAD_INST_DEL-NEXT: renamable $sgpr4_sgpr5_sgpr6_sgpr7 = S_LOAD_DWORDX4_IMM killed renamable $sgpr4_sgpr5, 0, 0 :: (invariant load (<4 x s32>), addrspace 4)
131+
; DEAD_INST_DEL-NEXT: renamable $vgpr1 = V_ADD_U32_e64 killed $sgpr0, killed $vgpr10, 0, implicit $exec
132+
; DEAD_INST_DEL-NEXT: BUFFER_STORE_DWORD_OFFEN_exact killed renamable $vgpr0, killed renamable $vgpr1, killed renamable $sgpr4_sgpr5_sgpr6_sgpr7, 0, 0, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 8)
133+
; DEAD_INST_DEL-NEXT: S_ENDPGM 0
134+
; DEAD_INST_DEL-NEXT: {{ $}}
135+
; DEAD_INST_DEL-NEXT: bb.5:
136+
; DEAD_INST_DEL-NEXT: successors: %bb.4(0x80000000)
137+
; DEAD_INST_DEL-NEXT: liveins: $sgpr0, $vgpr0, $vgpr10, $sgpr2_sgpr3, $sgpr4_sgpr5
138+
; DEAD_INST_DEL-NEXT: {{ $}}
139+
; DEAD_INST_DEL-NEXT: $exec = S_OR_B64 $exec, killed renamable $sgpr2_sgpr3, implicit-def $scc
140+
; DEAD_INST_DEL-NEXT: S_BRANCH %bb.4
141+
bb.0:
142+
successors: %bb.2(0x40000000), %bb.1(0x40000000)
143+
liveins: $sgpr0, $sgpr2_sgpr3, $sgpr4_sgpr5, $sgpr8_sgpr9_sgpr10_sgpr11, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr10
144+
145+
%0:sreg_32 = COPY $sgpr0
146+
%1:vgpr_32 = COPY $vgpr0
147+
%2:vgpr_32 = COPY $vgpr1
148+
%3:vgpr_32 = COPY $vgpr2
149+
%4:vgpr_32 = COPY $vgpr3
150+
%5:vgpr_32 = COPY $vgpr4
151+
%6:vgpr_32 = COPY $vgpr10
152+
%7:sreg_64 = COPY $sgpr2_sgpr3
153+
%8:sreg_64 = COPY $sgpr4_sgpr5
154+
%9:sgpr_128 = COPY $sgpr8_sgpr9_sgpr10_sgpr11
155+
undef %10.sub0_sub1_sub2_sub3:vreg_256 = BUFFER_LOAD_DWORDX4_OFFEN %1, %9, 0, 0, 0, 0, implicit $exec :: (dereferenceable load (<4 x s32>), align 1, addrspace 8)
156+
%10.sub4_sub5_sub6_sub7:vreg_256 = BUFFER_LOAD_DWORDX4_OFFEN %5, %9, 0, 0, 0, 0, implicit $exec :: (dereferenceable load (<4 x s32>), align 1, addrspace 8)
157+
%11:vreg_128 = BUFFER_LOAD_DWORDX4_OFFEN %4, %9, 0, 0, 0, 0, implicit $exec :: (dereferenceable load (<4 x s32>), align 1, addrspace 8)
158+
%12:vreg_128 = BUFFER_LOAD_DWORDX4_OFFEN %3, %9, 0, 0, 0, 0, implicit $exec :: (dereferenceable load (<4 x s32>), align 1, addrspace 8)
159+
undef %13.sub0:sgpr_256 = V_READFIRSTLANE_B32 %11.sub0, implicit $exec
160+
%13.sub1:sgpr_256 = V_READFIRSTLANE_B32 %12.sub0, implicit $exec
161+
S_CMP_EQ_U64 %13.sub0_sub1, %7, implicit-def $scc
162+
%14:sreg_32 = S_CSELECT_B32 1, 0, implicit killed $scc
163+
%15:sreg_64_xexec = V_CMP_NE_U32_e64 %2, 0, implicit $exec
164+
%16:vgpr_32 = IMPLICIT_DEF
165+
$exec = S_MOV_B64_term %15
166+
S_CBRANCH_EXECZ %bb.1, implicit $exec
167+
S_BRANCH %bb.2
168+
169+
bb.1:
170+
successors: %bb.3(0x40000000), %bb.5(0x40000000)
171+
172+
%17:sreg_64 = S_OR_SAVEEXEC_B64 %15, implicit-def $exec, implicit-def $scc, implicit $exec
173+
%18:sreg_64_xexec = S_AND_B64 $exec, %17, implicit-def $scc
174+
$exec = S_XOR_B64_term $exec, %18, implicit-def $scc
175+
S_CBRANCH_EXECZ %bb.5, implicit $exec
176+
S_BRANCH %bb.3
177+
178+
bb.2:
179+
successors: %bb.1(0x80000000)
180+
181+
%19:sreg_32 = S_OR_B32 %14, 2, implicit-def dead $scc
182+
%16:vgpr_32 = COPY %19
183+
undef %10.sub0_sub1:vreg_256 = IMPLICIT_DEF
184+
S_BRANCH %bb.1
185+
186+
bb.3:
187+
successors: %bb.5(0x80000000)
188+
189+
%20:sreg_32 = V_READFIRSTLANE_B32 %10.sub0, implicit $exec
190+
%21:sreg_32 = V_READFIRSTLANE_B32 %10.sub4, implicit $exec
191+
S_CMP_EQ_U32 %21, %20, implicit-def $scc
192+
%22:sreg_32 = S_CSELECT_B32 1, 0, implicit killed $scc
193+
%16:vgpr_32 = COPY %22
194+
S_BRANCH %bb.5
195+
196+
bb.4:
197+
%23:vgpr_32 = V_ADD_U32_e64 %0, %6, 0, implicit $exec
198+
%24:sgpr_128 = S_LOAD_DWORDX4_IMM %8, 0, 0 :: (invariant load (<4 x s32>), addrspace 4)
199+
BUFFER_STORE_DWORD_OFFEN_exact %16, %23, %24, 0, 0, 0, 0, implicit $exec :: (dereferenceable store (s32), align 1, addrspace 8)
200+
S_ENDPGM 0
201+
202+
bb.5:
203+
successors: %bb.4(0x80000000)
204+
205+
$exec = S_OR_B64 $exec, %18, implicit-def $scc
206+
S_BRANCH %bb.4
207+
...

0 commit comments

Comments
 (0)