Skip to content

Commit 9b909b8

Browse files
authored
[AMDGPU][NFC] Precommit tests representing agpr spills. (#115270)
Presently we are only marking implicit-def for the spilled AGPR tuple in the first spill instructions and not implicit.
1 parent 1469d82 commit 9b909b8

File tree

1 file changed

+94
-0
lines changed

1 file changed

+94
-0
lines changed
Lines changed: 94 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,94 @@
1+
# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
2+
# RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx908 %s -o - -run-pass prologepilog -verify-machineinstrs | FileCheck -check-prefix=GFX908-PEI %s
3+
# RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx908 %s -o - -run-pass prologepilog,machine-cp -verify-machineinstrs | FileCheck -check-prefix=GFX908-PEI-MACHINECP %s
4+
5+
# When VGPRs are available for spilling, prologepilog marks the tuple implicit-def as well as implicit in the first spill instruction.
6+
# As a consequence, machine-cp would NOT delete agpr2 copy here.
7+
8+
---
9+
name: agpr-spill-to-vgpr-machine-cp
10+
tracksRegLiveness: true
11+
stack:
12+
- { id: 0, name: '', type: spill-slot, offset: 0, size: 128, alignment: 4 }
13+
machineFunctionInfo:
14+
scratchRSrcReg: $sgpr0_sgpr1_sgpr2_sgpr3
15+
stackPtrOffsetReg: '$sgpr32'
16+
hasSpilledVGPRs: true
17+
body: |
18+
bb.0:
19+
successors:
20+
liveins: $vgpr0, $vgpr1
21+
; GFX908-PEI-LABEL: name: agpr-spill-to-vgpr-machine-cp
22+
; GFX908-PEI: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $vgpr9, $vgpr10, $vgpr11, $vgpr12, $vgpr13, $vgpr14, $vgpr15, $vgpr16, $vgpr17, $vgpr18, $vgpr19, $vgpr20, $vgpr21, $vgpr22, $vgpr23, $vgpr24, $vgpr25, $vgpr26, $vgpr27, $vgpr28, $vgpr29, $vgpr30, $vgpr31, $vgpr32, $vgpr33
23+
; GFX908-PEI-NEXT: {{ $}}
24+
; GFX908-PEI-NEXT: renamable $agpr0 = COPY renamable $vgpr0, implicit $exec
25+
; GFX908-PEI-NEXT: renamable $agpr2 = COPY renamable $vgpr1, implicit $exec
26+
; GFX908-PEI-NEXT: $vgpr33 = V_ACCVGPR_READ_B32_e64 $agpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2, implicit $agpr0_agpr1_agpr2
27+
; GFX908-PEI-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 $agpr1, implicit $exec
28+
; GFX908-PEI-NEXT: $vgpr31 = V_ACCVGPR_READ_B32_e64 $agpr2, implicit $exec, implicit $agpr0_agpr1_agpr2
29+
; GFX908-PEI-NEXT: S_ENDPGM 0
30+
;
31+
; GFX908-PEI-MACHINECP-LABEL: name: agpr-spill-to-vgpr-machine-cp
32+
; GFX908-PEI-MACHINECP: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4, $vgpr5, $vgpr6, $vgpr7, $vgpr8, $vgpr9, $vgpr10, $vgpr11, $vgpr12, $vgpr13, $vgpr14, $vgpr15, $vgpr16, $vgpr17, $vgpr18, $vgpr19, $vgpr20, $vgpr21, $vgpr22, $vgpr23, $vgpr24, $vgpr25, $vgpr26, $vgpr27, $vgpr28, $vgpr29, $vgpr30, $vgpr31, $vgpr32, $vgpr33
33+
; GFX908-PEI-MACHINECP-NEXT: {{ $}}
34+
; GFX908-PEI-MACHINECP-NEXT: renamable $agpr0 = COPY renamable $vgpr0, implicit $exec
35+
; GFX908-PEI-MACHINECP-NEXT: renamable $agpr2 = COPY renamable $vgpr1, implicit $exec
36+
; GFX908-PEI-MACHINECP-NEXT: $vgpr33 = V_ACCVGPR_READ_B32_e64 $agpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2, implicit $agpr0_agpr1_agpr2
37+
; GFX908-PEI-MACHINECP-NEXT: $vgpr32 = V_ACCVGPR_READ_B32_e64 $agpr1, implicit $exec
38+
; GFX908-PEI-MACHINECP-NEXT: $vgpr31 = V_ACCVGPR_READ_B32_e64 $agpr2, implicit $exec, implicit $agpr0_agpr1_agpr2
39+
; GFX908-PEI-MACHINECP-NEXT: S_ENDPGM 0
40+
renamable $agpr0 = COPY renamable $vgpr0, implicit $exec
41+
renamable $agpr2 = COPY renamable $vgpr1, implicit $exec
42+
SI_SPILL_AV96_SAVE $agpr0_agpr1_agpr2, %stack.0, $sgpr32, 0, implicit $exec :: (store (s96) into %stack.0, align 4, addrspace 5)
43+
S_ENDPGM 0
44+
...
45+
46+
# When VGPRs are NOT available for spilling (stack is used), prologepilog marks the tuple implicit-def only and NOT implicit.
47+
# As a consequence, machine-cp would delete agpr2 copy here. Presently, this is incorrect behavior.
48+
49+
---
50+
name: agpr-spill-to-vgpr-to-stack-machine-cp
51+
tracksRegLiveness: true
52+
stack:
53+
- { id: 0, name: '', type: spill-slot, offset: 0, size: 128, alignment: 4 }
54+
machineFunctionInfo:
55+
scratchRSrcReg: $sgpr0_sgpr1_sgpr2_sgpr3
56+
stackPtrOffsetReg: '$sgpr32'
57+
hasSpilledVGPRs: true
58+
body: |
59+
bb.0:
60+
successors:
61+
liveins: $vgpr0, $vgpr1
62+
; GFX908-PEI-LABEL: name: agpr-spill-to-vgpr-to-stack-machine-cp
63+
; GFX908-PEI: liveins: $vgpr0, $vgpr1, $vgpr18, $vgpr19, $vgpr20, $vgpr21, $vgpr22, $vgpr23, $vgpr24, $vgpr25, $vgpr26, $vgpr27, $vgpr28, $vgpr29, $vgpr30, $vgpr31, $vgpr32, $vgpr33, $vgpr34, $vgpr35, $vgpr36, $vgpr37, $vgpr38, $vgpr39, $vgpr48, $vgpr49, $vgpr50, $vgpr51, $vgpr52, $vgpr53, $vgpr54, $vgpr55
64+
; GFX908-PEI-NEXT: {{ $}}
65+
; GFX908-PEI-NEXT: renamable $agpr0 = COPY renamable $vgpr0, implicit $exec
66+
; GFX908-PEI-NEXT: renamable $agpr2 = COPY renamable $vgpr1, implicit $exec
67+
; GFX908-PEI-NEXT: $vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9 = IMPLICIT_DEF
68+
; GFX908-PEI-NEXT: $vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15_vgpr16_vgpr17 = IMPLICIT_DEF
69+
; GFX908-PEI-NEXT: $vgpr40 = V_ACCVGPR_READ_B32_e64 $agpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2
70+
; GFX908-PEI-NEXT: BUFFER_STORE_DWORD_OFFSET $vgpr40, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2 :: (store (s32) into %stack.0, addrspace 5)
71+
; GFX908-PEI-NEXT: $vgpr40 = V_ACCVGPR_READ_B32_e64 $agpr1, implicit $exec
72+
; GFX908-PEI-NEXT: BUFFER_STORE_DWORD_OFFSET $vgpr40, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, implicit $exec :: (store (s32) into %stack.0 + 4, addrspace 5)
73+
; GFX908-PEI-NEXT: $vgpr55 = V_ACCVGPR_READ_B32_e64 $agpr2, implicit $exec, implicit $agpr0_agpr1_agpr2
74+
; GFX908-PEI-NEXT: S_ENDPGM 0
75+
;
76+
; GFX908-PEI-MACHINECP-LABEL: name: agpr-spill-to-vgpr-to-stack-machine-cp
77+
; GFX908-PEI-MACHINECP: liveins: $vgpr0, $vgpr1, $vgpr18, $vgpr19, $vgpr20, $vgpr21, $vgpr22, $vgpr23, $vgpr24, $vgpr25, $vgpr26, $vgpr27, $vgpr28, $vgpr29, $vgpr30, $vgpr31, $vgpr32, $vgpr33, $vgpr34, $vgpr35, $vgpr36, $vgpr37, $vgpr38, $vgpr39, $vgpr48, $vgpr49, $vgpr50, $vgpr51, $vgpr52, $vgpr53, $vgpr54, $vgpr55
78+
; GFX908-PEI-MACHINECP-NEXT: {{ $}}
79+
; GFX908-PEI-MACHINECP-NEXT: renamable $agpr0 = COPY renamable $vgpr0, implicit $exec
80+
; GFX908-PEI-MACHINECP-NEXT: $vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9 = IMPLICIT_DEF
81+
; GFX908-PEI-MACHINECP-NEXT: $vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15_vgpr16_vgpr17 = IMPLICIT_DEF
82+
; GFX908-PEI-MACHINECP-NEXT: $vgpr40 = V_ACCVGPR_READ_B32_e64 $agpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2
83+
; GFX908-PEI-MACHINECP-NEXT: BUFFER_STORE_DWORD_OFFSET $vgpr40, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2 :: (store (s32) into %stack.0, addrspace 5)
84+
; GFX908-PEI-MACHINECP-NEXT: $vgpr40 = V_ACCVGPR_READ_B32_e64 $agpr1, implicit $exec
85+
; GFX908-PEI-MACHINECP-NEXT: BUFFER_STORE_DWORD_OFFSET $vgpr40, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, implicit $exec :: (store (s32) into %stack.0 + 4, addrspace 5)
86+
; GFX908-PEI-MACHINECP-NEXT: $vgpr55 = V_ACCVGPR_READ_B32_e64 $agpr2, implicit $exec, implicit $agpr0_agpr1_agpr2
87+
; GFX908-PEI-MACHINECP-NEXT: S_ENDPGM 0
88+
renamable $agpr0 = COPY renamable $vgpr0, implicit $exec
89+
renamable $agpr2 = COPY renamable $vgpr1, implicit $exec
90+
$vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9 = IMPLICIT_DEF
91+
$vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15_vgpr16_vgpr17 = IMPLICIT_DEF
92+
SI_SPILL_AV96_SAVE $agpr0_agpr1_agpr2, %stack.0, $sgpr32, 0, implicit $exec :: (store (s96) into %stack.0, align 4, addrspace 5)
93+
S_ENDPGM 0
94+
...

0 commit comments

Comments
 (0)