Skip to content

Commit 4fb7d19

Browse files
bababuckbababuck
authored
[AMDGPU] Remove implicit definition of register group when restoring the last sub-register after a spill (#133986)
Extension of https://reviews.llvm.org/D141101 to remove the define flag from the last stack memory access. Fixes case where COPY instructions are used for some of the stack restoration, but the copies get optimized away during the machine-cp pass. Prior to this change, was possible to produce the following code: $agpr16_agpr17_agpr18_agpr19 = SCRATCH_LOAD_DWORDX4_ST 64, 0, implicit $exec, implicit $flat_scr, implicit-def $agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (load (s128) from %stack.17, align 4, addrspace 5) $agpr20_agpr21_agpr22_agpr23 = SCRATCH_LOAD_DWORDX4_ST 80, 0, implicit $exec, implicit $flat_scr :: (load (s128) from %stack.17 + 16, align 4, addrspace 5) $agpr24_agpr25_agpr26_agpr27 = SCRATCH_LOAD_DWORDX4_ST 96, 0, implicit $exec, implicit $flat_scr :: (load (s128) from %stack.17 + 32, align 4, addrspace 5) $agpr31 = COPY $agpr112, implicit $agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 $agpr30 = COPY $agpr208, implicit $agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 $agpr28_agpr29 = SCRATCH_LOAD_DWORDX2_ST 112, 0, implicit $exec, implicit $flat_scr, implicit-def $agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 :: (load (s64) from %stack.17 + 48, align 4, addrspace 5) where `$agpr30 = COPY $agpr208` would be optimized away by `machine-cp` pass. Instead, change to: $agpr28_agpr29 = SCRATCH_LOAD_DWORDX2_ST 112, 0, implicit $exec, implicit $flat_scr :: (load (s64) from %stack.17 + 48, align 4, addrspace 5) Fixes #131386. Made the simple fix, but I'm not completely comfortable with this change since the reason for the previous inclusion of `IsLastSubReg` is unclear to me. @krzysz00
1 parent 7f5bf77 commit 4fb7d19

10 files changed

+339
-336
lines changed

‎llvm/lib/Target/AMDGPU/SIRegisterInfo.cpp

Lines changed: 3 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -1928,7 +1928,9 @@ void SIRegisterInfo::buildSpillLoadStore(
19281928
MIB->setAsmPrinterFlag(MachineInstr::ReloadReuse);
19291929
}
19301930

1931-
if (NeedSuperRegImpOperand && (IsFirstSubReg || IsLastSubReg))
1931+
bool IsSrcDstDef = SrcDstRegState & RegState::Define;
1932+
if (NeedSuperRegImpOperand &&
1933+
(IsFirstSubReg || (IsLastSubReg && !IsSrcDstDef)))
19321934
MIB.addReg(ValueReg, RegState::Implicit | SrcDstRegState);
19331935

19341936
// The epilog restore of a wwm-scratch register can cause undesired

‎llvm/test/CodeGen/AMDGPU/accvgpr-spill-scc-clobber.mir

Lines changed: 14 additions & 14 deletions
Original file line numberDiff line numberDiff line change
@@ -1050,7 +1050,7 @@ body: |
10501050
; GFX908-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFEN $vgpr1, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, implicit $exec :: (load (s32) from %stack.1, addrspace 5)
10511051
; GFX908-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1
10521052
; GFX908-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFEN $vgpr1, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, implicit $exec :: (load (s32) from %stack.1 + 4, addrspace 5)
1053-
; GFX908-NEXT: $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1
1053+
; GFX908-NEXT: $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec
10541054
; GFX908-NEXT: $vgpr1 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, implicit $exec :: (load (s32) from %stack.2, addrspace 5)
10551055
; GFX908-NEXT: S_CBRANCH_SCC1 %bb.2, implicit $scc
10561056
; GFX908-NEXT: {{ $}}
@@ -1297,7 +1297,7 @@ body: |
12971297
; GFX90A-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr40, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 704, 0, 0, implicit $exec :: (store (s32) into %stack.226, addrspace 5)
12981298
; GFX90A-NEXT: $vgpr40 = V_MOV_B32_e32 8904, implicit $exec
12991299
; GFX90A-NEXT: $agpr0 = BUFFER_LOAD_DWORD_OFFEN $vgpr40, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, implicit $exec, implicit-def $agpr0_agpr1 :: (load (s32) from %stack.1, addrspace 5)
1300-
; GFX90A-NEXT: $agpr1 = BUFFER_LOAD_DWORD_OFFEN killed $vgpr40, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, implicit $exec, implicit-def $agpr0_agpr1 :: (load (s32) from %stack.1 + 4, addrspace 5)
1300+
; GFX90A-NEXT: $agpr1 = BUFFER_LOAD_DWORD_OFFEN killed $vgpr40, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, implicit $exec :: (load (s32) from %stack.1 + 4, addrspace 5)
13011301
; GFX90A-NEXT: $vgpr40 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 704, 0, 0, implicit $exec :: (load (s32) from %stack.226, addrspace 5)
13021302
; GFX90A-NEXT: S_CBRANCH_SCC1 %bb.2, implicit $scc
13031303
; GFX90A-NEXT: {{ $}}
@@ -1548,7 +1548,7 @@ body: |
15481548
; GFX908-FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD $vgpr1, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.1, addrspace 5)
15491549
; GFX908-FLATSCR-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1
15501550
; GFX908-FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD $vgpr1, 4, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.1 + 4, addrspace 5)
1551-
; GFX908-FLATSCR-NEXT: $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1
1551+
; GFX908-FLATSCR-NEXT: $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec
15521552
; GFX908-FLATSCR-NEXT: $vgpr1 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.2, addrspace 5)
15531553
; GFX908-FLATSCR-NEXT: S_CBRANCH_SCC1 %bb.2, implicit $scc
15541554
; GFX908-FLATSCR-NEXT: {{ $}}
@@ -2075,7 +2075,7 @@ body: |
20752075
; GFX908-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFEN $vgpr1, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, implicit $exec :: (load (s32) from %stack.1 + 4, addrspace 5)
20762076
; GFX908-NEXT: $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec
20772077
; GFX908-NEXT: $vgpr0 = BUFFER_LOAD_DWORD_OFFEN $vgpr1, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, implicit $exec :: (load (s32) from %stack.1 + 8, addrspace 5)
2078-
; GFX908-NEXT: $agpr2 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2
2078+
; GFX908-NEXT: $agpr2 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec
20792079
; GFX908-NEXT: $vgpr1 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, implicit $exec :: (load (s32) from %stack.2, addrspace 5)
20802080
; GFX908-NEXT: S_CBRANCH_SCC1 %bb.2, implicit $scc
20812081
; GFX908-NEXT: {{ $}}
@@ -2323,7 +2323,7 @@ body: |
23232323
; GFX90A-NEXT: $vgpr40 = V_MOV_B32_e32 8904, implicit $exec
23242324
; GFX90A-NEXT: $agpr0 = BUFFER_LOAD_DWORD_OFFEN $vgpr40, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, implicit $exec, implicit-def $agpr0_agpr1_agpr2 :: (load (s32) from %stack.1, addrspace 5)
23252325
; GFX90A-NEXT: $agpr1 = BUFFER_LOAD_DWORD_OFFEN $vgpr40, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, implicit $exec :: (load (s32) from %stack.1 + 4, addrspace 5)
2326-
; GFX90A-NEXT: $agpr2 = BUFFER_LOAD_DWORD_OFFEN killed $vgpr40, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, implicit $exec, implicit-def $agpr0_agpr1_agpr2 :: (load (s32) from %stack.1 + 8, addrspace 5)
2326+
; GFX90A-NEXT: $agpr2 = BUFFER_LOAD_DWORD_OFFEN killed $vgpr40, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, implicit $exec :: (load (s32) from %stack.1 + 8, addrspace 5)
23272327
; GFX90A-NEXT: $vgpr40 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 704, 0, 0, implicit $exec :: (load (s32) from %stack.226, addrspace 5)
23282328
; GFX90A-NEXT: S_CBRANCH_SCC1 %bb.2, implicit $scc
23292329
; GFX90A-NEXT: {{ $}}
@@ -2576,7 +2576,7 @@ body: |
25762576
; GFX908-FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD $vgpr1, 4, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.1 + 4, addrspace 5)
25772577
; GFX908-FLATSCR-NEXT: $agpr1 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec
25782578
; GFX908-FLATSCR-NEXT: $vgpr0 = SCRATCH_LOAD_DWORD $vgpr1, 8, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.1 + 8, addrspace 5)
2579-
; GFX908-FLATSCR-NEXT: $agpr2 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2
2579+
; GFX908-FLATSCR-NEXT: $agpr2 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr0, implicit $exec
25802580
; GFX908-FLATSCR-NEXT: $vgpr1 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.2, addrspace 5)
25812581
; GFX908-FLATSCR-NEXT: S_CBRANCH_SCC1 %bb.2, implicit $scc
25822582
; GFX908-FLATSCR-NEXT: {{ $}}
@@ -4115,9 +4115,9 @@ body: |
41154115
; GFX908-NEXT: S_CMP_EQ_U32 0, 0, implicit-def $scc
41164116
; GFX908-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr1, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, implicit $exec :: (store (s32) into %stack.2, addrspace 5)
41174117
; GFX908-NEXT: $vgpr1 = V_MOV_B32_e32 8200, implicit $exec
4118-
; GFX908-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 $agpr0, implicit $exec, implicit-def $agpr0_agpr1
4118+
; GFX908-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 $agpr0, implicit $exec, implicit-def $agpr0_agpr1, implicit $agpr0_agpr1
41194119
; GFX908-NEXT: BUFFER_STORE_DWORD_OFFEN $vgpr0, $vgpr1, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1 :: (store (s32) into %stack.1, addrspace 5)
4120-
; GFX908-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 $agpr1, implicit $exec
4120+
; GFX908-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 $agpr1, implicit $exec, implicit $agpr0_agpr1
41214121
; GFX908-NEXT: BUFFER_STORE_DWORD_OFFEN $vgpr0, $vgpr1, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, implicit $exec, implicit $agpr0_agpr1 :: (store (s32) into %stack.1 + 4, addrspace 5)
41224122
; GFX908-NEXT: $vgpr1 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, implicit $exec :: (load (s32) from %stack.2, addrspace 5)
41234123
; GFX908-NEXT: S_CBRANCH_SCC1 %bb.2, implicit $scc
@@ -4613,9 +4613,9 @@ body: |
46134613
; GFX908-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr1, $sgpr32, 0, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.2, addrspace 5)
46144614
; GFX908-FLATSCR-NEXT: $vgpr1 = V_MOV_B32_e32 $sgpr32, implicit $exec
46154615
; GFX908-FLATSCR-NEXT: $vgpr1 = V_ADD_U32_e32 8200, $vgpr1, implicit $exec
4616-
; GFX908-FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 $agpr0, implicit $exec, implicit-def $agpr0_agpr1
4616+
; GFX908-FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 $agpr0, implicit $exec, implicit-def $agpr0_agpr1, implicit $agpr0_agpr1
46174617
; GFX908-FLATSCR-NEXT: SCRATCH_STORE_DWORD $vgpr0, $vgpr1, 0, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1 :: (store (s32) into %stack.1, addrspace 5)
4618-
; GFX908-FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 $agpr1, implicit $exec
4618+
; GFX908-FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 $agpr1, implicit $exec, implicit $agpr0_agpr1
46194619
; GFX908-FLATSCR-NEXT: SCRATCH_STORE_DWORD $vgpr0, $vgpr1, 4, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1 :: (store (s32) into %stack.1 + 4, addrspace 5)
46204620
; GFX908-FLATSCR-NEXT: $vgpr1 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.2, addrspace 5)
46214621
; GFX908-FLATSCR-NEXT: S_CBRANCH_SCC1 %bb.2, implicit $scc
@@ -5136,11 +5136,11 @@ body: |
51365136
; GFX908-NEXT: S_CMP_EQ_U32 0, 0, implicit-def $scc
51375137
; GFX908-NEXT: BUFFER_STORE_DWORD_OFFSET killed $vgpr1, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, implicit $exec :: (store (s32) into %stack.2, addrspace 5)
51385138
; GFX908-NEXT: $vgpr1 = V_MOV_B32_e32 8200, implicit $exec
5139-
; GFX908-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 $agpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2
5139+
; GFX908-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 $agpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2, implicit $agpr0_agpr1_agpr2
51405140
; GFX908-NEXT: BUFFER_STORE_DWORD_OFFEN $vgpr0, $vgpr1, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2 :: (store (s32) into %stack.1, addrspace 5)
51415141
; GFX908-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 $agpr1, implicit $exec
51425142
; GFX908-NEXT: BUFFER_STORE_DWORD_OFFEN $vgpr0, $vgpr1, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4, 0, 0, implicit $exec :: (store (s32) into %stack.1 + 4, addrspace 5)
5143-
; GFX908-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 $agpr2, implicit $exec
5143+
; GFX908-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 $agpr2, implicit $exec, implicit $agpr0_agpr1_agpr2
51445144
; GFX908-NEXT: BUFFER_STORE_DWORD_OFFEN $vgpr0, $vgpr1, $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 8, 0, 0, implicit $exec, implicit $agpr0_agpr1_agpr2 :: (store (s32) into %stack.1 + 8, addrspace 5)
51455145
; GFX908-NEXT: $vgpr1 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, implicit $exec :: (load (s32) from %stack.2, addrspace 5)
51465146
; GFX908-NEXT: S_CBRANCH_SCC1 %bb.2, implicit $scc
@@ -5637,11 +5637,11 @@ body: |
56375637
; GFX908-FLATSCR-NEXT: SCRATCH_STORE_DWORD_SADDR killed $vgpr1, $sgpr32, 0, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.2, addrspace 5)
56385638
; GFX908-FLATSCR-NEXT: $vgpr1 = V_MOV_B32_e32 $sgpr32, implicit $exec
56395639
; GFX908-FLATSCR-NEXT: $vgpr1 = V_ADD_U32_e32 8200, $vgpr1, implicit $exec
5640-
; GFX908-FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 $agpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2
5640+
; GFX908-FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 $agpr0, implicit $exec, implicit-def $agpr0_agpr1_agpr2, implicit $agpr0_agpr1_agpr2
56415641
; GFX908-FLATSCR-NEXT: SCRATCH_STORE_DWORD $vgpr0, $vgpr1, 0, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2 :: (store (s32) into %stack.1, addrspace 5)
56425642
; GFX908-FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 $agpr1, implicit $exec
56435643
; GFX908-FLATSCR-NEXT: SCRATCH_STORE_DWORD $vgpr0, $vgpr1, 4, 0, implicit $exec, implicit $flat_scr :: (store (s32) into %stack.1 + 4, addrspace 5)
5644-
; GFX908-FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 $agpr2, implicit $exec
5644+
; GFX908-FLATSCR-NEXT: $vgpr0 = V_ACCVGPR_READ_B32_e64 $agpr2, implicit $exec, implicit $agpr0_agpr1_agpr2
56455645
; GFX908-FLATSCR-NEXT: SCRATCH_STORE_DWORD $vgpr0, $vgpr1, 8, 0, implicit $exec, implicit $flat_scr, implicit $agpr0_agpr1_agpr2 :: (store (s32) into %stack.1 + 8, addrspace 5)
56465646
; GFX908-FLATSCR-NEXT: $vgpr1 = SCRATCH_LOAD_DWORD_SADDR $sgpr32, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %stack.2, addrspace 5)
56475647
; GFX908-FLATSCR-NEXT: S_CBRANCH_SCC1 %bb.2, implicit $scc

‎llvm/test/CodeGen/AMDGPU/agpr-spill-copy.mir

Lines changed: 43 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,43 @@
1+
# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 5
2+
# RUN: llc -mtriple=amdgcn--amdhsa -mcpu=gfx942 -run-pass prologepilog,machine-cp -o - %s | FileCheck -check-prefix=GFX942 %s
3+
4+
--- |
5+
define amdgpu_kernel void @agpr_spill_copy() #0 { ret void }
6+
7+
attributes #0 = { "amdgpu-waves-per-eu"="8,8" }
8+
---
9+
name: agpr_spill_copy
10+
tracksRegLiveness: true
11+
stack:
12+
- { id: 0, name: '', type: spill-slot, offset: 0, size: 64, alignment: 4 }
13+
machineFunctionInfo:
14+
stackPtrOffsetReg: '$sgpr32'
15+
hasSpilledVGPRs: true
16+
body: |
17+
bb.0:
18+
; GFX942-LABEL: name: agpr_spill_copy
19+
; GFX942: liveins: $agpr30, $agpr31
20+
; GFX942-NEXT: {{ $}}
21+
; GFX942-NEXT: renamable $agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27 = IMPLICIT_DEF
22+
; GFX942-NEXT: renamable $agpr28_agpr29 = IMPLICIT_DEF
23+
; GFX942-NEXT: renamable $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 = IMPLICIT_DEF
24+
; GFX942-NEXT: SCRATCH_STORE_DWORDX4_SADDR killed $vgpr0_vgpr1_vgpr2_vgpr3, $sgpr32, 0, 0, implicit $exec, implicit $flat_scr, implicit-def $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, implicit $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 :: (store (s128) into %stack.0, align 4, addrspace 5)
25+
; GFX942-NEXT: SCRATCH_STORE_DWORDX4_SADDR killed $vgpr4_vgpr5_vgpr6_vgpr7, $sgpr32, 16, 0, implicit $exec, implicit $flat_scr :: (store (s128) into %stack.0 + 16, align 4, addrspace 5)
26+
; GFX942-NEXT: SCRATCH_STORE_DWORDX4_SADDR killed $vgpr8_vgpr9_vgpr10_vgpr11, $sgpr32, 32, 0, implicit $exec, implicit $flat_scr :: (store (s128) into %stack.0 + 32, align 4, addrspace 5)
27+
; GFX942-NEXT: $agpr30 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr15, implicit $exec, implicit $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15
28+
; GFX942-NEXT: $agpr31 = V_ACCVGPR_WRITE_B32_e64 killed $vgpr14, implicit $exec, implicit $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15
29+
; GFX942-NEXT: SCRATCH_STORE_DWORDX2_SADDR killed $vgpr12_vgpr13, $sgpr32, 48, 0, implicit $exec, implicit $flat_scr, implicit killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 :: (store (s64) into %stack.0 + 48, align 4, addrspace 5)
30+
; GFX942-NEXT: $agpr0_agpr1_agpr2_agpr3 = SCRATCH_LOAD_DWORDX4_SADDR $sgpr32, 0, 0, implicit $exec, implicit $flat_scr, implicit-def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 :: (load (s128) from %stack.0, align 4, addrspace 5)
31+
; GFX942-NEXT: $agpr4_agpr5_agpr6_agpr7 = SCRATCH_LOAD_DWORDX4_SADDR $sgpr32, 16, 0, implicit $exec, implicit $flat_scr :: (load (s128) from %stack.0 + 16, align 4, addrspace 5)
32+
; GFX942-NEXT: $agpr8_agpr9_agpr10_agpr11 = SCRATCH_LOAD_DWORDX4_SADDR $sgpr32, 32, 0, implicit $exec, implicit $flat_scr :: (load (s128) from %stack.0 + 32, align 4, addrspace 5)
33+
; GFX942-NEXT: $agpr15 = COPY $agpr30, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
34+
; GFX942-NEXT: $agpr14 = COPY $agpr31, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
35+
; GFX942-NEXT: $agpr12_agpr13 = SCRATCH_LOAD_DWORDX2_SADDR $sgpr32, 48, 0, implicit $exec, implicit $flat_scr :: (load (s64) from %stack.0 + 48, align 4, addrspace 5)
36+
; GFX942-NEXT: S_ENDPGM 0, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
37+
renamable $agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27 = IMPLICIT_DEF
38+
renamable $agpr28_agpr29 = IMPLICIT_DEF
39+
renamable $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 = IMPLICIT_DEF
40+
SI_SPILL_AV512_SAVE killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, %stack.0, $sgpr32, 0, implicit $exec :: (store (s512) into %stack.0, align 4, addrspace 5)
41+
renamable $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15 = SI_SPILL_AV512_RESTORE %stack.0, $sgpr32, 0, implicit $exec :: (load (s512) from %stack.0, align 4, addrspace 5)
42+
S_ENDPGM 0, implicit $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15
43+
...

0 commit comments

Comments
 (0)