This is an archive of the discontinued LLVM Phabricator instance.

Differential D154816

[AMDGPU] Update test
Needs ReviewPublic

Authored by piotr on Jul 10 2023, 12:48 AM.
This revision needs review, but there are no reviewers specified.

Details

Reviewers
None
Summary

Update test/CodeGen/AMDGPU/remat-smrd.mir:

  • Convert a negative case of non-dereferenceable invariant load to positive one.
  • Add new cases for subreg.

Diff Detail

Event Timeline

piotr created this revision.Jul 10 2023, 12:48 AM
Herald added a project: Restricted Project. · View Herald TranscriptJul 10 2023, 12:48 AM
Herald added subscribers: kerbowa, tpr, dstuttard and 3 others. · View Herald Transcript
piotr requested review of this revision.Jul 10 2023, 12:48 AM
Herald added a project: Restricted Project. · View Herald TranscriptJul 10 2023, 12:48 AM
Herald added subscribers: llvm-commits, wdng. · View Herald Transcript
Harbormaster completed remote builds in B244062: Diff 538532.Jul 10 2023, 1:44 AM
piotr mentioned this in D154083: [AMDGPU] Rematerialize scalar loads.Jul 12 2023, 7:52 AM
piotr updated this revision to Diff 542231.Jul 19 2023, 4:05 PM

Added more tests.

Harbormaster completed remote builds in B246705: Diff 542231.Jul 19 2023, 6:41 PM
piotr updated this revision to Diff 542783.Jul 21 2023, 12:36 AM

Added new case with the same reg used for use and def: %0.sub0_sub1:sgpr_256 = S_LOAD_DWORDX2_IMM %0.sub0_sub1:sgpr_256.

piotr edited the summary of this revision. (Show Details)Jul 21 2023, 12:36 AM
Harbormaster completed remote builds in B247119: Diff 542783.Jul 21 2023, 3:21 AM

Revision Contents

PathSize
llvm/
test/
CodeGen/
AMDGPU/
192 lines

Diff 542783

llvm/test/CodeGen/AMDGPU/remat-smrd.mir

Show First 20 Lines • Show All 160 Lines • ▼ Show 20 Lines bb.0:
%3:sgpr_512 = S_LOAD_DWORDX16_IMM %0, 8, 0 :: (invariant dereferenceable load (s512), addrspace 4) %3:sgpr_512 = S_LOAD_DWORDX16_IMM %0, 8, 0 :: (invariant dereferenceable load (s512), addrspace 4)
S_NOP 0, implicit %1 S_NOP 0, implicit %1
S_NOP 0, implicit %2 S_NOP 0, implicit %2
S_NOP 0, implicit %3 S_NOP 0, implicit %3
S_ENDPGM 0, implicit %0 S_ENDPGM 0, implicit %0
... ...
--- ---
name: test_remat_s_load_dword_immx16_subreg
tracksRegLiveness: true
body: |
bb.0:
liveins: $sgpr8_sgpr9, $vgpr0_vgpr1
; GCN-LABEL: name: test_remat_s_load_dword_immx16_subreg
; GCN: liveins: $sgpr8_sgpr9, $vgpr0_vgpr1
; GCN-NEXT: {{ $}}
; GCN-NEXT: renamable $sgpr0_sgpr1 = COPY $sgpr8_sgpr9
; GCN-NEXT: renamable $sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19 = S_LOAD_DWORDX16_IMM renamable $sgpr0_sgpr1, 0, 0 :: (dereferenceable invariant load (s512), align 4, addrspace 4)
; GCN-NEXT: SI_SPILL_S512_SAVE killed renamable $sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19, %stack.0, implicit $exec, implicit $sp_reg :: (store (s512) into %stack.0, align 4, addrspace 5)
; GCN-NEXT: renamable $sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19 = S_LOAD_DWORDX16_IMM renamable $sgpr0_sgpr1, 128, 0 :: (dereferenceable invariant load (s512), align 4, addrspace 4)
; GCN-NEXT: renamable $sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19 = SI_SPILL_S512_RESTORE %stack.0, implicit $exec, implicit $sp_reg :: (load (s512) from %stack.0, align 4, addrspace 5)
; GCN-NEXT: S_NOP 0, implicit killed renamable $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19
; GCN-NEXT: renamable $sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19 = S_LOAD_DWORDX16_IMM renamable $sgpr0_sgpr1, 128, 0 :: (dereferenceable invariant load (s512), align 4, addrspace 4)
; GCN-NEXT: renamable $sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19 = SI_SPILL_S512_RESTORE %stack.0, implicit $exec, implicit $sp_reg :: (load (s512) from %stack.0, align 4, addrspace 5)
; GCN-NEXT: S_NOP 0, implicit killed renamable $sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11
; GCN-NEXT: S_ENDPGM 0, implicit killed renamable $sgpr0_sgpr1
%0:sreg_64_xexec = COPY $sgpr8_sgpr9
%1:sgpr_512 = S_LOAD_DWORDX16_IMM %0, 0, 0 :: (invariant dereferenceable load (s512), align 4, addrspace 4)
%2:sgpr_512 = S_LOAD_DWORDX16_IMM %0, 128, 0 :: (invariant dereferenceable load (s512), align 4, addrspace 4)
%2.sub8_sub9_sub10_sub11_sub12_sub13_sub14_sub15:sgpr_512 = COPY %1.sub8_sub9_sub10_sub11_sub12_sub13_sub14_sub15:sgpr_512
S_NOP 0, implicit %2.sub8_sub9_sub10_sub11_sub12_sub13_sub14_sub15
%3:sgpr_512 = S_LOAD_DWORDX16_IMM %0, 128, 0 :: (invariant dereferenceable load (s512), align 4, addrspace 4)
%3.sub0_sub1_sub2_sub3_sub4_sub5_sub6_sub7:sgpr_512 = COPY %1.sub0_sub1_sub2_sub3_sub4_sub5_sub6_sub7:sgpr_512
S_NOP 0, implicit %3.sub0_sub1_sub2_sub3_sub4_sub5_sub6_sub7:sgpr_512
S_ENDPGM 0, implicit %0
...
---
name: test_remat_s_load_dword_immx16_subreg_no_shrinking_double_use
tracksRegLiveness: true
body: |
bb.0:
liveins: $sgpr8_sgpr9, $vgpr0_vgpr1
; GCN-LABEL: name: test_remat_s_load_dword_immx16_subreg_no_shrinking_double_use
; GCN: liveins: $sgpr8_sgpr9, $vgpr0_vgpr1
; GCN-NEXT: {{ $}}
; GCN-NEXT: renamable $sgpr0_sgpr1 = COPY $sgpr8_sgpr9
; GCN-NEXT: renamable $sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19 = S_LOAD_DWORDX16_IMM renamable $sgpr0_sgpr1, 0, 0 :: (dereferenceable invariant load (s512), align 4, addrspace 4)
; GCN-NEXT: SI_SPILL_S512_SAVE killed renamable $sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19, %stack.0, implicit $exec, implicit $sp_reg :: (store (s512) into %stack.0, align 4, addrspace 5)
; GCN-NEXT: renamable $sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19 = S_LOAD_DWORDX16_IMM renamable $sgpr0_sgpr1, 128, 0 :: (dereferenceable invariant load (s512), align 4, addrspace 4)
; GCN-NEXT: renamable $sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19 = SI_SPILL_S512_RESTORE %stack.0, implicit $exec, implicit $sp_reg :: (load (s512) from %stack.0, align 4, addrspace 5)
; GCN-NEXT: renamable $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19 = KILL killed renamable $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19, implicit renamable $sgpr4_sgpr5_sgpr6_sgpr7
; GCN-NEXT: S_NOP 0, implicit killed renamable $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19
; GCN-NEXT: renamable $sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19 = S_LOAD_DWORDX16_IMM renamable $sgpr0_sgpr1, 128, 0 :: (dereferenceable invariant load (s512), align 4, addrspace 4)
; GCN-NEXT: renamable $sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19 = SI_SPILL_S512_RESTORE %stack.0, implicit $exec, implicit $sp_reg :: (load (s512) from %stack.0, align 4, addrspace 5)
; GCN-NEXT: renamable $sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11 = KILL killed renamable $sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11, implicit renamable $sgpr0_sgpr1
; GCN-NEXT: S_NOP 0, implicit killed renamable $sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11
; GCN-NEXT: S_ENDPGM 0, implicit killed renamable $sgpr0_sgpr1
%0:sreg_64_xexec = COPY $sgpr8_sgpr9
%1:sgpr_512 = S_LOAD_DWORDX16_IMM %0, 0, 0 :: (invariant dereferenceable load (s512), align 4, addrspace 4)
%2:sgpr_512 = S_LOAD_DWORDX16_IMM %0, 128, 0 :: (invariant dereferenceable load (s512), align 4, addrspace 4)
%2.sub8_sub9_sub10_sub11_sub12_sub13_sub14_sub15:sgpr_512 = COPY %1.sub8_sub9_sub10_sub11_sub12_sub13_sub14_sub15:sgpr_512, implicit %1.sub0_sub1_sub2_sub3:sgpr_512
S_NOP 0, implicit %2.sub8_sub9_sub10_sub11_sub12_sub13_sub14_sub15
%3:sgpr_512 = S_LOAD_DWORDX16_IMM %0, 128, 0 :: (invariant dereferenceable load (s512), align 4, addrspace 4)
%3.sub0_sub1_sub2_sub3_sub4_sub5_sub6_sub7:sgpr_512 = COPY %1.sub0_sub1_sub2_sub3_sub4_sub5_sub6_sub7:sgpr_512, implicit %0
S_NOP 0, implicit %3.sub0_sub1_sub2_sub3_sub4_sub5_sub6_sub7:sgpr_512
S_ENDPGM 0, implicit %0
...
---
name: test_remat_s_load_dword_immx16_subreg_no_shrinking_bundle
tracksRegLiveness: true
body: |
bb.0:
liveins: $sgpr8_sgpr9, $vgpr0_vgpr1
; GCN-LABEL: name: test_remat_s_load_dword_immx16_subreg_no_shrinking_bundle
; GCN: liveins: $sgpr8_sgpr9, $vgpr0_vgpr1
; GCN-NEXT: {{ $}}
; GCN-NEXT: renamable $sgpr0_sgpr1 = COPY $sgpr8_sgpr9
; GCN-NEXT: renamable $sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19 = S_LOAD_DWORDX16_IMM renamable $sgpr0_sgpr1, 0, 0 :: (dereferenceable invariant load (s512), align 4, addrspace 4)
; GCN-NEXT: SI_SPILL_S512_SAVE killed renamable $sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19, %stack.0, implicit $exec, implicit $sp_reg :: (store (s512) into %stack.0, align 4, addrspace 5)
; GCN-NEXT: renamable $sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19 = S_LOAD_DWORDX16_IMM renamable $sgpr0_sgpr1, 128, 0 :: (dereferenceable invariant load (s512), align 4, addrspace 4)
; GCN-NEXT: renamable $sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19 = SI_SPILL_S512_RESTORE %stack.0, implicit $exec, implicit $sp_reg :: (load (s512) from %stack.0, align 4, addrspace 5)
; GCN-NEXT: S_NOP 0, implicit killed renamable $sgpr12_sgpr13_sgpr14_sgpr15_sgpr16_sgpr17_sgpr18_sgpr19
; GCN-NEXT: S_ENDPGM 0, implicit killed renamable $sgpr0_sgpr1
%0:sreg_64_xexec = COPY $sgpr8_sgpr9
%1:sgpr_512 = S_LOAD_DWORDX16_IMM %0, 0, 0 :: (invariant dereferenceable load (s512), align 4, addrspace 4)
%2:sgpr_512 = S_LOAD_DWORDX16_IMM %0, 128, 0 :: (invariant dereferenceable load (s512), align 4, addrspace 4)
%2.sub8_sub9_sub10_sub11_sub12_sub13_sub14_sub15:sgpr_512 = COPY %1.sub8_sub9_sub10_sub11_sub12_sub13_sub14_sub15:sgpr_512 {
internal %2.sub8_sub9_sub10_sub11_sub12_sub13_sub14_sub15:sgpr_512 = COPY %1.sub8_sub9_sub10_sub11_sub12_sub13_sub14_sub15:sgpr_512
}
S_NOP 0, implicit %2.sub8_sub9_sub10_sub11_sub12_sub13_sub14_sub15
S_ENDPGM 0, implicit %0
...
---
name: test_remat_s_load_dword_immx8_subreg
tracksRegLiveness: true
body: |
bb.0:
liveins: $sgpr8_sgpr9, $vgpr0_vgpr1
; GCN-LABEL: name: test_remat_s_load_dword_immx8_subreg
; GCN: liveins: $sgpr8_sgpr9, $vgpr0_vgpr1
; GCN-NEXT: {{ $}}
; GCN-NEXT: renamable $sgpr0_sgpr1 = COPY $sgpr8_sgpr9
; GCN-NEXT: renamable $sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11 = S_LOAD_DWORDX8_IMM renamable $sgpr0_sgpr1, 0, 0 :: (dereferenceable invariant load (s256), align 4, addrspace 4)
; GCN-NEXT: SI_SPILL_S256_SAVE killed renamable $sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11, %stack.0, implicit $exec, implicit $sp_reg :: (store (s256) into %stack.0, align 4, addrspace 5)
; GCN-NEXT: renamable $sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11 = S_LOAD_DWORDX8_IMM renamable $sgpr0_sgpr1, 128, 0 :: (dereferenceable invariant load (s256), align 4, addrspace 4)
; GCN-NEXT: renamable $sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11 = SI_SPILL_S256_RESTORE %stack.0, implicit $exec, implicit $sp_reg :: (load (s256) from %stack.0, align 4, addrspace 5)
; GCN-NEXT: S_NOP 0, implicit killed renamable $sgpr8_sgpr9_sgpr10_sgpr11
; GCN-NEXT: renamable $sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11 = S_LOAD_DWORDX8_IMM renamable $sgpr0_sgpr1, 128, 0 :: (dereferenceable invariant load (s256), align 4, addrspace 4)
; GCN-NEXT: renamable $sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11 = SI_SPILL_S256_RESTORE %stack.0, implicit $exec, implicit $sp_reg :: (load (s256) from %stack.0, align 4, addrspace 5)
; GCN-NEXT: S_NOP 0, implicit killed renamable $sgpr4_sgpr5_sgpr6_sgpr7
; GCN-NEXT: S_ENDPGM 0, implicit killed renamable $sgpr0_sgpr1
%0:sreg_64_xexec = COPY $sgpr8_sgpr9
%1:sgpr_256 = S_LOAD_DWORDX8_IMM %0, 0, 0 :: (invariant dereferenceable load (s256), align 4, addrspace 4)
%2:sgpr_256 = S_LOAD_DWORDX8_IMM %0, 128, 0 :: (invariant dereferenceable load (s256), align 4, addrspace 4)
%2.sub4_sub5_sub6_sub7:sgpr_256 = COPY %1.sub4_sub5_sub6_sub7:sgpr_256
S_NOP 0, implicit %2.sub4_sub5_sub6_sub7
%3:sgpr_256 = S_LOAD_DWORDX8_IMM %0, 128, 0 :: (invariant dereferenceable load (s256), align 4, addrspace 4)
%3.sub0_sub1_sub2_sub3:sgpr_256 = COPY %1.sub0_sub1_sub2_sub3:sgpr_256
S_NOP 0, implicit %3.sub0_sub1_sub2_sub3:sgpr_256
S_ENDPGM 0, implicit %0
...
---
name: test_remat_s_get_waveid_in_workgroup name: test_remat_s_get_waveid_in_workgroup
tracksRegLiveness: true tracksRegLiveness: true
body: | body: |
bb.0: bb.0:
liveins: $sgpr8_sgpr9 liveins: $sgpr8_sgpr9
; GCN-LABEL: name: test_remat_s_get_waveid_in_workgroup ; GCN-LABEL: name: test_remat_s_get_waveid_in_workgroup
; GCN: liveins: $sgpr8_sgpr9 ; GCN: liveins: $sgpr8_sgpr9
; GCN-NEXT: {{ $}} ; GCN-NEXT: {{ $}}
▲ Show 20 Lines • Show All 178 Lines • ▼ Show 20 Lines bb.0:
%2:sreg_32_xm0_xexec = S_SCRATCH_LOAD_DWORD_IMM %0, 4, 0, implicit $flat_scr :: (invariant dereferenceable load (s32), addrspace 4) %2:sreg_32_xm0_xexec = S_SCRATCH_LOAD_DWORD_IMM %0, 4, 0, implicit $flat_scr :: (invariant dereferenceable load (s32), addrspace 4)
%3:sreg_32_xm0_xexec = S_SCRATCH_LOAD_DWORD_IMM %0, 8, 0, implicit $flat_scr :: (invariant dereferenceable load (s32), addrspace 4) %3:sreg_32_xm0_xexec = S_SCRATCH_LOAD_DWORD_IMM %0, 8, 0, implicit $flat_scr :: (invariant dereferenceable load (s32), addrspace 4)
S_NOP 0, implicit %1 S_NOP 0, implicit %1
S_NOP 0, implicit %2 S_NOP 0, implicit %2
S_NOP 0, implicit %3 S_NOP 0, implicit %3
S_ENDPGM 0, implicit %0 S_ENDPGM 0, implicit %0
... ...
#---------------------------------------------------------------------
# Negative tests
#---------------------------------------------------------------------
--- ---
name: test_no_remat_s_load_dword_not_invariant name: test_remat_s_load_dword_invariant_not_dereferenceable
tracksRegLiveness: true tracksRegLiveness: true
body: | body: |
bb.0: bb.0:
liveins: $sgpr8_sgpr9 liveins: $sgpr8_sgpr9
; GCN-LABEL: name: test_no_remat_s_load_dword_not_invariant ; GCN-LABEL: name: test_remat_s_load_dword_invariant_not_dereferenceable
; GCN: liveins: $sgpr8_sgpr9 ; GCN: liveins: $sgpr8_sgpr9
; GCN-NEXT: {{ $}} ; GCN-NEXT: {{ $}}
; GCN-NEXT: renamable $sgpr2_sgpr3 = COPY $sgpr8_sgpr9 ; GCN-NEXT: renamable $sgpr2_sgpr3 = COPY $sgpr8_sgpr9
; GCN-NEXT: renamable $sgpr0 = S_LOAD_DWORD_IMM renamable $sgpr2_sgpr3, 0, 0 :: (dereferenceable load (s32), addrspace 4) ; GCN-NEXT: renamable $sgpr0 = S_LOAD_DWORD_IMM renamable $sgpr2_sgpr3, 0, 0 :: (invariant load (s32), addrspace 4)
; GCN-NEXT: SI_SPILL_S32_SAVE killed renamable $sgpr0, %stack.0, implicit $exec, implicit $sp_reg :: (store (s32) into %stack.0, addrspace 5) ; GCN-NEXT: SI_SPILL_S32_SAVE killed renamable $sgpr0, %stack.0, implicit $exec, implicit $sp_reg :: (store (s32) into %stack.0, addrspace 5)
; GCN-NEXT: renamable $sgpr1 = S_LOAD_DWORD_IMM renamable $sgpr2_sgpr3, 4, 0 :: (dereferenceable load (s32), addrspace 4) ; GCN-NEXT: renamable $sgpr1 = S_LOAD_DWORD_IMM renamable $sgpr2_sgpr3, 4, 0 :: (invariant load (s32), addrspace 4)
; GCN-NEXT: renamable $sgpr0 = S_LOAD_DWORD_IMM renamable $sgpr2_sgpr3, 8, 0 :: (dereferenceable load (s32), addrspace 4) ; GCN-NEXT: renamable $sgpr0 = S_LOAD_DWORD_IMM renamable $sgpr2_sgpr3, 8, 0 :: (invariant load (s32), addrspace 4)
; GCN-NEXT: SI_SPILL_S32_SAVE killed renamable $sgpr0, %stack.1, implicit $exec, implicit $sp_reg :: (store (s32) into %stack.1, addrspace 5) ; GCN-NEXT: SI_SPILL_S32_SAVE killed renamable $sgpr0, %stack.1, implicit $exec, implicit $sp_reg :: (store (s32) into %stack.1, addrspace 5)
; GCN-NEXT: renamable $sgpr0 = SI_SPILL_S32_RESTORE %stack.0, implicit $exec, implicit $sp_reg :: (load (s32) from %stack.0, addrspace 5) ; GCN-NEXT: renamable $sgpr0 = SI_SPILL_S32_RESTORE %stack.0, implicit $exec, implicit $sp_reg :: (load (s32) from %stack.0, addrspace 5)
; GCN-NEXT: S_NOP 0, implicit killed renamable $sgpr0 ; GCN-NEXT: S_NOP 0, implicit killed renamable $sgpr0
; GCN-NEXT: S_NOP 0, implicit killed renamable $sgpr1 ; GCN-NEXT: S_NOP 0, implicit killed renamable $sgpr1
; GCN-NEXT: renamable $sgpr0 = SI_SPILL_S32_RESTORE %stack.1, implicit $exec, implicit $sp_reg :: (load (s32) from %stack.1, addrspace 5) ; GCN-NEXT: renamable $sgpr0 = SI_SPILL_S32_RESTORE %stack.1, implicit $exec, implicit $sp_reg :: (load (s32) from %stack.1, addrspace 5)
; GCN-NEXT: S_NOP 0, implicit killed renamable $sgpr0 ; GCN-NEXT: S_NOP 0, implicit killed renamable $sgpr0
; GCN-NEXT: S_ENDPGM 0, implicit killed renamable $sgpr2_sgpr3 ; GCN-NEXT: S_ENDPGM 0, implicit killed renamable $sgpr2_sgpr3
%0:sreg_64_xexec = COPY $sgpr8_sgpr9 %0:sreg_64_xexec = COPY $sgpr8_sgpr9
%1:sreg_32_xm0_xexec = S_LOAD_DWORD_IMM %0, 0, 0 :: (dereferenceable load (s32), addrspace 4) %1:sreg_32_xm0_xexec = S_LOAD_DWORD_IMM %0, 0, 0 :: (invariant load (s32), addrspace 4)
%2:sreg_32_xm0_xexec = S_LOAD_DWORD_IMM %0, 4, 0 :: (dereferenceable load (s32), addrspace 4) %2:sreg_32_xm0_xexec = S_LOAD_DWORD_IMM %0, 4, 0 :: (invariant load (s32), addrspace 4)
%3:sreg_32_xm0_xexec = S_LOAD_DWORD_IMM %0, 8, 0 :: (dereferenceable load (s32), addrspace 4) %3:sreg_32_xm0_xexec = S_LOAD_DWORD_IMM %0, 8, 0 :: (invariant load (s32), addrspace 4)
S_NOP 0, implicit %1 S_NOP 0, implicit %1
S_NOP 0, implicit %2 S_NOP 0, implicit %2
S_NOP 0, implicit %3 S_NOP 0, implicit %3
S_ENDPGM 0, implicit %0 S_ENDPGM 0, implicit %0
... ...
#---------------------------------------------------------------------
# Negative tests
#---------------------------------------------------------------------
--- ---
name: test_no_remat_s_load_dword_not_dereferenceable name: test_no_remat_s_load_dword_not_invariant
tracksRegLiveness: true tracksRegLiveness: true
body: | body: |
bb.0: bb.0:
liveins: $sgpr8_sgpr9 liveins: $sgpr8_sgpr9
; GCN-LABEL: name: test_no_remat_s_load_dword_not_dereferenceable ; GCN-LABEL: name: test_no_remat_s_load_dword_not_invariant
; GCN: liveins: $sgpr8_sgpr9 ; GCN: liveins: $sgpr8_sgpr9
; GCN-NEXT: {{ $}} ; GCN-NEXT: {{ $}}
; GCN-NEXT: renamable $sgpr2_sgpr3 = COPY $sgpr8_sgpr9 ; GCN-NEXT: renamable $sgpr2_sgpr3 = COPY $sgpr8_sgpr9
; GCN-NEXT: renamable $sgpr0 = S_LOAD_DWORD_IMM renamable $sgpr2_sgpr3, 0, 0 :: (invariant load (s32), addrspace 4) ; GCN-NEXT: renamable $sgpr0 = S_LOAD_DWORD_IMM renamable $sgpr2_sgpr3, 0, 0 :: (dereferenceable load (s32), addrspace 4)
; GCN-NEXT: SI_SPILL_S32_SAVE killed renamable $sgpr0, %stack.0, implicit $exec, implicit $sp_reg :: (store (s32) into %stack.0, addrspace 5) ; GCN-NEXT: SI_SPILL_S32_SAVE killed renamable $sgpr0, %stack.0, implicit $exec, implicit $sp_reg :: (store (s32) into %stack.0, addrspace 5)
; GCN-NEXT: renamable $sgpr1 = S_LOAD_DWORD_IMM renamable $sgpr2_sgpr3, 4, 0 :: (invariant load (s32), addrspace 4) ; GCN-NEXT: renamable $sgpr1 = S_LOAD_DWORD_IMM renamable $sgpr2_sgpr3, 4, 0 :: (dereferenceable load (s32), addrspace 4)
; GCN-NEXT: renamable $sgpr0 = S_LOAD_DWORD_IMM renamable $sgpr2_sgpr3, 8, 0 :: (invariant load (s32), addrspace 4) ; GCN-NEXT: renamable $sgpr0 = S_LOAD_DWORD_IMM renamable $sgpr2_sgpr3, 8, 0 :: (dereferenceable load (s32), addrspace 4)
; GCN-NEXT: SI_SPILL_S32_SAVE killed renamable $sgpr0, %stack.1, implicit $exec, implicit $sp_reg :: (store (s32) into %stack.1, addrspace 5) ; GCN-NEXT: SI_SPILL_S32_SAVE killed renamable $sgpr0, %stack.1, implicit $exec, implicit $sp_reg :: (store (s32) into %stack.1, addrspace 5)
; GCN-NEXT: renamable $sgpr0 = SI_SPILL_S32_RESTORE %stack.0, implicit $exec, implicit $sp_reg :: (load (s32) from %stack.0, addrspace 5) ; GCN-NEXT: renamable $sgpr0 = SI_SPILL_S32_RESTORE %stack.0, implicit $exec, implicit $sp_reg :: (load (s32) from %stack.0, addrspace 5)
; GCN-NEXT: S_NOP 0, implicit killed renamable $sgpr0 ; GCN-NEXT: S_NOP 0, implicit killed renamable $sgpr0
; GCN-NEXT: S_NOP 0, implicit killed renamable $sgpr1 ; GCN-NEXT: S_NOP 0, implicit killed renamable $sgpr1
; GCN-NEXT: renamable $sgpr0 = SI_SPILL_S32_RESTORE %stack.1, implicit $exec, implicit $sp_reg :: (load (s32) from %stack.1, addrspace 5) ; GCN-NEXT: renamable $sgpr0 = SI_SPILL_S32_RESTORE %stack.1, implicit $exec, implicit $sp_reg :: (load (s32) from %stack.1, addrspace 5)
; GCN-NEXT: S_NOP 0, implicit killed renamable $sgpr0 ; GCN-NEXT: S_NOP 0, implicit killed renamable $sgpr0
; GCN-NEXT: S_ENDPGM 0, implicit killed renamable $sgpr2_sgpr3 ; GCN-NEXT: S_ENDPGM 0, implicit killed renamable $sgpr2_sgpr3
%0:sreg_64_xexec = COPY $sgpr8_sgpr9 %0:sreg_64_xexec = COPY $sgpr8_sgpr9
%1:sreg_32_xm0_xexec = S_LOAD_DWORD_IMM %0, 0, 0 :: (invariant load (s32), addrspace 4) %1:sreg_32_xm0_xexec = S_LOAD_DWORD_IMM %0, 0, 0 :: (dereferenceable load (s32), addrspace 4)
%2:sreg_32_xm0_xexec = S_LOAD_DWORD_IMM %0, 4, 0 :: (invariant load (s32), addrspace 4) %2:sreg_32_xm0_xexec = S_LOAD_DWORD_IMM %0, 4, 0 :: (dereferenceable load (s32), addrspace 4)
%3:sreg_32_xm0_xexec = S_LOAD_DWORD_IMM %0, 8, 0 :: (invariant load (s32), addrspace 4) %3:sreg_32_xm0_xexec = S_LOAD_DWORD_IMM %0, 8, 0 :: (dereferenceable load (s32), addrspace 4)
S_NOP 0, implicit %1 S_NOP 0, implicit %1
S_NOP 0, implicit %2 S_NOP 0, implicit %2
S_NOP 0, implicit %3 S_NOP 0, implicit %3
S_ENDPGM 0, implicit %0 S_ENDPGM 0, implicit %0
... ...
--- ---
name: test_no_remat_s_memtime name: test_no_remat_s_memtime
▲ Show 20 Lines • Show All 53 Lines • ▼ Show 20 Lines bb.0:
%1:sreg_64_xexec = S_MEMREALTIME %1:sreg_64_xexec = S_MEMREALTIME
%2:sreg_64_xexec = S_MEMREALTIME %2:sreg_64_xexec = S_MEMREALTIME
%3:sreg_64_xexec = S_MEMREALTIME %3:sreg_64_xexec = S_MEMREALTIME
S_NOP 0, implicit %1 S_NOP 0, implicit %1
S_NOP 0, implicit %2 S_NOP 0, implicit %2
S_NOP 0, implicit %3 S_NOP 0, implicit %3
S_ENDPGM 0, implicit %0 S_ENDPGM 0, implicit %0
... ...
---
name: test_no_remat_s_load_dword_immx2_tied
tracksRegLiveness: true
body: |
bb.0:
liveins: $sgpr8_sgpr9
; GCN-LABEL: name: test_no_remat_s_load_dword_immx2_tied
; GCN: liveins: $sgpr8_sgpr9
; GCN-NEXT: {{ $}}
; GCN-NEXT: renamable $sgpr0_sgpr1 = COPY $sgpr8_sgpr9
; GCN-NEXT: renamable $sgpr2_sgpr3 = S_LOAD_DWORDX2_IMM renamable $sgpr0_sgpr1, 0, 0 :: (dereferenceable invariant load (s64), addrspace 4)
; GCN-NEXT: SI_SPILL_S64_SAVE killed renamable $sgpr2_sgpr3, %stack.1, implicit $exec, implicit $sp_reg :: (store (s64) into %stack.1, align 4, addrspace 5)
; GCN-NEXT: renamable $sgpr2_sgpr3 = S_LOAD_DWORDX2_IMM renamable $sgpr0_sgpr1, 4, 0 :: (dereferenceable invariant load (s64), addrspace 4)
; GCN-NEXT: SI_SPILL_S64_SAVE killed renamable $sgpr2_sgpr3, %stack.0, implicit $exec, implicit $sp_reg :: (store (s64) into %stack.0, align 4, addrspace 5)
; GCN-NEXT: renamable $sgpr0_sgpr1 = S_LOAD_DWORDX2_IMM renamable $sgpr0_sgpr1, 8, 0 :: (dereferenceable invariant load (s64), addrspace 4)
; GCN-NEXT: renamable $sgpr2_sgpr3 = SI_SPILL_S64_RESTORE %stack.1, implicit $exec, implicit $sp_reg :: (load (s64) from %stack.1, align 4, addrspace 5)
; GCN-NEXT: S_NOP 0, implicit killed renamable $sgpr2_sgpr3
; GCN-NEXT: renamable $sgpr2_sgpr3 = SI_SPILL_S64_RESTORE %stack.0, implicit $exec, implicit $sp_reg :: (load (s64) from %stack.0, align 4, addrspace 5)
; GCN-NEXT: S_NOP 0, implicit killed renamable $sgpr2_sgpr3
; GCN-NEXT: S_NOP 0, implicit killed renamable $sgpr0_sgpr1
undef %0.sub0_sub1:sgpr_256 = COPY $sgpr8_sgpr9
%1:sreg_64_xexec = S_LOAD_DWORDX2_IMM %0.sub0_sub1:sgpr_256, 0, 0 :: (invariant dereferenceable load (s64), addrspace 4)
%2:sreg_64_xexec = S_LOAD_DWORDX2_IMM %0.sub0_sub1:sgpr_256, 4, 0 :: (invariant dereferenceable load (s64), addrspace 4)
%0.sub0_sub1:sgpr_256 = S_LOAD_DWORDX2_IMM %0.sub0_sub1:sgpr_256, 8, 0 :: (invariant dereferenceable load (s64), addrspace 4)
S_NOP 0, implicit %1
S_NOP 0, implicit %2
S_NOP 0, implicit %0.sub0_sub1:sgpr_256
...