Page MenuHomePhabricator

kerbowa (Austin Kerbow)
User

Projects

User does not belong to any projects.

User Details

User Since
Dec 31 2018, 12:07 PM (108 w, 10 h)

Recent Activity

Sat, Jan 23

kerbowa updated the diff for D85882: [AMDGPU] Update subtarget features for new target ID support.

Rebase. Address comments.

Sat, Jan 23, 12:45 AM · Restricted Project

Thu, Jan 21

kerbowa added inline comments to D85882: [AMDGPU] Update subtarget features for new target ID support.
Thu, Jan 21, 11:18 PM · Restricted Project
kerbowa updated the diff for D85882: [AMDGPU] Update subtarget features for new target ID support.

Rebase after 560d7e04113bf.

Thu, Jan 21, 11:07 PM · Restricted Project

Sun, Jan 17

kerbowa added inline comments to D85882: [AMDGPU] Update subtarget features for new target ID support.
Sun, Jan 17, 10:39 PM · Restricted Project
kerbowa added inline comments to D85882: [AMDGPU] Update subtarget features for new target ID support.
Sun, Jan 17, 1:54 PM · Restricted Project
kerbowa added inline comments to D85882: [AMDGPU] Update subtarget features for new target ID support.
Sun, Jan 17, 12:52 PM · Restricted Project
kerbowa updated the diff for D85882: [AMDGPU] Update subtarget features for new target ID support.

Don't use Optional type for TargetID.

Sun, Jan 17, 11:55 AM · Restricted Project
kerbowa updated the diff for D85882: [AMDGPU] Update subtarget features for new target ID support.

Rebase.

Sun, Jan 17, 1:07 AM · Restricted Project

Mon, Jan 4

kerbowa added a comment to D93715: AMDGPU: Don't fold AGPR copy pairs that need a temp VGPR.

I'm working on eliminating some AGPR->VGPR->AGPR copies which commonly cause problems in loops where the VGPR is considered live across iterations by the register allocator even if the VGPR copy should be sunk. Why do we need this temp VGPR?

Mon, Jan 4, 8:43 AM · Restricted Project

Dec 8 2020

kerbowa committed rG4aa842a800b5: [AMDGPU] Add new pseudos for indirect addressing with VGPR Indexing (authored by kerbowa).
[AMDGPU] Add new pseudos for indirect addressing with VGPR Indexing
Dec 8 2020, 12:25 PM
kerbowa closed D91048: [AMDGPU] Add new pseudos for indirect addressing with VGPR Indexing.
Dec 8 2020, 12:24 PM · Restricted Project
kerbowa updated the diff for D91048: [AMDGPU] Add new pseudos for indirect addressing with VGPR Indexing.

Add M0 to defs.

Dec 8 2020, 12:08 PM · Restricted Project

Dec 7 2020

kerbowa added a comment to D91048: [AMDGPU] Add new pseudos for indirect addressing with VGPR Indexing.

JBTW, having a need to def M0 probably defeats the idea of potential rescheduling which justifies the separation of the indirect access methods.

Dec 7 2020, 8:34 PM · Restricted Project

Dec 6 2020

kerbowa added a comment to D91048: [AMDGPU] Add new pseudos for indirect addressing with VGPR Indexing.

ping

Dec 6 2020, 1:08 PM · Restricted Project

Nov 30 2020

kerbowa updated the diff for D91048: [AMDGPU] Add new pseudos for indirect addressing with VGPR Indexing.

Fix test.

Nov 30 2020, 10:22 PM · Restricted Project
kerbowa added inline comments to D85882: [AMDGPU] Update subtarget features for new target ID support.
Nov 30 2020, 11:34 AM · Restricted Project
kerbowa updated the diff for D85882: [AMDGPU] Update subtarget features for new target ID support.

Rebase. Move some utility functions to header.

Nov 30 2020, 11:21 AM · Restricted Project

Nov 29 2020

kerbowa added a comment to D91048: [AMDGPU] Add new pseudos for indirect addressing with VGPR Indexing.

EPSDB is passing.

Nov 29 2020, 12:29 AM · Restricted Project
kerbowa updated the diff for D91048: [AMDGPU] Add new pseudos for indirect addressing with VGPR Indexing.

Address comments.

Nov 29 2020, 12:27 AM · Restricted Project

Nov 9 2020

kerbowa added a comment to D91048: [AMDGPU] Add new pseudos for indirect addressing with VGPR Indexing.

Is there a practical reason to have different pseudos for set_idx and movrel methods and not just expand a common pseudo accordingly post RA?

It's a good question but in the end, I chose not to since m0 initializations can be rescheduled, and combing the pseudo's would disallow this. The other issue is that the movrel pseudo includes the scalar version which is used on targets with VGPR indexing as well. I'm open to trying it if you think it seems more practical.

Both methods use M0, aren't they?

Nov 9 2020, 7:39 PM · Restricted Project
kerbowa added a comment to D91048: [AMDGPU] Add new pseudos for indirect addressing with VGPR Indexing.

Is there a practical reason to have different pseudos for set_idx and movrel methods and not just expand a common pseudo accordingly post RA?

Nov 9 2020, 12:20 PM · Restricted Project

Nov 8 2020

kerbowa updated the diff for D91048: [AMDGPU] Add new pseudos for indirect addressing with VGPR Indexing.

Remove dead code.

Nov 8 2020, 11:40 PM · Restricted Project
kerbowa requested review of D91048: [AMDGPU] Add new pseudos for indirect addressing with VGPR Indexing.
Nov 8 2020, 11:28 PM · Restricted Project

Oct 28 2020

kerbowa committed rGde518673436e: [AMDGPU] Add Reset function to GCNHazardRecognizer (authored by kerbowa).
[AMDGPU] Add Reset function to GCNHazardRecognizer
Oct 28 2020, 4:33 PM
kerbowa closed D90347: [AMDGPU] Add Reset function to GCNHazardRecognizer.
Oct 28 2020, 4:33 PM · Restricted Project
kerbowa requested review of D90347: [AMDGPU] Add Reset function to GCNHazardRecognizer.
Oct 28 2020, 3:13 PM · Restricted Project
kerbowa committed rG8b127a8661a1: [AMDGPU] Fix inserting combined s_nop in bundles (authored by kerbowa).
[AMDGPU] Fix inserting combined s_nop in bundles
Oct 28 2020, 2:35 PM
kerbowa closed D90334: [AMDGPU] Fix inserting combined s_nop in bundles.
Oct 28 2020, 2:34 PM · Restricted Project
kerbowa added inline comments to D90334: [AMDGPU] Fix inserting combined s_nop in bundles.
Oct 28 2020, 2:16 PM · Restricted Project
kerbowa added inline comments to D90334: [AMDGPU] Fix inserting combined s_nop in bundles.
Oct 28 2020, 1:41 PM · Restricted Project
kerbowa requested review of D90334: [AMDGPU] Fix inserting combined s_nop in bundles.
Oct 28 2020, 12:23 PM · Restricted Project

Oct 26 2020

kerbowa updated the summary of D85882: [AMDGPU] Update subtarget features for new target ID support.
Oct 26 2020, 11:17 PM · Restricted Project
kerbowa updated the diff for D85882: [AMDGPU] Update subtarget features for new target ID support.

Update, rebase, and merge with change of default patch.

Oct 26 2020, 11:16 PM · Restricted Project

Oct 20 2020

kerbowa committed rGebdcef20ce29: [AMDGPU] Avoid inserting noops during scheduling (authored by kerbowa).
[AMDGPU] Avoid inserting noops during scheduling
Oct 20 2020, 5:12 PM
kerbowa closed D89754: [AMDGPU] Avoid inserting noops during scheduling.
Oct 20 2020, 5:12 PM · Restricted Project
kerbowa committed rG37d907899f49: [HazardRec] Allow inserting multiple wait-states simultaneously (authored by kerbowa).
[HazardRec] Allow inserting multiple wait-states simultaneously
Oct 20 2020, 5:06 PM
kerbowa closed D89753: [HazardRec] Allow inserting multiple wait-states simultaneously.
Oct 20 2020, 5:06 PM · Restricted Project
kerbowa added a comment to D89753: [HazardRec] Allow inserting multiple wait-states simultaneously.

Do we properly count number of pre-existing wait states if we have s_nop > 0?

Oct 20 2020, 11:52 AM · Restricted Project
kerbowa updated the diff for D89753: [HazardRec] Allow inserting multiple wait-states simultaneously.

Address comments.

Oct 20 2020, 11:42 AM · Restricted Project
kerbowa added inline comments to D89754: [AMDGPU] Avoid inserting noops during scheduling.
Oct 20 2020, 10:01 AM · Restricted Project

Oct 19 2020

kerbowa requested review of D89754: [AMDGPU] Avoid inserting noops during scheduling.
Oct 19 2020, 5:14 PM · Restricted Project
kerbowa requested review of D89753: [HazardRec] Allow inserting multiple wait-states simultaneously.
Oct 19 2020, 5:11 PM · Restricted Project
kerbowa abandoned D75030: [AMDGPU] Use conservative defaults for XNACK/SRAM ECC.
Oct 19 2020, 9:36 AM · Restricted Project
kerbowa abandoned D86042: [AMDGPU] Use correct defaults for xnack and sramecc.
Oct 19 2020, 9:35 AM · Restricted Project

Oct 16 2020

kerbowa committed rG978fbd8268ce: [AMDGPU] Run hazard recognizer pass later (authored by kerbowa).
[AMDGPU] Run hazard recognizer pass later
Oct 16 2020, 12:16 PM
kerbowa closed D89077: [AMDGPU] Run hazard recognizer pass later.
Oct 16 2020, 12:16 PM · Restricted Project

Oct 11 2020

kerbowa updated the diff for D89077: [AMDGPU] Run hazard recognizer pass later.

Add test.

Oct 11 2020, 10:28 PM · Restricted Project

Oct 8 2020

kerbowa added a comment to D89077: [AMDGPU] Run hazard recognizer pass later.

Is this now running after the waitcnt insertion pass? That would avoid the NOPs currently being inserted to split memory clauses that are not necessary as the waitcnt instructions will split the clauses.

Oct 8 2020, 5:42 PM · Restricted Project
kerbowa added a comment to D89077: [AMDGPU] Run hazard recognizer pass later.

Is this now running after the waitcnt insertion pass? That would avoid the NOPs currently being inserted to split memory clauses that are not necessary as the waitcnt instructions will split the clauses.

Oct 8 2020, 5:41 PM · Restricted Project
kerbowa committed rGa4f35ab23281: [AMDGPU] Fix mai hazard VALU to LD/ST (authored by kerbowa).
[AMDGPU] Fix mai hazard VALU to LD/ST
Oct 8 2020, 5:13 PM
kerbowa closed D89079: [AMDGPU] Fix mai hazard VALU to LD/ST.
Oct 8 2020, 5:13 PM · Restricted Project
kerbowa updated the diff for D89079: [AMDGPU] Fix mai hazard VALU to LD/ST.

Wt -> Wr

Oct 8 2020, 4:50 PM · Restricted Project
kerbowa requested review of D89079: [AMDGPU] Fix mai hazard VALU to LD/ST.
Oct 8 2020, 4:38 PM · Restricted Project
kerbowa requested review of D89077: [AMDGPU] Run hazard recognizer pass later.
Oct 8 2020, 4:10 PM · Restricted Project

Sep 16 2020

kerbowa added inline comments to D87621: [AMDGPU] Add XDL resource to scheduling model.
Sep 16 2020, 8:13 PM · Restricted Project

Sep 14 2020

kerbowa committed rGf859c30ecbbb: [AMDGPU] Add XDL resource to scheduling model (authored by kerbowa).
[AMDGPU] Add XDL resource to scheduling model
Sep 14 2020, 1:49 PM
kerbowa closed D87621: [AMDGPU] Add XDL resource to scheduling model.
Sep 14 2020, 1:49 PM · Restricted Project
kerbowa requested review of D87621: [AMDGPU] Add XDL resource to scheduling model.
Sep 14 2020, 10:58 AM · Restricted Project

Aug 16 2020

kerbowa requested review of D86042: [AMDGPU] Use correct defaults for xnack and sramecc.
Aug 16 2020, 2:21 PM · Restricted Project

Aug 14 2020

kerbowa added inline comments to D85882: [AMDGPU] Update subtarget features for new target ID support.
Aug 14 2020, 8:50 AM · Restricted Project

Aug 13 2020

kerbowa updated the summary of D85882: [AMDGPU] Update subtarget features for new target ID support.
Aug 13 2020, 10:51 PM · Restricted Project
kerbowa updated the summary of D85882: [AMDGPU] Update subtarget features for new target ID support.
Aug 13 2020, 7:24 PM · Restricted Project
kerbowa updated the diff for D85882: [AMDGPU] Update subtarget features for new target ID support.

Rename "Default" to "Any". Address comments.

Aug 13 2020, 7:21 PM · Restricted Project
kerbowa added inline comments to D85882: [AMDGPU] Update subtarget features for new target ID support.
Aug 13 2020, 4:21 PM · Restricted Project
kerbowa committed rG7d1cb187fbee: [AMDGPU] Fix FP/BP spills when MUBUF constant offset exceeded (authored by kerbowa).
[AMDGPU] Fix FP/BP spills when MUBUF constant offset exceeded
Aug 13 2020, 2:12 PM
kerbowa closed D85772: [AMDGPU] Fix FP/BP spills when MUBUF constant offset exceeded.
Aug 13 2020, 2:12 PM · Restricted Project
kerbowa updated the diff for D85882: [AMDGPU] Update subtarget features for new target ID support.

Re-add DoesNotSupport features. Use enum class. No fatal error on setting mismatch.

Aug 13 2020, 1:53 PM · Restricted Project
kerbowa updated the diff for D85772: [AMDGPU] Fix FP/BP spills when MUBUF constant offset exceeded.

Fix typo.

Aug 13 2020, 11:08 AM · Restricted Project
kerbowa added a reviewer for D85772: [AMDGPU] Fix FP/BP spills when MUBUF constant offset exceeded: hsmhsm.
Aug 13 2020, 10:53 AM · Restricted Project
kerbowa updated the diff for D85772: [AMDGPU] Fix FP/BP spills when MUBUF constant offset exceeded.

As discussed offline with Mahesha, this and https://reviews.llvm.org/D85767 are fixing the same issue, we decided I would take the fix forward.

Aug 13 2020, 10:52 AM · Restricted Project
kerbowa added inline comments to D85882: [AMDGPU] Update subtarget features for new target ID support.
Aug 13 2020, 8:39 AM · Restricted Project
kerbowa requested review of D85882: [AMDGPU] Update subtarget features for new target ID support.
Aug 13 2020, 12:21 AM · Restricted Project

Aug 11 2020

kerbowa requested review of D85772: [AMDGPU] Fix FP/BP spills when MUBUF constant offset exceeded.
Aug 11 2020, 12:51 PM · Restricted Project

Aug 7 2020

kerbowa committed rG04bd5b528620: [AMDGPU] Fix not rescheduling without clustering (authored by vangthao95).
[AMDGPU] Fix not rescheduling without clustering
Aug 7 2020, 11:18 AM
kerbowa closed D85498: [AMDGPU] Fix not rescheduling without clustering.
Aug 7 2020, 11:17 AM · Restricted Project

Aug 6 2020

kerbowa requested review of D85498: [AMDGPU] Fix not rescheduling without clustering.
Aug 6 2020, 9:09 PM · Restricted Project
kerbowa added a comment to D85456: Fix instruction counting in post-RA scheduler.

Seems wrong to have a bundle without BUNDLE opcode but the change seems fine obviously. Since it's allowed we should probably just account for the possibility.

Aug 6 2020, 2:00 PM · Restricted Project

Aug 5 2020

kerbowa accepted D84163: AMDGPU/GlobalISel: Fix trying to widen <3 x s1> boolean ops.

LGTM

Aug 5 2020, 8:06 PM · Restricted Project
kerbowa accepted D84249: AMDGPU/GlobalISel: Stop using G_EXTRACT in argument lowering.

LGTM

Aug 5 2020, 7:41 PM · Restricted Project

Jul 29 2020

kerbowa accepted D84883: [AMDGPU] prefer non-mfma in post-RA schedule.

LGTM either way

Jul 29 2020, 12:15 PM · Restricted Project
kerbowa added a comment to D84883: [AMDGPU] prefer non-mfma in post-RA schedule.

Actually MFMA would need more height to ensure post-RA tries to schedule them first and together. You could have their results be implicit uses on later instruction like S_ENDPGM or something.

Jul 29 2020, 12:13 PM · Restricted Project
kerbowa added a comment to D84883: [AMDGPU] prefer non-mfma in post-RA schedule.

Minor nit, but test should disable-power-sched otherwise I think MFMA will be scheduled apart anyway.

Jul 29 2020, 12:01 PM · Restricted Project

Jul 28 2020

kerbowa accepted D84198: AMDGPU/GlobalISel: Refactor special argument management.

LGTM

Jul 28 2020, 8:56 PM · Restricted Project
kerbowa committed rGadeeac9d5a32: [AMDGPU] Spill CSR VGPR which is reserved for SGPR spills (authored by kerbowa).
[AMDGPU] Spill CSR VGPR which is reserved for SGPR spills
Jul 28 2020, 11:54 AM
kerbowa closed D83730: [AMDGPU] Spill CSR VGPR which is reserved for SGPR spills.
Jul 28 2020, 11:53 AM · Restricted Project

Jul 26 2020

kerbowa accepted D84203: AMDGPU/GlobalISel: Select llvm.returnaddress.

LGTM

Jul 26 2020, 4:49 PM · Restricted Project

Jul 25 2020

kerbowa added inline comments to D84198: AMDGPU/GlobalISel: Refactor special argument management.
Jul 25 2020, 2:36 PM · Restricted Project
kerbowa accepted D83988: AMDGPU/GlobalISel: Start trying to handle AGPR bank.

LGTM with parent

Jul 25 2020, 2:24 PM · Restricted Project
kerbowa accepted D84164: AMDGPU/GlobalISel: Pack constant G_BUILD_VECTOR_TRUNCs when selecting.

LGTM

Jul 25 2020, 2:05 PM · Restricted Project

Jul 23 2020

kerbowa updated the diff for D83730: [AMDGPU] Spill CSR VGPR which is reserved for SGPR spills.

Create the stack object in lowerShiftReservedVGPR. Also handle this reserved VGPR not being live-in to all BB. Add more tests. Cleanup some of the existing logic.

Jul 23 2020, 11:42 PM · Restricted Project
kerbowa added inline comments to D83730: [AMDGPU] Spill CSR VGPR which is reserved for SGPR spills.
Jul 23 2020, 7:52 PM · Restricted Project
kerbowa accepted D83323: AMDGPU/GlobalISel: Handle call return values.

LGTM

Jul 23 2020, 1:20 AM · Restricted Project

Jul 19 2020

kerbowa accepted D83258: AMDGPU/GlobalISel: Initial Implementation of calls.

LGTM as far as i can tell

Jul 19 2020, 9:09 PM · Restricted Project

Jul 18 2020

kerbowa added inline comments to D83988: AMDGPU/GlobalISel: Start trying to handle AGPR bank.
Jul 18 2020, 8:46 PM · Restricted Project

Jul 15 2020

kerbowa updated the diff for D83730: [AMDGPU] Spill CSR VGPR which is reserved for SGPR spills.

Workaround MIR emitter issue.

Jul 15 2020, 10:45 AM · Restricted Project

Jul 13 2020

Herald added a project to D83730: [AMDGPU] Spill CSR VGPR which is reserved for SGPR spills: Restricted Project.
Jul 13 2020, 4:40 PM · Restricted Project

Jun 30 2020

kerbowa accepted D82463: [AMDGPU] Spill more than wavesize CSR SGPRs.
Jun 30 2020, 10:12 PM · Restricted Project
kerbowa added inline comments to D82463: [AMDGPU] Spill more than wavesize CSR SGPRs.
Jun 30 2020, 5:23 PM · Restricted Project

Jun 29 2020

kerbowa accepted D82463: [AMDGPU] Spill more than wavesize CSR SGPRs.

LGTM

Jun 29 2020, 2:12 PM · Restricted Project

Jun 26 2020

kerbowa added inline comments to D82463: [AMDGPU] Spill more than wavesize CSR SGPRs.
Jun 26 2020, 9:17 AM · Restricted Project