Page MenuHomePhabricator

critson (Carl Ritson)
User

Projects

User does not belong to any projects.

User Details

User Since
Sep 4 2018, 4:49 AM (144 w, 4 d)

Recent Activity

Thu, Jun 10

critson committed rG2c2d2922a24b: [ValueTypes] Define MVTs for v6i32, v6f32, v7i32, v7f32 (authored by critson).
[ValueTypes] Define MVTs for v6i32, v6f32, v7i32, v7f32
Thu, Jun 10, 4:59 PM
critson committed rGcfbb92441f17: [SDAG] Fix pow2 assumption when splitting vectors (authored by critson).
[SDAG] Fix pow2 assumption when splitting vectors
Thu, Jun 10, 4:59 PM
critson closed D103881: [ValueTypes] Define MVTs for v6i32, v6f32, v7i32, v7f32.
Thu, Jun 10, 4:59 PM · Restricted Project
critson closed D103880: [SDAG] Fix pow2 assumption when splitting vectors.
Thu, Jun 10, 4:59 PM · Restricted Project
critson updated the diff for D103880: [SDAG] Fix pow2 assumption when splitting vectors.
  • Replace getVectorMinNumElements with getVectorNumElements
Thu, Jun 10, 3:54 AM · Restricted Project

Wed, Jun 9

critson added a comment to D103880: [SDAG] Fix pow2 assumption when splitting vectors.

test case?

Wed, Jun 9, 8:03 PM · Restricted Project
critson accepted D102212: [AMDGPU] Add Optimize VGPR LiveRange Pass..

LGTM, with one nit

Wed, Jun 9, 6:59 PM · Restricted Project
critson updated the diff for D103881: [ValueTypes] Define MVTs for v6i32, v6f32, v7i32, v7f32.
  • Rebase
Wed, Jun 9, 4:47 AM · Restricted Project
critson added inline comments to D102212: [AMDGPU] Add Optimize VGPR LiveRange Pass..
Wed, Jun 9, 2:21 AM · Restricted Project

Tue, Jun 8

critson added a comment to D103881: [ValueTypes] Define MVTs for v6i32, v6f32, v7i32, v7f32.

Awfully brave to add new non-pow2 types; I like it. Unfortunate that we've both filed patches to this same file at the same time (D103884). One of us will have a fun time.

Tue, Jun 8, 8:14 PM · Restricted Project
critson added a comment to D102212: [AMDGPU] Add Optimize VGPR LiveRange Pass..

Sorry a few more comments, mostly minor.

Tue, Jun 8, 3:04 AM · Restricted Project
critson requested review of D103881: [ValueTypes] Define MVTs for v6i32, v6f32, v7i32, v7f32.
Tue, Jun 8, 2:57 AM · Restricted Project
critson requested review of D103880: [SDAG] Fix pow2 assumption when splitting vectors.
Tue, Jun 8, 2:56 AM · Restricted Project

Mon, Jun 7

critson committed rGc8bbfb8cf5ea: [AMDGPU] Allow oversize vaddr in GFX10 MIMG assembly (authored by critson).
[AMDGPU] Allow oversize vaddr in GFX10 MIMG assembly
Mon, Jun 7, 7:57 PM
critson closed D103733: [AMDGPU] Allow oversize vaddr in GFX10 MIMG assembly.
Mon, Jun 7, 7:57 PM · Restricted Project
critson committed rGf8816c740025: [AMDGPU] Add v5f32/VReg_160 support for MIMG instructions (authored by critson).
[AMDGPU] Add v5f32/VReg_160 support for MIMG instructions
Mon, Jun 7, 7:31 PM
critson closed D103672: [AMDGPU] Add v5f32/VReg_160 support for MIMG instructions.
Mon, Jun 7, 7:31 PM · Restricted Project
critson updated the diff for D103733: [AMDGPU] Allow oversize vaddr in GFX10 MIMG assembly.
  • Address reviewer comments
Mon, Jun 7, 4:11 AM · Restricted Project
critson requested review of D103800: [AMDGPU] Add VReg_192 support for MIMG instructions.
Mon, Jun 7, 3:40 AM · Restricted Project

Fri, Jun 4

critson requested review of D103733: [AMDGPU] Allow oversize vaddr in GFX10 MIMG assembly.
Fri, Jun 4, 7:33 PM · Restricted Project
critson updated the diff for D103348: [AMDGPU] Add maximum NSA size limit ISA feature.
  • Address reviewer comments
Fri, Jun 4, 6:44 PM · Restricted Project
critson added inline comments to D103348: [AMDGPU] Add maximum NSA size limit ISA feature.
Fri, Jun 4, 6:43 PM · Restricted Project
critson added inline comments to D103348: [AMDGPU] Add maximum NSA size limit ISA feature.
Fri, Jun 4, 4:19 AM · Restricted Project
critson updated the diff for D103348: [AMDGPU] Add maximum NSA size limit ISA feature.
  • GlobalISel: move vaddr widening from legalizer to selection
Fri, Jun 4, 4:18 AM · Restricted Project
critson added a comment to D103672: [AMDGPU] Add v5f32/VReg_160 support for MIMG instructions.

My only slight concern is if people have GFX10 disassemblies stored somewhere, they might find they can no longer reassemble them after your change. I don't know how serious that is.

Fri, Jun 4, 3:37 AM · Restricted Project
critson requested review of D103672: [AMDGPU] Add v5f32/VReg_160 support for MIMG instructions.
Fri, Jun 4, 12:56 AM · Restricted Project

Tue, Jun 1

critson added inline comments to D103348: [AMDGPU] Add maximum NSA size limit ISA feature.
Tue, Jun 1, 11:06 PM · Restricted Project

Fri, May 28

critson requested review of D103348: [AMDGPU] Add maximum NSA size limit ISA feature.
Fri, May 28, 5:30 PM · Restricted Project
critson added a comment to D103230: [AMDGPU] Add options to disable NSA for BVH instructions.

Any reason to force SA?

Fri, May 28, 5:25 PM · Restricted Project

Thu, May 27

critson requested review of D103230: [AMDGPU] Add options to disable NSA for BVH instructions.
Thu, May 27, 2:05 AM · Restricted Project

Fri, May 21

critson added inline comments to D102830: [AMDGPU] Avoid null export insertion when unifying exit blocks.
Fri, May 21, 3:16 AM · Restricted Project
critson added inline comments to D102212: [AMDGPU] Add Optimize VGPR LiveRange Pass..
Fri, May 21, 3:01 AM · Restricted Project

Thu, May 20

critson updated the diff for D102830: [AMDGPU] Avoid null export insertion when unifying exit blocks.
  • Address reviewer comments
Thu, May 20, 8:52 PM · Restricted Project
critson added a comment to D102830: [AMDGPU] Avoid null export insertion when unifying exit blocks.

Mostly minor comments inline but I'm concerned about the "multiple predecessors with different done exports" case.

Thu, May 20, 8:52 PM · Restricted Project
critson requested review of D102830: [AMDGPU] Avoid null export insertion when unifying exit blocks.
Thu, May 20, 1:25 AM · Restricted Project

Thu, May 13

critson committed rG9cf6ff7affad: [AMDGPU] Do not clause NSA instructions (authored by critson).
[AMDGPU] Do not clause NSA instructions
Thu, May 13, 8:55 PM
critson closed D102211: [AMDGPU] Do not clause NSA instructions.
Thu, May 13, 8:55 PM · Restricted Project

May 11 2021

critson updated the diff for D102211: [AMDGPU] Do not clause NSA instructions.
  • Rebase
  • Contain in GFX10.1 specific bug feature
May 11 2021, 6:20 PM · Restricted Project
critson added a comment to D102211: [AMDGPU] Do not clause NSA instructions.

As I understand it, this is only needed for GFX10.1, not for GFX10.3. So I think it probably deserves its own target feature which would be included in GFX10_1_Bugs.

May 11 2021, 5:47 AM · Restricted Project

May 10 2021

critson updated the diff for D102211: [AMDGPU] Do not clause NSA instructions.
  • Rebase on pre-commit tests.
May 10 2021, 8:20 PM · Restricted Project
critson committed rGad558a4ff7cd: [AMDGPU] Pre-commit tests for D102211 (authored by critson).
[AMDGPU] Pre-commit tests for D102211
May 10 2021, 8:18 PM
critson requested review of D102211: [AMDGPU] Do not clause NSA instructions.
May 10 2021, 7:58 PM · Restricted Project

May 6 2021

critson committed rG67cfefebbbbb: [AMDGPU] Fix WQM failure with single block inactive demote (authored by critson).
[AMDGPU] Fix WQM failure with single block inactive demote
May 6 2021, 5:03 AM
critson closed D101966: [AMDGPU] Fix WQM failure with single block inactive demote.
May 6 2021, 5:03 AM · Restricted Project

May 5 2021

critson requested review of D101966: [AMDGPU] Fix WQM failure with single block inactive demote.
May 5 2021, 6:22 PM · Restricted Project
critson added a comment to D101414: [AMDGPU] Disable the scalar IR, SDWA and load store vectorizer passes at -O1.

llc-pipeline.ll has problems with -DLLVM_ENABLE_EXPENSIVE_CHECKS=1 as instruction verification seems to still take place.

May 5 2021, 4:08 AM · Restricted Project, Restricted Project

Apr 29 2021

critson committed rG424f1f6f96d0: [AMDGPU][NFC] Refactor hazard recognition IsHazardFn and IsExpiredFn (authored by critson).
[AMDGPU][NFC] Refactor hazard recognition IsHazardFn and IsExpiredFn
Apr 29 2021, 5:19 PM
critson closed D101430: [AMDGPU][NFC] Refactor hazard recognition IsHazardFn and IsExpiredFn.
Apr 29 2021, 5:19 PM · Restricted Project
critson committed rG749702fc6ba2: [AMDGPU] Remove dead early-out in GCNHazardRecognizer (authored by critson).
[AMDGPU] Remove dead early-out in GCNHazardRecognizer
Apr 29 2021, 4:57 PM
critson closed D101520: [AMDGPU] Remove dead early-out in GCNHazardRecognizer.
Apr 29 2021, 4:57 PM · Restricted Project
critson updated the diff for D101430: [AMDGPU][NFC] Refactor hazard recognition IsHazardFn and IsExpiredFn.

Rebase.

Apr 29 2021, 5:00 AM · Restricted Project
critson retitled D101430: [AMDGPU][NFC] Refactor hazard recognition IsHazardFn and IsExpiredFn from [AMDGPU] Refactor hazard recognition IsHazardFn and IsExpiredFn to [AMDGPU][NFC] Refactor hazard recognition IsHazardFn and IsExpiredFn.
Apr 29 2021, 4:58 AM · Restricted Project
critson added a comment to D101430: [AMDGPU][NFC] Refactor hazard recognition IsHazardFn and IsExpiredFn.

Move the code removal change to D101520, making this NFC.

Apr 29 2021, 4:57 AM · Restricted Project
critson updated the diff for D101520: [AMDGPU] Remove dead early-out in GCNHazardRecognizer.

Remove extra line.

Apr 29 2021, 4:57 AM · Restricted Project
critson requested review of D101520: [AMDGPU] Remove dead early-out in GCNHazardRecognizer.
Apr 29 2021, 4:54 AM · Restricted Project

Apr 28 2021

critson added a comment to D101430: [AMDGPU][NFC] Refactor hazard recognition IsHazardFn and IsExpiredFn.
Apr 28 2021, 3:23 AM · Restricted Project
critson added a comment to D101430: [AMDGPU][NFC] Refactor hazard recognition IsHazardFn and IsExpiredFn.

I do not think it is broken per-say but definitely confusing.

I think it is broken because we are looping over predecessors trying to find the most recent hazard. The early-out allows the search to terminate if the hazard in some predecessor was too long ago, but then we would miss the fact that another predecessor might have a more recent hazard.

Apr 28 2021, 1:33 AM · Restricted Project
critson added inline comments to D56923: [AMDGPU] Fixed hazard recognizer to walk predecessors.
Apr 28 2021, 1:10 AM · Restricted Project
critson requested review of D101430: [AMDGPU][NFC] Refactor hazard recognition IsHazardFn and IsExpiredFn.
Apr 28 2021, 1:08 AM · Restricted Project

Apr 27 2021

critson added inline comments to D56923: [AMDGPU] Fixed hazard recognizer to walk predecessors.
Apr 27 2021, 7:27 PM · Restricted Project

Apr 21 2021

critson accepted D100969: [AMDGPU] SIWholeQuadMode: don't add duplicate implicit $exec operands.

I think this particular line of code is pretty old.
I take it this is largely NFC on compiler output (i.e. no test difference)?

Apr 21 2021, 4:59 PM · Restricted Project

Apr 6 2021

critson added a comment to D88287: [NARY-REASSOCIATE] Support reassociation of min/max.

Raising this here as well, as it seems the previous concern I raised with the commit was ignored.

Apr 6 2021, 7:30 PM · Restricted Project
critson raised a concern with rG2388aae401dc: [NARY-REASSOCIATE] Support reassociation of min/max.

This code goes into an infinite loop on AMDGPU backend.
See attached bugpoint test case:
llc -march=amdgcn -mcpu=gfx700 < bugpoint-reduced-simplified.ll

Apr 6 2021, 7:25 PM

Mar 29 2021

critson added a comment to D99507: [amdgpu] Add a pass to avoid jump into blocks with 0 exec mask..

Instead of having a fixup patch to avoid cases where this happens, we should have the infrastructure to stop this from happening in the first place

Mar 29 2021, 5:15 PM · Restricted Project

Mar 28 2021

critson added inline comments to D99429: [AMDGPU] Save WWM registers in functions.
Mar 28 2021, 8:24 PM · Restricted Project

Mar 23 2021

critson accepted D98953: [AMDGPU] Use reductions instead of scans in the atomic optimizer.

LGTM - this seems like a good use of GFX10 row_xmask.

Mar 23 2021, 1:14 AM · Restricted Project

Mar 22 2021

critson committed rG64db6b8d3755: [AMDGPU] Only unbundle memory accesses in SIMemoryLegalizer (authored by critson).
[AMDGPU] Only unbundle memory accesses in SIMemoryLegalizer
Mar 22 2021, 7:31 PM
critson closed D99061: [AMDGPU] Only unbundle memory accesses in SIMemoryLegalizer.
Mar 22 2021, 7:31 PM · Restricted Project
critson added a comment to D99061: [AMDGPU] Only unbundle memory accesses in SIMemoryLegalizer.

What was the decision on the philosophy on who should be doing unbundling in general? How is it known that it is sfe for the memory legalized to do this unbundling? Is the policy documented in some AMD GPU design page?

Mar 22 2021, 6:55 PM · Restricted Project
critson added a comment to D99061: [AMDGPU] Only unbundle memory accesses in SIMemoryLegalizer.

Thanks, I will leave this 12 hours before committing in case of further comments.

Mar 22 2021, 3:43 AM · Restricted Project
critson requested review of D99061: [AMDGPU] Only unbundle memory accesses in SIMemoryLegalizer.
Mar 22 2021, 2:23 AM · Restricted Project

Mar 19 2021

critson committed rG6c9cac5da1d1: [AMDGPU] Add MDT update missing from D98915 (authored by critson).
[AMDGPU] Add MDT update missing from D98915
Mar 19 2021, 9:40 PM
critson committed rGfe5f4c397f02: [AMDGPU] Rename SIInsertSkips Pass (authored by critson).
[AMDGPU] Rename SIInsertSkips Pass
Mar 19 2021, 7:56 PM
critson closed D98915: [AMDGPU] Rename SIInsertSkips Pass.
Mar 19 2021, 7:56 PM · Restricted Project
critson committed rG5df2af8b0ef3: [AMDGPU] Merge SIRemoveShortExecBranches into SIPreEmitPeephole (authored by critson).
[AMDGPU] Merge SIRemoveShortExecBranches into SIPreEmitPeephole
Mar 19 2021, 7:27 PM
critson closed D98917: [AMDGPU] Merge SIRemoveShortExecBranches into SIPreEmitPeephole.
Mar 19 2021, 7:27 PM · Restricted Project
critson added a comment to D98940: [AMDGPU] Allow index optimisation in SIPreEmitPeephole for bundles.

It would help if we had more clarity on what exactly we thought a bundle should be and eventually had more granularity to identify types of bundles like Jay/Stas were suggesting.

Mar 19 2021, 6:28 PM · Restricted Project
critson committed rGb76c09023d9a: [AMDGPU] Allow index optimisation in SIPreEmitPeephole for bundles (authored by critson).
[AMDGPU] Allow index optimisation in SIPreEmitPeephole for bundles
Mar 19 2021, 6:27 PM
critson closed D98940: [AMDGPU] Allow index optimisation in SIPreEmitPeephole for bundles.
Mar 19 2021, 6:26 PM · Restricted Project
critson added a comment to D98940: [AMDGPU] Allow index optimisation in SIPreEmitPeephole for bundles.

OK, but really this change should either have its own test, or it should be combined with the change to SIMemoryLegalizer.

Mar 19 2021, 5:46 AM · Restricted Project
critson updated the diff for D98940: [AMDGPU] Allow index optimisation in SIPreEmitPeephole for bundles.
  • Add basic tests (copies of unbundled ones)
  • Add comment w.r.t. need to process bundles
Mar 19 2021, 5:42 AM · Restricted Project
critson added a comment to D98940: [AMDGPU] Allow index optimisation in SIPreEmitPeephole for bundles.

Add code so duplication mode register changes

"duplicated"?

Needs a test. Also can you explain what bundles these are and why it's OK to modify them? In general I think we should err on the side of not modifying bundles, otherwise what's the point of bundling them?

Mar 19 2021, 5:15 AM · Restricted Project
critson retitled D98940: [AMDGPU] Allow index optimisation in SIPreEmitPeephole for bundles from [AMDGPU] Allow mode optimisation in SIPreEmitPeephole for bundles to [AMDGPU] Allow index optimisation in SIPreEmitPeephole for bundles.
Mar 19 2021, 5:09 AM · Restricted Project
critson requested review of D98940: [AMDGPU] Allow index optimisation in SIPreEmitPeephole for bundles.
Mar 19 2021, 4:49 AM · Restricted Project

Mar 18 2021

critson requested review of D98917: [AMDGPU] Merge SIRemoveShortExecBranches into SIPreEmitPeephole.
Mar 18 2021, 8:25 PM · Restricted Project
critson requested review of D98915: [AMDGPU] Rename SIInsertSkips Pass.
Mar 18 2021, 8:20 PM · Restricted Project

Mar 17 2021

critson committed rG1a4bc3aba360: [AMDGPU] Avoid unnecessary graph visits during WQM marking (authored by critson).
[AMDGPU] Avoid unnecessary graph visits during WQM marking
Mar 17 2021, 6:02 PM
critson closed D98772: [AMDGPU] Avoid unnecessary graph visits during WQM marking.
Mar 17 2021, 6:01 PM · Restricted Project
critson updated the diff for D98772: [AMDGPU] Avoid unnecessary graph visits during WQM marking.
  • Address reviewer comments.
Mar 17 2021, 4:59 AM · Restricted Project
critson requested review of D98772: [AMDGPU] Avoid unnecessary graph visits during WQM marking.
Mar 17 2021, 4:24 AM · Restricted Project

Mar 15 2021

critson committed rG13877db2fa85: [AMDGPU] Fix shortfalls in WQM marking (authored by critson).
[AMDGPU] Fix shortfalls in WQM marking
Mar 15 2021, 5:45 AM
critson closed D98614: [AMDGPU] Fix shortfalls in WQM marking.
Mar 15 2021, 5:45 AM · Restricted Project
critson updated the diff for D98614: [AMDGPU] Fix shortfalls in WQM marking.
  • Fix mistake in previous diff
Mar 15 2021, 5:43 AM · Restricted Project
critson added inline comments to D98614: [AMDGPU] Fix shortfalls in WQM marking.
Mar 15 2021, 5:40 AM · Restricted Project
critson updated the diff for D98614: [AMDGPU] Fix shortfalls in WQM marking.
  • Address comments
Mar 15 2021, 5:31 AM · Restricted Project
critson added inline comments to D98614: [AMDGPU] Fix shortfalls in WQM marking.
Mar 15 2021, 5:23 AM · Restricted Project
critson updated the diff for D98614: [AMDGPU] Fix shortfalls in WQM marking.
  • Add comment to NextPredIdx
Mar 15 2021, 4:41 AM · Restricted Project
critson added inline comments to D98614: [AMDGPU] Fix shortfalls in WQM marking.
Mar 15 2021, 4:36 AM · Restricted Project
critson updated the diff for D98614: [AMDGPU] Fix shortfalls in WQM marking.
  • Switch to SmallSetVector
  • Address other review comments
Mar 15 2021, 2:49 AM · Restricted Project

Mar 14 2021

critson requested review of D98614: [AMDGPU] Fix shortfalls in WQM marking.
Mar 14 2021, 5:59 PM · Restricted Project