Page MenuHomePhabricator

Please use GitHub pull requests for new patches. Phabricator shutdown timeline

piotr (Piotr Sobczak)
User

Projects

User does not belong to any projects.

User Details

User Since
Dec 4 2018, 6:02 AM (251 w, 3 h)

Recent Activity

Thu, Sep 21

piotr added a comment to D154083: [AMDGPU] Rematerialize scalar loads.

Tested on 10k pipelines from Vulkan games, this patch reduces the number of v_writelane instructions from 9842 to 6440 (at the expense of using more loads of course).

Thu, Sep 21, 5:30 AM · Restricted Project, Restricted Project

Mon, Sep 18

piotr added a comment to D158059: [AMDGPU/wmma] - Disable 3-address syntax for f16.

Then, as you say, our register allocation needs to be intelligent enough to keep the matrices packed.
How would you define the instructions for this to work?

Mon, Sep 18, 7:53 AM · Restricted Project, Restricted Project

Wed, Sep 13

piotr updated the diff for D156999: [Inline Spiller] Consider bundles when marking defs as dead.

Corrected the use of isCopyInstr.

Wed, Sep 13, 6:07 AM · Restricted Project, Restricted Project
piotr added a comment to D158463: [AMDGPU] Add IR-level pass to rewrite away address space 7.

I have done the first round of testing of the patch on graphics content, with LLPC PatchBufferOp disabled and small modification in LLPC to use ptr addrspace (7) in lieu of lgc.buffer.desc.to.ptr.

Wed, Sep 13, 3:28 AM · Restricted Project, Restricted Project

Tue, Sep 12

piotr added a comment to D158059: [AMDGPU/wmma] - Disable 3-address syntax for f16.

Thanks for the extra info; I understand the problem now: currently there seems to be no way to take advantage of the opsel bit to reuse the same destination matrix registers for two wmma instructions.

Tue, Sep 12, 12:24 AM · Restricted Project, Restricted Project

Mon, Sep 11

piotr updated the diff for D156999: [Inline Spiller] Consider bundles when marking defs as dead.

Delete only if all copies are dead, use isCopyInstr.

Mon, Sep 11, 9:12 AM · Restricted Project, Restricted Project

Fri, Sep 8

piotr added inline comments to D156999: [Inline Spiller] Consider bundles when marking defs as dead.
Fri, Sep 8, 7:14 AM · Restricted Project, Restricted Project

Thu, Sep 7

piotr added a comment to D158059: [AMDGPU/wmma] - Disable 3-address syntax for f16.

Thanks, that would avoid the regression. However, I still do not fully understand the failing mode - can you show the test case + extra code that triggers the issue?

Thu, Sep 7, 3:24 AM · Restricted Project, Restricted Project

Mon, Sep 4

piotr added a comment to D154083: [AMDGPU] Rematerialize scalar loads.

Ping.

Mon, Sep 4, 10:20 AM · Restricted Project, Restricted Project

Tue, Aug 29

piotr added a comment to D158463: [AMDGPU] Add IR-level pass to rewrite away address space 7.

Thanks for working on this. Just a few high-level comments:

Tue, Aug 29, 6:03 AM · Restricted Project, Restricted Project

Mon, Aug 28

piotr updated the diff for D156999: [Inline Spiller] Consider bundles when marking defs as dead.

Reworked to check for copy.

Mon, Aug 28, 7:11 AM · Restricted Project, Restricted Project

Aug 25 2023

piotr added a comment to D154083: [AMDGPU] Rematerialize scalar loads.

Ping - I think the only unresolved point is potential weakening of the generic check.

Aug 25 2023, 8:35 AM · Restricted Project, Restricted Project
piotr updated the diff for D154083: [AMDGPU] Rematerialize scalar loads.

Rebased.

Aug 25 2023, 8:27 AM · Restricted Project, Restricted Project
piotr added inline comments to D156999: [Inline Spiller] Consider bundles when marking defs as dead.
Aug 25 2023, 8:27 AM · Restricted Project, Restricted Project
piotr added a comment to D158059: [AMDGPU/wmma] - Disable 3-address syntax for f16.

I am trying to understand the failing case better. Can the issue only happen with the extra patch with packing? Is the issue only with zeroinitializers (constant matrices), or it is just where the problem was found?

Aug 25 2023, 6:46 AM · Restricted Project, Restricted Project

Aug 3 2023

piotr added inline comments to D156999: [Inline Spiller] Consider bundles when marking defs as dead.
Aug 3 2023, 7:00 AM · Restricted Project, Restricted Project
piotr added inline comments to D156999: [Inline Spiller] Consider bundles when marking defs as dead.
Aug 3 2023, 6:51 AM · Restricted Project, Restricted Project
piotr added reviewers for D156999: [Inline Spiller] Consider bundles when marking defs as dead: arsenm, MatzeB.
Aug 3 2023, 6:41 AM · Restricted Project, Restricted Project
piotr added a comment to D156999: [Inline Spiller] Consider bundles when marking defs as dead.

I am open to suggestions how to handle this more gracefully.

Aug 3 2023, 6:40 AM · Restricted Project, Restricted Project
piotr added a comment to D156999: [Inline Spiller] Consider bundles when marking defs as dead.

The test case is dependent on AMDGPU-specific D154083, which rematerializes instructions with wide registers. I was not able to observe the erroneous behaviour without it, but should be possible to trigger it somehow.

Aug 3 2023, 6:39 AM · Restricted Project, Restricted Project
piotr added a comment to D156998: [NFC] Pre-commit test for dead bundle bug.

Pre-committing the test for bug fix in D156999. The test is dependent on AMDGPU-specific D154083, which rematerializes instructions with wide registers.

Aug 3 2023, 6:37 AM · Restricted Project, Restricted Project
piotr requested review of D156999: [Inline Spiller] Consider bundles when marking defs as dead.
Aug 3 2023, 6:34 AM · Restricted Project, Restricted Project
piotr requested review of D156998: [NFC] Pre-commit test for dead bundle bug.
Aug 3 2023, 6:33 AM · Restricted Project, Restricted Project

Jul 28 2023

piotr added a comment to D156520: [TII] NFCI: Simplify the interface for isTriviallyReMaterializable.

Thanks, AMD changes LGTM. I spotted this weirdness recently when staring at some remat code for AMDGPU. (It's not a serious issue for us, because the generic check is really conservative for our target, and we typically do not want to block the remat when the generic check returns true).

Jul 28 2023, 6:52 AM · Restricted Project, Restricted Project

Jul 26 2023

piotr added inline comments to D154083: [AMDGPU] Rematerialize scalar loads.
Jul 26 2023, 11:59 PM · Restricted Project, Restricted Project

Jul 21 2023

piotr added inline comments to D154083: [AMDGPU] Rematerialize scalar loads.
Jul 21 2023, 1:01 AM · Restricted Project, Restricted Project
piotr added inline comments to D154083: [AMDGPU] Rematerialize scalar loads.
Jul 21 2023, 12:59 AM · Restricted Project, Restricted Project
piotr added a comment to D154083: [AMDGPU] Rematerialize scalar loads.

Could really use a MIR test that shows this. Also would be nice to have some evil cases, where the result register is tied to the input pointer register

This patch is now based on a test update (https://reviews.llvm.org/D154816), where I am also adding a new test that exercises the shrinking - test_remat_s_load_dword_immx16_subreg.

Can you describe the evil case(s) in more detail? Do you mean S_LOAD_DWORDX16_IMM with tied-def, or something else?

Yes. I don't think subregisters with tied operands are particularly well defined, but I was thinking something like %0:sreg_256 = S_LOAD_DWORDX16 %0.sub0_sub1

Jul 21 2023, 12:38 AM · Restricted Project, Restricted Project
piotr updated the summary of D154816: [AMDGPU] Update test.
Jul 21 2023, 12:36 AM · Restricted Project, Restricted Project
piotr updated the diff for D154816: [AMDGPU] Update test.

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.

Jul 21 2023, 12:36 AM · Restricted Project, Restricted Project

Jul 19 2023

piotr updated the diff for D154816: [AMDGPU] Update test.

Added more tests.

Jul 19 2023, 4:05 PM · Restricted Project, Restricted Project
piotr updated the diff for D154083: [AMDGPU] Rematerialize scalar loads.

Ran extensive testing on graphics workloads, which uncovered some bugs. Added fixes and more tests for those interesting cases in D154816.

Jul 19 2023, 4:04 PM · Restricted Project, Restricted Project

Jul 12 2023

piotr added a comment to D154083: [AMDGPU] Rematerialize scalar loads.

Could really use a MIR test that shows this. Also would be nice to have some evil cases, where the result register is tied to the input pointer register

Jul 12 2023, 7:52 AM · Restricted Project, Restricted Project
piotr added inline comments to D154825: [InstCombine][NFC] Rename check run label.
Jul 12 2023, 1:38 AM · Restricted Project, Restricted Project
piotr committed rG608cf17b734b: [InstCombine][NFC] Rename check run label (authored by piotr).
[InstCombine][NFC] Rename check run label
Jul 12 2023, 1:35 AM · Restricted Project, Restricted Project
piotr closed D154825: [InstCombine][NFC] Rename check run label.
Jul 12 2023, 1:34 AM · Restricted Project, Restricted Project

Jul 10 2023

piotr added inline comments to D154825: [InstCombine][NFC] Rename check run label.
Jul 10 2023, 3:07 AM · Restricted Project, Restricted Project
piotr added a reviewer for D154825: [InstCombine][NFC] Rename check run label: nikic.
Jul 10 2023, 2:40 AM · Restricted Project, Restricted Project
piotr added a comment to D154825: [InstCombine][NFC] Rename check run label.

Note: I am aware the "DEFAULT" prefix is currently unused, but was not sure if the intention was for it to be used in the near future - so I am leaving it there for now.

Jul 10 2023, 2:40 AM · Restricted Project, Restricted Project
piotr requested review of D154825: [InstCombine][NFC] Rename check run label.
Jul 10 2023, 2:38 AM · Restricted Project, Restricted Project
piotr added inline comments to D154083: [AMDGPU] Rematerialize scalar loads.
Jul 10 2023, 12:51 AM · Restricted Project, Restricted Project
piotr requested review of D154816: [AMDGPU] Update test.
Jul 10 2023, 12:48 AM · Restricted Project, Restricted Project
piotr updated the diff for D154083: [AMDGPU] Rematerialize scalar loads.
  • Addressed review comments.
  • Relaxed check to include all invariant loads, not only dereferenceable ones.
  • Rebased patch over the commit with new/changed tests.
  • Updated MMO with new size in the shrinking path.
Jul 10 2023, 12:44 AM · Restricted Project, Restricted Project

Jul 4 2023

piotr accepted D154432: [AMDGPU][AsmParser] Simplify the implementation of SWZ operands..
Jul 4 2023, 5:16 AM · Restricted Project, Restricted Project
piotr added inline comments to D68200: [AMDGPU] Extend buffer intrinsics with swizzling.
Jul 4 2023, 5:15 AM · Restricted Project, Restricted Project
piotr added inline comments to D68200: [AMDGPU] Extend buffer intrinsics with swizzling.
Jul 4 2023, 2:02 AM · Restricted Project, Restricted Project

Jun 30 2023

piotr added inline comments to D154083: [AMDGPU] Rematerialize scalar loads.
Jun 30 2023, 9:15 AM · Restricted Project, Restricted Project

Jun 29 2023

piotr added inline comments to D154083: [AMDGPU] Rematerialize scalar loads.
Jun 29 2023, 10:01 AM · Restricted Project, Restricted Project
piotr added reviewers for D154083: [AMDGPU] Rematerialize scalar loads: arsenm, rampitec, foad.
Jun 29 2023, 7:24 AM · Restricted Project, Restricted Project
piotr added inline comments to D154083: [AMDGPU] Rematerialize scalar loads.
Jun 29 2023, 7:23 AM · Restricted Project, Restricted Project
piotr requested review of D154083: [AMDGPU] Rematerialize scalar loads.
Jun 29 2023, 6:59 AM · Restricted Project, Restricted Project

Jun 6 2023

piotr added a comment to D149842: Scalarizer: limit scalarization for small element types.

Having looked at some real-world graphics content on AMDGPU (with ScalarizeMinBits = 32), I can confirm the usefulness of this patch. I can see more packed instructions generated (for example v_pk_add_f16, v_pk_mul_f16, v_pk_fma_f16).

Jun 6 2023, 7:16 AM · Restricted Project, Restricted Project
piotr accepted D11621: AMDGPU: Mark scalar loads as rematerializable.

Yes - this is NFC, but paves the way for other changes.

Jun 6 2023, 6:22 AM · Restricted Project, Restricted Project

Jun 2 2023

piotr added a comment to D147547: [AMDGPU] Add buffer intrinsics that take resources as pointers.

LGTM with a nit, but please wait for Matt's approval.

Jun 2 2023, 2:53 AM · Restricted Project, Restricted Project

May 30 2023

piotr added a comment to D151703: [AMDGPU][LSV] Restrict forming extra large vectors.

What we should do is teach rematerialization to split scalar loads

May 30 2023, 6:25 AM · Restricted Project, Restricted Project
piotr added a reviewer for D151703: [AMDGPU][LSV] Restrict forming extra large vectors: Restricted Project.
May 30 2023, 5:00 AM · Restricted Project, Restricted Project
piotr requested review of D151703: [AMDGPU][LSV] Restrict forming extra large vectors.
May 30 2023, 4:58 AM · Restricted Project, Restricted Project

May 15 2023

piotr committed rG7322d35476c3: [ValueTracking] Fix computeKnownFPClass with canonicalize (authored by piotr).
[ValueTracking] Fix computeKnownFPClass with canonicalize
May 15 2023, 5:51 AM · Restricted Project, Restricted Project
piotr closed D150556: [ValueTracking] Fix computeKnownFPClass with canonicalize.
May 15 2023, 5:51 AM · Restricted Project, Restricted Project
piotr added a reviewer for D150556: [ValueTracking] Fix computeKnownFPClass with canonicalize: dstuttard.
May 15 2023, 5:23 AM · Restricted Project, Restricted Project
piotr added reviewers for D150556: [ValueTracking] Fix computeKnownFPClass with canonicalize: arsenm, jcranmer-intel, nikic, foad.
May 15 2023, 5:23 AM · Restricted Project, Restricted Project
piotr requested review of D150556: [ValueTracking] Fix computeKnownFPClass with canonicalize.
May 15 2023, 5:22 AM · Restricted Project, Restricted Project

Apr 7 2023

piotr added a comment to D147547: [AMDGPU] Add buffer intrinsics that take resources as pointers.

Thanks for working on this. Just added a couple of nits.

Apr 7 2023, 4:01 AM · Restricted Project, Restricted Project

Mar 30 2023

piotr added a comment to D147216: [AMDGPU] Reduce repetition in SMEM instruction definitions.

LGTM - looks cleaner than the version before the change.

Mar 30 2023, 4:31 AM · Restricted Project, Restricted Project

Mar 16 2023

piotr abandoned D137066: [AMDGPU] Add amdgcn_s_buffer_load_imm intrinsic.
Mar 16 2023, 1:48 AM · Restricted Project, Restricted Project
piotr added a comment to D137066: [AMDGPU] Add amdgcn_s_buffer_load_imm intrinsic.

What is the current state of this?

Mar 16 2023, 1:48 AM · Restricted Project, Restricted Project

Mar 3 2023

piotr added a comment to D145216: [llvm][Uniformity] correctly use a vector as a set by uniqifying elements.

Would be good to add a test case, but I do realize this can be tricky.

Mar 3 2023, 1:52 AM · Restricted Project, Restricted Project

Mar 1 2023

piotr added a comment to rG46db90cc71d1: [SCEV] `MatchBinaryOp()`: try to recognize `or` as `add`-in-disguise (w/ no….

We're seeing some regression coming from this change for AMDGPU backend.

Mar 1 2023, 7:18 AM · Restricted Project, Restricted Project

Feb 27 2023

piotr accepted D144890: [AMDGPU][NFC] Eliminate the u32imm operand definition..
Feb 27 2023, 1:39 PM · Restricted Project, Restricted Project

Feb 23 2023

piotr committed rGab174c57f417: [AMDGPU] Add more tests for buffer intrinsics (authored by piotr).
[AMDGPU] Add more tests for buffer intrinsics
Feb 23 2023, 5:46 AM · Restricted Project, Restricted Project
piotr committed rG51a49ec52a08: [AMDGPU] Clean up MUBUF immediate offset (authored by piotr).
[AMDGPU] Clean up MUBUF immediate offset
Feb 23 2023, 4:30 AM · Restricted Project, Restricted Project
piotr closed D144628: [AMDGPU] Clean up MUBUF immediate offset.
Feb 23 2023, 4:30 AM · Restricted Project, Restricted Project
piotr updated the summary of D144628: [AMDGPU] Clean up MUBUF immediate offset.
Feb 23 2023, 2:59 AM · Restricted Project, Restricted Project
piotr updated the diff for D144628: [AMDGPU] Clean up MUBUF immediate offset.

(fixing the spelling error)

Feb 23 2023, 2:58 AM · Restricted Project, Restricted Project
piotr requested review of D144628: [AMDGPU] Clean up MUBUF immediate offset.
Feb 23 2023, 2:55 AM · Restricted Project, Restricted Project
piotr committed rGa3d7b3121cfd: [AMDGPU][NFC] Add getMaxMUBUFImmOffset (authored by piotr).
[AMDGPU][NFC] Add getMaxMUBUFImmOffset
Feb 23 2023, 2:35 AM · Restricted Project, Restricted Project
piotr closed D144623: [AMDGPU][NFC] Add getMaxMUBUFImmOffset.
Feb 23 2023, 2:35 AM · Restricted Project, Restricted Project
piotr added reviewers for D144623: [AMDGPU][NFC] Add getMaxMUBUFImmOffset: foad, arsenm, rampitec, mbrkusanin.
Feb 23 2023, 12:44 AM · Restricted Project, Restricted Project
piotr requested review of D144623: [AMDGPU][NFC] Add getMaxMUBUFImmOffset.
Feb 23 2023, 12:42 AM · Restricted Project, Restricted Project

Feb 22 2023

piotr committed rG1b9b4f3bfa8b: [AMDGPU][NFC] Convert llvm.amdgcn tests to autogen (authored by piotr).
[AMDGPU][NFC] Convert llvm.amdgcn tests to autogen
Feb 22 2023, 11:21 PM · Restricted Project, Restricted Project

Feb 14 2023

piotr updated subscribers of D143945: [AMDGPU] Add legalization case for PTR_ADD on buffer pointers.
Feb 14 2023, 12:40 AM · Restricted Project, Restricted Project

Feb 3 2023

piotr committed rG627aee2c7651: [AMDGPU] Make offset a 32-bit operand (authored by piotr).
[AMDGPU] Make offset a 32-bit operand
Feb 3 2023, 2:45 AM · Restricted Project, Restricted Project
piotr closed D143174: [AMDGPU] Make offset a 32-bit operand.
Feb 3 2023, 2:45 AM · Restricted Project, Restricted Project

Feb 2 2023

piotr added a comment to D143174: [AMDGPU] Make offset a 32-bit operand.

Will do extra testing.

Feb 2 2023, 6:38 AM · Restricted Project, Restricted Project
piotr added inline comments to D143174: [AMDGPU] Make offset a 32-bit operand.
Feb 2 2023, 6:03 AM · Restricted Project, Restricted Project
piotr added reviewers for D143174: [AMDGPU] Make offset a 32-bit operand: foad, arsenm, mbrkusanin, matejam, rampitec.
Feb 2 2023, 5:30 AM · Restricted Project, Restricted Project
piotr requested review of D143174: [AMDGPU] Make offset a 32-bit operand.
Feb 2 2023, 5:29 AM · Restricted Project, Restricted Project

Jan 25 2023

piotr committed rGcefc8428ed3d: [AMDGPU][MC][NFC] MUBUF code cleanup (authored by piotr).
[AMDGPU][MC][NFC] MUBUF code cleanup
Jan 25 2023, 12:53 AM · Restricted Project, Restricted Project

Dec 21 2022

piotr committed rGcce3cd203e66: [AMDGPU][MC][NFC] MUBUF/MTBUF code cleanup (authored by piotr).
[AMDGPU][MC][NFC] MUBUF/MTBUF code cleanup
Dec 21 2022, 1:09 AM · Restricted Project, Restricted Project
piotr closed D140410: [AMDGPU][MC][NFC] MUBUF/MTBUF code cleanup.
Dec 21 2022, 1:09 AM · Restricted Project, Restricted Project

Dec 20 2022

piotr added reviewers for D140410: [AMDGPU][MC][NFC] MUBUF/MTBUF code cleanup: dp, rampitec, foad, s.egerton, Joe_Nash.
Dec 20 2022, 8:53 AM · Restricted Project, Restricted Project
piotr requested review of D140410: [AMDGPU][MC][NFC] MUBUF/MTBUF code cleanup.
Dec 20 2022, 8:50 AM · Restricted Project, Restricted Project

Dec 7 2022

piotr committed rG1e3abd82b9ed: [AMDGPU] Fix wide spills (authored by piotr).
[AMDGPU] Fix wide spills
Dec 7 2022, 1:24 AM · Restricted Project, Restricted Project
piotr closed D139203: [AMDGPU] Fix wide spills.
Dec 7 2022, 1:24 AM · Restricted Project, Restricted Project

Dec 6 2022

piotr updated the diff for D139203: [AMDGPU] Fix wide spills.

Updating patch to address two more issues detected by running tests in debug mode:

  • wrong opcodes used in getVGPRSpillSaveOpcode in the original patch
  • tests using insufficient spill slot size in my patch
Dec 6 2022, 2:14 AM · Restricted Project, Restricted Project

Dec 5 2022

piotr updated the diff for D139203: [AMDGPU] Fix wide spills.

Thanks - the extra set of tests has uncovered another issue in code.

Dec 5 2022, 6:09 AM · Restricted Project, Restricted Project

Dec 2 2022

piotr added a reviewer for D139203: [AMDGPU] Fix wide spills: matejam.
Dec 2 2022, 8:04 AM · Restricted Project, Restricted Project
piotr added reviewers for D139203: [AMDGPU] Fix wide spills: mbrkusanin, foad, dstuttard, arsenm, rampitec.
Dec 2 2022, 7:48 AM · Restricted Project, Restricted Project
piotr requested review of D139203: [AMDGPU] Fix wide spills.
Dec 2 2022, 7:47 AM · Restricted Project, Restricted Project

Nov 22 2022

piotr added a comment to D137066: [AMDGPU] Add amdgcn_s_buffer_load_imm intrinsic.

Ping.

Nov 22 2022, 3:16 AM · Restricted Project, Restricted Project