Page MenuHomePhabricator

krzysz00 (Krzysztof Drewniak)
User

Projects

User does not belong to any projects.

User Details

User Since
May 3 2021, 10:02 AM (66 w, 1 d)

Recent Activity

Yesterday

krzysz00 abandoned D131406: [mlir][Affine] Fix non-converging empty loop pattern.

Ok, looks like this wasn't the bug.

Mon, Aug 8, 8:44 AM · Restricted Project, Restricted Project
krzysz00 requested review of D131406: [mlir][Affine] Fix non-converging empty loop pattern.
Mon, Aug 8, 8:36 AM · Restricted Project, Restricted Project

Thu, Aug 4

krzysz00 committed rG63295622491a: [mlir][AMDGPU] Explicitly truncate memory addresses in buffer ops (authored by krzysz00).
[mlir][AMDGPU] Explicitly truncate memory addresses in buffer ops
Thu, Aug 4, 12:43 PM · Restricted Project, Restricted Project
krzysz00 closed D131016: [mlir][AMDGPU] Explicitly truncate memory addresses in buffer ops.
Thu, Aug 4, 12:42 PM · Restricted Project, Restricted Project
krzysz00 updated the diff for D131016: [mlir][AMDGPU] Explicitly truncate memory addresses in buffer ops.

Added test

Thu, Aug 4, 9:40 AM · Restricted Project, Restricted Project
krzysz00 added a reviewer for D131016: [mlir][AMDGPU] Explicitly truncate memory addresses in buffer ops: ThomasRaoux.
Thu, Aug 4, 8:40 AM · Restricted Project, Restricted Project

Tue, Aug 2

krzysz00 committed rGc2fc8d9b95bd: [mlir][GPU] Allow bare pointer memrefs when calling GPU kernels (authored by krzysz00).
[mlir][GPU] Allow bare pointer memrefs when calling GPU kernels
Tue, Aug 2, 1:59 PM · Restricted Project, Restricted Project
krzysz00 closed D130716: [mlir][GPU] Allow bare pointer memrefs when calling GPU kernels.
Tue, Aug 2, 1:58 PM · Restricted Project, Restricted Project
krzysz00 requested review of D131016: [mlir][AMDGPU] Explicitly truncate memory addresses in buffer ops.
Tue, Aug 2, 1:31 PM · Restricted Project, Restricted Project
krzysz00 added a comment to D130716: [mlir][GPU] Allow bare pointer memrefs when calling GPU kernels.

(Also , @ThomasRaoux - should this be added to the NVVM conversion? I'm not doing it right now because I can't test the change, but I thought I'd raise the possibility)

Tue, Aug 2, 9:53 AM · Restricted Project, Restricted Project
krzysz00 updated the diff for D130716: [mlir][GPU] Allow bare pointer memrefs when calling GPU kernels.

Adress review comments

Tue, Aug 2, 9:51 AM · Restricted Project, Restricted Project

Mon, Aug 1

krzysz00 committed rG938fe9f277c6: [mlir][Arith] Fix up integer range inference for truncation (authored by krzysz00).
[mlir][Arith] Fix up integer range inference for truncation
Mon, Aug 1, 12:30 PM · Restricted Project, Restricted Project
krzysz00 closed D130501: [mlir][Arith] Fix up integer range inference for truncation.
Mon, Aug 1, 12:30 PM · Restricted Project, Restricted Project
krzysz00 updated the diff for D130501: [mlir][Arith] Fix up integer range inference for truncation.

Added comments, fixed bug, updated tests

Mon, Aug 1, 9:37 AM · Restricted Project, Restricted Project

Thu, Jul 28

krzysz00 requested review of D130716: [mlir][GPU] Allow bare pointer memrefs when calling GPU kernels.
Thu, Jul 28, 9:50 AM · Restricted Project, Restricted Project

Mon, Jul 25

krzysz00 added inline comments to D130501: [mlir][Arith] Fix up integer range inference for truncation.
Mon, Jul 25, 11:08 AM · Restricted Project, Restricted Project
krzysz00 requested review of D130501: [mlir][Arith] Fix up integer range inference for truncation.
Mon, Jul 25, 10:25 AM · Restricted Project, Restricted Project

Thu, Jul 14

krzysz00 committed rGbc61cc9a2db5: [mlir][AMDGPU] Add lds_barrier op (authored by krzysz00).
[mlir][AMDGPU] Add lds_barrier op
Thu, Jul 14, 1:45 PM · Restricted Project, Restricted Project
krzysz00 closed D129522: [mlir][AMDGPU] Add lds_barrier op.
Thu, Jul 14, 1:45 PM · Restricted Project, Restricted Project
krzysz00 added a comment to D129522: [mlir][AMDGPU] Add lds_barrier op.

@kerbowa I'd like to still land this because the lds_barrier is a useful abstraction over whatever combination of (fence + barrier)/inline assembly/... you end up needing to use to implement it.

Thu, Jul 14, 8:22 AM · Restricted Project, Restricted Project

Tue, Jul 12

krzysz00 added inline comments to D129522: [mlir][AMDGPU] Add lds_barrier op.
Tue, Jul 12, 11:51 AM · Restricted Project, Restricted Project
krzysz00 committed rGd6ef3d20b4e3: [mlir] Remove VectorToROCDL (authored by krzysz00).
[mlir] Remove VectorToROCDL
Tue, Jul 12, 8:21 AM · Restricted Project, Restricted Project, Restricted Project
krzysz00 closed D129308: [mlir] Remove VectorToROCDL.
Tue, Jul 12, 8:21 AM · Restricted Project, Restricted Project, Restricted Project
krzysz00 added a comment to D129522: [mlir][AMDGPU] Add lds_barrier op.

Re the above, you mentioned gfx90a+ ... what about gfx908?

Tue, Jul 12, 6:56 AM · Restricted Project, Restricted Project

Mon, Jul 11

krzysz00 updated subscribers of D129522: [mlir][AMDGPU] Add lds_barrier op.

This is a single op that expands to both a waitcnt on LDS and a barrier. I
can go digging tomorrow for what we used to lower this to some sort of
fence and a barrier, but I recall ( @whchung who may have more detail) that
using this bit of inline assembly to work around what I think was the lack
of an LDS-only fence gave a noticable performance increase

Mon, Jul 11, 4:24 PM · Restricted Project, Restricted Project
krzysz00 requested review of D129522: [mlir][AMDGPU] Add lds_barrier op.
Mon, Jul 11, 4:12 PM · Restricted Project, Restricted Project
krzysz00 added a comment to D128630: [MLIR][TOSA] Fix converting tosa.clamp and tosa.relu to linalg.

And I do have the bits, so landed. Thanks for the external pair of eyes!

Mon, Jul 11, 10:19 AM · Restricted Project, Restricted Project
krzysz00 committed rG6e8e91a7b63c: [MLIR][TOSA] Fix converting tosa.clamp and tosa.relu to linalg (authored by jungpark-mlir).
[MLIR][TOSA] Fix converting tosa.clamp and tosa.relu to linalg
Mon, Jul 11, 10:19 AM · Restricted Project, Restricted Project
krzysz00 closed D128630: [MLIR][TOSA] Fix converting tosa.clamp and tosa.relu to linalg.
Mon, Jul 11, 10:18 AM · Restricted Project, Restricted Project

Jul 7 2022

krzysz00 committed rGdb590549a990: [mlir][AMDGPU] Use the correct values for OOB_SELECT on gfx10 (authored by krzysz00).
[mlir][AMDGPU] Use the correct values for OOB_SELECT on gfx10
Jul 7 2022, 2:23 PM · Restricted Project, Restricted Project
krzysz00 closed D129320: [mlir][AMDGPU] Use the correct values for OOB_SELECT on gfx10.
Jul 7 2022, 2:23 PM · Restricted Project, Restricted Project
krzysz00 requested review of D129320: [mlir][AMDGPU] Use the correct values for OOB_SELECT on gfx10.
Jul 7 2022, 12:10 PM · Restricted Project, Restricted Project
krzysz00 added a comment to D128630: [MLIR][TOSA] Fix converting tosa.clamp and tosa.relu to linalg.

@eric-k256 Would you be willing to approve here?

Jul 7 2022, 10:59 AM · Restricted Project, Restricted Project
krzysz00 requested review of D129308: [mlir] Remove VectorToROCDL.
Jul 7 2022, 10:41 AM · Restricted Project, Restricted Project, Restricted Project
krzysz00 committed rGcab44c515c63: [mlir][AMDGPU] Add --chipset option to AMDGPUToROCDL (authored by krzysz00).
[mlir][AMDGPU] Add --chipset option to AMDGPUToROCDL
Jul 7 2022, 7:58 AM · Restricted Project, Restricted Project
krzysz00 closed D129228: [mlir][AMDGPU] Add --chipset option to AMDGPUToROCDL.
Jul 7 2022, 7:58 AM · Restricted Project, Restricted Project
krzysz00 added inline comments to D129228: [mlir][AMDGPU] Add --chipset option to AMDGPUToROCDL.
Jul 7 2022, 7:57 AM · Restricted Project, Restricted Project

Jul 6 2022

krzysz00 requested review of D129228: [mlir][AMDGPU] Add --chipset option to AMDGPUToROCDL.
Jul 6 2022, 2:29 PM · Restricted Project, Restricted Project

Jul 5 2022

krzysz00 abandoned D126911: [mlir] Add integer range analysis to SCCP.

Since @Mogball is handling this in his dataflow analysis rework, abandoning.

Jul 5 2022, 8:44 AM · Restricted Project, Restricted Project

Jul 4 2022

krzysz00 accepted D129036: [mlir] Add InferIntRangeInterface to gpu.launch.

Thanks for doing all the research on this!

Jul 4 2022, 10:37 AM · Restricted Project, Restricted Project
krzysz00 added a comment to D129036: [mlir] Add InferIntRangeInterface to gpu.launch.

Ah, I missed that the one with 128 was laneId, not any of the other ones - my bad for not reading carefully.

Jul 4 2022, 9:09 AM · Restricted Project, Restricted Project
krzysz00 requested changes to D129036: [mlir] Add InferIntRangeInterface to gpu.launch.

I agree that getting anything more than very conservative bounds in an outlined kernel will be tricky and so it might not be worth doing (right now).

Jul 4 2022, 7:57 AM · Restricted Project, Restricted Project

Jul 2 2022

krzysz00 added a comment to D129036: [mlir] Add InferIntRangeInterface to gpu.launch.

Thanks for digging up the sources on those bounds!

Jul 2 2022, 8:55 AM · Restricted Project, Restricted Project
krzysz00 added inline comments to D129036: [mlir] Add InferIntRangeInterface to gpu.launch.
Jul 2 2022, 7:14 AM · Restricted Project, Restricted Project
krzysz00 added a comment to D129036: [mlir] Add InferIntRangeInterface to gpu.launch.

Two initial comments;

  • Do we actually know that the number of blocks/threads is bounded above by 2^31? For one thing, the HIP (and presumably CUDA) launch APIs take uint32_t and so the bound could be 2^32
  • Could you find a way to propagate these into (outlined) kernels, so that we get bounds on gpu.block_id and friends?
Jul 2 2022, 7:11 AM · Restricted Project, Restricted Project

Jun 30 2022

krzysz00 accepted D128866: [mlir] Swap integer range inference to the new framework.
Jun 30 2022, 8:59 AM · Restricted Project, Restricted Project
krzysz00 added a comment to D128866: [mlir] Swap integer range inference to the new framework.

LGTM, at least as far as the integer analysis parts go.

Jun 30 2022, 8:58 AM · Restricted Project, Restricted Project

Jun 29 2022

krzysz00 added a comment to D128866: [mlir] Swap integer range inference to the new framework.

So, overall comments - especially since it looks like all the details are basically fine ... wouldn't it still be a good idea to have a wrapper around IntRangeAnalysis that hides the fact that it's a dataflow analysis? Or are we abandoning that wrapper here?

Jun 29 2022, 9:04 PM · Restricted Project, Restricted Project

Jun 28 2022

krzysz00 added a comment to D128630: [MLIR][TOSA] Fix converting tosa.clamp and tosa.relu to linalg.

@jpienaar Would you be willing to approve this, or do you not make sense as a reviewer?

Jun 28 2022, 8:25 AM · Restricted Project, Restricted Project

Jun 27 2022

krzysz00 added a comment to rG0ecea5b2d34b: [gdb-scripts] Fix PointerIntPairPrinter.to_string after D127969.

Thanks!

Jun 27 2022, 1:13 PM · Restricted Project, Restricted Project
krzysz00 added a comment to D128630: [MLIR][TOSA] Fix converting tosa.clamp and tosa.relu to linalg.

In "things I'm just now noticing now that we're trying to send this upstream", should we add a test?

Jun 27 2022, 12:43 PM · Restricted Project, Restricted Project

Jun 21 2022

krzysz00 committed rG7c5c4e781b8c: [gdb-scripts] Add to_string methods to printer implementations (authored by krzysz00).
[gdb-scripts] Add to_string methods to printer implementations
Jun 21 2022, 9:09 AM · Restricted Project, Restricted Project, Restricted Project
krzysz00 closed D127969: [gdb-scripts] Add to_string methods to printer implementations.
Jun 21 2022, 9:09 AM · Restricted Project, Restricted Project, Restricted Project

Jun 20 2022

krzysz00 added a reviewer for D127969: [gdb-scripts] Add to_string methods to printer implementations: dblaikie.
Jun 20 2022, 8:05 AM · Restricted Project, Restricted Project, Restricted Project
krzysz00 committed rG8e61fdc72780: [mlir][ROCDL] Define MLIR wrappers around new MFMA intrinsics (authored by krzysz00).
[mlir][ROCDL] Define MLIR wrappers around new MFMA intrinsics
Jun 20 2022, 8:04 AM · Restricted Project, Restricted Project
krzysz00 closed D128079: [mlir][ROCDL] Define MLIR wrappers around new MFMA intrinsics.
Jun 20 2022, 8:03 AM · Restricted Project, Restricted Project
krzysz00 committed rGe49ae6284c38: [mlir][Arith] Make --unsigned-when-equivalent use dialect conversion (authored by krzysz00).
[mlir][Arith] Make --unsigned-when-equivalent use dialect conversion
Jun 20 2022, 8:03 AM · Restricted Project, Restricted Project
krzysz00 closed D128096: [mlir][Arith] Make --unsigned-when-equivalent use dialect conversion.
Jun 20 2022, 8:03 AM · Restricted Project, Restricted Project
krzysz00 updated the diff for D128096: [mlir][Arith] Make --unsigned-when-equivalent use dialect conversion.

Review comments

Jun 20 2022, 8:02 AM · Restricted Project, Restricted Project

Jun 17 2022

krzysz00 requested review of D128096: [mlir][Arith] Make --unsigned-when-equivalent use dialect conversion.
Jun 17 2022, 1:41 PM · Restricted Project, Restricted Project
krzysz00 requested review of D128079: [mlir][ROCDL] Define MLIR wrappers around new MFMA intrinsics.
Jun 17 2022, 11:34 AM · Restricted Project, Restricted Project
krzysz00 abandoned D127244: [mlir][AMDGPU] Add `mfma` operation to wrap mfma intrinsics..

Per feedback here, I'm abandoning this revision in favor of

  1. A new revision that adds the new ROCDL intrinsics from LLVM but doesn't touch AMDGPU
  2. Going back downstream to design a better mfma operation that looks like something like mfma {k = K, m = M, n = N, ...} %c = %a * %b.
Jun 17 2022, 8:38 AM · Restricted Project, Restricted Project

Jun 16 2022

krzysz00 requested review of D127969: [gdb-scripts] Add to_string methods to printer implementations.
Jun 16 2022, 7:41 AM · Restricted Project, Restricted Project, Restricted Project

Jun 15 2022

krzysz00 added a comment to D124024: [mlir][Arith] Pass to switch signed ops for equivalent unsigned ones.

Should cleaning this up be a new revision or is there a way to re-open this one?

Jun 15 2022, 11:05 AM · Restricted Project, Restricted Project
krzysz00 added inline comments to D124024: [mlir][Arith] Pass to switch signed ops for equivalent unsigned ones.
Jun 15 2022, 8:28 AM · Restricted Project, Restricted Project

Jun 14 2022

krzysz00 committed rGb0b00432093b: [mlir][Arith] Pass to switch signed ops for equivalent unsigned ones (authored by krzysz00).
[mlir][Arith] Pass to switch signed ops for equivalent unsigned ones
Jun 14 2022, 2:18 PM · Restricted Project, Restricted Project
krzysz00 closed D124024: [mlir][Arith] Pass to switch signed ops for equivalent unsigned ones.
Jun 14 2022, 2:18 PM · Restricted Project, Restricted Project
krzysz00 updated the diff for D124024: [mlir][Arith] Pass to switch signed ops for equivalent unsigned ones.

Rebase

Jun 14 2022, 11:48 AM · Restricted Project, Restricted Project
krzysz00 committed rG75bfc6f29579: [mlir][Arith] Implement InferIntRangeInterface for arithmetic ops (authored by krzysz00).
[mlir][Arith] Implement InferIntRangeInterface for arithmetic ops
Jun 14 2022, 11:30 AM · Restricted Project, Restricted Project
krzysz00 closed D124022: [mlir][Arith] Implement InferIntRangeInterface for arithmetic ops.
Jun 14 2022, 11:30 AM · Restricted Project, Restricted Project
krzysz00 updated the diff for D124022: [mlir][Arith] Implement InferIntRangeInterface for arithmetic ops.

unsigned int -> unsigned throughought

Jun 14 2022, 9:47 AM · Restricted Project, Restricted Project
krzysz00 updated the diff for D124022: [mlir][Arith] Implement InferIntRangeInterface for arithmetic ops.

Apply style comments, send off last build pre-land

Jun 14 2022, 9:05 AM · Restricted Project, Restricted Project
krzysz00 updated subscribers of D127244: [mlir][AMDGPU] Add `mfma` operation to wrap mfma intrinsics..
Jun 14 2022, 8:17 AM · Restricted Project, Restricted Project

Jun 10 2022

krzysz00 added inline comments to D127244: [mlir][AMDGPU] Add `mfma` operation to wrap mfma intrinsics..
Jun 10 2022, 12:05 PM · Restricted Project, Restricted Project
krzysz00 committed rGa2cdb9791bf8: [mlir][AMDGPU] Set ABI version constant when linking device libs (authored by krzysz00).
[mlir][AMDGPU] Set ABI version constant when linking device libs
Jun 10 2022, 11:41 AM · Restricted Project, Restricted Project
krzysz00 closed D126913: [mlir][AMDGPU] Set ABI version constant when linking device libs.
Jun 10 2022, 11:41 AM · Restricted Project, Restricted Project
krzysz00 added a comment to D126913: [mlir][AMDGPU] Set ABI version constant when linking device libs.

The only test I can think of that could catch this is an integration test, and our buildbot isn't testing against pre-release versions of the backend

Jun 10 2022, 10:54 AM · Restricted Project, Restricted Project
krzysz00 added a comment to D124022: [mlir][Arith] Implement InferIntRangeInterface for arithmetic ops.

@Mogball , I know you LGTM'd this a while back, but it got more-or-less rewritten since then, so it could probably use another style/sanity check

Jun 10 2022, 9:01 AM · Restricted Project, Restricted Project

Jun 8 2022

krzysz00 updated the diff for D124022: [mlir][Arith] Implement InferIntRangeInterface for arithmetic ops.

Add a test closer in intent to the original while loop test

Jun 8 2022, 3:10 PM · Restricted Project, Restricted Project
krzysz00 updated the diff for D124022: [mlir][Arith] Implement InferIntRangeInterface for arithmetic ops.

Rebase onto main, fix tests to account for arith.constant inference

Jun 8 2022, 2:49 PM · Restricted Project, Restricted Project
krzysz00 added a comment to D127261: [mlir] Fix handling of some region branch terminator successors.

So yeah, from the perspective of the integer range stuff, feel free to land
this.

Jun 8 2022, 9:16 AM · Restricted Project, Restricted Project
krzysz00 accepted D127261: [mlir] Fix handling of some region branch terminator successors.

I think the semantics of DataFlowAnalysis::getSuccessorsForOperands() could be clarified ... but I'm not sure it matters, since there're plans to scrap that entire framework.

Jun 8 2022, 8:35 AM · Restricted Project, Restricted Project

Jun 7 2022

krzysz00 requested review of D127244: [mlir][AMDGPU] Add `mfma` operation to wrap mfma intrinsics..
Jun 7 2022, 12:30 PM · Restricted Project, Restricted Project
krzysz00 added a comment to D124022: [mlir][Arith] Implement InferIntRangeInterface for arithmetic ops.

(Submitting inline comments)

Jun 7 2022, 11:09 AM · Restricted Project, Restricted Project
krzysz00 updated the diff for D124022: [mlir][Arith] Implement InferIntRangeInterface for arithmetic ops.

Add block comments, remove IntPair

Jun 7 2022, 11:09 AM · Restricted Project, Restricted Project

Jun 6 2022

krzysz00 added a comment to D126444: [mlir]Implement SoftwareBF16 to handle the bf16 type.

To re-raise things, what's the right venue for discussing bf16 emulation in LLVM?

Jun 6 2022, 8:11 AM · Restricted Project, Restricted Project
krzysz00 added a comment to D124022: [mlir][Arith] Implement InferIntRangeInterface for arithmetic ops.

@rriddle @Mogball ?

Jun 6 2022, 8:09 AM · Restricted Project, Restricted Project

Jun 3 2022

krzysz00 added a reverting change for rG4e5ce2056e3e: Revert "[mlir] Add integer range inference analysis": rG95aff23e2921: Re-land "[mlir] Add integer range inference analysis"".
Jun 3 2022, 10:14 AM · Restricted Project, Restricted Project
krzysz00 committed rG95aff23e2921: Re-land "[mlir] Add integer range inference analysis"" (authored by krzysz00).
Re-land "[mlir] Add integer range inference analysis""
Jun 3 2022, 10:14 AM · Restricted Project, Restricted Project
krzysz00 closed D126926: Re-land "[mlir] Add integer range inference analysis"".
Jun 3 2022, 10:13 AM · Restricted Project, Restricted Project
krzysz00 added a comment to D126911: [mlir] Add integer range analysis to SCCP.

I don't think we can delete old SCCP - range analysis doesn't handle vectors, for example

Jun 3 2022, 10:07 AM · Restricted Project, Restricted Project
krzysz00 added reviewers for D126926: Re-land "[mlir] Add integer range inference analysis"": mehdi_amini, rriddle.

r

Jun 3 2022, 8:57 AM · Restricted Project, Restricted Project
krzysz00 added a reverting change for rG4e5ce2056e3e: Revert "[mlir] Add integer range inference analysis": D126926: Re-land "[mlir] Add integer range inference analysis"".
Jun 3 2022, 8:56 AM · Restricted Project, Restricted Project
krzysz00 updated the diff for D126926: Re-land "[mlir] Add integer range inference analysis"".

Update commit message

Jun 3 2022, 8:56 AM · Restricted Project, Restricted Project

Jun 2 2022

krzysz00 added a comment to D124023: [mlir] Add integer range inference analysis.

I did get an email but the email only mentioned the warning, not the error.

Jun 2 2022, 4:53 PM · Restricted Project, Restricted Project
krzysz00 abandoned D126920: [NFC] Fix -Wunused-result in test dialect.

Abandoning, I thought the warnings were just warnings, there was also a linker error - there'll be a new revision to re-land this

Jun 2 2022, 2:58 PM · Restricted Project, Restricted Project
krzysz00 requested review of D126926: Re-land "[mlir] Add integer range inference analysis"".
Jun 2 2022, 2:57 PM · Restricted Project, Restricted Project
krzysz00 requested review of D126920: [NFC] Fix -Wunused-result in test dialect.
Jun 2 2022, 1:55 PM · Restricted Project, Restricted Project
krzysz00 committed rG1350c9887dca: [mlir] Add integer range inference analysis (authored by krzysz00).
[mlir] Add integer range inference analysis
Jun 2 2022, 1:24 PM · Restricted Project, Restricted Project
krzysz00 closed D124023: [mlir] Add integer range inference analysis.
Jun 2 2022, 1:24 PM · Restricted Project, Restricted Project