Page MenuHomePhabricator

ezhulenev (Eugene Zhulenev)
User

Projects

User does not belong to any projects.

User Details

User Since
Mar 6 2020, 7:42 PM (97 w, 4 d)

Recent Activity

Yesterday

ezhulenev accepted D117542: [Coroutines] Offering llvm.coro.align intrinsic.

thank you! For our MLIR use case this is exactly what we need.

Tue, Jan 18, 3:49 AM · Restricted Project

Mon, Jan 17

ezhulenev accepted D117501: [AsyncToLLVM] Align frames to 64 bytes.
Mon, Jan 17, 9:47 AM · Restricted Project

Fri, Jan 14

ezhulenev committed rG69bc334be569: [mlir] Remove getNumberOfExecutions from RegionBranchOpInterface (authored by ezhulenev).
[mlir] Remove getNumberOfExecutions from RegionBranchOpInterface
Fri, Jan 14, 1:15 PM
ezhulenev closed D117347: [mlir] Remove getNumberOfExecutions from RegionBranchOpInterface.
Fri, Jan 14, 1:15 PM · Restricted Project
ezhulenev updated the summary of D117347: [mlir] Remove getNumberOfExecutions from RegionBranchOpInterface.
Fri, Jan 14, 11:11 AM · Restricted Project
ezhulenev requested review of D117347: [mlir] Remove getNumberOfExecutions from RegionBranchOpInterface.
Fri, Jan 14, 11:08 AM · Restricted Project

Thu, Jan 13

ezhulenev committed rG764e52f0d4b6: [DebugInfo][InstrRef] Short-circuit unnecessary preferred location map… (authored by ezhulenev).
[DebugInfo][InstrRef] Short-circuit unnecessary preferred location map…
Thu, Jan 13, 6:25 AM
ezhulenev closed D117162: [DebugInfo][InstrRef] Short-circuit unnecessary preferred location map construction.
Thu, Jan 13, 6:25 AM · Restricted Project
ezhulenev updated the diff for D117162: [DebugInfo][InstrRef] Short-circuit unnecessary preferred location map construction.

Rebase

Thu, Jan 13, 3:35 AM · Restricted Project
ezhulenev added a reviewer for D117162: [DebugInfo][InstrRef] Short-circuit unnecessary preferred location map construction: cota.
Thu, Jan 13, 2:44 AM · Restricted Project
ezhulenev updated the diff for D117162: [DebugInfo][InstrRef] Short-circuit unnecessary preferred location map construction.

Fix bugs

Thu, Jan 13, 2:20 AM · Restricted Project

Wed, Jan 12

ezhulenev added a reviewer for D117162: [DebugInfo][InstrRef] Short-circuit unnecessary preferred location map construction: mehdi_amini.
Wed, Jan 12, 2:45 PM · Restricted Project
ezhulenev requested review of D117162: [DebugInfo][InstrRef] Short-circuit unnecessary preferred location map construction.
Wed, Jan 12, 2:44 PM · Restricted Project

Fri, Dec 31

ezhulenev updated the summary of D116440: [mlir] Async: update condition for dispatching block-aligned compute function.
Fri, Dec 31, 5:12 AM · Restricted Project
ezhulenev requested review of D116440: [mlir] Async: update condition for dispatching block-aligned compute function.
Fri, Dec 31, 5:10 AM · Restricted Project

Fri, Dec 24

ezhulenev accepted D115790: [Coroutines] Set presplit attribute in Clang and mlir.
Fri, Dec 24, 8:32 AM · Restricted Project, Restricted Project, Restricted Project

Dec 19 2021

ezhulenev committed rGec0e4545caa1: Make AsyncParallelForRewrite parameterizable with a cost model which drives… (authored by bakhtiyar <bakhtiyar@x.team>).
Make AsyncParallelForRewrite parameterizable with a cost model which drives…
Dec 19 2021, 8:41 AM
ezhulenev closed D115423: Make AsyncParallelForRewrite parameterizable with a cost model which drives deciding the parallelization granularity..
Dec 19 2021, 8:41 AM · Restricted Project

Dec 17 2021

ezhulenev updated subscribers of D115790: [Coroutines] Set presplit attribute in Clang and mlir.

There are two places where in MLIR you can put an attribute to coroutine functions:

Dec 17 2021, 7:25 AM · Restricted Project, Restricted Project, Restricted Project

Dec 15 2021

ezhulenev added inline comments to D115423: Make AsyncParallelForRewrite parameterizable with a cost model which drives deciding the parallelization granularity..
Dec 15 2021, 2:14 AM · Restricted Project
ezhulenev added a comment to D115423: Make AsyncParallelForRewrite parameterizable with a cost model which drives deciding the parallelization granularity..

Build is failing: https://buildkite.com/llvm-project/premerge-checks/builds/70123#b97ba1d4-10ba-4287-899a-0bcf5260aefe

Dec 15 2021, 2:10 AM · Restricted Project
ezhulenev accepted D115423: Make AsyncParallelForRewrite parameterizable with a cost model which drives deciding the parallelization granularity..
Dec 15 2021, 2:05 AM · Restricted Project

Dec 14 2021

ezhulenev added inline comments to D115423: Make AsyncParallelForRewrite parameterizable with a cost model which drives deciding the parallelization granularity..
Dec 14 2021, 4:28 AM · Restricted Project

Dec 9 2021

ezhulenev accepted D115423: Make AsyncParallelForRewrite parameterizable with a cost model which drives deciding the parallelization granularity..
Dec 9 2021, 3:19 PM · Restricted Project
ezhulenev committed rG49ce40e9ab25: [mlir] AsyncParallelFor: align block size to be a multiple of inner loops… (authored by ezhulenev).
[mlir] AsyncParallelFor: align block size to be a multiple of inner loops…
Dec 9 2021, 6:51 AM
ezhulenev closed D115436: [mlir] AsyncParallelFor: align block size to be a multiple of inner loops iterations.
Dec 9 2021, 6:51 AM · Restricted Project
ezhulenev committed rG9f151b784be0: [mlir] AsyncParallelFor: sink constants into the parallel compute function (authored by ezhulenev).
[mlir] AsyncParallelFor: sink constants into the parallel compute function
Dec 9 2021, 6:48 AM
ezhulenev closed D115263: [mlir] AsyncParallelFor: sink constants into the parallel compute function.
Dec 9 2021, 6:48 AM · Restricted Project
ezhulenev updated the diff for D115436: [mlir] AsyncParallelFor: align block size to be a multiple of inner loops iterations.

Reset number of unrallable loops if decide not to uroll

Dec 9 2021, 3:21 AM · Restricted Project
ezhulenev updated the summary of D115436: [mlir] AsyncParallelFor: align block size to be a multiple of inner loops iterations.
Dec 9 2021, 2:53 AM · Restricted Project
ezhulenev requested review of D115436: [mlir] AsyncParallelFor: align block size to be a multiple of inner loops iterations.
Dec 9 2021, 2:52 AM · Restricted Project
ezhulenev added a reviewer for D115263: [mlir] AsyncParallelFor: sink constants into the parallel compute function: bkramer.
Dec 9 2021, 2:50 AM · Restricted Project

Dec 7 2021

ezhulenev updated the diff for D115263: [mlir] AsyncParallelFor: sink constants into the parallel compute function.

Cleanup

Dec 7 2021, 6:45 PM · Restricted Project
ezhulenev updated the summary of D115263: [mlir] AsyncParallelFor: sink constants into the parallel compute function.
Dec 7 2021, 3:21 PM · Restricted Project
ezhulenev added a reviewer for D115263: [mlir] AsyncParallelFor: sink constants into the parallel compute function: mehdi_amini.
Dec 7 2021, 3:19 PM · Restricted Project
ezhulenev updated the diff for D115263: [mlir] AsyncParallelFor: sink constants into the parallel compute function.

Fix LLVM style violations

Dec 7 2021, 12:34 PM · Restricted Project
ezhulenev added a reviewer for D115263: [mlir] AsyncParallelFor: sink constants into the parallel compute function: cota.
Dec 7 2021, 9:45 AM · Restricted Project
ezhulenev requested review of D115263: [mlir] AsyncParallelFor: sink constants into the parallel compute function.
Dec 7 2021, 9:44 AM · Restricted Project

Dec 6 2021

ezhulenev committed rG68a7c001ad9a: [mlir] Improve async parallel for tests + fix typos (authored by ezhulenev).
[mlir] Improve async parallel for tests + fix typos
Dec 6 2021, 1:28 PM
ezhulenev closed D115152: [mlir] Improve async parallel for tests + fix typos.
Dec 6 2021, 1:28 PM · Restricted Project
ezhulenev updated the diff for D115152: [mlir] Improve async parallel for tests + fix typos.

Address comments

Dec 6 2021, 11:24 AM · Restricted Project
ezhulenev updated the summary of D115152: [mlir] Improve async parallel for tests + fix typos.
Dec 6 2021, 6:20 AM · Restricted Project
ezhulenev added a reviewer for D115152: [mlir] Improve async parallel for tests + fix typos: cota.
Dec 6 2021, 6:19 AM · Restricted Project
ezhulenev updated the diff for D115152: [mlir] Improve async parallel for tests + fix typos.

Revert accidental changes

Dec 6 2021, 6:19 AM · Restricted Project
ezhulenev requested review of D115152: [mlir] Improve async parallel for tests + fix typos.
Dec 6 2021, 6:17 AM · Restricted Project

Nov 28 2021

ezhulenev added a comment to D114496: Implement a cost model to drive the lowering of scf.parallel..

It would be great to have the cost model to be a part of core MLIR, defined with op interface, but for it to be usable for all the potential targets it will require a lot of discussions (e.g. how cost model can capture GPU memory coalescing, should it be the concern of the cost model at all, or it should just track compute?). Let's start an RFC to figure out how it should look like. In the meantime we can hide the current crude implementation behind the AsyncBlockSizeComputationFunction API, and keep implementation inside the Tensorflow compiler. We can tune it for the compilation pipeline we have right now, without need to worry about the general case.

Nov 28 2021, 10:24 AM · Restricted Project

Nov 24 2021

ezhulenev committed rG7bd87a03fdf1: Promote readability by factoring out creation of min/max operation. Remove… (authored by bakhtiyar <bakhtiyar@x.team>).
Promote readability by factoring out creation of min/max operation. Remove…
Nov 24 2021, 4:17 PM
ezhulenev closed D110680: Promote readability by factoring out creation of min/max operation. Remove unnecessary divisions..
Nov 24 2021, 4:17 PM · Restricted Project

Oct 29 2021

ezhulenev added a comment to D110680: Promote readability by factoring out creation of min/max operation. Remove unnecessary divisions..

I have a failing test for this patch: Integration/Dialect/Async/CPU/test-async-parallel-for-2d.mlir

Oct 29 2021, 4:32 AM · Restricted Project
ezhulenev committed rGec32d540f808: [mlir] MathApproximations: scalars shape must be 0-rank (authored by ezhulenev).
[mlir] MathApproximations: scalars shape must be 0-rank
Oct 29 2021, 4:03 AM
ezhulenev closed D112764: [mlir] MathApproximations: scalars shape must be 0-rank.
Oct 29 2021, 4:02 AM · Restricted Project

Oct 28 2021

ezhulenev updated the summary of D112764: [mlir] MathApproximations: scalars shape must be 0-rank.
Oct 28 2021, 3:36 PM · Restricted Project
ezhulenev added a reviewer for D112764: [mlir] MathApproximations: scalars shape must be 0-rank: pifon2a.
Oct 28 2021, 3:34 PM · Restricted Project
ezhulenev updated the summary of D112764: [mlir] MathApproximations: scalars shape must be 0-rank.
Oct 28 2021, 3:34 PM · Restricted Project
ezhulenev requested review of D112764: [mlir] MathApproximations: scalars shape must be 0-rank.
Oct 28 2021, 3:34 PM · Restricted Project
ezhulenev added a comment to D112736: [mlir] MathApproximations: unroll virtual vectors into hardware vectors for ISA specific operation.

Basically add a pass something like X86vectorMathApproximationsPass?

Oct 28 2021, 2:48 PM · Restricted Project
ezhulenev added a comment to D112736: [mlir] MathApproximations: unroll virtual vectors into hardware vectors for ISA specific operation.

But avx and avx512 rsqrt have different accuracy

Oct 28 2021, 2:46 PM · Restricted Project
ezhulenev added a comment to D112736: [mlir] MathApproximations: unroll virtual vectors into hardware vectors for ISA specific operation.

Maybe we can have generic x86vector.rsqrt operation (without avx or avx512 prefix), and then it can be unrolled and lowered to ISA specific intrinsics by the x86dialect passes? /cc @aartbik

Oct 28 2021, 2:41 PM · Restricted Project
ezhulenev added a comment to D112736: [mlir] MathApproximations: unroll virtual vectors into hardware vectors for ISA specific operation.

Can we move this RsqrtOp lowering to the x86vector dialect? It seems really out of place in the math dialect, which should remain target agnostic. (also, since this approximation is newton-raphson based instead of polynomial based, it seems out of place in PolynomialApproximation.cpp as well)

Oct 28 2021, 2:39 PM · Restricted Project
ezhulenev added a comment to D112736: [mlir] MathApproximations: unroll virtual vectors into hardware vectors for ISA specific operation.

This doesn't seem to be the right pass to do target-specific lowerings. Can you add math.rsqrt and move the vector unrolling to a generic pass in the vector dialect? Then this pass wouldn't depend on target-details.

Oct 28 2021, 2:14 PM · Restricted Project
ezhulenev committed rG627fa0b9a897: [mlir] MathApproximations: unroll virtual vectors into hardware vectors for ISA… (authored by ezhulenev).
[mlir] MathApproximations: unroll virtual vectors into hardware vectors for ISA…
Oct 28 2021, 12:52 PM
ezhulenev closed D112736: [mlir] MathApproximations: unroll virtual vectors into hardware vectors for ISA specific operation.
Oct 28 2021, 12:52 PM · Restricted Project
ezhulenev updated the diff for D112736: [mlir] MathApproximations: unroll virtual vectors into hardware vectors for ISA specific operation.

Address comments

Oct 28 2021, 12:37 PM · Restricted Project
ezhulenev accepted D112729: [mlir] fix shape check in Rsqrt approximation.

Included this fix in: https://reviews.llvm.org/D112736

Oct 28 2021, 10:52 AM · Restricted Project
ezhulenev added a reviewer for D112736: [mlir] MathApproximations: unroll virtual vectors into hardware vectors for ISA specific operation: cota.
Oct 28 2021, 10:40 AM · Restricted Project
ezhulenev requested review of D112736: [mlir] MathApproximations: unroll virtual vectors into hardware vectors for ISA specific operation.
Oct 28 2021, 10:38 AM · Restricted Project

Oct 27 2021

ezhulenev committed rG0d9b47893281: [mlir] Reduce the number of iterations in async microbenchmarks (authored by ezhulenev).
[mlir] Reduce the number of iterations in async microbenchmarks
Oct 27 2021, 3:20 AM
ezhulenev closed D112609: [mlir] Reduce the number of iterations in async microbenchmarks.
Oct 27 2021, 3:20 AM · Restricted Project
ezhulenev added a reviewer for D112609: [mlir] Reduce the number of iterations in async microbenchmarks: nicolasvasilache.
Oct 27 2021, 3:19 AM · Restricted Project
ezhulenev requested review of D112609: [mlir] Reduce the number of iterations in async microbenchmarks.
Oct 27 2021, 3:18 AM · Restricted Project

Oct 23 2021

ezhulenev committed rG35553d452b32: [mlir] Add polynomial approximation for vectorized math::Rsqrt (authored by cota).
[mlir] Add polynomial approximation for vectorized math::Rsqrt
Oct 23 2021, 4:56 AM
ezhulenev closed D112192: [mlir] Add polynomial approximation for vectorized math::Rsqrt.
Oct 23 2021, 4:56 AM · Restricted Project
ezhulenev accepted D112192: [mlir] Add polynomial approximation for vectorized math::Rsqrt.
Oct 23 2021, 4:47 AM · Restricted Project

Oct 21 2021

ezhulenev accepted D112200: [MLIR][Math] Add erf to math dialect.
Oct 21 2021, 2:41 PM · Restricted Project
ezhulenev added inline comments to D112192: [mlir] Add polynomial approximation for vectorized math::Rsqrt.
Oct 21 2021, 1:28 PM · Restricted Project
ezhulenev added inline comments to D112192: [mlir] Add polynomial approximation for vectorized math::Rsqrt.
Oct 21 2021, 5:01 AM · Restricted Project
ezhulenev added inline comments to D112192: [mlir] Add polynomial approximation for vectorized math::Rsqrt.
Oct 21 2021, 4:51 AM · Restricted Project

Oct 20 2021

ezhulenev accepted D112115: Avoid infinity arithmetics when computing exp approximations.
Oct 20 2021, 4:34 PM · Restricted Project
ezhulenev added a comment to D110680: Promote readability by factoring out creation of min/max operation. Remove unnecessary divisions..

There is one failed test

Oct 20 2021, 7:23 AM · Restricted Project

Oct 19 2021

ezhulenev added a comment to D110680: Promote readability by factoring out creation of min/max operation. Remove unnecessary divisions..

Can you please sync to head and re-upload, can't apply the patch to the head

Oct 19 2021, 5:32 AM · Restricted Project
ezhulenev accepted D110680: Promote readability by factoring out creation of min/max operation. Remove unnecessary divisions..
Oct 19 2021, 5:27 AM · Restricted Project
ezhulenev committed rGf97f946839d1: Canonicalize max/min operations on integers. (authored by bakhtiyar <bakhtiyar@x.team>).
Canonicalize max/min operations on integers.
Oct 19 2021, 5:26 AM
ezhulenev closed D112051: Canonicalize max/min operations on integers..
Oct 19 2021, 5:26 AM · Restricted Project
ezhulenev accepted D112051: Canonicalize max/min operations on integers..
Oct 19 2021, 5:19 AM · Restricted Project

Oct 18 2021

ezhulenev committed rGbf32bb7e056c: [mlir] Update approximation range for Tanh operation (authored by ezhulenev).
[mlir] Update approximation range for Tanh operation
Oct 18 2021, 10:58 AM
ezhulenev closed D112011: [mlir] Update approximation range for Tanh operation.
Oct 18 2021, 10:57 AM · Restricted Project
ezhulenev updated the summary of D112011: [mlir] Update approximation range for Tanh operation.
Oct 18 2021, 9:09 AM · Restricted Project
ezhulenev requested review of D112011: [mlir] Update approximation range for Tanh operation.
Oct 18 2021, 9:06 AM · Restricted Project

Oct 8 2021

ezhulenev committed rGe2a37bb5407e: [mlir] Add alignment option to constant tensor bufferization pass (authored by ezhulenev).
[mlir] Add alignment option to constant tensor bufferization pass
Oct 8 2021, 3:17 AM
ezhulenev closed D111364: [mlir] Add alignment option to constant tensor bufferization pass.
Oct 8 2021, 3:17 AM · Restricted Project

Oct 7 2021

ezhulenev added a reviewer for D111364: [mlir] Add alignment option to constant tensor bufferization pass: bkramer.
Oct 7 2021, 3:53 PM · Restricted Project
ezhulenev requested review of D111364: [mlir] Add alignment option to constant tensor bufferization pass.
Oct 7 2021, 3:52 PM · Restricted Project
ezhulenev committed rG8276ac13e91b: [mlir] Add alignment attribute to memref.global (authored by ezhulenev).
[mlir] Add alignment attribute to memref.global
Oct 7 2021, 6:22 AM
ezhulenev closed D111309: [mlir] Add alignment attribute to memref.global.
Oct 7 2021, 6:22 AM · Restricted Project
ezhulenev retitled D111309: [mlir] Add alignment attribute to memref.global from [mlir] Add alignment attribute to the memref.global to [mlir] Add alignment attribute to memref.global.
Oct 7 2021, 5:52 AM · Restricted Project
ezhulenev updated the summary of D111309: [mlir] Add alignment attribute to memref.global.
Oct 7 2021, 5:50 AM · Restricted Project
ezhulenev requested review of D111309: [mlir] Add alignment attribute to memref.global.
Oct 7 2021, 5:50 AM · Restricted Project

Oct 2 2021

ezhulenev added a comment to D110680: Promote readability by factoring out creation of min/max operation. Remove unnecessary divisions..

We should migrate it to min/max after https://github.com/llvm/llvm-project/commit/9fb57c8c1dd87df36daf7b6f7dee3b7423475afc

Oct 2 2021, 7:31 AM · Restricted Project

Sep 28 2021

ezhulenev committed rGbdde959533f0: Remove unnecessary async group creates and awaits. (authored by bakhtiyar <bakhtiyar@x.team>).
Remove unnecessary async group creates and awaits.
Sep 28 2021, 2:52 PM
ezhulenev committed rG55dfab39a249: Rename target block size to min task size for clarity. (authored by bakhtiyar <bakhtiyar@x.team>).
Rename target block size to min task size for clarity.
Sep 28 2021, 2:52 PM