Page MenuHomePhabricator

nicolasvasilache (Nicolas Vasilache)
User

Projects

User does not belong to any projects.

User Details

User Since
Feb 2 2017, 2:24 AM (206 w, 4 d)

Recent Activity

Today

nicolasvasilache requested changes to D94843: [mlir][affine] Fix crash when composing affine_expr with symbol multiplying dim.

This is an unreasonable relaxation IMO.
I'm looking into rewriting the part that requires AffineNormalizer, stay tuned.

Mon, Jan 18, 3:40 AM · Restricted Project
nicolasvasilache accepted D94812: [mlir][python] Swap shape and element_type order for MemRefType..
Mon, Jan 18, 1:06 AM · Restricted Project

Sat, Jan 16

nicolasvasilache accepted D94598: [mlir] Fix cross-compilation (Linalg ODS gen).
Sat, Jan 16, 5:50 AM · Restricted Project
nicolasvasilache accepted D94852: [mlir][sparse] improved sparse runtime support library.
Sat, Jan 16, 5:00 AM · Restricted Project

Thu, Jan 14

nicolasvasilache accepted D93076: [mlir][Linalg] Add canonicalization of linalg op -> dim op..

Thanks!

Thu, Jan 14, 11:33 AM · Restricted Project
nicolasvasilache accepted D94548: [mlir][linalg] Add docstring support for named op spec.
Thu, Jan 14, 6:40 AM · Restricted Project

Wed, Jan 13

nicolasvasilache accepted D94632: [mlir][PatternRewriter] Add a new hook to selectively replace uses of an operation.

Thanks!

Wed, Jan 13, 1:46 PM · Restricted Project
nicolasvasilache added a comment to D93076: [mlir][Linalg] Add canonicalization of linalg op -> dim op..

ok, my apologies, this looks great, I'll do a proper review in a bit but consider it landed :)

Wed, Jan 13, 7:50 AM · Restricted Project
nicolasvasilache added a comment to D94551: [mlir][sparse] add vectorization strategies to sparse compiler.

On one hand, it's great to see minimal additional patterns translate to new capabilities, very nice!

Wed, Jan 13, 1:01 AM · Restricted Project
nicolasvasilache accepted D94335: [mlir][linalg] Use attributes in named ops' indexing maps.
Wed, Jan 13, 12:17 AM · Restricted Project

Tue, Jan 12

nicolasvasilache committed rG80f078548868: [mlir][Linalg] NFC - Refactor fusion APIs (authored by nicolasvasilache).
[mlir][Linalg] NFC - Refactor fusion APIs
Tue, Jan 12, 6:32 AM
nicolasvasilache closed D94493: [mlir][Linalg] NFC - Refactor fusion APIs.
Tue, Jan 12, 6:32 AM · Restricted Project
nicolasvasilache retitled D94493: [mlir][Linalg] NFC - Refactor fusion APIs from BEGIN_PUBLIC [mlir][Linalg] NFC - Refactor fusion APIs to [mlir][Linalg] NFC - Refactor fusion APIs.
Tue, Jan 12, 6:28 AM · Restricted Project
nicolasvasilache requested review of D94493: [mlir][Linalg] NFC - Refactor fusion APIs.
Tue, Jan 12, 6:02 AM · Restricted Project
nicolasvasilache added a comment to D94441: [mlir][vector] verify memref of vector memory ops.

LGTM, thanks!

Tue, Jan 12, 2:11 AM · Restricted Project

Sun, Jan 10

nicolasvasilache committed rGa92248600ec4: [mlir] NFC - Drop spurious assertion on symbols during… (authored by nicolasvasilache).
[mlir] NFC - Drop spurious assertion on symbols during…
Sun, Jan 10, 6:07 AM

Sat, Jan 9

nicolasvasilache accepted D94352: [mlir][vector] modified scatter/gather syntax, pass_thru mandatory.
Sat, Jan 9, 9:20 AM · Restricted Project
nicolasvasilache added a comment to D94335: [mlir][linalg] Use attributes in named ops' indexing maps.

There are a few typos in the commit message.

Sat, Jan 9, 8:45 AM · Restricted Project
nicolasvasilache accepted D94240: [mlir][linalg] Support parsing attributes in named op spec.
Sat, Jan 9, 8:42 AM · Restricted Project

Fri, Jan 8

nicolasvasilache accepted D94318: [mlir][vector] Add memory effects to transfer_read transfer_write ops.
Fri, Jan 8, 11:31 AM · Restricted Project

Wed, Jan 6

nicolasvasilache accepted D94148: [mlir][vector] Support transfer op on tensor optimizations.
Wed, Jan 6, 10:01 AM · Restricted Project
nicolasvasilache accepted D94115: [mlir] Add hoisting transformation for transfer ops on tensor.
Wed, Jan 6, 8:21 AM · Restricted Project
nicolasvasilache accepted D93904: [mlir][vector] Support unrolling for transfer ops using tensors.

Looks good but let's please simplify the Value source part.

Wed, Jan 6, 7:41 AM · Restricted Project
nicolasvasilache added inline comments to D94148: [mlir][vector] Support transfer op on tensor optimizations.
Wed, Jan 6, 4:59 AM · Restricted Project
nicolasvasilache accepted D93774: [mlir][Linalg] Fix reshape fusion to reshape the outs instead of creating new tensors..

Great, much nicer, thanks!

Wed, Jan 6, 3:20 AM · Restricted Project

Tue, Jan 5

nicolasvasilache accepted D93077: [mlir][Linalg] NFC: Verify tiling on linalg.generic operation on tensors..

+1 for the extra test.

Tue, Jan 5, 12:23 PM · Restricted Project
nicolasvasilache added a comment to D93076: [mlir][Linalg] Add canonicalization of linalg op -> dim op..

Going through my unreviewed stack, note that with the revamp of linalg on tensors @pifon2a sent a diff today that implements the functionality with out tensors,

Tue, Jan 5, 11:58 AM · Restricted Project
nicolasvasilache accepted D94093: [mlir][Linalg] Add a test case that consumer has "reduction" loops..
Tue, Jan 5, 8:52 AM · Restricted Project
nicolasvasilache accepted D94079: [mlir] Add canonicalization pattern out_tensor->linalg->dim to out_tensor->dim..

Thanks @pifon2a !

Tue, Jan 5, 5:17 AM · Restricted Project

Wed, Dec 30

nicolasvasilache accepted D93831: fix affine_map compose with multi-symbols.
Wed, Dec 30, 9:09 AM · Restricted Project
nicolasvasilache committed rG9b5a3d67b496: [mlir] Fix indexing of first offset operand in ops that implement… (authored by nicolasvasilache).
[mlir] Fix indexing of first offset operand in ops that implement…
Wed, Dec 30, 8:58 AM
nicolasvasilache closed D93947: [mlir] Fix indexing of first offset operand in ops that implement OffsetSizeAndStrideOpInterface.
Wed, Dec 30, 8:57 AM · Restricted Project
nicolasvasilache requested review of D93947: [mlir] Fix indexing of first offset operand in ops that implement OffsetSizeAndStrideOpInterface.
Wed, Dec 30, 8:44 AM · Restricted Project
nicolasvasilache committed rGed507bc4d5ee: [mlir] NFC - Fix SubViewOp printing (authored by nicolasvasilache).
[mlir] NFC - Fix SubViewOp printing
Wed, Dec 30, 8:39 AM

Tue, Dec 29

nicolasvasilache accepted D93890: [mlir][linalg] Add vectorization for linalg on tensor ops.

LGTM, thanks!

Tue, Dec 29, 12:38 AM · Restricted Project

Dec 18 2020

nicolasvasilache updated the diff for D93469: [mlir][Linalg] Revisit the Linalg on tensors abstraction.

Apply fix.

Dec 18 2020, 11:34 AM · Restricted Project
nicolasvasilache updated the diff for D93469: [mlir][Linalg] Revisit the Linalg on tensors abstraction.

clang-tidy

Dec 18 2020, 10:10 AM · Restricted Project
nicolasvasilache updated the diff for D93469: [mlir][Linalg] Revisit the Linalg on tensors abstraction.

Address.

Dec 18 2020, 9:44 AM · Restricted Project
nicolasvasilache added inline comments to D93469: [mlir][Linalg] Revisit the Linalg on tensors abstraction.
Dec 18 2020, 9:28 AM · Restricted Project
nicolasvasilache added inline comments to D93469: [mlir][Linalg] Revisit the Linalg on tensors abstraction.
Dec 18 2020, 8:25 AM · Restricted Project
nicolasvasilache updated the diff for D93469: [mlir][Linalg] Revisit the Linalg on tensors abstraction.

Rebase on a better Linalg.md

Dec 18 2020, 8:24 AM · Restricted Project
nicolasvasilache committed rGb88ed4ec8e7d: [mlir][Linlag] Reflow Linalg.md - NFC (authored by nicolasvasilache).
[mlir][Linlag] Reflow Linalg.md - NFC
Dec 18 2020, 8:16 AM
nicolasvasilache updated the diff for D93469: [mlir][Linalg] Revisit the Linalg on tensors abstraction.

Rebase

Dec 18 2020, 5:19 AM · Restricted Project
nicolasvasilache updated the summary of D93469: [mlir][Linalg] Revisit the Linalg on tensors abstraction.
Dec 18 2020, 3:25 AM · Restricted Project
nicolasvasilache updated the diff for D93469: [mlir][Linalg] Revisit the Linalg on tensors abstraction.

Drop debug spew.

Dec 18 2020, 3:24 AM · Restricted Project
nicolasvasilache updated the diff for D93469: [mlir][Linalg] Revisit the Linalg on tensors abstraction.

Rebase

Dec 18 2020, 3:05 AM · Restricted Project
nicolasvasilache added reviewers for D93469: [mlir][Linalg] Revisit the Linalg on tensors abstraction: pifon2a, mravishankar, silvas, ftynse.
Dec 18 2020, 2:34 AM · Restricted Project
nicolasvasilache updated the diff for D93469: [mlir][Linalg] Revisit the Linalg on tensors abstraction.

Finish the impl.

Dec 18 2020, 2:34 AM · Restricted Project

Dec 17 2020

nicolasvasilache accepted D93500: [mlir[[vector] Extend Transfer read/write ops to support tensor types..

Thanks @ThomasRaoux

Dec 17 2020, 11:49 PM · Restricted Project
nicolasvasilache accepted D93492: [mlir] Add canonicalization from `tensor_cast` to `dim` op..
Dec 17 2020, 2:40 PM · Restricted Project
nicolasvasilache accepted D93374: [mlir][Linalg] Define a linalg.init_tensor operation..
Dec 17 2020, 11:53 AM · Restricted Project
nicolasvasilache added a comment to D93374: [mlir][Linalg] Define a linalg.init_tensor operation..

+1 it's fine for now and unlocks a bigger piece of the puzzle.
We can refactor and cleanup as followups: subview crams together 3 variadic static / dynamic + canonicalizations.
There is a nicer future where we can define a generic pair of variadic operand + attribute + canonicalizations for all such things.

Dec 17 2020, 11:53 AM · Restricted Project
nicolasvasilache requested review of D93469: [mlir][Linalg] Revisit the Linalg on tensors abstraction.
Dec 17 2020, 10:22 AM · Restricted Project
nicolasvasilache accepted D93315: [mlir] partially update LLVM dialect documentation.
Dec 17 2020, 1:33 AM · Restricted Project

Dec 16 2020

nicolasvasilache accepted D93418: [mlir] Modify linalg loops test to have nested regions.

Thanks!

Dec 16 2020, 1:49 PM · Restricted Project

Dec 15 2020

nicolasvasilache added a comment to D93308: [mlir] Allow nested regions in inlineRegionAndEmitStore.

well you need at least some test here :p

Dec 15 2020, 12:08 PM · Restricted Project

Dec 11 2020

nicolasvasilache committed rG7310501f7403: [mlir][ArmNeon][RFC] Add a Neon dialect (authored by nicolasvasilache).
[mlir][ArmNeon][RFC] Add a Neon dialect
Dec 11 2020, 5:51 AM
nicolasvasilache closed D92171: [mlir][ArmNeon][RFC] Add an ArmNeon dialect.
Dec 11 2020, 5:51 AM · Restricted Project
nicolasvasilache updated the diff for D92171: [mlir][ArmNeon][RFC] Add an ArmNeon dialect.

Line length.

Dec 11 2020, 5:50 AM · Restricted Project
nicolasvasilache added a comment to D92171: [mlir][ArmNeon][RFC] Add an ArmNeon dialect.

.

Dec 11 2020, 5:14 AM · Restricted Project
nicolasvasilache updated the diff for D92171: [mlir][ArmNeon][RFC] Add an ArmNeon dialect.

CMake

Dec 11 2020, 5:13 AM · Restricted Project
nicolasvasilache updated the diff for D92171: [mlir][ArmNeon][RFC] Add an ArmNeon dialect.

Fixes post rebase.

Dec 11 2020, 5:11 AM · Restricted Project
nicolasvasilache added inline comments to D92171: [mlir][ArmNeon][RFC] Add an ArmNeon dialect.
Dec 11 2020, 4:58 AM · Restricted Project
nicolasvasilache updated the diff for D92171: [mlir][ArmNeon][RFC] Add an ArmNeon dialect.

Address and rebase

Dec 11 2020, 4:57 AM · Restricted Project
nicolasvasilache accepted D93086: [mlir][Linalg] Extend tile+fuse to work on Linalg operation on tensors..

Looks good!

Dec 11 2020, 3:54 AM · Restricted Project
nicolasvasilache requested changes to D93077: [mlir][Linalg] NFC: Verify tiling on linalg.generic operation on tensors..

It's nice that we actually don't need to change op semantics to support "init_tensor form" and that all semantics can be dealt with just in tiling.
However we need a better abstraction and documentation to avoid leaking the op + init or result tensor implementation detail into the tiling implementation.

Dec 11 2020, 3:45 AM · Restricted Project
nicolasvasilache requested changes to D93076: [mlir][Linalg] Add canonicalization of linalg op -> dim op..

Also @pifon2a

Dec 11 2020, 2:58 AM · Restricted Project

Dec 10 2020

nicolasvasilache added a comment to D92980: [mlir] Remove the dependency on StandardOps from FoldUtils.

LGTM

Dec 10 2020, 2:25 PM · Restricted Project

Dec 9 2020

nicolasvasilache accepted D92991: [mlir] Start splitting the `tensor` dialect out of `std`..

I'm -1 on the proposed name change: LLVM has {insert|extract}{element|value}, op names are globally consistent with that terminology.

Dec 9 2020, 11:38 PM · Restricted Project
nicolasvasilache accepted D92963: [mlir] Allow RegionBranchOps in dependence analysis..

As discussed offline, this will be revamped in a relatively near future.
Fine to unblock for now assuming presubmits pass.

Dec 9 2020, 1:14 PM · Restricted Project

Dec 8 2020

nicolasvasilache added inline comments to D92172: [mlir][ArmSVE][RFC] Add an ArmSVE dialect.
Dec 8 2020, 1:06 AM · Restricted Project

Dec 7 2020

nicolasvasilache added inline comments to D92172: [mlir][ArmSVE][RFC] Add an ArmSVE dialect.
Dec 7 2020, 6:50 AM · Restricted Project
nicolasvasilache added inline comments to D92172: [mlir][ArmSVE][RFC] Add an ArmSVE dialect.
Dec 7 2020, 6:43 AM · Restricted Project
nicolasvasilache added a comment to D92172: [mlir][ArmSVE][RFC] Add an ArmSVE dialect.

Once my comments are addressed I am happy to land this.
I think the additional layer of tablegen discussed in https://llvm.discourse.group/t/exposing-llvm-ops-to-the-mlir-type-system/2290/11 and having everything in the LLVM dialect is a cross-cutting change that should be done separately and unified across all 3 dialects (AVX512, ArmNeon, ArmSVE).

Dec 7 2020, 6:38 AM · Restricted Project
nicolasvasilache updated the diff for D92171: [mlir][ArmNeon][RFC] Add an ArmNeon dialect.

CMake

Dec 7 2020, 6:24 AM · Restricted Project
nicolasvasilache retitled D92171: [mlir][ArmNeon][RFC] Add an ArmNeon dialect from [mlir][Neon][RFC] Add a Neon dialect to [mlir][ArmNeon][RFC] Add an ArmNeon dialect.
Dec 7 2020, 6:12 AM · Restricted Project
nicolasvasilache updated the diff for D92171: [mlir][ArmNeon][RFC] Add an ArmNeon dialect.

Update commit message

Dec 7 2020, 6:11 AM · Restricted Project
nicolasvasilache updated the diff for D92171: [mlir][ArmNeon][RFC] Add an ArmNeon dialect.

Address and rebase

Dec 7 2020, 6:11 AM · Restricted Project
nicolasvasilache abandoned D91427: [mlir][Linalg] NFC - Add bufferization options.
Dec 7 2020, 6:02 AM · Restricted Project
nicolasvasilache abandoned D92216: [mlir] Add InlineAsmOp to standard.
Dec 7 2020, 6:02 AM · Restricted Project

Dec 4 2020

nicolasvasilache accepted D92664: [mlir][vector] rephrased description.
Dec 4 2020, 1:40 PM · Restricted Project
nicolasvasilache requested changes to D92664: [mlir][vector] rephrased description.
Dec 4 2020, 11:41 AM · Restricted Project
nicolasvasilache accepted D92635: [mlir][vector][NFC] Change UnrollVectorPattern to not be statically dependent on an op type.
Dec 4 2020, 8:41 AM · Restricted Project
nicolasvasilache committed rG2c66b6ec09dd: [mlir][Linalg] NFC - Expose tiling canonicalization patterns through a populate… (authored by nicolasvasilache).
[mlir][Linalg] NFC - Expose tiling canonicalization patterns through a populate…
Dec 4 2020, 7:02 AM
nicolasvasilache committed rGa1cd559ce500: [mlir][Linalg] Properly use distribution options. (authored by nicolasvasilache).
[mlir][Linalg] Properly use distribution options.
Dec 4 2020, 6:05 AM
nicolasvasilache closed D92653: [mlir][Linalg] Properly use distribution options..
Dec 4 2020, 6:05 AM · Restricted Project
nicolasvasilache requested review of D92653: [mlir][Linalg] Properly use distribution options..
Dec 4 2020, 5:52 AM · Restricted Project
nicolasvasilache added a comment to D92614: [mlir][vector][avx512] move avx512 lowering pass into general vector lowering.

Great, thanks for refactoring this!

Dec 4 2020, 12:01 AM · Restricted Project

Dec 3 2020

nicolasvasilache added a comment to D92343: [mlir] Add initial support for an alias analysis framework in MLIR.

Very cool, thanks River!
I have a bunch of use cases for this :)

Dec 3 2020, 1:32 PM · Restricted Project
nicolasvasilache accepted D92540: [mlir][linalg] Add vectorization for element-wise linalg ops.

Very cool, thanks for pushing this @ThomasRaoux !

Dec 3 2020, 3:46 AM · Restricted Project

Nov 30 2020

nicolasvasilache added a comment to D92166: [mlir][LLVMIR] Add support for InlineAsmOp.

Poking a bit around I see this build: http://lab.llvm.org:8011/#/builders/135/builds/110

Nov 30 2020, 8:31 AM · Restricted Project
nicolasvasilache added a comment to D92166: [mlir][LLVMIR] Add support for InlineAsmOp.

Hi,

This seems to break several buildbots that don't build the x86 backend, e.g. http://lab.llvm.org:8011/#/builders/135/builds/90.
Could you please take a look?

Thanks!

Nov 30 2020, 7:12 AM · Restricted Project
nicolasvasilache committed rG78c71187465a: [mlir] Make mlir-cpu-runner depend on native instead of X86 (authored by nicolasvasilache).
[mlir] Make mlir-cpu-runner depend on native instead of X86
Nov 30 2020, 7:12 AM
nicolasvasilache abandoned D92234: [mlir] Relax sign extend to work with signed or signless integers..
Nov 30 2020, 1:14 AM · Restricted Project
nicolasvasilache committed rG047400ed8204: [mlir][LLVMIR] Add support for InlineAsmOp (authored by nicolasvasilache).
[mlir][LLVMIR] Add support for InlineAsmOp
Nov 30 2020, 12:37 AM
nicolasvasilache closed D92166: [mlir][LLVMIR] Add support for InlineAsmOp.
Nov 30 2020, 12:36 AM · Restricted Project
nicolasvasilache added inline comments to D92166: [mlir][LLVMIR] Add support for InlineAsmOp.
Nov 30 2020, 12:30 AM · Restricted Project
nicolasvasilache added a comment to D92166: [mlir][LLVMIR] Add support for InlineAsmOp.

I'm a bit concerned about bringing this into MLIR without stronger constraints than what is allowed in LLVM. Unfortunately even if we document restriction, it isn't clear how we can enforce anything? The way inline assembly is an open-door to "anything" seems quite scary to me.

Nov 30 2020, 12:29 AM · Restricted Project
nicolasvasilache updated the summary of D92166: [mlir][LLVMIR] Add support for InlineAsmOp.
Nov 30 2020, 12:12 AM · Restricted Project

Nov 29 2020

nicolasvasilache added a comment to D92216: [mlir] Add InlineAsmOp to standard.

"I didn't care enough to model this as an intrinsic or op/my machine scheduler modelling isn't good enough/I need an escape hatch to do things which are invalid but work for my case".

Nov 29 2020, 11:39 PM · Restricted Project