Page MenuHomePhabricator
Feed Advanced Search

Jan 31 2020

ftynse closed D73614: [mlir] Linalg tiling: generate code avoding out-of-bounds accesses.
Jan 31 2020, 10:45 AM · Restricted Project
ftynse accepted D73733: [mlir] [VectorOps] fixed bug in vector.insert_strided_slice lowering.

LGTM after changing the test to capture %arg0 and %arg1. Thanks!

Jan 31 2020, 5:37 AM · Restricted Project
ftynse accepted D73769: Make gpu.launch implicitly capture uses of values defined above..
Jan 31 2020, 5:32 AM · Restricted Project
ftynse accepted D73770: [mlir] Fix link to 'Getting started with MLIR'.
Jan 31 2020, 2:54 AM · Restricted Project
ftynse added inline comments to D72802: [mlir] Introduce bare ptr calling convention for MemRefs in LLVM dialect.
Jan 31 2020, 12:46 AM · Restricted Project
ftynse requested changes to D73764: [mlir] Added RankedIntElementsAttr class.
Jan 31 2020, 12:37 AM · Restricted Project
ftynse accepted D72802: [mlir] Introduce bare ptr calling convention for MemRefs in LLVM dialect.

This is a summary of the pending items that I would suggest addressing separately:
<...>

Jan 31 2020, 12:37 AM · Restricted Project

Jan 30 2020

ftynse committed rGeb67bd78dc18: [mlir] LLVM dialect: Generate conversions between EnumAttrCase and LLVM API (authored by ftynse).
[mlir] LLVM dialect: Generate conversions between EnumAttrCase and LLVM API
Jan 30 2020, 12:55 PM
ftynse closed D73468: [mlir] LLVM dialect: Generate conversions between EnumAttrCase and LLVM API.
Jan 30 2020, 12:55 PM · Restricted Project
ftynse updated the diff for D73468: [mlir] LLVM dialect: Generate conversions between EnumAttrCase and LLVM API.

Split "from" and "to" conversion generation and added more doc as requested.

Jan 30 2020, 12:54 PM · Restricted Project
ftynse added inline comments to D73702: [WIP][mlir] enable conversion materialization for 1-1 type conversions.
Jan 30 2020, 11:03 AM · Restricted Project
ftynse accepted D73684: [MLIR][Linalg] Lower linalg.generic to ploops..
Jan 30 2020, 8:41 AM · Restricted Project
ftynse updated the diff for D73468: [mlir] LLVM dialect: Generate conversions between EnumAttrCase and LLVM API.

rebase

Jan 30 2020, 8:13 AM · Restricted Project
ftynse committed rGfdc496a3d30d: [mlir] EnumsGen: dissociate string form of integer enum from C++ symbol name (authored by ftynse).
[mlir] EnumsGen: dissociate string form of integer enum from C++ symbol name
Jan 30 2020, 8:04 AM
ftynse closed D73362: [mlir] EnumsGen: dissociate string form of integer enum from C++ symbol name.
Jan 30 2020, 8:04 AM · Restricted Project
ftynse added a comment to D73468: [mlir] LLVM dialect: Generate conversions between EnumAttrCase and LLVM API.

This is a follow-up to that.

Jan 30 2020, 8:04 AM · Restricted Project
ftynse updated the summary of D73702: [WIP][mlir] enable conversion materialization for 1-1 type conversions.
Jan 30 2020, 6:22 AM · Restricted Project
ftynse created D73702: [WIP][mlir] enable conversion materialization for 1-1 type conversions.
Jan 30 2020, 6:22 AM · Restricted Project
ftynse committed rGd242aa245ccf: [MLIR] Added llvm.invoke and llvm.landingpad (authored by shraiysh).
[MLIR] Added llvm.invoke and llvm.landingpad
Jan 30 2020, 3:56 AM
ftynse closed D72006: [MLIR] Added llvm.invoke and llvm.landingpad.
Jan 30 2020, 3:56 AM · Restricted Project
ftynse added inline comments to D73685: Fix conversion of loops to GPU with no block/thread dimensions..
Jan 30 2020, 3:42 AM · Restricted Project
ftynse added a comment to D73684: [MLIR][Linalg] Lower linalg.generic to ploops..

What happens when you have non-parallel loops in the generic op? Can we have a test for this case (even if there's no transform happening, just make sure we bail out instead of generating incorrectly parallel loop?)

Jan 30 2020, 3:05 AM · Restricted Project
ftynse accepted D73654: [MLIR] break cyclic dependency between VectorOps and Analysis.
Jan 30 2020, 2:53 AM · Restricted Project
ftynse added a comment to D73671: [MLIR][Standard] Add folding for indexCast(indexCast(x)) -> x.

Looks good in general, please follow @rriddle's recommendation.

Jan 30 2020, 2:50 AM · Restricted Project
ftynse added a reviewer for D73671: [MLIR][Standard] Add folding for indexCast(indexCast(x)) -> x: ftynse.
Jan 30 2020, 2:50 AM · Restricted Project
ftynse added a comment to D73571: [MLIR] Add the sqrt operation to mlir..

Please LMK if you need somebody to commit this for you

Jan 30 2020, 2:50 AM · Restricted Project
ftynse accepted D73685: Fix conversion of loops to GPU with no block/thread dimensions..

Good catch, thanks!

Jan 30 2020, 2:47 AM · Restricted Project
ftynse added a comment to D73620: Add 'gpu.terminator' operation..

D71961 is going in the direction of generalizing std.return ; what is the guideline for adding a new op vs re-using std.return?

Jan 30 2020, 2:35 AM · Restricted Project

Jan 29 2020

ftynse accepted D72802: [mlir] Introduce bare ptr calling convention for MemRefs in LLVM dialect.

I'm fine with this, but please make sure @rriddle has no objections either.

Jan 29 2020, 2:53 PM · Restricted Project
ftynse accepted D73571: [MLIR] Add the sqrt operation to mlir..
Jan 29 2020, 2:35 PM · Restricted Project
ftynse added a comment to D73571: [MLIR] Add the sqrt operation to mlir..

One more nit, thanks!

Jan 29 2020, 12:17 PM · Restricted Project
ftynse added a comment to D73595: [mlir][Linalg][doc] Add Design Document for the Linalg Dialect.

@ftynse I fully agree hence my pushing back on the github comments.

If you had started submitting this incrementally, you wouldn't have this problem ;)

Jan 29 2020, 8:44 AM · Restricted Project
ftynse accepted D73407: [mlir] Update various operations to declaratively specify their assembly format..
Jan 29 2020, 8:25 AM · Restricted Project
ftynse accepted D73503: [mlir][NFC] Use declarative format for several operations in LLVM and Linalg dialects.
Jan 29 2020, 8:17 AM · Restricted Project
ftynse added a comment to D73595: [mlir][Linalg][doc] Add Design Document for the Linalg Dialect.

@ftynse it seems approppriate for you to review parts you haven't touched yet though.
I think your review would still be very valuable for sections 3 - 8 (Core Guiding Principles - end).

Thanks!

Jan 29 2020, 6:26 AM · Restricted Project
ftynse added inline comments to D73620: Add 'gpu.terminator' operation..
Jan 29 2020, 6:17 AM · Restricted Project
ftynse accepted D73620: Add 'gpu.terminator' operation..

Thanks!

Jan 29 2020, 5:31 AM · Restricted Project
ftynse added a comment to D72802: [mlir] Introduce bare ptr calling convention for MemRefs in LLVM dialect.

This looks good to me. I don't fully understand the problem @rriddle mentioned with replaceUsesOfWith, and I wonder if we could have a solution that also avoids the need for the placeholder operation.

Jan 29 2020, 5:22 AM · Restricted Project
ftynse accepted D73145: [mlir][Linalg] Introduce folding patterns to remove certain MemRefCastOp.
Jan 29 2020, 4:55 AM · Restricted Project
ftynse added a comment to D73362: [mlir] EnumsGen: dissociate string form of integer enum from C++ symbol name.

I meant that instead of having both I32EnumAttrCase (taking two parameters) and I32StrEnumAttrCase (taking three parameters), what about just having I32EnumAttrCase and making it taking in string cSym, int cVal, string strVal = cSym? Similarly for other *EnumAttrCases.

Jan 29 2020, 2:58 AM · Restricted Project
ftynse updated the diff for D73362: [mlir] EnumsGen: dissociate string form of integer enum from C++ symbol name.

Use default value for parameter and drop IntStr*

Jan 29 2020, 2:58 AM · Restricted Project
ftynse created D73614: [mlir] Linalg tiling: generate code avoding out-of-bounds accesses.
Jan 29 2020, 2:39 AM · Restricted Project
ftynse added a comment to D73595: [mlir][Linalg][doc] Add Design Document for the Linalg Dialect.

I don't feel it's appropriate for me to review this since I wrote some parts of the text.

Jan 29 2020, 12:50 AM · Restricted Project
ftynse accepted D73296: [mlir] Add MemRefTypeBuilder and refactor some MemRefType::get()..
Jan 29 2020, 12:50 AM · Restricted Project

Jan 28 2020

ftynse requested changes to D73571: [MLIR] Add the sqrt operation to mlir..
Jan 28 2020, 2:43 PM · Restricted Project
ftynse committed rG8ed47b74300e: [mlir] NFC: use ValueRange in AffineToStandard conversion (authored by ftynse).
[mlir] NFC: use ValueRange in AffineToStandard conversion
Jan 28 2020, 3:00 AM
ftynse committed rG6895a1c37e44: [mlir] NFC: use doxygen-style comments in AffineToStandard.cpp (authored by ftynse).
[mlir] NFC: use doxygen-style comments in AffineToStandard.cpp
Jan 28 2020, 1:29 AM

Jan 27 2020

ftynse added inline comments to D73488: [mlir] add lowering from affine.min to std.
Jan 27 2020, 1:57 PM · Restricted Project
ftynse committed rG51ba5b528a98: [mlir] add lowering from affine.min to std (authored by ftynse).
[mlir] add lowering from affine.min to std
Jan 27 2020, 1:39 PM
ftynse committed rG74df89f67f17: [NFC][mlir][linalg] Merge Utils/Intrinsics.h into EDSC/Intrinsics.h (authored by kernhanda).
[NFC][mlir][linalg] Merge Utils/Intrinsics.h into EDSC/Intrinsics.h
Jan 27 2020, 1:39 PM
ftynse closed D73377: [NFC] [mlir] [linalg] Merge Utils/Intrinsics.h into EDSC/Intrinsics.h.
Jan 27 2020, 1:39 PM · Restricted Project
ftynse closed D73488: [mlir] add lowering from affine.min to std.
Jan 27 2020, 1:39 PM · Restricted Project
ftynse added inline comments to D73465: Add gpu::LaunchOp::addKernelArgument..
Jan 27 2020, 1:38 PM · Restricted Project
ftynse added a comment to D73377: [NFC] [mlir] [linalg] Merge Utils/Intrinsics.h into EDSC/Intrinsics.h.

Could you plz rebase on master? It does not seem to apply anymore. And fix clang-format while you are there.

Jan 27 2020, 8:21 AM · Restricted Project
ftynse created D73488: [mlir] add lowering from affine.min to std.
Jan 27 2020, 8:21 AM · Restricted Project
ftynse accepted D73470: [mlir] fixed invalid LLVM intrinsics in LLVMOPs.td and llvmir-intrinsics.mlir..

Please note that we tend to prefix the header of the commit message with [mlir] so that people on llvm-commits@ can understand the changes are restricted to MLIR (many more people might have a say if actual LLVM IR intrinsics were broken)

Jan 27 2020, 7:53 AM · Restricted Project
ftynse retitled D73470: [mlir] fixed invalid LLVM intrinsics in LLVMOPs.td and llvmir-intrinsics.mlir. from Fixed invalid LLVM intrinsics in LLVMOPs.td and llvmir-intrinsics.mlir. to [mlir] fixed invalid LLVM intrinsics in LLVMOPs.td and llvmir-intrinsics.mlir..
Jan 27 2020, 7:46 AM · Restricted Project
ftynse committed rG84c3f05c8e3e: [mlir] Harden error propagation in LLVM import (authored by ftynse).
[mlir] Harden error propagation in LLVM import
Jan 27 2020, 7:17 AM
ftynse committed rG07328944efb6: [mlir] LLVM import: handle constant data and array/vector aggregates (authored by ftynse).
[mlir] LLVM import: handle constant data and array/vector aggregates
Jan 27 2020, 7:17 AM
ftynse closed D72834: [mlir] LLVM import: handle constant data and array/vector aggregates.
Jan 27 2020, 7:16 AM · Restricted Project
ftynse closed D72839: [mlir] Harden error propagation in LLVM import.
Jan 27 2020, 7:16 AM · Restricted Project
ftynse accepted D73471: Add tanh lowering from Standard dialect to NVVM and ROCDL..

Please update the commit message from "fixed tanh lowering" to "add tanh lowering". It was not broken (which would mandate a fix), it was never implemented to start with. Good to go otherwise.

Jan 27 2020, 6:40 AM · Restricted Project
ftynse accepted D73348: Add lowering for loop.parallel to cfg..
Jan 27 2020, 6:40 AM · Restricted Project
ftynse requested changes to D73470: [mlir] fixed invalid LLVM intrinsics in LLVMOPs.td and llvmir-intrinsics.mlir..

I suppose that, in general, there is no strict correspondance between the number of type suffixes in an overloaded intrinsic and the number of operands. So having one type may make sense for fmuladd, but may not make sense for some platform-specific vector select-like intrinsic with three arguments that would require two types. I would suggest renaming BinaryIntrinsicOp to BinarySameArgsIntrinsicOp if you do this change. Or even dropping it entirely and keeping an explicit definition for copysign and fmuladd since I don't see any other users of the Binary and TernaryIntrinsicOp classes.

Jan 27 2020, 6:31 AM · Restricted Project
ftynse accepted D73465: Add gpu::LaunchOp::addKernelArgument..
Jan 27 2020, 6:03 AM · Restricted Project
ftynse created D73468: [mlir] LLVM dialect: Generate conversions between EnumAttrCase and LLVM API.
Jan 27 2020, 5:56 AM · Restricted Project
ftynse requested changes to D72962: [MLIR, OpenMP] Translation of OpenMP barrier construct to LLVM IR.

Thanks! I have a couple of comments.

Jan 27 2020, 2:27 AM · Restricted Project
ftynse added a comment to D73348: Add lowering for loop.parallel to cfg..

Looks generally okay to me. I don't see the point of cloning before erasing, but can live with it pending a clean-up.

Jan 27 2020, 2:15 AM · Restricted Project
ftynse added a comment to D72834: [mlir] LLVM import: handle constant data and array/vector aggregates.

Anything else @nicolasvasilache ?

Jan 27 2020, 1:47 AM · Restricted Project
Herald added a reviewer for D73362: [mlir] EnumsGen: dissociate string form of integer enum from C++ symbol name: rriddle.

Thanks Alex for adding this! Have you considered adding an additional parameter (defaulting to C++ enumerant symbol) for the string representation instead of duplicating all the cases?

Jan 27 2020, 1:47 AM · Restricted Project

Jan 24 2020

ftynse added a reviewer for D73362: [mlir] EnumsGen: dissociate string form of integer enum from C++ symbol name: antiagainst.
Jan 24 2020, 8:58 AM · Restricted Project
ftynse created D73362: [mlir] EnumsGen: dissociate string form of integer enum from C++ symbol name.
Jan 24 2020, 8:58 AM · Restricted Project
ftynse committed rGb9013351936d: [mlir] Use all_of instead of a manual loop in IntrinsicGen. NFC (authored by ftynse).
[mlir] Use all_of instead of a manual loop in IntrinsicGen. NFC
Jan 24 2020, 2:33 AM
ftynse requested changes to D73296: [mlir] Add MemRefTypeBuilder and refactor some MemRefType::get()..

Thanks, I like the simplification.

Jan 24 2020, 1:15 AM · Restricted Project
ftynse added a comment to rGc8695ba9cdeb: Revert "[mlir] Add baseAttr to TypedArrayAttrBase.".

The change looks innocuous, I would have landed on approval, too.

Jan 24 2020, 12:58 AM

Jan 23 2020

ftynse accepted D73233: [mlir] Add option to use custom base class for dialect in LLVMIRIntrinsicGen..

Thanks!

Jan 23 2020, 11:05 AM · Restricted Project
ftynse added inline comments to D73233: [mlir] Add option to use custom base class for dialect in LLVMIRIntrinsicGen..
Jan 23 2020, 9:43 AM · Restricted Project
ftynse requested changes to D73145: [mlir][Linalg] Introduce folding patterns to remove certain MemRefCastOp.
Jan 23 2020, 8:01 AM · Restricted Project
ftynse added a comment to D72802: [mlir] Introduce bare ptr calling convention for MemRefs in LLVM dialect.

I'm supportive of this change as long as Mehdi's and River's concerns are addressed.

Jan 23 2020, 7:43 AM · Restricted Project
ftynse requested changes to D73233: [mlir] Add option to use custom base class for dialect in LLVMIRIntrinsicGen..

Looks good in general, I am suggesting to push it one step further.

Jan 23 2020, 7:24 AM · Restricted Project
ftynse added a comment to D72888: [mlir] Add baseAttr to TypedArrayAttrBase..

I reverted this change since it broke dependent projects, and the modeling generally looks incorrect.

Jan 23 2020, 6:03 AM · Restricted Project
ftynse committed rGc8695ba9cdeb: Revert "[mlir] Add baseAttr to TypedArrayAttrBase." (authored by ftynse).
Revert "[mlir] Add baseAttr to TypedArrayAttrBase."
Jan 23 2020, 5:54 AM
ftynse added a reverting change for rGeec36909c18b: [mlir] Add baseAttr to TypedArrayAttrBase.: rGc8695ba9cdeb: Revert "[mlir] Add baseAttr to TypedArrayAttrBase.".
Jan 23 2020, 5:54 AM

Jan 22 2020

ftynse committed rG7984b47401f7: [mlir][orc] unbreak MLIR ExecutionEngine after ORC changes (authored by ftynse).
[mlir][orc] unbreak MLIR ExecutionEngine after ORC changes
Jan 22 2020, 1:18 AM

Jan 20 2020

ftynse committed rGf63f5a228f30: [mlir] clarify LangRef wording around control flow in regions (authored by ftynse).
[mlir] clarify LangRef wording around control flow in regions
Jan 20 2020, 5:31 AM
ftynse accepted D72995: [MLIR] LLVM Dialect: add llvm.cmpxchg and improve llvm.atomicrmw custom parser.

Minor comments only.

Jan 20 2020, 1:30 AM · Restricted Project
ftynse accepted D72129: [mlir] Add in-dialect lowering of gpu.all_reduce..
Jan 20 2020, 1:21 AM · Restricted Project

Jan 17 2020

ftynse committed rG60a0c612df51: [MLIR] LLVM dialect: Add llvm.atomicrmw (authored by flaub).
[MLIR] LLVM dialect: Add llvm.atomicrmw
Jan 17 2020, 12:19 PM
ftynse committed rG37e2560d3d24: [Flang][mlir] add a band-aid to support the creation of mutually recursive… (authored by schweitz).
[Flang][mlir] add a band-aid to support the creation of mutually recursive…
Jan 17 2020, 12:19 PM
ftynse closed D72741: [MLIR] LLVM dialect: Add llvm.atomicrmw.
Jan 17 2020, 12:19 PM · Restricted Project
ftynse closed D72542: [Flang][mlir] add a band-aid to support the creation of mutually recursive types when lowering to LLVM IR.
Jan 17 2020, 12:18 PM · Restricted Project, Restricted Project
ftynse accepted D72741: [MLIR] LLVM dialect: Add llvm.atomicrmw.
Jan 17 2020, 12:09 PM · Restricted Project
ftynse updated subscribers of D72741: [MLIR] LLVM dialect: Add llvm.atomicrmw.

@tra, your rule adds you as blocking on all of MLIR :)

Jan 17 2020, 12:09 PM · Restricted Project
ftynse added a comment to D72942: [mlir][Linalg] Extend linalg vectorization to MatmulOp.

The change is mostly renaming/clean-up. I only have minor comments, feel free to land after addressing.

Jan 17 2020, 11:58 AM · Restricted Project
ftynse accepted D72942: [mlir][Linalg] Extend linalg vectorization to MatmulOp.
Jan 17 2020, 11:58 AM · Restricted Project
ftynse added a comment to D72741: [MLIR] LLVM dialect: Add llvm.atomicrmw.

To clarify further, LLVM dialect in MLIR is not a new IR, it's essentially a representation of LLVM IR using MLIR. It's supposed to minimize the cost of translation from MLIR to LLVM IR. While it does diverge from LLVM IR proper, it does so for modeling reasons: there's no first-class constants in MLIR, neither are there phi nodes. Changing the name of an enumerand locally to the LLVM dialect will increase the cost of translation, which goes against the goal of this dialect. That being said, I do agree that we should use this opportunity to reflect on LLVM IR design and update it following the proper process.

Jan 17 2020, 11:21 AM · Restricted Project
ftynse added a comment to D72741: [MLIR] LLVM dialect: Add llvm.atomicrmw.

Changing LLVM IR should _not_ be in this commit, it's semantically different and requires separate discussion with relevant people involved. I also don't see a compelling reason to block this from landing until LLVM IR changes (or decides not to). LLVM dialect models LLVM IR in its current state, whatever the state is.

Jan 17 2020, 11:21 AM · Restricted Project
ftynse committed rGf343544b8138: [mlir] Generator converting LLVM intrinsics defs to MLIR ODS (authored by ftynse).
[mlir] Generator converting LLVM intrinsics defs to MLIR ODS
Jan 17 2020, 9:25 AM
ftynse closed D72926: [mlir] Generator converting LLVM intrinsics defs to MLIR ODS.
Jan 17 2020, 9:25 AM · Restricted Project
ftynse updated the diff for D72926: [mlir] Generator converting LLVM intrinsics defs to MLIR ODS.

address review

Jan 17 2020, 9:25 AM · Restricted Project