Page MenuHomePhabricator
Feed Advanced Search

Feb 22 2021

bondhugula accepted D95330: [MLIR][GPU][NVVM] Add warp synchronous matrix-multiply accumulate ops.

Thanks for addressing the large number of comments. Some additional minor ones and one that was missed (or not pushed). This overall looks great to me!

Feb 22 2021, 11:02 AM · Restricted Project
bondhugula added a comment to D95330: [MLIR][GPU][NVVM] Add warp synchronous matrix-multiply accumulate ops.

Hi, Thanks for the comments.

A high-level design question: why does the element type of mmafragment have to be a vector type? I'd just use 2D indexing for the fragment, it's not like we are going to extract vectors from it.

I have tried to keep the types as close to what is expected by the corresponding LLVM intrinsics. As the mma.compute intrinsic expects operands in <2 x half> form and also returns things in a similar form, I have used the vector type.

This is an anti-argument for me, I see very little value in just lifting the low-level LLVM abstractions to higher levels. Hardcoding NVVM-specific modeling in the GPU dialect that is
supposed to abstract that away defies the purpose of the GPU dialect. It sounds like memfragment<AxBxf16> would make all of the code, except for a tiny part of the conversion, simpler.

Feb 22 2021, 10:55 AM · Restricted Project
bondhugula committed rG15332982c3d8: [MLIR][affine] Prevent fusion when ops with memory effect free are present… (authored by vinayaka-polymage).
[MLIR][affine] Prevent fusion when ops with memory effect free are present…
Feb 22 2021, 9:52 AM
bondhugula closed D97032: [MLIR][affine] Prevent fusion when ops with memory effect free are present between producer and consumer.
Feb 22 2021, 9:51 AM · Restricted Project
bondhugula accepted D96821: make Affine parallel and yield ops MemRefsNormalizable .

LGTM.

Feb 22 2021, 9:50 AM · Restricted Project
bondhugula added a comment to D97030: [MLIR][affine-loop-fusion] Handle defining ops between the source and dest loops.

If you have fixed for the iter_args handling, please also update the commit summary to reflect that.

Feb 22 2021, 8:48 AM · Restricted Project
bondhugula accepted D97030: [MLIR][affine-loop-fusion] Handle defining ops between the source and dest loops.

LGTM.

Feb 22 2021, 8:47 AM · Restricted Project
bondhugula accepted D97032: [MLIR][affine] Prevent fusion when ops with memory effect free are present between producer and consumer.
Feb 22 2021, 7:57 AM · Restricted Project
bondhugula added a comment to D97030: [MLIR][affine-loop-fusion] Handle defining ops between the source and dest loops.

We should prevent loops with iter_args to be fused

Hi Diego, do you prefer to have it in this patch? I could do it by simply checking the source and dest affine.for, if one of them uses iter_args (or returns a SSA value), we prevent them to be fused.

Feb 22 2021, 5:25 AM · Restricted Project

Feb 20 2021

bondhugula requested changes to D97030: [MLIR][affine-loop-fusion] Handle defining ops between the source and dest loops.

Some comments on clarifying doc / code comments.

Feb 20 2021, 8:52 AM · Restricted Project
bondhugula added a comment to D97030: [MLIR][affine-loop-fusion] Handle defining ops between the source and dest loops.

Hey tungld,

Thanks for the patch! There is a different but somehow related issue that is being address here: https://reviews.llvm.org/D97032.
I would like to know what @bondhugula and @andydavis1 think but I think we might want to add the proper %0 -> %i1 dependence to the MDG,
instead of special-casing the load scenario. Even though it's not strictly a memref dependence, it's a memref-related dependence between two graph nodes.

Feb 20 2021, 8:43 AM · Restricted Project
bondhugula accepted D97032: [MLIR][affine] Prevent fusion when ops with memory effect free are present between producer and consumer.

LGTM.

Feb 20 2021, 8:30 AM · Restricted Project
bondhugula accepted D96821: make Affine parallel and yield ops MemRefsNormalizable .

Nit: ... also be made to be MemRefsNormalizable -> should have the MemRefsNormalizable trait.

Feb 20 2021, 8:21 AM · Restricted Project

Feb 13 2021

bondhugula added inline comments to D96397: Add C++ helpers to manage Unranked Memref in native code.
Feb 13 2021, 3:44 AM · Restricted Project
bondhugula added inline comments to D96461: [mlir][vector] Add missing support for contract of integer lowering..
Feb 13 2021, 3:33 AM · Restricted Project
bondhugula added inline comments to D96635: [mlir][Linalg] Add utility method to reshape ops to express output shape in terms of input shape..
Feb 13 2021, 3:27 AM · Restricted Project
bondhugula accepted D96284: separate AffineMapAccessInterface from AffineRead/WriteOpInterface .

LGTM, thanks. Please do resolve the remaining comments.

Feb 13 2021, 12:47 AM · Restricted Project
bondhugula requested changes to D96284: separate AffineMapAccessInterface from AffineRead/WriteOpInterface .
Feb 13 2021, 12:46 AM · Restricted Project

Feb 11 2021

bondhugula added inline comments to D96397: Add C++ helpers to manage Unranked Memref in native code.
Feb 11 2021, 6:14 PM · Restricted Project
bondhugula requested changes to D96284: separate AffineMapAccessInterface from AffineRead/WriteOpInterface .
Feb 11 2021, 6:09 PM · Restricted Project
bondhugula added inline comments to D96284: separate AffineMapAccessInterface from AffineRead/WriteOpInterface .
Feb 11 2021, 6:09 PM · Restricted Project
bondhugula resigned from D96145: [mlir] Model MemRef memory space as Attribute.
Feb 11 2021, 6:05 PM · Restricted Project
bondhugula added a comment to D96145: [mlir] Model MemRef memory space as Attribute.

Looks fine for the minor comments part I had.

Feb 11 2021, 6:05 PM · Restricted Project

Feb 10 2021

bondhugula added inline comments to D95847: [mlir][sparse] sparse tensor storage implementation.
Feb 10 2021, 11:30 AM · Restricted Project
bondhugula added a comment to D96284: separate AffineMapAccessInterface from AffineRead/WriteOpInterface .

Please also make the commit title more descriptive.

Feb 10 2021, 11:18 AM · Restricted Project
bondhugula added a comment to D96284: separate AffineMapAccessInterface from AffineRead/WriteOpInterface .

Thanks for making the "memref deferencing ops" thing more systematic here. This is looking good to me. Mostly polishing and doc related comments.

Feb 10 2021, 11:18 AM · Restricted Project
bondhugula requested changes to D96284: separate AffineMapAccessInterface from AffineRead/WriteOpInterface .

Mostly minor comments to address.

Feb 10 2021, 11:11 AM · Restricted Project
bondhugula added inline comments to D96192: Add convenience C++ helper to manipulate ranked strided memref.
Feb 10 2021, 11:07 AM · Restricted Project

Feb 9 2021

bondhugula committed rGfdfd6478372c: [MLIR] NFC Fix vector transforms build warnings (authored by bondhugula).
[MLIR] NFC Fix vector transforms build warnings
Feb 9 2021, 9:14 PM
bondhugula committed rG5400f602cd5a: [MLIR] Update affine.for unroll utility for iter_args support (authored by bondhugula).
[MLIR] Update affine.for unroll utility for iter_args support
Feb 9 2021, 9:09 PM
bondhugula closed D96383: [MLIR] Update affine.for unroll utility for iter_args support.
Feb 9 2021, 9:09 PM · Restricted Project
bondhugula added inline comments to D96383: [MLIR] Update affine.for unroll utility for iter_args support.
Feb 9 2021, 9:08 PM · Restricted Project
bondhugula updated the diff for D96383: [MLIR] Update affine.for unroll utility for iter_args support.

Address review comment on auto.

Feb 9 2021, 9:08 PM · Restricted Project
bondhugula requested changes to D96185: [mlir][Vector] Introduce 'vector.load' and 'vector.store' ops.
Feb 9 2021, 6:32 PM · Restricted Project
bondhugula updated the diff for D96383: [MLIR] Update affine.for unroll utility for iter_args support.

Rebase on main tip. Minor doc comment and commit summary update.

Feb 9 2021, 5:36 PM · Restricted Project
bondhugula requested review of D96383: [MLIR] Update affine.for unroll utility for iter_args support.
Feb 9 2021, 4:31 PM · Restricted Project

Feb 8 2021

bondhugula accepted D96192: Add convenience C++ helper to manipulate ranked strided memref.

A few more minor comments. Please also address the commit summary comment above. This functionality is looking great to me. Please do go with Alex's deeper review comments.

Feb 8 2021, 11:30 PM · Restricted Project
Herald added a reviewer for D75837: [MLIR] Introduce std.execute_region op: bollu.

This op will have to be moved to the right dialect once the std dialect split completes - mostly scf.

Feb 8 2021, 4:30 PM · Restricted Project, Restricted Project
bondhugula committed rG333d2cfc707d: [MLIR][NFC] Fix std.copysign op documentation (authored by bondhugula).
[MLIR][NFC] Fix std.copysign op documentation
Feb 8 2021, 4:29 PM
bondhugula closed D96217: [MLIR][NFC] Fix std.copysign op documentation.
Feb 8 2021, 4:29 PM · Restricted Project

Feb 7 2021

bondhugula requested changes to D96191: [MLIR] Add a generic VerifyFullyConverted pass.
Feb 7 2021, 12:19 AM · Restricted Project
bondhugula added inline comments to D96169: [mlir][spirv] Add support for OpTypeSampledImage.
Feb 7 2021, 12:11 AM · Restricted Project
bondhugula requested changes to D96145: [mlir] Model MemRef memory space as Attribute.

Please remove RFC from the commit title. Also,
"MemRef memory space as Attribute" -> "Model MemRef memory space as Attribute" or "Switch MemRef memory space: unsigned -> Attribute".

Feb 7 2021, 12:06 AM · Restricted Project

Feb 6 2021

bondhugula added a comment to D96192: Add convenience C++ helper to manipulate ranked strided memref.

Can you please add a commit summary as well - "manipulate ranked strided memref" doesn't capture all of this support accurately.

Feb 6 2021, 10:53 PM · Restricted Project
bondhugula requested changes to D96192: Add convenience C++ helper to manipulate ranked strided memref.
Feb 6 2021, 10:51 PM · Restricted Project
bondhugula requested review of D96217: [MLIR][NFC] Fix std.copysign op documentation.
Feb 6 2021, 10:31 PM · Restricted Project
bondhugula committed rG05c6c648ece3: [MLIR] [affine-loop-fusion] Fix a bug about non-result ops in affine-loop-fusion (authored by tungld).
[MLIR] [affine-loop-fusion] Fix a bug about non-result ops in affine-loop-fusion
Feb 6 2021, 12:06 AM
bondhugula closed D95668: [MLIR] [affine-loop-fusion] Fix a bug about non-result ops in affine-loop-fusion.
Feb 6 2021, 12:06 AM · Restricted Project

Feb 5 2021

bondhugula added inline comments to D96185: [mlir][Vector] Introduce 'vector.load' and 'vector.store' ops.
Feb 5 2021, 9:12 PM · Restricted Project
bondhugula accepted D96026: [mlir] Add `const` qualifiers to `AffineMap` methods.
Feb 5 2021, 3:02 AM · Restricted Project
bondhugula added a comment to D95668: [MLIR] [affine-loop-fusion] Fix a bug about non-result ops in affine-loop-fusion.

Thanks for improving this. Let me know if you want me to commit this for you.

Feb 5 2021, 2:59 AM · Restricted Project
bondhugula accepted D95668: [MLIR] [affine-loop-fusion] Fix a bug about non-result ops in affine-loop-fusion.

LGTM

Feb 5 2021, 2:59 AM · Restricted Project

Feb 4 2021

bondhugula added a comment to D96093: [mlir] Introduce more intuitive wording for attributes..

"inherent" (or "core") and "external" are looking good to me. Thanks!

Feb 4 2021, 6:29 PM · Restricted Project
bondhugula added inline comments to D95668: [MLIR] [affine-loop-fusion] Fix a bug about non-result ops in affine-loop-fusion.
Feb 4 2021, 7:31 AM · Restricted Project
bondhugula added a comment to D95668: [MLIR] [affine-loop-fusion] Fix a bug about non-result ops in affine-loop-fusion.

@bondhugula any additional modification needed? thanks!

Feb 4 2021, 7:28 AM · Restricted Project
bondhugula added inline comments to D95330: [MLIR][GPU][NVVM] Add warp synchronous matrix-multiply accumulate ops.
Feb 4 2021, 12:52 AM · Restricted Project
bondhugula added inline comments to D95330: [MLIR][GPU][NVVM] Add warp synchronous matrix-multiply accumulate ops.
Feb 4 2021, 12:50 AM · Restricted Project
bondhugula added a comment to D95330: [MLIR][GPU][NVVM] Add warp synchronous matrix-multiply accumulate ops.

Can you use mma_fragment instead of mmafragment for better readability?

Feb 4 2021, 12:42 AM · Restricted Project
bondhugula accepted D95334: [MLIR][CUDA-RUNNER] Add WMMA Tensor core matmul test.
Feb 4 2021, 12:41 AM · Restricted Project

Feb 2 2021

bondhugula added inline comments to D95797: [mlir][Linalg] Refactor Linalg vectorization for better reuse and extensibility..
Feb 2 2021, 12:35 AM · Restricted Project
bondhugula requested changes to D95797: [mlir][Linalg] Refactor Linalg vectorization for better reuse and extensibility..
Feb 2 2021, 12:18 AM · Restricted Project

Jan 30 2021

bondhugula added inline comments to D95668: [MLIR] [affine-loop-fusion] Fix a bug about non-result ops in affine-loop-fusion.
Jan 30 2021, 5:21 AM · Restricted Project
bondhugula added inline comments to D95668: [MLIR] [affine-loop-fusion] Fix a bug about non-result ops in affine-loop-fusion.
Jan 30 2021, 5:16 AM · Restricted Project
bondhugula requested changes to D95668: [MLIR] [affine-loop-fusion] Fix a bug about non-result ops in affine-loop-fusion.

Thanks for noticing this issue. A couple of comments.

Jan 30 2021, 5:15 AM · Restricted Project

Jan 26 2021

bondhugula added inline comments to D95334: [MLIR][CUDA-RUNNER] Add WMMA Tensor core matmul test.
Jan 26 2021, 3:59 AM · Restricted Project
bondhugula added inline comments to D95330: [MLIR][GPU][NVVM] Add warp synchronous matrix-multiply accumulate ops.
Jan 26 2021, 3:10 AM · Restricted Project
bondhugula accepted D95333: [MLIR][NVVM] Add test cases to check translation of matrix-multiply accumulate ops to the corresponding intrinsics in NVPTX backend.
Jan 26 2021, 2:48 AM · Restricted Project
bondhugula accepted D95332: [MLIR][CUDA-RUNNER] Add CL options to pass SM version and index-bitwidth.
Jan 26 2021, 2:43 AM · Restricted Project
bondhugula accepted D95334: [MLIR][CUDA-RUNNER] Add WMMA Tensor core matmul test.

Looks great.

Jan 26 2021, 2:42 AM · Restricted Project
bondhugula added a comment to D95394: Add a new Dialect interface to control implicit attribute propagation..

This seems to me like a 'tip of the iceberg' kind of solution. What happens if the attribute needs to be transformed? What happens if there is not a 1:1 correspondance between input operations and output operations in a transformation? etc. etc. I'm not sure what the right solution is, but this seems to be a narrowly scoped infrastructure solution to a tensorflow problem. I'd be much happier if this solved other obvious problems.

Jan 26 2021, 12:49 AM · Restricted Project

Jan 25 2021

bondhugula added inline comments to D95394: Add a new Dialect interface to control implicit attribute propagation..
Jan 25 2021, 7:56 PM · Restricted Project
bondhugula added inline comments to D95394: Add a new Dialect interface to control implicit attribute propagation..
Jan 25 2021, 6:56 PM · Restricted Project
bondhugula requested changes to D95394: Add a new Dialect interface to control implicit attribute propagation..
Jan 25 2021, 6:46 PM · Restricted Project
bondhugula added inline comments to D95394: Add a new Dialect interface to control implicit attribute propagation..
Jan 25 2021, 6:43 PM · Restricted Project

Jan 23 2021

bondhugula added inline comments to D95047: [MLIR] Add support for extracting an integer sample point (if one exists) from an unbounded FlatAffineConstraints..
Jan 23 2021, 3:52 AM · Restricted Project

Jan 20 2021

bondhugula added inline comments to D92876: [mlir][Affine] Add support for multi-store producer fusion.
Jan 20 2021, 9:39 AM · Restricted Project

Jan 19 2021

bondhugula accepted D92876: [mlir][Affine] Add support for multi-store producer fusion.

This looks really great! A bunch of minor comments to address.

Jan 19 2021, 4:26 PM · Restricted Project
bondhugula added inline comments to D92876: [mlir][Affine] Add support for multi-store producer fusion.
Jan 19 2021, 4:17 PM · Restricted Project
bondhugula added inline comments to D92876: [mlir][Affine] Add support for multi-store producer fusion.
Jan 19 2021, 4:07 PM · Restricted Project

Jan 6 2021

bondhugula requested changes to D94116: Update the maximum integer bitwidth in MLIR. .

Can you please add a proper commit summary with the rationale (basically whatever is in the discussion now)? Also, as @jpienaar points out, the commit title isn't really accurate.

Jan 6 2021, 6:14 PM · Restricted Project

Jan 1 2021

bondhugula added a comment to D93947: [mlir] Fix indexing of first offset operand in ops that implement OffsetSizeAndStrideOpInterface.

Missing a test case?

Jan 1 2021, 5:36 PM · Restricted Project
bondhugula committed rG3bcca6b12d4c: [MLIR] Fix affine_map compose with multi-symbols (authored by yaochengji).
[MLIR] Fix affine_map compose with multi-symbols
Jan 1 2021, 5:31 PM
bondhugula closed D93831: fix affine_map compose with multi-symbols.
Jan 1 2021, 5:31 PM · Restricted Project

Dec 31 2020

bondhugula added a comment to D93831: fix affine_map compose with multi-symbols.

Thanks very much for fixing this!

Dec 31 2020, 7:36 AM · Restricted Project

Dec 24 2020

bondhugula requested changes to D93704: [mlir][Linalg] Introduce linalg.pad_tensor op..
Dec 24 2020, 10:30 AM · Restricted Project

Dec 23 2020

bondhugula added a comment to D93120: [mlir][vector] Add more vector Ops canonicalization.

@ThomasRaoux, not on a directly related note, but this file VectorOps.cpp is taking nearly 12-15s to build on a fast workstation -- directly increasing build times / critical path towards mlir-opt and other targets for everyone. It'd be great to consider refactoring among next steps.

Dec 23 2020, 11:20 PM · Restricted Project
bondhugula added inline comments to D93717: [IR] Add an ImplicitLocOpBuilder helper class for building IR with the same loc..
Dec 23 2020, 10:12 AM · Restricted Project
bondhugula added inline comments to D93717: [IR] Add an ImplicitLocOpBuilder helper class for building IR with the same loc..
Dec 23 2020, 10:12 AM · Restricted Project

Dec 22 2020

bondhugula committed rG3e07b0b9d336: [MLIR] Fix lowering of affine operations with return values (authored by pr4tgpt).
[MLIR] Fix lowering of affine operations with return values
Dec 22 2020, 8:16 AM
bondhugula closed D93090: [MLIR] Fix lowering of affine operations with return values..
Dec 22 2020, 8:16 AM · Restricted Project
bondhugula accepted D93090: [MLIR] Fix lowering of affine operations with return values..
Dec 22 2020, 3:36 AM · Restricted Project
bondhugula accepted D93090: [MLIR] Fix lowering of affine operations with return values..
Dec 22 2020, 3:11 AM · Restricted Project

Dec 21 2020

bondhugula resigned from D93500: [mlir[[vector] Extend Transfer read/write ops to support tensor types..
Dec 21 2020, 7:39 AM · Restricted Project

Dec 19 2020

bondhugula requested changes to D93500: [mlir[[vector] Extend Transfer read/write ops to support tensor types..
Dec 19 2020, 10:17 AM · Restricted Project
bondhugula requested changes to D93090: [MLIR] Fix lowering of affine operations with return values..

Mostly minor comments.

Dec 19 2020, 12:47 AM · Restricted Project

Dec 11 2020

bondhugula added inline comments to D93090: [MLIR] Fix lowering of affine operations with return values..
Dec 11 2020, 12:14 AM · Restricted Project
bondhugula requested changes to D93090: [MLIR] Fix lowering of affine operations with return values..
Dec 11 2020, 12:11 AM · Restricted Project
bondhugula added a reviewer for D93090: [MLIR] Fix lowering of affine operations with return values.: dcaballe.
Dec 11 2020, 12:08 AM · Restricted Project

Dec 10 2020

bondhugula added inline comments to D92876: [mlir][Affine] Add support for multi-store producer fusion.
Dec 10 2020, 1:03 AM · Restricted Project
bondhugula requested changes to D92876: [mlir][Affine] Add support for multi-store producer fusion.

Thanks for this significant update! Some superficial comments to start with. It may be good to handle the pass documentation update as well, starting with an update to the summary. It's currently emptly: https://mlir.llvm.org/docs/Passes/#-affine-loop-fusion-fuse-affine-loop-nests
You could mention about the fusion strategies producer/consumer and sibling ones as well. Here's the previous doc paragraph that disappeared (was overwritten) when pass documentation was migrated to auto-generated stuff.

Dec 10 2020, 12:55 AM · Restricted Project