Page MenuHomePhabricator
Feed Advanced Search

Thu, Dec 2

pifon2a accepted D114824: [mlir][bufferization] fixed typo in to_memref doc.

Thank you, Aart!

Thu, Dec 2, 10:14 AM · Restricted Project

Wed, Dec 1

pifon2a committed rG3a6c4f307bd0: [mlir] Add a helper for TiledLoopOp to get an operand tied to the bbArg. (authored by pifon2a).
[mlir] Add a helper for TiledLoopOp to get an operand tied to the bbArg.
Wed, Dec 1, 12:32 AM
pifon2a closed D114852: [mlir] Add a helper for TiledLoopOp to get an operand tied to the bbArg..
Wed, Dec 1, 12:32 AM · Restricted Project
pifon2a updated the diff for D114852: [mlir] Add a helper for TiledLoopOp to get an operand tied to the bbArg..

nit

Wed, Dec 1, 12:30 AM · Restricted Project
pifon2a updated the diff for D114852: [mlir] Add a helper for TiledLoopOp to get an operand tied to the bbArg..

format

Wed, Dec 1, 12:26 AM · Restricted Project
pifon2a requested review of D114852: [mlir] Add a helper for TiledLoopOp to get an operand tied to the bbArg..
Wed, Dec 1, 12:24 AM · Restricted Project

Tue, Nov 30

pifon2a committed rG97db64082eb0: [mlir] Add bazel build for BufferizationToMemRef. (authored by pifon2a).
[mlir] Add bazel build for BufferizationToMemRef.
Tue, Nov 30, 4:28 AM
pifon2a added a comment to D114780: Added missing link references..

I think only Pass and Transforms are needed. I already submitted it. Thanks!

Tue, Nov 30, 4:14 AM · Restricted Project
pifon2a committed rGf910aa910555: [mlir] Fix BufferizationToMemRef build. (authored by pifon2a).
[mlir] Fix BufferizationToMemRef build.
Tue, Nov 30, 4:11 AM
pifon2a added a comment to D114698: [mlir] Move bufferization-related passes to `bufferization` dialect..

@mehdi_amini @springerm Yes, the passes like tensor-bufferize will stay in their dialects. It might be possible to reuse BufferizableOpInterface for the "gradual" bufferization instead of using the patterns. I think that this is how these two bufferization approaches will converge in the end.

I am confused by your answer: tensor-buffering isn’t part of the gradual bufferization right now?

Tue, Nov 30, 1:34 AM · Restricted Project
pifon2a committed rGf89bb3c012b4: [mlir] Move bufferization-related passes to `bufferization` dialect. (authored by pifon2a).
[mlir] Move bufferization-related passes to `bufferization` dialect.
Tue, Nov 30, 12:59 AM
pifon2a closed D114698: [mlir] Move bufferization-related passes to `bufferization` dialect..
Tue, Nov 30, 12:59 AM · Restricted Project
pifon2a added a comment to D114698: [mlir] Move bufferization-related passes to `bufferization` dialect..

@mehdi_amini @springerm Yes, the passes like tensor-bufferize will stay in their dialects. It might be possible to reuse BufferizableOpInterface for the "gradual" bufferization instead of using the patterns. I think that this is how these two bufferization approaches will converge in the end.

Tue, Nov 30, 12:30 AM · Restricted Project

Mon, Nov 29

pifon2a accepted D114730: [mlir][sparse] some leftover cleanup from migration to bufferization dialect.

Thank you!

Mon, Nov 29, 12:00 PM · Restricted Project
pifon2a added a comment to D114552: [mlir] Move memref.[tensor_load|buffer_cast|clone] to "bufferization" dialect..

LGTM for the sparse part (there are some dangling references in the doc now to "memref" but since this is submitted, I will fix that in a follow up).

Mon, Nov 29, 10:02 AM · Restricted Project
pifon2a accepted D114711: [mlir] Handle an edge case when folding reshapes with multiple trailing 1 dimensions.
Mon, Nov 29, 9:27 AM · Restricted Project
pifon2a updated the diff for D114698: [mlir] Move bufferization-related passes to `bufferization` dialect..
  1. The first line is used as subject, next lines as comment. #

Update

Mon, Nov 29, 4:45 AM · Restricted Project
pifon2a updated the diff for D114698: [mlir] Move bufferization-related passes to `bufferization` dialect..

Update the names of the ops.

Mon, Nov 29, 4:42 AM · Restricted Project
pifon2a requested review of D114698: [mlir] Move bufferization-related passes to `bufferization` dialect..
Mon, Nov 29, 4:37 AM · Restricted Project
pifon2a accepted D114233: [mlir] Decompose Bufferization Clone operation into Memref Alloc and Copy..
Mon, Nov 29, 2:38 AM · Restricted Project

Thu, Nov 25

pifon2a committed rG57470abc4125: [mlir] Move memref.[tensor_load|buffer_cast|clone] to "bufferization" dialect. (authored by pifon2a).
[mlir] Move memref.[tensor_load|buffer_cast|clone] to "bufferization" dialect.
Thu, Nov 25, 2:58 AM
pifon2a closed D114552: [mlir] Move memref.[tensor_load|buffer_cast|clone] to "bufferization" dialect..
Thu, Nov 25, 2:57 AM · Restricted Project
pifon2a updated the diff for D114552: [mlir] Move memref.[tensor_load|buffer_cast|clone] to "bufferization" dialect..

address the comments.

Thu, Nov 25, 2:44 AM · Restricted Project
pifon2a resigned from D114552: [mlir] Move memref.[tensor_load|buffer_cast|clone] to "bufferization" dialect..
Thu, Nov 25, 2:43 AM · Restricted Project
pifon2a accepted D114552: [mlir] Move memref.[tensor_load|buffer_cast|clone] to "bufferization" dialect..
Thu, Nov 25, 2:43 AM · Restricted Project
pifon2a added a reverting change for rGee1bf186723a: [mlir][SCF] Further simplify affine maps during `for-loop-canonicalization`: rG3c228573bcb8: Revert "[mlir][SCF] Further simplify affine maps during `for-loop….
Thu, Nov 25, 1:55 AM
pifon2a committed rG3c228573bcb8: Revert "[mlir][SCF] Further simplify affine maps during `for-loop… (authored by pifon2a).
Revert "[mlir][SCF] Further simplify affine maps during `for-loop…
Thu, Nov 25, 1:55 AM
pifon2a added a reverting change for D114138: [mlir][SCF] Further simplify affine maps during `for-loop-canonicalization`: rG3c228573bcb8: Revert "[mlir][SCF] Further simplify affine maps during `for-loop….
Thu, Nov 25, 1:55 AM · Restricted Project

Wed, Nov 24

pifon2a updated the diff for D114552: [mlir] Move memref.[tensor_load|buffer_cast|clone] to "bufferization" dialect..

Clean-up

Wed, Nov 24, 11:14 AM · Restricted Project
pifon2a requested review of D114552: [mlir] Move memref.[tensor_load|buffer_cast|clone] to "bufferization" dialect..
Wed, Nov 24, 11:10 AM · Restricted Project

Tue, Nov 23

pifon2a committed rGc7cc70c8f877: Revert "Revert "[mlir] Move AllocationOpInterface to… (authored by pifon2a).
Revert "Revert "[mlir] Move AllocationOpInterface to…
Tue, Nov 23, 1:50 AM
pifon2a accepted D114388: [mlir][Vector] Thread 0-d vectors through InsertElementOp..
Tue, Nov 23, 1:09 AM · Restricted Project

Mon, Nov 22

pifon2a accepted D114391: [mlir][memref] Fix expanded shape ops memref.cast folding with changed type.
Mon, Nov 22, 1:45 PM · Restricted Project
pifon2a added a reverting change for rG3028bca6a987: [mlir] Move AllocationOpInterface to Bufferize/IR/AllocationOpInterface.td.: rGde18b7dee6a8: Revert "[mlir] Move AllocationOpInterface to Bufferize/IR/AllocationOpInterface..
Mon, Nov 22, 12:37 PM
pifon2a committed rGde18b7dee6a8: Revert "[mlir] Move AllocationOpInterface to Bufferize/IR/AllocationOpInterface. (authored by pifon2a).
Revert "[mlir] Move AllocationOpInterface to Bufferize/IR/AllocationOpInterface.
Mon, Nov 22, 12:37 PM
pifon2a added a reverting change for D114147: [mlir] Move AllocationOpInterface to Bufferize/IR/AllocationOpInterface.td.: rGde18b7dee6a8: Revert "[mlir] Move AllocationOpInterface to Bufferize/IR/AllocationOpInterface..
Mon, Nov 22, 12:37 PM · Restricted Project
pifon2a committed rG3028bca6a987: [mlir] Move AllocationOpInterface to Bufferize/IR/AllocationOpInterface.td. (authored by pifon2a).
[mlir] Move AllocationOpInterface to Bufferize/IR/AllocationOpInterface.td.
Mon, Nov 22, 12:01 PM
pifon2a closed D114147: [mlir] Move AllocationOpInterface to Bufferize/IR/AllocationOpInterface.td..
Mon, Nov 22, 12:01 PM · Restricted Project
pifon2a added inline comments to D114233: [mlir] Decompose Bufferization Clone operation into Memref Alloc and Copy..
Mon, Nov 22, 4:31 AM · Restricted Project
pifon2a added inline comments to D114147: [mlir] Move AllocationOpInterface to Bufferize/IR/AllocationOpInterface.td..
Mon, Nov 22, 3:07 AM · Restricted Project
pifon2a updated the diff for D114147: [mlir] Move AllocationOpInterface to Bufferize/IR/AllocationOpInterface.td..

Address Stephan's comments.

Mon, Nov 22, 2:52 AM · Restricted Project
pifon2a added a comment to D114147: [mlir] Move AllocationOpInterface to Bufferize/IR/AllocationOpInterface.td..

It wasn't clear to me that an agreement had been reached in the RFC to start populating anything here.

Mon, Nov 22, 1:38 AM · Restricted Project
pifon2a requested changes to D114233: [mlir] Decompose Bufferization Clone operation into Memref Alloc and Copy..
Mon, Nov 22, 1:36 AM · Restricted Project
pifon2a accepted D114175: [mlir][linalg] Add makeComposedPadHighOp..

Nice!

Mon, Nov 22, 1:33 AM · Restricted Project
pifon2a updated the diff for D114147: [mlir] Move AllocationOpInterface to Bufferize/IR/AllocationOpInterface.td..

Rename the dialect to "bufferization".

Mon, Nov 22, 1:13 AM · Restricted Project

Thu, Nov 18

pifon2a requested review of D114147: [mlir] Move AllocationOpInterface to Bufferize/IR/AllocationOpInterface.td..
Thu, Nov 18, 1:37 AM · Restricted Project

Mon, Nov 15

pifon2a committed rG9b1d90e8ac9c: [mlir] Move min/max ops from Std to Arith. (authored by pifon2a).
[mlir] Move min/max ops from Std to Arith.
Mon, Nov 15, 4:20 AM
pifon2a closed D113881: [mlir] Move min/max ops from Std to Arith..
Mon, Nov 15, 4:19 AM · Restricted Project
pifon2a requested review of D113881: [mlir] Move min/max ops from Std to Arith..
Mon, Nov 15, 3:53 AM · Restricted Project

Nov 3 2021

pifon2a added a comment to D113117: [mlir] Remove `tensor::DimOp` from `AffineOps.cpp`..

Can you provide a description please?

Nov 3 2021, 11:05 AM · Restricted Project
pifon2a updated the summary of D113117: [mlir] Remove `tensor::DimOp` from `AffineOps.cpp`..
Nov 3 2021, 11:05 AM · Restricted Project
pifon2a requested review of D113117: [mlir] Remove `tensor::DimOp` from `AffineOps.cpp`..
Nov 3 2021, 8:49 AM · Restricted Project

Oct 27 2021

pifon2a accepted D112640: [mlir][Linalg] NFC - Make more option names consistent..
Oct 27 2021, 11:27 AM · Restricted Project

Oct 26 2021

pifon2a added a comment to D112570: Use vectorShape instead of vectorWidth.

Thank you, Ahmed. I submitted smth like that already https://github.com/llvm/llvm-project/commit/a9a0ea92d13c20227c03b212d2dfa6a41a21ea6f

Oct 26 2021, 12:38 PM · Restricted Project
pifon2a committed rG7d79a2576942: [mlir] Use float literals to make Windows build happy. (authored by pifon2a).
[mlir] Use float literals to make Windows build happy.
Oct 26 2021, 12:37 PM
pifon2a committed rGa9a0ea92d13c: [mlir] Update Erf approximation. (authored by pifon2a).
[mlir] Update Erf approximation.
Oct 26 2021, 12:28 PM
pifon2a committed rG96cee29762f0: [mlir] Allow polynomial approximations for N-d vectors. (authored by pifon2a).
[mlir] Allow polynomial approximations for N-d vectors.
Oct 26 2021, 11:52 AM
pifon2a closed D112566: [mlir] Allow polynomial approximations for N-d vectors..
Oct 26 2021, 11:52 AM · Restricted Project
pifon2a requested review of D112566: [mlir] Allow polynomial approximations for N-d vectors..
Oct 26 2021, 11:39 AM · Restricted Project

Oct 25 2021

pifon2a accepted D112414: [mlr][Linalg] NFC - Add option to hook vector.multi_reduction lowering to strategies..
Oct 25 2021, 2:58 AM · Restricted Project

Oct 22 2021

pifon2a accepted D112332: [mlir][Linalg] NFC - Drop Optional in favor of FailureOr.
Oct 22 2021, 9:50 AM · Restricted Project

Oct 18 2021

pifon2a accepted D111974: Fix bazel build..
Oct 18 2021, 1:09 AM · Restricted Project

Oct 14 2021

pifon2a accepted D111825: [mlir][NFC] Refactor linalg vectorization for reduction ops.
Oct 14 2021, 11:20 AM · Restricted Project

Oct 12 2021

pifon2a accepted D111565: [mlir][Linalg] Enable vectorization of 'mul', 'and', 'or' and 'xor' reductions.

I think this is ready to be landed.

Oct 12 2021, 1:33 PM · Restricted Project
pifon2a accepted D111626: [mlir][Linalg] Refactor and improve vectorization to add support for reduction into 0-d tensors..
Oct 12 2021, 3:57 AM · Restricted Project
pifon2a accepted D111558: [mlir][Linalg] NFC - Refactor vector.broadcast op verification logic and make it available as a precondition in Linalg vectorization..

LGTM, once the comments are addressed.

Oct 12 2021, 12:57 AM · Restricted Project
pifon2a accepted D111603: [mlir][Vector] Add support for lowering 0-d transfers to load/store..
Oct 12 2021, 12:19 AM · Restricted Project
pifon2a accepted D111563: [mlir][Linalg] Enable vectorization of explicit broadcasts.
Oct 12 2021, 12:14 AM · Restricted Project

Oct 11 2021

pifon2a added a comment to D111563: [mlir][Linalg] Enable vectorization of explicit broadcasts.

Could you please verify if all tests still pass? For example, we might need to modify applyPermutationMap in AffineMap.h.

template <typename T>
SmallVector<T> applyPermutationMap(AffineMap map, llvm::ArrayRef<T> source) {
  assert(map.isProjectedPermutation());
  assert(map.getNumInputs() == source.size());
  SmallVector<T> result;
  result.reserve(map.getNumResults());
  for (unsigned i = 0, e = map.getNumResults(); i < e; ++i) {
    /* fix
    if (auto expr = map.getResult(i).dyn_cast<AffineConstantExpr>()) {
      result.push_back(0);
      continue;
    }
    */
    unsigned dim = map.getDimPosition(i);
    result.push_back(source[dim]);
  }
  return result;
}
Oct 11 2021, 12:24 PM · Restricted Project
pifon2a accepted D111565: [mlir][Linalg] Enable vectorization of 'mul', 'and', 'or' and 'xor' reductions.
Oct 11 2021, 11:05 AM · Restricted Project
pifon2a accepted D111442: [mlir][Vector] Let vector.multi_reduction reduce down to a scalar..

Just a general comment. This is quite interesting that this kind of transformation/canonicalization is happening now on different levels of the lowering stack. There are transforms on HLO level, that convert reductions to row/column reductions or to a 1D reduction. We can do the same transformation in Linalg, which would not prevent tile-n-fuse of input producers happening. Or we might need both.

Oct 11 2021, 1:24 AM · Restricted Project

Oct 5 2021

pifon2a accepted D110854: [mlir][Linalg] Add support for min/max reduction vectorization in linalg.generic.

Great, thank you, Diego!

Oct 5 2021, 4:29 AM · Restricted Project
pifon2a abandoned D110832: [mlir] Rewrite transfer_read as transfer_read + broadcast + transpose..

Actually, I think the current TransferReadPermutationLowering and TransferOpReduceRank are good enough. I wish I knew about them before implementing this.

Oct 5 2021, 4:27 AM · Restricted Project
pifon2a accepted D111127: [mlir] Convert ConstShapeOp to a static tensor type..
Oct 5 2021, 3:02 AM · Restricted Project

Oct 1 2021

pifon2a committed rG693c61b2e050: [mlir] Enable loop peeling for "reduction" dimensions of tiled_loop. (authored by pifon2a).
[mlir] Enable loop peeling for "reduction" dimensions of tiled_loop.
Oct 1 2021, 4:08 AM
pifon2a closed D110919: [mlir] Enable loop peeling for "reduction" dimensions of tiled_loop..
Oct 1 2021, 4:08 AM · Restricted Project
pifon2a requested review of D110919: [mlir] Enable loop peeling for "reduction" dimensions of tiled_loop..
Oct 1 2021, 4:03 AM · Restricted Project
pifon2a added inline comments to D110854: [mlir][Linalg] Add support for min/max reduction vectorization in linalg.generic.
Oct 1 2021, 12:17 AM · Restricted Project

Sep 30 2021

pifon2a added inline comments to D110854: [mlir][Linalg] Add support for min/max reduction vectorization in linalg.generic.
Sep 30 2021, 1:14 PM · Restricted Project
pifon2a requested review of D110832: [mlir] Rewrite transfer_read as transfer_read + broadcast + transpose..
Sep 30 2021, 6:52 AM · Restricted Project

Sep 29 2021

pifon2a accepted D110678: [mlir][Linalg] Rewrite CodegenStrategy to populate a pass pipeline..

So much nicer now!

Sep 29 2021, 1:43 AM · Restricted Project

Sep 28 2021

pifon2a committed rG9fb57c8c1dd8: [mlir] Add min/max operations to Standard. (authored by pifon2a).
[mlir] Add min/max operations to Standard.
Sep 28 2021, 12:41 AM
pifon2a closed D110540: [mlir] Add min/max operations to Standard..
Sep 28 2021, 12:40 AM · Restricted Project
pifon2a added a comment to D110540: [mlir] Add min/max operations to Standard..

Thanks, Alex! LGTM.
Could you please add some tests to Dialect/Standard/ops.mlir and Dialect/Standard/invalid.mlir.

Sep 28 2021, 12:26 AM · Restricted Project
pifon2a updated the diff for D110540: [mlir] Add min/max operations to Standard..

Address the comments.

Sep 28 2021, 12:24 AM · Restricted Project
pifon2a added inline comments to D110540: [mlir] Add min/max operations to Standard..
Sep 28 2021, 12:24 AM · Restricted Project

Sep 27 2021

pifon2a updated the diff for D110540: [mlir] Add min/max operations to Standard..

Last typo

Sep 27 2021, 12:33 PM · Restricted Project
pifon2a added inline comments to D110540: [mlir] Add min/max operations to Standard..
Sep 27 2021, 12:17 PM · Restricted Project
pifon2a updated the diff for D110540: [mlir] Add min/max operations to Standard..

Addressed the comments. Thank you Mehdi and Aart!

Sep 27 2021, 12:16 PM · Restricted Project
pifon2a updated the diff for D110540: [mlir] Add min/max operations to Standard..

Updated documentation

Sep 27 2021, 6:48 AM · Restricted Project
pifon2a abandoned D110544: [mlir] Add min/max operations to Standard..
Sep 27 2021, 5:58 AM · Restricted Project
pifon2a requested review of D110544: [mlir] Add min/max operations to Standard..
Sep 27 2021, 5:57 AM · Restricted Project
pifon2a requested review of D110540: [mlir] Add min/max operations to Standard..
Sep 27 2021, 5:47 AM · Restricted Project

Sep 24 2021

pifon2a accepted D110303: [mlir] Create a generic reduction detection utility.

Thank you, Diego!

Sep 24 2021, 7:08 AM · Restricted Project

Sep 22 2021

pifon2a accepted D110220: [mlir][linalg] Support tile+peel with TiledLoopOp.
Sep 22 2021, 5:42 AM · Restricted Project

Sep 10 2021

pifon2a accepted D109581: [mlir][linalg] Fix bufferize pattern to allow unknown operations in body of generic.
Sep 10 2021, 3:40 AM · Restricted Project
pifon2a accepted D109581: [mlir][linalg] Fix bufferize pattern to allow unknown operations in body of generic.

Thank you, Stephan!

Sep 10 2021, 1:36 AM · Restricted Project
pifon2a accepted D109569: [mlir][linalg] Pass all operands to tile to the tile loop region builder (NFC)..

thank you a lot, Tobias!

Sep 10 2021, 12:03 AM · Restricted Project

Sep 6 2021

pifon2a committed rG58c188507f72: [mlir][linalg] Fix `FoldInitTensorWithDimOp` if dim(init_tensor) is static. (authored by pifon2a).
[mlir][linalg] Fix `FoldInitTensorWithDimOp` if dim(init_tensor) is static.
Sep 6 2021, 1:48 AM
pifon2a closed D109247: [mlir][linalg] Fix `FoldInitTensorWithDimOp` if dim(init_tensor) is static..
Sep 6 2021, 1:48 AM · Restricted Project