Integrate llvm-project and bump dependencies. #12562

vmurali · 2023-03-08T18:30:23Z

llvm-project: e510d0bda0876c4baa3a270dca39b95da7ec6d9e
mlir-hlo: e86610442f58b889a57bf814d75c4b50c769c2a3
tensorflow: 67ba341c869e30ee4a89e040cd875d12b9bc666e

Cherry-picked from LLVM:

commit 80074d5fc0ab3f165865b15f5bf55ffac0917bcd (HEAD -> integrate-3-8-2023, fork/integrate-3-8-2023)
Author: Matthias Springer <me@m-sp.org>
Date:   Fri Mar 10 11:25:15 2023 +0100

    [mlir][NFC] reifyResultShapes: Add extra error checking
    
    This change adds a new helper function `mlir::reifyResultShapes` that calls the corresponding interface method and also checks the result produced by the implementation when running in debug mode. Bugs due to incorrect interface implementations can be difficult to debug.
    
    This helper function also reduces the amount of code needed at call sites: the cast to `ReifyRankedShapedTypeOpInterface` is done in the helper function.
    
    Differential Revision: https://reviews.llvm.org/D145777

commit 32b15f601de173e9511f470f7423108d3154e582
Author: Matthias Springer <me@m-sp.org>
Date:   Fri Mar 10 11:24:43 2023 +0100

    [mlir][tensor/linalg] Fix bug in reifyResultShapes
    
    `reifyResultShapes` should return an IntegerAttr if and only if the corresponding dimension is static.
    
    Differential Revision: https://reviews.llvm.org/D145702

commit 894555cd6adf2e0faffe713373a266650b40bb4e
Author: David Green <david.green@arm.com>
Date:   Wed Mar 8 12:48:21 2023 +0000

    [AArch64] Fix load-insert-zero patterns with i8 and negative offsets.
    
    These should have been using the LDURBi instructions where the offset is
    negative, as reported from the reproducer in D144086.

Created a new commit on iree-mlir-hlo fork:

https://github.com/iree-org/iree-mhlo-fork/commit/b14e9d9b06255e4476f5698e3bfc531dec793ded

matthias-springer · 2023-03-09T17:01:01Z

compiler/src/iree/compiler/Dialect/Flow/IR/FlowOps.cpp

      } else {
-        shape.push_back(b.create<arith::ConstantIndexOp>(getLoc(), dim));
+        shape.push_back((Value)b.create<arith::ConstantIndexOp>(getLoc(), dim));


This must return an OpFoldResult.

From the documentation of reifyResultShapes:

`reifiedReturnShapes` is populated with one vector per op result. Each of those vectors contains an OpFoldResult for each dimension of the shaped type. In case a dimension in the type is static, the corresponding entry is an IntegerAttr. Otherwise, it is a Value. The given builder may be used to insert ops that compute result shapes.

In this particular case here:

shape.push_back(b.getIndexAttr(dim));

(same for a few more places below)

So

shape.push_back((Value)b.create<arith::ConstantIndexOp>(getLoc(), dim));

should be changed to

shape.push_back(b.getIndexAttr(dim));

?

* llvm-project: e510d0bda0876c4baa3a270dca39b95da7ec6d9e * mlir-hlo: e86610442f58b889a57bf814d75c4b50c769c2a3 * tensorflow: 67ba341c869e30ee4a89e040cd875d12b9bc666e

compiler/src/iree/compiler/Dialect/Flow/IR/FlowOps.cpp

llvm-external-projects/iree-dialects/lib/Dialect/LinalgExt/IR/LinalgExtInterfaces.cpp

vmurali · 2023-03-09T22:01:54Z

@hanhanW the tests fail unless I do what I had done. It was @matthias-springer's suggestion:

for every source dimension {
  if (dimension is static) {
    reifiedReturnShapes[0].push_back(builder.getIndexAttr(...));
  } else {
    (whatever getDims is doing)
  }
}

I am going to mark the comments as resolved.

MaheshRavishankar · 2023-03-09T22:04:47Z

@hanhanW the tests fail unless I do what I had done. It was @matthias-springer's suggestion:
for every source dimension {
  if (dimension is static) {
    reifiedReturnShapes[0].push_back(builder.getIndexAttr(...));
  } else {
    (whatever getDims is doing)
  }
}
I am going to mark the comments as resolved.

I think you can use createDimValues https://github.com/llvm/llvm-project/blob/e1b3c5c403b89172a3346ca8d781ef5ee82c1c01/mlir/include/mlir/Dialect/Tensor/Utils/Utils.h#L31

hanhanW · 2023-03-09T22:28:07Z

@hanhanW the tests fail unless I do what I had done. It was @matthias-springer's suggestion:
for every source dimension {
  if (dimension is static) {
    reifiedReturnShapes[0].push_back(builder.getIndexAttr(...));
  } else {
    (whatever getDims is doing)
  }
}
I am going to mark the comments as resolved.

This is very surprising. Does the getMixedTile() one also fail? According to the td file, the DispatchTensorLoadOp implements OffsetSizeAndStrideOpInterface interface.

def FLOW_DispatchTensorLoadOp : FLOW_PureOp<"dispatch.tensor.load", [
  AttrSizedOperandSegments,
  OffsetSizeAndStrideOpInterface,
  DeclareOpInterfaceMethods<ReifyRankedShapedTypeOpInterface>,
  DeclareOpInterfaceMethods<Util_TiedOpInterface, [
      "getTiedResult",
      "getTiedResultOperandIndex",
      "getTiedResultOperandIndices",
  ]>,

The getMixedTile method is part of OffsetSizeAndStrideOpInterface, which returns vector of OpFoldValue...

https://github.com/llvm/llvm-project/blob/54225c457a336b1609c6d064b2b606a9238a28b9/mlir/include/mlir/Interfaces/ViewLikeInterface.td#L173-L184

IMO, the fix should not be using casting like (Value).... We should fix it in a proper way...

vmurali · 2023-03-09T22:31:24Z

@hanhanW the tests fail unless I do what I had done. It was @matthias-springer's suggestion:
for every source dimension {
  if (dimension is static) {
    reifiedReturnShapes[0].push_back(builder.getIndexAttr(...));
  } else {
    (whatever getDims is doing)
  }
}
I am going to mark the comments as resolved.
This is very surprising. Does the getMixedTile() one also fail? According to the td file, the DispatchTensorLoadOp implements OffsetSizeAndStrideOpInterface interface.
def FLOW_DispatchTensorLoadOp : FLOW_PureOp<"dispatch.tensor.load", [
  AttrSizedOperandSegments,
  OffsetSizeAndStrideOpInterface,
  DeclareOpInterfaceMethods<ReifyRankedShapedTypeOpInterface>,
  DeclareOpInterfaceMethods<Util_TiedOpInterface, [
      "getTiedResult",
      "getTiedResultOperandIndex",
      "getTiedResultOperandIndices",
  ]>,
The getMixedTile method is part of OffsetSizeAndStrideOpInterface, which returns vector of OpFoldValue...

https://github.com/llvm/llvm-project/blob/54225c457a336b1609c6d064b2b606a9238a28b9/mlir/include/mlir/Interfaces/ViewLikeInterface.td#L173-L184

IMO, the fix should not be using casting like (Value).... We should fix it in a proper way...

No I fixed that. There's the one unresolved comment that fails if I remove the cast.

looks good so far, unblocking for now

hanhanW · 2023-03-10T05:28:40Z

I'm worried that the next round integrator does not know these information. Can you add the commits that are cherry-picked to the PR description?

I noticed that there are commits that are not cherry-picked from upstream, e.g., https://github.com/iree-org/iree-llvm-fork/commit/87c11e2f9c56ed302657b930dce1b8553becb8b8 and https://github.com/iree-org/iree-mhlo-fork/commit/b14e9d9b06255e4476f5698e3bfc531dec793ded

Are they sent out for review yet?

matthias-springer · 2023-03-10T10:46:25Z

Upstream fix for reifyResultShapes has landed. Also an additional change with extra debug mode checks to debug issues like this more easily in the future.

vmurali · 2023-03-10T16:16:50Z

Upstream fix for reifyResultShapes has landed. Also an additional change with extra debug mode checks to debug issues like this more easily in the future.

Should I pull these two: 0a9f6b8ca3fa9449cc0accba1bc11e98d6dbc6b6, 758329dc7cd3b0da835a4f865b89003263050080 ?

vmurali · 2023-03-10T16:39:57Z

https://github.com/openxla/iree/actions/runs/4380626838/jobs/7668127668 -

******************** TEST 'TENSORFLOW_TESTS :: iree_tfl_tests/llvmcpu_posenet_i8.run' FAILED ********************
Script:
--
: 'RUN: at line 2';   /usr/bin/python3 -m iree_tfl_tests.posenet_i8_test --target_backend=llvmcpu --artifacts_dir=/tmp/lit/iree_tfl_tests/Output/llvmcpu_posenet_i8.run.tmp
--
Exit Code: 1

Command Output (stdout):
--
$ ":" "RUN: at line 2"
$ "/usr/bin/python3" "-m" "iree_tfl_tests.posenet_i8_test" "--target_backend=llvmcpu" "--artifacts_dir=/tmp/lit/iree_tfl_tests/Output/llvmcpu_posenet_i8.run.tmp"
# command output:
TMPDIR = /tmp/lit/iree_tfl_tests/Output/llvmcpu_posenet_i8.run.tmp/download

# command stderr:
2023-03-10 02:39:54.077060: W tensorflow/stream_executor/platform/default/dso_loader.cc:64] Could not load dynamic library 'libcudart.so.11.0'; dlerror: libcudart.so.11.0: cannot open shared object file: No such file or directory
2023-03-10 02:39:54.077160: I tensorflow/stream_executor/cuda/cudart_stub.cc:29] Ignore above cudart dlerror if you do not have a GPU set up on your machine.
Running tests under Python 3.7.5: /usr/bin/python3
2023-03-10 02:39:56.311971: W tensorflow/stream_executor/platform/default/dso_loader.cc:64] Could not load dynamic library 'libcuda.so.1'; dlerror: libcuda.so.1: cannot open shared object file: No such file or directory
2023-03-10 02:39:56.312022: W tensorflow/stream_executor/cuda/cuda_driver.cc:269] failed call to cuInit: UNKNOWN ERROR (303)
2023-03-10 02:39:56.312052: I tensorflow/stream_executor/cuda/cuda_diagnostics.cc:156] kernel driver does not appear to be running on this host (gh-runner-presubmit-cpu-us-central1-gw6m): /proc/driver/nvidia/version does not exist
[ RUN      ] PosenetI8Test.test_compile_tflite
I0310 02:39:56.315438 139756582582080 test_util.py:50] Saving compilation artifacts and traces to '/tmp/lit/iree_tfl_tests/Output/llvmcpu_posenet_i8.run.tmp/download'
I0310 02:39:56.388114 139756582582080 test_util.py:141] Setting up for IREE
I0310 02:39:56.389639 139756582582080 binaries.py:216] Invoke IREE Pipeline:
  /work/integrations/tensorflow/python_projects/iree_tflite/iree/tools/tflite/iree-import-tflite /tmp/lit/iree_tfl_tests/Output/llvmcpu_posenet_i8.run.tmp/download/model.mlirbc --mlir-print-debuginfo --save-temp-tfl-input=/tmp/lit/iree_tfl_tests/Output/llvmcpu_posenet_i8.run.tmp/download/tflite.mlirbc --save-temp-iree-input=/tmp/lit/iree_tfl_tests/Output/llvmcpu_posenet_i8.run.tmp/download/tosa.mlirbc
  /work/full-build-dir/compiler/bindings/python/iree/compiler/tools/../_mlir_libs/iree-compile - --iree-input-type=tosa --iree-vm-bytecode-module-output-format=flatbuffer-binary --iree-hal-target-backends=llvm-cpu -o=/tmp/lit/iree_tfl_tests/Output/llvmcpu_posenet_i8.run.tmp/download/module.vmfb --iree-llvmcpu-embedded-linker-path=/work/full-build-dir/compiler/bindings/python/iree/compiler/tools/../_mlir_libs/iree-lld --mlir-print-debuginfo --mlir-print-op-on-diagnostic=false
[  FAILED  ] PosenetI8Test.test_compile_tflite
======================================================================
ERROR: test_compile_tflite (__main__.PosenetI8Test)
PosenetI8Test.test_compile_tflite
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/work/integrations/tensorflow/test/python/iree_tfl_tests/posenet_i8_test.py", line 47, in test_compile_tflite
    self.compile_and_execute()
  File "/work/integrations/tensorflow/test/python/iree_tfl_tests/test_util.py", line 149, in compile_and_execute
    import_only=False)
  File "/work/full-build-dir/compiler/bindings/python/iree/compiler/tools/tflite.py", line 154, in compile_file
    result = invoke_pipeline([import_cl, compile_cl])
  File "/work/full-build-dir/compiler/bindings/python/iree/compiler/tools/binaries.py", line 259, in invoke_pipeline
    raise CompilerToolError(stage.completed)
iree.compiler.tools.binaries.CompilerToolError: Error invoking IREE compiler tool iree-import-tflite

Debugging this. If anyone has any insights, let me know.

hanhanW · 2023-03-10T18:24:56Z

Thanks for tracking them.

(also, could you please add the other comment instead of editing my comment?)

vmurali · 2023-03-10T18:29:01Z

Thanks for tracking them.

(also, could you please add the other comment instead of editing my comment?)

Sorry I thought I quoted it.

ScottTodd · 2023-03-10T21:39:25Z

integrations/tensorflow/WORKSPACE

-TENSORFLOW_COMMIT = "eece4dba1fc65f7977085581852b0d6e6d42f04e"
+TENSORFLOW_COMMIT = "67ba341c869e30ee4a89e040cd875d12b9bc666e"


(just commenting here for threading)

The change to mlir-hlo includes the stablehlo test suite (https://github.com/tensorflow/mlir-hlo/tree/master/stablehlo/stablehlo/testdata), which we might want to slice out if it increases git checkout time... this is a lot of files:

hanhanW

LGTM, thanks for the hard work!

As mentioned on discord, I think we can land it on Monday. It's Friday. I think we don't want to have a situation where we are dealing with breakages on Friday afternoon/weekend.

* llvm-project: e510d0bda0876c4baa3a270dca39b95da7ec6d9e * mlir-hlo: e86610442f58b889a57bf814d75c4b50c769c2a3 * tensorflow: 67ba341c869e30ee4a89e040cd875d12b9bc666e Cherry-picked from LLVM: ``` commit 80074d5fc0ab3f165865b15f5bf55ffac0917bcd (HEAD -> integrate-3-8-2023, fork/integrate-3-8-2023) Author: Matthias Springer <me@m-sp.org> Date: Fri Mar 10 11:25:15 2023 +0100 [mlir][NFC] reifyResultShapes: Add extra error checking This change adds a new helper function `mlir::reifyResultShapes` that calls the corresponding interface method and also checks the result produced by the implementation when running in debug mode. Bugs due to incorrect interface implementations can be difficult to debug. This helper function also reduces the amount of code needed at call sites: the cast to `ReifyRankedShapedTypeOpInterface` is done in the helper function. Differential Revision: https://reviews.llvm.org/D145777 commit 32b15f601de173e9511f470f7423108d3154e582 Author: Matthias Springer <me@m-sp.org> Date: Fri Mar 10 11:24:43 2023 +0100 [mlir][tensor/linalg] Fix bug in reifyResultShapes `reifyResultShapes` should return an IntegerAttr if and only if the corresponding dimension is static. Differential Revision: https://reviews.llvm.org/D145702 commit 894555cd6adf2e0faffe713373a266650b40bb4e Author: David Green <david.green@arm.com> Date: Wed Mar 8 12:48:21 2023 +0000 [AArch64] Fix load-insert-zero patterns with i8 and negative offsets. These should have been using the LDURBi instructions where the offset is negative, as reported from the reproducer in D144086. ``` Created a new commit on iree-mlir-hlo fork: https://github.com/iree-org/iree-mhlo-fork/commit/b14e9d9b06255e4476f5698e3bfc531dec793ded

matthias-springer reviewed Mar 9, 2023

View reviewed changes

Integrate llvm-project and bump dependencies.

0bccfce

* llvm-project: e510d0bda0876c4baa3a270dca39b95da7ec6d9e * mlir-hlo: e86610442f58b889a57bf814d75c4b50c769c2a3 * tensorflow: 67ba341c869e30ee4a89e040cd875d12b9bc666e

vmurali force-pushed the llvm-bump-3-8-2023 branch from b72407d to 0bccfce Compare March 9, 2023 19:46

Fixing the merge conflict

d32c51b

vmurali marked this pull request as ready for review March 9, 2023 19:51

vmurali requested review from hanhanW, MaheshRavishankar, ThomasRaoux, benvanik and stellaraccident as code owners March 9, 2023 19:51

clang format

3c8f3f3

vmurali enabled auto-merge (squash) March 9, 2023 19:54

hanhanW previously requested changes Mar 9, 2023

View reviewed changes

Fixed as per Hanhan's suggestions

f6717f4

Fixed as per Hanhan's suggestions

18abde3

vmurali added 6 commits March 9, 2023 23:21

fixed tensorflow directory issue

0d6c19c

buildifier

a1c3d5e

fix from mhlo

4985c71

fix from mhlo

13365ad

mhlo fix to bazel_skylib

c0d801f

more mhlo fixes

a8be8af

hanhanW disabled auto-merge March 10, 2023 05:27

cherry-pick instead of custom llvm-project patches

8300156

removed a test because of frontend issue

e9248cd

vmurali requested a review from rsuderman as a code owner March 10, 2023 20:27

vmurali mentioned this pull request Mar 10, 2023

Failed to import posenet_i8 model after 3-8-2023 integrate #12593

Closed

ScottTodd reviewed Mar 10, 2023

View reviewed changes

hanhanW approved these changes Mar 10, 2023

View reviewed changes

vmurali merged commit 2eb6450 into main Mar 13, 2023

vmurali deleted the llvm-bump-3-8-2023 branch March 13, 2023 05:12

jpienaar mentioned this pull request Mar 23, 2023

[NFC] Update from deprecated llvm::Optional #12734

Merged

ScottTodd mentioned this pull request Mar 29, 2023

Add build/test Windows CI #11009

Closed

jpienaar mentioned this pull request Apr 3, 2023

Fix XLA import path #12898

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Integrate llvm-project and bump dependencies. #12562

Integrate llvm-project and bump dependencies. #12562

vmurali commented Mar 8, 2023 •

edited

Loading

matthias-springer Mar 9, 2023

vmurali Mar 9, 2023

vmurali commented Mar 9, 2023

MaheshRavishankar commented Mar 9, 2023

hanhanW commented Mar 9, 2023

vmurali commented Mar 9, 2023

hanhanW commented Mar 10, 2023 •

edited

Loading

matthias-springer commented Mar 10, 2023

vmurali commented Mar 10, 2023

vmurali commented Mar 10, 2023

hanhanW commented Mar 10, 2023

vmurali commented Mar 10, 2023

ScottTodd Mar 10, 2023

hanhanW left a comment

		TENSORFLOW_COMMIT = "eece4dba1fc65f7977085581852b0d6e6d42f04e"
		TENSORFLOW_COMMIT = "67ba341c869e30ee4a89e040cd875d12b9bc666e"

Integrate llvm-project and bump dependencies. #12562

Integrate llvm-project and bump dependencies. #12562

Conversation

vmurali commented Mar 8, 2023 • edited Loading

matthias-springer Mar 9, 2023

Choose a reason for hiding this comment

vmurali Mar 9, 2023

Choose a reason for hiding this comment

vmurali commented Mar 9, 2023

MaheshRavishankar commented Mar 9, 2023

hanhanW commented Mar 9, 2023

vmurali commented Mar 9, 2023

hanhanW commented Mar 10, 2023 • edited Loading

matthias-springer commented Mar 10, 2023

vmurali commented Mar 10, 2023

vmurali commented Mar 10, 2023

hanhanW commented Mar 10, 2023

vmurali commented Mar 10, 2023

ScottTodd Mar 10, 2023

Choose a reason for hiding this comment

hanhanW left a comment

Choose a reason for hiding this comment

vmurali commented Mar 8, 2023 •

edited

Loading

hanhanW commented Mar 10, 2023 •

edited

Loading