Create a neura canonicalize pass #86

ShangkunLi · 2025-07-20T15:42:30Z

In this pr:

promote live-in values in each bb to the block arguments
promote the func arguments to neura.constants so that we can predicate and map these func arguments

ShangkunLi · 2025-07-20T18:18:19Z

Hi @tancheng, it seems we don't have a unified version of clang right now, which is why this test failed. Do we need to build the latest version of clang in the workflow? Or just sync the clang version to 14 as mentioned in #19.

tancheng · 2025-07-20T18:38:33Z

Hi @tancheng, it seems we don't have a unified version of clang right now, which is why this test failed. Do we need to build the latest version of clang in the workflow? Or just sync the clang version to 14 as mentioned in #19.

Error msg:

/home/runner/work/dataflow/dataflow/test/c2llvm2mlir/Output/test.mlir.script: line 1: /home/runner/work/dataflow/dataflow/llvm-project/build/./bin/clang++: No such file or directory

Why these tests succeed on current master? There should be clang++ installed in github action, right?

ShangkunLi · 2025-07-20T18:51:18Z

Hi @tancheng, it seems we don't have a unified version of clang right now, which is why this test failed. Do we need to build the latest version of clang in the workflow? Or just sync the clang version to 14 as mentioned in #19.

Error msg:
/home/runner/work/dataflow/dataflow/test/c2llvm2mlir/Output/test.mlir.script: line 1: /home/runner/work/dataflow/dataflow/llvm-project/build/./bin/clang++: No such file or directory
Why these tests succeed on current master? There should be clang++ installed in github action, right?

Because I changed the test/lit.cfg.in file to make tests use the clang built directly from source code, and I added the built scripts to build the clang in the latest commit.

In previous tests, clang++ uses the clang installed in github action.

I prefer to use the clang built from the same source code as the mlir to keep consistency, WDYT? (Let's see how many hours it would take to build both clang & mlir and then make a decision)

include/NeuraDialect/NeuraPasses.h

lib/Conversion/ArithToNeura/ArithToNeuraPass.cpp

lib/NeuraDialect/Transforms/TransformCtrlToDataFlowPass.cpp

test/neura/ctrl/branch_for.mlir

lib/NeuraDialect/Transforms/CanonicalizeLiveInPass.cpp

tancheng · 2025-07-21T06:01:42Z

Hi @tancheng, it seems we don't have a unified version of clang right now, which is why this test failed. Do we need to build the latest version of clang in the workflow? Or just sync the clang version to 14 as mentioned in #19.

Error msg:
/home/runner/work/dataflow/dataflow/test/c2llvm2mlir/Output/test.mlir.script: line 1: /home/runner/work/dataflow/dataflow/llvm-project/build/./bin/clang++: No such file or directory
Why these tests succeed on current master? There should be clang++ installed in github action, right?
Because I changed the test/lit.cfg.in file to make tests use the clang built directly from source code, and I added the built scripts to build the clang in the latest commit.

In previous tests, clang++ uses the clang installed in github action.

I prefer to use the clang built from the same source code as the mlir to keep consistency, WDYT? (Let's see how many hours it would take to build both clang & mlir and then make a decision)

3h looks fine to me.. Then what should I do on my local docker? Do I need to install a specific llvm/clang version to build mlir? Or do i need that specific llvm/clang to run our test?

ShangkunLi · 2025-07-21T06:06:41Z

Hi @tancheng, it seems we don't have a unified version of clang right now, which is why this test failed. Do we need to build the latest version of clang in the workflow? Or just sync the clang version to 14 as mentioned in #19.

Error msg:
/home/runner/work/dataflow/dataflow/test/c2llvm2mlir/Output/test.mlir.script: line 1: /home/runner/work/dataflow/dataflow/llvm-project/build/./bin/clang++: No such file or directory
Why these tests succeed on current master? There should be clang++ installed in github action, right?
Because I changed the test/lit.cfg.in file to make tests use the clang built directly from source code, and I added the built scripts to build the clang in the latest commit.
In previous tests, clang++ uses the clang installed in github action.
I prefer to use the clang built from the same source code as the mlir to keep consistency, WDYT? (Let's see how many hours it would take to build both clang & mlir and then make a decision)
3h looks fine to me.. Then what should I do on my local docker? Do I need to install a specific llvm/clang version to build mlir? Or do i need that specific llvm/clang to run our test?

Use this cmake cripts to build the clang. And then use the clang++ in build/bin/clang++ instead.

git clone https://github.com/llvm/llvm-project.git
cd llvm-project
git checkout 6146a88
mkdir build && cd build
cmake -G Ninja ../llvm \
-DLLVM_ENABLE_PROJECTS="mlir;clang" \
-DLLVM_BUILD_EXAMPLES=OFF \
-DLLVM_TARGETS_TO_BUILD="Native" \
-DCMAKE_BUILD_TYPE=Release \
-DLLVM_ENABLE_ASSERTIONS=ON \
-DCMAKE_C_COMPILER=clang \
-DCMAKE_CXX_COMPILER=clang++ \
-DCMAKE_CXX_STANDARD=17 \
-DCMAKE_CXX_FLAGS="-std=c++17 -frtti" \
-DLLVM_ENABLE_LLD=ON \
-DMLIR_INSTALL_AGGREGATE_OBJECTS=ON \
-DLLVM_ENABLE_RTTI=ON \
-DCMAKE_C_COMPILER_LAUNCHER=ccache \
-DCMAKE_CXX_COMPILER_LAUNCHER=ccache

cmake --build . --target check-mlir
cmake --build . --target check-clang

tancheng · 2025-07-21T06:08:01Z

Hi @tancheng, it seems we don't have a unified version of clang right now, which is why this test failed. Do we need to build the latest version of clang in the workflow? Or just sync the clang version to 14 as mentioned in #19.

Error msg:
/home/runner/work/dataflow/dataflow/test/c2llvm2mlir/Output/test.mlir.script: line 1: /home/runner/work/dataflow/dataflow/llvm-project/build/./bin/clang++: No such file or directory
Why these tests succeed on current master? There should be clang++ installed in github action, right?
Because I changed the test/lit.cfg.in file to make tests use the clang built directly from source code, and I added the built scripts to build the clang in the latest commit.
In previous tests, clang++ uses the clang installed in github action.
I prefer to use the clang built from the same source code as the mlir to keep consistency, WDYT? (Let's see how many hours it would take to build both clang & mlir and then make a decision)
3h looks fine to me.. Then what should I do on my local docker? Do I need to install a specific llvm/clang version to build mlir? Or do i need that specific llvm/clang to run our test?
Use this cmake cripts to build the clang. And then use the clang++ in build/bin/clang++ instead.
git clone https://github.com/llvm/llvm-project.git
cd llvm-project
git checkout 6146a88
mkdir build && cd build
cmake -G Ninja ../llvm \
-DLLVM_ENABLE_PROJECTS="mlir;clang" \
-DLLVM_BUILD_EXAMPLES=OFF \
-DLLVM_TARGETS_TO_BUILD="Native" \
-DCMAKE_BUILD_TYPE=Release \
-DLLVM_ENABLE_ASSERTIONS=ON \
-DCMAKE_C_COMPILER=clang \
-DCMAKE_CXX_COMPILER=clang++ \
-DCMAKE_CXX_STANDARD=17 \
-DCMAKE_CXX_FLAGS="-std=c++17 -frtti" \
-DLLVM_ENABLE_LLD=ON \
-DMLIR_INSTALL_AGGREGATE_OBJECTS=ON \
-DLLVM_ENABLE_RTTI=ON \
-DCMAKE_C_COMPILER_LAUNCHER=ccache \
-DCMAKE_CXX_COMPILER_LAUNCHER=ccache

cmake --build . --target check-mlir
cmake --build . --target check-clang

Plz update README.md then.

lib/NeuraDialect/Transforms/CanonicalizeLiveInPass.cpp

ShangkunLi added 3 commits July 20, 2025 14:31

promote live-in values to block args

5b8f212

promote func args to constants

fb75076

[fix] add more changed files

bea4965

[fix] support clang build

7a971aa

tancheng reviewed Jul 20, 2025

View reviewed changes

include/NeuraDialect/NeuraPasses.h Outdated Show resolved Hide resolved

lib/Conversion/ArithToNeura/ArithToNeuraPass.cpp Show resolved Hide resolved

lib/NeuraDialect/Transforms/TransformCtrlToDataFlowPass.cpp Outdated Show resolved Hide resolved

tancheng reviewed Jul 20, 2025

View reviewed changes

test/neura/ctrl/branch_for.mlir Outdated Show resolved Hide resolved

test/neura/ctrl/branch_for.mlir Show resolved Hide resolved

rename the pass & add test

b38cbd3

ShangkunLi marked this pull request as ready for review July 21, 2025 05:19

tancheng reviewed Jul 21, 2025

View reviewed changes

ShangkunLi added 2 commits July 21, 2025 14:59

change README and add sort the bbs

ccaa737

change the test scripts

c728fea

tancheng reviewed Jul 21, 2025

View reviewed changes

lib/NeuraDialect/Transforms/CanonicalizeLiveInPass.cpp Outdated Show resolved Hide resolved

remove block sort function

e143058

tancheng approved these changes Jul 21, 2025

View reviewed changes

tancheng assigned ShangkunLi Jul 21, 2025

tancheng added the new feature New feature or request label Jul 21, 2025

ShangkunLi merged commit 8ff1cec into coredac:main Jul 22, 2025
1 check passed

This was referenced Jul 22, 2025

[P1] Create a canonicalization pass #83

Closed

[P1] Fix leverage-predicated-value pass for c++ lowering to neura #84

Closed

Create a neura canonicalize pass #86

Create a neura canonicalize pass #86

Uh oh!

Conversation

ShangkunLi commented Jul 20, 2025

Uh oh!

ShangkunLi commented Jul 20, 2025

Uh oh!

tancheng commented Jul 20, 2025

Uh oh!

ShangkunLi commented Jul 20, 2025

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

tancheng commented Jul 21, 2025

Uh oh!

ShangkunLi commented Jul 21, 2025

Uh oh!

tancheng commented Jul 21, 2025

Uh oh!

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants