[Issue]: LLVM/MLIR build step isn't working

[brunomazzottiamd]

Problem Description

FlyDSL build step 1 (Build LLVM/MLIR) isn't working as expected. LLVM compilation fails:

root@smci355-ccs-aus-m02-13:/workspace/FlyDSL# bash scripts/build_llvm.sh -j64
Base directory: /workspace
LLVM Source:    /workspace/llvm-project
LLVM Build:     /workspace/llvm-project/build-flydsl
LLVM Install:   /workspace/llvm-project/mlir_install
LLVM Tarball:   /workspace/llvm-project/mlir_install.tgz
LLVM Commit:    ac5dc54d509169d387fcfd495d71853d81c46484
Cloning llvm-project...
Cloning into '/workspace/llvm-project'...
remote: Enumerating objects: 7514979, done.
remote: Counting objects: 100% (3230/3230), done.
remote: Compressing objects: 100% (2409/2409), done.
remote: Total 7514979 (delta 3103), reused 821 (delta 821), pack-reused 7511749 (from 2)
Receiving objects: 100% (7514979/7514979), 3.42 GiB | 2.26 MiB/s, done.
Resolving deltas: 100% (6165036/6165036), done.
Updating files: 100% (173287/173287), done.
Checking out llvm-project commit ac5dc54d509169d387fcfd495d71853d81c46484 (amd-staging)...
/workspace/llvm-project /workspace/FlyDSL
remote: Enumerating objects: 1, done.
remote: Counting objects: 100% (1/1), done.
remote: Total 1 (delta 0), reused 0 (delta 0), pack-reused 0 (from 0)
Unpacking objects: 100% (1/1), 876 bytes | 876.00 KiB/s, done.
From https://github.com/ROCm/llvm-project
* branch                      amd-staging -> FETCH_HEAD
   9da81213f232..8895e6bc7506  amd-staging -> origin/amd-staging
Updating files: 100% (17775/17775), done.7775)
Note: switching to 'ac5dc54d509169d387fcfd495d71853d81c46484'.

You are in 'detached HEAD' state. You can look around, make experimental
changes and commit them, and you can discard any commits you make in this
state without impacting any branches by switching back to a branch.

If you want to create a new branch to retain commits you create, you may
do so (now or later) by using -c with the switch command. Example:

  git switch -c <new-branch-name>

Or undo this operation with:

  git switch -

Turn off this advice by setting config variable advice.detachedHead to false

HEAD is now at ac5dc54d5091 [SLP][NFC]Add disjoint or forms of tests, which actually should lead to scalar ident
ity/bswap, NFC
/workspace/FlyDSL
Configuring LLVM...
Installing Python dependencies...
Requirement already satisfied: nanobind in /opt/venv/lib/python3.12/site-packages (2.12.0)
Requirement already satisfied: numpy in /opt/venv/lib/python3.12/site-packages (2.4.2)
Requirement already satisfied: pybind11 in /opt/venv/lib/python3.12/site-packages (3.0.2)
Using Ninja generator.
-- The C compiler identification is GNU 13.3.0
-- The CXX compiler identification is GNU 13.3.0
.
.
[CMake output omitted for brevity]
.
.
/workspace/llvm-project/mlir/include/mlir/Bindings/Python/Nanobind.h:22:32: note: did you mean _-Wno-switch-default_?
cc1plus: note: unrecognized command-line option _-Wno-zero-length-array_ may have been intended to silence earlier diagnostics
cc1plus: note: unrecognized command-line option _-Wno-nested-anon-types_ may have been intended to silence earlier diagnostics
cc1plus: note: unrecognized command-line option _-Wno-deprecated-literal-operator_ may have been intended to silence earlier diagnostics
cc1plus: note: unrecognized command-line option _-Wno-covered-switch-default_ may have been intended to silence earlier diagnostics
cc1plus: note: unrecognized command-line option _-Wno-c++98-compat-extra-semi_ may have been intended to silence earlier diagnostics
[7044/7569] Building CXX object tools/llvm-dlang-deman...vm-dlang-demangle-fuzzer.dir/DummyDemanglerFuzzer.cpp.oFAILED: [code=1] runtimes/runtimes-stamps/runtimes-build /workspace/llvm-project/build-flydsl/runtimes/runtimes-stamps/runtimes-build
cd /workspace/llvm-project/build-flydsl/runtimes/runtimes-bins && /opt/venv/lib/python3.12/site-packages/cmake/data/bin/cmake --build .
[7107/7569] Building CXX object tools/clang/tools/clan...deps/CMakeFiles/clang-scan-deps.dir/ClangScanDeps.cpp.o[7108/7569] Building CXX object tools/clang/lib/ASTMat...akeFiles/obj.clangDynamicASTMatchers.dir/Registry.cpp.o
ninja: build stopped: subcommand failed.

Operating System

Ubuntu 24.04.4 LTS (Noble Numbat)

CPU

AMD EPYC 9575F 64-Core Processor

GPU

AMD Instinct MI355X

ROCm Version

ROCm 7.2.0

Steps to Reproduce

Run bash scripts/build_llvm.sh -j64, as documented in README.md file.

Additional Information

FlyDSL commit: 99ccb1e

[coderfeli]

fixed in latest main. @brunomazzottiamd could you try again with latest codes?

[brunomazzottiamd]

I'm at commit 36dd3fe and I'm still getting the same error when running bash scripts/build_llvm.sh -j64:

.
.
.
/workspace/llvm-project/mlir/include/mlir/Bindings/Python/Nanobind.h:22:32: note: did you mean _-Wno-switch-default_?
cc1plus: note: unrecognized command-line option _-Wno-zero-length-array_ may have been intended to silence earlier diagnostics
cc1plus: note: unrecognized command-line option _-Wno-nested-anon-types_ may have been intended to silence earlier diagnostics
cc1plus: note: unrecognized command-line option _-Wno-deprecated-literal-operator_ may have been intended to silence earlier diagnostics
cc1plus: note: unrecognized command-line option _-Wno-covered-switch-default_ may have been intended to silence earlier diagnostics
cc1plus: note: unrecognized command-line option _-Wno-c++98-compat-extra-semi_ may have been intended to silence earlier diagnostics
[7103/7569] Building CXX object tools/clang/lib/ASTMat...akeFiles/obj.clangDynamicASTMatchers.dir/Registry.cpp.o
ninja: build stopped: subcommand failed.

I was missing the apt-get update && apt-get install -y cmake build-essential part in my container. I did it but I'm still getting the same LLVM/MLIR compilation error: ninja: build stopped: subcommand failed..

I'm going to reproduce the exact same steps of flydsl.yaml, as in CI.

[brunomazzottiamd]

@coderfeli, I followed the CI steps flydsl.yaml and everything worked:

# Start Docker container:
docker run -dt --network=host --user root --device=/dev/kfd --device /dev/dri \
    --ipc=host --group-add video --shm-size 16g --cap-add=SYS_PTRACE \
    --security-opt seccomp=unconfined -w /flydsl-test --name flydsl_test \
    rocm/pytorch:rocm7.1_ubuntu24.04_py3.12_pytorch_release_2.9.1
docker exec -it flydsl_test bash

# Clone FlyDSL:
git clone https://github.com/ROCm/FlyDSL.git .

# Install FlyDSL dependencies:
apt-get update && apt-get install -y cmake build-essential
python3 -m pip install -U pip setuptools wheel
python3 -m pip install 'ninja>=1.11.1'
python3 -m pip install -U 'hypothesis>=6.82.0'

# Install FlyDSL:
bash scripts/build_llvm.sh
# bash scripts/build_llvm.sh -j64 also works
export MLIR_PATH=/llvm-project/mlir_install
python3 -m pip install -e .

# Install AITER:
git clone --recursive https://github.com/ROCm/aiter.git /tmp/aiter
cd /tmp/aiter
python3 setup.py develop

# Run FlyDSL tests:
cd /flydsl-test
RUN_TESTS_FULL=1 bash scripts/run_tests.sh

I was using rocm/pytorch:latest before, with ROCm 7.2. CI uses rocm/pytorch:rocm7.1_ubuntu24.04_py3.12_pytorch_release_2.9.1, with ROCm 7.1. I think I'll stick to this Docker image to learn FlyDSL. FYI: @rahulbatra85

[rahulbatra85]

@brunomazzottiamd ok this works. Thanks!