[Mlir-commits] [mlir] [MLIR][Linalg] Scalable Vectorization of Reduction (PR #97788)

[Zhaoshi Zheng]

zhaoshiz wrote:

> Sorry for not getting back to you earlier. Could you clarify - are you asking how to run this on x86_64 or AArch64? The former will require cross-compiling. The latter, should just work ™️ 😅 If it didn't, could you share more details?

We should be able to run the integration tests on SVE/SME in both ways: 1. qemu-aarch64 on x86_64; 2 native aarch64-linux
For 2, I'm assuming they just work with `-shared-libs=%libmlir_runner_utils,...`.
For 1, I can build llvm/mlir on x86_64 with addtional cmake variables specifying pre-built aarch64 mlir-cpu-runner and shared libs:
```
  -DARM_EMULATOR_EXECUTABLE="<path_to_qemu_bin>/qemu-aarch64" \
  -DARM_EMULATOR_OPTIONS="-L /usr/aarch64-linux-gnu" \
  -DARM_EMULATOR_MLIR_CPU_RUNNER_EXECUTABLE="<path_to_llvm_arm64_build>/bin/mlir-cpu-runner-arm64" \
  -DARM_EMULATOR_UTILS_LIB_DIR="<path_to_llvm_arm64_build>/lib"
```
and write integration tests with `-shared-libs=%mlir_native_utils_lib_dir/libmlir_runner_utils%shlibext,...` which expands to: `-shared-libs=config.arm_emulator_utils_lib_dir/libmlir_runner_utils.so,...` where "config.arm_emulator_utils_lib_dir" is set by "-DARM_EMULATOR_UTILS_LIB_DIR".

My question is will `-shared-libs=%mlir_native_utils_lib_dir/libmlir_runner_utils%shlibext,...` work in native aarch64-linux environment?



https://github.com/llvm/llvm-project/pull/97788