TEST 'IMEX :: Integration/Dialect/XeGPU/gemm_4kx4kx4k_f16_f16_f16_w_8x32xf16_stores.mlir' FAILED
Opened this issue · 3 comments
Hi,
Building IMEX as in README and tests fail. Here's one of the error:
FAIL: IMEX :: Integration/Dialect/XeGPU/gemm_4kx4kx4k_f16_f16_f16_w_8x32xf16_stores.mlir (97 of 234)
******************** TEST 'IMEX :: Integration/Dialect/XeGPU/gemm_4kx4kx4k_f16_f16_f16_w_8x32xf16_stores.mlir' FAILED ********************
Exit Code: 2
Command Output (stderr):
--
RUN: at line 1: IMEX_ENABLE_LARGE_REG_FILE=1 /home/sasank/miniconda3/bin/python3 /home/sasank/code/llvm-project/build/bin/imex-runner.py --requires=l0-runtime -i /home/sasank/code/mlir-extensions/test/Integratio
n/Dialect/XeGPU/gemm_4kx4kx4k_f16_f16_f16_w_8x32xf16_stores.mlir --pass-pipeline-file=/home/sasank/code/mlir-extensions/test/Integration/Dialect/XeGPU/xegpu-to-llvm.pp --ru
nner imex-cpu-runner -e main --entry-point-result=void --shared-libs=/home/sasank/code/llvm-project/build/lib/libimex_runner_utils.so
,/home/sasank/code/llvm-project/build/lib/libmlir_runner_utils.so,/home/sasank/code/llvm-project/build/lib/libmlir_c_runner_utils.so,/home/sasank/code/llvm-project/build/lib/liblevel-zero-runtime.so --filecheck
+ IMEX_ENABLE_LARGE_REG_FILE=1
+ /home/sasank/miniconda3/bin/python3 /home/sasank/code/llvm-project/build/bin/imex-runner.py --requires=l0-runtime -i /home/sasank/code/mlir-extensions/test/Integration/Dialect/XeGPU/gemm_4kx4kx4k_f16_f16_f16_w
_8x32xf16_stores.mlir --pass-pipeline-file=/home/sasank/code/mlir-extensions/test/Integration/Dialect/XeGPU/xegpu-to-llvm.pp --runner imex-cpu-runner -e main --entry-point-result=void --shared-libs=/home/sasank/
code/llvm-project/build/lib/libimex_runner_utils.so,/home/sasank/code/llvm-project/build/lib/libmlir_runner_utils.so,/home/sasank/code/llvm-project/build/lib/libmlir_c_runner_utils.so,/home/sasank/code/llvm-proj
ect/build/lib/liblevel-zero-runtime.so --filecheck
error: LLVM ERROR: VISA builder API call failed: CisaBuilder->Compile( BC->isaDumpsEnabled() && BC->hasShaderDumper() ? BC->getShaderDumper().composeDumpPath("final.isaasm").c_str() : "", BC->emitVisaOnly())
PLEASE submit a bug report to https://github.com/llvm/llvm-project/issues/ and include the crash backtrace.
Stack dump:
0. Program arguments: /home/sasank/code/llvm-project/build/bin/imex-cpu-runner -e main --entry-point-result=void --shared-libs=/home/sasank/code/llvm-project/build/lib/libimex_runner_utils.so,/home/sasank/c
ode/llvm-project/build/lib/libmlir_runner_utils.so,/home/sasank/code/llvm-project/build/lib/libmlir_c_runner_utils.so,/home/sasank/code/llvm-project/build/lib/liblevel-zero-runtime.so
Stack dump without symbol names (ensure you have llvm-symbolizer in your PATH or set the environment var `LLVM_SYMBOLIZER_PATH` to point to it):
0 imex-cpu-runner 0x0000564daf2765a0
1 imex-cpu-runner 0x0000564daf27369f
2 imex-cpu-runner 0x0000564daf2737f5
3 libc.so.6 0x00007fb9361ef520
4 libc.so.6 0x00007fb9362439fc pthread_kill + 300
5 libc.so.6 0x00007fb9361ef476 raise + 22
6 libc.so.6 0x00007fb9361d57f3 abort + 211
7 libigc.so.1 0x00007fb9141347e7 llvm::report_fatal_error(llvm::Twine const&, bool) + 151
8 libigc.so.1 0x00007fb914134928
9 libigc.so.1 0x00007fb9146b5014
10 libigc.so.1 0x00007fb913c0d48f llvm::LLVMContext::diagnose(llvm::DiagnosticInfo const&) + 431
11 libigc.so.1 0x00007fb9146e5749
12 libigc.so.1 0x00007fb91475ba3f
13 libigc.so.1 0x00007fb9147715a9
14 libigc.so.1 0x00007fb913c2895c llvm::legacy::PassManagerImpl::run(llvm::Module&) + 812
15 libigc.so.1 0x00007fb914677aa0
16 libigc.so.1 0x00007fb9137ab16d
17 libigc.so.1 0x00007fb912e5a9c7
18 libigc.so.1 0x00007fb912ec67f5
19 libze_intel_gpu.so.1 0x00007fb91853cf27
20 libze_intel_gpu.so.1 0x00007fb91821fe80
21 libze_intel_gpu.so.1 0x00007fb918222433
22 libze_intel_gpu.so.1 0x00007fb918227662
23 libze_intel_gpu.so.1 0x00007fb918228d4e
24 libze_intel_gpu.so.1 0x00007fb91818798e
25 liblevel-zero-runtime.so 0x00007fb93618202a gpuModuleLoad + 266
26 liblevel-zero-runtime.so 0x00007fb93674b153 gpuModuleLoad + 6066739
27 liblevel-zero-runtime.so 0x00007fb93674b632 gpuModuleLoad + 6067986
28 liblevel-zero-runtime.so 0x00007fb93674bb51 gpuModuleLoad + 6069297
29 imex-cpu-runner 0x0000564daf80c884
30 imex-cpu-runner 0x0000564daf80cc7d
31 imex-cpu-runner 0x0000564daf80fb35
32 imex-cpu-runner 0x0000564daf1c9ecb
33 libc.so.6 0x00007fb9361d6d90
34 libc.so.6 0x00007fb9361d6e40 __libc_start_main + 128
35 imex-cpu-runner 0x0000564daf25e1e5
FileCheck error: '<stdin>' is empty.
FileCheck command line: /home/sasank/code/llvm-project/build/bin/FileCheck /home/sasank/code/mlir-extensions/test/Integration/Dialect/XeGPU/gemm_4kx4kx4k_f16_f16_f16_w_8x32xf16_stores.mlir
--
Other tests that failed for me:
- IMEX :: Integration/Dialect/XeGPU/large_stores_8x32xf16_w_constant_vector_shuffle.vc.mlir
- MEX :: Integration/Dialect/XeGPU/large_stores_8x32xf16_w_2d_vector_shuffle.vc.mlir
- IMEX :: Integration/Dialect/XeGPU/large_stores_8x32xf16_w_1d_vector_shuffle.vc.mlir
- IMEX :: Integration/Dialect/XeGPU/load_with_block_array_8_16_2.vc.mlir
Can you tell me how do I fix these?
These steps don't end up with that error for me:
activate imex conda env
Make sure llvm is updated and clean - git pull origin main
Make sure mlir-project is updated - git pull origin main
cd llvm-project
git checkout ``cat ../mlir-extensions/build_tools/llvm_version.txt``
git apply ../mlir-extensions/build_tools/patches/*
rm -rf build/*;cmake -G Ninja -B build -S llvm -DLLVM_ENABLE_PROJECTS=mlir -DLLVM_BUILD_EXAMPLES=ON -DLLVM_TARGETS_TO_BUILD="X86" -DCMAKE_BUILD_TYPE=Release -DLLVM_ENABLE_ASSERTIONS=ON -DLLVM_EXTERNAL_PROJECTS="Imex" -DLLVM_EXTERNAL_IMEX_SOURCE_DIR=../mlir-extensions/ -DIMEX_ENABLE_L0_RUNTIME=1 -DIMEX_ENABLE_SYCL_RUNTIME=1
cmake --build build --target check-imex | tee build/tests.txt
My command line to run the test
IMEX_ENABLE_LARGE_REG_FILE=1 python3.9 /mydir/llvm-project/build/bin/imex-runner.py --requires=l0-runtime -i /mydir/mlir-extensions/test/Integration/Dialect/XeGPU/gemm_4kx4kx4k_f16_f16_f16_w_8x32xf16_stores.mlir --pass-pipeline-file=/mydir/mlir-extensions/test/Integration/Dialect/XeGPU/xegpu-to-llvm.pp --runner imex-cpu-runner -e main --entry-point-result=void --shared-libs=/mydir/llvm-project/build/lib/libimex_runner_utils.so,/mydir/llvm-project/build/lib/libmlir_runner_utils.so,/mydir/llvm-project/build/lib/libmlir_c_runner_utils.so,/mydir/llvm-project/build/lib/liblevel-zero-runtime.so
I will retry again. For reference, I have installed drivers/compiler/libs etc from debian repos as described here: https://chsasank.com/intel-arc-gpu-driver-oneapi-installation.html
I have not used conda environment to install dependencies and didn't run pre-commit config as well.
The command you shared - is that how I run these benchmarks on Intel GPU?
The command you shared - is that how I run these benchmarks on Intel GPU?
Yes, I'm on the machine with IntelGPU. Note that gemm_4kx4kx4k_f16_f16_f16_w_8x32xf16_stores.mlir is PVC-specific