illegal hardware instruction (core dumped)

linuxlonelyeagle commented 2 years ago

Describe the bug

when i input ./image-processing-benchmark ../../benchmarks/ImageProcessing/Images/YuTu.png laplacianKernelAlign in the shell. I encountered this situation that illegal hardware instruction (core dumped)

To Reproduce

Steps to reproduce the behavior.
cd buddy-benchmark
cd build  
cd bin
./image-processing-benchmark ../../benchmarks/ImageProcessing/Images/YuTu.png laplacianKernelAlign
Running ./image-processing-benchmark
Run on (16 X 2900 MHz CPU s)
CPU Caches:
L1 Data 32 KiB (x8)
L1 Instruction 32 KiB (x8)
L2 Unified 512 KiB (x8)
L3 Unified 4096 KiB (x2)
Load Average: 1.13, 0.91, 0.74
WARNING  CPU scaling is enabled, the benchmark real time measurements may be noisy and will incur extra overhead.
WARNING  Library was built as DEBUG. Timings may be affected.
-------------------------------------------------------------
Benchmark                   Time             CPU   Iterations
-------------------------------------------------------------
OpenCV_Filter2D/1     3471609 ns      3460337 ns          201
Eigen_Convolve2D/1   76930880 ns     76863287 ns            9
MLIR_Conv2D/1        12705419 ns     12695988 ns           55
[1]    4396 illegal hardware instruction (core dumped)  ./image-processing-benchmark ../../benchmarks/ImageProcessing/Images/YuTu.png

Desktop :

OS: arch linux

Version [5.17.4]

Additional context

My computer cpu is amd4800 $lscpu

Model name：AMD Ryzen 7 4800H with Radeon Graphics
Flags：              fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ht 
                     syscall nx mmxext fxsr_opt pdpe1gb rdtscp lm constant_tsc rep_good nopl nonstop_tsc cpuid extd_apicid
                      aperfmperf rapl pni pclmulqdq monitor ssse3 fma cx16 sse4_1 sse4_2 movbe popcnt aes xsave avx f16c r
                     drand lahf_lm cmp_legacy svm extapic cr8_legacy abm sse4a misalignsse 3dnowprefetch osvw ibs skinit w
                     dt tce topoext perfctr_core perfctr_nb bpext perfctr_llc mwaitx cpb cat_l3 cdp_l3 hw_pstate ssbd mba 
                     ibrs ibpb stibp vmmcall fsgsbase bmi1 avx2 smep bmi2 cqm rdt_a rdseed adx smap clflushopt clwb sha_ni
                      xsaveopt xsavec xgetbv1 xsaves cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local clzero irperf xsave
                     erptr rdpru wbnoinvd cppc arat npt lbrv svm_lock nrip_save tsc_scale vmcb_clean flushbyasid decodeass
                     ists pausefilter pfthreshold avic v_vmsave_vmload vgif v_spec_ctrl umip rdpid overflow_recov succor s
                     mca sme sev sev_es

zhanghb97 commented 2 years ago

From your description, the issue is from the DIP Corr2D case.

Please try to run the DIP Corr2D example (“Lowering DIP Dialect” section in the buddy-mlir README) and see if you can reproduce the issue again.

Note: your device seems not support AVX512, please use right configuration (avx2) for the ISA vector extension.

zhanghb97 commented 2 years ago

Ping @meshtag for this DIP related issue.

meshtag commented 2 years ago

@linuxlonelyeagle .mlir files relevant to IP benchmark use buddy-opt to lower appropriate IP related MLIR Ops for generating the required .o file. As mentioned by @zhanghb97 , please ensure that you are able to build the DIP example by following instructions mentioned in this readme file.

linuxlonelyeagle commented 2 years ago

I have solved the problem, here is what i do.When i need use the cmake command.I make the CMake Options what is -DBUDDY_OPT_ATTR equal to avx2,because my computer don‘t support avx512f. For example,when i build the Image Processing Benchmark，my input is below:

$ cd buddy-benchmark
$ mkdir build && cd build
$ cmake -G Ninja .. \
    -DIMAGE_PROCESSING_BENCHMARKS=ON \
    -DOpenCV_DIR=/PATH/TO/OPENCV/BUILD/ \
    -DEIGEN_DIR=/PATH/TO/EIGEN/SOURCE/CODE \
    -DBUDDY_OPT_BUILD_DIR=/PATH/TO/BUDDY-MLIR/BUILD/
    -DBUDDY_OPT_ATTR=avx2     
$ ninja image-processing-benchmark

So we should point out -DBUDDY_OPT_ATTR=avx2

meshtag commented 2 years ago

Glad you solved the issue! I think the buddy-mlir doc mentions it clearly that ISA vector extension should be specified by the user as per their machine. Quoting the following line from budd-mlir's readme file :

You should specify the strip mining size (e.g. 256) and ISA vector extension (e.g. avx512f).

Feel free to create a PR if you think we should mention this more clearly in either buddy-mlir or buddy-benchmark's readme file(s).

PS : Please close the issue if you think your concerns are addressed.

buddy-compiler / buddy-benchmark