Building from source with MKL-DNN fails on Fedora37

[comrade62]

Description

Building from source with MKL-DNN fails on Fedora37 using (at least) these branches: master and v2.0.0.beta0.rc1 when using the following steps in the offical mxnet doc pages:

• https://mxnet.incubator.apache.org/versions/1.9.1/api/python/docs/tutorials/performance/backend/mkldnn/mkldnn_readme.html#6

Error Message

In file included from /usr/lib/gcc/x86_64-redhat-linux/12/include/immintrin.h:49,
from /repos/mxnet/3rdparty/intgemm/intgemm/types.h:3,
from /repos/mxnet/3rdparty/intgemm/intgemm/intgemm.h:45,
from /repos/mxnet/3rdparty/intgemm/intgemm/intgemm.cc:1:
In function ‘__m256d _mm512_extractf64x4_pd(__m512d, int)’,
inlined from ‘float intgemm::AddFloat32(__m512)’ at /home/owner/repos/mxnet/3rdparty/intgemm/intgemm/stats.h:51:34,
inlined from ‘intgemm::MeanStd intgemm::avx512bw::VectorMeanStd(const float*, const float*, bool)’ at /home/owner/repos/mxnet/3rdparty/intgemm/intgemm/stats.inl:86:33:
/usr/lib/gcc/x86_64-redhat-linux/12/include/avx512fintrin.h:5946:10: error: ‘__Y’ is used uninitialized [-Werror=uninitialized]
5946 | return (__m256d) __builtin_ia32_extractf64x4_mask ((__v8df) __A,
| ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
5947 | __imm,
| ~~~~~~
5948 | (__v4df)
| ~~~~~~~~
5949 | _mm256_undefined_pd (),
| ~~~~~~~~~~~~~~~~~~~~~~~
5950 | (__mmask8) -1);
| ~~~~~~~~~~~~~~

../mxnet/3rdparty/onednn/src/cpu/x64/shuffle/jit_uni_shuffle_kernel.cpp: In member function ‘void dnnl::impl::cpu::x64::jit_uni_shuffle_kernel_t::shuffle_blocked_format() [with dnnl::impl::cpu::x64::cpu_isa_t isa = dnnl::impl::cpu::x64::sse41]’:
/repos/mxnet/3rdparty/onednn/src/cpu/x64/shuffle/jit_uni_shuffle_kernel.cpp:297:15: note: object ‘vmm_tmp’ of size 32
297 | const Vmm vmm_tmp[4] = {Vmm(5), Vmm(6), Vmm(7), Vmm(8)};
| ^~~~~~~
[ 46%] Built target dnnl_cpu_x64
make: *** [Makefile:146: all] Error 2

To Reproduce

Build from source.

Steps to reproduce

1. Download the source (git clone --recursive https://github.com/apache/mxnet.git) and switch to dir mxnet
2. Use dev branch (master) or v2.0.0.beta0.rc1
3. mkdir build && cd build
4. cmake -DUSE_CUDA=OFF -DUSE_MKL_IF_AVAILABLE=ON -DUSE_MKLDNN=ON -DUSE_OPENMP=ON -DUSE_OPENCV=ON
5. make -j $(nproc)

What have you tried to solve it?

Nothing, I have no idea where to start.

Environment

Fedora 37
Kernel: Linux fedoravm 6.1.15-200.fc37.x86_64 #1 SMP PREEMPT_DYNAMIC Fri Mar 3 17:29:44 UTC 2023 x86_64 x86_64 x86_64 GNU/Linux

Environment Information

# Paste the diagnose.py command output here
----------Python Info----------
Version      : 3.11.2
Compiler     : GCC 12.2.1 20221121 (Red Hat 12.2.1-4)
Build        : ('main', 'Feb  8 2023 00:00:00')
Arch         : ('64bit', 'ELF')
------------Pip Info-----------
Version      : 23.0.1
Directory    : /home/owner/.local/lib/python3.11/site-packages/pip
----------MXNet Info-----------
No MXNet installed.
----------System Info----------
Platform     : Linux-6.1.15-200.fc37.x86_64-x86_64-with-glibc2.36
system       : Linux
node         : fedoravm
release      : 6.1.15-200.fc37.x86_64
version      : #1 SMP PREEMPT_DYNAMIC Fri Mar  3 17:29:44 UTC 2023
----------Hardware Info----------
machine      : x86_64
processor    : x86_64
Architecture:            x86_64
  CPU op-mode(s):        32-bit, 64-bit
  Address sizes:         46 bits physical, 48 bits virtual
  Byte Order:            Little Endian
CPU(s):                  15
  On-line CPU(s) list:   0-14
Vendor ID:               GenuineIntel
  Model name:            Genuine Intel(R) CPU 0000%@
    CPU family:          6
    Model:               85
    Thread(s) per core:  1
    Core(s) per socket:  15
    Socket(s):           1
    Stepping:            5
    BogoMIPS:            4199.95
    Flags:               fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ss ht syscall nx pdpe1gb rdtscp lm c
                         onstant_tsc arch_perfmon rep_good nopl xtopology nonstop_tsc cpuid pni pclmulqdq vmx ssse3 fma cx16 pdcm pcid sse4_1 sse4_2 x2apic mov
                         be popcnt tsc_deadline_timer aes xsave avx f16c rdrand lahf_lm abm 3dnowprefetch cpuid_fault invpcid_single ssbd ibrs ibpb stibp tpr_s
                         hadow vnmi ept vpid ept_ad fsgsbase tsc_adjust bmi1 hle avx2 smep bmi2 erms invpcid rtm mpx avx512f avx512dq rdseed adx smap clflushop
                         t clwb avx512cd avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves arat umip pku ospke md_clear arch_capabilities
Virtualization features: 
  Virtualization:        VT-x
Caches (sum of all):     
  L1d:                   256 KiB (8 instances)
  L1i:                   256 KiB (8 instances)
  L2:                    8 MiB (8 instances)
  L3:                    22 MiB (1 instance)
NUMA:                    
  NUMA node(s):          1
  NUMA node0 CPU(s):     0-14
Vulnerabilities:         
  Itlb multihit:         Not affected
  L1tf:                  Not affected
  Mds:                   Mitigation; Clear CPU buffers; SMT disabled
  Meltdown:              Not affected
  Mmio stale data:       Vulnerable: Clear CPU buffers attempted, no microcode; SMT disabled
  Retbleed:              Mitigation; IBRS
  Spec store bypass:     Mitigation; Speculative Store Bypass disabled via prctl
  Spectre v1:            Mitigation; usercopy/swapgs barriers and __user pointer sanitization
  Spectre v2:            Mitigation; IBRS, IBPB conditional, RSB filling, PBRSB-eIBRS Not affected
  Srbds:                 Not affected
  Tsx async abort:       Mitigation; Clear CPU buffers; SMT disabled

</details>

[github-actions]

Welcome to Apache MXNet (incubating)! We are on a mission to democratize AI, and we are glad that you are contributing to it by opening this issue.
Please make sure to include all the relevant context, and one of the @apache/mxnet-committers will be here shortly.
If you are interested in contributing to our project, let us know! Also, be sure to check out our guide on contributing to MXNet and our development guides wiki.

[comrade62]

If this can be fixed by downgrading gcc, please advise to which version I should downgrade to as this page (https://mxnet.apache.org/get_started/build_from_source) suggests that all versions above gcc7 should work when building the dev or 2.x versions.

[vpirogov]

@comrade62, looks like GCC is barking at it's own avx512fintrin.h header included from 3rdparty/intgemm. This module is built with -Werror unlike the rest of MXNet. Try disabling it or adding -Wno-error=unused-variable like it's done for MXNet iteslf.

A note issued from oneDNN code is benign and should not affect build.

[comrade62]

@vpirogov thank you for your response, I mistakenly included just the one error where there was in fact two but once I added to the compile options "-Wno-error=maybe-uninitialized" it fixed the above error and then also added "-Wno-error=uninitialized" (both) to the file "3rdparty/intgemm/CMakeLists.txt" the build completed successfully, thank you again for your assistance it is very much appreciated.