Disable numpy AVX512 when compiled with Intel classic

[DavidHuber-NOAA]

This sets the environmental variable NPY_DISABLE_CPU_FEATURES=AVX512F to disable AVX512 features when the py-numpy module is loaded. The AVX512 features produce undefined behavior when executing certain ndarray methods such as min().

Partly resolves JCSDA/spack-stack#1276 numpy/numpy#27840 matplotlib/matplotlib#28762

[climbfuji]

Can you add the URLS of the github issues/discussions as hyperlinks above the archs = line`, for example:

    # https://linktomatplotlibissue
    # https://linktonumpyissue
    archs = ...

Thanks!

[DavidHuber-NOAA]

Done, thanks!

[DavidHuber-NOAA]

Per a comment from @seberg, I will try disabling the AVX512 features at build time.

[DavidHuber-NOAA]

@seberg @climbfuji @AlexanderRichert-NOAA I tried adding this to the package.py with this patch. This produces the following lines at build time:

  Configuring npy_cpu_dispatch_config.h using configuration
  Message:
  CPU Optimization Options
    baseline:
      Requested : min
      Enabled   : SSE SSE2 SSE3
    dispatch:
      Requested : MAX -AVX512F -AVX512CD -AVX512_KNL -AVX512_KNM -AVX512_SKX -AVX512_CLX -AVX512_CNL -AVX512_ICL
      Enabled   : SSSE3 SSE41 POPCNT SSE42 AVX F16C FMA3 AVX2 AVX512F AVX512CD AVX512_SKX AVX512_CLX AVX512_CNL AVX512_ICL

Indicating that AVX512 instructions are still being used. I have also tried this all lower case, but it still produces AVX512 instructions. What command should I use in the spack recipe to disable these?

[seberg]

I'll have to look it up myself, let me pin @seiko2plus in case he sees it, he would know exactly immediately.

[r-devulap]

Indicating that AVX512 instructions are still being used. I have also tried this all lower case, but it still produces AVX512 instructions. What command should I use in the spack recipe to disable these?

Using NPY_DISABLE_CPU_FEATURES=AVX512F disables dynamic dispatch to these features, they still get build into NumPy library. But AVX512 code paths are never taken. Does that suffice in your case or do you not even want to build AVX512?

[seiko2plus]

Requested : MAX -AVX512F -AVX512CD -AVX512_KNL -AVX512_KNM -AVX512_SKX -AVX512_CLX -AVX512_CNL -AVX512_ICL

MAX also holds the value AVX512_SPR, if one feature is enabled it will leads to enable all implied features.

"setup-args": {
   "-Dcpu-dispatch": (
        settings.get("setup-args", {}).get("-Dcpu-dispatch", "") +
            "MAX -AVX512F -AVX512CD -AVX512_KNL -AVX512_KNM -AVX512_SKX -AVX512_CLX -AVX512_CNL -AVX512_ICL -AVX512_SPR"
       )
}

[DavidHuber-NOAA]

@r-devulap Yes, I think that would be sufficient (tested sufficiently at 26dd56f), though @seberg suggested that disabling the builds may be desirable. Since we would always be disabling the feature anyway, skipping the build may be desirable.

@seiko2plus Thank you!! Adding -AVX512_SP worked.