openblas: fix compilation with mips x4k series CPU

[cotequeiroz]

Maintainer: @commodo
Compile tested: ath79/TP-Link Archer C6, built with CONFIG_CPU_TYPE="74kc", on master.
Run tested: same arch, using a small test program from https://www.hpc.ntnu.no/idun/software-on-idun/openblas/

/*  blas_test.c : Example using BLAS matrix multiply code sgemm */

#include <stdio.h>
#include <cblas.h>

int main(void)
{
  const int lda=3, ldb=3, ldc=3;
  int m, n, k;
  float alpha, beta;

  float a[] = { 0.11, 0.21, 0.12,
                0.15, 0.13, 0.17,
                0.22, 0.13, 0.23 };

  float b[] = { 1011, 1021, 1018,
                1031, 1012, 1021,
                1022, 1032, 1015 };

  float c[] = { 0.00, 0.00, 0,00,
                0.00, 0.00, 0,00,
                0.00, 0.00, 0,00 };

  m = 3; n = 3; k = 3;

  alpha = 1.0; beta = 0.0;

  cblas_sgemm (CblasColMajor, CblasNoTrans, CblasNoTrans,
                m, n, k, alpha, a, lda, b, ldb, beta, c, ldc);

  printf ("[ %g, %g, %g\n", c[0], c[1], c[2]);
  printf ("  %g, %g, %g\n", c[3], c[4], c[5]);
  printf ("  %g, %g, %g ]\n", c[6], c[7], c[8]);

  return 0;
}

Description:
Anything later than MIPS 24k can run MIPS 24k code. Set OPENBLAS_TARGET
to MIPS24K in those cases.

Signed-off-by: Eneas U de Queiroz cotequeiroz@gmail.com

It should fix current build errors for mipsel_74kc, mipsel_24kc_24kf, and of course, if you have a custom build on mips_74kc, like I just did 😜

[neheb]

@cotequeiroz can you look into this failure as well? https://downloads.openwrt.org/snapshots/faillogs/mipsel_24kc_24kf/packages/openblas/compile.txt

[cotequeiroz]

This should take care of it as well, although I haven't actually tried it. That's why I used 24k% in the filter.
When you select the pistachio target, you get CONFIG_CPU_TYPE="24kc+24kf", and that's what we are using to select OPENBLAS_TARGET.

[cotequeiroz]

Would anyone mind if I add a commit here to avoid the package setting its own make -j flag? I hope not. I've just noticed this on top of the fail log:

make[5]: warning: -j32 forced in submake: resetting jobserver mode.

[commodo]

Would anyone mind if I add a commit here to avoid the package setting its own make -j flag? I hope not. I've just noticed this on top of the fail log:

make[5]: warning: -j32 forced in submake: resetting jobserver mode.

fine from me;

notifying @bhack and @brada4 ;
it may be of interest for them

[brada4]

you can use MAKE_NB_JOBS=-1 to respect jobserver setting from outside. Otherwise OpenBLAS would try all visible cores, no matter -j .

[brada4]

REF:Makefile.rule

[cotequeiroz]

you can use MAKE_NB_JOBS=-1 to respect jobserver setting from outside. Otherwise OpenBLAS would try all visible cores, no matter -j .

Yes, that's exactly what was done here.

packages/libs/openblas/Makefile

Line 84 in 0dd482f

MAKE_NB_JOBS=-1 \

Thanks.

[neheb]

ping @cotequeiroz https://downloads.openwrt.org/snapshots/faillogs/mips_mips32/packages/openblas/compile.txt

[commodo]

ping @cotequeiroz https://downloads.openwrt.org/snapshots/faillogs/mips_mips32/packages/openblas/compile.txt

here we probably need to make it GENERIC

[cotequeiroz]

It is set to GENERIC, but generic is not compiling for mips CPUs for some reason.

[brada4]

If CPU is supposed to have MSA then MIPS24K is OK, otherwise there is no kernel/$ARCH/KERNEL.GENERIC

[commodo]

fixed with PR #17077

apologies for the lateness;
was busy with other stuff;