OpenBLAS on LoongArch64: openblas_config.h: error: '_Float16' is not supported on this target

[wegank]

I encountered the same error as in #5700 when building FLINT on LoongArch64:

configure:15030: checking for cblas.h
configure:15030: gcc -c   -I/nix/store/s3qryikb131hghk6qnakyhrvhwn2v5vp-gmp-with-cxx-6.3.0/include -I/nix/store/29b1k4wdvn1ig9rwnrhjvgp8k5n4njw5-mpfr-4.2.2/include -I/nix/store/lmzx27k35xrg1rdpwc3m>
In file included from /nix/store/h52mz21p1k513nyfvxv40gsllqw29dqi-openblas-0.3.31-dev/include/cblas.h:33,
                 from conftest.c:58:
/nix/store/h52mz21p1k513nyfvxv40gsllqw29dqi-openblas-0.3.31-dev/include/openblas_config.h:63:9: error: '_Float16' is not supported on this target
   63 | typedef _Float16 hfloat16;
      |         ^~~~~~~~
configure:15030: $? = 1

This issue didn't happen before NixOS/nixpkgs#450106 where OpenBLAS was updated to 0.3.31.
@martin-frbg Could you help me, please?

[martin-frbg]

Ah, sorry. FP16 support is new in 0.3.31 and I did not expect this would create a problem for code compiled against OpenBLAS.
I think the PPC solution from #5670 should work for Loongarch too, that is, change openblas_config_template.h around line 43 to have

#if defined(OPENBLAS_ARCH_POWER) || defined(OPENBLAS_ARCH_LOONGARCH64)
typedef bfloat16 hfloat16
#else
typedef __Float16 hfloat16
#endif

instead of having just

typedef __Float16 hfloat16 

and/or make the same change around line 63 of openblas_config.h in your already compiled OpenBLAS