Cleanup handling of march=native

[Dead2]

Simplify handling of native in CMake.

As a side note, ARM support in configure is quite messy. There is a lot of feature tests being defined directly in the final "Set arch specific flags" part (line ~1500+). All the other arches have their test functions defined well before that, and only have function calls in that part. Better readability and easier to reuse the test code.

[codecov]

Codecov Report

Patch coverage has no change and project coverage change: +0.02% 🎉

Comparison is base (0cb2b37) 83.87% compared to head (4336b89) 83.90%.

❗ Current head 4336b89 differs from pull request most recent head f377948. Consider uploading reports for the commit f377948 to get more accurate results

Additional details and impacted files

@@             Coverage Diff             @@
##           develop    #1544      +/-   ##
===========================================
+ Coverage    83.87%   83.90%   +0.02%     
===========================================
  Files          132      132              
  Lines        10843    10843              
  Branches      2801     2801              
===========================================
+ Hits          9095     9098       +3     
+ Misses        1049     1046       -3     
  Partials       699      699              

Flag	Coverage Δ
macos_clang	42.97% <ø> (ø)
macos_gcc	73.61% <ø> (ø)
ubuntu_clang	83.53% <ø> (+1.02%)	⬆️
ubuntu_clang_debug	82.14% <ø> (ø)
ubuntu_clang_inflate_allow_invalid_dist	82.15% <ø> (ø)
ubuntu_clang_inflate_strict	82.49% <ø> (ø)
ubuntu_clang_mmap	82.82% <ø> (ø)
ubuntu_clang_pigz	13.96% <ø> (ø)
ubuntu_clang_pigz_no_optim	11.51% <ø> (ø)
ubuntu_clang_pigz_no_threads	13.78% <ø> (ø)
ubuntu_clang_reduced_mem	83.92% <ø> (+1.02%)	⬆️
ubuntu_clang_toolchain_riscv	∅ <ø> (∅)
ubuntu_gcc	75.30% <ø> (+0.07%)	⬆️
ubuntu_gcc_aarch64	77.40% <ø> (ø)
ubuntu_gcc_aarch64_compat_no_opt	75.65% <ø> (ø)
ubuntu_gcc_aarch64_no_acle	76.16% <ø> (ø)
ubuntu_gcc_aarch64_no_neon	76.16% <ø> (ø)
ubuntu_gcc_armhf	77.47% <ø> (ø)
ubuntu_gcc_armhf_compat_no_opt	75.62% <ø> (+0.01%)	⬆️
ubuntu_gcc_armhf_no_acle	77.42% <ø> (ø)
ubuntu_gcc_armhf_no_neon	77.32% <ø> (ø)
ubuntu_gcc_armsf	74.65% <ø> (ø)
ubuntu_gcc_armsf_compat_no_opt	74.11% <ø> (+0.01%)	⬆️
ubuntu_gcc_benchmark	73.82% <ø> (+0.41%)	⬆️
ubuntu_gcc_compat_no_opt	76.85% <ø> (+0.01%)	⬆️
ubuntu_gcc_compat_sprefix	73.73% <ø> (ø)
ubuntu_gcc_m32	73.23% <ø> (-0.16%)	⬇️
ubuntu_gcc_mingw_i686	73.50% <ø> (ø)
ubuntu_gcc_mingw_x86_64	73.51% <ø> (ø)
ubuntu_gcc_mips	74.97% <ø> (ø)
ubuntu_gcc_mips64	74.98% <ø> (ø)
ubuntu_gcc_no_avx2	74.28% <ø> (ø)
ubuntu_gcc_no_ctz	74.65% <ø> (ø)
ubuntu_gcc_no_ctzll	74.64% <ø> (ø)
ubuntu_gcc_no_pclmulqdq	74.28% <ø> (ø)
ubuntu_gcc_no_sse2	74.54% <ø> (ø)
ubuntu_gcc_no_sse42	74.73% <ø> (ø)
ubuntu_gcc_o1	74.18% <ø> (+0.05%)	⬆️
ubuntu_gcc_osb	∅ <ø> (∅)
ubuntu_gcc_pigz	38.17% <ø> (+0.02%)	⬆️
ubuntu_gcc_pigz_aarch64	39.02% <ø> (-0.03%)	⬇️
ubuntu_gcc_ppc	73.92% <ø> (ø)
ubuntu_gcc_ppc64	74.36% <ø> (ø)
ubuntu_gcc_ppc64_power9	74.53% <ø> (ø)
ubuntu_gcc_ppc64le	74.43% <ø> (ø)
ubuntu_gcc_ppc64le_novsx	74.75% <ø> (ø)
ubuntu_gcc_ppc64le_power9	74.31% <ø> (ø)
ubuntu_gcc_ppc_no_power8	74.63% <ø> (ø)
ubuntu_gcc_s390x	74.80% <ø> (ø)
ubuntu_gcc_s390x_dfltcc	71.92% <ø> (ø)
ubuntu_gcc_s390x_dfltcc_compat	73.98% <ø> (ø)
ubuntu_gcc_s390x_no_crc32	74.59% <ø> (ø)
ubuntu_gcc_sparc64	74.79% <ø> (ø)
ubuntu_gcc_sprefix	73.24% <ø> (ø)
win64_gcc	73.95% <ø> (-0.05%)	⬇️
win64_gcc_compat_no_opt	74.71% <ø> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.

see 4 files with indirect coverage changes

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[nmoinvaz]

I have questions whether check_neon_compiler_flag and check_acle_compiler_flag still need to be run if native. I can't really tell.. Otherwise looks gtm.

[Dead2]

Rebased.

[Dead2]

I have questions whether check_neon_compiler_flag and check_acle_compiler_flag still need to be run if native. I can't really tell.. Otherwise looks gtm.

I think the correct thing to do is to rewrite those checks to instead use the native flag, and then verify that the neon/acle/etc code still compiles with native. And let that decide whether we attempt to include those optimized functions in the library or not.

Arm "native" was completely broken in configure before this PR, and with this it is less broken.
Before it would never compile any of the instruction set optimized functions.
With this PR it will enable all optimized functions that the compiler can support, not only the ones the cpu can support, different breakage is expected since much of the now-enabled code has never been tested like this before, only assumed to work.

I am tempted to just rip out the native support from configure completely, for all arches, rather than attempt to fix it and maintain it, and only support such advanced usage in CMake. It is a niche feature, and it is not trivial to get right with all the arch-specific code we have.

[nmoinvaz]

Regarding my question, I don't really expect it to be resolved with this PR. It is just a strange scenario.

[Dead2]

Rebased and removed commit making changes to Configure. CMake changes remain unchanged.