Graviton2-b

[accopeland]

Another attempt at an aarch64 support. I made some changes to CMakelists.txt to prevent errors at config time. Added some ifdefs around the arch-specific parts so you should get identical x86_64 performance now and working binaries on aarch64 (only graviton2 tested so far).

[voutcn]

Thank you for the PR.

Where is ../simde/x86/avx2.h?

[pkubaj]

Also, why limit it to aarch64? ppc64* could probably be fixed by this as well.

[accopeland]

Apologies, I forgot that bit. I could update the branch either via a git submodule or local copy (from https://github.com/simd-everywhere/simde). Do you have a preference?

…

On Mon, Apr 4, 2022 at 9:48 PM Dinghua Li ***@***.***> wrote: ***@***.**** commented on this pull request. ------------------------------ In src/kmlib/kmrns.h <#330 (comment)>: > @@ -8,7 +8,12 @@ #include <stdint.h> #include <stdio.h> #include <stdlib.h> -#include <x86intrin.h> +#if defined(__GNUC__) && defined(__aarch64__) + #define SIMDE_ENABLE_NATIVE_ALIASES + #include "../simde/x86/avx2.h" Thank you for the PR. Where is ../simde/x86/avx2.h? — Reply to this email directly, view it on GitHub <#330 (review)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAE7ERRJN45FKO3V6RIIABLVDPAYPANCNFSM5SRCFAEA> . You are receiving this because you authored the thread.Message ID: ***@***.***>

[voutcn]

Found that with this PR HasPopCnt and HasBmi2 will always return false on aarch64. And the megahit python script will always choose the megahit_core binary without popcnt and BMI2. Thus you don't need to include any simde headers.

#if !defined(__aarch64__)
  #include <x86intrin.h>
#endif

and change

megahit/src/kmlib/kmrns.h

Line 268 in 0abdd48

#if defined(__BMI2__) && defined(USE_BMI2)

to

defined(__BMI2__) && defined(USE_BMI2) && !defined(__aarch64__)

should be sufficient to get it compiled.

[voutcn]

FYI. We have three megahit_core binary compiled: one for popcnt + BMI2, one for popcnt only and one for none of them. And the megahit python script will choose the correct binary (through the functions in cpu_dispatch.h) here:

megahit/src/megahit

Lines 613 to 630 in f8afe5d

	def cpu_dispatch():
	if not opt.hw_accel:
	logger.info('Using megahit_core without POPCNT and BMI2 support, '
	'because --no-hw-accel option manually specified')
	else:
	has_hw_accel = check_output([opt.megahit_core, 'checkcpu'])
	if has_hw_accel == '1':
	logger.info('Using megahit_core with POPCNT and BMI2 support')
	else:
	opt.hw_accel = False
	has_popcnt = check_output([opt.megahit_core, 'checkpopcnt'])
	if has_popcnt == '1':
	opt.has_popcnt = True
	logger.info('Using megahit_core with POPCNT support')
	else:
	logger.info('Using megahit_core without POPCNT and BMI2 support, '
	'because the features not detected by CPUID ')

If we have a good enough way to check POPCNT or BMI2 support on ARM we can revisit the POPCNT and BMI2 hardware acceleration options for ARM.

[pkubaj]

Again, why only aarch64?