sse2neon -> SIMDe and added non-SIMD version

[mbrcic]

Substituted sse2neon by SIMDe library to convert SSE2 to ARM Neon.
Added option to build non-SIMD version of minimap2 using SIMDe.

SIMDe is added as a git submodule.

[junaruga]

This could be -Ilib/simde rather than -Ilib/simde/simde, modifying the loaded header paths. Because for example #include <simde/x86/sse2.h> is the right call by the simde system library (such as libsimde-dev in Debian) or simde-devel in Fedora and EPEL (CentOS).

[mbrcic]

I've changed to this in the following commit.

[junaruga]

I really appreciate this PR. Maybe we can try to run the code later on Travis ppc64le and s390x.

[junaruga]

Unnecessary difference. 1 space here.

[mbrcic]

I've removed the space in the following commit.

[lh3]

Thanks. Could you make SIMDe an option, not the default? For example:

make simde=1

[mbrcic]

All the comments so far have been implemented.

[junaruga]

Thanks for the update!

I think this part could like like this. And other parts are as well.
Because it's natural for me such as "if simde is used, include the simde header, otherwise include the default header.".

But let's wait @lh3 's response.

#ifdef USE_SIMDE
#include <simde/x86/sse2.h>
#else
#include <emmintrin.h>
#endif

[mbrcic]

As there was no other response, I have changed C preprocessor conditionals to the above stated form.

[junaruga]

Thanks. It looks good to me.

[mr-c]

FYI, I personally always use #define SIMDE_ENABLE_NATIVE_ALIASES + #include <simde/x86/sse2.h> as the SIMDe library has fallback that get compiled away to the plain intrinsic calls when on x86 with the regular compilation flag (-msse4.1, etc).

[junaruga]

I tested your PR on my forked repository adding new use cases tests in .travis.yml.
junaruga@7c0f286

simde=1 test on arch: arm64 is failed with the following error.
https://travis-ci.org/github/junaruga/minimap2/jobs/678368841#L162

gcc -c -g -Wall -O2 -Wc++-compat -DSIMDE_ENABLE_NATIVE_ALIASES -DUSE_SIMDE -msse2 -DHAVE_KALLOC -Ilib/simde ksw2_ll_sse.c -o ksw2_ll_sse.o
gcc: error: unrecognized command line option ‘-msse2’

Perhaps it might be good to add the tests in `.travis.yml in this PR, to guarantee the modification.
When Heng wants to modify your commits to adjust, he might do cherry-pick your commit modifying it.

[mbrcic]

simde=1 just enables SIMDe, but for current Makefile you have to enable also arm_neon=1 for arm and aarch64=1 for aarch64. I have tested with such setting on arm64.

The failed test used SIMDe, but pushed compiling with sse2 flag even though code contains no SSE2 code. arm64 compilers have no such flag.

[mbrcic]

simde=1 just enables SIMDe, but for current Makefile you have to enable also arm_neon=1 for arm and aarch64=1 for aarch64. I have tested with such setting on arm64.

The failed test used SIMDe, but pushed compiling with sse2 flag even though code contains no SSE2 code. arm64 compilers have no such flag.

Another option is to put just activation of simde=1 and put into all compiler calls in Makefile -march=native. Then SIMDe and gcc would work for all combinations by auto-detect, including x86 / x86_64. But, this is high-level decision that I did not feel I could make.

[junaruga]

Okay. In my humble opinion, I would prefer that make simde=1 works on non-x86_64 environment without arm_neon=1 or aarch64=1 if it is no problem for the performance and maintainability.
But I am just a contributor. So, let's wait Heng's response.

By the way, here is simde's .travis.yml. We can see some compiler flags tested on non-x86_64 environments.

[lh3]

@mbrcic In the end, I decide to create a new Makefile.simde for simde; otherwise Makefile would become too complex. Makefile.simde doesn't support CPU dispatch for simplicity. CPU dispatch is very rarely needed these days anyway.

@junaruga Currently minimap2 doesn't have test for Makefile.simde.

[junaruga]

@lh3 I have a suggestion. @mbrcic Could you send a pull-request to add a arch: arm64 case using make -f Makfile.simde updating .travis.yml, to test Makfile.simde and to show the reproductive command steps to minimap2 developers or users? I think it's useful.

[mbrcic]

Hi @junaruga , I've just made a PR with the requested changes.