[Bug libstdc++/107929] New: std::experimental::simd needs to be reimplemented with GCC's vector extension

[Bug libstdc++/107929] New: std::experimental::simd needs to be reimplemented with GCC's vector extension

[unlvsur at live dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107929

            Bug ID: 107929
           Summary: std::experimental::simd needs to be reimplemented with
                    GCC's vector extension
           Product: gcc
           Version: 13.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: libstdc++
          Assignee: unassigned at gcc dot gnu.org
          Reporter: unlvsur at live dot com
  Target Milestone: ---

https://gcc.gnu.org/onlinedocs/gcc/Vector-Extensions.html

It is just too slow to compile by including immintrin.h headers. Using vector
extension is extremely easy to implement and more direct.

[Bug libstdc++/107929] std::experimental::simd needs to be reimplemented with GCC's vector extension

[unlvsur at live dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107929

cqwrteur <unlvsur at live dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |unlvsur at live dot com

--- Comment #1 from cqwrteur <unlvsur at live dot com> ---
copy_from and copy_to can be implemented with __builtin_memcpy

https://github.com/cppfastio/fast_io/blob/b55c940d59ea70eb6c644d2c71d157a7d53a8df6/include/fast_io_core_impl/simd/gcc_clang.h#L52

[Bug libstdc++/107929] std::experimental::simd needs to be reimplemented with GCC's vector extension

[mkretz at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107929

Matthias Kretz (Vir) <mkretz at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
         Resolution|---                         |INVALID
                 CC|                            |mkretz at gcc dot gnu.org
             Status|UNCONFIRMED                 |RESOLVED

--- Comment #2 from Matthias Kretz (Vir) <mkretz at gcc dot gnu.org> ---
std::experimental::simd has always been implemented with GCC's vector
extension. Also copy_from and copy_to are implemented with __builtin_memcpy.

[Bug libstdc++/107929] std::experimental::simd needs to be reimplemented with GCC's vector extension

[unlvsur at live dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107929

--- Comment #3 from cqwrteur <unlvsur at live dot com> ---
(In reply to Matthias Kretz (Vir) from comment #2)
> std::experimental::simd has always been implemented with GCC's vector
> extension. Also copy_from and copy_to are implemented with __builtin_memcpy.

The problem is that the compilation speed are very slow. immintrin.h header
should not be included

[Bug libstdc++/107929] std::experimental::simd needs to be reimplemented with GCC's vector extension

[mkretz at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107929

--- Comment #4 from Matthias Kretz (Vir) <mkretz at gcc dot gnu.org> ---
I agree that compilation speed is a problem. However,

1. GCC's vector extensions are not sufficient (especially for use of AVX-512
masks)

2. I see 0.6s difference between calling `g++ -O2 -march=skylake` on an empty
file vs. a file that #includes <immintrin.h>. Whereas #include
<experimental/simd> takes about 2.7s.

[Bug libstdc++/107929] std::experimental::simd needs to be reimplemented with GCC's vector extension

[unlvsur at live dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107929

--- Comment #5 from cqwrteur <unlvsur at live dot com> ---
1. the real gap is much larger due to larger preprocessor space.
2. that is just an excuse. the mm_xxxx intrinsic is implemented with builtin
functions on gcc and clang. You can always avoid include that header by using
builtin functions. There is no reason to include headers.
3. You really need that to support none x86 none arm platforms

Get Outlook for Android<https://aka.ms/AAb9ysg>
________________________________
From: mkretz at gcc dot gnu.org <gcc-bugzilla@gcc.gnu.org>
Sent: Thursday, December 1, 2022 5:07:42 AM
To: unlvsur@live.com <unlvsur@live.com>
Subject: [Bug libstdc++/107929] std::experimental::simd needs to be
reimplemented with GCC's vector extension

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107929

--- Comment #4 from Matthias Kretz (Vir) <mkretz at gcc dot gnu.org> ---
I agree that compilation speed is a problem. However,

1. GCC's vector extensions are not sufficient (especially for use of AVX-512
masks)

2. I see 0.6s difference between calling `g++ -O2 -march=skylake` on an empty
file vs. a file that #includes <immintrin.h>. Whereas #include
<experimental/simd> takes about 2.7s.

--
You are receiving this mail because:
You are on the CC list for the bug.
You reported the bug.