Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > comp.lang.forth > #59743

Re: ANN: matmul package available

Show all headers | View raw

pahihu <pahihu@gmail.com> writes:
>2017. m=C3=A1jus 22., h=C3=A9tf=C5=91 13:22:09 UTC+2 id=C5=91pontban Anton =
>Ertl a k=C3=B6vetkez=C5=91t =C3=ADrta:
>> And of course auto-vectorization looks attractive to managers with
>> lots of existing code.  It promises speedups for free, while the real
>> cure requires an investment in programmers who know how to vectorize.
>>=20
>
>I've got GCC 4.9.4 on my machine and tried auto vectorization.
>I compiled this dumb C routine with it:
>
>void daxpy(double ra,double *x,double *y,unsigned n)
>{
>        unsigned i;
>
>        for (i =3D 0; i < n; i++)
>            y[i] +=3D ra * x[i];
>}

That's actually daxpy_nostride; a real daxpy is

daxpy (N, DA, DX, INCX, DY, INCY)

<http://www.netlib.org/lapack/explore-html/d9/dcd/daxpy_8f.html>

On AMD64, already gcc -O3 produces vectorized SSE2 code for
this function; the function is compiled into a 229-byte machine-code
routine, with the inner loop being:

  83:   66 0f 10 0c 08          movupd (%rax,%rcx,1),%xmm1
  88:   41 83 c0 01             add    $0x1,%r8d
  8c:   66 0f 59 ca             mulpd  %xmm2,%xmm1
  90:   66 41 0f 58 0c 0a       addpd  (%r10,%rcx,1),%xmm1
  96:   41 0f 29 0c 0a          movaps %xmm1,(%r10,%rcx,1)
  9b:   48 83 c1 10             add    $0x10,%rcx
  9f:   45 39 c8                cmp    %r9d,%r8d
  a2:   72 df                   jb     83 <xdaxpy+0x83>

If I compile it with gcc -O3 -mavx2, I get a 385-byte routine, with
the inner code being

  ca:   c4 a1 79 10 0c 00       vmovupd (%rax,%r8,1),%xmm1
  d0:   41 83 c3 01             add    $0x1,%r11d
  d4:   c4 a3 75 18 4c 00 10    vinsertf128 $0x1,0x10(%rax,%r8,1),%ymm1,%ymm1
  db:   01 
  dc:   c5 f5 59 ca             vmulpd %ymm2,%ymm1,%ymm1
  e0:   c4 81 75 58 0c 04       vaddpd (%r12,%r8,1),%ymm1,%ymm1
  e6:   c4 81 7d 29 0c 04       vmovapd %ymm1,(%r12,%r8,1)
  ec:   49 83 c0 20             add    $0x20,%r8
  f0:   45 39 cb                cmp    %r9d,%r11d
  f3:   72 d5                   jb     ca <xdaxpy+0xca>

The xmm registers are 128-bits (2 C doubles) wide, the ymm registers
256 bits (4 C doubles).

However, wrt to the part of my posting you cited, the question is how
much of the code is like the C code above, and how much contains
various stumbling blocks for vectorization.  E.g., gcc 4.9 does not
vectorize faxpy (which is much closer to the real daxpy than the
function above).  Another question is how much of the code where the
stumbling blocks can be overcome is actually vectorized by gcc.

- anton
-- 
M. Anton Ertl  http://www.complang.tuwien.ac.at/anton/home.html
comp.lang.forth FAQs: http://www.complang.tuwien.ac.at/forth/faq/toc.html
     New standard: http://www.forth200x.org/forth200x.html
   EuroForth 2017: http://www.euroforth.org/ef17/

Back to comp.lang.forth | Previous | Next — Previous in thread | Next in thread | Find similar | Unroll thread

Thread

ANN: matmul package available anton@mips.complang.tuwien.ac.at (Anton Ertl) - 2017-05-19 10:43 +0000
  Re: ANN: matmul package available krishna.myneni@ccreweb.org - 2017-05-19 05:20 -0700
    Re: ANN: matmul package available anton@mips.complang.tuwien.ac.at (Anton Ertl) - 2017-05-19 14:16 +0000
      Re: ANN: matmul package available pahihu <pahihu@gmail.com> - 2017-05-19 08:28 -0700
        Re: ANN: matmul package available pahihu <pahihu@gmail.com> - 2017-05-20 05:25 -0700
        Re: ANN: matmul package available anton@mips.complang.tuwien.ac.at (Anton Ertl) - 2017-05-20 13:23 +0000
          Re: ANN: matmul package available pahihu <pahihu@gmail.com> - 2017-05-20 08:43 -0700
            Re: ANN: matmul package available anton@mips.complang.tuwien.ac.at (Anton Ertl) - 2017-05-20 16:02 +0000
              Re: ANN: matmul package available pahihu <pahihu@gmail.com> - 2017-05-20 11:36 -0700
                Re: ANN: matmul package available anton@mips.complang.tuwien.ac.at (Anton Ertl) - 2017-05-21 07:52 +0000
                Re: ANN: matmul package available mhx@iae.nl - 2017-05-21 03:28 -0700
                Re: ANN: matmul package available pahihu <pahihu@gmail.com> - 2017-05-21 05:32 -0700
                Re: ANN: matmul package available mhx@iae.nl - 2017-05-21 05:52 -0700
                Re: ANN: matmul package available pahihu <pahihu@gmail.com> - 2017-05-21 07:54 -0700
                Re: ANN: matmul package available mhx@iae.nl - 2017-05-21 08:32 -0700
                Re: ANN: matmul package available pahihu <pahihu@gmail.com> - 2017-05-21 11:24 -0700
                Re: ANN: matmul package available anton@mips.complang.tuwien.ac.at (Anton Ertl) - 2017-05-21 13:52 +0000
                Re: ANN: matmul package available mhx@iae.nl - 2017-05-21 08:47 -0700
                Re: ANN: matmul package available anton@mips.complang.tuwien.ac.at (Anton Ertl) - 2017-05-22 09:50 +0000
                Re: ANN: matmul package available pahihu <pahihu@gmail.com> - 2017-05-22 07:14 -0700
                Re: ANN: matmul package available mhx@iae.nl - 2017-05-22 07:39 -0700
                Re: ANN: matmul package available anton@mips.complang.tuwien.ac.at (Anton Ertl) - 2017-05-23 10:06 +0000
  Re: ANN: matmul package available albert@cherry.spenarnc.xs4all.nl (Albert van der Horst) - 2017-05-19 17:00 +0200
    Re: ANN: matmul package available anton@mips.complang.tuwien.ac.at (Anton Ertl) - 2017-05-20 13:59 +0000

csiph-web