Re: [pooma-dev] Good News. Intel's ICC 8.0 Beta looks promising, now.

[Richard Guenther]

On Tue, 3 Jun 2003, Paul A. Renard wrote:

> Richard:
>
> From your message:
>       Unfortunately my tests show its better, but still worse than with gcc.
>       Your test is 1d, try 3d and it starts to suck. Inlining is still the
>       culprit, as is CSE with f.i. Loc<n> (where n>1) objects.
>
>
> Actually, my test is 2D.  Do you have a 3D test you can send?  Were you 
> comparing
> icc 8.0?  I'd like to try your test on my machine with KCC and icc 8.0

Yes, attached. I tested icc8.0 and gcc3.3 (patched). I'd be interested in
KCC results, too.

With gcc I get

Benchmark size 262144:
  ET:               5.55688e-08
  Stencil:          6.05278e-08
  ScalarCode (int): 7.5695e-08
  ScalarCode (Loc): 1.15906e-07
Benchmark size 2097152:
  ET:               5.74374e-08
  Stencil:          6.38685e-08
  ScalarCode (int): 7.94697e-08
  ScalarCode (Loc): 1.19308e-07
Benchmark size 262144:
  ET:               7.75644e-08
  Stencil:          7.78923e-08
  ScalarCode (int): 6.76191e-08
  ScalarCode (Loc): 1.55674e-07
Benchmark size 2097152:
  ET:               6.99201e-08
  Stencil:          7.7395e-08
  ScalarCode (int): 6.24175e-08
  ScalarCode (Loc): 1.54993e-07
Total (sum) s/iteration 1.37126e-06

with icc

Benchmark size 262144:
  ET:               7.37382e-08
  Stencil:          7.42148e-08
  ScalarCode (int): 8.37249e-08
  ScalarCode (Loc): 9.26857e-08
Benchmark size 2097152:
  ET:               8.0122e-08
  Stencil:          7.84069e-08
  ScalarCode (int): 8.49171e-08
  ScalarCode (Loc): 9.70053e-08
Benchmark size 262144:
  ET:               1.14643e-07
  Stencil:          9.76029e-08
  ScalarCode (int): 6.61776e-08
  ScalarCode (Loc): 1.42822e-07
Benchmark size 2097152:
  ET:               1.13272e-07
  Stencil:          9.84888e-08
  ScalarCode (int): 5.80321e-08
  ScalarCode (Loc): 1.41148e-07
Total (sum) s/iteration 1.497e-06


While the 1d Loc using ScalarCode are better with icc, the 3d expression
template versions are awfully slow (filed a PR already).

Richard.

Bench.cpp
Description: Text document