Speed comparison with Intel compiler

[Tim Prince]

Ignacio Fernández Galván wrote:
> --- Ray Nachlinger <ray@ultramarine.com> wrote:
>
>   
>>    Silly question, but why read the values into String at all? It
>>    appears that String is never used and it looks like you are 
>>    spending a lot of time blanking string.
>>     
>
> Of course, in this case it's useless, the whole program is useless, I
> was just reducing the original program (which *does* something, I was
> actually passing the string as an argument to a subroutine...) to
> something that would reproduce the difference in speed between both
> compilers. This example was to compare the speed when reading directly
> from a file or from a character variable, keeping the rest as constant
> as possible.
>
>   
As you are spending most of the time in library functions, it would be 
useful to include those in the gfortran profile.  I didn't find your 
information on compile/link versions/options there; I guess you are 
depending on our memory of your previous posts.
I suppose you used shared libraries with no profiling symbols.  
Profiling static libraries should give a lot more information.
You could confirm or deny the suggestion that your use of MATMUL is 
inefficient without in-line expansion, either by including it in the 
profile, or by checking the generated code to see whether either 
compiler in-lined.