[Rd] complex.h in R

[Prof Brian Ripley]

On Mon, 26 Sep 2005, Jennifer Lai wrote:

> Hi,
> I checked config.log and BLAS_LIBS was not set. However, I have set 
> BLAS_LIBS='-L/usr/lib64 -lblas' in config.site file.
> I can't figure out why BLAS_LIBS is not set, when PGI compiler is used. When 
> gcc is used, BLAS_LIBS need not be set in config.site and automatically get 
> picked up by the configuration script.
>
> Here is a snapshot of the configuration script output (for buidling R with 
> PGI compiler):
> checking for complex.h... yes
> checking for double complex... no
> checking for sgemm_ in -L/usr/lib64 -lblas... yes
> checking whether double complex BLAS can be used... no
>
> Is there other thing I should look into in the config.log?

You need to look for the evidence for the line

> checking for sgemm_ in -L/usr/lib64 -lblas... yes

in config.log.  I very much doubt that -L/usr/lib64 helps you: surely that 
should be in your library path.  But you will probably find you cannot mix 
code compiled under different compilers, especially as -lblas is likely to 
be Fortran compiled with g77.  (You may need -lblas -lg2c, but you would 
be better off using the BLAS built into R.)


> Your help is very much appreicated.
>
>
> Thanks,
> Jennifer
>
>
> Prof Brian Ripley wrote:
>
>> On Sat, 24 Sep 2005, Jennifer Lai wrote:
>> 
>> 
>>> Hi,
>>>    How does complex.h used in R? Whether a compiler support complex.h
>>> or not, does it affect R's performance?
>>> 
>> 
>> complex.h will only used in future (2.2.0-to-be) versions of R, and only
>> if configure finds enough C99 support. Otherwise R's own C-level complex
>> support is used, as it always was.  One would expect the OS's support to
>> be faster and more accurate (but one could be disappointed, we have
>> found).
>> 
>> 
>>> I used PGI compiler to build
>>> R-devel on AMD Opteron, but the configuration file failed to link BLAS
>>> library despite the fact it is located in the usual location,
>>> /usr/lib64.
>>> 
>> 
>> Look in config.log to find out why.
>> 
>> 
>>> PGI said they don't support complex.h. R configuration
>>> script printed out that doublecomplex is not supported.
>>> 
>> 
>> That is *FORTRAN* DOUBLE COMPLEX, and that does affect the operation of R,
>> as without it you will not have complex linear algebra support.
>> 
>> 
>>> Are "supporting
>>> complex.h" and "linking BLAS library" related?
>>> 
>> 
>> No.  They are completely orthoogonal concepts.
>> 
>> 
>
>
>

-- 
Brian D. Ripley,                  ripley at stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595