[CMake] CMake FindBLAS and FindLAPACK patches

[don la dieu]

That was intentional. .dll Should already be in CMAKE_FIND_LIBRARY_SUFFIXES on WIN32.

Since your commits haven't hit git/gitweb yet I'm not sure what you didn't commit.

As far as the ACML changes go, the file globbing is way too greedy, and breaks if more than one ACML or ACML-GPU package is installed. ACML-GPU can be found when you want ACML and vice versa. So, that clearly needs to change. Also, in the check_fortran_libraries() for ACML you're try trying to search the the multithreaded library (libacml_mp) when you need to search the single threaded library (libacml). And, in ACML-GPU you also search the multithreaded library (libacml_mp) which doesn't exist in the ACML-GPU package provided by AMD. See http://goo.gl/fQKq6 and http://goo.gl/2qCeW. So, clearly those fixes need to be in there.

Don

On Sep 1, 2011, at 3:02 AM, Alexey Ozeritsky wrote:

> Hello,
> 
> I've commited all the changes except ACML.
> 
> Please check your patches more careful next time:
> 
> -      set(CMAKE_FIND_LIBRARY_SUFFIXES ".lib;.dll")
> +      set(CMAKE_FIND_LIBRARY_SUFFIXES .lib ${CMAKE_FIND_LIBRRAY_SUFFIXES})
> .......................................................................................................^^^^^^
> 
> 
> On Sat, Aug 27, 2011 at 9:21 PM, don la dieu <nega at icecube.umd.edu> wrote:
>> Hi Alexey-
>> 
>> Sorry for the delay in getting back to you.
>> 
>> Attached are updates to the FindBLAS you sent me (well, actually trunk.)
>> FindLAPACK was fine. It tightens up the default directory detection (it
>> prefers the the most current version it can find) and fixes a couple of
>> typos. I didn't add the recently released ACML v5.0.0. AMD changed
>> things up with that release, and it will probably require an additional
>> BLA_VENDOR.
>> 
>> Also, I've attached 2 patches that improve "Generic" detection on Ubuntu
>> (and I assume Debian), and works with the libblas3gf and liblapack3gf
>> packages. It does this by saving/changing/restoring
>> CMAKE_FIND_LIBRARY_SUFFIXES. I added BLAS_STATIC to that logic too.
>> Note, these are are against 2.8.5. I can update them to trunk if you want.
>> 
>> I found 2 minor bugs which I haven't patched as I'm not sure how you
>> wanted to address it. BLA_VENDORs "Apple" (Accelerate Framework) and
>> "NAS" (vecLib Framework) fail as 'cblas_dgemm_' doesn't exist.
>> 'clbas_dgemm' and I think 'dgemm_' do though.
>> 
>> CC'd the list as this note might be relevant to others.
>> 
>> Don
>> 
>> On 08/13/2011 09:09 AM, Alexey Ozeritsky wrote:
>>> Hi,
>>> 
>>> You can test attached modules.
>>> There are 3 ways to set acml library paths
>>> 1. install libraries to LD_LIBRARY_PATH or LIB (for windows)
>>> 2. set variable BLAS_ACML_LIB_DIRS or BLAS_ACML_MP_LIB_DIRS or
>>> BLAS_ACML_GPU_LIB_DIRS
>>> 3. install libraries to /opt/acml* or c:/amd/acml (for windows)
>>> 
>>> On Fri, Aug 12, 2011 at 12:28 AM, don la dieu <nega at icecube.umd.edu> wrote:
>>>> Hi Alexey-
>>>> 
>>>> It probably makes more sense to have ACML-GPU as a separate BLAS_VENDOR, especially since that's what you're already doing with ACML and ACML_MP. It would also make the code clearer.
>>>> 
>>>> Don
>>>> 
>>>> On Aug 11, 2011, at 4:31 AM, Alexey Ozeritsky wrote:
>>>> 
>>>>> Hello,
>>>>> 
>>>>> Thanks for patches.
>>>>> I've just pushed GotoBLAS2 into upstream.
>>>>> 
>>>>> I think that ACML-GPU support should be done as separate BLAS_VENDOR.
>>>>> What do you think about it ?
>>>>> 
>>>>> 11.08.2011, 00:43, "don la dieu" <nega at icecube.umd.edu>:
>>>>>> Alexey-
>>>>>> 
>>>>>> Attached are two patches for adding detection support for GotoBLAS2 and
>>>>>> ACML-GPU. Both patches are against CMake 2.8.5.
>>>>>> 
>>>>>> Thanks,
>>>>>> Don
>>>> 
>>>> 
>> 
>>