Re: Compiling ATLAS with MinGW

[Michael Goffioul]

On Sun, Feb 19, 2012 at 3:46 AM, Tatsuro MATSUOKA <address@hidden> wrote:
> Hello
>
>
> --- On Sun, 2012/2/19, Michael Goffioul wrote:
>
>> On Sun, Feb 19, 2012 at 12:03 AM, Michael Goffioul
>> <address@hidden> wrote:
>> > On Sat, Feb 18, 2012 at 11:08 PM, Tatsuro MATSUOKA
>> > <address@hidden> wrote:
>> >> Hello
>> >>
>> >> I have complied the ATLAS self-build GCC on Cygwin compiler (gcc-4.5.3).
>> >> From the libatlats.a and liff77blas.a, I have create libblas.dll.a and 
>> >> libblas.dll using MinGW GCC as linker.
>> >>
>> >> This ATLAS can be linked with Octave on MinGW.
>> >>
>> >> The reason why I use self build GCC on cygwin, I have not create 
>> >> libblas.dll.a and libblas.dll with MinGW linker from libatlats.a and 
>> >> libff77blas.a built with the GCC bundled with Cygwin.
>> >
>> > Thanks Tatsuro. So you're basically doing the same as me. The reason
>> > I'm asking is that I've been trying to recompile ATLAS with
>> > multi-thread support, but went into problems. The multi-threaded ATLAS
>> > version is compiled against pthread library. However when applying the
>> > same method as above, you're using cygwin-GCC to compile ATLAS; hence
>> > you're using the cygwin version of pthread headers. At link stage,
>> > you're using MinGW-GCC and link against the MinGW version of pthread
>> > library (which is basically this one:
>> > http://sourceware.org/pthreads-win32/).
>> >
>> > But the catch is that the pthread types in the cywin version and MinGW
>> > version are not the same, especially they don't have the same size. In
>> > other words: sizeof(pthread_t) in cygwin is not the same as
>> > sizeof(pthread_t) in MinGW. And you end up with stack corruption.
>> >
>> > I tried to compile ATLAS with the mingw-xxx compilers that are
>> > available in cygwin, but teh build process stops almost immediately
>> > with an error (ERROR 435 DURING Stage 1 install).
>>
>> Actually, it seems to work fine when I keep the XCC compiler to be
>> cygwin-GCC. I'll give that a shot when I get some time.
>>
>> Michael.
>>
>
>
> Thank you for your information.  As you described I could not build 
> multi-threaded atlas because of pthread issue.  I hope that your trial will 
> be successful.

I managed to complete the compilation process, but it wasn't really a
success. I specified MinGW as main compiler, but kept cygwin-GCC as
the XCC compiler. I had to hack a bit the Makefile in tune/sysinfo/,
because some support executables are compiled with the ICC compiler
(interface compiler), mingw-in-cygwin in this case, but are then
called with cygwin-style paths and fail. I also noticed a similar
issue that prevented a proper creation of files like
atlas_dtrsmXover.h (empty files are created), although these errors
were ignored by make.

I created the DLL and tested the result with octave with this snippet:

n=2000; A=randn(n); B=randn(n);tic; C=A*B; t=toc, MFLOPS=2*n^3/t*1e-6

and verified that both CPU's were used (in the task manager). However
it seems that the generated multi-threaded ATLAS is suboptimal and
even with the 2 CPU's being used, it was slower than a single-threaded
ATLAS. I don't know the reason, but it may be related to the fact that
some files couldn't be properly generated. In the end, it seems like
trying to compile ATLAS with MinGW is a dead-end...

Michael.