Re: [pooma-dev] [RFC] Change Pooma/tests

[Richard Guenther]

On 6 Jan 2003, Tarjei Knapstad wrote:

> > I do not have access to Intel's compiler so I cannot check the status
> > of the regression tests.
> >
> I've just succesfully built POOMA (current CVS) with the latest Intel
> compiler on Linux together with all the tests except for the Connect
> directory. If you'd like test results I'd be happy to compile some for
> you tomorrow.
>
> My current setup:
>
> Intel(R) C++ Compiler for 32-bit applications, Version 7.0   Build
> 20021021Z

With

Intel(R) C++ Compiler for 32-bit applications, Version 7.0   Build
20021212Z

I get those additional failures (compared to a run with a gcc3.2 that
still contains the loop optimizer bug fixed around Dec20) (-PASSED lines
usually mean compile failures or segfaults of the tests):

-FAILED ... array_test27
+PASSED ... array_test27
+PASSED ... MP DynamicArray w/ shared layouts
 FAILED ... DynamicArray dynamic ops w/views
 FAILED ... NearestNeighbors
-PASSED ... DiskLayoutTest1
-PASSED ... FileSetWriterTest2

Unfortunately gcc from todays gcc-3_2 branch fixes none of the above bugs
(I didnt see array_test27 failure on powerpc, it seems a floating point
comparison failure which might be fixed using -ffloat-store or changing
precision to 64bits rather than 80 which is default for gcc but not icc).

One may also note that compiling poom and running the complete testsuite
takes around the same time with gcc-3.2 and icc 7.0 when using -O2 on both
builds (this is of course mainly compile time and IO).

Richard.

--
Richard Guenther <address@hidden>
WWW: http://www.tat.physik.uni-tuebingen.de/~rguenth/