| [eigen] Eigen beta test on PowerPC/ARM |
[ Thread Index |
Date Index
| More lists.tuxfamily.org/eigen Archives
]
Hi all,
here are the results of the beta1 tests on both PowerPC (with and without
vectorization), and ARM (without vectorization, I got some ICE with the
current gcc (4.4.1) on some tests):
PowerPC:
99% tests passed, 6 tests failed out of 452
Total Test time (real) = 344.90 sec
The following tests FAILED:
37 - integer_types_3 (Failed)
252 - qr_3 (Failed)
254 - qr_5 (Failed)
302 - eigensolver_selfadjoint_2 (Failed)
307 - eigensolver_selfadjoint_7 (Failed)
410 - householder_4 (Failed)
Errors while running CTest
PowerPC/Altivec:
99% tests passed, 4 tests failed out of 452
Total Test time (real) = 416.48 sec
The following tests FAILED:
37 - integer_types_3 (Failed)
302 - eigensolver_selfadjoint_2 (Failed)
443 - nullary_7 (Failed)
444 - nullary_8 (Failed)
Errors while running CTest
PowerPC/Altivec+ROW_MAJOR:
99% tests passed, 3 tests failed out of 452
Total Test time (real) = 386.65 sec
The following tests FAILED:
37 - integer_types_3 (Failed)
443 - nullary_7 (Failed)
444 - nullary_8 (Failed)
Errors while running CTest
PowerPC+ROW_MAJOR:
98% tests passed, 10 tests failed out of 452
Total Test time (real) = 346.70 sec
The following tests FAILED:
37 - integer_types_3 (Failed)
140 - array_4 (Failed)
225 - cholesky_6 (Failed)
235 - lu_7 (Failed)
252 - qr_3 (Failed)
254 - qr_5 (Failed)
302 - eigensolver_selfadjoint_2 (Failed)
307 - eigensolver_selfadjoint_7 (Failed)
400 - umeyama_2 (Failed)
410 - householder_4 (Failed)
Errors while running CTest
PowerPC/Altivec+OPENMP:
99% tests passed, 6 tests failed out of 452
Total Test time (real) = 359.50 sec
The following tests FAILED:
37 - integer_types_3 (Failed)
264 - qr_colpivoting_3 (Failed)
266 - qr_colpivoting_5 (Failed)
271 - qr_fullpivoting_1 (Failed)
443 - nullary_7 (Failed)
444 - nullary_8 (Failed)
Errors while running CTest
PowerPC+OPENMP:
ts passed, 6 tests failed out of 452
Total Test time (real) = 438.42 sec
The following tests FAILED:
37 - integer_types_3 (Failed)
252 - qr_3 (Failed)
254 - qr_5 (Failed)
302 - eigensolver_selfadjoint_2 (Failed)
307 - eigensolver_selfadjoint_7 (Failed)
410 - householder_4 (Failed)
Errors while running CTest
ARM:
99% tests passed, 4 tests failed out of 452
Total Test time (real) = 2686.07 sec
The following tests FAILED:
3 - dynalloc (Failed)
37 - integer_types_3 (Failed)
345 - geo_quaternion_1 (Failed)
346 - geo_quaternion_2 (Failed)
Errors while running CTest
ARM+ROW_MAJOR:
t time (real) = 2187.20 sec
The following tests FAILED:
3 - dynalloc (Failed)
37 - integer_types_3 (Failed)
345 - geo_quaternion_1 (Failed)
346 - geo_quaternion_2 (Failed)
Errors while running CTest
ARM+OPENMP:
99% tests passed, 4 tests failed out of 452
Total Test time (real) = 2299.18 sec
The following tests FAILED:
3 - dynalloc (Failed)
37 - integer_types_3 (Failed)
345 - geo_quaternion_1 (Failed)
346 - geo_quaternion_2 (Failed)
Errors while running CTest
I hope this helps you track down some remaining errors, I'd gladly try
anything on these machines. Anyway, nothing looks like a show stopper -and
both those arches are secondary anyway so 3.0 could be released even with a
few tests failing on ARM/PowerPC.
Regards
Konstantinos
PS. I see that in some cases PowerPC without vectorization is faster than
using vectorization. Might I suggest we have a benchmark that measures each
optimized trait separately so that we can see which function behaves badly -I
already found one on powerpc, integer pmadd on PowerPC, which I disabled).