clmagma 1.0.0 testing_xgemm errors (nvidia gpu)

Open discussion for MAGMA

clmagma 1.0.0 testing_xgemm errors (nvidia gpu)

Postby mb_infn_it » Tue Oct 30, 2012 11:07 am

I would like to report a strange ./testing_xgemm behaviour on my machine running clmagma 1.0.0, also seen with clmagma 0.3.
I used to got this kind of error when passing wrong arguments or data to the OCL kernel, making it terminate with wrong results and quite-null computational time.
I can reproduce this behaviour only on xgemm kernels: as you can see from the output below, single precision test kernel seems to work well before -M 6102 -N 6102 test, while dgemm, zgemm and cgemm gives very similar output on almost every work dimension.

I made some hacks on make.inc and Makefile.internal files to make work NVIDIA gpu with clAmdBlas (i wanted an opencl-only setup), so maybe i did some configuration mistake..
Can you please check them and point me a solution? Thank you in advance.

Environment (Linux Fedora 17):
. Intel i5 quad-core
. NVIDIA Quadro 600 (NVIDIA UNIX x86_64 Kernel Module 304.51)
. GCC version: gcc version 4.7.2 20120921 (Red Hat 4.7.2-2) (GCC)

. clAmdBlas 1.8.291
. AMD-APP-SDK 2.7 lnx64
. clmagma-1.0.0

Code: Select all
[xxx@yyy testing]$ ./testing_dgemm && ./testing_sgemm
Initializing clMAGMA runtime ...

Usage:
  testing_dgemm [-NN|NT|TN|TT] [-N 1024]


Testing transA = o  transB = o
    M    N    K   clAmdBlas GFLop/s (sec)    CPU GFlop/s (sec)     error
===========================================================================
 1024  1024  1024    16930.83 (  0.00)      6.59 (  0.33)    8.340644e+01
 1280  1280  1280    37430.18 (  0.00)      6.64 (  0.63)    1.803064e+02
 1600  1600  1600    73105.83 (  0.00)      6.64 (  1.23)    3.013679e+02
 2000  2000  2000    134486.70 (  0.00)      6.68 (  2.40)    4.507322e+02
 2500  2500  2500    276523.21 (  0.00)      6.69 (  4.67)    6.353319e+02
 3125  3125  3125    507936.51 (  0.00)      6.68 (  9.13)    8.665594e+02
 3906  3906  3906    1028609.07 (  0.00)      6.71 ( 17.77)    1.154436e+03
 4882  4882  4882    2059230.13 (  0.00)      6.69 ( 34.78)    1.515831e+03
 6102  6102  6102    4020945.33 (  0.00)      6.69 ( 67.88)    1.963820e+03
Initializing clMAGMA runtime ...

Usage:
  testing_sgemm [-NN|NT|TN|TT] [-N 1024]


Testing transA = o  transB = o
    M    N    K   clAmdBlas GFLop/s (sec)    CPU GFlop/s (sec)     error
===========================================================================
 1024  1024  1024       42.50 (  0.05)     13.30 (  0.16)    1.754761e-04
 1280  1280  1280       42.76 (  0.10)     13.46 (  0.31)    2.746582e-04
 1600  1600  1600       42.99 (  0.19)     13.48 (  0.61)    3.280640e-04
 2000  2000  2000       39.73 (  0.40)     13.55 (  1.18)    4.882812e-04
 2500  2500  2500       38.87 (  0.80)     13.56 (  2.30)    7.781982e-04
 3125  3125  3125       39.64 (  1.54)     13.58 (  4.49)    9.918213e-04
 3906  3906  3906       39.00 (  3.06)     13.63 (  8.75)    1.312256e-03
 4882  4882  4882       39.59 (  5.88)     13.57 ( 17.15)    2.075195e-03
 6102  6102  6102    2950353.08 (  0.00)     13.60 ( 33.40)    4.695436e+02
Attachments
make.inc.txt
(1.38 KiB) Downloaded 159 times
Makefile.internal.txt
(2.01 KiB) Downloaded 157 times
mb_infn_it
 
Posts: 7
Joined: Wed Oct 10, 2012 4:39 am

Return to User discussion

Who is online

Users browsing this forum: Bing [Bot], Google [Bot] and 1 guest

cron