[parsec-users] Acceptable precision level for facesim log.txt output and new workload with different input is needed.
Raghavendra Swamy
raghu.heng at gmail.com
Sun Nov 11 14:14:51 EST 2012
Dear members,
1. Could you let me know the acceptable precision level difference for
facesim log.txt output, Here the comparison between two different compilers
outputs are noted.
a. gcc at -O level
b. pgi at -O2 level
>>
0157: Newton iteration residual after 5 iterations =
126.544 <-----gcc at -O level
Newton iteration residual after 5 iterations =
126.486 <-----pgi at -O2 level
^
0185: Newton iteration residual after 6 iterations = 124.302
Newton iteration residual after 6 iterations = 125.501
^
0213: Newton iteration residual after 7 iterations = 99.1778
Newton iteration residual after 7 iterations = 98.1436
^
0241: Newton iteration residual after 8 iterations = 95.2348
Newton iteration residual after 8 iterations = 95.0616
^
0269: Newton iteration residual after 9 iterations = 96.1028
Newton iteration residual after 9 iterations = 96.1557
^
0297: Newton iteration residual after 10 iterations = 81.6889
Newton iteration residual after 10 iterations = 82.48
^
0325: Newton iteration residual after 11 iterations = 72.2564
Newton iteration residual after 11 iterations = 70.0978
^
0353: Newton iteration residual after 12 iterations = 46.7435
Newton iteration residual after 12 iterations = 46.7358
^
0381: Newton iteration residual after 13 iterations = 47.5774
Newton iteration residual after 13 iterations = 47.4629
^
0409: Newton iteration residual after 14 iterations = 38.9028
Newton iteration residual after 14 iterations = 38.3283
<<
2. If anyone can point to different workload for facesim other than the
shipped one with the parsec/facesim package would be great. [ Specifically
looking for the input data workload which have a differing inputs.]. The
reason being, wanted to have a different frame for each computation
instead of same frame being computed 100 times and generating same output.
Note : Relevant reference
>>
PARSEC defines native input set intended for performance measurements on
real machines. In case of facesim this input set merely repeats the same
frame animation of the same model a hundred times. Since there is no change
of the model state or the input from frame to frame, the computations are
exactly the same in each of the 100 repetitions.
Reference link:
https://docs.google.com/viewer?a=v&q=cache:cFSuxsp_eEcJ:ppl.stanford.edu/cs315a/pub/Main/CS315a/facesim.pdf+facesim+pdf&hl=en&gl=in&pid=bl&srcid=ADGEESgaESfcpZRpgbUKYehIoYQa7LTEgKl3INNeY5KDnig5f1hKf3aMF1lySEVV2U2Qplju0yluhxrZfC1unqqtEiOjHG6GsmWHHbPCw6KiRXTMPo2GbAeieCHTwHon5bDmtxZM1-wB&sig=AHIEtbQa7Ww-T0kxYkW7qSfXSDjOtZDTRw
<<
Thanks and regards,
Raghu
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.cs.princeton.edu/pipermail/parsec-users/attachments/20121112/d71dc540/attachment.html>
More information about the parsec-users
mailing list