Google Groups no longer supports new Usenet posts or subscriptions. Historical content remains viewable.
Dismiss
UPC Timing on Millennium
0 views
Skip to first unread message
gennari
Mar 27, 2004, 1:24:42 AM3/27/04
to
When I compile and run knap-upc-mpi on millennium, the timing functions
print out large and potentially negative numbers for the runtimes. I haven't
changed the code at all, and it seems like the program is not properly
writing to the timing struct data fields. Is this some struct alignment
problem/bug in UPC? Maybe UPC is compiling using 64-bit integer timing in
the header and 32-bit integers in a library? Anyone else see this problem
(I'm not sure which other programs use timer.c and timer.h)?
print out large and potentially negative numbers for the runtimes. I haven't
changed the code at all, and it seems like the program is not properly
writing to the timing struct data fields. Is this some struct alignment
problem/bug in UPC? Maybe UPC is compiling using 64-bit integer timing in
the header and 32-bit integers in a library? Anyone else see this problem
(I'm not sure which other programs use timer.c and timer.h)?
Frank
Edward Jason Riedy
Mar 29, 2004, 2:39:47 PM3/29/04
to
And Frank Gennari writes:
- [...], and it seems like the program is not properly
- writing to the timing struct data fields. Is this some struct alignment
- problem/bug in UPC?
- [...], and it seems like the program is not properly
- writing to the timing struct data fields. Is this some struct alignment
- problem/bug in UPC?
It's setting both the tv_sec and tv_usec fields to the value
written into tv_usec. Great. Looks like the "older" upc
cannot handle nested structures. (They were using an "unstable"
translator which handled it correctly, but switched back and I
didn't test after the switch. My fault. urgh.)
Updated the code, and it seems to work.
http://www.cs.berkeley.edu/~ejr/GSI/cs267-s04/homework-3/knap/knap/timer.c
http://www.cs.berkeley.edu/~ejr/GSI/cs267-s04/homework-3/knap/knap/timer.h
Jason
--
0 new messages