We are currently trying to migrate our code from MKL 5.2 to MKL 8.0.1. Comparing the results we realized, that on functions that operate with sparse vectors (like zgthr() or zdotui_sub()) the index which select the entries to be used in x is starting from 0 in MKL 8.0.1 and from 1 in MKL 5.2.
For example, in order to access the first element of x in zgthr( nz, y, x, indx ), the first entry of indx would be 0 using MKL 8.0.1 and 1 using MKL 5.2.
We wrote a small program to demonstrate this behavior (see attachment). Its output is:
-------------------------
vector a:
1.44 + i*6.93
2.64 + i*10.23
3.84 + i*13.53
indices to be gathered from a:
1
2
Here are the results of MKL 5.2
vector b:
1.44 + i*6.93
2.64 + i*10.23
Here are the results of MKL 8.0.1
vector b:
2.64 + i*10.23
3.84 + i*13.53
-------------------------
We are using the following system:
gcc version 3.3.3 (SuSE Linux 9.1)
For MKL 8.0.1 we link:
libmkl
libguide
For MKL 5.2 we link:
libmkl_p3
libguide
Is there a compile flag or runtime variable which can be used to set the starting index 0 or 1 for sparse operations?
Thanks
Michael Junge
Message Edited by mdajunge on 02-10-2006 09:38 AM
