User talk:Pzgesvd: Difference between revisions
Jump to navigation
Jump to search
(low cost stack for function call?) |
(clapack3.1.1 blas testing handtune) |
||
| Line 29: | Line 29: | ||
can the stacking operation of the dag chasing be reduced so the overhead is minimized? | can the stacking operation of the dag chasing be reduced so the overhead is minimized? | ||
== clapack3.1.1 blas testing handtune == | |||
in the testing routines, eps is wrongly calculated as 1e-19, which should be 1e-7. In each files (dblat3.c, for example), a new piece of eps code is inserted. | |||
For double precision (complex16), problem exists for not accurate element result. | |||
Revision as of 19:58, 8 September 2008
bidiagonal reduction code function matching
DGEQRT --- LQR1
DTSQRT --- LQR2
DLARTB --- LUP1
DSSRFT --- LUP2
DGEQRT --- RQR1
DTSQRT --- RQR2
DLARTB --- RUP1
DSSRFT --- RUP2
delete lines from files using sed
sed -ie '1,11d' dgetrf.c
original files are backed up in the .ce files
refer to [1]
low cost stack for function call?
can the stacking operation of the dag chasing be reduced so the overhead is minimized?
clapack3.1.1 blas testing handtune
in the testing routines, eps is wrongly calculated as 1e-19, which should be 1e-7. In each files (dblat3.c, for example), a new piece of eps code is inserted.
For double precision (complex16), problem exists for not accurate element result.