Confusion about the check of incx and incy
Posted:
Fri Nov 24, 2006 5:59 amby ttrr
I found in BLAS 1, you don't check the incx and incy and give the correct result. But in BLAS 2, you check the incx and incy and if incx or incy is zero, stop and return.
Can you tell me why you act like this?
and in the future, will you support the zero case of incx and incy in BLAS 2?
Thanks!
Posted:
Fri Nov 24, 2006 7:17 amby sven
Dear ttrr,
In the specification for the Level 2 BLAS, zero increments are considered invalid. See:
@ARTICLE{DDHH:TOMS:88,
AUTHOR = {Dongarra, J. J. and Du Croz, J. and Hammarling, S. and
Hanson, R. J.},
TITLE = {An Extended Set of {FORTRAN Basic Linear Algebra
Subprograms}},
JOURNAL = j:TOMS,
YEAR = {1988},
VOLUME = {14},
PAGES = {1--32, 399} }
pages 7 and 8. This is also true for the more recent specifications, see:
@ARTICLE{BDDDHHHKLPRRW:TOMS:02,
AUTHOR = {Blackford, L. S. and Demmel, J. and Dongarra, J. J. and
Duff, I. S. and Hammarling, S. and Henry, G. and Heroux,
M. and Kaufman, L. and Lumsdaine, A. and Petitet, A. and
Pozo, R. and Remington, K. and Whaley, R. C.},
TITLE = {An Updated Set of {Basic Linear Algebra Subprograms
(BLAS)}},
JOURNAL = j:TOMS,
YEAR = {2002},
VOLUME = {28},
PAGES = {135--151} }
I hope that clarifies the matter (even if you do not like the decision!). Best wishes,
Sven Hammarling.