tune&fat: run-time vs. compile-time constant thresholds
tg at gmplib.org
Sun May 6 11:51:34 CEST 2012
bodrato at mail.dm.unipi.it writes:
I pushed a few changes to the MAYBE_ mechanism of lower degree toom
functions, and to the ABOVE_THRESHOLD macro (only for GCC).
The effect of the change can be measured looking at Hydra's coverage report:
- before the change we had a total of 43613 branches
- after it, we have 42003; we avoided 1610 branches, most of them in mpn
Eh, that was a very impressive reduction, an order of magnitude or two
larger that I would have anticipated!
(Should really the Hydra coverage report include the test suite? I feel
that it just adds noise to the summary, since we certainly don't aim to
"cover" the code there. I hope its error printing code will remain
Let's see if there are other undesired effects. E.g. tuneup should be
affected, but I hope it will benefit of this changes too.
Why would tuneup be affected?
More information about the gmp-devel