C implementation of mod_1_1

Torbjorn Granlund tg at gmplib.org
Wed Mar 2 15:18:48 CET 2011


nisse at lysator.liu.se (Niels Möller) writes:

  That should lower overhead on those targets. But it doesn't explain
  lowered thresholds in, e.g., mpn/powerpc64/mode64/p5/gmp-mparam.h...
  
I completely agree.  :-)

  > Perhaps we should suppress the measuring, or at least avoid putting
  > ignored parameters in the gmp-mparam.h files?
  
  I think there's some value in measuring all available variants and warn
  for anomalies such as an assembly implementation not being fastest. But I
  have no strong feelings either way.
  
I think various sanity measuring makes a lot of sense:

(1) check that the chosen code is not slower than any alternative
    implementation of the same function

(2) check that the chosen code is not slower than any implementation of
    a more powerful function (such as addmul_1 being faster than
    addlsh*_n)

(3) check that we don't cause short-time or long-term slowdown of any
    function for any operand size

I don't think any of these results should be put into tuneup.c and
reported as for inclusion gmp-mparam.h, though.

-- 
Torbjörn


More information about the gmp-devel mailing list