make tune fails on a Loongson processor

Torbjorn Granlund tg at gmplib.org
Mon Oct 18 17:41:22 CEST 2010


Sylvain CHEVILLARD <sylvain.chevillard at loria.fr> writes:

  I am encountering a problem when trying to perform a "make tune" with
  GMP 5.0.1 on a Loongson-2 processor (a kind of MIPS64).
  Namely, I performed my tests on the following machine (which you might
  have access): gcc42.fsffrance.org (on port 9092).
  
  * GMP version: 5.0.1
  * Test program:
      make
      cd tune
      make tune
  
  * I get a crash with the following message:
  ./tuneup
  Parameters for ./mpn/mips64/gmp-mparam.h
  Using: microsecond accurate getrusage()
  speed_precision 1000, speed_unittime 1.00e-06 secs, CPU freq 528.38 MHz
  DEFAULT_MAX_SIZE 1000, fft_max_size 50000
  
  /* Generated by tuneup.c, 2010-10-18, gcc 4.3 */
  
  Fatal error: too many (11) failed measurements (0.0)
  make: *** [tune] Aborted
  
I am afraid I also experience tuneup failures on this and other mips
processors that host any variant of the kernel Linux.

It is trying a use a claimed-accurate getrusage, but this fails.

It would be nice if somebody could find a solution, a cycle counter
handled through a new tune/mips.asm might work, but why isn't getrusage
working?  Is is a kernel problem (Linux's handling of anything with time
is poor)?  Could GMP be fixed to work anyway?

I am sorry for not being more helpful.

-- 
Torbjörn


More information about the gmp-bugs mailing list