[PATCH] T3/T4 sparc shifts, plus more timings

Torbjorn Granlund tg at gmplib.org
Sun Mar 31 22:57:36 CEST 2013

David Miller <davem at davemloft.net> writes:

  I'm going to play around with some things to try and fix this.
  Interestingly, UltraSPARC-1 and UltraSPARC-2 would not group the
  final cycle of the loop this way, because of it's requirement that
  integer operations must occur in the first three instructions of
  a group.
But the code might not run well on US1/2 because shift needs to come
before any other inter artithmetic op.

I think I'll commit the code to the top-level sparc64 directory now, then
remove ultrasparct1/[rl]shift{c,}.asm and but leave the files in
ultrasparc1234 for now.

US1/2 is not very important, but we should check the lshift and rshift
performance before we remove the special ultrasparc1234 code, not to
cause outright slowdown.

Reducing the function variants is a nice thing, but it is not the most
important goal.


More information about the gmp-devel mailing list