Messages in this thread | | | Date | Wed, 19 May 1999 23:03:16 +0200 | From | Kurt Garloff <> | Subject | Re: Kernel checksum routime (>=5)x86 optimization |
| |
On Tue, May 18, 1999 at 07:22:24AM +0200, Jamie Lokier wrote: > Kurt Garloff wrote: > > Well, I hacked a test. It seems my theory fails. Can someone explain, why > > the PPro does not perform better using two regs? > > Because each instruction still depends on the previous one: > `adcl' means to use the carry flag from the previous instruction.
Well, it does not in my test proggie.
By the way: Is it correct to use adcl in this piece of code??? Why do we want to add 1 instead of 1<<32 in case of overflow???
> For instructions this simple the integer units are probably faster than > the instruction decoder anyway.
That might be the reason, yes. it might also be that only one load can be executed per cycle.
> You might get better results with the PIII's 128-bit SIMD instructions ;-)
-- Kurt Garloff <garloff@suse.de> SuSE GmbH, Nürnberg, FRG Linux kernel development; SCSI driver: DC390 (tmscsim/AM53C974) [unhandled content-type:application/pgp-signature] | |