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

Torbjorn Granlund tg at gmplib.org
Sun Mar 31 05:03:10 CEST 2013


David Miller <davem at davemloft.net> writes:

  I'd suggest checking this code in at this point.

I decided to leave lshift, rshift, and lshiftc all separate for now.
But I whacked the post loop code somewhat more.

The lshiftc code runs at 3 c/l on US3, not the claimed 2.5 c/l.  I
suspect also the US1 claim if 2 c/l is invalid.

It is not hard to see why it is 3 c/l.  I haven't tried swapping things
around to fix this, since that would risk hurting the T3/T4 performance.

Perhaps you could look into that?

My current versions of the files are attached. 

-------------- next part --------------
A non-text attachment was scrubbed...
Name: sparc64-lshift.asm
Type: application/octet-stream
Size: 2681 bytes
Desc: not available
URL: <http://gmplib.org/list-archives/gmp-devel/attachments/20130331/6d382127/attachment.obj>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: sparc64-rshift.asm
Type: application/octet-stream
Size: 2705 bytes
Desc: not available
URL: <http://gmplib.org/list-archives/gmp-devel/attachments/20130331/6d382127/attachment-0001.obj>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: sparc64-lshiftc.asm
Type: application/octet-stream
Size: 2768 bytes
Desc: not available
URL: <http://gmplib.org/list-archives/gmp-devel/attachments/20130331/6d382127/attachment-0002.obj>
-------------- next part --------------


-- 
Torbj?rn


More information about the gmp-devel mailing list