[PATCH, MIPS] Use dmult in 64-bit signed widening multiplication

Richard Sandiford rdsandiford@googlemail.com
Thu Oct 15 18:49:00 GMT 2009


Adam Nemet <anemet@caviumnetworks.com> writes:
> Richard Sandiford writes:
>> Adam Nemet <anemet@caviumnetworks.com> writes:
>> > Taking this one step further I am not sure there is any benefit of
>> > implementing mulsidi3 for 64-bit with DMULT.  We're really just reiterating
>> > what's already optimized in the backend that sign-extension is a move.
>> > I.e. if we FAILed in the expander for this case, the sign-extension on the
>> > input operands should be removed most of the time.  I'll benchmark this a
>> > little more.
>> 
>> Yeah, I wondered about suggesting that too, but when I last looked
>> at it, we weren't getting rid of sign extensions as often as we'd like.
>> That was with the old register allocators though; IRA could well have
>> made things better.
>
> Yes, it seems the difference is still not negligible on CSiBE:

Oh well!  Thanks for testing it.

Richard



More information about the Gcc-patches mailing list