This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: RFC: Prefetch instruction support
- To: Michael Meissner <meissner at cygnus dot com>
- Subject: Re: RFC: Prefetch instruction support
- From: Richard Henderson <rth at cygnus dot com>
- Date: Tue, 11 Apr 2000 22:20:00 -0700
- Cc: Jan Hubicka <hubicka at atrey dot karlin dot mff dot cuni dot cz>, egcs at egcs dot cygnus dot com, alex dot dreyzen at amd dot com, norbert dot juffa at amd dot com
- References: <20000411164306.B12560@atrey.karlin.mff.cuni.cz> <20000411190853.56561@cse.cygnus.com>
On Tue, Apr 11, 2000 at 07:08:53PM -0400, Michael Meissner wrote:
> (insn 92 28 30 (prefetch (mem:SI (plus:SI (plus:SI (mult:SI (reg/v:SI 26)
> (const_int 4))
> (reg:SI 29))
> (const_int 128)))))
That might be helpful.
For testing, Jan, you might have better luck with (use (unspec [])),
than with an unspec_volatile. The later severely inhibits scheduling,
which you don't want.
Also, you might want to get rid of the mem, since you probably don't
want the prefetch to restrict the schedule of a store. Retain only
the address in the pattern via
(match_operand:QI N "address_operand" "p")
or such.
r~