[PATCH] Reduce cost of Athlon multiplication sequences
Roger Sayle
roger@eyesopen.com
Thu Jan 5 18:21:00 GMT 2006
On Thu, 5 Jan 2006, Richard Henderson wrote:
> Even if we don't add the cost check, I wonder if we should
> be checking TARGET_DECOMPOSE_LEA here...
Interestingly, that's the follow-up patch that I mentioned in
my posting. Currently, TARGET_DECOMPOSE_LEA has absolutely no
effect in the i386 backend, other than to tweak the cost of
ix86_cost->lea in ix86_rtx_costs. This of course, is fairly
pointless, as the subtargets can now directly control the
ix86_cost->lea themselves.
My follow-up patch was therefore to remove all vestiges of
TARGET_DECOMPOSE_LEA and x86_decompose_lea from the config/i386
directory. Instead, the one place where it currently used
instead falls back to the "ix86_cost->lea <= ix86_cost->shift_const"
on the following line.
Roger
--
More information about the Gcc-patches
mailing list