This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [4.0 PATCH, i386]: Fix PR target/20421


On Wed, Mar 16, 2005 at 08:54:36AM +0100, Uros Bizjak wrote:
> +;; Following rounding expanders clobber FLAGS_REG, when logic instructions
> +;; are emitted in emit_i387_cw_initialization () function.  The "if"
> +;; condition in this function and the "if" condition in the body of round
> +;; expanders should be kept in sync to select correct FLAGS_REG clobbering
> +;; pattern.
> +
> +(define_insn_and_split "frndintxf2_floor_1"

Any reason to not just add the clobber unconditionally, and let it
vanish in those cases where we don't wind up killing the flags?
Given that we do the split before sched2, it doesn't seem like we're
buying anything at all with this extra code.

Same comment applies to your mainline patch.


r~


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]