This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: 2.95, x86: severe performance problems with short arithmetic
- To: Zack Weinberg <zack at bitmover dot com>
- Subject: Re: 2.95, x86: severe performance problems with short arithmetic
- From: Richard Henderson <rth at cygnus dot com>
- Date: Tue, 10 Aug 1999 23:33:43 -0700
- Cc: gcc at gcc dot gnu dot org
- References: <199908101852.LAA11338@zack.bitmover.com>
On Tue, Aug 10, 1999 at 11:52:32AM -0700, Zack Weinberg wrote:
> /* Use a 32-bit operation when possible, to avoid the prefix penalty. */
> if (REG_P (operands[0])
> && i386_aligned_p (operands[2])
> && i386_cc_probably_useless_p (insn))
> {
The code in question is a performance improvement on P1 and
previous cpus. IMO, we should add a !TARGET_PENTIUMPRO to
this conditional -- as noted in a similar test in the block
above, partial reg stalls are just too painful to risk.
r~