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: Patch: Scheduler support for IA-64 speculation [5/5]


Ian Lance Taylor wrote:
Maxim Kuvyrkov <mkuvyrkov@ispras.ru> writes:

This patch adds support for IA-64 data/control speculation to haifa
scheduler.

I haven't really looked at this patch, but a few quick notes.


This patch doesn't include documentation patches for the new target
hooks or the new parameters.
There is no separate documentation for *any* of the target hooks. Still,
comments for the new hooks present in target.h. As for the two new
parameters, submitted patch contains hunks for patching
gcc/doc/invoke.texi .

The patch is pretty substantial. Can you say something about how it affects scheduling for non-ia64 targets?
For i686-pc-linux-gnu user times are:

bootstrap (c, c++):
mainline: 47m18s
patched:  47m42s (+0.8%)

make check:
mainline: 58m30s
patched:  58m32s

SPECint base compilation (excluding gcc and eon - they fail to compile):
mainline: 3m16s
patched:  3m15s

SPECint peak compilation (excluding gcc and eon):
mainline: 3m55s
patched:  3m54s

Why does the speculate_insn hook return int rather than bool?
First, I don't like bool. And, the important one, second: this hook is
used in conjunction with function haifa-sched.c:speculate_insn (). This function can return three different values, two of which have the same meaning as values returned by this hook.



-- Maxim




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