| From: | Thomas Lockhart <lockhart(at)fourpalms(dot)org> |
|---|---|
| To: | Bruce Momjian <pgman(at)candle(dot)pha(dot)pa(dot)us> |
| Cc: | Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us>, Tatsuo Ishii <t-ishii(at)sra(dot)co(dot)jp>, pgsql-hackers(at)postgresql(dot)org |
| Subject: | Re: Intermediate report for AIX 5L port |
| Date: | 2001-12-13 15:29:22 |
| Message-ID: | 3C18C952.D2460588@fourpalms.org |
| Views: | Raw Message | Whole Thread | Download mbox | Resend email |
| Thread: | |
| Lists: | pgsql-hackers |
...
> It can see the lock values don't have any effect on 'a'. What actually
> does keep this stuff from moving around?
Lack of ambition?
I'm pretty sure that the only reasons *to* reorder instructions are:
1) there could be a performance gain, as in
a) loop unrolling
b) pipeline fill considerations
c) unnecessary assignment (e.g. result is ignored, or only used on one
path)
2) the optimization level allows it (-O0 does not reorder at all)
I vaguely recall that the gcc docs discuss the kinds of optimizations
allowed at each level. Presumably IBM's AIX compiler was a bit more
aggressive in evaluating costs or pipeline fills than is gcc on other
processors.
- Thomas
| From | Date | Subject | |
|---|---|---|---|
| Next Message | Patrick Welche | 2001-12-13 15:29:57 | Re: Bulkloading using COPY - ignore duplicates? |
| Previous Message | Thomas Lockhart | 2001-12-13 15:16:15 | Re: [HACKERS] Platform Testing - Cygwin |