| From: | Михаил Бахтерев <mob(at)k(dot)imm(dot)uran(dot)ru> |
|---|---|
| To: | Andrey Borodin <borodin(at)octonica(dot)com> |
| Cc: | amborodin(at)acm(dot)org, Heikki Linnakangas <hlinnaka(at)iki(dot)fi>, Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us>, pgsql-hackers <pgsql-hackers(at)postgresql(dot)org>, Alexander Korotkov <a(dot)korotkov(at)postgrespro(dot)ru> |
| Subject: | Re: GiST penalty functions [PoC] |
| Date: | 2016-09-09 06:41:40 |
| Message-ID: | 20160909064140.GA2506@kite |
| Views: | Raw Message | Whole Thread | Download mbox | Resend email |
| Thread: | |
| Lists: | pgsql-hackers |
Yes. You are right, ANSI C allows only load-time initializers. Attached
ANSI compatible version leads to the same assembly.
And let me suggest a bit-twiddling version as well. It gives 12
instructions, instead of 13. 12 is better, as modern x86 CPU will fetch
them at most in 3 cycles, one less than for 13 instructions. Also this
bit-twiddling is more parallel at instruction level.
And for ARM, which is unsurpassed at bit-twiddling this code is a way
better.
Of course speed is influenced by a lot of factors as always, so it needs
to be tested on some datasets.
- Mikhail, respectfully
On Fri, Sep 09, 2016 at 08:50:53AM +0500, Andrey Borodin wrote:
> Thank you for your attention to details, Mikhail.
>
> pack_float_good() looks good. But I'm not sure inline strict init is allowed under ansi C. Converting to regular ancient form b.fp = v; won't change compile result, would it?
>
> Regards, Andrey Borodin.
| Attachment | Content-Type | Size |
|---|---|---|
| pack-float.c | text/x-c | 1.2 KB |
| pack-float.s | text/x-asm | 2.2 KB |
| pack-float-arm.s | text/x-asm | 2.6 KB |
| From | Date | Subject | |
|---|---|---|---|
| Next Message | Kuntal Ghosh | 2016-09-09 07:01:48 | Re: WAL consistency check facility |
| Previous Message | Vik Fearing | 2016-09-09 06:23:06 | Re: Quorum commit for multiple synchronous replication. |