Re: faster version of AllocSetFreeIndex for x86 architecture

On Tue, 2009-06-02 at 23:53 +0900, Atsushi Ogawa wrote:

> I made a faster version of AllocSetFreeIndex for x86 architecture.

> Results of benchmark script:
> Xeon(Core architecture), RedHat EL4, gcc 3.4.6
>  bytes   :     4     8    16    32    64   128   256   512  1024   mix
>  original: 0.780 0.780 0.820 0.870 0.930 0.970 1.030 1.080 1.130 0.950
>  patched : 0.380 0.170 0.170 0.170 0.170 0.180 0.170 0.180 0.180 0.280

> The effect of the patch that I measured by oprofile is:
> - test program: pgbench -c 1 -t 50000 (fsync=off)
> 
> original:
> CPU: P4 / Xeon with 2 hyper-threads, speed 2793.55 MHz (estimated)
> Counted GLOBAL_POWER_EVENTS events
> with a unit mask of 0x01 (mandatory) count 100000
> samples  %        symbol name
> 66854     6.6725  AllocSetAlloc

> patched:
> CPU: P4 / Xeon with 2 hyper-threads, speed 2793.55 MHz (estimated)
> Counted GLOBAL_POWER_EVENTS events
> with a unit mask of 0x01 (mandatory) count 100000
> samples  %        symbol name
> 47610     4.9333  AllocSetAlloc

> I think this patch improves AllocSetAlloc/AllocSetFree performance.

Looks like very good work. Much appreciated.

-- Simon Riggs           www.2ndQuadrant.comPostgreSQL Training, Services and Support