On Tue, 23 Sep 2008, Szak�ts Viktor wrote:

Hi Viktor,
> Pls find attached some speedtst/memtst outputs.
> Looks like dlalloc is well suited for MSVC with
> a significant speedup in ST mode, and mostly
> better results in MT mode.
> Tested with MSVS2008 SP1, XPPro SP3, Intel 865 + P4HT 2.6, 3GB RAM.
[...]
> ascan(a,{|x|x==i%ARR_LEN}) ->               4.45     4.91    17.34    13.84
> =============================================================================
> total application time:                    32.52    44.42    96.91    92.02

I'm surprised with this results.
In ascan(a,{|x|x==i%ARR_LEN}) only two one small memory blocks are
allocated but the speed difference is huge.
Anyhow the MT overhead is too big to say sth more now.
Let's try to reduce it.
MSVC builds does not use native TLS support by default.
Please enable it be recompiling Harbour with HB_USE_TLS
macro. If native TLS supports works well in MSVC then
it should give noticeable speed improvement. Please send
the results here.
If everything works as expected and you can also test other MSVC
versions then please enable native TLS support in default MSVC
builds in hbthread.h

best regards,
Przemek
_______________________________________________
Harbour mailing list
Harbour@harbour-project.org
http://lists.harbour-project.org/mailman/listinfo/harbour

Reply via email to