Re: [Harbour] Darwin: 64-bit problem

[Szakáts Viktor]

Hi Przemek,

Yes, Thank you.
The most visible is the cost of mutexes in reference counters.
If there is no collision then all is executed quite fast:

ST mode (no mutexes at all):
[ T001: x := L_C ]..............................................0.07

MT mode and single thread:
[ T001: x := L_C ]..............................................0.13

Such cost is high though still accpetable but this:


shows that concurrent access to the same complex item is too slow.

[ T001: x := L_C ]____________________________________ 0.67 18.20 - > 0.04 We will have to add atomic inc/dec assembler inline functions to PPC builds.

Also to Intel 64-bit builds I guess. (test above was x86/64-bit)

[ I've yet to create a PPC 32-bit speed/MT test. ]

Anyhow in DARWIN you should have <libkern/OSAtomic.h> with set of functions

for atomic operations. Can you look for 32 and 64 bit atomic inc/dec?
They can be used by us directly.

Please see the whole file attached, there are multiple
versions of these functions.

Brgds,
Viktor

OSAtomic.h
Description: Binary data

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