Page 2 of 2
Posted: Thu Mar 04, 2004 3:26 pm
by coma
if was an example to show how things may be "curious".
sometimes, the empty loop take more time than the other one.
Posted: Thu Mar 04, 2004 3:35 pm
by Psychophanta
coma:
result for your last question was: 156 / 109 , other times 140 / 110 ...
Freak:
Yes, i know, but i prefer real ASM better than inlined
Posted: Thu Mar 04, 2004 4:13 pm
by Dare2
hehe. I had the empty loop come 2nd a few times as well.
Deduction - Bloat your code! It may run faster!
Posted: Thu Mar 04, 2004 4:27 pm
by techjunkie
Psychophanta wrote:result is: 110 / 94 / 125 / 46
The result on my Dell Dimension 8200 (P4 2GHz) is,
203 / 141 / 1140 / 32
Why is the t2 example so slow? Hmmmm...
Posted: Thu Mar 04, 2004 7:05 pm
by Psychophanta
techjunkie wrote
Why is the t2 example so slow? Hmmmm...
I don't know, but it seem one more of the Intel incoherences; now with XCHG instruction, and AMD seem to be less incoherent with this instruction.
Have you tried Intel i386 LOOP asm instruction? it is a shame
Posted: Thu Mar 04, 2004 7:17 pm
by coma
XCHG eax,ebx is very fast.
it's "XCHG eax,mem" that is slow.
Posted: Thu Mar 04, 2004 9:11 pm
by Psychophanta
XCHG eax,ebx is very fast.
it's "XCHG eax,mem" that is slow.
Which demonstrates still more incoherence
Posted: Fri Aug 28, 2009 12:22 pm
by Trond
XHCG with memory operand always asserts a bus lock to make the instruction atomic. That's why it's slow.
Since we have the new structure assignment, I request this feature to work for structured types as well.
Posted: Sun Aug 30, 2009 10:39 am
by talisman
Trond wrote:XHCG with memory operand always asserts a bus lock to make the instruction atomic. That's why it's slow.
Since we have the new structure assignment, I request this feature to work for structured types as well.
...and your reply came in just 5 years slow
Posted: Sun Aug 30, 2009 10:57 am
by DarkDragon
talisman wrote:Trond wrote:XHCG with memory operand always asserts a bus lock to make the instruction atomic. That's why it's slow.
Since we have the new structure assignment, I request this feature to work for structured types as well.
...and your reply came in just 5 years slow
He wants it for structured types
. So thats the right topic.
Posted: Tue Sep 01, 2009 5:38 am
by John Puccio
Trond wrote:Since we have the new structure assignment, I request this feature to work for structured types as well.
+1 @Trond's request. I think that would be really neat!
Posted: Tue Sep 01, 2009 8:22 am
by blueznl
Me too!