1. Ports
- Posted by Robert Craig <rds at EMAIL.MSN.COM>
Jan 03, 1998
-
Last edited Jan 04, 1998
Peter Blue and Ralf have suggested that Port I/O
routines be added to ex.exe.
I'm wondering if there are any applications
(real or imagined) that would need better
port I/O speed than is provided in Jacques Deschenes
ports.e.
I timed his 8-bit Input() routine on a 486 DX-50.
I wanted to know the extra "overhead" from
executing his Euphoria statements.
I measured (simulated) how long it would take to read
56K bits (at 8-bits per read). I concluded it would
be about 0.32 seconds. (I left out the actual call() that does the
port access - it should be pretty fast and would be
needed in an ex.exe implementation anyway.).
If you replace:
poke(InputAsm+3, int_to_bytes(port))
with:
poke4(InputAsm+3, port)
(poke4 is new in 2.0 alpha)
the time drops to just .08 seconds. (Remember to say:
"without type_check" - he has some user-defined types.)
You can reduce the time a bit further by removing
his test:
if not InputAsm then
return -1
end if
You could also avoid adding 3 to InputAsm each time
etc.
So my feeling is that you could handle a 56K / sec. modem
and only 8% or less of the time would go into
"Euphoria overhead for Port I/O". With the remaining 92%
you could examine the incoming characters, store them
in a sequence etc. Actually, if you need to handle
hardware interrupts at a very high rate, that could be your
ultimate downfall - not the port I/O.
I like to avoid putting stuff into ex.exe (exw.exe)
unless it is used by a lot of programs and performance
is critical. I was tempted to add ports before, but there
was always something higher up on my wish list.
Regards,
Rob Craig
Rapid Deployment Software
