On Feb 6, 2006, at 3:11 PM, Tomas Larsson wrote:

> You need precise timing (at least sort of) if you are supposed
> to program a PIC.

It's important to remember that "sort of" part.


>> Real-time control involving the host CPU, involves to flush all cash 
>> memory both data and instruction, and since today's cpus are heavily 
>> pipelined it takes quit long time to do this, even on a 3000+ cpu.

IO instructions to control external bits are quite slow.  The
pipeline gets stalled, all the out-of-order execution stuff has
to come to a halt, and other stuff slows.  I don't THINK that
caches are flushed; cache coherency is handled separately and
the IO space and memory space are separate.  But an IO instruction
to set a UART register takes literally hundreds to thousands of
cycles on a machine that normally executes somewhat more than one
instruction in each cycle, so it's incredibly inefficient.

However, that's still ~1000 cycles at 3GHz, which is still
significantly less than a microsecond, which is plenty small
a time interval for programming PICs.  (But you'll not that
it's not much faster than a PIC can twiddle pins... :-)
And no one really cares how inefficient a computer is these
days, as longs as it's fast...

Of more serious concern is that sometime during that 50uS pulse
you wanted, something like a legacy USB bios or power management
interrupt will come along and do something (or nothing) for 30uS
or so and screw up everything...

BillW

-- 
http://www.piclist.com PIC/SX FAQ & list archive
View/change your membership options at
http://mailman.mit.edu/mailman/listinfo/piclist