avrx_notes.txt

建立一个属于自己的AVR的RTOS Proteus6.7 可以用来模拟仿真avr系列的单片机 WinAVR v2.0.5.48 基于GCC AVR的编译环境

Notes to myself:

6/20/02 - all suggestions, below, implemented as of 5/30/02  Actual performance
testing not done, but responding to an interrupt from IDLE should be about twice
as fast and SetSemaphore should be double to triple the speed when no or a lower
priority task has been queue (relative to the caller).

-------

AvrXSetSemaphore
	AvrXIntSetSemaphore works because "Running" isn't changed until Epilog.
	So setting a semaphore and queueing a pid doesn't affect the state until
	the running task is interrupted.  So, a further optimization would be:

	rcall	AvrXIntSetSemapore

	BeginCritical
	lds	tmp0, Running+NextL	; 16 cycles if no taskswap.
	lds	tmp1, Running+NextH
	lds	tmp2, RunQueue+NextL
	lds	tmp3, RunQueue+NextH
	cp	tmp0, tmp2
	cpc	tmp1, tmp3
	breq	exit
	rcall	IntProlog
	rjmp	_Epilog
exit:
	EndCriticalReturn

A better way would be to have some sort of flag set by QueuePid, as it "knows"
when something has changed the head of a queue.

In general, checking for task swap (e.g. top of queue changed) before doing actual
task swap...

This could be a general re-write of AvrX -> General scheduler in C.  Just use
regular C constructs and swap contexts as needed.  This would probably be pretty
fast at the expense of more task stack space needed.

Epilog/Prolog/IdleTask
    Have a bit somewhere that says we are in IDLE state and skip saving context
    for first entry into kernel or restoring context for 1->0 transistion.

Tasking, in general:
        Critical kernel data (Running, RunQUeue, Syslevel, AvrXC_stack (IAR only)
        could be a struct and all routines re-written to use that data via a pointer.
        This would make for more instructions to load/reload, but faster because LDD
        is only 2 cycles vs 3 for LDS.  

	ldi	Zl, lo8(_kernel)
	ldi	zh, hi8(_kernel)
	ldd
	...
	cp
	...

I have an experimental version of Avrx_tasking that uses this technique (not tested)
and it looks like a push.