Add lazy register push to PAUSE. Thats a big win when you usually don…

…'t switch tasks.

pause.fth

@@ -1,46 +1,71 @@
-\ This version of the scheduler has three mods - 
-\ Counting task runs with a TCB run counter
-\ Using bit-banded addresses for atomic status word updates
-\ It issues a WFI if there are no runnable tasks.
+\ An improved scheduler.  Its intended for a system that is interrupt-driven
+\ and spends most of it's time asleep waiting for an interrupt that may or
+\ may not may not make a task runnable.
+\ The typical workload is wfi -> Look for a runnable task -> run it.
+\ In a system that mostly waits around a responds to timer ticks, we'll tend
+\ to run the same tasks over and over.  So saving state is often a waste of time.
+
+\ This version of the scheduler has mods - 
+\ 1.  Counting task runs with a TCB run counter
+\ 2.  Using bit-banded addresses for atomic status word updates
+\ 3.  It issues a WFI if there are no runnable tasks.
+\ 4.  Don't bother doing the single-multi check.
+\ 5.  Implement lazy register push.  Only push registers if switching.
+
+\ Switching to this approach from a dedicated idle task resulted in the 
+\ same CPU Utilization, but better responsiveness.
 
 CODE pause	\ -- ; the scheduler itself
 \ *G The software scheduler itself.
-l: [schedule]
-  mvl     r0, # ' multi? >body		\ inspect contents of MULTI?
-  ldr     r1, [ r0 ]
-  cmp     r1, # 0			\ single tasking if 0
-  it .eq
-    bx      lr				\ single tasking so exit
-\
-\ Save the task state
-\
-  push    { r7, r9, r12, link }		\ stack used registers ; SFP001
-  str     rsp, [ up, # 0 tcb.ssp ]	\ save SP in TCB
+\ Register Usage 
+\  R0 Scratch
+\  R1 Scratch
+\  R5 Status word of current task under consideration
+\  R6 Saved UP
+\ 
+\ l: [schedule]
+\  mvl     r0, # ' multi? >body		\ inspect contents of MULTI?
+\  ldr     r1, [ r0 ]
+\  cmp     r1, # 0			\ single tasking if 0
+\  it .eq
+\    bx      lr				\ single tasking so exit
+
+  mov r6, up   \ Remember who called PAUSE. 
 
-  mov r2, up   \ Remember.  We will loop around the task ring.
 \
 \ select next task to run
 \
 l: [schedule]next
-  ldr     up, [ up, # 0 tcb.link ]	\ get next task
-  ldr     r3, [ up, # 0 tcb.status ]	\ inspect status
-  cmp     r3, # 0			\ 0 = not running
+  ldr     r6, [ r6, # 0 tcb.link ]	\ get next task
+  ldr     r5, [ r6, # 0 tcb.status ]	\ inspect status
+  cmp     r5, # 0			\ 0 = not running
   b .ne   [schedule]run
 
-  cmp     up, r4 \ If we made it around the ring without running, WFI.
+  cmp     r6, up \ If we made it around the task list without running, WFI.
   it .eq 
     wfi
   b  [schedule]next
 
 l: [schedule]run
 
+  \ First order of business - If we've circled around, just return
+  cmp r6, up
+  it .eq 
+    bx lr 
+
+  \ Save the prior task state
+  push    { r7, r9, r12, link }		\ stack used registers ; SFP001
+  str     rsp, [ up, # 0 tcb.ssp ]	\ save SP in TCB
+
+  mov up, r6 \ Load up the new task pointer.
+
   \ We know who it is.   Load up the MPU entries.
   add r0, up, # 0 tcb.mputab   \ Calculate the position of the mpu table.
   svc # SAPI_VEC_MPULOAD       \ Install it.
 
-  ldr r1, [ up, # 0 tcb.runcount ]      \ Bump the run counter 
-  add r1, # 1 
-  str r1, [ up, # 0 tcb.runcount ]
+  ldr r0, [ up, # 0 tcb.runcount ]      \ Bump the run counter 
+  add r0, # 1 
+  str r0, [ up, # 0 tcb.runcount ]
 
 \
 \ run selected task - sp, up, rp, ip
@@ -52,7 +77,7 @@ event-handler? [if]	\ if user wants the event handler
 \
 \ event handler exit
 \
-  and .s  r1, r3, # trg-mask		\ inspect event trigger bit
+  and .s  r0, r5, # trg-mask		\ inspect event trigger bit
   ne, if,
     ldr     r0, [ up, # 0 tcb.bbstatus ]
     mov     r1, # 0