Execf call skip behaviour
macca
Posts: 806
in Propeller 2
Hello,
I have an issue with a program that uses execf with call instructions.
See this example:
CON _clkfreq = 160_000_000 DAT org $000 call #test_execf call #test_execf_nop jmp #$ test_execf debug(udec(test_execf)) push #$+2 execf op1 debug(udec(t1)) push #$+2 execf op2 debug(udec(t1)) push #$+2 execf op3 debug(udec(t1)) ret test_execf_nop debug(udec(test_execf_nop)) push #$+2 execf op1_nop debug(udec(t1)) push #$+2 execf op2_nop debug(udec(t1)) push #$+2 execf op3_nop debug(udec(t1)) ret op1 long op | %1_1_0 << 10 op2 long op | %1_0_1 << 10 op3 long op | %0_1_1 << 10 op1_nop long op_nop | %01_01_00 << 10 op2_nop long op_nop | %01_00_01 << 10 op3_nop long op_nop | %00_01_01 << 10 op call #sub1 call #sub2 call #sub3 ret op_nop call #sub1 nop call #sub2 nop call #sub3 nop ret sub1 _ret_ mov t1, #1 sub2 _ret_ mov t1, #2 sub3 _ret_ mov t1, #3 t1 res 1 t2 res 1
This is an image of the debug output:
You can see that, apparently, contiguous calls can't skipped but if there is another instruction in between (a nop in this case) the skip pattern works.
Is that the correct behaviour or I'm missing something ?
Comments
CALL must use absolute addressing with SKIPF/EXECF, e.g CALL #\routine or CALL reg, if the instruction after the call might be skipped. See "Special SKIPF Branching Rules" in the doc.
It is safest to use absolute addressing for every CALL in a SKIPF sequence. I do that for non-skipping CALLs, too, just in case they become part of a skip sequence later.
Isn't that what I'm using ?
No, just noticed the backslash !!!
(some rants and really bad words deleted before posting)
New postit: P2 CALL/JMP ARE RELATIVE !!!
Thanks for your help!