Posted on October 18, 2017 at 06:23Ok. In my opinion, the main problem of the SDCC for STM8 is too wasteful use of the stack. I wanted to show you an example, but in the comment field the text is heavily distorted, and there is no possibility of transferring the attachment. In is snippet:
0001F9 460 00104$:
461 ; /home/master/Desktop/kernel.c: 134: base.counter++;
0001F9 1E 07 [ 2] 462 ldw x, (0x07, sp)
0001FB 5C [ 1] 463 incw x
0001FC 16 03 [ 2] 464 ldw y, (0x03, sp)
0001FE 90 FF [ 2] 465 ldw (y), x
466 ; /home/master/Desktop/kernel.c: 135: }
000200 5B 0A [ 2] 467 addw sp, &sharp10
000202 81 [ 4] 468 ret
; sdcc -c -mstm8 --peep-file /home/master/peeph.def -max-allocs-per-node --opt-code-size %f
; SDCC : mcs51/z80/z180/r2k/r3ka/gbz80/tlcs90/ds390/pic16/pic14/TININative/ds400/hc08/s08/stm8 3.6.9 &sharp10062 (Linux)
Variable base it is simple static structure, but SDCC for some reason, it generates indirect addressing, storing pointer variables on the stack. Saving RAM is very important, for micro-controllers. I would like to have a stack saving mode. The second problem concerns prologues and epilogues of system functions declared as __interrupt and __critical. It would be desirable to have, what or possibility, independently to describe their code and insert it before automatic reservation of a stack. For example, I need use 'push cc\n push &sharpmylevel\n ... pop cc\n ret\n' instead 'sim\n ... rim\n ret\n'. Also, in the inline assembler, the meta-constants indicating offsets of the variables in the stack would be very useful. I hope that soon, this compiler will become good enough for real use.
Vyacheslav
