"Notes on SDCC" on website about sduino

[tenbaht]

Hi Philipp,

issue 1

Unfortunately, I didn't keep the files when I found that problem. I can't find any examples anymore, so I assume it was fixed in one of the later compiler revisions.

issue 2

minimal example code:

int array[4];
void array_access(unsigned char index)
{
    array[index] = 0;
}

compiled using the latest snapshot build 9998:

$ /opt/sdcc/bin/sdcc -c -mstm8 issue14-2.c
$ /opt/sdcc/bin/sdcc -v
SDCC : mcs51/z80/z180/r2k/r3ka/gbz80/tlcs90/ds390/pic16/pic14/TININative/ds400/hc08/s08/stm8 3.6.9 #9998 (Linux)
published under GNU General Public License (GPL)

results in this code:

      000000                         47 _array_access:
      000000 52 02            [ 2]   48 	sub	sp, #2
                                     49 ;	issue14.c: 6: array[index] = 0;
      000002 AEr00r00         [ 2]   50 	ldw	x, #_array+0
      000005 1F 01            [ 2]   51 	ldw	(0x01, sp), x
      000007 7B 05            [ 1]   52 	ld	a, (0x05, sp)
      000009 97               [ 1]   53 	ld	xl, a
      00000A A6 02            [ 1]   54 	ld	a, #0x02
      00000C 42               [ 4]   55 	mul	x, a
      00000D 72 FB 01         [ 2]   56 	addw	x, (0x01, sp)
      000010 6F 01            [ 1]   57 	clr	(0x1, x)
      000012 7F               [ 1]   58 	clr	(x)
      000013 5B 02            [ 2]   59 	addw	sp, #2
      000015 81               [ 4]   60 	ret

When I look at it I see that the mul command already includes the transfer of the result into x which needs to happen anyway. It would be great if the temporary variable (0x01,sp) could be avoided in the first place.

issues 3-5

Thank you for your explanation!

I noticed the recent changes in the backend, the optimizer and the simulator. I would like to compare the development of the code size compared to an older version. Thank you for all the work there!

Michael

[spth]

Issue 2: Thanks. I can see how using sllw could save 1 byte and 2 cycles vs the current solution using mul. I'll probably look into implementing it next week.

Philipp

P.S.: Eliminating the temporary at (1,sp) would take more effort, since it needs an optimization capable of folding ADDRESS_OF into '+'.

[spth]

As of revision #10000, a sllw is used instead of the mul. The mcs51 backend can optimize out the temporary that holds the array address; I'll look into porting the solution to stm8 (and maybe z80).

Philipp

P.S.: See also https://sourceforge.net/p/sdcc/feature-requests/536/

[tenbaht]

sllw works with rev. 10088 and saves one byte.

I am looking forward to see the temporary optimized out, that could save 7 bytes on every array access. I will follow the discussion on SF, no need to duplicate it here.