On 18/02/2021 15:32, Royce Pereira (T-star Instrumentation) wrote:
Its curious why the "PD_ODR |= 1<<1" produced the "ld a ...." in the
1st example,
while just the 'bset ...' in the second, when both cases were absolute
addresses.
The answer is simple: SDCC does not currently have any peephole
optimiser rules that cater to the first example.
You can see what rules are applied for optimising to 'bset/bres/bcpl'
instructions by looking at src/stm8/peeph.def file in the SDCC source.
Rules 18-*, 19-*, and 20-*, if memory serves me correctly.
You can easily formulate some additional rules for yourself and create a
custom .def file that can be specified at compile-time with --peep-file
argument.
For your example, a rule like the following may be appropriate (note:
untested!):
replace restart {
ld a, %1+%3
or a, #0x02
ld %1+%3, a
} by {
bset %1+%3, #1
} if notUsed('a'), notUsed('n'), notUsed('z'), operandsLiteral(%3)
The rule would need to be repeated to cover every bit position - e.g.
#0x01, #0x02, #0x04, etc. for 'or' instruction second operand; #0, #1,
#2, etc. for 'bset' second operand.
Regards,
Basil Hussain
_______________________________________________
Sdcc-user mailing list
Sdcc-user@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/sdcc-user
