#3814 Regression test failures on macppc

[Philipp Klaus Krause]

Since a few days, we get snapshots for macOS on PowerPC again. But there are regression test failures. AFAIK, the machine used is a Mac Mini G4. I also have an old Mac Mini G4, and just installed OpenBSD 7.6 on it, and also see regression test failures. The failures are reproducible, and a subset of the failures we see on MacOS on ppc. AFAIR, there is also some overlap with the remaining sporadic failures seen on Windows snapshots built from the nativecrosstools branch [bugs:#3797].

Apparently the test failures only affect the z80-related, hc08-related and mos6502-related ports.

AFAIK, macppc is the only 32-bit big-endian architecture for which current SDCC got tested (ppc64 is 64-bit, and all other archs are little-endian).

[Philipp Klaus Krause]

I tracked a z80 codegen bug as part of this issue, which is now fixed in the next branch in [r15215]. For me, test-ucz80 now passes on OpenBSD 7.6 on macppc.

[Philipp Klaus Krause]

At least some of the failures for the mos6502 port are due to an issue with binary or. Here is the simplest case I found so far (diff from asm generated on amd64 to asm generated on macppc):

        and     #0x0f   ;  A:U-:???  X:--:???  Y:--:???  F:a C:? 
-       ora     #0x20   ;  A:U-:???  X:--:???  Y:--:???  F:a C:? 
+; Peephole or 1 - eliminated unnecessary ora #0x00
        sta     _sb_bf  ;  A:U-:???  X:--:???  Y:--:???  F:a C:?

[Philipp Klaus Krause]

This part should be fixed in [r15222]: Regression tests on GNU/Linux amd64 pass (after lib rebuild) for uc6502 and uc65c02. I'm also running uc-6502 tests on OpenBSD on macppc now, and so far I see no failures (and thus some tests pass that failed before).
If we're lucky, the last bit of this bug is fixed now.

P.S.: Yes, this make test-uc6502 pass for me on OpenBSD/macppc.