Hello Philipp, Monday, December 27, 2021, 12:12:08 PM, you wrote:
Oh, THANK YOU SO MUCH for clarification! So, __banked forces __sdcccall(0), that's nice. I am starting with the GBDK-2020 library update now, so, i will write if i find some problems. PKK> What SDCC currently does for z80 and related: PKK> 1) __banked can be combined with either __sdcccall(0) or __sdcccall(1). PKK> 2) Calls to __sdcccall(0) or __sdcccall(1) __banked functions always pass all parameters on the stack. PKK> 3) Calls to __banked __z88dk_fastcall function pass the only 8/16/32 bit parameter in registers. PKK> 4) In calls to __banked functions, the stack is always cleaned up by the caller. PKK> 5) The return values is passed as the caling convention (__sdcccall(0) or __sdcccall(1) pr __z88dk_fastcall or __smallc) specify. PKK> 6) ___sdcc_bcall trampoline is used for "legacy banking" (i.e. when compiling with --legacy-banking). PKK> 7) __sdcc_bcall_abc trampoline is used for __z88dk_fastcall __banked functions. PKK> 8) __sdcc_bcall_ehl is used for all other __banked functions. PKK> 9) Default trampolines are provided in the library, and could be overridden by the user. PKK> 10) The default trampolines call user-specified helper functions set_bank and get_bank that get / set the bank in register a, but leave other registers unchanged. PKK> Most of this should work ok, but there are two bugs: PKK> a) The combination of 5) and 8) is a problem: The __sdcc_bcall_ehl assumes that a is not used for the return value, which results in a bug when that is not true (__sdcccall(1) returns 8-bit values in a). I've opened bug report #3298 for this. PKK> b) All this is apparently not documented in the manual (looks like the manual only documents how __banked works on mcs51 / ds390). I've opened bug report #3299 for this. PKK> Philipp -- Best regards, Tony mailto:unt...@mail.ru _______________________________________________ Sdcc-user mailing list Sdcc-user@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/sdcc-user
