On: Ask HN: Show me your half baked project

[mysterymath]

I'm porting LLVM/Clang to the 6502. Notoriously difficult project, attempted dozens of times by as many folks.

https://github.com/mysterymath/clang6502

My take generates pretty darn good assembly for the cases it handles, but it's absurdly incomplete. Still, a huge amount of risk factors have already been addressed, and there's only a few big known unknowns left.

Example input:

  void print_int(char x) {
    if (x < 10) {
      x += '0';
      asm volatile ("JSR\t$FFD2" : "+a"(x));
      return;
    }
    print_int(x / 10);
    print_int(x % 10);
  }

Example output:

  .code
  .global print__int                      ; -- Begin function print_int
  print__int:                             ; @print_int
  ; %bb.0:                                ; %entry
    CMP #10
    BMI LBB0__2
  ; %bb.1:                                ; %if.end.preheader
    LDX #10
    PHA                                     ; 1-byte Folded Spill
    JSR ____udivqi3
    JSR print__int
    PLA                                     ; 1-byte Folded Reload
    LDX #10
    JSR ____umodqi3
  LBB0__2:                                ; %if.then
    CLC
    ADC #48
    ;APP
    JSR $FFD2
    ;NO_APP
    RTS
                                        ; -- End function
  .global ____udivqi3
  .global ____umodqi3

[rzzzt → mysterymath]

OMF:2097 profile picture spotted! Loved the game as a child.

Will the compiler support 65C02 instructions?

[mysterymath → rzzzt]

Shouldn't be too hard to add later, since they're mostly just swizzles onto the existing instructions, and LLVM already has good support for registering slightly different versions of a target (subtargeting).

For now, I'm targeting the plain-jane MOS chip, since it's the buggiest and least capable, and thus the most difficult.