The fastcall combined with invoke is also a powerful tool in assembly ;-)
> Conventions entitled fastcall have not been standardized, and have been
> implemented differently, depending on the compiler vendor.
The standards in this case is C/C++, and in the case of MS stdcalls will be used?
Watcom allows the user to specify his own calling convention, and uses
EAX, EDX, EBX, ECX - ax, dx, bx, cx (-zf1).
The assembler calling convention is register calls, usually using (E)AX, (E)DX and (E)CX.
EBX is preserved in Win32, but BX used in 16-bit.
in v2.08, support for MS fastcall has been improved. The proposed optimization has been implemented, and loading the arguments into the registers is now more fool-proved.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
code segment para 'code' assume cs:code,ds:code org 100h start: jmp around fast1 proc fastcall a1:word mov ax,a1 ret fast1 endp around: invoke fast1, 1 sub ax,1 mov ah,4Ch int 21h code ends end startDX is used as argument to fast1:
JWasm v2.07pre, Nov 2 2011 fast1.asm 00000000 code segment para 'code' assume cs:code,ds:code org 100h 00000100 EB03 start: jmp around 00000102 fast1 proc fastcall a1:word 00000102 8BC1 mov ax,cx 00000104 ret 00000104 C3 * retn 00000105 fast1 endp 00000105 around: invoke fast1, 1 00000105 BA0100 * mov dx, 1 00000108 E8F7FF * call fast1issue opened in bug tracker:
https://sourceforge.net/tracker/?func=detail&aid=3538021&group_id=255677&atid=1126895
Please report bugs in the bug tracker!
There are 2 problems with support of MS 16-bit fastcalls:
- the registers used to hold arguments in JWasm v2.06 were CX and DX. However, the correct registers are AX, DX and BX.
- Additional parameters are pushed from LEFT to RIGHT ( like the PASCAL calling convention )
The fastcall combined with invoke is also a powerful tool in assembly ;-)
> Conventions entitled fastcall have not been standardized, and have been
> implemented differently, depending on the compiler vendor.
The standards in this case is C/C++, and in the case of MS stdcalls will be used?
Watcom allows the user to specify his own calling convention, and uses
EAX, EDX, EBX, ECX - ax, dx, bx, cx (-zf1).
The assembler calling convention is register calls, usually using (E)AX, (E)DX and (E)CX.
EBX is preserved in Win32, but BX used in 16-bit.
fast4 proc _CType a1:size_t, a2:size_t, a3:size_t, a4:size_t mov ax?,a1 mov dx?,a2 mov bx?,a3 mov cx?,a4 ret fast4 endp invoke fast4, 1, dx?, bx?, cx?MS: (-zf0) ECX and EDX - cx and dx
Watcom: (-zf1)
It would be nice then if this..
..could be reduced to this:
Yes, I agree the proposed optimization is useful. But it won't happen for v2.07, this version is due to be released soon.
in v2.08, support for MS fastcall has been improved. The proposed optimization has been implemented, and loading the arguments into the registers is now more fool-proved.