Change definition of "weak" in threats calculation about asmfish

[tthsqe12]

If you want to learn the instructions, you can see my latest changes.
Keep in mind that some variables like PiecesThem, PiecesPawn, ect. are in registers already. The full list is in the beginning of the EvalThreats macro. I will give you the operations of some of the instructions here.

xor a, b ; same as a = a^b;
and a, b ; same as a = a&b;
 or a, b ; same as a = a|b;
andn a, b, c ; same as a = (~b) & c, uses a macro for cpu without this instruction, see BasicMacros.asm

[lantonov]

	mov   r8, PiecesThem
	mov   r9, PiecesPawn
	and   r9, r8         ;r9 = pos.pieces(Them, PAWN)
	xor   r8, r9          ;r8 = (pos.pieces(Them) ^ pos.pieces(Them, PAWN))
        mov   r9, qword[.ei.attackedBy+8*(8*Them+Pawn)]
       andn  r10,  qword[.ei.attackedBy2+8*Us], qword[.ei.attackedBy2+8*Them]
         or  r9, r10          ; r9 = stronglyProtected  		
        and   r8, r9
; r8 = defended
       andn   r9, r9, PiecesThem           ; r9 = ~stronglyProtected & pos.pieces(Them)
	and   r9, AttackedByUs              ; r9 = ~stronglyProtected & pos.pieces(Them) & ei.attackedBy[Us][ALL_PIECES]
; r9 = weak
	 or   r8, r9
; r8 = defended | weak

I tried per your explanations which are clear enough and without looking at your code. Afraid to test it though, not to blow my computer.

[tthsqe12]

Ah, don't be afraid of blowing up your computer. Your computer runs the program like it runs any other program. The worst you can get here is probably a segfault, which will be unhandled and thus passed to the os. There are several ways you can lock up your system:

• set the program to real time priority and start a lot of threads that get stuck in an infinite loop
• allocate a ton of memory and then start to fill it up

I think you will see there is no danger of that in this section of code.

[lantonov]

Ok, I'll "try this at home" and then try to find my errors.

[tthsqe12]

So your code will assemble on popcnt and base versions (and be correct!) but it will not assemble on bmi2 version. This is due to the andn macro.

line:
 andn  r10,  qword[.ei.attackedBy2+8*Us], qword[.ei.attackedBy2+8*Them]

andn is defined as a macro in line 295 of Basic macros.asm
This means that the preprocessor is going to either of these things

when CPU_HAS_BMI1=0
 mov r10, qword[.ei.attackedBy2+8*Us]
 not r10
 and r10, qword[.ei.attackedBy2+8*Them]


when CPU_HAS_BMI1=1
 andn  r10,  qword[.ei.attackedBy2+8*Us], qword[.ei.attackedBy2+8*Them]

This last instruction is not encodable because only the last argument is allowed to be a memory reference.
http://www.felixcloutier.com/x86/ANDN.html

Before you start wishing that that it were possible to encode two memory references into a single instruction, keep in mind that the cpu actually has to decode all these instructions. That is, if you believe the power consumption myth.
https://www.usenix.org/system/files/conference/cooldc16/cooldc16-paper-hirki.pdf

[lantonov]

Thanks for the detailed explanation and reference. Wow, I almost did it !

[lantonov]

	mov   r8, PiecesThem
	mov   r9, PiecesPawn
	and   r9, r8         ;r9 = pos.pieces(Them, PAWN)
	xor   r8, r9          ;r8 = (pos.pieces(Them) ^ pos.pieces(Them, PAWN))
        mov   r9, qword[.ei.attackedBy+8*(8*Them+Pawn)]
        mov r10, qword[.ei.attackedBy2+8*Us] 
       andn  r10,  r10, qword[.ei.attackedBy2+8*Them]
         or  r9, r10          ; r9 = stronglyProtected  		
        and   r8, r9
        ; r8 = defended
       andn   r9, r9, PiecesThem           ; r9 = ~stronglyProtected & pos.pieces(Them)
         and   r9, AttackedByUs              ; r9 = ~stronglyProtected & pos.pieces(Them) & ei.attackedBy[Us][ALL_PIECES]
       ; r9 = weak
	 or   r8, r9
       ; r8 = defended | weak

This should work then (I am away from my 64 bit to try it)