Re: [9front] Weird >>= operator behaviour in 5c/5l

[ori@eigenstate.org]

> This will affect any assembler source that uses
> 
> 	ROR $0, R1, R2
> 
> to get at the special RRX arm instruction alias
> for a rotate right by one bit with extension.
> 
> That instruction would become
> 
> 	SRA $0, R1, R2
> 
> for a arithmetic shift right by 32 bits. It's
> probably not used anywhere but I wanted to make
> sure you knew about it.
> 
> For reference, here's my understanding of the
> possible shift values on arm:
> 
> 	a	i	stype	imm5
> 	SLL	LSL	00	0..31
> 	SRL	LSR	01	1..32 mod 32
> 	SRA	ASR	10	1..32 mod 32
> 	ROR	ROR	11	1..31
> 	ROR	RRX	11	0
> 
> Hope that helps.
> 
> Cheers,
>   Anthony
> 

Yeah -- I was thinking that a 0-byte ROR would be
a no-op, and it wouldn't matter -- but I think I
made a mistake. It seems like I should just make
check explicitly for the opcodes, and substitute
in oprrr().

diff -r 01125acb5565 sys/src/cmd/5l/asm.c
--- a/sys/src/cmd/5l/asm.c	Tue Jan 19 19:56:38 2021 -0800
+++ b/sys/src/cmd/5l/asm.c	Wed Jan 20 15:38:43 2021 -0800
@@ -785,7 +785,10 @@
 
 	case 8:		/* sll $c,[R],R -> mov (R<<$c),R */
 		aclass(&p->from);
-		o1 = oprrr(p->as, p->scond);
+		if((p->as == ASRL || p->as == ASRA) && instoffset == 0)
+			o1 = oprrr(ASLL, p->scond);
+		else
+			o1 = oprrr(p->as, p->scond);
 		r = p->reg;
 		if(r == NREG)
 			r = p->to.reg;