[Openmcl-devel] A little more LAP stuff...

[R. Matthew Emerson]

On Feb 16, 2010, at 2:18 PM, Jon S. Anthony wrote:

> Looking over some LAP functions and disassembling some user level lisp
> functions, my question(s) is what (or whether) there is any difference
> between the two approaches used in these for the X86 register toggle
> tagging between immediate/node indications.
> 
> In user level lisp, this typically seems to use btrl to reset bit
> indicating immediate and btsl to set bit indicating node.  For example:
> 
> 
>  [13]    (btrl ($ 1) (@ (% fs) 8)) ; indicate immo is immediate
>  [23]    (movl (@ -2 (% arg_z)) (% imm0)) ; load it up...
> ...
>  [46]    (btsl ($ 1) (@ (% fs) 8)) ; indicate immo is node
> 
> 
> This same thing in LAP functions where (mark-as-imm ...) and
> (mark-as-node ...) are used is:
> 
>          ;; (mark-as-imm immo)
>  [30]    (andb ($ 254) (@ (% fs) 8)) ; indicate immo is immediate
> ...
>          ;; (mark-as-node imm0)
>  [84]    (xorl (% imm0) (% imm0)) ; always there, but unclear about it
>  [86]    (orb ($ 1) (@ (% fs) 8)) ; indicate immo is node
> 
> Is there any real difference between these for the intended task?

They do the same thing (set/clear a bit in the node-regs-mask word stored in the TCR).

I'm not sure why they're different.  I like the btr/bts for clarity, though I think it may be the case that the btr/bts instructions are a byte longer than equivalent logical operations would be.  I think the lap macro uses logical operations because I had a sort of vague intention to make it support reclassifying the category of multiple registers at once, i.e., I thought that one might say

(mark-as-imm temp0 temp1)
...
(mark-as-node temp0 temp1)

and have that expand to

(andb ($ 249) (@ (% fs) 8))
...
(xorl (% temp0) (% temp0))
(xorl (% temp1) (% temp1))
(orb ($ 6) (@ (% fs) 8))

but I never got around to doing that.