Fix multiply. Seems to work.

Probably won't test too much more. It will be easier to test once there
is a jet.
This commit is contained in:
Jared Hance 2014-05-30 15:34:43 -04:00
parent 33f7d69414
commit 86b871e395

View File

@ -1184,7 +1184,8 @@
::
:: Law: =((met 0 (ari p m)) +(p))
++ ari |= [p=@ m=@] ^- @
(lia p (mix (lsh 0 (met 0 m) 1) m))
:: (lia p (mix (lsh 0 (met 0 m) 1) m))
(mix (lsh 0 p 1) m)
:: bex base a to power p (call w/ b=0 c=1). very naive (need to replace)
:: or jet
@ -1221,18 +1222,24 @@
:: limit ari to precision p
++ lia |= [p=@ a=@] ^- @
=+ al=(met 0 a)
=+ p2=(add p 1)
?: (lte al p2)
(lsh 0 (sub p2 al) a)
(rnd p a (end 0 (sub al p2) a) (sub al p2))
?: (lte (met 0 a) (add p 1))
(lsh 0 (sub (add p 1) (met 0 a)) a)
(rnd p a)
:: round to nearest or even based on r (which has length n)
:: n should be the actual length of r, as it exists within a
:: The result is either (rhs 0 n a) or +(rsh 0 n a)
++ rnd |= [p=@ a=@ r=@ n=@]
++ rnd |= [p=@ a=@]
?: (lte (met 0 a) (add p 1))
a :: avoid overflow
=+ n=(sub (met 0 a) (add p 1))
=+ r=(end 0 n a)
(rne p a r n)
:: the real rnd
++ rne |= [p=@ a=@ r=@ n=@]
=+ b=(rsh 0 n a)
?: !=((met 0 r) n) :: starts with 0
?: !=((met 0 r) n) :: starts with 0 => not same distance
b
?: =((mod r 2) 0)
$(r (lsh 0 1 r)) :: ending 0s have no effect
@ -1245,12 +1252,15 @@
::::::::::::
++ mul |= [p=@ n=[s=? e=@ a=@] m=[s=? e=@ a=@]] ^- [s=? e=@ a=@]
=+ a2=(^mul a.n a.m)
=+ a3=(mix (lsh 0 (^mul p 2) 1) (end 0 (^mul p 2) a2))
~& [%mult a.n a.m]
~& [%res a2 a3]
=+ e2=(met 0 (rsh 0 (^mul p 2) a2))
:: =+ a3=(mix (lsh 0 (^mul p 2) 1) (end 0 (^mul p 2) a2))
~& [%mult `@ub`a.n `@ub`a.m]
~& [%res `@ub`a2]
=+ e2=(met 0 (rsh 0 (add 1 (^mul p 2)) a2))
:: =+ a4=(rnd p (rsh 0 e2 a3))
=+ a4=(rnd p (rsh 0 e2 a2))
~& [%fin `@ub`a4]
=+ s2=|(s.n s.m)
[s=s2 e=:(add e.n e.m e2) a=a3]
[s=s2 e=:(add e.n e.m e2) a=a4]
--
:: Real interface for @rd