Add LSHIFT, RSHIFT primitives

[dvdkhlng]

Hi,

I ported the logarithmic floating point numbers from my euro-forth presentation for use with volksForth. However these rely on having efficient bit-shift instructions available.

Here is an implementation of LSHIFT and RSHIFT that appears to be pretty efficient. Is there any reason e.g. WRT memory resource constraints that would prevent us from adding these to vf-cbm-core.fth?

\ 6502 optimized bit-shift   dk06aug26
Code lshift ( x1 b8 -- x2)
 iny  SP )Y lda  N sta
 SP X)  lda  0<> ?[
   tax iny
   SP )Y lda
   [[ N asl  .a rol   dex 0= ?]
   SP )Y sta  dey
 ]?  dey
 SP 2inc  N lda  Puta jmp
end-code

Code rshift ( x1 0..15 -- x2)
 iny  SP )Y lda  N sta
 SP X)  lda  0<> ?[
   tax iny
   SP )Y lda
   [[ .a lsr  N ror   dex 0= ?]
   SP )Y sta   dey
 ]?  dey
 SP 2inc  N lda  Puta jmp
end-code

[pzembrod]

I am indeed a bit wary of binary size; e.g. when building cc64 on the X16, I currently have maybe 100 bytes of leeway.

I wouldn't be opposed to configuring additional versions of the kernel, minimal vs more feature-rich. We could also just provide added loadable sources, and add convenient make targets and prebuilt binaries with a richer word set e.g. including LSHIFT and RSHIFT.

Once we get closer to ANS compatibility further down the road, then I'd see including them in the kernel in a different light again. Though, still, having a minimal kernel on which to build further would probably still have its attraction then.