riscv: kernel: Shared NEXT saves about 256 bytes out of 1 KiB

So it might be worth doing. The key parts of the runtime - doconst,
docolon etc - have their copies of NEXT for speed, but the rest of the
code words that make up the kernel share a single copy and simply jump
to it instead of including it inline. For every jump that replaces NEXT,
12 bytes are saved.

It's also very easy to turn this on and off.

mu/target/RISC-V/kernel-itc.mu4

@@ -69,7 +69,8 @@ forth
 -- ------------------------------------------------------------------------
 assembler
 
-: next   asm{  0 ip w lw   4 ip ip addi   -4 w x lw   0 x jr  } ;
+( Compile NEXT inline.)
+: next,   asm{  0 ip w lw   4 ip ip addi   -4 w x lw   0 x jr  } ;
 
 ( XXX Macros or subroutines?)
 ( Data stack macros.)
@@ -118,15 +119,16 @@ meta: constant  new  ,  ;code  label doconst
 
 meta: create    new     ;code  label dovar
                                label wpush
-                               top push1   w top mv  ( move pfa to top)   next  ;c
+                               top push1   w top mv  ( move pfa to top)
+                               label shared-next  next,  ;c
 
                                label dodoes
                                top push1   w top mv  ( move pfa to top)
                                x w mv  ( x is parent ip)  ( fall thru)  ;c
 
 meta: :         new   __target-colon
                         ;code  label docolon
-                               ip rpush1   w ip mv  ( pfa is new ip)   next  ;c
+                               ip rpush1   w ip mv  ( pfa is new ip)   next,  ;c
 
 definer
 meta: does>   <;code>  asm{  dodoes x jal  }  __target-colon ;
@@ -143,7 +145,7 @@ meta: variable     cell    buffer ;  ( A variable is a cell-sized buffer!)
 meta: 2variable    cell 2* buffer ;
 
 
-code*  ^    ( unnest)   ip rpop1   here ( begin)   next  ;c
+code*  ^    ( unnest)   ip rpop1   here ( begin)   next,  ;c
 codes nope
 
 code* (lit)   0 ip w lw   4 ip ip addi  ( fetch literal)   wpush j  ;c
@@ -153,6 +155,10 @@ code execute   ( pfa)      top w mv   here ( begin)
 code @execute  ( 'pfa)   0 top w lw   j ( again)  ;c
 
 
+( Compile a jump to a previous inline NEXT.)
+meta: next   asm{  shared-next j  } ;
+
+
 .ifdef notyet  ( skip stuff that is still MSP430 code ;-)
 
 code*    (branch)   0 ip ip lw  ( follow branch)   next  ;c