guile/guile-lightning README

[Marius Vollmer]

CVSROOT:        /cvs
Module name:    guile
Changes by:     Marius Vollmer <address@hidden> 01/04/01 09:13:34

Modified files:
        guile-lightning: README 

Log message:
        

CVSWeb URLs:
http://subversions.gnu.org/cgi-bin/cvsweb/guile/guile-lightning/README.diff?r1=1.2&r2=1.3

Patches:
Index: guile/guile-lightning/README
diff -u guile/guile-lightning/README:1.2 guile/guile-lightning/README:1.3
--- guile/guile-lightning/README:1.2    Mon Mar 19 17:54:21 2001
+++ guile/guile-lightning/README        Sun Apr  1 09:13:34 2001
@@ -3,3 +3,75 @@
 
 Guile-lightning makes the JIT compiler GNU Lightning available to
 Guile.
+
+
+Random notes:
+- - - - - - -
+
+Tail-callable calling convention.  I want to have a tail-callable
+calling convention that supports a large number of arguments and
+multiple return values with cheap ignoring of additional values.
+
+The convention laid out here is intended for the x86 and is stack
+based.  It is expressed in Lightning terms, tho.  I want to modify
+Lightning to offer this calling convention as an abstraction, which
+can then be ported to other backends.
+
+    Caller saved registers: r0, r1, r2
+    Callee saved registers: v0, v1, v2
+
+    On entry:  r0 = closure
+               r1 = number of args (scaled)
+
+              sp[0]  = return address
+              sp[1]  = first arg
+               sp[r1] = last arg
+
+    On return: r0       = first val
+              r1       = number return values (scaled)
+               sp[0]    = last val
+               sp[r1-1] = second val
+
+This requires a lot of stack shuffling.
+
+Prolog/Epilog variants
+
+ Fixed, known number of args
+
+    push  v0
+    push  v1
+    push  v2
+    push  fp
+    mov   fp sp
+     :
+    mov   sp fp
+    pop   fp
+    pop          v2
+    pop          v1
+    pop          v0
+    pop          r2
+    add          sp sp num_args*sizeof(SCM)
+    mov   r1 4
+    jmp   r2
+
+ Variable number of args
+
+    push  r1
+    push  v1
+    push  v2
+    push  fp
+    mov   fp sp
+     :
+    mov   sp fp
+    pop   fp
+    pop          v2
+    pop          v1
+    pop          v0
+    pop   r1
+    pop          r2
+    add          sp sp r1
+    mov   r1 4
+    jmp   r2
+
+ Maybe have a way to declare which of v0, v1, v2 really needs to be
+ saved.  Maybe allow frame-pointer less operation.