For compressed marshaling, use absolute shared object references

xavierleroy · xavierleroy · commit 6f0b636c0697 · 2023-02-12T14:59:11.000+01:00
Instead of the default relative references. As noted in #4056, absolute references compress better. (We get a 15% size reduction on typical .cmt files.) Closes: #4056
diff --git a/runtime/extern.c b/runtime/extern.c
@@ -805,7 +805,10 @@ static void extern_rec(struct caml_extern_state* s, value v)
     /* Check if object already seen */
     if (! (s->extern_flags & NO_SHARING)) {
       if (extern_lookup_position(s, v, &pos, &h)) {
-        extern_shared_reference(s, s->obj_counter - pos);
+        /* #4056: using absolute references for shared objects improves
+           compressibility. */
+        uintnat d = s->extern_flags & COMPRESSED ? pos : s->obj_counter - pos;
+        extern_shared_reference(s, d);
         goto next_item;
       }
     }
diff --git a/runtime/intern.c b/runtime/intern.c
@@ -527,10 +527,10 @@ static void intern_rec(struct caml_intern_state* s,
       case CODE_SHARED8:
         ofs = read8u(s);
       read_shared:
-        CAMLassert (ofs > 0);
-        CAMLassert (ofs <= s->obj_counter);
+        if (!s->compressed) ofs = s->obj_counter - ofs;
+        CAMLassert (ofs < s->obj_counter);
         CAMLassert (s->intern_obj_table != NULL);
-        v = s->intern_obj_table[s->obj_counter - ofs];
+        v = s->intern_obj_table[ofs];
         break;
       case CODE_SHARED16:
         ofs = read16u(s);

Original file line number	Diff line number	Diff line change
`@@ -805,7 +805,10 @@ static void extern_rec(struct caml_extern_state* s, value v)`
`805`	`805`	`/* Check if object already seen */`
`806`	`806`	`if (! (s->extern_flags & NO_SHARING)) {`
`807`	`807`	`if (extern_lookup_position(s, v, &pos, &h)) {`
`808`		`- extern_shared_reference(s, s->obj_counter - pos);`
	`808`	`+ /* #4056: using absolute references for shared objects improves`
	`809`	`+ compressibility. */`
	`810`	`+ uintnat d = s->extern_flags & COMPRESSED ? pos : s->obj_counter - pos;`
	`811`	`+ extern_shared_reference(s, d);`
`809`	`812`	`goto next_item;`
`810`	`813`	`}`
`811`	`814`	`}`