staticdata: encode link_id in tagged linkage (#48673)

On 64-bit, we have enough space to encode (1) the tag, (2) the
`depmods` index, and (3) the offset all in a single 64-bit pointer
field. This means we don't need the external `link_id` arrays,
which reduces the size of many pkgimages by ~5%.

On 32-bit, we don't have enough bits to implement this strategy.
However, most linkages seem to be against the sysimage, and so
by giving that a separate tag we can achieve similar compression
because the `link_id` lists will be much shorter.

Co-authored-by: Tim Holy <tim.holy@gmail.com>

src/codegen.cpp

@@ -4251,7 +4251,7 @@ static jl_cgval_t emit_invoke(jl_codectx_t &ctx, const jl_cgval_t &lival, const
                     bool cache_valid = ctx.use_cache;
                     bool external = false;
                     if (ctx.external_linkage) {
-                       if (jl_object_in_image((jl_value_t*)codeinst)) {
+                       if (0 && jl_object_in_image((jl_value_t*)codeinst)) {
                            // Target is present in another pkgimage
                            cache_valid = true;
                            external = true;
@@ -5617,7 +5617,7 @@ static Function *emit_tojlinvoke(jl_code_instance_t *codeinst, Module *M, jl_cod
 
     bool cache_valid = params.cache;
     if (params.external_linkage) {
-        if (jl_object_in_image((jl_value_t*)codeinst)) {
+        if (0 && jl_object_in_image((jl_value_t*)codeinst)) {
             // Target is present in another pkgimage
             cache_valid = true;
         }
@@ -8529,7 +8529,7 @@ void jl_compile_workqueue(
         auto invoke = jl_atomic_load_relaxed(&codeinst->invoke);
         bool cache_valid = params.cache;
         if (params.external_linkage) {
-            cache_valid = jl_object_in_image((jl_value_t*)codeinst);
+            cache_valid = 0 && jl_object_in_image((jl_value_t*)codeinst);
         }
         // WARNING: isspecsig is protected by the codegen-lock. If that lock is removed, then the isspecsig load needs to be properly atomically sequenced with this.
         if (cache_valid && invoke != NULL) {

src/gc.c

@@ -1985,6 +1985,7 @@ STATIC_INLINE void gc_mark_array8(jl_ptls_t ptls, jl_value_t *ary8_parent, jl_va
     jl_gc_markqueue_t *mq = &ptls->mark_queue;
     jl_value_t *new_obj;
     size_t elsize = ((jl_array_t *)ary8_parent)->elsize / sizeof(jl_value_t *);
+    assert(elsize > 0);
     // Decide whether need to chunk ary8
     size_t nrefs = (ary8_end - ary8_begin) / elsize;
     if (nrefs > MAX_REFS_AT_ONCE) {
@@ -2016,6 +2017,7 @@ STATIC_INLINE void gc_mark_array16(jl_ptls_t ptls, jl_value_t *ary16_parent, jl_
     jl_gc_markqueue_t *mq = &ptls->mark_queue;
     jl_value_t *new_obj;
     size_t elsize = ((jl_array_t *)ary16_parent)->elsize / sizeof(jl_value_t *);
+    assert(elsize > 0);
     // Decide whether need to chunk ary16
     size_t nrefs = (ary16_end - ary16_begin) / elsize;
     if (nrefs > MAX_REFS_AT_ONCE) {
@@ -2668,7 +2670,6 @@ static void gc_mark_roots(jl_gc_markqueue_t *mq)
     }
     gc_try_claim_and_push(mq, jl_all_methods, NULL);
     gc_try_claim_and_push(mq, _jl_debug_method_invalidation, NULL);
-    gc_try_claim_and_push(mq, jl_build_ids, NULL);
     // constants
     gc_try_claim_and_push(mq, jl_emptytuple_type, NULL);
     gc_try_claim_and_push(mq, cmpswap_names, NULL);

src/jl_exported_data.inc

@@ -126,8 +126,6 @@
     XX(jl_voidpointer_type) \
     XX(jl_void_type) \
     XX(jl_weakref_type) \
-    XX(jl_build_ids) \
-    XX(jl_linkage_blobs) \
 
 // Data symbols that are defined inside the public libjulia
 #define JL_EXPORTED_DATA_SYMBOLS(XX) \

src/julia_internal.h

@@ -313,9 +313,8 @@ extern tracer_cb jl_newmeth_tracer;
 void jl_call_tracer(tracer_cb callback, jl_value_t *tracee);
 void print_func_loc(JL_STREAM *s, jl_method_t *m);
 extern jl_array_t *_jl_debug_method_invalidation JL_GLOBALLY_ROOTED;
-extern arraylist_t jl_linkage_blobs;                        // external linkage: sysimg/pkgimages
-extern jl_array_t *jl_build_ids JL_GLOBALLY_ROOTED;         // external linkage: corresponding build_ids
-extern arraylist_t jl_image_relocs;                        // external linkage: sysimg/pkgimages
+JL_DLLEXPORT extern arraylist_t jl_linkage_blobs; // external linkage: sysimg/pkgimages
+JL_DLLEXPORT extern arraylist_t jl_image_relocs;  // external linkage: sysimg/pkgimages
 
 extern JL_DLLEXPORT size_t jl_page_size;
 extern jl_function_t *jl_typeinf_func JL_GLOBALLY_ROOTED;
@@ -951,10 +950,7 @@ static inline void jl_set_gc_and_wait(void)
 // Query if a Julia object is if a permalloc region (due to part of a sys- pkg-image)
 STATIC_INLINE size_t n_linkage_blobs(void) JL_NOTSAFEPOINT
 {
-    if (!jl_build_ids)
-        return 0;
-    assert(jl_is_array(jl_build_ids));
-    return jl_array_len(jl_build_ids);
+    return jl_image_relocs.len;
 }
 
 // TODO: Makes this a binary search