From 1c0c0bb315661e22472bbf5770b63330316dce25 Mon Sep 17 00:00:00 2001
From: Justin Ethier <justin.ethier@gmail.com>
Date: Fri, 15 Jul 2016 23:14:36 -0400
Subject: [PATCH] Change how mutations are stored in memory

Use a reallocated memory buffer instead of malloc'd pairs. This should speed things up by reducing the number of allocations and by keeping mutations in contiguous sections of memory.
---
 gc.c                    | 27 ++++------------------
 include/cyclone/types.h |  4 +++-
 runtime.c               | 51 ++++++++++++++++++++++++++---------------
 3 files changed, 40 insertions(+), 42 deletions(-)

diff --git a/gc.c b/gc.c
index 30464566..89b42dda 100644
--- a/gc.c
+++ b/gc.c
@@ -831,27 +831,6 @@ void gc_thr_add_to_move_buffer(gc_thread_data * d, int *alloci, object obj)
   (*alloci)++;
 }
 
-// Generic buffer functions
-void **vpbuffer_realloc(void **buf, int *len)
-{
-  return realloc(buf, (*len) * sizeof(void *));
-}
-
-void **vpbuffer_add(void **buf, int *len, int i, void *obj)
-{
-  if (i == *len) {
-    *len *= 2;
-    buf = vpbuffer_realloc(buf, len);
-  }
-  buf[i] = obj;
-  return buf;
-}
-
-void vpbuffer_free(void **buf)
-{
-  free(buf);
-}
-
 // END heap definitions
 
 // Tri-color GC section
@@ -1458,6 +1437,10 @@ void gc_thread_data_init(gc_thread_data * thd, int mut_num, char *stack_base,
   thd->stack_trace_idx = 0;
   thd->stack_prev_frame = NULL;
   thd->mutations = NULL;
+  thd->mutation_buflen = 128;
+  thd->mutation_count = 0;
+  thd->mutations = 
+      vpbuffer_realloc(thd->mutations, &(thd->mutation_buflen));
   thd->exception_handler_stack = NULL;
 //  thd->thread = NULL;
   thd->thread_state = CYC_THREAD_STATE_NEW;
@@ -1503,7 +1486,7 @@ void gc_thread_data_free(gc_thread_data * thd)
     if (thd->stack_traces)
       free(thd->stack_traces);
     if (thd->mutations) {
-      clear_mutations(thd);
+      free(thd->mutations);
     }
     free(thd);
   }
diff --git a/include/cyclone/types.h b/include/cyclone/types.h
index f18aa1c6..0041415f 100644
--- a/include/cyclone/types.h
+++ b/include/cyclone/types.h
@@ -118,7 +118,9 @@ struct gc_thread_data_t {
   char *stack_start;
   char *stack_limit;
   // Minor GC write barrier
-  void *mutations;
+  void **mutations;
+  int mutation_buflen;
+  int mutation_count;
   // List of objects moved to heap during minor GC
   void **moveBuf;
   int moveBufLen;
diff --git a/runtime.c b/runtime.c
index 9f187d25..7a888ece 100644
--- a/runtime.c
+++ b/runtime.c
@@ -357,33 +357,26 @@ void debug_dump_globals()
  * Note these functions and underlying data structure are only used by
  * the calling thread, so locking is not required.
  */
-
 void add_mutation(void *data, object var, int index, object value)
 {
   gc_thread_data *thd = (gc_thread_data *) data;
   if (is_object_type(value)) {
+    thd->mutations = vpbuffer_add(thd->mutations, &(thd->mutation_buflen), thd->mutation_count, var);
+    thd->mutation_count++;
     if (index >= 0) {
-      // For vectors only, malloc_make_pair index as another var. That way
+      // For vectors only, add index as another var. That way
       // the write barrier only needs to inspect the mutated index.
-      thd->mutations = malloc_make_pair(obj_int2obj(index), thd->mutations);
+      thd->mutations = vpbuffer_add(thd->mutations, &(thd->mutation_buflen), thd->mutation_count, obj_int2obj(index));
+      thd->mutation_count++;
     }
-    thd->mutations = malloc_make_pair(var, thd->mutations);
   }
 }
 
-/* TODO: consider a more efficient implementation, such as reusing old nodes
-         instead of reclaiming them each time
- */
 void clear_mutations(void *data)
 {
+  // Not clearing memory, just resetting count
   gc_thread_data *thd = (gc_thread_data *) data;
-  list l = thd->mutations, next;
-  while (l != NULL) {
-    next = cdr(l);
-    free(l);
-    l = next;
-  }
-  thd->mutations = NULL;
+  thd->mutation_count = 0;
 }
 
 /* END mutation table */
@@ -3791,9 +3784,9 @@ int gc_minor(void *data, object low_limit, object high_limit, closure cont,
 
   // Transport mutations
   {
-    list l;
-    for (l = ((gc_thread_data *) data)->mutations; l != NULL; l = cdr(l)) {
-      object o = car(l);
+    int l = 0;
+    while (l < ((gc_thread_data *) data)->mutation_count) {
+      object o = ((gc_thread_data *) data)->mutations[l++]; //car(l);
       if (is_value_type(o)) {
         // Can happen if a vector element was already
         // moved and we found an index. Just ignore it
@@ -3804,8 +3797,7 @@ int gc_minor(void *data, object low_limit, object high_limit, closure cont,
         int i;
         object idx;
         // For vectors, index is encoded as the next mutation
-        l = cdr(l);
-        idx = car(l);
+        idx = ((gc_thread_data *) data)->mutations[l++];
         i = obj_obj2int(idx);
         gc_move2heap(((vector) o)->elements[i]);
       } else if (type_of(o) == forward_tag) {
@@ -4446,6 +4438,27 @@ object copy2heap(void *data, object obj)
                   &on_stack);
 }
 
+// Generic buffer functions
+void **vpbuffer_realloc(void **buf, int *len)
+{
+  return realloc(buf, (*len) * sizeof(void *));
+}
+
+void **vpbuffer_add(void **buf, int *len, int i, void *obj)
+{
+  if (i == *len) {
+    *len *= 2;
+    buf = vpbuffer_realloc(buf, len);
+  }
+  buf[i] = obj;
+  return buf;
+}
+
+void vpbuffer_free(void **buf)
+{
+  free(buf);
+}
+
 /* RNG section */
 #define norm 2.328306549295728e-10
 #define m1   4294967087.0