about summary refs log tree commit diff stats
diff options
context:
space:
mode:
authorKartik K. Agaram <vc@akkartik.com>2016-05-17 12:35:06 -0700
committerKartik K. Agaram <vc@akkartik.com>2016-05-17 12:50:43 -0700
commit51b53b13f9d88b42ca81510b41a91340bab72ccc (patch)
tree31083eaa868d2ee8b409fff9b11055e87d44d7cf
parentcaa4275c9a56f4e64ea3bcaf572f76445d305d93 (diff)
downloadmu-51b53b13f9d88b42ca81510b41a91340bab72ccc.tar.gz
2968
More reorganization in preparation for implementing recursive abandon().

Refcounts are getting incredibly hairy. I need to juggle containers
containing other containers, and containers *pointing* to other
containers. For a while I considered getting rid of address_element_info
entirely and just going by types for every single
update_refcount. But that's definitely more work, and it's unclear that
things will be cleaner/shorter/simpler. I haven't measured the speedup,
but it seems worth optimizing every pointer copy to make sure we aren't
manipulating types at runtime.

The key insight now is a) to continue to compute information about
nested containers at load time, because that's the common case when
updating refcounts, but b) to compute information about *pointed* values
at run-time, because that's the uncommon case.

As a result, we're going to cheat in the interpreter and use type
information at runtime just for abandon(), just because the
corresponding task when we get to a compiler will be radically
different. It will still be tractable, though.
-rw-r--r--010vm.cc2
-rw-r--r--030container.cc2
-rw-r--r--036refcount.cc51
-rw-r--r--037abandon.cc4
4 files changed, 42 insertions, 17 deletions
diff --git a/010vm.cc b/010vm.cc
index 7c2ccba1..ee18a171 100644
--- a/010vm.cc
+++ b/010vm.cc
@@ -98,6 +98,8 @@ struct string_tree {
   string_tree(string_tree* l, string_tree* r) :left(l), right(r) {}
 };
 
+// End type_tree Definition
+
 :(before "End Globals")
 // Locations refer to a common 'memory'. Each location can store a number.
 map<int, double> Memory;
diff --git a/030container.cc b/030container.cc
index 902afb21..387693ef 100644
--- a/030container.cc
+++ b/030container.cc
@@ -96,7 +96,7 @@ def main [
 //: Can't put this in type_info because later layers will add support for more
 //: complex type trees where metadata depends on *combinations* of types.
 
-:(after "Types")
+:(before "struct reagent")
 struct container_metadata {
   int size;
   vector<int> offset;
diff --git a/036refcount.cc b/036refcount.cc
index 8c593943..a7da07b8 100644
--- a/036refcount.cc
+++ b/036refcount.cc
@@ -22,10 +22,16 @@ if (is_mu_address(x)) {
   assert(scalar(data));
   assert(x.value);
   assert(!x.metadata.size);
-  update_refcounts(get_or_insert(Memory, x.value), data.at(0), payload_size(x));
+  update_refcounts(x, data.at(0));
 }
+
 :(code)
-void update_refcounts(int old_address, int new_address, int size) {
+void update_refcounts(const reagent& old, int new_address) {
+  assert(is_mu_address(old));
+  update_refcounts(get_or_insert(Memory, old.value), new_address, old.type->right, payload_size(old));
+}
+
+void update_refcounts(int old_address, int new_address, const type_tree* payload_type, int /*just in case it's an array*/payload_size) {
   if (old_address == new_address) {
     trace(9999, "mem") << "copying address to itself; refcount unchanged" << end();
     return;
@@ -43,7 +49,7 @@ void update_refcounts(int old_address, int new_address, int size) {
       cerr << "Negative refcount: " << old_address << ' ' << old_refcount << '\n';
       exit(0);
     }
-    // End Decrement Reference Count(old_address, size)
+    // End Decrement Reference Count(old_address, payload_type, payload_size)
   }
   // increment refcount of new address
   if (new_address) {
@@ -114,7 +120,7 @@ reagent/*copy*/ element = element_type(base.type, offset);
 assert(!has_property(element, "lookup"));
 element.value = address;
 if (is_mu_address(element))
-  update_refcounts(get_or_insert(Memory, element.value), ingredients.at(2).at(0), payload_size(element));
+  update_refcounts(element, ingredients.at(2).at(0));
 // End Update Refcounts in PUT
 
 :(scenario refcounts_put_index)
@@ -136,7 +142,7 @@ def main [
 
 :(after "Write Memory in PUT_INDEX in Run")
 if (is_mu_address(element))
-  update_refcounts(get_or_insert(Memory, element.value), value.at(0), payload_size(element));
+  update_refcounts(element, value.at(0));
 // End Update Refcounts in PUT_INDEX
 
 :(scenario refcounts_maybe_convert)
@@ -160,7 +166,7 @@ def main [
 
 :(after "Write Memory in Successful MAYBE_CONVERT")
 if (is_mu_address(product))
-  update_refcounts(get_or_insert(Memory, product.value), get_or_insert(Memory, base_address+/*skip tag*/1), payload_size(product));
+  update_refcounts(product, get_or_insert(Memory, base_address+/*skip tag*/1));
 // End Update Refcounts in Successful MAYBE_CONVERT
 
 //:: manage refcounts in instructions that copy multiple locations at a time
@@ -186,13 +192,29 @@ def main [
 +run: {3: "foo"} <- copy {2: ("address" "foo"), "lookup": ()}
 +mem: incrementing refcount of 1000: 2 -> 3
 
-:(after "Types")
+:(after "End type_tree Definition")
 struct address_element_info {
   int offset;  // where inside a container type (after flattening nested containers!) the address lies
-  int payload_size;  // size of type it points to
-  address_element_info(int o, int p) {
+  const type_tree* payload_type;  // all the information we need to compute sizes of items inside an address inside a container. Doesn't need to be a full-scale reagent, since an address inside a container can never be an array, and arrays are the only type that need to know their location to compute their size.
+  address_element_info(int o, const type_tree* p) {
     offset = o;
-    payload_size = p;
+    payload_type = p;
+  }
+  address_element_info(const address_element_info& other) {
+    offset = other.offset;
+    payload_type = other.payload_type ? new type_tree(*other.payload_type) : NULL;
+  }
+  ~address_element_info() {
+    if (payload_type) {
+      delete payload_type;
+      payload_type = NULL;
+    }
+  }
+  address_element_info& operator=(const address_element_info& other) {
+    offset = other.offset;
+    if (payload_type) delete payload_type;
+    payload_type = other.payload_type ? new type_tree(*other.payload_type) : NULL;
+    return *this;
   }
 };
 :(before "struct container_metadata ")
@@ -255,7 +277,8 @@ void compute_container_address_offsets(type_tree* type) {
 bool append_addresses(int base_offset, const type_tree* type, vector<address_element_info>& out, container_metadata& out_metadata) {
   const type_info& info = get(Type, type->value);
   if (type->name == "address") {
-    out.push_back(address_element_info(base_offset, payload_size(type)));
+    assert(type->right && type->right->name != "array");  // array types can't be handled without a full reagent and its value
+    out.push_back(address_element_info(base_offset, new type_tree(*type)));
     return true;
   }
   if (info.kind == PRIMITIVE) return true;
@@ -265,7 +288,7 @@ bool append_addresses(int base_offset, const type_tree* type, vector<address_ele
     // Compute Container Address Offset(element)
     if (is_mu_address(element)) {
       trace(9993, "transform") << "address at offset " << curr_offset << end();
-      out.push_back(address_element_info(curr_offset, payload_size(element)));
+      out.push_back(address_element_info(curr_offset, new type_tree(*element.type)));
       ++curr_offset;
     }
     else if (is_mu_container(element)) {
@@ -323,13 +346,13 @@ void update_container_refcounts(const reagent& x, const vector<double>& data) {
   const container_metadata& metadata = get(Container_metadata, x.type);
   for (int i = 0; i < SIZE(metadata.address); ++i) {
     const address_element_info& info = metadata.address.at(i);
-    update_refcounts(get_or_insert(Memory, x.value + info.offset), data.at(info.offset), info.payload_size);
+    update_refcounts(get_or_insert(Memory, x.value + info.offset), data.at(info.offset), info.payload_type, size_of(info.payload_type));
   }
   for (map<pair<int, int>, vector<address_element_info> >::const_iterator p = metadata.maybe_address.begin(); p != metadata.maybe_address.end(); ++p) {
     if (data.at(p->first.first) != p->first.second) continue;
     for (int i = 0; i < SIZE(p->second); ++i) {
       const address_element_info& info = p->second.at(i);
-      update_refcounts(get_or_insert(Memory, x.value + info.offset), data.at(info.offset), info.payload_size);
+      update_refcounts(get_or_insert(Memory, x.value + info.offset), data.at(info.offset), info.payload_type, size_of(info.payload_type));
     }
   }
 }
diff --git a/037abandon.cc b/037abandon.cc
index 05842702..98901568 100644
--- a/037abandon.cc
+++ b/037abandon.cc
@@ -13,10 +13,10 @@ def main [
 # both allocations should have returned the same address
 +mem: storing 1 in location 5
 
-:(before "End Decrement Reference Count(old_address, size)")
+:(before "End Decrement Reference Count(old_address, payload_type, payload_size)")
 if (old_refcount == 0) {
   trace(9999, "mem") << "automatically abandoning " << old_address << end();
-  abandon(old_address, size);
+  abandon(old_address, payload_size);
 }
 
 //: When abandoning addresses we'll save them to a 'free list', segregated by size.