From 4563f51c98f07126de32fda1a5da356602ce865f Mon Sep 17 00:00:00 2001
From: Alex Crichton <alex@alexcrichton.com>
Date: Fri, 28 Jan 2022 12:25:34 -0800
Subject: [PATCH] Lazily allocate the bump-alloc chunk in the externref table

This commit updates the allocation of a `VMExternRefActivationsTable`
structure to perform zero malloc memory allocations. Previously it would
allocate a page-size of `chunk` plus some space in hash sets for future
insertions. The main trick here implemented is that after the first gc
during the slow path the fast chunk allocation is allocated and
configured.

The motivation for this PR is that given our recent work to further
refine and optimize the instantiation process this allocation started to
show up in a nontrivial fashion. Most modules today never touch this
table anyway as almost none of them use reference types, so the time
spent allocation and deallocating the table per-store was largely wasted
time.

Concretely on a microbenchmark this PR speeds up instantiation of a
module with one function by 30%, decreasing the instantiation cost from
1.8us to 1.2us. Overall a pretty minor win but when the instantiation
times we're measuring start being in the single-digit microseconds this
win ends up getting magnified!
---
 crates/runtime/src/externref.rs | 24 +++++++++++++++++++-----
 1 file changed, 19 insertions(+), 5 deletions(-)

diff --git a/crates/runtime/src/externref.rs b/crates/runtime/src/externref.rs
index 573dd4b24f5a..6dd59ad747af 100644
--- a/crates/runtime/src/externref.rs
+++ b/crates/runtime/src/externref.rs
@@ -566,8 +566,13 @@ impl VMExternRefActivationsTable {
 
     /// Create a new `VMExternRefActivationsTable`.
     pub fn new() -> Self {
-        let mut chunk = Self::new_chunk(Self::CHUNK_SIZE);
-        let next = chunk.as_mut_ptr().cast::<TableElem>();
+        // Start with an empty chunk in case this activations table isn't used.
+        // This means that there's no space in the bump-allocation area which
+        // will force any path trying to use this to the slow gc path. The first
+        // time this happens, though, the slow gc path will allocate a new chunk
+        // for actual fast-bumping.
+        let mut chunk: Box<[TableElem]> = Box::new([]);
+        let next = chunk.as_mut_ptr();
         let end = unsafe { next.add(chunk.len()) };
 
         VMExternRefActivationsTable {
@@ -576,8 +581,8 @@ impl VMExternRefActivationsTable {
                 end: NonNull::new(end).unwrap(),
                 chunk,
             },
-            over_approximated_stack_roots: HashSet::with_capacity(Self::CHUNK_SIZE),
-            precise_stack_roots: HashSet::with_capacity(Self::CHUNK_SIZE),
+            over_approximated_stack_roots: HashSet::new(),
+            precise_stack_roots: HashSet::new(),
             stack_canary: None,
             #[cfg(debug_assertions)]
             gc_okay: true,
@@ -728,9 +733,18 @@ impl VMExternRefActivationsTable {
             "after sweeping the bump chunk, all slots should be `None`"
         );
 
+        // If this is the first instance of gc then the initial chunk is empty,
+        // so we lazily allocate space for fast bump-allocation in the future.
+        if self.alloc.chunk.is_empty() {
+            self.alloc.chunk = Self::new_chunk(Self::CHUNK_SIZE);
+            self.alloc.end =
+                NonNull::new(unsafe { self.alloc.chunk.as_mut_ptr().add(self.alloc.chunk.len()) })
+                    .unwrap();
+        }
+
         // Reset our `next` finger to the start of the bump allocation chunk.
         unsafe {
-            let next = self.alloc.chunk.as_mut_ptr().cast::<TableElem>();
+            let next = self.alloc.chunk.as_mut_ptr();
             debug_assert!(!next.is_null());
             *self.alloc.next.get() = NonNull::new_unchecked(next);
         }