ps-st3: bulk steal transfers, and a test that reaches them

.github/workflows/ci.yml

@@ -11,7 +11,7 @@ env:
 jobs:
   check:
     name: Check
-    runs-on: ubuntu-latest
+    runs-on: ubicloud-standard-2
     strategy:
       fail-fast: false
       matrix:
@@ -32,7 +32,7 @@ jobs:
 
   test:
     name: Test suite
-    runs-on: ubuntu-latest
+    runs-on: ubicloud-standard-2
     steps:
       - name: Checkout sources
         uses: actions/checkout@v4
@@ -45,7 +45,7 @@ jobs:
 
   loom:
     name: Test suite (Loom)
-    runs-on: ubuntu-latest
+    runs-on: ubicloud-standard-2
     steps:
       - name: Checkout sources
         uses: actions/checkout@v4
@@ -61,7 +61,7 @@ jobs:
 
   miri:
     name: Miri
-    runs-on: ubuntu-latest
+    runs-on: ubicloud-standard-2
     steps:
       - name: Checkout sources
         uses: actions/checkout@v4
@@ -78,7 +78,7 @@ jobs:
 
   lints:
     name: Lints
-    runs-on: ubuntu-latest
+    runs-on: ubicloud-standard-2
     steps:
       - name: Checkout sources
         uses: actions/checkout@v4
@@ -96,7 +96,7 @@ jobs:
 
   docs:
     name: Docs
-    runs-on: ubuntu-latest
+    runs-on: ubicloud-standard-2
     steps:
       - name: Checkout sources
         uses: actions/checkout@v4

.github/workflows/publish.yml

@@ -0,0 +1,40 @@
+name: Publish
+
+# House scheme: a version bump on the default branch is the release. No tags,
+# no manual step. Publishing is irreversible, so the trigger has to be
+# something you cannot do by accident, and editing the version field is that.
+on:
+  push:
+    branches: [main]
+    paths: ['Cargo.toml']
+
+jobs:
+  publish:
+    runs-on: ubicloud-standard-2
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          fetch-depth: 2
+      - uses: dtolnay/rust-toolchain@stable
+
+      - id: version
+        run: |
+          new=$(sed -nE '0,/^version = "(.*)"/s//\1/p' Cargo.toml)
+          old=$(git show HEAD^:Cargo.toml | sed -nE '0,/^version = "(.*)"/s//\1/p')
+          if [ "$new" = "$old" ]; then
+            echo "version unchanged at $new, nothing to publish"
+            echo "changed=false" >> "$GITHUB_OUTPUT"
+          else
+            echo "version $old -> $new"
+            echo "changed=true" >> "$GITHUB_OUTPUT"
+          fi
+
+      # The gate runs on this commit too, but a release must not depend on job
+      # ordering, so it is re-run here rather than assumed.
+      - if: steps.version.outputs.changed == 'true'
+        run: cargo test --release
+
+      - if: steps.version.outputs.changed == 'true'
+        run: cargo publish --token "$CARGO_REGISTRY_TOKEN"
+        env:
+          CARGO_REGISTRY_TOKEN: ${{ secrets.CARGO_REGISTRY_TOKEN }}

Cargo.toml

@@ -1,16 +1,16 @@
 [package]
-name = "st3"
+name = "ps-st3"
 # When incrementing version and releasing to crates.io:
 # - Update crate version in README.md
 # - Update CHANGELOG.md
 # - Update if necessary copyright notice in LICENSE-MIT
 # - Create a "vX.Y.Z" git tag
-version = "0.4.1"
+version = "0.4.2"
 authors = ["Serge Barral <serge.barral@asynchronics.com>"]
 edition = "2021"
 rust-version = "1.60"
 license = "MIT OR Apache-2.0"
-repository = "https://github.com/asynchronics/st3"
+repository = "https://github.com/pathscale/st3"
 readme = "README.md"
 description = """
 A very fast lock-free, bounded, work-stealing LIFO queue.
@@ -20,6 +20,11 @@ keywords = ["queue", "stack", "bounded", "lock-free", "work-stealing"]
 autotests = false
 autobenches = false
 
+[lib]
+# The crate is renamed but the library is not, so `use st3::...` and the
+# `st3_loom` cfg both keep working unchanged against the fork.
+name = "st3"
+
 [dependencies]
 crossbeam-utils = { version = "0.8.14", default-features = false }
 

README.md

@@ -1,5 +1,10 @@
 # St³ — the Stealing Static Stack
 
+> A PathScale fork of [asynchronics/st3](https://github.com/asynchronics/st3), published as
+> `ps-st3`. The library is still named `st3`, so `use st3::lifo::Worker` is unchanged.
+> It carries upstream PR #12 (bulk segmented copies on steal) ahead of a release, plus a
+> differential test holding that bulk path against the per-element reference.
+
 Lock-free, bounded, work-stealing queues with FIFO stealing and LIFO or FIFO
 semantic for the worker thread.
 
@@ -34,7 +39,7 @@ Add this to your `Cargo.toml`:
 
 ```toml
 [dependencies]
-st3 = "0.4.1"
+ps-st3 = "0.4.1"
 ```
 
 

src/fifo.rs

@@ -585,11 +585,16 @@ impl<T> Stealer<T> {
         debug_or_loom_assert!(transfer_count <= dest_free_capacity);
 
         // Move all items but the last to the destination queue.
-        for offset in 0..transfer_count {
-            unsafe {
-                let item = self.queue.read_at(stealer_head.wrapping_add(offset));
-                dest.queue.write_at(dest_tail.wrapping_add(offset), item);
-            }
+        unsafe {
+            crate::transfer::transfer_items(
+                &self.queue.buffer,
+                self.queue.mask,
+                stealer_head,
+                &dest.queue.buffer,
+                dest.queue.mask,
+                dest_tail,
+                transfer_count,
+            );
         }
 
         // Make the moved items visible by updating the destination tail position.
@@ -664,11 +669,18 @@ impl<T> Stealer<T> {
         debug_or_loom_assert!(transfer_count <= dest_free_capacity);
 
         // Move all items but the last to the destination queue.
-        for offset in 0..transfer_count {
-            unsafe {
-                let item = self.queue.read_at(stealer_head.wrapping_add(offset));
-                dest.queue.write_at(dest_tail.wrapping_add(offset), item);
-            }
+        //
+        // Safety: see the `steal()` method.
+        unsafe {
+            crate::transfer::transfer_items(
+                &self.queue.buffer,
+                self.queue.mask,
+                stealer_head,
+                &dest.queue.buffer,
+                dest.queue.mask,
+                dest_tail,
+                transfer_count,
+            );
         }
 
         // Read the last item.

src/lib.rs

@@ -66,6 +66,7 @@ mod config;
 pub mod fifo;
 pub mod lifo;
 mod loom_exports;
+mod transfer;
 
 /// Error returned when stealing is unsuccessful.
 #[derive(Debug, Clone, PartialEq, Eq)]

src/lifo.rs

@@ -661,11 +661,16 @@ impl<T> Stealer<T> {
         debug_or_loom_assert!(transfer_count <= dest_free_capacity);
 
         // Move all items but the last to the destination queue.
-        for offset in 0..transfer_count {
-            unsafe {
-                let item = self.queue.read_at(old_head.wrapping_add(offset));
-                dest.queue.write_at(dest_tail.wrapping_add(offset), item);
-            }
+        unsafe {
+            crate::transfer::transfer_items(
+                &self.queue.buffer,
+                self.queue.mask,
+                old_head,
+                &dest.queue.buffer,
+                dest.queue.mask,
+                dest_tail,
+                transfer_count,
+            );
         }
 
         // Make the moved items visible by updating the destination tail position.
@@ -732,11 +737,18 @@ impl<T> Stealer<T> {
         debug_or_loom_assert!(transfer_count <= dest_free_capacity);
 
         // Move all items but the last to the destination queue.
-        for offset in 0..transfer_count {
-            unsafe {
-                let item = self.queue.read_at(old_head.wrapping_add(offset));
-                dest.queue.write_at(dest_tail.wrapping_add(offset), item);
-            }
+        //
+        // Safety: see the `steal()` method.
+        unsafe {
+            crate::transfer::transfer_items(
+                &self.queue.buffer,
+                self.queue.mask,
+                old_head,
+                &dest.queue.buffer,
+                dest.queue.mask,
+                dest_tail,
+                transfer_count,
+            );
         }
 
         // Read the last item.