Merge pull request #1 from warpdotdev/v0.20

Merge `v0.20` into `warpdotdev/v0.20`
warpdotdev · Jul 8, 2024 · 51c125c · 51c125c
2 parents 44b2170 + 14a7698
commit 51c125c
Show file tree

Hide file tree

Showing 33 changed files with 854 additions and 358 deletions.
diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -37,7 +37,47 @@ Bottom level categories:
 - Hal
 -->
 
-## Unreleased
+## Unreleased 
+
+## v0.20.2 (2024-06-12)
+
+This release force-bumps transitive dependencies of `wgpu` on `wgpu-core` and `wgpu-hal` to 0.21.1, to resolve some undefined behavior observable in the DX12 backend after upgrading to Rust 1.79 or later.
+
+### Bug Fixes
+
+#### General
+
+* Fix a `CommandBuffer` leak. By @cwfitzgerald and @nical in [#5141](https://github.com/gfx-rs/wgpu/pull/5141)
+
+#### DX12
+
+* Do not feed `&""` to `D3DCompile`, by @workingjubilee in [#5812](https://github.com/gfx-rs/wgpu/issues/5812).
+
+## v0.20.1 (2024-06-12)
+
+This release included v0.21.0 of `wgpu-core` and `wgpu-hal`, due to breaking changes needed to solve vulkan validation issues.
+
+### Bug Fixes
+
+This release fixes the validation errors whenever a surface is used with the vulkan backend. By @cwfitzgerald in [#5681](https://github.com/gfx-rs/wgpu/pull/5681).
+
+#### General
+
+- Clean up weak references to texture views and bind groups to prevent memory leaks. By @xiaopengli89 in [#5595](https://github.com/gfx-rs/wgpu/pull/5595).
+- Fix segfault on exit is queue & device are dropped before surface. By @sagudev in [#5640](https://github.com/gfx-rs/wgpu/pull/5640).
+
+#### Metal
+
+- Fix unrecognized selector crash on iOS 12. By @vladasz in [#5744](https://github.com/gfx-rs/wgpu/pull/5744).
+
+#### Vulkan
+
+- Fix enablement of subgroup ops extension on Vulkan devices that don't support Vulkan 1.3. By @cwfitzgerald in [#5624](https://github.com/gfx-rs/wgpu/pull/5624).
+
+
+#### GLES / OpenGL
+
+-  Fix regression on OpenGL (EGL) where non-sRGB still used sRGB [#5642](https://github.com/gfx-rs/wgpu/pull/5642)
 
 ## v0.20.0 (2024-04-28)
 

diff --git a/Cargo.lock b/Cargo.lock
diff --git a/Cargo.toml b/Cargo.toml
@@ -45,13 +45,13 @@ keywords = ["graphics"]
 license = "MIT OR Apache-2.0"
 homepage = "https://wgpu.rs/"
 repository = "https://github.com/gfx-rs/wgpu"
-version = "0.20.0"
+version = "0.20.2"
 authors = ["gfx-rs developers"]
 
 [workspace.dependencies.wgc]
 package = "wgpu-core"
 path = "./wgpu-core"
-version = "0.20.0"
+version = "0.21.0"
 
 [workspace.dependencies.wgt]
 package = "wgpu-types"
@@ -61,7 +61,7 @@ version = "0.20.0"
 [workspace.dependencies.hal]
 package = "wgpu-hal"
 path = "./wgpu-hal"
-version = "0.20.0"
+version = "0.21.0"
 
 [workspace.dependencies.naga]
 path = "./naga"
@@ -119,7 +119,7 @@ smallvec = "1"
 static_assertions = "1.1.0"
 thiserror = "1"
 wgpu = { version = "0.20.0", path = "./wgpu" }
-wgpu-core = { version = "0.20.0", path = "./wgpu-core" }
+wgpu-core = { version = "0.21.0", path = "./wgpu-core" }
 wgpu-example = { version = "0.20.0", path = "./examples/common" }
 wgpu-macros = { version = "0.20.0", path = "./wgpu-macros" }
 wgpu-test = { version = "0.20.0", path = "./tests" }

diff --git a/deno_webgpu/Cargo.toml b/deno_webgpu/Cargo.toml
@@ -45,7 +45,7 @@ workspace = true
 features = ["dx12"]
 
 [target.'cfg(windows)'.dependencies.wgpu-hal]
-version = "0.20.0"
+version = "0.21.0"
 path = "../wgpu-hal"
 features = ["windows_rs"]
 

diff --git a/tests/tests/subgroup_operations/mod.rs b/tests/tests/subgroup_operations/mod.rs
@@ -11,10 +11,22 @@ static SUBGROUP_OPERATIONS: GpuTestConfiguration = GpuTestConfiguration::new()
         TestParameters::default()
             .features(wgpu::Features::SUBGROUP)
             .limits(wgpu::Limits::downlevel_defaults())
-            .expect_fail(wgpu_test::FailureCase::molten_vk())
+            // Expect metal to fail on tests involving operations in divergent control flow
+            //
+            // Newlines are included in the panic message to ensure that _additional_ failures
+            // are not matched against.
+            .expect_fail(
+                wgpu_test::FailureCase::molten_vk()
+                    // 14.3 doesn't fail test 29
+                    .panic("thread 0 failed tests: 27,\nthread 1 failed tests: 27, 28,\n")
+                    // Prior versions do.
+                    .panic("thread 0 failed tests: 27, 29,\nthread 1 failed tests: 27, 28, 29,\n"),
+            )
             .expect_fail(
-                // Expect metal to fail on tests involving operations in divergent control flow
                 wgpu_test::FailureCase::backend(wgpu::Backends::METAL)
+                    // 14.3 doesn't fail test 29
+                    .panic("thread 0 failed tests: 27,\nthread 1 failed tests: 27, 28,\n")
+                    // Prior versions do.
                     .panic("thread 0 failed tests: 27, 29,\nthread 1 failed tests: 27, 28, 29,\n"),
             ),
     )

diff --git a/tests/tests/subgroup_operations/shader.wgsl b/tests/tests/subgroup_operations/shader.wgsl
@@ -111,6 +111,7 @@ fn main(
     add_result_to_mask(&passed, 25u, subgroup_invocation_id == 0u || subgroupShuffleUp(subgroup_invocation_id, 1u) == subgroup_invocation_id - 1u);
     add_result_to_mask(&passed, 26u, subgroupShuffleXor(subgroup_invocation_id, subgroup_size - 1u) == (subgroup_invocation_id ^ (subgroup_size - 1u)));
 
+    // Mac/Apple will fail this test.
     var passed_27 = false;
     if subgroup_invocation_id % 2u == 0u {
         passed_27 |= subgroupAdd(1u) == (subgroup_size / 2u);
@@ -119,6 +120,7 @@ fn main(
     }
     add_result_to_mask(&passed, 27u, passed_27);
 
+    // Mac/Apple will fail this test.
     var passed_28 = false;
     switch subgroup_invocation_id % 3u {
         case 0u: {
@@ -134,6 +136,7 @@ fn main(
     }
     add_result_to_mask(&passed, 28u, passed_28);
 
+    // Mac/Apple will sometimes fail this test. MacOS 14.3 passes it, so the bug in the metal compiler seems to be fixed.
     expected = 0u;
     for (var i = subgroup_size; i >= 0u; i -= 1u) {
         expected = subgroupAdd(1u);

diff --git a/wgpu-core/Cargo.toml b/wgpu-core/Cargo.toml
@@ -1,6 +1,6 @@
 [package]
 name = "wgpu-core"
-version = "0.20.0"
+version = "0.21.1"
 authors = ["gfx-rs developers"]
 edition = "2021"
 description = "WebGPU core logic on wgpu-hal"
@@ -127,7 +127,7 @@ version = "0.20.0"
 [dependencies.hal]
 package = "wgpu-hal"
 path = "../wgpu-hal"
-version = "0.20.0"
+version = "0.21.0"
 default_features = false
 
 [target.'cfg(all(target_arch = "wasm32", not(target_os = "emscripten")))'.dependencies]

diff --git a/wgpu-core/src/device/any_device.rs b/wgpu-core/src/device/any_device.rs
@@ -34,7 +34,7 @@ impl AnyDevice {
         unsafe fn drop_glue<A: HalApi>(ptr: *mut ()) {
             // Drop the arc this instance is holding.
             unsafe {
-                _ = Arc::from_raw(ptr.cast::<A::Surface>());
+                _ = Arc::from_raw(ptr.cast::<A::Device>());
             }
         }
 

diff --git a/wgpu-core/src/device/life.rs b/wgpu-core/src/device/life.rs
@@ -596,6 +596,18 @@ impl<A: HalApi> LifetimeTracker<A> {
             &mut trackers.textures,
             |maps| &mut maps.textures,
         );
+
+        // We may have been suspected because a texture view or bind group
+        // referring to us was dropped. Remove stale weak references, so that
+        // the backlink table doesn't grow without bound.
+        for texture in self.suspected_resources.textures.values() {
+            texture.views.lock().retain(|view| view.strong_count() > 0);
+            texture
+                .bind_groups
+                .lock()
+                .retain(|bg| bg.strong_count() > 0);
+        }
+
         self
     }
 
@@ -621,6 +633,13 @@ impl<A: HalApi> LifetimeTracker<A> {
             |maps| &mut maps.buffers,
         );
 
+        // We may have been suspected because a bind group referring to us was
+        // dropped. Remove stale weak references, so that the backlink table
+        // doesn't grow without bound.
+        for buffer in self.suspected_resources.buffers.values() {
+            buffer.bind_groups.lock().retain(|bg| bg.strong_count() > 0);
+        }
+
         self
     }
 

diff --git a/wgpu-core/src/device/queue.rs b/wgpu-core/src/device/queue.rs
@@ -1172,7 +1172,9 @@ impl Global {
 
             let mut used_surface_textures = track::TextureUsageScope::default();
 
-            let mut submit_surface_textures_owned = SmallVec::<[_; 2]>::new();
+            // Use a hashmap here to deduplicate the surface textures that are used in the command buffers.
+            // This avoids vulkan deadlocking from the same surface texture being submitted multiple times.
+            let mut submit_surface_textures_owned = FastHashMap::default();
 
             {
                 let mut command_buffer_guard = hub.command_buffers.write();
@@ -1279,7 +1281,9 @@ impl Global {
                                     Some(TextureInner::Native { .. }) => false,
                                     Some(TextureInner::Surface { ref raw, .. }) => {
                                         if raw.is_some() {
-                                            submit_surface_textures_owned.push(texture.clone());
+                                            // Compare the Arcs by pointer as Textures don't implement Eq.
+                                            submit_surface_textures_owned
+                                                .insert(Arc::as_ptr(&texture), texture.clone());
                                         }
 
                                         true
@@ -1472,7 +1476,9 @@ impl Global {
                         Some(TextureInner::Native { .. }) => {}
                         Some(TextureInner::Surface { ref raw, .. }) => {
                             if raw.is_some() {
-                                submit_surface_textures_owned.push(texture.clone());
+                                // Compare the Arcs by pointer as Textures don't implement Eq
+                                submit_surface_textures_owned
+                                    .insert(Arc::as_ptr(texture), texture.clone());
                             }
 
                             unsafe {
@@ -1517,7 +1523,7 @@ impl Global {
             let mut submit_surface_textures =
                 SmallVec::<[_; 2]>::with_capacity(submit_surface_textures_owned.len());
 
-            for texture in &submit_surface_textures_owned {
+            for texture in submit_surface_textures_owned.values() {
                 submit_surface_textures.extend(match texture.inner.get(&snatch_guard) {
                     Some(TextureInner::Surface { raw, .. }) => raw.as_ref(),
                     _ => None,
@@ -1529,7 +1535,7 @@ impl Global {
                     .raw
                     .as_ref()
                     .unwrap()
-                    .submit(&refs, &submit_surface_textures, Some((fence, submit_index)))
+                    .submit(&refs, &submit_surface_textures, (fence, submit_index))
                     .map_err(DeviceError::from)?;
             }
 

diff --git a/wgpu-core/src/present.rs b/wgpu-core/src/present.rs
@@ -154,17 +154,20 @@ impl Global {
                 parent_id: surface_id,
             });
         }
-        #[cfg(not(feature = "trace"))]
-        let _ = device;
+
+        let fence_guard = device.fence.read();
+        let fence = fence_guard.as_ref().unwrap();
 
         let suf = A::surface_as_hal(surface.as_ref());
         let (texture_id, status) = match unsafe {
-            suf.unwrap()
-                .acquire_texture(Some(std::time::Duration::from_millis(
-                    FRAME_TIMEOUT_MS as u64,
-                )))
+            suf.unwrap().acquire_texture(
+                Some(std::time::Duration::from_millis(FRAME_TIMEOUT_MS as u64)),
+                fence,
+            )
         } {
             Ok(Some(ast)) => {
+                drop(fence_guard);
+
                 let texture_desc = wgt::TextureDescriptor {
                     label: (),
                     size: wgt::Extent3d {

diff --git a/wgpu-hal/Cargo.toml b/wgpu-hal/Cargo.toml
@@ -1,6 +1,6 @@
 [package]
 name = "wgpu-hal"
-version = "0.20.0"
+version = "0.21.1"
 authors = ["gfx-rs developers"]
 edition = "2021"
 description = "WebGPU hardware abstraction layer"