Jim Blandy

Push constants in render bundles are half-implemented

3

`RenderBundleEncoder::finish` propagates `SetPushConstant` commands from the encoder to the final `RenderBundle`, but the bundle's `BasePass`'s `push_constant_data` vector is always empty.

[core] Documentation and cleanups for `LifetimeTracker`

2

- Use a `for` loop in `LifetimeTracker` triage code. A `for` loop is less noisy here than a `drain`, which requires: - a `mut` qualifier for a variable whose modified...

Failure to fail in `Vulkan/llvmpipe (LLVM 17.0.6, 256 bits)/1] wgpu_test::shader::struct_layout::uniform_input`

7

This test fails on Fedora 39 because the expected error message is not generated: ``` [2024-05-05T10:00:11Z ERROR wgpu_test::expectations] Expected to fail due to [FailureReason { kind: Some(ValidationError), message: Some("a matrix...

`cargo doc` doesn't check for broken links in bitflags documentation

2

Running `cargo doc` on the following complains about the broken link in the docs for `Legit`, but not the one in the docs fo `Flags`: ```rust bitflags::bitflags! { /// This...

`wgpu_hal::dx12` should not allocate a 256KiB buffer just to hold zeroes

2

In `wgpu_hal`, `dx12::Device` should not spend 256KiB of GPU memory on `Device::zero_buffer`, whose only reason for existence is to serve as the source of `CopyBufferRegion` calls in `::clear_buffer`. It seems...

When generating code for a `naga::Module`, Naga currently evaluates every override-expression it can, regardless of whether that expression is statically used by the entry point(s) for which we are generating...

I think it is the case that Naga generates bounds checks on storage texture writes, but this is unnecessary: https://github.com/gpuweb/gpuweb/pull/4194

Support dynamic indexing of by-value matrices and arrays, per WGSL

6

WGSL permits dynamically indexing matrices that are not stored in variables: - A function parameter may be a matrix. - A formal parameter expression does not evaluate to a pointer,...

`wgpu_test::compute_pass_resource_ownership` causes Linux kernel crash on `gl` backend

2

Running the `wgpu_test::compute_pass_resource_ownership` test on Linux using the GL backend causes a GPU reset and eventual kernel instability. The same test passes when run using the AMD or llvmpipe Vulkan...

I missed this reviewing #5383, but in `PhysicalDeviceFeatures::from_extensions_and_requested_features` in `wgpu-hal/src/vulkan/adapter.rs`, when we populate `shader_atomic_int64`, we should set both `shader_buffer_int64_atomics` *and* `shader_shared_int64_atomics`, since we require both in `PhysicalDeviceFeatures::to_wgpu`. cc @atlv24 @teoxoy