This vector's contents always ended up identical to the
`RenderBundleEncoder`'s `BasePass`'s `dynamic_offsets` vector, so
we can just take values from there instead of copying them.
The `dynamic_offsets` and `is_dirty` flags only make sense when the
slot is occupied, so they should be inside the `Option`. This makes
`State::bind` into an `ArrayVec<Option<BindState>>`, and cleans up
various other bits.
In parking_lot 0.12 and parking_lot_core 0.9.0, those crates switched
from the winapi crate to the official Microsoft windows-sys crate.
This is fine, except that windows-sys and its dependencies are even
larger than winapi. Some users may wish to stick with winapi for the
time being; this change allows wgpu to accommodate them.
It's very odd to have almost all the render pass and compute pass ffi
functions in `wgpu` except for the `set_index_buffer` functions, which
live in Firefox. I'd like to remove these from Firefox and put them
back next to their companions.
These functions were originally removed from wgpu in #1077, because
wgpu-native has its own incompatible version of IndexFormat (see that
PR for details). However, with wgpu-native#85, that code was removed,
so having these functions in `wgpu` should be no longer be a problem
for wgpu-native.
* fix: don't panic on invalid id in Storage::get
* formatting
* removed double matches
* more match removal
* fix formatting
* add fix to Storage::label_for_invalid_id
This is intended to help developers new to wgpu or to graphics debugging
quickly recognize in a debugging tool which items are wgpu-generated, as
opposed to either part of their program or part of the platform graphics
system.
I also removed existing marker-like text such as leading underscores and
angle brackets.
Without this change, `LifetimeTracker::triage_suspected` never notices
that compute or render pipelines are free, and they stick around until
the hub is destroyed.
Fixes#2564.
Go ahead and call `global.device_drop` from `direct::Context::device_drop`.
Let `Global::device_drop` simply drop the life guard's `RefCount`, and
put off everything else entailed in freeing a device until
`Device::maintain` says its queue is empty and there are no more
references to it. (The user can reach that function, even after
dropping their `Device`, by calling `wgpu::Instance::poll_all`.)
Fixes#2563.
`RefCount::rich_drop_inner` is no longer used by anything other than `RefCount::drop`. It's simpler to just handle it directly in `drop`.
`MultiRefCount` has no need to heap-allocate the count, since it's
never cloned.
* Support to create surface from visual on Windows, add mpo(Multiple Plane Overlay) feature to AdapterInfo
* Expose create_surface_from_visual method
* Fix code style
* Make code more concise
* Revert mpo from AdapterInfo
The `min_index` parameter has no useful effect. Because the range 0..n is pushed
on the free list, index values less than `min_index` do get allocated.
Document `IdentityManager` and `TypedId`.
* CLEAR_COMMANDS extension is now more of a window into wgpu zero-init
this has mostly implications on the constraints, but also allows a more leaky documentation which makes sense for this non-standard function as there is no other place to look it up
* clear_texture via renderpasses wip
* 3D depth textures are no longer allowed, volumes are always cleared via CPY_DST
* cleanup texture's clear_views
* rename CLEAR_COMMANDS to CLEAR_TEXTURE
* separate clear_texture into reusable & more descriptive parts
* texture clear views are now created ahead of time
* discarded surface fixup goes through new clear_texture method now
* onq ueue texture initialization now goes threw clear_texture
pending inits need to store Stored textures now though, causing more ref count bumping
* texture init on queue_write_texture now also goes through new clear_texture
* transfer functions on commandbuffer use now new texture init route
* merge collect_zero_buffer_copies_for_clear_texture into clear_texture_via_buffer_copies
* clear functions now take TextureInitRange
* Fix clippy lints
* command_encoder_clear_texture no longer takes write lock on texture
* TextureClearMode encodes now is_color
* code cleanup, mostly about `use`
* Handle volume textures in clear_texture_via_render_passes properly
* texture clear no longer requires id::Stored
* init tracking fixes for volumes and init on partial subresource writes
* texture creation enforces COPY_DST only if absolutely necessary
* unrolled functional chain, reduce unsafe scope size
* fix clippy lints
* clear_texture test no longer creates 1D textures
see #2323
* 3D textures are no longer cleared as render target since this isn't supported on Metal
* fix deno building issue, fix formatting
* TextureInner::Surface can now be zero initialized
* Add feature gated 16-bit normalized texture support
Fixes#1934
* Query format properties only once
* Prevent supports_format from erroneously reporting false if the format wasn't queried
* Assert that 16bit norm formats also support on vulkan
* Add storage to TextureFormatInfo for 16-bit norm formats now that we check for support
* Introduce SamplerBindingType enum and fix bug in validation
This matches the WebGPU spec more closely and also lets us implement the
validation as it's written in the spec[1]. This fixes a bug which
previously prevented the "shadow" example from running in Firefox.
Previously the validation would check whether the filtering was
appropiate even if it was a comparison sampler. This didn't matter when
running the shadow example natively since the example was setting
`filtering: true` in addition to `comparison: true`. But this failed
when running through a browser since there the used WebIDL has the
proper enum representation and eventually resulted in `filtering: false`
being passed to wgpu-core which would then fail validation.[2]
[1]: https://gpuweb.github.io/gpuweb/#bind-group-creation
[2]: 674b6582ba/dom/webgpu/ipc/WebGPUChild.cpp (L502-L513)
* Fix remaining examples
* Fix deno_webgpu
* Renamed clear_buffer to fill_buffer (keeping up with spec)
* fill_buffer does not require CLEAR_BUFFER feature
* Changed ClearBuffer to FillBuffer for `trace` feature
* Renamed clearBuffer for web backend
* Creation of max compute workgroup size limits
* Implemented for Vulkan and Dx12
* Follow up of gfx-rs/wgpu#1808
* Check if current dispatch operation is under the limit
* Typos and some forgotten fixes
* Port to GLES
* Port to Metal with default limits
* Change Vulkan max_workroup_size_per_dims to pick the min of the reported dimensions
Co-authored-by: Dzmitry Malyshau <kvark@fastmail.com>
Co-authored-by: Dzmitry Malyshau <kvarkus@gmail.com>
* Added tests for texture zero init
sadly, they are typically passing even if texture zero init isn't doing its job
However, they form nice isolated examples for testing out texture initialization
It could be possible to dirty texture memory prior to ensure zero init did the job
* texture init tracker
* tracking texture init requirements for bind group, transfer and rendertarget
* texture clears for texture init
* queue submit
* write_texture
* Enforce presence of either render target or copy_dst flag
* clear render targets also with using buffer copies
enforce COPY_DST usage now on all textures
* adjust ImageSubresourceRange.layer_range calculation for 3D textures
init_tracker has now a `discard` function to get single data points back to uninitialized
use new standardized partition_point function
* track init state for discarded textures from renderpasses
missing:
* init on the fly if discarded is found within command buffer
* handle discarding only stencil or only depth
* added tests for zero init after discard
* tracking discarded surfaces now in separate struct, piping all inits through utility function
allows to resolve discard/init_action interactions
* Move various memory init code to separate mod
CommandBufferTextureMemoryActions is now fully encapsulated
* implemented discard init fixups for everything but renderpass
* render passes also cause now discard fixups
* fixup_discarded_surfaces takes now an iterator instead of Drain
* Add memory init test for discarding depth targets
* handle divergently discarded depth/stencil target
* comment & clippy fixes
* fix collect_zero_buffer_copies_for_clear_texture yielding block breaking copies
* [pr feedback] minor cleanup in zero_init_texture_after_discard, `use` hygenie
* [pr feedback] fix bug in ImageSubresourceRange range utils
* [pr feedback] fix texture tracker check, bundle transition_texture on init, cleanups
* Implemented drop for InitTrackerDrain
* remove incorrect comment about extents in add_pass_texture_init_actions
* Fix unit test & clippy issues in init_tracker
* init_tracker has now a `discard` function to get single data points back to uninitialized
use new standardized partition_point function
* allow unused InitTracker.discard function for time being
* new test for clearing textures of all formats
* clear_texture is now always implemented in wgpu-core instead of wgpu-hal
* add 1D/array/volume texture to clear_texture test
* set mip_level_count to 1 for D1 textures in clear_texture test
* fix warning & clippy lints
* Fix web build
* [pr feedback] fix potentially skipped resource transition, fix wrong mip size for texture clear
* collect_zero_buffer_copies_for_clear_texture aligns rows now properly
* Move validation from `Device::create_texture_from_hal()` to `Device::create_texture()`
---
Users encountered a segfault upon creating a texture when mip_level_count was set to 0 with (at least) Vulkan.
This is due to wgpu's validation taking place after the API's texture is created.
This change moves wgpu's validation to before the API's texture is created, introducing a friendly error instead of a segfault.
* Make `Device::create_texture_from_hal()` infallible
* run cargo fmt
* enable CLEAR_COMMANDS feature for all adapters
* replaced fill_buffer with clear_buffer (minus variable value)
* Improve clear buffer/texture test
and remove unused clear-buffer.ron test
* renamed clear_image to clear_texture
* skeleton for new wgpu-hal clear_texture
* clear_texture implementation for vulkan
* clear_texture now restricts usage
* clear_texture implementation for dx12
* Implemented clear_texture for Metal backend
* Clean up GLES clear_buffer and leave note on how to implement clear_texture in the future
* fix linux compilation & formatting issues
* comment & namespace fixes
* Extent3d has now a simple mipmap calculating function
Fix incorrect use of texture.size in clear_texture for metal/dx12
Fix incorrect mip/layer ranges in clear_texture for metal/dx12
* WIP: add cts_runner and deno_webgpu crate
* add test
* remove Cargo.lock
* review comment
* simplify
* fix bugs
* improve cts_runner to work with crowlKats/webgpu-examples
* fix
* remove build.rs
cts_runner binaries are now not portable anymore.
Also startup will now print a bunch of cargo:rerun-if-changed=. This
will be fixed in deno_core.
* remove d.ts
* add original deno license file
* hal/metal: support Features::NON_FILL_POLYGON_MODE
* Document Metal support for NON_FILL_POLYGON_MODE
* Split Features::NON_FILL_POLYGON_MODE in two
Adds two new features: `LINE_POLYGON_MODE` and `POINT_POLYGON_MODE`.
Allows metal to support `PolygonMode::Line` while not being required to
support `PolygonMode::Point`.
* Formatting
* Rename {LINE,POINT}_POLYGON_MODE to POLYGON_MODE_{LINE,POINT}
* split out init_tracker module
with explicit buffer init tracking types
* Added shortcut methods for checked init action handling
Fixed a few unchecked sites in the process
* fix clippy warnings
due to previous changes there was only a no-op buffer collapse
The ranges it tried collapsing on were already guranteed to be merged by the init-trackers drain method. However, accross several init actions there can be opportunity for reducing fill_buffer calls.
1609: [Vulkan] Initialize wgpu objects from raw handles r=kvark a=zarik5
**Connections**
This PR is a successor of https://github.com/gfx-rs/gfx/pull/3762
**Description**
The `handle_is_external` flag mechanism was not enough to ensure safety, when for example the Instance is wrapped in `Arc<>` and a library cannot keep track of all its clones. This is the case with the bevy render rewrite. The solution was to let wgpu be in charge of destroying the handles, but it also has to keep a reference to a "drop guard" which is always dropped after the handle is destroyed. For the OpenXR integration, this drop guard will be `xr::Instance`.
For now this is just a proof of concept. Only instance creation is handled, and there is even type error in `wgc::Instance::from_hal()`.
I have a few concerns:
* Is it ok to expose `hal::Instance` from the wgpu crate? Or should the user pass all the parameters so `hal::Instance` can be constructed internally? This second options is more disruptive, since the wgpu-types crate would need to import all platform-specific crates to define the structure/enum to hold the raw handles.
* Without counting the call to create the `hal::Instance`, there are 4 indirection calls to save the raw instance. Can this be optimized in any way?
Do you think it is possible to merge wgpu-hal into wgpu-core? This could help with reducing the distance from the surface level API to the platform specific APIs even more.
**Testing**
Vulkan/Android (Oculus Quest) using [this sample](https://github.com/zarik5/openxrs/blob/wgpu-test/openxr/examples/vulkan.rs).
Co-authored-by: Riccardo Zaglia <riccardo.zaglia5@gmail.com>
1704: Fix buffer memory initialization to use the tracker consistently r=kvark a=kvark
**Connections**
Depends on https://github.com/gfx-rs/naga/pull/1125
**Description**
Contains a number of dx12 fixes in addition to the buffer initialization fix.
The problem there was - we used the "pending writes" command buffer to initialize buffers, but also used the device's resource tracker and its state to figure out the transitions *after* a command buffer is merged its states.
**Testing**
Ran the examples on Vulkan.
Co-authored-by: Dzmitry Malyshau <kvarkus@gmail.com>
1602: New DX12 backend (unfinished) r=kvark a=kvark
**Connections**
Continuation of #1471
**Description**
Mostly fresh implementation on DX12 backend for wgpu-hal
Core changes:
- texture usage is filtered before passing to `create_texture_view`, so that wgpu-hal can assume every bit is important.
- all the copies involving textures are limited to one array layer at a time
- remove render pass boundary usages
**Testing**
Examples (not running yet)
Co-authored-by: Dzmitry Malyshau <dmalyshau@mozilla.com>
Co-authored-by: Dzmitry Malyshau <kvarkus@gmail.com>
1607: Fix Downlevel Vertex Stage Storage Buffer Check r=kvark a=cwfitzgerald
**Connections**
Fixes a bug in #1599. Also follows up on #1583 marking all of hello-compute as a failure.
**Testing**
Tests now pass on rpi4.
Co-authored-by: Connor Fitzgerald <connorwadefitzgerald@gmail.com>
1599: Fix Limits for lava/llvmpipe and re-enable and re-work CI r=kvark,groves,wumpf a=cwfitzgerald
**Connections**
No longer fixes#1551.
**Description**
This PR has a couple things going on at once.
- Fixes limits for llvmpipe, lavapipe, and rpi4.
- Added a downlevel limit to express that RPI4 does not allow storage buffers in vertex shaders on GL.
- Added a `Limits::downlevel_default()` that takes minimum limits from GLES and adds some more documentation on how to choose limits.
- Moved all examples to the new downlevel limits
- Reworks CI to re-enable software testing and unify instructions.
**Testing**
It is
Co-authored-by: Connor Fitzgerald <connorwadefitzgerald@gmail.com>
1592: Detailed limits check and error r=cwfitzgerald a=kvark
**Connections**
Related to #1590
**Description**
The old comparison "A < B" was simply wrong, because it would do every field in order and could bail out mid-way without reaching for all limits.
**Testing**
Untested
Co-authored-by: Dzmitry Malyshau <kvarkus@gmail.com>
