The wire's device is externally owned so reference/release were no-ops.
To unify the code paths, remove the special casing and instead
take an extra ref on the device the wire server is created with. This
is functionally equivalent and will allow both the current wire code,
and the incoming change to allow multiple device/adapter creation to
both work.
This CL also makes it possible for the client to destroy the device
before child objects.
A follow-up CL will mitigate this on the server side.
Bug: dawn:384
Change-Id: Ic5427074469012dccf8689ec95a848e6ba2c1fc2
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/37001
Commit-Queue: Austin Eng <enga@chromium.org>
Reviewed-by: Jiawei Shao <jiawei.shao@intel.com>
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
This removes the logic where the Client owns the Device and the
Device owns all other objects. Ownership should be tracked in
dawn_native either with refcounting or validation to disallow
operations after an object's parent has been destroyed.
This simplifies the wire client code in that the client only
tracks allocated handles and does not manage parent/child lifetimes.
This is an important simplification so we can support multiple WebGPU
instances, adapters, and devices on a single wire.
Bug: dawn:384
Change-Id: I8ecc7c368130b8917202150c467b5f0e7d4b753e
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/37000
Commit-Queue: Austin Eng <enga@chromium.org>
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
According to the http://www.eel.is/c++draft/temp.expl.spec:
An explicit specialization shall not use a storage-class-specifier
other than thread_local.
Clang doesn't claims about it, but GCC does.
An error example for GCC 8.4.0:
gen/third_party/dawn/src/dawn_wire/client/ApiObjects_autogen.h:25:5:
error: explicit template specialization cannot have a storage class
Bug: dawn:384
Change-Id: Iaf86722a943d19c9796a7f112885666ac88f20ca
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/33480
Commit-Queue: Corentin Wallez <cwallez@chromium.org>
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
When the wire is disconnected, the client will not receive any
messages from the server. We need to manually reject all callbacks.
Bug: dawn:556
Change-Id: Ia03456b3209dbe0e1e54543d344180d11d4c6f1e
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/31162
Commit-Queue: Austin Eng <enga@chromium.org>
Reviewed-by: Stephen White <senorblanco@chromium.org>
This is needed so that:
1. We can support multiple devices in the wire. The device will need
to know how to destroy its child objects.
2. The wire needs to be aware of all objects and their in-flight
callbacks so that it can reject them if the wire is disconnnected.
A future change will handle this.
3. Fix leaks of objects on page teardown. When the page is torn down,
the wire client is destroyed, and we skip calling release() for all
objects since the object holding the proc table was also destroyed.
Bug: dawn:384, dawn:556
Change-Id: Ie23afe4e515b02e924fcfc2db92b749fd2257c9c
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/31160
Reviewed-by: Austin Eng <enga@chromium.org>
Commit-Queue: Austin Eng <enga@chromium.org>
This CL adds support for chunking large commands by first serializing
large commands first into a separate buffer, and then sending the
buffer data chunk by chunk.
This code path is used for large writeBuffer and writeTexture, as well
as the inline memory transfer service for buffer mapping. The transfer
for writeBuffer and writeTexture will be optimized further in Chrome,
and the inline memory transfer service is currently used only in tests.
Bug: chromium:1123861, chromium:951558
Change-Id: I02491a44e653e2383174958d9c3d4a4db6fd7bde
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/28882
Commit-Queue: Austin Eng <enga@chromium.org>
Reviewed-by: Stephen White <senorblanco@chromium.org>
This is a reland of b04a92f01b
with the deletion of a duplicate exported function in dawn_wire that
was causing a compilation failure on Windows.
Original change's description:
> Add a per-thread proc table using thread local storage
>
> In situations where both dawn_wire and dawn_native are used on separate
> threads (Chrome with --single-process or --in-process-gpu), it's
> desirable to have a per-thread proc table so that the WebGPU C++ API can
> still be used. This eliminates classes of bugs with manual
> reference/release errors.
>
> This also changes many of the GetProcs functions to return const
> references to the static proc tables known at compile time, instead of a
> copy.
>
> Bug: none
> Change-Id: I8775bb715b312dd9476a1903fbd797d4b1302614
> Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/29240
> Reviewed-by: Stephen White <senorblanco@chromium.org>
> Reviewed-by: Corentin Wallez <cwallez@chromium.org>
> Commit-Queue: Austin Eng <enga@chromium.org>
Bug: none
Change-Id: Id90e5372132cd93a2f8631c8185d0e71b01bc1af
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/29443
Commit-Queue: Austin Eng <enga@chromium.org>
Reviewed-by: Kai Ninomiya <kainino@chromium.org>
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Reviewed-by: Stephen White <senorblanco@chromium.org>
This reverts commit b04a92f01b.
Reason for revert: Build failed on win-archive-dbg; see:
https://chromium-review.googlesource.com/c/chromium/src/+/2450792
Original change's description:
> Add a per-thread proc table using thread local storage
>
> In situations where both dawn_wire and dawn_native are used on separate
> threads (Chrome with --single-process or --in-process-gpu), it's
> desirable to have a per-thread proc table so that the WebGPU C++ API can
> still be used. This eliminates classes of bugs with manual
> reference/release errors.
>
> This also changes many of the GetProcs functions to return const
> references to the static proc tables known at compile time, instead of a
> copy.
>
> Bug: none
> Change-Id: I8775bb715b312dd9476a1903fbd797d4b1302614
> Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/29240
> Reviewed-by: Stephen White <senorblanco@chromium.org>
> Reviewed-by: Corentin Wallez <cwallez@chromium.org>
> Commit-Queue: Austin Eng <enga@chromium.org>
TBR=cwallez@chromium.org,senorblanco@chromium.org,enga@chromium.org
Change-Id: I4587b457b7b2dd5d3c7457065bf4e77b95af59d7
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Bug: none
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/29442
Reviewed-by: Kai Ninomiya <kainino@chromium.org>
Commit-Queue: Kai Ninomiya <kainino@chromium.org>
In situations where both dawn_wire and dawn_native are used on separate
threads (Chrome with --single-process or --in-process-gpu), it's
desirable to have a per-thread proc table so that the WebGPU C++ API can
still be used. This eliminates classes of bugs with manual
reference/release errors.
This also changes many of the GetProcs functions to return const
references to the static proc tables known at compile time, instead of a
copy.
Bug: none
Change-Id: I8775bb715b312dd9476a1903fbd797d4b1302614
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/29240
Reviewed-by: Stephen White <senorblanco@chromium.org>
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Commit-Queue: Austin Eng <enga@chromium.org>
In the client code, we often need to translate between WGPUObject (the
API type) to Object* (the internal client type). This added a bunch of
reinterpret_casts that make the code less readable and more fragile.
This CL adds FromAPI and ToAPI helpers in the autogenerated
ApiObjects_autogen.h header, that convert between API and internal types
in a type-safe way.
Bug: dawn:445
Change-Id: Ia1bf624f0315ced496b95cb660adf88abd916d71
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/24063
Commit-Queue: Corentin Wallez <cwallez@chromium.org>
Reviewed-by: Stephen White <senorblanco@chromium.org>
Reviewed-by: Austin Eng <enga@chromium.org>
This header was only used to call DeviceCreateErrorBuffer and
DeviceInjectError that are used in some handwritten client commands.
We remove the need for the header by making these two commands
handwritten.
This also improves readability, previously injecting errors read:
ClientDeviceInjectError(reinterpret_cast<WGPUDevice>(device),
WGPUErrorType_Validation,
"Some validation message");
And now reads:
device->InjectError(WGPUErrorType_Validation, "Some validation message");
Bug: dawn:445
Change-Id: Ie11570aacf3036e13abd174d91670ecb84661226
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/24080
Commit-Queue: Corentin Wallez <cwallez@chromium.org>
Reviewed-by: Stephen White <senorblanco@chromium.org>
Reviewed-by: Austin Eng <enga@chromium.org>
This CL modifies code generation so that the generated client procs call
the handwritten methods on client objects directly.
Previously the flow was:
- wgpuBarDoStuff
- ClientBarDoStuff
- ClientHandwrittenBarDoStuff
- client::Bar::DoStuff
With this CL the flow is:
- wgpuBarDoStuff
- ClientBarDoStuff
- client::Bar::DoStuff
This required adding Buffer creation methods on client::Device instead
of calling client::Buffer static methods directly.
Bug: dawn:445
Change-Id: I1b332b71ac7a03685afcf8fd0617d3d27da468cf
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/24062
Commit-Queue: Corentin Wallez <cwallez@chromium.org>
Reviewed-by: Stephen White <senorblanco@chromium.org>
Reviewed-by: Austin Eng <enga@chromium.org>
This is a temporary fix until dawn_wire can support multiple devices.
When using objects from different devices, the wire will inject an
error into the receiver object's device. Methods that return objects
will return a dummy object. Using the dummy object will cause a fatal
error on the server.
Without this fix, the server would blindly lookup an ObjectId which
could point to some other object. This would bypass same-device
validation and have incorrect results.
Bug: dawn:383
Change-Id: I898f07d4b26f2a97ef952b82af488e7f807c36f0
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/19261
Commit-Queue: Austin Eng <enga@chromium.org>
Reviewed-by: Kai Ninomiya <kainino@chromium.org>
This is a more accurate name and conflicts less with the callback
request serials.
Bug: none
Change-Id: I0f9660c24468064dadffb3ab9b3392d403f93c41
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/19260
Commit-Queue: Austin Eng <enga@chromium.org>
Reviewed-by: Kai Ninomiya <kainino@chromium.org>
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Also fixes some warnings when compiling with GCC
Bug: dawn:333
Change-Id: Ib597bb3b950476279a1e20e3556765ec9f1db697
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/15960
Commit-Queue: Kai Ninomiya <kainino@chromium.org>
Reviewed-by: Kai Ninomiya <kainino@chromium.org>
This is the first step in making the API before WGPUDevice creation
match webgpu.h and is necessary to implement WGPUSwapChain.
BUG=dawn:269
Change-Id: If92ced42d7683d79e67c02738949ff8b483d22c4
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/14061
Commit-Queue: Corentin Wallez <cwallez@chromium.org>
Reviewed-by: Kai Ninomiya <kainino@chromium.org>
This was used to make the distinction between native-only methods that
were those manipulating "natively defined" types, and the rest. Now that
all "natively defined" objects are "callback" instead this name didn't
make sense.
The only relevant thing is that in C there are the Reference and Release
methods that don't appear in dawn.json and shouldn't be exposed on C++
objects. Hence most of the native_methods() calls in the templates are
updated to be c_methods() calls except for the webgpu_cpp templates that
use type.methods directly.
BUG=dawn:22
Change-Id: I65c160b8b8a829e4728862c65bc67268a46f445e
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/13902
Commit-Queue: Corentin Wallez <cwallez@chromium.org>
Reviewed-by: Jiawei Shao <jiawei.shao@intel.com>
webgpu.h is the "official" header for WebGPU in native and is being
developed in https://github.com/webgpu-native/webgpu-headers
dawn.h and dawncpp.h are changed to become webgpu.h and webgpu_cpp.h
respectively and use the new naming convention. New dawn.h and dawncpp.h
headers are created that just proxy the types, constants and functions
to their WebGPU counterpart.
Almost no naming change is done in Dawn in this commit, which help check
that the proxying headers work correctly. A couple changes were
necessary, in particular for tests of the internal of dawncpp.h, and a
workaround for a standard library bug for std::underlying_type was
removed because it is no longer needed and got in the way.
Finally since some templates were renamed to match the name of the file
they create instead of using the generic "api" name.
BUG=dawn:22
Change-Id: I12ee22d0b02ccb5b8a52ceccabb3e63ce74da007
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/12480
Commit-Queue: Corentin Wallez <cwallez@chromium.org>
Reviewed-by: Kai Ninomiya <kainino@chromium.org>
This will become wgpuGetProcAddress that is part of the webgpu.h and the
last gap in functionality for dawn.h to match webgpu.h.
BUG=dawn:22
Change-Id: I0dcb3b5e6bd99cb10db273fc101d3ec0161b7da0
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/12120
Commit-Queue: Corentin Wallez <cwallez@chromium.org>
Reviewed-by: Kai Ninomiya <kainino@chromium.org>
This prevents bugs where the compiler assumes a piece of memory
will be the same if read from twice.
Bug: dawn:230
Change-Id: Ib3358e56b6cf8f1fbf449c5d564ef85c969d695b
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/11840
Reviewed-by: Kai Ninomiya <kainino@chromium.org>
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Commit-Queue: Austin Eng <enga@chromium.org>
This same callback will be used for push/pop error scope.
Bug: dawn:153
Change-Id: I2771539e13f8a4e6a59f13c8082689d25ba44905
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/10460
Commit-Queue: Austin Eng <enga@chromium.org>
Reviewed-by: Kai Ninomiya <kainino@chromium.org>
This removes blocks of code that were obviously builder-specific but
also removes the ObjectStorage::valid member that was used to implement
the maybe monad on the wire server side. This is no longer needed since
dawn_native handles the maybe monad internally now.
BUG=dawn:125
Change-Id: I8c30daae9fc70853bc1996d85a860b4877c5976c
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/6161
Reviewed-by: Austin Eng <enga@chromium.org>
Reviewed-by: Kai Ninomiya <kainino@chromium.org>
Commit-Queue: Corentin Wallez <cwallez@chromium.org>
This is to match Chromium style.
Change-Id: Ic97cc03e2291c653ade9662ba3d5e629872b10ad
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/5482
Reviewed-by: Kai Ninomiya <kainino@chromium.org>
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Commit-Queue: Austin Eng <enga@chromium.org>
This moves ownership of ObjectAllocators and CommandSerializers
from the Device to the Client. There may be also be multiple
Devices, so New() now takes the Device the object belongs to.
Device allocation specializes New() to take the owning Client so
that we can get a pointer to the Client from any Dawn API object.
Bug: dawn:88
Change-Id: Ie4274d46313884c44a857159e95d236dc1141c0c
Reviewed-on: https://dawn-review.googlesource.com/c/4001
Commit-Queue: Austin Eng <enga@chromium.org>
Reviewed-by: Kai Ninomiya <kainino@chromium.org>
This patch copies methods and classes from WireClient.cpp and
distributes them over multiple files. Headers and forward declarations
are added as necessary, but no logical changes are made.
Bug: dawn:88
Change-Id: I9d0fcaeb67e789c3ad0597eaa3839adefc3ac5f9
Reviewed-on: https://dawn-review.googlesource.com/c/3980
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Commit-Queue: Austin Eng <enga@chromium.org>
Austin Eng
Ivan Murashov
Kai Ninomiya
Corentin Wallez
Stephen White