For Dawn/Chrome interop, Chrome OS differs from Linux as it uses the
NativePixmap-based SharedImageBacking which stores image memory as
dma-bufs as opposed to opaque FDs. Likewise, for synchronization,
Chrome OS wants to use sync obj FDs as opposed to opaque FDs for
more flexibility.
The motivating difference between sync obj FDs and their opaque
counterparts is how they are created. As their name suggests, Opaque
FDs are mostly meaningless outside of the Vulkan ecosystem and must
be created from a VkDevice. As a result, Linux’s
ExternalVkImageBacking needs the Vulkan implementation to create the
FD even when accessing the SharedImage as a GL texture [1]. On Chrome
OS, however, we don’t guarantee Vulkan outside of Dawn, so we aren’t
able to create the opaque FD directly in Chrome.
Instead, we are always able to create sync objs (e.g. via a
fence [2]) which can be imported as VkSemaphores by simply changing
VkImportSemaphoreFdInfoKHR::handleType. Similarly, we can export
signal VkSemaphores as sync objs as well by updating
VkSemaphoreGetFdInfoKHR::handleType.
This CL adds conditional support for using SYNC_FDs on Chrome OS
when importing/exporting VkSemaphores and renames
SemaphoreServiceOpaqueFD accordingly. With this, we can properly
wait on reads/writes on a GL SharedImage representation before
accessing the same memory in Dawn [3].
[1] https://source.chromium.org/chromium/chromium/src/+/main:gpu/command_buffer/service/external_vk_image_gl_representation.cc;l=75;drc=f887797b3b527feabd5dfe9b3f2cc7f6deade49f
[2] https://source.chromium.org/chromium/chromium/src/+/main:gpu/command_buffer/service/shared_image_backing_gl_image.cc;l=681;drc=f887797b3b527feabd5dfe9b3f2cc7f6deade49f
[3] https://chromium-review.googlesource.com/c/chromium/src/+/3042460
BUG=b:172208313
Change-Id: I5357847fea40e41d1b982054e3573d363e17530c
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/59080
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Reviewed-by: Austin Eng <enga@chromium.org>
Commit-Queue: Brian Ho <hob@chromium.org>
--disable_workgroup_init will disable the workgroup memory zero initiailization. Useful for benchmarking.
--disable_symbol_renaming will disable tint's symbol renamer. Useful for debugging output.
Bug: tint:1003
Fixed: dawn:1016
Change-Id: I92486ef88a2c1112d9ccb40f7920947bd2011c70
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/58861
Commit-Queue: Ben Clayton <bclayton@google.com>
Commit-Queue: Corentin Wallez <cwallez@chromium.org>
Auto-Submit: Ben Clayton <bclayton@google.com>
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
This requires restructuring the logic around MTLCommandBuffer allocation
so that GetPendingCommandContext is guaranteed to never fail. Logic in
the Metal backend is now similar to the Vulkan backend: the
MTLCommandBuffer is prepared at device initialization time, or after a
submission, such that it is always valid.
A new mUsed boolean is added to CommandRecordingContext to say whether
any commands have been recording. Previously mCommandBuffer was used for
that purpose, but it is now always non-null.
Bug: dawn:801
Change-Id: I5dc6747d1e6d538054010cc50533a03a49af921a
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/58720
Auto-Submit: Corentin Wallez <cwallez@chromium.org>
Reviewed-by: Stephen White <senorblanco@chromium.org>
Reviewed-by: Austin Eng <enga@chromium.org>
Commit-Queue: Corentin Wallez <cwallez@chromium.org>
This includes OOM as well as internal driver failures when
creating a view of an MTLTexture. This required changing the code to use
the Create-Initialize pattern used everywhere else.
Bug: dawn:801
Change-Id: Ib8a8dec74141aacfa58a55bb8201a83351b3b739
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/58721
Commit-Queue: Corentin Wallez <cwallez@chromium.org>
Auto-Submit: Corentin Wallez <cwallez@chromium.org>
Reviewed-by: Stephen White <senorblanco@chromium.org>
1. Trigger all deferred callback tasks before registering a new
device-level callback function, making sure that these tasks won't be
invalided due to callback function changing;
2. Fix the end to end testsuit DeviceLostTests, setting the device lost
callback at the beginning of each test so that callback tasks created
during the test will not be triggered unexpectedly.
Bug: chromium:1223390
Bug: chromium:1223603
Bug: chromium:1228134
Change-Id: I2530e938d8fbb2920f3cc6fc78baa01c5d18ad5d
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/56040
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Reviewed-by: Jiawei Shao <jiawei.shao@intel.com>
Commit-Queue: Zhaoming Jiang <zhaoming.jiang@intel.com>
This patch adds the validation rules on the maximum inter-shader
stage variable locations. According to Vulkan SPEC, "Each effective
Location must have a value less than the number of location avaiable for
the given interface", and according to Table 18 "Shader Input and Output
Locations", the number of available vertex output locations is
maxVertexOutputComponents / 4, and the number of available fragment input
locations is maxFragmentInputComponents / 4.
BUG=dawn:802
TEST=dawn_unittests
Change-Id: I73a66ad69c808fbd0b128abaed1ca84f19a7925d
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/58640
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Reviewed-by: Austin Eng <enga@chromium.org>
Commit-Queue: Jiawei Shao <jiawei.shao@intel.com>
Fuzzers like to generate silly long source, and formatting large spans of these can take considerable time.
Only format the diagnostic if it is going to be displayed.
Significantly speeds up some fuzzing tests, fixing some timeouts.
Also add a minor optimization to the formatter repeat() implementation.
Fixed: chromium:1230313
Change-Id: Ib1f6ac0b31010f86cb7f4e1432dc703ecbe52cb0
Reviewed-on: https://dawn-review.googlesource.com/c/tint/+/58841
Auto-Submit: Ben Clayton <bclayton@google.com>
Commit-Queue: Ryan Harrison <rharrison@chromium.org>
Kokoro: Kokoro <noreply+kokoro@google.com>
Reviewed-by: Ryan Harrison <rharrison@chromium.org>
Spvtools recently added support for the vkdebuginfo instructions,
causing roll failures since Tint build files didn't generate the
headers for them. Instead of adding more generator stuff in Tint's
BUILD.gn files, this commit removes everything in favor of directly
referencing spvtools' targets.
In follow-up work, the references to spvtools_internal_config will
be changed to a more narrowly scoped target so that spvtools'
warning suppressions don't get propagated to Tint.
Bug:chromium:1228274
Change-Id: I5e28c9cd978afd3cfbc941c09decf5a6e7e1554a
Reviewed-on: https://dawn-review.googlesource.com/c/tint/+/58840
Commit-Queue: Corentin Wallez <cwallez@chromium.org>
Kokoro: Kokoro <noreply+kokoro@google.com>
Reviewed-by: Ben Clayton <bclayton@google.com>
Change the ShaderModuleBase::MakeError to return a Ref<ShaderModuleBase>
rather than a naked pointer. This Ref will be used and finally detached in
DeviceBase::APICreateShaderModule.
Bug: chromium:1223546
Change-Id: Ie4213f091e37fdc61c6e1f0bb742e8787bcb4eb8
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/58660
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Reviewed-by: Austin Eng <enga@chromium.org>
Commit-Queue: Zhaoming Jiang <zhaoming.jiang@intel.com>
Treat color target format with "float" capabilities as
blendable format and validate when blending is enabled.
Add helpers for checking float16 texture values.
Bug: dawn:726
Change-Id: Icf8c0182e5e9a13523970c84b5af91f395a089af
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/57744
Commit-Queue: Shrek Shao <shrekshao@google.com>
Reviewed-by: Austin Eng <enga@chromium.org>
Adds some handling that ensures correct reflection of texture_external now
that Dawn uses Tint by default. Enables the previously written tests.
dawn: 1004
Change-Id: I8fd43292e9fe243aee037c70fe47b79ace8b672f
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/58540
Commit-Queue: Brandon Jones (Intel) <brandon1.jones@intel.com>
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Previously Dawn was using acquireKey + 1 for the releaseKey but we
changed the code transition to requiring a specific key. We supported
both options for a time but now that Chromium uses a releaseKey, we can
remove the deprecated option.
Bug: chromium:1213977
Change-Id: Iab253673c15b83b107dcc3635ee165cfa09efeb6
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/58383
Commit-Queue: Corentin Wallez <cwallez@chromium.org>
Auto-Submit: Corentin Wallez <cwallez@chromium.org>
Reviewed-by: Austin Eng <enga@chromium.org>
Rework the clamping so that it unifies the logic for arrays, matricies
and vectors. Try to preserve constant signess, and only clamp the values
if they're possibly out of bounds.
Use ConstantValue() instead of scanning for ScalarConstantExpressions.
As ConstantValue() improves, so will the performance of robustness.
Change-Id: I013a67e15f43350d0a57bcd8ba9ae0c1bcb1eaec
Reviewed-on: https://dawn-review.googlesource.com/c/tint/+/58280
Kokoro: Kokoro <noreply+kokoro@google.com>
Reviewed-by: David Neto <dneto@google.com>
Auto-Submit: Ben Clayton <bclayton@google.com>
Commit-Queue: David Neto <dneto@google.com>
