In order to preserve padding properly for MSL, we need to use its
packed_vec type for all vec3 types in storage buffers, not just struct
members. This commit includes a complete rewrite of the PackedVec3
transform to achieve this. The key details are:
* An internal `__packed_vec3<>` type was added, which corresponds to a
`type::Vector` with an additional flag to indicate that it will be
emitted as packed vector.
* The `PackedVec3` transform replaces all vec3 types used in
host-shareable address spaces with the internal `__packed_vec3`
type. This includes vec3 types that appear as the store type of a
pointer.
* When used as an array element, these `__packed_vec3` types are
wrapped in a struct that contains a single `__packed_vec3`
member. This allows us to add an `@align()` attribute that ensures
that `array<vec3<T>>` still has the correct array element stride.
* When the `vec3<T>` appears as a struct member in the input program,
we apply the `@align()` to that member to ensure that we do not
change its offset.
* Matrix types with three rows that are used in memory are replaced
with an array of columns, where each column uses a `__packed_vec3`
inside an aligned wrapper structure as above.
* Accesses to host-shareable memory that involve any of these types
invoke a "pack" or "unpack" helper function to convert them to the
equivalent type that uses `__packed_vec3` or a regular `vec3` as
required.
* The `chromium_internal_relaxed_uniform_layout` extension is used to
avoid issues where modifying a type in the uniform address space
triggers stricter layout validation rules.
Bug: tint:1571
Fixed: tint:1837
Change-Id: Idaf2da2f5bcb2be00c85ec657edfb614186476bb
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/121200
Reviewed-by: Ben Clayton <bclayton@google.com>
Commit-Queue: James Price <jrprice@google.com>
Kokoro: Kokoro <noreply+kokoro@google.com>
Only load extension entry points if the extension is supported.
From the eglGetProcAddress manpage:
"A non-NULL return value does not guarantee that an extension function
is actually supported at runtime. The client must also make a
corresponding query, such as glGetString(GL_EXTENSIONS) for OpenGL and
OpenGL ES extensions [...] to determine if a function is supported by
EGL or a specific client API context."
This required moving extension initialization from OpenGLFunctions
into OpenGLFunctionsBase.
Change-Id: Ib4e8360ba455818701990b4476689b651d097ca8
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/121760
Commit-Queue: Stephen White <senorblanco@chromium.org>
Kokoro: Kokoro <noreply+kokoro@google.com>
Reviewed-by: Austin Eng <enga@chromium.org>
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Also increased stack size for windows + debug GN builds of
tint_unittests to 4 MB, which matches what we do in the CMake build.
Required, otherwise some of my new tests stack overflow.
Bug: tint:1209
Change-Id: I3b98000a989aa8b42b20cc4e2219c91887e52451
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/121360
Reviewed-by: Ben Clayton <bclayton@google.com>
Kokoro: Kokoro <noreply+kokoro@google.com>
If buffer is released and the external refcount reaches 0 while
buffer map state is pending map in Wire, the map async callback
is fired with destroyed-before-callback status from the buffer
destructor.
It is possible to call another MapAsync in the callback. At that
time the pointer is still valid because the internal refcount is
not 0 yet.
The behavior of MapAsync should be undefined if external refcount
is 0. This commit adds an assert to check whether external
refcount is 0 in Buffer::MapAsync() in Wire. Ending up with
assertion error may be reasonable.
This commit also adds protected GetRefcount() method to
ObjectBase to allow derived classes to check the refcount.
bug: dawn:1624
Change-Id: I95411a7be2093ba7bb2bb45b466f17f1ebac0ca9
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/119961
Kokoro: Kokoro <noreply+kokoro@google.com>
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Commit-Queue: Corentin Wallez <cwallez@chromium.org>
Avoid linearly searching for prefix_i. This becomes very slow when
adding the same prefix many times. Instead, cache each prefix and the
last index that was assigned, and start searching from there.
Note that this locally speeds up MaxLimitTests.MaxBufferBindingSize on
GL backends from about 10 minutes to a few seconds. This test creates a
very large struct, and Tint's GLES backend runs the PadStructs
transform, which produces a struct with 16384 members. Creating these
members triggers the behaviour described above when creating names for
each member.
Also replaced std::unordered_maps with utils::Hashmap.
Change-Id: I86678a049ac229ce617d500971e9c077ff5c10a1
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/121440
Reviewed-by: Ben Clayton <bclayton@google.com>
Commit-Queue: Antonio Maiorano <amaiorano@google.com>
Kokoro: Kokoro <noreply+kokoro@google.com>
Change the DecomposeMemoryAccess to behave more like the DirectVariableAccess transform, in that it'll inline the access of buffer variable into the load / store helper functions, instead of passing the array down.
This avoids large array copies observed with FXC, which can have *severe* performance costs.
Fixed: tint:1819
Change-Id: I52eb3f908813f72ab9da446743e24a2637158309
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/121460
Kokoro: Kokoro <noreply+kokoro@google.com>
Auto-Submit: Ben Clayton <bclayton@google.com>
Reviewed-by: James Price <jrprice@google.com>
Commit-Queue: James Price <jrprice@google.com>
Instead, compile pipelines synchronously.
When backend validation is on with Mac AMD, the driver crashes during
pipeline compilation. This seems to only occur when partition alloc
is enabled.
Bug: dawn:1200
Change-Id: I520087eeae9c356990962d0c4a96eba017735af1
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/121400
Commit-Queue: Austin Eng <enga@chromium.org>
Kokoro: Kokoro <noreply+kokoro@google.com>
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
ExternalTexture has active, expired and destroyed states.
Only active state external texture is valid to submit.
Expired state external texture can be refresh to active but
destroyed external texture cannot be refresh.
Bug: chromium:1412338
Change-Id: Ic7f12d274d27b644f19ec3ef8b46c110610afa2b
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/120982
Kokoro: Kokoro <noreply+kokoro@google.com>
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Commit-Queue: Shaobo Yan <shaobo.yan@intel.com>
On Intel Gen9 and Gen9.5 GPUs there is a D3D12 driver bug about
using DstAlpha as the source blend factors in both color and alpha
blending.
Although we cannot add workaround for all such cases, we have
confirmed that when the blend operation is 'add', and the
destination blend factors are ‘zero', we can workaround this driver
bug by replacing 'add' to 'minus'. We cannot add workaround when
destination blend factor is another value.
This patch adds such workaround on the affected platforms.
Bug: dawn:1579
Test: dawn_end2end_tests
Change-Id: If7c82b9d559b876e42fb36e2f539f9ff2f71ea91
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/120981
Commit-Queue: Jiawei Shao <jiawei.shao@intel.com>
Kokoro: Kokoro <noreply+kokoro@google.com>
Reviewed-by: Austin Eng <enga@chromium.org>
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Validate as the bind group lazy aspect at each dispatch/draw
call.
Use nested loops to iterate through each bind group and binding
to find if any aliasing exists, which has time complexity of
O(N^2) and can be further optimized to use O(NlogN) algorithm.
Bug: dawn:1642
Change-Id: I8c43128cdeea75352c194752fb22258b6a73430e
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/118440
Commit-Queue: Shrek Shao <shrekshao@google.com>
Reviewed-by: Loko Kung <lokokung@google.com>
Kokoro: Kokoro <noreply+kokoro@google.com>
- Plumbs all descriptors through so that the objects act much more like
a real implementation.
- Lots of changes to DestroyObjectTests.cpp to update usages and test
through the APIs when applicable.
- Put in defaults for the mock device and make it as easy to use as
possible.
Change-Id: I85b243a18ec1872aff0172549aec0f599967ea0e
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/119821
Reviewed-by: Austin Eng <enga@chromium.org>
Kokoro: Kokoro <noreply+kokoro@google.com>
Commit-Queue: Loko Kung <lokokung@google.com>
The dependency graph no longer errors if a symbol cannot be resolved, instead the ResolvedIdentifier now has an unresolved variant.
This is required as the second resolve phase only has the full context of the identifier usage, to provide the hints.
Also: Split Slice out of the utils/vector.h, so it can be used as a lightweight view over static data.
Fixed: tint:1842
Change-Id: I31fa7697790be24c35b7e4fab5ca903c8a7afbba
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/121020
Commit-Queue: Ben Clayton <bclayton@chromium.org>
Kokoro: Kokoro <noreply+kokoro@google.com>
Commit-Queue: Ben Clayton <bclayton@google.com>
Reviewed-by: Dan Sinclair <dsinclair@chromium.org>
Query D3D12_FEATURE_DATA_D3D12_OPTIONS13 to determine whether the device
supports copies between textures of different dimensions. If it does
not, or if D3D12_FEATURE_D3D12_OPTIONS13 is not supported, force enable
toggle
D3D12UseTempBufferInTextureToTextureCopyBetweenDifferentDimensions.
Fixed: dawn:1216
Change-Id: I976e7fad291126f8dcee31ce6b681314d38e69e7
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/120882
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Commit-Queue: Austin Eng <enga@chromium.org>
Kokoro: Kokoro <noreply+kokoro@google.com>
Auto-Submit: Austin Eng <enga@chromium.org>
