Currently we use offset to calculate the index of the queries in
timestamp compute shader, which is incorrect. The offset is the buffer
offset where we start to write the query results, and has nothing to
do with query index. In the query availability detection, the query
index should be based on the parameter firstQuery.
Add new test for resolving a timestamp query twice to the same
destination buffer with potentially overlapping ranges.
Bug: dawn:434
Change-Id: I2b5c5b192cf5d987ac48187e8240a25937957f51
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/50760
Reviewed-by: Austin Eng <enga@chromium.org>
Commit-Queue: Hao Li <hao.x.li@intel.com>
Previously all uses of reentrant object creation in Dawn native
needed to manually AcquireRef. Change them to use CreateFooInternal that
returns a ResultOrError<Ref<>> and are renamed to CreateFoo.
Bug: dawn:723
Change-Id: Ifcda3659d02cc5a4c63c248dc53af7fee7c4a61d
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/46626
Commit-Queue: Corentin Wallez <cwallez@chromium.org>
Reviewed-by: Stephen White <senorblanco@chromium.org>
In timestamp compute shader, we will create an internal buffer for
resolving QuerySet and use it as input buffer in compute shader,
the user-provided resolve buffer is used as output buffer.
This will cause the buffer zero initialization to be called twice,
one is the internal buffer is zero initialized in ResolveQuerySet,
antoher is the user-provided buffer is tracked as pass resource
and need to be initialized. But for ResolveQuerySet(), we expect
there is only once.
We have no special requirements to have an internal buffer. It is
possible to directly use the user-provided buffer for read and
write becuase it will get STORAGE_INTERNAL usage.
Bug: dawn:434
Change-Id: Ia8c8ac6e9ba23fea31468a6d9b4580eece189be2
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/36201
Commit-Queue: Hao Li <hao.x.li@intel.com>
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Because the uint64 is not supported on all GPU drivers, we use uint32
and float to simulate the multiplication of uint64, but there is
accuracy loss between the results and the expected results computed by
uint64. This test checks that the accuracy loss is less than 0.2%.
Bug: dawn:434
Change-Id: I6f5c842b6915f101441886bdfa4f9feb2827d174
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/34120
Commit-Queue: Hao Li <hao.x.li@intel.com>
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
