Fraser Cormack

Fix race condition in OpenCL kernel

1

Description ----------- Without the barrier at the end of barrierOR, it is possible for work-item 0 to start the next loop iteration and update predicates[0] while other work-items are still...

Clarification on zero sizes in clEnqueueXXXBuffer APIs

2

This might be a simple question to answer as it's arguably the status quo in the specification, but because it has caused a discussion internally I'd like to ask for...

[HIP] Implement urDeviceGetNativeHandle

1

This is mostly just a copy of the CUDA version of this implementation.

When configured without match files, the CTS was trying to run a command and its arguments as a single command, as it was represented by a single string variable. It...

[HIP] Add support for querying program binary type

2

Also trim down a number of HIP 'program' match entries which appear to be passing now.

[SYCL][NVPTX] Emit 'grid_constant' annotations for by-val kernel params

1

Also fix up the `DeadArgumentElimination` passes to correctly preserve the annotations; when removing arguments from functions, dead parameters need pruned and alive ones may need their values shifted down by...

We need this to successfully compile the subgroup device_code test.

Copies code for the pool descriptor from the CUDA adapter. Fixes https://github.com/oneapi-src/unified-runtime/issues/1479.

[SYCL] Add max work-group size kernel properties

4

This patch adds two kernel properties to allow users to specify the maximum work-group size that a kernel will be invoked with. The `max_work_group_size` property corresponds to the `intel::max_work_group_size` function...