Georgi Mirazchiyski

[SYCL][Ext] Query kernel maximum active work-groups based on occupancy

1

The currently proposed and implemented query is `max_num_work_group_occupancy_per_cu` which retrieves the maximum actively executing workgroups based on compute unit occupancy granularity. This commit also overloads the `max_num_num_work_group_sync` query to take...

[CUDA] Implement urKernelSuggestMaxCooperativeGroupCountExp for Cuda

2

This commit implements the experimental `urKernelSuggestMaxCooperativeGroupCountExp`, for the Cuda adapter, to retrieve the maximum number of cooperative groups that can be launched on the device. Additionally, the changes also cache...

[Cuda] Reintroduce catching and reporting of bad_alloc for event object creation

2

Reintroduces the changes from commit https://github.com/oneapi-src/unified-runtime/commit/c4ae460f779021aa6840ea47a373f6ac336bc589, which were reverted in related merged commit https://github.com/oneapi-src/unified-runtime/commit/1b4a8b852c6b86545b42a71927f88c4fed107217 due to being mistakenly deleted and omitted on rebasing. This happened because the former changes got...

[SYCL] Limit allocation size in atomic_memory_order_seq_cst test to a safe maximum size

1

The `atomic_memory_order_seq_cst.cpp` test only checks `info::device::max_mem_alloc_size` to limit the maximum allocation. It does not check the actual available on device. This lead to allocating exceeding amount of memory that I...

This change simplifies the `getMaxRegistersJitOptionValue` function as it was previously unclear what the return bool value actually meant, because it doesn't really signify whether the value was valid or not...

This `ErrorAdapterNativeCode` can be used to propage the native error code values up the stack from unified-runtime when calling `urAdapterGetLastError`, if needed. This can be helpful in DPC++ sycl runtime...

The three added `gfx7` architectures were previously omitted from the list of supported ones for SYCL targeting AMDGPU. There is no rational for them to be excluded rather than a...