Victor Lomuller

Some devices can manage global values to push and pull different values. In the CUDA runtime, this is covered by cudaMemcpyToSymbol/cudaMemcpyFromSymbol functions. But other APIs may offer similar functionalities. This...

Update the specialization constant extension proposal to align with the module proposal. This PR introduce a `kernel_handler` which is an optional argument to parallel_for functors and allow the user to...

The document present the description of the address space representation in SYCL 1.2.1 and then present the list of current ambiguities the address space deduction rules. A concrete proposal on...

[SYCL] Don't set PI_USM_INDIRECT_ACCESS if platform don't support it

1

If the OpenCL platform doesn't support USM, don't set PI_USM_INDIRECT_ACCESS exec info. This will avoid SYCL program to fail when they don't use USM. If the program do need USM...

[SYCL] Add a CUDA compatibilty mode

3

This patch enables CUDA mode at the same time as the SYCL mode. This allows the compiler to define CUDA macros and add implicit defines. To enable the mode the...

Hierarchical: LowerWG pass ignores convergent functions

1

When I compile this code: ``` #include using namespace cl::sycl; int main() { queue q(default_selector().select_device()); auto buf = cl::sycl::buffer(cl::sycl::range(1)); q.submit([&](handler &cgh) { auto globalAcc = buf.get_access(cgh); auto sizeAcc = buf.get_access(cgh);...

https://github.com/intel/llvm/pull/15061 introduce a new property `work_group_static_memory` which allow the user to set a given amount of local memory to be used. In order to pass this information to the adaptor,...

The patch partially implements `work_group_static` and update proposal. Implemented: - `work_group_static` to handle static allocation in kernel. - `get_dynamic_work_group_memory` to handle runtime allocation, but only on CUDA `work_group_static` is implemented...