This repository has been archived by the owner. It is now read-only.
Tags: yuriykoch/llvm
Tags
[Driver][SYCL] Update default SPIR device arch to correlate with host (… …intel#4936) The default target arch for the device compilation when using -fsycl should match the target arch of the host compilation. x86_64 (64-bit) -> spir64 i686 (32-bit) -> spir Adjust locations where this is set (offload and device-only compiles) and modify the corresponding tests.
[SYCL] Eliminate leak of programs from program/kernel in-memory cache (… …intel#4949) Signed-off-by: Sergey Kanaev <sergey.kanaev@intel.com>
[SYCL][DOC] Proposal to add number of HW threads per EU query (intel#… …4876) It follows https://spec.oneapi.io/level-zero/latest/core/api.html#ze-device-thread-t Also the patch removes several 'supported only on L0 disclaimers' since recently an OpenCL extension was released: https://github.com/KhronosGroup/OpenCL-Docs/blob/master/extensions/cl_intel_device_attribute_query.asciidoc Signed-off-by: Dmitry Sidorov <dmitry.sidorov@intel.com>
[SYCL] Fix the unused parameter warnings in the level_zero backend (i… …ntel#4938) When the compiler is built with -Werror argument, the following errors occur: include/sycl/ext/oneapi/backend/level_zero.hpp:125:26: error: unused parameter 'Handler' [-Werror,-Wunused-parameter] const async_handler &Handler) { ^ include/sycl/ext/oneapi/backend/level_zero.hpp:136:55: error: unused parameter 'Handler' [-Werror,-Wunused-parameter] const context &TargetContext, const async_handler Handler) { ^ 2 errors generated. In order to not break the API, the parameters is still declared in the function's signatures but marked as unused via the (void)Handler expressions.
[SYCL] Fix for undefined Result on some code paths. (intel#4924) Signed-off-by: Rajiv Deodhar <rajiv.deodhar@intel.com>
[SYCL][CUDA] Set cuda_piextUSMEnqueuePrefetch to fail for windows (in… …tel#4908) `cuMemPrefetchAsync` currently has issues on Windows machines, this appears to be a problem with CUDA and not the plugins usage of the function. I think `cuda_piextUSMEnqueuePrefetch` which uses `cuMemPrefetchAsync` should be set to fail for windows and send a message to the user that this functionality is not available on windows devices. More details: intel#4809
[SYCL][CUDA][Matrix] Add initial support for Tensorcore matrix ext (i… …ntel#4696) Initial Implementation based on the new matrix extension supporting Nvidia Tensorcore, intel#4695, that is adapted from the AMX matrix extension. Only double data type matrix elements are initially supported. Signed-off-by: jack.kirk <jack.kirk@codeplay.com>
[ESIMD] Update vc-intrinsics repo hash tag. (intel#4894) Signed-off-by: Konstantin S Bobrovsky <konstantin.s.bobrovsky@intel.com>
[NFC][SYCL] Update testing dependencies for append-file (intel#4890) Certain tests are dependent on the append-file tool for a successful run. Add the dependencies so the binary is available when needed.
[ESIMD] Refactor vc-intrinsics build for easier sync with custom buil… …ds. (intel#4879) Signed-off-by: Konstantin S Bobrovsky <konstantin.s.bobrovsky@intel.com>
PreviousNext