Kealan Barbieri
Kealan Barbieri
`CL_DEVICE_HALF_FP_ATOMIC_CAPABILITIES_EXT` is the limiting factor for `get_ocl_device_enabled_native_float_atomics`. My understanding is it requires OpenCL 2.0 driver support (per [extension doc](https://registry.khronos.org/OpenCL/extensions/ext/cl_ext_float_atomics.html)), which is in line with the requirements in our README. I'm...
Submitted a trivial change that can eliminate this kind of issue without degrading the associated functionality. Thanks @rjoursler for the suggestion!
This is due to limited support for gen9 platforms in the optimized conv implementation. I submitted a small PR to provide a workaround for such cases.
@Simonsays095 there are some cases where `kInterleave` is set but `kInterleaveChunk` is not so in `driverInfo` unroll for `K` dim actually gets set to 0 after its been initialized to...
make test set test_scope=NIGHTLY disable test_device_cpu disable benchdnn_all enable benchdnn_matmul
make test perf-gpu set primitive=matmul ip
make test set test_scope=NIGHTLY disable test_device_cpu disable benchdnn_all enable benchdnn_matmul
make test perf-gpu set primitive=matmul ip
make test set test_scope=NIGHTLY disable test_device_cpu disable benchdnn_all enable benchdnn_matmul
make test set test_scope=NIGHTLY disable test_device_cpu disable benchdnn_all enable benchdnn_matmul