jinge90
jinge90
Signed-off-by: jinge90
Signed-off-by: jinge90
Signed-off-by: jinge90
Some deep learning framework uses '__nv_rcp64h' in CUDA backend. We need to provide equivalent functionality in DPC++ compiler.
Adds new simd emulate functions.
Currently, unified runtime L0 adapter will only expose bfloat16 conversion extension on PVC device: https://github.com/oneapi-src/unified-runtime/blob/095e8464124a48c8ed4b995403e754254c072143/source/adapters/level_zero/device.cpp#L292 For other device supporting native bfloat16 conversions(BMG, DG2, LunarLake), checking bfloat16 conversion extension will fail....
I went through the source and found jemalloc has already included some code for riscv support but couldn't find any doc/description to build it for riscv target on a x86_64...
We decide to remove sycl device library jit link path, the 1st step is to mark all related driver option as 'deprecated' and cleared all tests using these options.