diff options
Diffstat (limited to 'src/core/gpu/cl/kernels/ClSoftmaxKernel.cpp')
-rw-r--r-- | src/core/gpu/cl/kernels/ClSoftmaxKernel.cpp | 10 |
1 files changed, 10 insertions, 0 deletions
diff --git a/src/core/gpu/cl/kernels/ClSoftmaxKernel.cpp b/src/core/gpu/cl/kernels/ClSoftmaxKernel.cpp index 000c9ad04d..1dd905d66e 100644 --- a/src/core/gpu/cl/kernels/ClSoftmaxKernel.cpp +++ b/src/core/gpu/cl/kernels/ClSoftmaxKernel.cpp @@ -154,6 +154,11 @@ const unsigned int ClLogits1DMaxShiftExpSumKernel::_serial_vector_size = 8; /**< Vector size in the parallel case (obtained through auto-tuning, enables the best memory access pattern for Bifrost) .*/ const unsigned int ClLogits1DMaxShiftExpSumKernel::_parallel_vector_size = 4; +ClLogits1DMaxShiftExpSumKernel::ClLogits1DMaxShiftExpSumKernel() +{ + _type = CLKernelType::ELEMENTWISE; +} + void ClLogits1DMaxShiftExpSumKernel::configure(const CLCompileContext &compile_context, const ITensorInfo &src, ITensorInfo &max, ITensorInfo &dst, ITensorInfo &sum, const SoftmaxKernelInfo &info) { auto padding_info = get_padding_info({ &src, &max, &dst, &sum }); @@ -273,6 +278,11 @@ void ClLogits1DMaxShiftExpSumKernel::run_op(ITensorPack &tensors, const Window & while(window_collapsed.slide_window_slice_3D(slice)); } +ClLogits1DNormKernel::ClLogits1DNormKernel() +{ + _type = CLKernelType::ELEMENTWISE; +} + void ClLogits1DNormKernel::configure(const CLCompileContext &compile_context, const ITensorInfo &src, const ITensorInfo &sum, ITensorInfo &dst, const SoftmaxKernelInfo &info) { auto padding_info = get_padding_info({ &src, &dst, &sum }); |