diff options
| author | Vidhya Sudhan Loganathan <vidhyasudhan.loganathan@arm.com> | 2018-11-16 11:33:12 +0000 |
|---|---|---|
| committer | Georgios Pinitas <georgios.pinitas@arm.com> | 2018-11-16 17:37:40 +0000 |
| commit | a25d16c86f0d870408bc8b941aa755093417b0f0 (patch) | |
| tree | b62d145a4e5009d894262a7ffa66cdba8260bb03 /tests/SimpleTensor.h | |
| parent | a7b54f44e2bf133179f24a34007bc93237dd2265 (diff) | |
| download | ComputeLibrary-a25d16c86f0d870408bc8b941aa755093417b0f0.tar.gz | |
COMPMID-1266 : Add support for FP16 in CLWinogradConvolutionLayer: 5x5 kernels
Introduced F32 accumulation for F16 winograd gemm and output transform
WinogradConvolution will be available for F16 only if fast math flag is enabled
Change-Id: I215593c205236a0f9669218437bb40b184ec6a4f
Diffstat (limited to 'tests/SimpleTensor.h')
| -rw-r--r-- | tests/SimpleTensor.h | 39 |
1 files changed, 39 insertions, 0 deletions
diff --git a/tests/SimpleTensor.h b/tests/SimpleTensor.h index 335ef9130a..dd4a8bee2c 100644 --- a/tests/SimpleTensor.h +++ b/tests/SimpleTensor.h @@ -220,6 +220,45 @@ protected: DataLayout _data_layout{ DataLayout::UNKNOWN }; }; +template <typename T1, typename T2> +SimpleTensor<T1> copy_tensor(const SimpleTensor<T2> &tensor) +{ + SimpleTensor<T1> st(tensor.shape(), tensor.data_type(), + tensor.num_channels(), + tensor.quantization_info(), + tensor.data_layout()); + for(size_t n = 0; n < size_t(st.num_elements()); n++) + { + st.data()[n] = static_cast<T1>(tensor.data()[n]); + } + return st; +} + +template <typename T1, typename T2, typename std::enable_if<std::is_same<T1, T2>::value, int>::type = 0> +SimpleTensor<T1> copy_tensor(const SimpleTensor<half> &tensor) +{ + SimpleTensor<T1> st(tensor.shape(), tensor.data_type(), + tensor.num_channels(), + tensor.quantization_info(), + tensor.data_layout()); + memcpy((void *)st.data(), (const void *)tensor.data(), size_t(st.num_elements() * sizeof(T1))); + return st; +} + +template < typename T1, typename T2, typename std::enable_if < (std::is_same<T1, half>::value || std::is_same<T2, half>::value), int >::type = 0 > +SimpleTensor<T1> copy_tensor(const SimpleTensor<half> &tensor) +{ + SimpleTensor<T1> st(tensor.shape(), tensor.data_type(), + tensor.num_channels(), + tensor.quantization_info(), + tensor.data_layout()); + for(size_t n = 0; n < size_t(st.num_elements()); n++) + { + st.data()[n] = half_float::detail::half_cast<T1, T2>(tensor.data()[n]); + } + return st; +} + template <typename T> SimpleTensor<T>::SimpleTensor(TensorShape shape, Format format) : _buffer(nullptr), |