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 /arm_compute/core/CL | |
| 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 'arm_compute/core/CL')
| -rw-r--r-- | arm_compute/core/CL/kernels/CLGEMMMatrixMultiplyKernel.h | 7 |
1 files changed, 5 insertions, 2 deletions
diff --git a/arm_compute/core/CL/kernels/CLGEMMMatrixMultiplyKernel.h b/arm_compute/core/CL/kernels/CLGEMMMatrixMultiplyKernel.h index e030fa2d2a..f61c330de6 100644 --- a/arm_compute/core/CL/kernels/CLGEMMMatrixMultiplyKernel.h +++ b/arm_compute/core/CL/kernels/CLGEMMMatrixMultiplyKernel.h @@ -59,9 +59,11 @@ public: * @param[in] alpha Weight of the matrix product * @param[in] is_interleaved_transposed (Optional) True if input0 and input1 have been reshaped respectively using @ref CLGEMMInterleave4x4Kernel and @ref CLGEMMTranspose1xWKernel * @param[in] reshape_info (Optional) GEMM reshape info. If is_interleaved_transposed = true, this object must contain the information to understand how the matrix A and matrix B have been reshaped + * @param[in] fp_mixed_precision (Optional) Use wider accumulators (32 bit instead of 16 for FP16) to improve accuracy * */ - void configure(const ICLTensor *input0, const ICLTensor *input1, ICLTensor *output, float alpha, bool is_interleaved_transposed = true, const GEMMReshapeInfo &reshape_info = GEMMReshapeInfo()); + void configure(const ICLTensor *input0, const ICLTensor *input1, ICLTensor *output, float alpha, bool is_interleaved_transposed = true, const GEMMReshapeInfo &reshape_info = GEMMReshapeInfo(), + bool fp_mixed_precision = false); /** Static function to check if given info will lead to a valid configuration of @ref CLGEMMMatrixMultiplyKernel * * @param[in] input0 Input tensor containing the Matrix A. Data types supported: F16/F32 @@ -71,11 +73,12 @@ public: * @param[in] is_interleaved_transposed True if input0 and input1 have been reshaped respectively using @ref CLGEMMInterleave4x4Kernel and @ref CLGEMMTranspose1xWKernel * @param[in] reshape_info GEMM reshape info. If is_interleaved_transposed = true, this object must contain the information to understand how the matrix A and matrix B have been reshaped * @param[in] gpu_target GPU Target + * @param[in] fp_mixed_precision (Optional) Use wider accumulators (32 bit instead of 16 for FP16) to improve accuracy * * @return a status */ static Status validate(const ITensorInfo *input0, const ITensorInfo *input1, const ITensorInfo *output, float alpha, bool is_interleaved_transposed, const GEMMReshapeInfo &reshape_info, - GPUTarget gpu_target); + GPUTarget gpu_target, bool fp_mixed_precision = false); // Inherited methods overridden: void run(const Window &window, cl::CommandQueue &queue) override; |