From d6afedc775220f17317f1835a4d18b72a54525de Mon Sep 17 00:00:00 2001 From: Chunosov Date: Mon, 6 Nov 2017 22:09:45 +0700 Subject: COMPMID-661: softmax-fp32 optimisation (#14) Change-Id: I2007af1ed9dcf68065cf412aa50f73a2025b31a6 Reviewed-on: http://mpd-gerrit.cambridge.arm.com/94605 Reviewed-by: Gian Marco Iodice Tested-by: Kaizen --- arm_compute/runtime/CL/functions/CLSoftmaxLayer.h | 18 ++++++++++-------- 1 file changed, 10 insertions(+), 8 deletions(-) (limited to 'arm_compute/runtime/CL/functions') diff --git a/arm_compute/runtime/CL/functions/CLSoftmaxLayer.h b/arm_compute/runtime/CL/functions/CLSoftmaxLayer.h index d84297e9a1..72ef679d6a 100644 --- a/arm_compute/runtime/CL/functions/CLSoftmaxLayer.h +++ b/arm_compute/runtime/CL/functions/CLSoftmaxLayer.h @@ -54,8 +54,8 @@ public: /** Set the input and output tensors. * * @param[in] input Source tensor. Data types supported: QS8/QS16/F16/F32 - * @param[in] beta A scaling factor for the exponent. * @param[out] output Destination tensor. Data types supported: same as @p input + * @param[in] beta (Optional) A scaling factor for the exponent. Defaults to 1.f */ void configure(const ICLTensor *input, ICLTensor *output, float beta = 1.0f); @@ -63,13 +63,15 @@ public: void run() override; private: - CLMemoryGroup _memory_group; - CLLogits1DMaxKernel _max_kernel; - CLLogits1DShiftExpSumKernel _shift_exp_sum_kernel; - CLLogits1DNormKernel _norm_kernel; - CLTensor _max; - CLTensor _sum; - CLTensor _tmp; + CLMemoryGroup _memory_group; + CLLogits1DMaxKernel _max_kernel; + CLLogits1DShiftExpSumKernel _shift_exp_sum_kernel; + CLLogits1DMaxShiftExpSumKernel _max_shift_exp_sum_kernel; + CLLogits1DNormKernel _norm_kernel; + CLTensor _max; + CLTensor _sum; + CLTensor _tmp; + bool _run_legacy_path; }; } #endif /* __ARM_COMPUTE_CLSOFTMAXLAYER_H__ */ -- cgit v1.2.1