diff options
Diffstat (limited to 'arm_compute/runtime/CL/functions/CLSoftmaxLayer.h')
| -rw-r--r-- | arm_compute/runtime/CL/functions/CLSoftmaxLayer.h | 65 |
1 files changed, 37 insertions, 28 deletions
diff --git a/arm_compute/runtime/CL/functions/CLSoftmaxLayer.h b/arm_compute/runtime/CL/functions/CLSoftmaxLayer.h index fadbc430e6..231a56f712 100644 --- a/arm_compute/runtime/CL/functions/CLSoftmaxLayer.h +++ b/arm_compute/runtime/CL/functions/CLSoftmaxLayer.h @@ -50,6 +50,10 @@ class ICLTensor; * -# @ref CLLogits1DMaxKernel * -# @ref CLLogits1DShiftExpSumKernel * -# @ref CLLogits1DNormKernel + * And if the reduce_end_axis is not 0, the function will use one of the the following kernels to reshape the input and + * perform softmax on the reshaped input: + * -# @ref CLFlattenLayerKernel + * -# @ref CLReshapeLayerKernel */ template <bool IS_LOG = false> class CLSoftmaxLayerGeneric : public IFunction @@ -59,36 +63,39 @@ public: CLSoftmaxLayerGeneric(std::shared_ptr<IMemoryManager> memory_manager = nullptr); /** Set the input and output tensors. * - * @param[in] input Source tensor. Data types supported: QASYMM8/F16/F32 - * @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 - * @param[in] axis (Optional) Reduction axis. It has the purpose of squashing the first @p axis - * dimensions together. For instance, given a [4x4x4x4] image, - * when @p axis is 2, the Softmax reduction will be applied on each of the [4x4] planes of the input image. + * @param[in] input Source tensor. Data types supported: QASYMM8/F16/F32 + * @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 + * @param[in] reduce_end_axis (Optional) The last axis of the first n dimensions (inclusive)to reduce. Defaults to 0. + * It has the purpose of squashing together the first n dimensions till (including) the @p reduce_end_axis. For instance, given a [2x3x4x5] image, + * when @p reduce_end_axis is 1, the reduction will be applied to axes 0 and 1, and the Softmax op will be applied on each of the [2x3] planes of the input image. + * Must be in range [0, input_num_dimensions). */ - void configure(const ICLTensor *input, ICLTensor *output, float beta = 1.0f, size_t axis = 1); + void configure(const ICLTensor *input, ICLTensor *output, float beta = 1.0f, size_t reduce_end_axis = 0); /** Set the input and output tensors. * * @param[in] compile_context The compile context to be used. * @param[in] input Source tensor. Data types supported: QASYMM8/F16/F32 * @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 - * @param[in] axis (Optional) Reduction axis. It has the purpose of squashing the first @p axis - * dimensions together. For instance, given a [4x4x4x4] image, - * when @p axis is 2, the Softmax reduction will be applied on each of the [4x4] planes of the input image. + * @param[in] reduce_end_axis (Optional) The last axis of the first n dimensions (inclusive)to reduce. Defaults to 0. + * It has the purpose of squashing together the first n dimensions till (including) the @p reduce_end_axis. For instance, given a [2x3x4x5] image, + * when @p reduce_end_axis is 1, the reduction will be applied to axes 0 and 1, and the Softmax op will be applied on each of the [2x3] planes of the input image. + * Must be in range [0, input_num_dimensions). */ - void configure(const CLCompileContext &compile_context, const ICLTensor *input, ICLTensor *output, float beta = 1.0f, size_t axis = 1); + void configure(const CLCompileContext &compile_context, const ICLTensor *input, ICLTensor *output, float beta = 1.0f, size_t reduce_end_axis = 0); /** Static function to check if given info will lead to a valid configuration of @ref CLSoftmaxLayer * - * @param[in] input Source tensor. Data types supported: QASYMM8/F16/F32 - * @param[in] output Destination tensor. Data types supported: same as @p input - * @param[in] beta (Optional) A scaling factor for the exponent. Defaults to 1.f - * @param[in] axis (Optional) Reduction axis. It has the purpose of squashing the first @p axis - * dimensions together. For instance, given a [4x4x4x4] image, - * when @p axis is 2, the Softmax reduction will be applied on each of the [4x4] planes of the input image. + * @param[in] input Source tensor. Data types supported: QASYMM8/F16/F32 + * @param[in] output Destination tensor. Data types supported: same as @p input + * @param[in] beta (Optional) A scaling factor for the exponent. Defaults to 1.f + * @param[in] reduce_end_axis (Optional) The last axis of the first n dimensions (inclusive)to reduce. Defaults to 0. + * It has the purpose of squashing together the first n dimensions till (including) the @p reduce_end_axis. For instance, given a [2x3x4x5] image, + * when @p reduce_end_axis is 1, the reduction will be applied to axes 0 and 1, and the Softmax op will be applied on each of the [2x3] planes of the input image. + * Must be in range [0, input_num_dimensions). * @return a status */ - static Status validate(const ITensorInfo *input, const ITensorInfo *output, float beta = 1.0f, size_t axis = 1); + static Status validate(const ITensorInfo *input, const ITensorInfo *output, float beta = 1.0f, size_t reduce_end_axis = 0); // Inherited methods overridden: void run() override; @@ -101,13 +108,14 @@ private: * it initializes the kernel @p _flatten_kernel and the tensors @p _input_flat and * @p _output_flat * - * @param[in] input Original source tensor. - * @param[in] output Original destination tensor. - * @param[in] axis (Optional) Reduction axis. It has the purpose of squashing the first @p axis - * dimensions together. For instance, given a [4x4x4x4] image, - * when @p axis is 2, the Softmax reduction will be applied on each of the [4x4] planes of the input image. + * @param[in] input Original source tensor. + * @param[in] output Original destination tensor. + * @param[in] reduce_end_axis (Optional) The last axis of the first n dimensions (inclusive)to reduce. Defaults to 0. + * It has the purpose of squashing together the first n dimensions till (including) the @p reduce_end_axis. For instance, given a [2x3x4x5] image, + * when @p reduce_end_axis is 1, the reduction will be applied to axes 0 and 1, and the Softmax op will be applied on each of the [2x3] planes of the input image. + * Must be in range [0, input_num_dimensions). */ - void configure_reshape_input_kernel(const ICLTensor *input, const ICLTensor *output, size_t axis); + void configure_reshape_input_kernel(const ICLTensor *input, const ICLTensor *output, size_t reduce_end_axis); /** Utility method to configure the kernels needed to flatten the input * tensor. * @@ -118,11 +126,12 @@ private: * @param[in] compile_context The compile context to be used. * @param[in] input Original source tensor. * @param[in] output Original destination tensor. - * @param[in] axis (Optional) Reduction axis. It has the purpose of squashing the first @p axis - * dimensions together. For instance, given a [4x4x4x4] image, - * when @p axis is 2, the Softmax reduction will be applied on each of the [4x4] planes of the input image. + * @param[in] reduce_end_axis (Optional) The last axis of the first n dimensions (inclusive)to reduce. Defaults to 0. + * It has the purpose of squashing together the first n dimensions till (including) the @p reduce_end_axis. For instance, given a [2x3x4x5] image, + * when @p reduce_end_axis is 1, the reduction will be applied to axes 0 and 1, and the Softmax op will be applied on each of the [2x3] planes of the input image. + * Must be in range [0, input_num_dimensions). */ - void configure_reshape_input_kernel(const CLCompileContext &compile_context, const ICLTensor *input, const ICLTensor *output, size_t axis); + void configure_reshape_input_kernel(const CLCompileContext &compile_context, const ICLTensor *input, const ICLTensor *output, size_t reduce_end_axis); MemoryGroup _memory_group; CLLogits1DMaxShiftExpSumKernel _max_shift_exp_sum_kernel; |