COMPMID-1451: Perform CLOutputStage using floats.
Change-Id: Ic8312a5b6790aa7cd4468d42f08d557ad40e9441
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/154570
Tested-by: bsgcomp <bsgcomp@arm.com>
Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
diff --git a/arm_compute/core/CL/CLKernels.h b/arm_compute/core/CL/CLKernels.h
index 298cf52..1e456fa 100644
--- a/arm_compute/core/CL/CLKernels.h
+++ b/arm_compute/core/CL/CLKernels.h
@@ -71,6 +71,7 @@
#include "arm_compute/core/CL/kernels/CLGEMMLowpMatrixMultiplyKernel.h"
#include "arm_compute/core/CL/kernels/CLGEMMLowpOffsetContributionKernel.h"
#include "arm_compute/core/CL/kernels/CLGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel.h"
+#include "arm_compute/core/CL/kernels/CLGEMMLowpQuantizeDownInt32ToUint8ScaleByFloatKernel.h"
#include "arm_compute/core/CL/kernels/CLGEMMLowpQuantizeDownInt32ToUint8ScaleKernel.h"
#include "arm_compute/core/CL/kernels/CLGEMMLowpReductionKernel.h"
#include "arm_compute/core/CL/kernels/CLGEMMMatrixAccumulateBiasesKernel.h"
diff --git a/arm_compute/core/CL/kernels/CLGEMMLowpQuantizeDownInt32ToUint8ScaleByFloatKernel.h b/arm_compute/core/CL/kernels/CLGEMMLowpQuantizeDownInt32ToUint8ScaleByFloatKernel.h
new file mode 100644
index 0000000..5a5d393
--- /dev/null
+++ b/arm_compute/core/CL/kernels/CLGEMMLowpQuantizeDownInt32ToUint8ScaleByFloatKernel.h
@@ -0,0 +1,100 @@
+/*
+ * Copyright (c) 2018 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef __ARM_COMPUTE_CLGEMMLOWPQUANTIZEDOWNINT32TOUINT8SCALEBYFLOATKERNEL_H__
+#define __ARM_COMPUTE_CLGEMMLOWPQUANTIZEDOWNINT32TOUINT8SCALEBYFLOATKERNEL_H__
+
+#include "arm_compute/core/CL/ICLKernel.h"
+
+namespace arm_compute
+{
+// Forward declarations
+class ICLTensor;
+
+/** OpenCL kernel used to quantize down the int32 accumulator values of GEMMLowp to QASYMM8
+ *
+ * This kernel takes a final int32 accumulator value (the output of @ref CLGEMMLowpMatrixMultiplyKernel), and processes it to obtain the final QASYMM8 value.
+ * The following computations will be performed by the kernel:
+ *
+ * -# Compute fixed point multiplication between each entry of input by result_fixedpoint_multiplier
+ * -# Add bias to final result if bias tensor is not a nullptr
+ * -# Requantize
+ * -# Add offset to each result
+ * -# Clamp the value between the specified min and max bounds
+ * -# Clamp the resulting int32 values to the [0..255] range and cast to QASYMM8.
+ */
+class CLGEMMLowpQuantizeDownInt32ToUint8ScaleByFloatKernel : public ICLKernel
+{
+public:
+ /** Constructor */
+ CLGEMMLowpQuantizeDownInt32ToUint8ScaleByFloatKernel();
+ /** Prevent instances of this class from being copied (As this class contains pointers)*/
+ CLGEMMLowpQuantizeDownInt32ToUint8ScaleByFloatKernel(const CLGEMMLowpQuantizeDownInt32ToUint8ScaleByFloatKernel &) = delete;
+ /** Prevent instances of this class from being copied (As this class contains pointers)*/
+ CLGEMMLowpQuantizeDownInt32ToUint8ScaleByFloatKernel &operator=(const CLGEMMLowpQuantizeDownInt32ToUint8ScaleByFloatKernel &) = delete;
+ /** Allow instances of this class to be moved */
+ CLGEMMLowpQuantizeDownInt32ToUint8ScaleByFloatKernel(CLGEMMLowpQuantizeDownInt32ToUint8ScaleByFloatKernel &&) = default;
+ /** Allow instances of this class to be moved */
+ CLGEMMLowpQuantizeDownInt32ToUint8ScaleByFloatKernel &operator=(CLGEMMLowpQuantizeDownInt32ToUint8ScaleByFloatKernel &&) = default;
+ /** Initialise the kernel's input and output.
+ *
+ * @param[in] input Input tensor. Data type supported: S32
+ * @param[in] bias Biases tensor. Only shared biases supported and it can be a nullptr if the biases addition is not required.
+ * Biases are 1D tensor with dimensions [OFM]. Data type supported: Same as @p input.
+ * @param[out] output Output tensor. Data type supported: Data type supported: QASYMM8
+ * @param[in] multiplier Float multiplier to be multiplied to each element of the input matrix
+ * @param[in] offset Offset to be applied to result before converting it back to QASYMM8
+ * @param[in] min (Optional) Min value used to saturate down the output result before converting back to QASYMM8
+ * @param[in] max (Optional) Max value used to saturate up the output result before converting back to QASYMM8,
+ * Along with @p min, this value can be used to implement "rectified linear unit" activation functions
+ * @param[in] output_3d_depth (Optional) Depth of output in 3D (Defaults to 1)
+ */
+ void configure(const ICLTensor *input, const ICLTensor *bias, ICLTensor *output, float multiplier, int offset,
+ int min = 0, int max = 0, unsigned int output_3d_depth = 1);
+ /** Static function to check if given info will lead to a valid configuration of @ref CLGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel
+ *
+ * @param[in] input Input tensor. Data type supported: S32
+ * @param[in] bias Biases tensor. Only shared biases supported and it can be a nullptr if the biases addition is not required.
+ * Biases are 1D tensor with dimensions [OFM]. Data type supported: Same as @p input.
+ * @param[in] output Output tensor. Data type supported: Data type supported: QASYMM8
+ * @param[in] min (Optional) Min value used to saturate down the output result before converting back to QASYMM8
+ * @param[in] max (Optional) Max value used to saturate up the output result before converting back to QASYMM8,
+ * Along with @p min, this value can be used to implement "rectified linear unit" activation functions
+ * @param[in] output_3d_depth (Optional) Depth of output in 3D (Defaults to 1)
+ *
+ * @return a status
+ */
+ static Status validate(const ITensorInfo *input, const ITensorInfo *bias, const ITensorInfo *output,
+ int min = 0, int max = 0, unsigned int output_3d_depth = 1);
+
+ // Inherited methods overridden:
+ void run(const Window &window, cl::CommandQueue &queue) override;
+
+private:
+ const ICLTensor *_input;
+ const ICLTensor *_bias;
+ ICLTensor *_output;
+ bool _reinterpret_as_3d;
+};
+} // namespace arm_compute
+#endif /* __ARM_COMPUTE_CLGEMMLOWPQUANTIZEDOWNINT32TOUINT8SCALEBYFLOATKERNEL_H__ */
diff --git a/arm_compute/runtime/CL/functions/CLGEMMConvolutionLayer.h b/arm_compute/runtime/CL/functions/CLGEMMConvolutionLayer.h
index 958e70f..48b8801 100644
--- a/arm_compute/runtime/CL/functions/CLGEMMConvolutionLayer.h
+++ b/arm_compute/runtime/CL/functions/CLGEMMConvolutionLayer.h
@@ -178,15 +178,15 @@
static Status validate_mm(const ITensorInfo *input, const ITensorInfo *weights, const ITensorInfo *output, int gemm_3d_depth = 1, bool skip_im2col = false);
private:
- CLMemoryGroup _memory_group;
- CLConvolutionLayerReshapeWeights _reshape_weights;
- CLIm2ColKernel _im2col_kernel;
- CLGEMM _mm_gemm;
- CLGEMMLowpMatrixMultiplyCore _mm_gemmlowp;
- CLGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPoint _gemmlowp_output_stage;
- CLCol2ImKernel _col2im_kernel;
- CLActivationLayer _activationlayer_function;
- CLArithmeticAdditionKernel _add_bias_kernel;
+ CLMemoryGroup _memory_group;
+ CLConvolutionLayerReshapeWeights _reshape_weights;
+ CLIm2ColKernel _im2col_kernel;
+ CLGEMM _mm_gemm;
+ CLGEMMLowpMatrixMultiplyCore _mm_gemmlowp;
+ CLGEMMLowpQuantizeDownInt32ToUint8ScaleByFloat _gemmlowp_output_stage;
+ CLCol2ImKernel _col2im_kernel;
+ CLActivationLayer _activationlayer_function;
+ CLArithmeticAdditionKernel _add_bias_kernel;
const ICLTensor *_original_weights;
diff --git a/arm_compute/runtime/CL/functions/CLGEMMLowpOutputStage.h b/arm_compute/runtime/CL/functions/CLGEMMLowpOutputStage.h
index dca00f0..51fcbe9 100644
--- a/arm_compute/runtime/CL/functions/CLGEMMLowpOutputStage.h
+++ b/arm_compute/runtime/CL/functions/CLGEMMLowpOutputStage.h
@@ -150,5 +150,47 @@
*/
static Status validate(const ITensorInfo *input, const ITensorInfo *bias, const ITensorInfo *output, int min = 0, int max = 0, unsigned int output_3d_depth = 1);
};
+
+/** Basic function to execute CLGEMMLowpQuantizeDownInt32ToUint8ScaleByFloat on OpenCL.
+ *
+ * This function calls the following OpenCL kernels:
+ *
+ * -# @ref CLGEMMLowpQuantizeDownInt32ToUint8ScaleByFloatKernel
+ *
+ * @note The function accepts also 2 optional input arguments (min and max) which can be used to implement "rectified linear unit" activation functions
+ * after the result is shifted right by result_shift
+*/
+class CLGEMMLowpQuantizeDownInt32ToUint8ScaleByFloat : public ICLSimpleFunction
+{
+public:
+ /** Initialise the kernel's inputs, output
+ *
+ * @param[in] input Input tensor. Data type supported: S32
+ * @param[in] bias Biases tensor. Only shared biases supported and it can be a nullptr if the biases addition is not required.
+ * Biases are 1D tensor with dimensions [OFM]. Data type supported: Same as @p input.
+ * @param[out] output Output tensor. Data type supported: Data type supported: QASYMM8
+ * @param[in] multiplier Float multiplier to be multiplied to each element of the input matrix
+ * @param[in] offset Offset to be applied to result before converting it back to QASYMM8
+ * @param[in] min (Optional) Min value used to saturate down the output result before converting back to QASYMM8
+ * @param[in] max (Optional) Max value used to saturate up the output result before converting back to QASYMM8,
+ * Along with @p min, this value can be used to implement "rectified linear unit" activation functions
+ * @param[in] output_3d_depth (Optional) Depth of output in 3D (Defaults to 1)
+ */
+ void configure(const ICLTensor *input, const ICLTensor *bias, ICLTensor *output, float multiplier, int offset, int min = 0, int max = 0, unsigned int output_3d_depth = 1);
+ /** Static function to check if given info will lead to a valid configuration of @ref CLGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPoint
+ *
+ * @param[in] input Input tensor. It is the output of @ref CLGEMMLowpMatrixMultiplyCore function. Data type supported: S32
+ * @param[in] bias Biases tensor. Only shared biases supported and it can be a nullptr if the addition of biases is not required.
+ * Biases are 1D tensor with dimensions [OFM]. Data type supported: Same as @p input.
+ * @param[in] output Output tensor. Data type supported: Data type supported: QASYMM8
+ * @param[in] min (Optional) Min value used to saturate down the output result before converting back to QASYMM8
+ * @param[in] max (Optional) Max value used to saturate up the output result before converting back to QASYMM8,
+ * Along with @p min, this value can be used to implement "rectified linear unit" activation functions
+ * @param[in] output_3d_depth (Optional) Depth of output in 3D (Defaults to 1)
+ *
+ * @return a status
+ */
+ static Status validate(const ITensorInfo *input, const ITensorInfo *bias, const ITensorInfo *output, int min = 0, int max = 0, unsigned int output_3d_depth = 1);
+};
} // namespace arm_compute
#endif /*__ARM_COMPUTE_CLGEMMLOWPOUTPUTSTAGE_H__ */
\ No newline at end of file