COMPMID-3829: Create CLGEMMLowpQuantizeDownInt32ScaleByFixedPointKernel and remove padding from related OpenCL kernels
Change-Id: I0b0be8fcccf511c7214e83ba6aa8d0e901bc4f3c
Signed-off-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/4146
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
diff --git a/arm_compute/core/CL/CLKernels.h b/arm_compute/core/CL/CLKernels.h
index dbda0db..231534f 100644
--- a/arm_compute/core/CL/CLKernels.h
+++ b/arm_compute/core/CL/CLKernels.h
@@ -78,11 +78,9 @@
#include "arm_compute/core/CL/kernels/CLGEMMLowpMatrixMultiplyReshapedOnlyRHSKernel.h"
#include "arm_compute/core/CL/kernels/CLGEMMLowpOffsetContributionKernel.h"
#include "arm_compute/core/CL/kernels/CLGEMMLowpOffsetContributionOutputStageKernel.h"
+#include "arm_compute/core/CL/kernels/CLGEMMLowpQuantizeDownInt32ScaleByFixedPointKernel.h"
#include "arm_compute/core/CL/kernels/CLGEMMLowpQuantizeDownInt32ScaleByFloatKernel.h"
#include "arm_compute/core/CL/kernels/CLGEMMLowpQuantizeDownInt32ScaleKernel.h"
-#include "arm_compute/core/CL/kernels/CLGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel.h"
-#include "arm_compute/core/CL/kernels/CLGEMMLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel.h"
-#include "arm_compute/core/CL/kernels/CLGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel.h"
#include "arm_compute/core/CL/kernels/CLGEMMLowpReductionKernel.h"
#include "arm_compute/core/CL/kernels/CLGEMMMatrixMultiplyKernel.h"
#include "arm_compute/core/CL/kernels/CLGEMMMatrixMultiplyNativeKernel.h"
diff --git a/arm_compute/core/CL/kernels/CLGEMMLowpQuantizeDownInt32ScaleByFixedPointKernel.h b/arm_compute/core/CL/kernels/CLGEMMLowpQuantizeDownInt32ScaleByFixedPointKernel.h
new file mode 100644
index 0000000..23040e7
--- /dev/null
+++ b/arm_compute/core/CL/kernels/CLGEMMLowpQuantizeDownInt32ScaleByFixedPointKernel.h
@@ -0,0 +1,89 @@
+/*
+ * Copyright (c) 2020 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_CLGEMMLOWPQUANTIZEDOWNINT32SCALEBYFIXEDPOINTKERNEL_H
+#define ARM_COMPUTE_CLGEMMLOWPQUANTIZEDOWNINT32SCALEBYFIXEDPOINTKERNEL_H
+
+#include "arm_compute/core/CL/ICLKernel.h"
+
+namespace arm_compute
+{
+class ICLTensor;
+
+/** OpenCL kernel used to quantize down the int32 accumulator values of GEMMLowp to QASYMM8/QASYMM8_SIGNED/QSYMM16
+ *
+ * This kernel takes a final int32 accumulator value (the output of the matrix multiplication), and processes it to obtain the final quantized value.
+ * The following computations will be performed by the kernel:
+ *
+ * -# Compute fixed point multiplication between each entry of input by gemmlowp_multiplier
+ * -# Add bias to final result if bias tensor is not a nullptr
+ * -# Round to nearest division by a power-of-two using result_shift
+ * -# Add offset to each result
+ * -# Clamp the value between the specified min and max bounds
+ * -# Clamp the resulting int32 values to the proper quantized range and cast to QASYMM8/QASYMM8_SIGNED/QSYMM16.
+ */
+class CLGEMMLowpQuantizeDownInt32ScaleByFixedPointKernel : public ICLKernel
+{
+public:
+ /** Constructor */
+ CLGEMMLowpQuantizeDownInt32ScaleByFixedPointKernel();
+ /** Prevent instances of this class from being copied (As this class contains pointers)*/
+ CLGEMMLowpQuantizeDownInt32ScaleByFixedPointKernel(const CLGEMMLowpQuantizeDownInt32ScaleByFixedPointKernel &) = delete;
+ /** Prevent instances of this class from being copied (As this class contains pointers)*/
+ CLGEMMLowpQuantizeDownInt32ScaleByFixedPointKernel &operator=(const CLGEMMLowpQuantizeDownInt32ScaleByFixedPointKernel &) = delete;
+ /** Allow instances of this class to be moved */
+ CLGEMMLowpQuantizeDownInt32ScaleByFixedPointKernel(CLGEMMLowpQuantizeDownInt32ScaleByFixedPointKernel &&) = default;
+ /** Allow instances of this class to be moved */
+ CLGEMMLowpQuantizeDownInt32ScaleByFixedPointKernel &operator=(CLGEMMLowpQuantizeDownInt32ScaleByFixedPointKernel &&) = default;
+ /** Initialise the kernel's input and output.
+ *
+ * @param[in] compile_context The compile context to be used.
+ * @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/QASYMM8_SIGNED/QSYMM16.
+ * @param[in] info Output stage info. Used to pass the quantized output data type
+ */
+ void configure(const CLCompileContext &compile_context, const ICLTensor *input, const ICLTensor *bias, ICLTensor *output, const GEMMLowpOutputStageInfo *info);
+ /** Static function to check if given info will lead to a valid configuration of @ref CLGEMMLowpQuantizeDownInt32ScaleByFixedPointKernel
+ *
+ * @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: QSYMM8/QASYMM8_SIGNED/QSYMM16.
+ * @param[in] info Output stage info. Used to pass the quantized output data type
+ *
+ * @return a status
+ */
+ static Status validate(const ITensorInfo *input, const ITensorInfo *bias, const ITensorInfo *output, const GEMMLowpOutputStageInfo *info);
+
+ // Inherited methods overridden:
+ void run(const Window &window, cl::CommandQueue &queue) override;
+
+private:
+ const ICLTensor *_input;
+ const ICLTensor *_bias;
+ ICLTensor *_output;
+};
+} // namespace arm_compute
+#endif /* ARM_COMPUTE_CLGEMMLOWPQUANTIZEDOWNINT32SCALEBYFIXEDPOINTKERNEL_H */
diff --git a/arm_compute/core/CL/kernels/CLGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel.h b/arm_compute/core/CL/kernels/CLGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel.h
deleted file mode 100644
index 16990c5..0000000
--- a/arm_compute/core/CL/kernels/CLGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel.h
+++ /dev/null
@@ -1,108 +0,0 @@
-/*
- * Copyright (c) 2019-2020 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_CLGEMMLOWPQUANTIZEDOWNINT32TOINT16SCALEBYFIXEDPOINTKERNEL_H
-#define ARM_COMPUTE_CLGEMMLOWPQUANTIZEDOWNINT32TOINT16SCALEBYFIXEDPOINTKERNEL_H
-
-#include "arm_compute/core/CL/ICLKernel.h"
-
-namespace arm_compute
-{
-class ICLTensor;
-
-/** CL kernel used to quantize down the int32 accumulator values of GEMMLowp to QSYMM16
- *
- * This kernel takes a final int32 accumulator value (the output of the matrix multiplication), and processes it to obtain the final QSYMM16 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
- * -# Round to nearest division by a power-of-two using result_shift
- * -# Clamp the value between the specified min and max bounds
- * -# Clamp the resulting int32 values to the [-32768, 32767] range and cast to QSYMM16.
- *
- */
-class CLGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel : public ICLKernel
-{
-public:
- /** Constructor */
- CLGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel();
- /** Prevent instances of this class from being copied (As this class contains pointers)*/
- CLGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel(const CLGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel &) = delete;
- /** Prevent instances of this class from being copied (As this class contains pointers)*/
- CLGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel &operator=(const CLGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel &) = delete;
- /** Allow instances of this class to be moved */
- CLGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel(CLGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel &&) = default;
- /** Allow instances of this class to be moved */
- CLGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel &operator=(CLGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel &&) = 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: QSYMM16
- * @param[in] result_fixedpoint_multiplier Fixed point value to be multiplied to each element of the input matrix when once the result_offset has been add
- * @param[in] result_shift Integer value used to round to nearest division by a power-of-two the result after the fixed point multiplication
- * @param[in] min (Optional) Min value used to saturate down the output result before converting back to QSYMM16. Defaults to 0.
- * @param[in] max (Optional) Max value used to saturate up the output result before converting back to QSYMM16.
- * Along with @p min, this value can be used to implement "rectified linear unit" activation functions. Defaults to 0.
- */
- void configure(const ICLTensor *input, const ICLTensor *bias, ICLTensor *output, int result_fixedpoint_multiplier, int result_shift, int min = 0, int max = 0);
- /** Initialise the kernel's input and output.
- *
- * @param[in] compile_context The compile context to be used.
- * @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: QSYMM16
- * @param[in] result_fixedpoint_multiplier Fixed point value to be multiplied to each element of the input matrix when once the result_offset has been add
- * @param[in] result_shift Integer value used to round to nearest division by a power-of-two the result after the fixed point multiplication
- * @param[in] min (Optional) Min value used to saturate down the output result before converting back to QSYMM16. Defaults to 0.
- * @param[in] max (Optional) Max value used to saturate up the output result before converting back to QSYMM16.
- * Along with @p min, this value can be used to implement "rectified linear unit" activation functions. Defaults to 0.
- */
- void configure(const CLCompileContext &compile_context, const ICLTensor *input, const ICLTensor *bias, ICLTensor *output, int result_fixedpoint_multiplier, int result_shift, int min = 0, int max = 0);
- /** Static function to check if given info will lead to a valid configuration of @ref CLGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel
- *
- * @param[in] input Input tensor info. Data type supported: S32
- * @param[in] bias Biases tensor info. Only shared biases supported and it can be a nullptr if the biases addition is not required.
- * Biases are 1D tensor info with dimensions [OFM]. Data type supported: Same as @p input.
- * @param[in] output Output tensor info. Data type supported: Data type supported: QSYMM16
- * @param[in] min (Optional) Min value used to saturate down the output result before converting back to QSYMM16. Defaults to 0.
- * @param[in] max (Optional) Max value used to saturate up the output result before converting back to QSYMM16,
- * Along with @p min, this value can be used to implement "rectified linear unit" activation functions. Defaults to 0.
- *
- * @return a status
- */
- static Status validate(const ITensorInfo *input, const ITensorInfo *bias, const ITensorInfo *output, int min = 0, int max = 0);
-
- // Inherited methods overridden:
- void run(const Window &window, cl::CommandQueue &queue) override;
-
-private:
- const ICLTensor *_input;
- const ICLTensor *_bias;
- ICLTensor *_output;
-};
-} // namespace arm_compute
-#endif /* ARM_COMPUTE_CLGEMMLOWPQUANTIZEDOWNINT32TOINT16SCALEBYFIXEDPOINTKERNEL_H */
diff --git a/arm_compute/core/CL/kernels/CLGEMMLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel.h b/arm_compute/core/CL/kernels/CLGEMMLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel.h
deleted file mode 100644
index ef962d8..0000000
--- a/arm_compute/core/CL/kernels/CLGEMMLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel.h
+++ /dev/null
@@ -1,112 +0,0 @@
-/*
- * Copyright (c) 2019-2020 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_CLGEMMLOWPQUANTIZEDOWNINT32TOINT8SCALEBYFIXEDPOINTKERNEL_H
-#define ARM_COMPUTE_CLGEMMLOWPQUANTIZEDOWNINT32TOINT8SCALEBYFIXEDPOINTKERNEL_H
-
-#include "arm_compute/core/CL/ICLKernel.h"
-
-namespace arm_compute
-{
-class ICLTensor;
-
-/** OpenCL kernel used to quantize down the int32 accumulator values of GEMMLowp to QASYMM8_SIGNED
- *
- * This kernel takes a final int32 accumulator value (the output of the matrix multiplication), and processes it to obtain the final QASYMM8_SIGNED 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
- * -# Round to nearest division by a power-of-two using result_shift
- * -# Add offset to each result
- * -# Clamp the value between the specified min and max bounds
- * -# Clamp the resulting int32 values to the [-128..127] range and cast to QASYMM8_SIGNED.
- */
-class CLGEMMLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel : public ICLKernel
-{
-public:
- /** Constructor */
- CLGEMMLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel();
- /** Prevent instances of this class from being copied (As this class contains pointers)*/
- CLGEMMLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel(const CLGEMMLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel &) = delete;
- /** Prevent instances of this class from being copied (As this class contains pointers)*/
- CLGEMMLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel &operator=(const CLGEMMLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel &) = delete;
- /** Allow instances of this class to be moved */
- CLGEMMLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel(CLGEMMLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel &&) = default;
- /** Allow instances of this class to be moved */
- CLGEMMLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel &operator=(CLGEMMLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel &&) = 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_SIGNED
- * @param[in] result_fixedpoint_multiplier Fixed point value to be multiplied to each element of the input matrix when once the result_offset has been add
- * @param[in] result_shift Integer value used to round to nearest division by a power-of-two the result after the fixed point multiplication
- * @param[in] result_offset_after_shift Offset to be applied to result before converting it back to QASYMM8_SIGNED
- * @param[in] min (Optional) Min value used to saturate down the output result before converting back to QASYMM8_SIGNED. Defaults to 0
- * @param[in] max (Optional) Max value used to saturate up the output result before converting back to QASYMM8_SIGNED. Defaults to 0
- * Along with @p min, this value can be used to implement "rectified linear unit" activation functions
- */
- void configure(const ICLTensor *input, const ICLTensor *bias, ICLTensor *output, int result_fixedpoint_multiplier, int result_shift, int result_offset_after_shift,
- int min = 0, int max = 0);
- /** Initialise the kernel's input and output.
- *
- * @param[in] compile_context The compile context to be used.
- * @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_SIGNED
- * @param[in] result_fixedpoint_multiplier Fixed point value to be multiplied to each element of the input matrix when once the result_offset has been add
- * @param[in] result_shift Integer value used to round to nearest division by a power-of-two the result after the fixed point multiplication
- * @param[in] result_offset_after_shift Offset to be applied to result before converting it back to QASYMM8_SIGNED
- * @param[in] min (Optional) Min value used to saturate down the output result before converting back to QASYMM8_SIGNED. Defaults to 0
- * @param[in] max (Optional) Max value used to saturate up the output result before converting back to QASYMM8_SIGNED. Defaults to 0
- * Along with @p min, this value can be used to implement "rectified linear unit" activation functions
- */
- void configure(const CLCompileContext &compile_context, const ICLTensor *input, const ICLTensor *bias, ICLTensor *output, int result_fixedpoint_multiplier, int result_shift, int result_offset_after_shift,
- int min = 0, int max = 0);
- /** 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_SIGNED
- * @param[in] min (Optional) Min value used to saturate down the output result before converting back to QASYMM8_SIGNED
- * @param[in] max (Optional) Max value used to saturate up the output result before converting back to QASYMM8_SIGNED,
- * Along with @p min, this value can be used to implement "rectified linear unit" activation functions
- *
- * @return a status
- */
- static Status validate(const ITensorInfo *input, const ITensorInfo *bias, const ITensorInfo *output, int min = 0, int max = 0);
-
- // Inherited methods overridden:
- void run(const Window &window, cl::CommandQueue &queue) override;
-
-private:
- const ICLTensor *_input;
- const ICLTensor *_bias;
- ICLTensor *_output;
-};
-} // namespace arm_compute
-#endif /* ARM_COMPUTE_CLGEMMLOWPQUANTIZEDOWNINT32TOINT8SCALEBYFIXEDPOINTKERNEL_H */
diff --git a/arm_compute/core/CL/kernels/CLGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel.h b/arm_compute/core/CL/kernels/CLGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel.h
deleted file mode 100644
index ca13b2f..0000000
--- a/arm_compute/core/CL/kernels/CLGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel.h
+++ /dev/null
@@ -1,112 +0,0 @@
-/*
- * Copyright (c) 2017-2020 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_CLGEMMLOWPQUANTIZEDOWNINT32TOUINT8SCALEBYFIXEDPOINTKERNEL_H
-#define ARM_COMPUTE_CLGEMMLOWPQUANTIZEDOWNINT32TOUINT8SCALEBYFIXEDPOINTKERNEL_H
-
-#include "arm_compute/core/CL/ICLKernel.h"
-
-namespace arm_compute
-{
-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 the matrix multiplication), 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
- * -# Round to nearest division by a power-of-two using result_shift
- * -# 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 CLGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel : public ICLKernel
-{
-public:
- /** Constructor */
- CLGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel();
- /** Prevent instances of this class from being copied (As this class contains pointers)*/
- CLGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel(const CLGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel &) = delete;
- /** Prevent instances of this class from being copied (As this class contains pointers)*/
- CLGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel &operator=(const CLGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel &) = delete;
- /** Allow instances of this class to be moved */
- CLGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel(CLGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel &&) = default;
- /** Allow instances of this class to be moved */
- CLGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel &operator=(CLGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel &&) = 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] result_fixedpoint_multiplier Fixed point value to be multiplied to each element of the input matrix when once the result_offset has been add
- * @param[in] result_shift Integer value used to round to nearest division by a power-of-two the result after the fixed point multiplication
- * @param[in] result_offset_after_shift 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
- */
- void configure(const ICLTensor *input, const ICLTensor *bias, ICLTensor *output, int result_fixedpoint_multiplier, int result_shift, int result_offset_after_shift,
- int min = 0, int max = 0);
- /** Initialise the kernel's input and output.
- *
- * @param[in] compile_context The compile context to be used.
- * @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] result_fixedpoint_multiplier Fixed point value to be multiplied to each element of the input matrix when once the result_offset has been add
- * @param[in] result_shift Integer value used to round to nearest division by a power-of-two the result after the fixed point multiplication
- * @param[in] result_offset_after_shift 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
- */
- void configure(const CLCompileContext &compile_context, const ICLTensor *input, const ICLTensor *bias, ICLTensor *output, int result_fixedpoint_multiplier, int result_shift, int result_offset_after_shift,
- int min = 0, int max = 0);
- /** 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
- *
- * @return a status
- */
- static Status validate(const ITensorInfo *input, const ITensorInfo *bias, const ITensorInfo *output, int min = 0, int max = 0);
-
- // Inherited methods overridden:
- void run(const Window &window, cl::CommandQueue &queue) override;
-
-private:
- const ICLTensor *_input;
- const ICLTensor *_bias;
- ICLTensor *_output;
-};
-} // namespace arm_compute
-#endif /* ARM_COMPUTE_CLGEMMLOWPQUANTIZEDOWNINT32TOUINT8SCALEBYFIXEDPOINTKERNEL_H */
diff --git a/arm_compute/runtime/CL/functions/CLGEMMConvolutionLayer.h b/arm_compute/runtime/CL/functions/CLGEMMConvolutionLayer.h
index 277b27f..467045c 100644
--- a/arm_compute/runtime/CL/functions/CLGEMMConvolutionLayer.h
+++ b/arm_compute/runtime/CL/functions/CLGEMMConvolutionLayer.h
@@ -158,8 +158,8 @@
*
* -# @ref CLIm2ColKernel
* -# @ref CLGEMM (if the data type is FP32 or FP16)
- * -# @ref CLGEMMLowpMatrixMultiplyCore (if the data type is QASYMM8)
- * -# @ref CLGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPoint (if the data type is QASYMM8)
+ * -# @ref CLGEMMLowpMatrixMultiplyCore (if the data type is QASYMM8/QASYMM8_SIGNED)
+ * -# @ref CLGEMMLowpOutputStage with QUANTIZE_DOWN_FIXEDPOINT type of quantization (if the data type is QASYMM8/QASYMM8_SIGNED)
* -# @ref CLCol2ImKernel (if NCHW data layout)
*/
class CLGEMMConvolutionLayer : public IFunction
diff --git a/arm_compute/runtime/CL/functions/CLGEMMLowpOutputStage.h b/arm_compute/runtime/CL/functions/CLGEMMLowpOutputStage.h
index c6e9588..44c52ff 100644
--- a/arm_compute/runtime/CL/functions/CLGEMMLowpOutputStage.h
+++ b/arm_compute/runtime/CL/functions/CLGEMMLowpOutputStage.h
@@ -61,7 +61,7 @@
*
* This function calls the following OpenCL kernels:
*
- * -# @ref CLGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel
+ * -# @ref CLGEMMLowpQuantizeDownInt32ScaleByFixedPointKernel
*
* @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
@@ -139,7 +139,7 @@
*
* This function calls the following OpenCL kernels:
*
- * -# @ref CLGEMMLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel
+ * -# @ref CLGEMMLowpQuantizeDownInt32ScaleByFixedPointKernel
*
* @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
@@ -217,7 +217,7 @@
*
* This function calls the following NEON kernels:
*
- * -# @ref CLGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel
+ * -# @ref CLGEMMLowpQuantizeDownInt32ScaleByFixedPointKernel
*
* @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
@@ -274,9 +274,7 @@
*
* -# @ref CLGEMMLowpQuantizeDownInt32ScaleKernel
* -# @ref CLGEMMLowpQuantizeDownInt32ScaleByFloatKernel
- * -# @ref CLGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel
- * -# @ref CLGEMMLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel
- * -# @ref CLGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel
+ * -# @ref CLGEMMLowpQuantizeDownInt32ScaleByFixedPointKernel
*/
class CLGEMMLowpOutputStage : public ICLSimpleFunction
{
@@ -300,7 +298,7 @@
* @param[in] info GEMMLowp output stage metadata.
*/
void configure(const CLCompileContext &compile_context, const ICLTensor *input, const ICLTensor *bias, ICLTensor *output, const GEMMLowpOutputStageInfo &info);
- /** Static function to check if given info will lead to a valid configuration of @ref CLGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel
+ /** Static function to check if given info will lead to a valid configuration of @ref CLGEMMLowpQuantizeDownInt32ScaleByFixedPointKernel
*
* @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.