COMPMID-2967 Add support for QASYMM8_SIGNED in CLGEMMLowpQuantizeDownInt32ToUint8ScaleKernel
Signed-off-by: Luca Foschiani <luca.foschiani@arm.com>
Change-Id: I4f7918630ea95fc28597b3d7b189f3d8fd35aef8
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/2890
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@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 d070d6a..f2e16ca 100644
--- a/arm_compute/core/CL/CLKernels.h
+++ b/arm_compute/core/CL/CLKernels.h
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2016-2019 ARM Limited.
+ * Copyright (c) 2016-2020 ARM Limited.
*
* SPDX-License-Identifier: MIT
*
@@ -79,11 +79,11 @@
#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/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/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"
#include "arm_compute/core/CL/kernels/CLGEMMMatrixMultiplyKernel.h"
diff --git a/arm_compute/core/CL/kernels/CLGEMMLowpQuantizeDownInt32ScaleKernel.h b/arm_compute/core/CL/kernels/CLGEMMLowpQuantizeDownInt32ScaleKernel.h
new file mode 100644
index 0000000..f9599b5
--- /dev/null
+++ b/arm_compute/core/CL/kernels/CLGEMMLowpQuantizeDownInt32ScaleKernel.h
@@ -0,0 +1,93 @@
+/*
+ * 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_CLGEMMLOWPQUANTIZEDOWNINT32SCALEKERNEL_H
+#define ARM_COMPUTE_CLGEMMLOWPQUANTIZEDOWNINT32SCALEKERNEL_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
+ *
+ * This kernel takes a final int32 accumulator value (the output of @ref CLGEMMLowpMatrixMultiplyKernel), and processes it to obtain the final QASYMM8/QASYMM8_SIGNED value.
+ * The following computations will be performed by the kernel:
+ *
+ * -# Add offset terms to final result
+ * -# Multiply each entry of result by result_mult_int
+ * -# Add bias to final result if bias tensor is not a nullptr
+ * -# Shift the int32 accumulator by result_shift
+ * -# Clamp the value between the specified min and max bounds
+ * -# Clamp the resulting int32 values:
+ * -# -to the [0..255] range and cast to QASYMM8.
+ * -# -to the [-128..127] range and cast to QASYMM8/SIGNED.
+ *
+ */
+class CLGEMMLowpQuantizeDownInt32ScaleKernel : public ICLKernel
+{
+public:
+ /** Constructor */
+ CLGEMMLowpQuantizeDownInt32ScaleKernel();
+ /** Prevent instances of this class from being copied (As this class contains pointers)*/
+ CLGEMMLowpQuantizeDownInt32ScaleKernel(const CLGEMMLowpQuantizeDownInt32ScaleKernel &) = delete;
+ /** Prevent instances of this class from being copied (As this class contains pointers)*/
+ CLGEMMLowpQuantizeDownInt32ScaleKernel &operator=(const CLGEMMLowpQuantizeDownInt32ScaleKernel &) = delete;
+ /** Allow instances of this class to be moved */
+ CLGEMMLowpQuantizeDownInt32ScaleKernel(CLGEMMLowpQuantizeDownInt32ScaleKernel &&) = default;
+ /** Allow instances of this class to be moved */
+ CLGEMMLowpQuantizeDownInt32ScaleKernel &operator=(CLGEMMLowpQuantizeDownInt32ScaleKernel &&) = 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/QASYMM8_SIGNED
+ * @param[in] output_stage Output stage info. Used to pass the quantized output data type
+ */
+ void configure(const ICLTensor *input, const ICLTensor *bias, ICLTensor *output, const GEMMLowpOutputStageInfo *output_stage);
+ /** Static function to check if given info will lead to a valid configuration of @ref CLGEMMLowpQuantizeDownInt32ScaleKernel
+ *
+ * @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/QASYMM8_SIGNED
+ * @param[in] output_stage 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 *output_stage);
+
+ // Inherited methods overridden:
+ void run(const Window &window, cl::CommandQueue &queue) override;
+
+private:
+ const ICLTensor *_input;
+ const ICLTensor *_bias;
+ ICLTensor *_output;
+ const GEMMLowpOutputStageInfo *_output_stage;
+};
+} // namespace arm_compute
+
+#endif /* ARM_COMPUTE_CLGEMMLOWPQUANTIZEDOWNINT32SCALEKERNEL_H */
\ No newline at end of file
diff --git a/arm_compute/core/CL/kernels/CLGEMMLowpQuantizeDownInt32ToUint8ScaleKernel.h b/arm_compute/core/CL/kernels/CLGEMMLowpQuantizeDownInt32ToUint8ScaleKernel.h
deleted file mode 100644
index 9cd726a..0000000
--- a/arm_compute/core/CL/kernels/CLGEMMLowpQuantizeDownInt32ToUint8ScaleKernel.h
+++ /dev/null
@@ -1,97 +0,0 @@
-/*
- * Copyright (c) 2017-2019 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_CLGEMMLOWPQUANTIZEDOWNINT32TOUINT8SCALEKERNEL_H
-#define ARM_COMPUTE_CLGEMMLOWPQUANTIZEDOWNINT32TOUINT8SCALEKERNEL_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 @ref CLGEMMLowpMatrixMultiplyKernel), and processes it to obtain the final QASYMM8 value.
- * The following computations will be performed by the kernel:
- *
- * -# Add offset terms to final result
- * -# Multiply each entry of result by result_mult_int
- * -# Add bias to final result if bias tensor is not a nullptr
- * -# Shift the int32 accumulator by result_shift
- * -# 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 CLGEMMLowpQuantizeDownInt32ToUint8ScaleKernel : public ICLKernel
-{
-public:
- /** Constructor */
- CLGEMMLowpQuantizeDownInt32ToUint8ScaleKernel();
- /** Prevent instances of this class from being copied (As this class contains pointers)*/
- CLGEMMLowpQuantizeDownInt32ToUint8ScaleKernel(const CLGEMMLowpQuantizeDownInt32ToUint8ScaleKernel &) = delete;
- /** Prevent instances of this class from being copied (As this class contains pointers)*/
- CLGEMMLowpQuantizeDownInt32ToUint8ScaleKernel &operator=(const CLGEMMLowpQuantizeDownInt32ToUint8ScaleKernel &) = delete;
- /** Allow instances of this class to be moved */
- CLGEMMLowpQuantizeDownInt32ToUint8ScaleKernel(CLGEMMLowpQuantizeDownInt32ToUint8ScaleKernel &&) = default;
- /** Allow instances of this class to be moved */
- CLGEMMLowpQuantizeDownInt32ToUint8ScaleKernel &operator=(CLGEMMLowpQuantizeDownInt32ToUint8ScaleKernel &&) = 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_offset Offset to be added to each element of the input matrix
- * @param[in] result_mult_int Value to be multiplied to each element of the input matrix when once the result_offset has been add
- * @param[in] result_shift Number of bits to shift right the result before converting 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_offset, int result_mult_int, int result_shift, int min = 0, int max = 0);
- /** Static function to check if given info will lead to a valid configuration of @ref CLGEMMLowpQuantizeDownInt32ToUint8ScaleKernel
- *
- * @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_CLGEMMLOWPQUANTIZEDOWNINT32TOUINT8SCALEKERNEL_H */
\ No newline at end of file
diff --git a/arm_compute/runtime/CL/functions/CLGEMMLowpOutputStage.h b/arm_compute/runtime/CL/functions/CLGEMMLowpOutputStage.h
index b6619da..184d827 100644
--- a/arm_compute/runtime/CL/functions/CLGEMMLowpOutputStage.h
+++ b/arm_compute/runtime/CL/functions/CLGEMMLowpOutputStage.h
@@ -51,7 +51,7 @@
*
* This function calls the following OpenCL kernels:
*
- * -# @ref CLGEMMLowpQuantizeDownInt32ToUint8ScaleKernel
+ * -# @ref CLGEMMLowpQuantizeDownInt32ScaleKernel
*
* @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
@@ -72,6 +72,7 @@
* @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. Defaults to the maximum possible 32-bit signed integer.
*/
+ ARM_COMPUTE_DEPRECATED_REL(20.05)
void configure(const ICLTensor *input, const ICLTensor *bias, ICLTensor *output, int result_offset, int result_mult_int, int result_shift, int min = std::numeric_limits<int32_t>::lowest(),
int max = std::numeric_limits<int32_t>::max());
/** Static function to check if given info will lead to a valid configuration of @ref CLGEMMLowpQuantizeDownInt32ToUint8Scale
@@ -86,6 +87,7 @@
*
* @return a status
*/
+ ARM_COMPUTE_DEPRECATED_REL(20.05)
static Status validate(const ITensorInfo *input, const ITensorInfo *bias, const ITensorInfo *output, int min = std::numeric_limits<int32_t>::lowest(), int max = std::numeric_limits<int32_t>::max());
};
@@ -314,6 +316,7 @@
*
* This function calls the following CL kernels:
*
+ * -# @ref CLGEMMLowpQuantizeDownInt32ScaleKernel
* -# @ref CLGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel
* -# @ref CLGEMMLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel
*/