Port NEGEMMLowp Part 1

Details:
Port NEGEMMLowpQuantizeDownInt32ScaleKernel to CpuGemmLowpQuantizeDownInt32ScaleKernel
Port NEGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel to CpuGemmLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel
Port NEGEMMLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel to CpuGemmLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel
Port NEGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel to CpuGemmLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel
Port NEGEMMLowpOutputStage functions to CpuGemmLowpOutputStage operators

Partially Resolves: COMPMID-4403

Change-Id: I6d5f45e43f35d731d564ed3b5c0e804d2a318fb1
Signed-off-by: Manuel Bottini <manuel.bottini@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/5833
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>

2 files changed, 237 insertions, 0 deletions

diff --git a/src/runtime/cpu/operators/CpuGemmLowpOutputStage.cpp b/src/runtime/cpu/operators/CpuGemmLowpOutputStage.cpp
new file mode 100644
index 0000000000..e17f854a21
--- /dev/null
+++ b/src/runtime/cpu/operators/CpuGemmLowpOutputStage.cpp
@@ -0,0 +1,147 @@
+/*
+ * Copyright (c) 2021 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.
+ */
+#include "src/runtime/cpu/operators/CpuGemmLowpOutputStage.h"
+
+#include "arm_compute/core/ITensor.h"
+#include "arm_compute/core/Validate.h"
+#include "arm_compute/runtime/NEON/NEScheduler.h"
+#include "src/core/cpu/kernels/CpuGemmLowpQuantizeDownInt32ScaleKernel.h"
+#include "src/core/cpu/kernels/CpuGemmLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel.h"
+#include "src/core/cpu/kernels/CpuGemmLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel.h"
+#include "src/core/cpu/kernels/CpuGemmLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel.h"
+
+namespace arm_compute
+{
+namespace cpu
+{
+void CpuGemmLowpOutputStage::configure(ITensorInfo *src, ITensorInfo *bias, ITensorInfo *dst, const GEMMLowpOutputStageInfo &info)
+{
+    // Perform validate step
+    ARM_COMPUTE_ERROR_THROW_ON(CpuGemmLowpOutputStage::validate(src, bias, dst, info));
+
+    switch(info.type)
+    {
+        case GEMMLowpOutputStageType::QUANTIZE_DOWN_FIXEDPOINT:
+        {
+            switch(info.output_data_type)
+            {
+                case DataType::QASYMM8:
+                {
+                    auto k = std::make_unique<kernels::CpuGemmLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel>();
+                    k->configure(src, bias, dst, info.gemmlowp_multiplier, info.gemmlowp_shift, info.gemmlowp_offset, info.gemmlowp_min_bound, info.gemmlowp_max_bound);
+                    _kernel = std::move(k);
+                    break;
+                }
+                case DataType::QASYMM8_SIGNED:
+                {
+                    auto k = std::make_unique<kernels::CpuGemmLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel>();
+                    k->configure(src, bias, dst, info.gemmlowp_multiplier, info.gemmlowp_shift, info.gemmlowp_offset, info.gemmlowp_min_bound, info.gemmlowp_max_bound);
+                    _kernel = std::move(k);
+                    break;
+                }
+                case DataType::QSYMM16:
+                {
+                    auto k = std::make_unique<kernels::CpuGemmLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel>();
+                    k->configure(src, bias, dst, info.gemmlowp_multiplier, info.gemmlowp_shift, info.gemmlowp_min_bound, info.gemmlowp_max_bound);
+                    _kernel = std::move(k);
+                    break;
+                }
+                default:
+                {
+                    ARM_COMPUTE_ERROR("Unsupported output data type.");
+                    break;
+                }
+            }
+            break;
+        }
+        case GEMMLowpOutputStageType::QUANTIZE_DOWN:
+        {
+            switch(info.output_data_type)
+            {
+                case DataType::QASYMM8:
+                case DataType::QASYMM8_SIGNED:
+                {
+                    auto k = std::make_unique<kernels::CpuGemmLowpQuantizeDownInt32ScaleKernel>();
+                    k->configure(src, bias, dst, &info);
+                    _kernel = std::move(k);
+                    break;
+                }
+                default:
+                {
+                    ARM_COMPUTE_ERROR("Unsupported output data type.");
+                    break;
+                }
+            }
+            break;
+        }
+        default:
+            ARM_COMPUTE_ERROR("Unsupported GEMMLowpOutputStage type.");
+    }
+}
+
+Status CpuGemmLowpOutputStage::validate(const ITensorInfo *src, const ITensorInfo *bias, const ITensorInfo *dst, const GEMMLowpOutputStageInfo &info)
+{
+    ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(src, dst);
+    ARM_COMPUTE_RETURN_ERROR_ON_MSG(dst->data_type() == DataType::UNKNOWN, "CpuGemmLowpOutputStage cannot be used with UNKNOWN output data type.");
+    ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(dst, 1, DataType::QASYMM8, DataType::QASYMM8_SIGNED, DataType::QSYMM16);
+    ARM_COMPUTE_RETURN_ERROR_ON((info.type != GEMMLowpOutputStageType::QUANTIZE_DOWN) && (info.type != GEMMLowpOutputStageType::QUANTIZE_DOWN_FIXEDPOINT));
+
+    switch(info.type)
+    {
+        case GEMMLowpOutputStageType::QUANTIZE_DOWN_FIXEDPOINT:
+        {
+            switch(dst->data_type())
+            {
+                case DataType::QASYMM8:
+                    return kernels::CpuGemmLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel::validate(src, bias, dst, info.gemmlowp_min_bound, info.gemmlowp_max_bound);
+                case DataType::QASYMM8_SIGNED:
+                    return kernels::CpuGemmLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel::validate(src, bias, dst, info.gemmlowp_min_bound, info.gemmlowp_max_bound);
+                case DataType::QSYMM16:
+                    return kernels::CpuGemmLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel::validate(src, bias, dst, info.gemmlowp_min_bound, info.gemmlowp_max_bound);
+                default:
+                    return ARM_COMPUTE_CREATE_ERROR(ErrorCode::RUNTIME_ERROR, "Unsupported output data type.");
+            }
+        }
+        case GEMMLowpOutputStageType::QUANTIZE_DOWN:
+        {
+            switch(dst->data_type())
+            {
+                case DataType::QASYMM8:
+                case DataType::QASYMM8_SIGNED:
+                    return kernels::CpuGemmLowpQuantizeDownInt32ScaleKernel::validate(src, bias, dst, &info);
+                default:
+                    return ARM_COMPUTE_CREATE_ERROR(ErrorCode::RUNTIME_ERROR, "Unsupported output data type.");
+            }
+        }
+        default:
+            return ARM_COMPUTE_CREATE_ERROR(ErrorCode::RUNTIME_ERROR, "Unsupported GEMMLowpOutputStage type.");
+    }
+}
+
+void CpuGemmLowpOutputStage::run(ITensorPack &tensors)
+{
+    NEScheduler::get().schedule_op(_kernel.get(), Window::DimY, _kernel->window(), tensors);
+}
+} // namespace cpu
+} // namespace arm_compute
\ No newline at end of file
diff --git a/src/runtime/cpu/operators/CpuGemmLowpOutputStage.h b/src/runtime/cpu/operators/CpuGemmLowpOutputStage.h
new file mode 100644
index 0000000000..848aaea3e7
--- /dev/null
+++ b/src/runtime/cpu/operators/CpuGemmLowpOutputStage.h
@@ -0,0 +1,90 @@
+/*
+ * Copyright (c) 2021 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_CPU_GEMMLOWP_OUTPUTSTAGE_H
+#define ARM_COMPUTE_CPU_GEMMLOWP_OUTPUTSTAGE_H
+
+#include "arm_compute/core/Types.h"
+#include "src/runtime/cpu/ICpuOperator.h"
+
+/** This file contains all available output stages for GEMMLowp.
+ *
+ *  In gemmlowp, the "output stage" is the process that takes a final int32 accumulator value (the output of @ref NEGEMMLowpMatrixMultiplyCore),
+ *  and processes it to obtain the final ASYMM8 value.
+ *
+ *  More information about the GEMMLowp output stage can be found at https://github.com/google/gemmlowp/blob/master/doc/output.md
+ */
+
+namespace arm_compute
+{
+namespace cpu
+{
+/** Basic function to execute GEMMLowpQuantizeDown kernels.
+ *
+ *  This function calls the following kernels:
+ *
+ * -# @ref kernels::CpuGemmLowpQuantizeDownInt32ScaleKernel
+ * -# @ref kernels::CpuGemmLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel
+ * -# @ref kernels::CpuGemmLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel
+ * -# @ref kernels::CpuGemmLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel
+*/
+class CpuGemmLowpOutputStage : public ICpuOperator
+{
+public:
+    /** Constructor */
+    CpuGemmLowpOutputStage() = default;
+    /** Default destructor */
+    ~CpuGemmLowpOutputStage() = default;
+    /** Initialise the kernel's inputs, output
+     *
+     * Valid data layouts:
+     * - All
+     *
+     * Valid data type configurations:
+     * |src0           |src1          |dst           |
+     * |:--------------|:-------------|:-------------|
+     * |S32            |S32           |QASYMM8       |
+     * |S32            |S32           |QASYMM8_SIGNED|
+     * |S32            |S32           |QSYMM16       |
+     *
+     * @param[in]  src  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 with dimensions [OFM]. Data type supported: Same as @p input.
+     * @param[out] dst  Output tensor info. Data type supported: Data type supported: QASYMM8/QASYMM8_SIGNED/QSYMM16
+     * @param[in]  info GEMMLowp output stage metadata.
+     */
+    void configure(ITensorInfo *src, ITensorInfo *bias, ITensorInfo *dst, const GEMMLowpOutputStageInfo &info);
+    /** Static function to check if given info will lead to a valid configuration
+     *
+     * Similar to CpuGemmLowpOutputStage::configure()
+     *
+     * @return a status
+     */
+    static Status validate(const ITensorInfo *src, const ITensorInfo *bias, const ITensorInfo *dst, const GEMMLowpOutputStageInfo &info);
+
+    // Inherited methods overridden:
+    void run(ITensorPack &tensors) override;
+};
+} // namespace cpu
+} // namespace arm_compute
+#endif /* ARM_COMPUTE_CPU_GEMMLOWP_OUTPUTSTAGE_H */