COMPMID-2096: Refactor the CLGEMMLowp function selection (heuristic)

Change-Id: I15a8b39e0354d3b6686ed4cc8c361782c0512037
Signed-off-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
Reviewed-on: https://review.mlplatform.org/c/1410
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: VidhyaSudhan Loganathan <vidhyasudhan.loganathan@arm.com>

3 files changed, 139 insertions, 15 deletions

diff --git a/arm_compute/core/CL/gemm/native/CLGEMMNativeKernelConfiguration.h b/arm_compute/core/CL/gemm/native/CLGEMMNativeKernelConfiguration.h
new file mode 100644
index 0000000000..7d0e7c97d4
--- /dev/null
+++ b/arm_compute/core/CL/gemm/native/CLGEMMNativeKernelConfiguration.h
@@ -0,0 +1,59 @@
+/*
+ * Copyright (c) 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_CLGEMMNATIVEKERNELCONFIGURATION_H__
+#define __ARM_COMPUTE_CLGEMMNATIVEKERNELCONFIGURATION_H__
+
+#include "arm_compute/core/CL/ICLGEMMKernelConfiguration.h"
+#include "arm_compute/core/CL/gemm/native/CLGEMMNativeKernelConfigurationBifrost.h"
+
+#include <memory>
+
+namespace arm_compute
+{
+namespace cl_gemm
+{
+/** CLGEMMNative factory class */
+class CLGEMMNativeKernelConfigurationFactory final
+{
+public:
+    /** Static method to construct CLGEMMNative kernel object accordingly with the GPU architecture
+     *
+     * @param[in] arch GPU target
+     *
+     * @return CLGEMMNative kernel configuration class
+     */
+    static std::unique_ptr<ICLGEMMKernelConfiguration> create(GPUTarget arch)
+    {
+        switch(get_arch_from_target(arch))
+        {
+            case GPUTarget::BIFROST:
+                return support::cpp14::make_unique<CLGEMMNativeKernelConfigurationBifrost>(arch);
+            default:
+                return nullptr;
+        }
+    }
+};
+} // namespace cl_gemm
+} // namespace arm_compute
+#endif /*__ARM_COMPUTE_CLGEMMNATIVEKERNELCONFIGURATION_H__ */
diff --git a/arm_compute/core/CL/gemm/native/CLGEMMNativeKernelConfigurationBifrost.h b/arm_compute/core/CL/gemm/native/CLGEMMNativeKernelConfigurationBifrost.h
new file mode 100644
index 0000000000..ea46818750
--- /dev/null
+++ b/arm_compute/core/CL/gemm/native/CLGEMMNativeKernelConfigurationBifrost.h
@@ -0,0 +1,64 @@
+/*
+ * Copyright (c) 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_CLGEMMNATIVEKERNELCONFIGURATIONBIFROST_H__
+#define __ARM_COMPUTE_CLGEMMNATIVEKERNELCONFIGURATIONBIFROST_H__
+
+#include "arm_compute/core/CL/ICLGEMMKernelConfiguration.h"
+
+namespace arm_compute
+{
+namespace cl_gemm
+{
+/** Bifrost based OpenCL GEMMNative configuration */
+class CLGEMMNativeKernelConfigurationBifrost final : public ICLGEMMKernelConfiguration
+{
+public:
+    /** Constructor
+     *
+     * @param[in] arch GPU target
+     */
+    CLGEMMNativeKernelConfigurationBifrost(GPUTarget arch);
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    CLGEMMNativeKernelConfigurationBifrost(const CLGEMMNativeKernelConfigurationBifrost &) = delete;
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    CLGEMMNativeKernelConfigurationBifrost &operator=(const CLGEMMNativeKernelConfigurationBifrost &) = delete;
+    /** Default Move Constructor. */
+    CLGEMMNativeKernelConfigurationBifrost(CLGEMMNativeKernelConfigurationBifrost &&) = default;
+    /** Default move assignment operator */
+    CLGEMMNativeKernelConfigurationBifrost &operator=(CLGEMMNativeKernelConfigurationBifrost &&) = default;
+
+    // Inherited overridden method
+    std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> configure(unsigned int m, unsigned int n, unsigned int k, unsigned int b, DataType data_type) override;
+
+private:
+    std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> configure_G71_f32(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
+    std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> configure_G71_u8(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
+    std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> configure_G76_f32(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
+    std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> configure_G76_u8(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
+    std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> configure_default_f32(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
+    std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> configure_default_u8(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
+};
+} // namespace cl_gemm
+} // namespace arm_compute
+#endif /*__ARM_COMPUTE_CLGEMMNATIVEKERNELCONFIGURATIONBIFROST_H__ */
diff --git a/arm_compute/core/CL/kernels/CLGEMMLowpMatrixMultiplyKernel.h b/arm_compute/core/CL/kernels/CLGEMMLowpMatrixMultiplyKernel.h
index e576271780..409ed1bec2 100644
--- a/arm_compute/core/CL/kernels/CLGEMMLowpMatrixMultiplyKernel.h
+++ b/arm_compute/core/CL/kernels/CLGEMMLowpMatrixMultiplyKernel.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017-2018 ARM Limited.
+ * Copyright (c) 2017-2019 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -32,8 +32,9 @@ class ICLTensor;
 
 /** OpenCL kernel to multiply matrices
  *
- * @note @ref CLGEMMLowpMatrixMultiplyKernel low precision matrix product kernel
- *  This kernel performs the following computation:
+ * @note This kernel should be used ONLY for Midgard architectures
+ *
+ * This kernel performs the following computation:
  *
  *  -# Convert a values from int8 to int32
  *  -# Convert b values from int8 to int32
@@ -55,24 +56,24 @@ public:
     CLGEMMLowpMatrixMultiplyKernel &operator=(CLGEMMLowpMatrixMultiplyKernel &&) = default;
     /** Initialise the kernel's input and output.
      *
-     * @param[in]  input0                    Input tensor containing the interleaved Matrix A. Data type supported: QASYMM8
-     * @param[in]  input1                    Input tensor containing the transposed1xW Matrix B. Data type supported: same as @p input0
-     * @param[out] output                    Output tensor to store the result of matrix multiplication. Data type supported: S32
-     * @param[in]  is_interleaved_transposed (Optional) True if input0 and input1 have been reshaped respectively using @ref CLGEMMReshapeLHSMatrixKernel and @ref CLGEMMReshapeRHSMatrixKernel
-     * @param[in]  reshape_info              (Optional) GEMM reshape info. If is_interleaved_transposed = true, this object must contain the information to understand how the matrix A and matrix B have been reshaped
+     * @note This kernel should be used ONLY for Midgard architectures
+     *
+     * @param[in]  input0    Input tensor containing the LHS matrix. Data type supported: QASYMM8
+     * @param[in]  input1    Input tensor containing the RHS matrix. Data type supported: same as @p input0
+     * @param[out] output    Output tensor to store the result of matrix multiplication. Data type supported: S32
+     * @param[in]  gemm_info (Optional) GEMM information used to retrieve the original dimensions of the input matrices
      */
-    void configure(const ICLTensor *input0, const ICLTensor *input1, ICLTensor *output, bool is_interleaved_transposed = true, const GEMMReshapeInfo &reshape_info = GEMMReshapeInfo());
+    void configure(const ICLTensor *input0, const ICLTensor *input1, ICLTensor *output, const GEMMReshapeInfo &gemm_info = GEMMReshapeInfo());
     /** Static function to check if given info will lead to a valid configuration of @ref CLGEMMLowpMatrixMultiplyKernel
      *
-     * @param[in] input0                    Input tensor info containing the interleaved Matrix A. Data type supported: QASYMM8
-     * @param[in] input1                    Input tensor info containing the transposed Matrix B. Data type supported: same as @p input0
-     * @param[in] output                    Output tensor info to store the result of matrix multiplication. Data type supported: S32
-     * @param[in] is_interleaved_transposed True if input0 and input1 have been reshaped respectively using @ref CLGEMMReshapeLHSMatrixKernel and @ref CLGEMMReshapeRHSMatrixKernel
-     * @param[in] reshape_info              GEMM reshape info. If is_interleaved_transposed = true, this object must contain the information to understand how the matrix A and matrix B have been reshaped
+     * @param[in] input0    Input tensor containing the LHS matrix. Data type supported: QASYMM8
+     * @param[in] input1    Input tensor containing the RHS matrix. Data type supported: same as @p input0
+     * @param[in] output    Output tensor to store the result of matrix multiplication. Data type supported: S32
+     * @param[in] gemm_info (Optional) GEMM information used to retrieve the original dimensions of the input matrices
      *
      * @return a status
      */
-    static Status validate(const ITensorInfo *input0, const ITensorInfo *input1, const ITensorInfo *output, bool is_interleaved_transposed, const GEMMReshapeInfo &reshape_info);
+    static Status validate(const ITensorInfo *input0, const ITensorInfo *input1, const ITensorInfo *output, const GEMMReshapeInfo &gemm_info = GEMMReshapeInfo());
 
     // Inherited methods overridden:
     void run(const Window &window, cl::CommandQueue &queue) override;