Implement MatMul Function

Resolves: COMPMID-5949
Signed-off-by: Ramy Elgammal <ramy.elgammal@arm.com>
Change-Id: Idd8cfe6ea94a14f0b23178f6781251b5f0955563
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/9390
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Gunes Bayir <gunes.bayir@arm.com>
Benchmark: Arm Jenkins <bsgcomp@arm.com>

2 files changed, 101 insertions, 1 deletions

diff --git a/arm_compute/runtime/CL/CLFunctions.h b/arm_compute/runtime/CL/CLFunctions.h
index e37134d454..26e459680c 100644
--- a/arm_compute/runtime/CL/CLFunctions.h
+++ b/arm_compute/runtime/CL/CLFunctions.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2016-2022 Arm Limited.
+ * Copyright (c) 2016-2023 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -77,6 +77,7 @@
 #include "arm_compute/runtime/CL/functions/CLLogicalAnd.h"
 #include "arm_compute/runtime/CL/functions/CLLogicalNot.h"
 #include "arm_compute/runtime/CL/functions/CLLogicalOr.h"
+#include "arm_compute/runtime/CL/functions/CLMatMul.h"
 #include "arm_compute/runtime/CL/functions/CLMaxUnpoolingLayer.h"
 #include "arm_compute/runtime/CL/functions/CLMeanStdDevNormalizationLayer.h"
 #include "arm_compute/runtime/CL/functions/CLNormalizationLayer.h"
diff --git a/arm_compute/runtime/CL/functions/CLMatMul.h b/arm_compute/runtime/CL/functions/CLMatMul.h
new file mode 100644
index 0000000000..56dd9c5655
--- /dev/null
+++ b/arm_compute/runtime/CL/functions/CLMatMul.h
@@ -0,0 +1,99 @@
+/*
+ * Copyright (c) 2023 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_RUNTIME_CL_FUNCTIONS_CLMATMUL
+#define ARM_COMPUTE_RUNTIME_CL_FUNCTIONS_CLMATMUL
+
+#include "arm_compute/runtime/IFunction.h"
+#include <memory>
+namespace arm_compute
+{
+// Forward declarations for used types instead of including their header, that could minimize compile time
+class CLCompileContext;
+class ICLTensor;
+class ITensorInfo;
+class MatMulInfo;
+class Status;
+
+/** Basic function to execute MatMul (Matrix Multiplication) on OpenCL */
+class CLMatMul : public IFunction
+{
+public:
+    /** Default constructor.*/
+    CLMatMul();
+    /** Default destructor */
+    ~CLMatMul();
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    CLMatMul(const CLMatMul &) = delete;
+    /** Default move constructor */
+    CLMatMul(CLMatMul &&);
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    CLMatMul &operator=(const CLMatMul &) = delete;
+    /** Default move assignment operator */
+    CLMatMul &operator=(CLMatMul &&);
+    /** Initialise the kernel's inputs and output
+     *
+     * Valid data layouts:
+     * - All
+     *
+     * Valid data type configurations:
+     * |lhs          |rhs          |output         |
+     * |:------------|:------------|:--------------|
+     * |F32          |F32          |F32            |
+     * |F16          |F16          |F16            |
+     *
+     * @note BatchMatMul: Batched Matrix Multiply - [A * B], Multiplies all slices (slice is an element of a batch) of Tensors A and B
+     *                    and stores the result in the dst tensor of the same batch size.
+     *                    Batch here is number of slices from A and B multiplied at a time, do not confuse with the batch dimension 'N' of NHWC/NCHW
+     *                    For NHWC for example: the batch is the higher dimensions H * N, and in general it is H * all higher dimensions.
+     * @note All tensors must have the same data type.
+     *
+     * @param[in]  compile_context The compile context to be used.
+     * @param[in]  lhs             LHS input tensor (Matrix or Vector A). Data types supported: F16/F32
+     * @param[in]  rhs             RHS input tensor (Matrix B). Data type supported: same as @p lhs.
+     * @param[out] output          Output tensor. Data type supported: same as @p lhs.
+     * @param[in]  matmul_info     Attributes for MatMul
+     */
+    void configure(const CLCompileContext &compile_context, ICLTensor *rhs, ICLTensor *lhs, ICLTensor *output, const MatMulInfo &matmul_info);
+    /** Initialise the kernel's inputs and output
+     *
+     * Similar to @ref CLMatMul::configure()
+     */
+    void configure(ICLTensor *lhs, ICLTensor *rhs, ICLTensor *output, const MatMulInfo &matmul_info);
+    /** Static function to check if given info will lead to a valid configuration of @ref CLMatMul.
+     *
+     * Similar to @ref CLMatMul::configure()
+     *
+     * @return a status
+     */
+    static Status validate(const ITensorInfo *lhs, const ITensorInfo *rhs, const ITensorInfo *output, const MatMulInfo &matmul_info);
+    // Inherited methods overridden:
+    void run() override;
+
+private:
+    struct Impl;
+    std::unique_ptr<Impl> _impl;
+};
+} // namespace arm_compute
+
+#endif /* ARM_COMPUTE_RUNTIME_CL_FUNCTIONS_CLMATMUL */