Implement MatMul Function and Operator with Floating Point support for CPU

- Implements MatMul function and operator for floating point datatype FP16/FP32
- Includes support for transposing dynamic tensors prior to matrix multiplication.
- Adds tests for 2D/3D/4D+ tensors in MatMul with F32/F16 datatype (with all combinations of transposed/not-transposed tensors)
- Updates fixture to allow for testing fused activation in MatMul
- Adds tests for matmul with and without fused activation

Resolved: [COMPMID-5898]
Signed-off-by: Mohammed Suhail Munshi <MohammedSuhail.Munshi@arm.com>
Change-Id: Iefa84b26dd723c9a51e6c3f91023152c6c31ace2
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/9411
Reviewed-by: SiCong Li <sicong.li@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Benchmark: Arm Jenkins <bsgcomp@arm.com>

4 files changed, 125 insertions, 2 deletions

diff --git a/arm_compute/core/ITensorInfo.h b/arm_compute/core/ITensorInfo.h
index 1382649e74..7b0fd1c2b7 100644
--- a/arm_compute/core/ITensorInfo.h
+++ b/arm_compute/core/ITensorInfo.h
@@ -36,6 +36,9 @@
 
 namespace arm_compute
 {
+// Note: Any changes to the fields of the class below that have setters should be mirrored
+// (if possible) in the auto_init_if_empty function in AutoConfiguration.h
+
 /** Store the tensor's metadata */
 class ITensorInfo : public misc::ICloneable<ITensorInfo>
 {
diff --git a/arm_compute/runtime/CL/functions/CLMatMul.h b/arm_compute/runtime/CL/functions/CLMatMul.h
index 56dd9c5655..712bac06bf 100644
--- a/arm_compute/runtime/CL/functions/CLMatMul.h
+++ b/arm_compute/runtime/CL/functions/CLMatMul.h
@@ -35,6 +35,13 @@ class ITensorInfo;
 class MatMulInfo;
 class Status;
 
+/** Settings for MatMul OpenCL implementation */
+class GpuMatMulSettings
+{
+public:
+    /* Placeholder for operator parity between CPU/GPU */
+};
+
 /** Basic function to execute MatMul (Matrix Multiplication) on OpenCL */
 class CLMatMul : public IFunction
 {
@@ -73,13 +80,14 @@ public:
      * @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
+     * @param[in]  settings        Class containing flags for function level settings
      */
-    void configure(const CLCompileContext &compile_context, ICLTensor *rhs, ICLTensor *lhs, ICLTensor *output, const MatMulInfo &matmul_info);
+    void configure(const CLCompileContext &compile_context, ICLTensor *rhs, ICLTensor *lhs, ICLTensor *output, const MatMulInfo &matmul_info, const GpuMatMulSettings &settings = GpuMatMulSettings{});
     /** Initialise the kernel's inputs and output
      *
      * Similar to @ref CLMatMul::configure()
      */
-    void configure(ICLTensor *lhs, ICLTensor *rhs, ICLTensor *output, const MatMulInfo &matmul_info);
+    void configure(ICLTensor *lhs, ICLTensor *rhs, ICLTensor *output, const MatMulInfo &matmul_info, const GpuMatMulSettings &settings = GpuMatMulSettings{});
     /** Static function to check if given info will lead to a valid configuration of @ref CLMatMul.
      *
      * Similar to @ref CLMatMul::configure()
diff --git a/arm_compute/runtime/NEON/NEFunctions.h b/arm_compute/runtime/NEON/NEFunctions.h
index 8dbe6c59c3..836cba7699 100644
--- a/arm_compute/runtime/NEON/NEFunctions.h
+++ b/arm_compute/runtime/NEON/NEFunctions.h
@@ -74,6 +74,7 @@
 #include "arm_compute/runtime/NEON/functions/NELSTMLayer.h"
 #include "arm_compute/runtime/NEON/functions/NELSTMLayerQuantized.h"
 #include "arm_compute/runtime/NEON/functions/NELogical.h"
+#include "arm_compute/runtime/NEON/functions/NEMatMul.h"
 #include "arm_compute/runtime/NEON/functions/NEMaxUnpoolingLayer.h"
 #include "arm_compute/runtime/NEON/functions/NEMeanStdDevNormalizationLayer.h"
 #include "arm_compute/runtime/NEON/functions/NENormalizationLayer.h"
diff --git a/arm_compute/runtime/NEON/functions/NEMatMul.h b/arm_compute/runtime/NEON/functions/NEMatMul.h
new file mode 100644
index 0000000000..0f3e3adacc
--- /dev/null
+++ b/arm_compute/runtime/NEON/functions/NEMatMul.h
@@ -0,0 +1,111 @@
+/*
+ * 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_NEON_FUNCTIONS_NEMATMUL
+#define ARM_COMPUTE_RUNTIME_NEON_FUNCTIONS_NEMATMUL
+
+#include "arm_compute/runtime/IFunction.h"
+#include <memory>
+
+namespace arm_compute
+{
+/** Settings for MatMul Cpu implementation*/
+class CpuMatMulSettings
+{
+public:
+    // get fast math flag
+    bool fast_math() const
+    {
+        return _fast_math;
+    }
+    // Set fast math flag
+    CpuMatMulSettings &fast_math(bool fmath)
+    {
+        _fast_math = fmath;
+        return *this;
+    };
+
+private:
+    bool _fast_math{ false };
+};
+
+// Forward declarations
+class ITensor;
+class ITensorInfo;
+class MatMulInfo;
+class Status;
+
+/** Basic function to run the following operators:
+ *
+ * -# @ref cpu::CpuMatMul
+ */
+class NEMatMul : public IFunction
+{
+public:
+    /** Constructor */
+    NEMatMul();
+    /** Destructor */
+    ~NEMatMul();
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    NEMatMul(const NEMatMul &) = delete;
+    /** Default move constructor */
+    NEMatMul(NEMatMul &&) = default;
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    NEMatMul &operator=(const NEMatMul &) = delete;
+    /** Default move assignment operator */
+    NEMatMul &operator=(NEMatMul &&) = default;
+    /** Initialize
+     *
+     * Valid data layouts:
+     * - Any
+     *
+     * Valid data type configurations:
+     * |src0           |src1               |dst            |
+     * |:--------------|:------------------|:--------------|
+     * |F32            |F32                |F32            |
+     * |F16            |F16                |F16            |
+     *
+     * @param[in]  lhs      Input source tensor.
+     * @param[in]  rhs      Input source tensor.
+     * @param[out] output   Output tensor. Data type supported: same as @p lhs/rhs
+     * @param[in]  info     Class containing flags to transpose lhs/rhs
+     * @param[in]  settings Class containing flags for function level settings i.e fast math
+     */
+    void configure(ITensor *lhs, ITensor *rhs, ITensor *output, const MatMulInfo &info, const CpuMatMulSettings &settings);
+    /** Static function to check if given info will lead to a valid configuration of @ref NEMatMul
+     *
+     * Parameters are similar to @ref NEMatMul::configure()
+     *
+     * @return Status
+     */
+    static Status validate(const ITensorInfo *lhs, const ITensorInfo *rhs, const ITensorInfo *output, const MatMulInfo &info, const CpuMatMulSettings &settings);
+
+    // Inherited methods overridden
+    void run() override;
+
+private:
+    struct Impl;
+    std::unique_ptr<Impl> _impl;
+};
+}
+#endif /* ARM_COMPUTE_RUNTIME_NEON_FUNCTIONS_NEMATMUL */