IVGCVSW-6494 Add CpuAcc Batch MatMul Workload Fp32

Signed-off-by: Teresa Charlin <teresa.charlinreyes@arm.com>
Change-Id: I2def6995f81d33e68f1ea45d8d19a1e6294049b1

1 files changed, 41 insertions, 0 deletions

diff --git a/src/backends/neon/workloads/NeonBatchMatMulWorkload.hpp b/src/backends/neon/workloads/NeonBatchMatMulWorkload.hpp
new file mode 100644
index 0000000000..cb004d2478
--- /dev/null
+++ b/src/backends/neon/workloads/NeonBatchMatMulWorkload.hpp
@@ -0,0 +1,41 @@
+//
+// Copyright © 2022 Arm Ltd and Contributors. All rights reserved.
+// SPDX-License-Identifier: MIT
+//
+
+#pragma once
+
+#include "NeonBaseWorkload.hpp"
+
+#include <arm_compute/runtime/IFunction.h>
+#include <arm_compute/runtime/Tensor.h>
+
+#include <memory>
+
+namespace armnn
+{
+    arm_compute::Status NeonBatchMatMulValidate(const TensorInfo& inputX,
+                                                const TensorInfo& inputY,
+                                                const TensorInfo& output,
+                                                const BatchMatMulDescriptor& descriptor);
+
+    class NeonBatchMatMulWorkload : public NeonBaseWorkload<BatchMatMulQueueDescriptor>
+    {
+    public:
+        NeonBatchMatMulWorkload(const BatchMatMulQueueDescriptor& descriptor,
+                                const WorkloadInfo& info);
+        virtual void Execute() const override;
+
+    private:
+        // ACL layers required to fully form a Batch Mat Mul layer.
+        std::unique_ptr<arm_compute::IFunction> m_GEMMLayer;
+        std::unique_ptr<arm_compute::IFunction> m_PermuteLayerX;
+        std::unique_ptr<arm_compute::IFunction> m_PermuteLayerY;
+
+        // Additional ACL arm_compute::Tensors.
+        // Required to perform permutations.
+        arm_compute::Tensor m_PermutedTensorX;
+        arm_compute::Tensor m_PermutedTensorY;
+
+    };
+} //namespace armnn