IVGCVSW-6493 Add GpuAcc Batch MatMul workload Fp32

* GpuAcc only supports up to 3D, so no 4D test have been added

Signed-off-by: Teresa Charlin <teresa.charlinreyes@arm.com>
Change-Id: Ie926cd45c350be624cbdc6cb27c89d2d3f60884b

1 files changed, 41 insertions, 0 deletions

diff --git a/src/backends/cl/workloads/ClBatchMatMulWorkload.hpp b/src/backends/cl/workloads/ClBatchMatMulWorkload.hpp
new file mode 100644
index 0000000000..5277efc947
--- /dev/null
+++ b/src/backends/cl/workloads/ClBatchMatMulWorkload.hpp
@@ -0,0 +1,41 @@
+//
+// Copyright © 2022 Arm Ltd and Contributors. All rights reserved.
+// SPDX-License-Identifier: MIT
+//
+
+#pragma once
+
+#include "ClBaseWorkload.hpp"
+
+#include <arm_compute/runtime/IFunction.h>
+#include <arm_compute/runtime/CL/CLTensor.h>
+#include <memory>
+
+namespace armnn
+{
+    arm_compute::Status ClBatchMatMulValidate(const TensorInfo& inputX,
+                                              const TensorInfo& inputY,
+                                              const TensorInfo& output,
+                                              const BatchMatMulDescriptor& descriptor);
+
+    class ClBatchMatMulWorkload : public ClBaseWorkload<BatchMatMulQueueDescriptor>
+    {
+    public:
+        ClBatchMatMulWorkload(const BatchMatMulQueueDescriptor& descriptor,
+                              const WorkloadInfo& info,
+                              const arm_compute::CLCompileContext& clCompileContext);
+        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 CL arm_compute::Tensors.
+        // Required to perform permutations.
+        arm_compute::CLTensor m_PermutedTensorX;
+        arm_compute::CLTensor m_PermutedTensorY;
+
+    };
+} //namespace armnn