IVGCVSW-2291 TILE Operator CL Implementation

 * Added Tile Operator Implementation to CL
 * Added calls to the existing UnitTests
 * Added Documentation

Signed-off-by: Cian McGriskin <cian.mcgriskin@arm.com>
Change-Id: If7d25c7aa669c24e7816e5d445c7a3b9ce6972d4

14 files changed, 216 insertions, 35 deletions

diff --git a/delegate/test/TileTest.cpp b/delegate/test/TileTest.cpp
index 2e20859f5b..32f5610bce 100644
--- a/delegate/test/TileTest.cpp
+++ b/delegate/test/TileTest.cpp
@@ -53,7 +53,6 @@ void TileFloat32Test(std::vector<armnn::BackendId>& backends)
                      expectedOutputShape);
 }
 
-#if defined(TILE_GPUACC)
 TEST_SUITE("TileTests_GpuAccTests")
 {
 
@@ -64,7 +63,6 @@ TEST_SUITE("TileTests_GpuAccTests")
     }
 
 } // TEST_SUITE("Tile_Float32_GpuAcc_Test")
-#endif
 
 TEST_SUITE("TileTests_CpuAccTests")
 {
diff --git a/docs/02_operator_list.dox b/docs/02_operator_list.dox
index e3c9f8c1f2..a37b6fa3a7 100644
--- a/docs/02_operator_list.dox
+++ b/docs/02_operator_list.dox
@@ -3328,6 +3328,7 @@ where N = batches, C = channels, H = height, W = width
        <tr><td>FLOAT32
        <tr><td>QASYMMS8
        <tr><td>QASYMMU8
+       <tr><td>QSYMMS8
        <tr><td>QSYMMS16
        <tr><td>SIGNED32
       </table>
@@ -3352,12 +3353,18 @@ where N = batches, C = channels, H = height, W = width
   <td>GpuAcc
   <td>
       <ul>
-       <li>None
+       <li>All
       </ul>
   <td>
     <table>
     <tr><th>
-    <tr><td>None
+    <tr><td>FLOAT16
+    <tr><td>FLOAT32
+    <tr><td>QASYMMS8
+    <tr><td>QASYMMU8
+    <tr><td>QSYMMS8
+    <tr><td>QSYMMS16
+    <tr><td>SIGNED32
     </table>
 <tr>
   <td rowspan="3">TransposeConvolution2dLayer
diff --git a/src/backends/backendsCommon/test/TileEndToEndTestImpl.hpp b/src/backends/backendsCommon/test/TileEndToEndTestImpl.hpp
index 4047e5ad8e..03b76849e1 100644
--- a/src/backends/backendsCommon/test/TileEndToEndTestImpl.hpp
+++ b/src/backends/backendsCommon/test/TileEndToEndTestImpl.hpp
@@ -32,25 +32,25 @@ void TileEndToEnd(const std::vector<BackendId>& backends)
     int32_t qOffset = 0;
     bool    qConst  = true;
 
-    const TensorShape inputTensorShape =  { 2, 3 };
-    const TensorShape outputTensorShape = { 4, 6 };
+    const TensorShape inputTensorShape =  { 6 };
+    const TensorShape outputTensorShape = { 30 };
 
     TensorInfo inputInfo  (inputTensorShape, ArmnnType, qScale, qOffset, qConst);
     TensorInfo outputInfo (outputTensorShape, ArmnnType,qScale, qOffset);
 
     std::vector<T> inputData = armnnUtils::QuantizedVector<T>({
-        0.f, 1.f, 2.f,
-        3.f, 4.f, 5.f
+        65, 144, 91, 161, 56, 73
     }, qScale, qOffset);
 
     std::vector<T> expectedOutputData = armnnUtils::QuantizedVector<T>({
-        0.f, 1.f, 2.f, 0.f, 1.f, 2.f,
-        3.f, 4.f, 5.f, 3.f, 4.f, 5.f,
-        0.f, 1.f, 2.f, 0.f, 1.f, 2.f,
-        3.f, 4.f, 5.f, 3.f, 4.f, 5.f
+        65, 144, 91, 161, 56, 73,
+        65, 144, 91, 161, 56, 73,
+        65, 144, 91, 161, 56, 73,
+        65, 144, 91, 161, 56, 73,
+        65, 144, 91, 161, 56, 73
     }, qScale, qOffset);
 
-    auto descriptor = armnn::TileDescriptor(std::vector<uint32_t>{ 2, 2 });
+    auto descriptor = armnn::TileDescriptor(std::vector<uint32_t>{ 5 });
     INetworkPtr network = CreateTileNetwork(descriptor, inputInfo, outputInfo);
 
     std::map<int, std::vector<T>> inputTensor          = { { 0, inputData  } };
diff --git a/src/backends/backendsCommon/test/layerTests/TileTestImpl.cpp b/src/backends/backendsCommon/test/layerTests/TileTestImpl.cpp
index 0b13bba425..c61da26bb2 100644
--- a/src/backends/backendsCommon/test/layerTests/TileTestImpl.cpp
+++ b/src/backends/backendsCommon/test/layerTests/TileTestImpl.cpp
@@ -110,13 +110,13 @@ LayerTestResult<T, 2> Tile2dTest(armnn::IWorkloadFactory& workloadFactory,
                                  const armnn::IBackendInternal::IMemoryManagerSharedPtr& memoryManager,
                                  const armnn::ITensorHandleFactory& tensorHandleFactory)
 {
-    auto descriptor = armnn::TileDescriptor(std::vector<uint32_t>{ 2, 2 });
+    auto descriptor = armnn::TileDescriptor(std::vector<uint32_t>{ 2, 5 });
 
     float qScale = 1.0f;
     int32_t qOffset = 0;
 
     armnn::TensorShape inputShape  = { 2, 3 };
-    armnn::TensorShape outputShape = { 4, 6 };
+    armnn::TensorShape outputShape = { 4, 15 };
 
     armnn::TensorInfo inputInfo(inputShape, ArmnnType);
     armnn::TensorInfo outputInfo(outputShape, ArmnnType);
@@ -130,11 +130,11 @@ LayerTestResult<T, 2> Tile2dTest(armnn::IWorkloadFactory& workloadFactory,
 
     std::vector<T> expectedOutput = armnnUtils::QuantizedVector<T>(
     {
-        0.f, 1.f, 2.f, 0.f, 1.f, 2.f,
-        3.f, 4.f, 5.f, 3.f, 4.f, 5.f,
+        0.f, 1.f, 2.f, 0.f, 1.f, 2.f, 0.f, 1.f, 2.f, 0.f, 1.f, 2.f, 0.f, 1.f, 2.f,
+        3.f, 4.f, 5.f, 3.f, 4.f, 5.f, 3.f, 4.f, 5.f, 3.f, 4.f, 5.f, 3.f, 4.f, 5.f,
 
-        0.f, 1.f, 2.f, 0.f, 1.f, 2.f,
-        3.f, 4.f, 5.f, 3.f, 4.f, 5.f
+        0.f, 1.f, 2.f, 0.f, 1.f, 2.f, 0.f, 1.f, 2.f, 0.f, 1.f, 2.f, 0.f, 1.f, 2.f,
+        3.f, 4.f, 5.f, 3.f, 4.f, 5.f, 3.f, 4.f, 5.f, 3.f, 4.f, 5.f, 3.f, 4.f, 5.f
     }, qScale, qOffset);
 
     return TileTestImpl<T, 2>(workloadFactory,
@@ -152,37 +152,49 @@ LayerTestResult<T, 3> Tile3dTest(armnn::IWorkloadFactory& workloadFactory,
                                  const armnn::IBackendInternal::IMemoryManagerSharedPtr& memoryManager,
                                  const armnn::ITensorHandleFactory& tensorHandleFactory)
 {
-    auto descriptor = armnn::TileDescriptor(std::vector<uint32_t>{ 1, 2, 1 });
+    auto descriptor = armnn::TileDescriptor(std::vector<uint32_t>{ 1, 5, 5 });
 
     float qScale = 1.0f;
     int32_t qOffset = 0;
 
-    armnn::TensorShape inputShape  = { 2, 2, 3 };
-    armnn::TensorShape outputShape = { 2, 4, 3 };
+    armnn::TensorShape inputShape  = { 2, 2, 2 };
+    armnn::TensorShape outputShape = { 2, 10, 10 };
 
     armnn::TensorInfo inputInfo(inputShape, ArmnnType);
     armnn::TensorInfo outputInfo(outputShape, ArmnnType);
     std::vector<T> input = armnnUtils::QuantizedVector<T>(
     {
-        0.f, 1.f, 2.f,
-        3.f, 4.f, 5.f,
+        1.1f, 2.12f,
+        3.3f, 4.12234f,
 
-        6.f,  7.f,  8.f,
-        9.f, 10.f, 11.f
+        1.1f, 2.12f,
+        3.3f, 4.12234f,
     }, qScale, qOffset);
 
 
     std::vector<T> expectedOutput = armnnUtils::QuantizedVector<T>(
     {
-        0.f, 1.f, 2.f,
-        3.f, 4.f, 5.f,
-        0.f, 1.f, 2.f,
-        3.f, 4.f, 5.f,
-
-        6.f,  7.f,  8.f,
-        9.f, 10.f, 11.f,
-        6.f,  7.f,  8.f,
-        9.f, 10.f, 11.f
+        1.1f, 2.12f, 1.1f, 2.12f, 1.1f, 2.12f, 1.1f, 2.12f, 1.1f, 2.12f,
+        3.3f, 4.12234f, 3.3f, 4.12234f, 3.3f, 4.12234f, 3.3f, 4.12234f, 3.3f, 4.12234f,
+        1.1f, 2.12f, 1.1f, 2.12f, 1.1f, 2.12f, 1.1f, 2.12f, 1.1f, 2.12f,
+        3.3f, 4.12234f, 3.3f, 4.12234f, 3.3f, 4.12234f, 3.3f, 4.12234f, 3.3f, 4.12234f,
+        1.1f, 2.12f, 1.1f, 2.12f, 1.1f, 2.12f, 1.1f, 2.12f, 1.1f, 2.12f,
+        3.3f, 4.12234f, 3.3f, 4.12234f, 3.3f, 4.12234f, 3.3f, 4.12234f, 3.3f, 4.12234f,
+        1.1f, 2.12f, 1.1f, 2.12f, 1.1f, 2.12f, 1.1f, 2.12f, 1.1f, 2.12f,
+        3.3f, 4.12234f, 3.3f, 4.12234f, 3.3f, 4.12234f, 3.3f, 4.12234f, 3.3f, 4.12234f,
+        1.1f, 2.12f, 1.1f, 2.12f, 1.1f, 2.12f, 1.1f, 2.12f, 1.1f, 2.12f,
+        3.3f, 4.12234f, 3.3f, 4.12234f, 3.3f, 4.12234f, 3.3f, 4.12234f, 3.3f, 4.12234f,
+
+        1.1f, 2.12f, 1.1f, 2.12f, 1.1f, 2.12f, 1.1f, 2.12f, 1.1f, 2.12f,
+        3.3f, 4.12234f, 3.3f, 4.12234f, 3.3f, 4.12234f, 3.3f, 4.12234f, 3.3f, 4.12234f,
+        1.1f, 2.12f, 1.1f, 2.12f, 1.1f, 2.12f, 1.1f, 2.12f, 1.1f, 2.12f,
+        3.3f, 4.12234f, 3.3f, 4.12234f, 3.3f, 4.12234f, 3.3f, 4.12234f, 3.3f, 4.12234f,
+        1.1f, 2.12f, 1.1f, 2.12f, 1.1f, 2.12f, 1.1f, 2.12f, 1.1f, 2.12f,
+        3.3f, 4.12234f, 3.3f, 4.12234f, 3.3f, 4.12234f, 3.3f, 4.12234f, 3.3f, 4.12234f,
+        1.1f, 2.12f, 1.1f, 2.12f, 1.1f, 2.12f, 1.1f, 2.12f, 1.1f, 2.12f,
+        3.3f, 4.12234f, 3.3f, 4.12234f, 3.3f, 4.12234f, 3.3f, 4.12234f, 3.3f, 4.12234f,
+        1.1f, 2.12f, 1.1f, 2.12f, 1.1f, 2.12f, 1.1f, 2.12f, 1.1f, 2.12f,
+        3.3f, 4.12234f, 3.3f, 4.12234f, 3.3f, 4.12234f, 3.3f, 4.12234f, 3.3f, 4.12234f
     }, qScale, qOffset);
 
     return TileTestImpl<T, 3>(workloadFactory,
diff --git a/src/backends/cl/ClLayerSupport.cpp b/src/backends/cl/ClLayerSupport.cpp
index ff2b576f3d..60dab0538a 100644
--- a/src/backends/cl/ClLayerSupport.cpp
+++ b/src/backends/cl/ClLayerSupport.cpp
@@ -81,6 +81,7 @@
 #include "workloads/ClStackWorkload.hpp"
 #include "workloads/ClStridedSliceWorkload.hpp"
 #include "workloads/ClSubtractionWorkload.hpp"
+#include "workloads/ClTileWorkload.hpp"
 #include "workloads/ClTransposeConvolution2dWorkload.hpp"
 #include "workloads/ClTransposeWorkload.hpp"
 #include "workloads/ClUnidirectionalSequenceLstmFloatWorkload.hpp"
@@ -627,6 +628,11 @@ bool ClLayerSupport::IsLayerSupported(const LayerType& type,
             ARMNN_NO_DEPRECATE_WARN_BEGIN
             return IsSubtractionSupported(infos[0], infos[1], infos[2], reasonIfUnsupported);
             ARMNN_NO_DEPRECATE_WARN_END
+        case LayerType::Tile:
+            return IsTileSupported(infos[0],
+                                   infos[1],
+                                   *(PolymorphicDowncast<const TileDescriptor*>(&descriptor)),
+                                   reasonIfUnsupported);
         case LayerType::Transpose:
             return IsTransposeSupported(infos[0],
                                         infos[1],
@@ -1527,6 +1533,18 @@ bool ClLayerSupport::IsSubtractionSupported(const TensorInfo& input0,
                                    nullptr);
 }
 
+bool ClLayerSupport::IsTileSupported(const TensorInfo& input,
+                                     const TensorInfo& output,
+                                     const TileDescriptor& descriptor,
+                                     Optional<std::string&> reasonIfUnsupported) const
+{
+    FORWARD_WORKLOAD_VALIDATE_FUNC(ClTileWorkloadValidate,
+                                   reasonIfUnsupported,
+                                   input,
+                                   output,
+                                   descriptor);
+}
+
 bool ClLayerSupport::IsTransposeConvolution2dSupported(const TensorInfo& input,
                                                        const TensorInfo& output,
                                                        const TransposeConvolution2dDescriptor& descriptor,
diff --git a/src/backends/cl/ClLayerSupport.hpp b/src/backends/cl/ClLayerSupport.hpp
index 737ab7a850..9a8f896cab 100644
--- a/src/backends/cl/ClLayerSupport.hpp
+++ b/src/backends/cl/ClLayerSupport.hpp
@@ -343,6 +343,11 @@ public:
                                            const Optional<TensorInfo>& biases,
                                            Optional<std::string&> reasonIfUnsupported = EmptyOptional()) const;
 
+    bool IsTileSupported(const TensorInfo& input,
+                         const TensorInfo& output,
+                         const TileDescriptor& descriptor,
+                         Optional<std::string&> reasonIfUnsupported = EmptyOptional()) const;
+
     bool IsTransposeSupported(const TensorInfo& input,
                               const TensorInfo& output,
                               const TransposeDescriptor& descriptor,
diff --git a/src/backends/cl/ClWorkloadFactory.cpp b/src/backends/cl/ClWorkloadFactory.cpp
index 7b4e9be908..da05f67711 100644
--- a/src/backends/cl/ClWorkloadFactory.cpp
+++ b/src/backends/cl/ClWorkloadFactory.cpp
@@ -755,6 +755,11 @@ std::unique_ptr<IWorkload> ClWorkloadFactory::CreateWorkload(LayerType type,
             auto subtractionQueueDescriptor = PolymorphicDowncast<const SubtractionQueueDescriptor*>(&descriptor);
             return MakeWorkload<ClSubtractionWorkload>(*subtractionQueueDescriptor, info, m_CLCompileContext);
         }
+        case LayerType::Tile:
+        {
+            auto tileQueueDescriptor = PolymorphicDowncast<const TileQueueDescriptor*>(&descriptor);
+            return MakeWorkload<ClTileWorkload>(*tileQueueDescriptor, info, m_CLCompileContext);
+        }
         case LayerType::Transpose :
         {
             auto transposeQueueDescriptor = PolymorphicDowncast<const TransposeQueueDescriptor*>(&descriptor);
diff --git a/src/backends/cl/backend.mk b/src/backends/cl/backend.mk
index 5ceab700e4..34993059c2 100644
--- a/src/backends/cl/backend.mk
+++ b/src/backends/cl/backend.mk
@@ -90,6 +90,7 @@ BACKEND_SOURCES := \
         workloads/ClStackWorkload.cpp \
         workloads/ClStridedSliceWorkload.cpp \
         workloads/ClSubtractionWorkload.cpp \
+        workloads/ClTileWorkload.cpp \
         workloads/ClTransposeConvolution2dWorkload.cpp \
         workloads/ClTransposeWorkload.cpp \
         workloads/ClUnidirectionalSequenceLstmFloatWorkload.cpp
diff --git a/src/backends/cl/test/ClEndToEndTests.cpp b/src/backends/cl/test/ClEndToEndTests.cpp
index 2d7a813e60..50ba1dbb99 100644
--- a/src/backends/cl/test/ClEndToEndTests.cpp
+++ b/src/backends/cl/test/ClEndToEndTests.cpp
@@ -26,6 +26,7 @@
 #include <backendsCommon/test/SpaceToDepthEndToEndTestImpl.hpp>
 #include <backendsCommon/test/SplitterEndToEndTestImpl.hpp>
 #include <backendsCommon/test/SubgraphUtilsTest.hpp>
+#include <backendsCommon/test/TileEndToEndTestImpl.hpp>
 #include <backendsCommon/test/TransposeConvolution2dEndToEndTestImpl.hpp>
 #include <backendsCommon/test/TransposeEndToEndTestImpl.hpp>
 
@@ -466,6 +467,42 @@ TEST_CASE("ClSplitter4dDim3EndToEndUint8Test")
     Splitter4dDim3EndToEnd<armnn::DataType::QAsymmU8>(clDefaultBackends);
 }
 
+// Tile
+TEST_CASE("ClTileEndToEndFloat32")
+{
+    TileEndToEnd<armnn::DataType::Float32>(clDefaultBackends);
+}
+
+TEST_CASE("ClTileEndToEndFloat16")
+{
+    TileEndToEnd<armnn::DataType::Float16>(clDefaultBackends);
+}
+
+TEST_CASE("ClTileEndToEndQAsymmS8")
+{
+    TileEndToEnd<armnn::DataType::QAsymmS8>(clDefaultBackends);
+}
+
+TEST_CASE("ClTileEndToEndQAsymmU8")
+{
+    TileEndToEnd<armnn::DataType::QAsymmU8>(clDefaultBackends);
+}
+
+TEST_CASE("ClTileEndToEndQSymmS8")
+{
+    TileEndToEnd<armnn::DataType::QSymmS8>(clDefaultBackends);
+}
+
+TEST_CASE("ClTileEndToEndQSymmS16")
+{
+    TileEndToEnd<armnn::DataType::QSymmS16>(clDefaultBackends);
+}
+
+TEST_CASE("ClTileEndToEndSigned32")
+{
+    TileEndToEnd<armnn::DataType::Signed32>(clDefaultBackends);
+}
+
 // TransposeConvolution2d
 TEST_CASE("ClTransposeConvolution2dEndToEndFloatNchwTest")
 {
diff --git a/src/backends/cl/test/ClLayerTests.cpp b/src/backends/cl/test/ClLayerTests.cpp
index 93ad653441..d8d451ead3 100644
--- a/src/backends/cl/test/ClLayerTests.cpp
+++ b/src/backends/cl/test/ClLayerTests.cpp
@@ -2033,6 +2033,24 @@ ARMNN_AUTO_TEST_FIXTURE(DequantizeSimpleInt16, ClContextControlFixture, Dequanti
 ARMNN_AUTO_TEST_FIXTURE(DequantizeSimpleUint8ToFp16, ClContextControlFixture, DequantizeSimpleUint8ToFp16Test)
 ARMNN_AUTO_TEST_FIXTURE(DequantizeSimpleInt16ToFp16, ClContextControlFixture, DequantizeSimpleInt16ToFp16Test)
 
+//Tile
+ARMNN_AUTO_TEST_FIXTURE_WITH_THF(Tile1dTestFloat32, ClContextControlFixture, Tile1dTest<DataType::Float32>)
+ARMNN_AUTO_TEST_FIXTURE_WITH_THF(Tile2dTestFloat32, ClContextControlFixture, Tile2dTest<DataType::Float32>)
+ARMNN_AUTO_TEST_FIXTURE_WITH_THF(Tile3dTestFloat32, ClContextControlFixture, Tile3dTest<DataType::Float32>)
+ARMNN_AUTO_TEST_FIXTURE_WITH_THF(Tile4dTestFloat32, ClContextControlFixture, Tile4dTest<DataType::Float32>)
+ARMNN_AUTO_TEST_FIXTURE_WITH_THF(Tile1dTestFloat16, ClContextControlFixture, Tile1dTest<DataType::Float16>)
+ARMNN_AUTO_TEST_FIXTURE_WITH_THF(Tile2dTestFloat16, ClContextControlFixture, Tile2dTest<DataType::Float16>)
+ARMNN_AUTO_TEST_FIXTURE_WITH_THF(Tile3dTestFloat16, ClContextControlFixture, Tile3dTest<DataType::Float16>)
+ARMNN_AUTO_TEST_FIXTURE_WITH_THF(Tile4dTestFloat16, ClContextControlFixture, Tile4dTest<DataType::Float16>)
+ARMNN_AUTO_TEST_FIXTURE_WITH_THF(Tile1dTestInt8, ClContextControlFixture, Tile1dTest<DataType::QAsymmS8>)
+ARMNN_AUTO_TEST_FIXTURE_WITH_THF(Tile2dTestInt8, ClContextControlFixture, Tile2dTest<DataType::QAsymmS8>)
+ARMNN_AUTO_TEST_FIXTURE_WITH_THF(Tile3dTestInt8, ClContextControlFixture, Tile3dTest<DataType::QAsymmS8>)
+ARMNN_AUTO_TEST_FIXTURE_WITH_THF(Tile4dTestInt8, ClContextControlFixture, Tile4dTest<DataType::QAsymmS8>)
+ARMNN_AUTO_TEST_FIXTURE_WITH_THF(Tile1dTestUint8, ClContextControlFixture, Tile1dTest<DataType::QAsymmU8>)
+ARMNN_AUTO_TEST_FIXTURE_WITH_THF(Tile2dTestUint8, ClContextControlFixture, Tile2dTest<DataType::QAsymmU8>)
+ARMNN_AUTO_TEST_FIXTURE_WITH_THF(Tile3dTestUint8, ClContextControlFixture, Tile3dTest<DataType::QAsymmU8>)
+ARMNN_AUTO_TEST_FIXTURE_WITH_THF(Tile4dTestUint8, ClContextControlFixture, Tile4dTest<DataType::QAsymmU8>)
+
 // Transpose
 ARMNN_AUTO_TEST_FIXTURE_WITH_THF(
     SimpleTransposeFloat32, ClContextControlFixture, SimpleTransposeTest<DataType::Float32>)
diff --git a/src/backends/cl/workloads/CMakeLists.txt b/src/backends/cl/workloads/CMakeLists.txt
index 030d71988f..cb16ab19e8 100644
--- a/src/backends/cl/workloads/CMakeLists.txt
+++ b/src/backends/cl/workloads/CMakeLists.txt
@@ -131,6 +131,8 @@ list(APPEND armnnClBackendWorkloads_sources
     ClStridedSliceWorkload.hpp
     ClSubtractionWorkload.cpp
     ClSubtractionWorkload.hpp
+    ClTileWorkload.cpp
+    ClTileWorkload.hpp
     ClTransposeConvolution2dWorkload.cpp
     ClTransposeConvolution2dWorkload.hpp
     ClTransposeWorkload.cpp
diff --git a/src/backends/cl/workloads/ClTileWorkload.cpp b/src/backends/cl/workloads/ClTileWorkload.cpp
new file mode 100644
index 0000000000..2c2f63faac
--- /dev/null
+++ b/src/backends/cl/workloads/ClTileWorkload.cpp
@@ -0,0 +1,50 @@
+//
+// Copyright © 2023 Arm Ltd and Contributors. All rights reserved.
+// SPDX-License-Identifier: MIT
+//
+
+#include "ClTileWorkload.hpp"
+#include "ClWorkloadUtils.hpp"
+#include <aclCommon/ArmComputeUtils.hpp>
+#include <cl/ClTensorHandle.hpp>
+#include <vector>
+#include <algorithm>
+
+using namespace armnn::armcomputetensorutils;
+namespace armnn
+{
+arm_compute::Status ClTileWorkloadValidate(const TensorInfo& input,
+                                           const TensorInfo& output,
+                                           const TileDescriptor& descriptor)
+{
+    const arm_compute::TensorInfo aclInput = BuildArmComputeTensorInfo(input);
+    const arm_compute::TensorInfo aclOutput = BuildArmComputeTensorInfo(output);
+
+    std::vector<uint32_t> aclMultiples = descriptor.m_Multiples;
+    std::reverse(aclMultiples.begin(),aclMultiples.end());
+
+    return arm_compute::CLTile::validate(&aclInput, &aclOutput, aclMultiples);
+}
+
+ClTileWorkload::ClTileWorkload(const armnn::TileQueueDescriptor& descriptor,
+                               const armnn::WorkloadInfo& info,
+                               const arm_compute::CLCompileContext& clCompileContext)
+    : BaseWorkload<TileQueueDescriptor>(descriptor, info)
+{
+    m_Data.ValidateInputsOutputs("ClTileWorkload", 1, 1);
+
+    std::vector<uint32_t> aclMultiples = descriptor.m_Parameters.m_Multiples;
+    std::reverse(aclMultiples.begin(),aclMultiples.end());
+
+    arm_compute::ICLTensor& input = static_cast<IClTensorHandle*>(m_Data.m_Inputs[0])->GetTensor();
+    arm_compute::ICLTensor& output = static_cast<IClTensorHandle*>(m_Data.m_Outputs[0])->GetTensor();
+    m_Layer.configure(clCompileContext, &input, &output, aclMultiples);
+}
+
+void ClTileWorkload::Execute() const
+{
+    ARMNN_SCOPED_PROFILING_EVENT_CL_GUID("ClTileWorkload_Execute", this->GetGuid());
+    m_Layer.run();
+}
+
+} //namespace armnn
\ No newline at end of file
diff --git a/src/backends/cl/workloads/ClTileWorkload.hpp b/src/backends/cl/workloads/ClTileWorkload.hpp
new file mode 100644
index 0000000000..0f68559f85
--- /dev/null
+++ b/src/backends/cl/workloads/ClTileWorkload.hpp
@@ -0,0 +1,27 @@
+//
+// Copyright © 2023 Arm Ltd and Contributors. All rights reserved.
+// SPDX-License-Identifier: MIT
+//
+
+#pragma once
+#include "ClBaseWorkload.hpp"
+#include <arm_compute/runtime/CL/functions/CLTile.h>
+
+namespace armnn
+{
+arm_compute::Status ClTileWorkloadValidate(const TensorInfo& input,
+                                           const TensorInfo& output,
+                                           const TileDescriptor& descriptor);
+
+class ClTileWorkload : public BaseWorkload<TileQueueDescriptor> {
+public:
+    ClTileWorkload(const TileQueueDescriptor &descriptor,
+                   const WorkloadInfo &info,
+                   const arm_compute::CLCompileContext& clCompileContext);
+    void Execute() const override;
+
+private:
+    mutable arm_compute::CLTile m_Layer;
+};
+
+} //namespace armnn
\ No newline at end of file
diff --git a/src/backends/cl/workloads/ClWorkloads.hpp b/src/backends/cl/workloads/ClWorkloads.hpp
index d862aab949..cec8706911 100644
--- a/src/backends/cl/workloads/ClWorkloads.hpp
+++ b/src/backends/cl/workloads/ClWorkloads.hpp
@@ -68,6 +68,7 @@
 #include "ClSubtractionWorkload.hpp"
 #include "ClConvertFp16ToFp32Workload.hpp"
 #include "ClConvertFp32ToFp16Workload.hpp"
+#include "ClTileWorkload.hpp"
 #include "ClTransposeConvolution2dWorkload.hpp"
 #include "ClTransposeWorkload.hpp"
 #include "ClUnidirectionalSequenceLstmFloatWorkload.hpp"