Add Multiplication operator (FP only) to Dynamic Fusion Interface

Note: we use a separate test fixture for Multiplication op instead of reusing ElementwiseBinaryFixture to avoid exposing the internal enum ElementwiseOp to the public utils/TypePrinters.h as required by the data test case macros to print the test data. We also do not consider modifying the enum ArithmeticOp in the standard interface to include MUL without an implementation. Future work should consider refactoring this test fixture into the ElementwiseBinaryFixture to reduce the total number of fixtures/code duplication.

Resolves: COMPMID-5779

Change-Id: I84207658ce0407095b028fca0ab7bfa2950255ec
Signed-off-by: Jakub Sujak <jakub.sujak@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/9013
Reviewed-by: Viet-Hoa Do <viet-hoa.do@arm.com>
Reviewed-by: SiCong Li <sicong.li@arm.com>
Reviewed-by: Gunes Bayir <gunes.bayir@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Benchmark: Arm Jenkins <bsgcomp@arm.com>

4 files changed, 578 insertions, 78 deletions

diff --git a/tests/validation/dynamic_fusion/gpu/cl/Add.cpp b/tests/validation/dynamic_fusion/gpu/cl/Add.cpp
index 0385407ad2..52ba0520ad 100644
--- a/tests/validation/dynamic_fusion/gpu/cl/Add.cpp
+++ b/tests/validation/dynamic_fusion/gpu/cl/Add.cpp
@@ -34,7 +34,6 @@
 #include "tests/datasets/DynamicFusionDataset.h"
 #include "tests/datasets/ShapeDatasets.h"
 #include "tests/validation/fixtures/dynamic_fusion/gpu/cl/ElementwiseBinaryFixture.h"
-#include "tests/validation/reference/ElementwiseOperations.h"
 
 namespace arm_compute
 {
@@ -42,6 +41,14 @@ namespace test
 {
 namespace validation
 {
+/* Synced with tests/validation/CL/ArithmeticAddition.cpp from the standard interface.
+ *
+ * Difference | Why the difference
+ * No quantized tests  | Not supported yet
+ * No in place tests   | Not supported yet
+ * No activation tests | Not needed in dynamic fusion interface
+ *
+ */
 TEST_SUITE(CL)
 TEST_SUITE(DYNAMIC_FUSION)
 TEST_SUITE(ADD)
@@ -49,29 +56,33 @@ TEST_SUITE(ADD)
 // *INDENT-OFF*
 // clang-format off
 DATA_TEST_CASE(Validate, framework::DatasetMode::ALL, zip(zip(
-               framework::dataset::make("Input1Info", { TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::F32),
+               framework::dataset::make("LhsInfo", { TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::F32),
                                                         TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::F32),    // Invalid data type combination
                                                         TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::S16),    // S16 is valid data type for Add
                                                         TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::S32),    // S32 is valid data type for Add
                                                         TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::F32),    // Mismatching shapes
                                                         TensorInfo(TensorShape(32U,  1U, 1U), 1, DataType::F32),    // Broadcasting allowed for lhs
+                                                        TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::QASYMM8),    // Unsupported data type QASYMM8
+                                                        TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::QASYMM8_SIGNED),    // Unsupported data type QASYMM8
                                                         TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::F32),
                                                         TensorInfo(TensorShape(15U, 23U, 3U), 1, DataType::F32),    // Broadcast Y dimension is not allowed
                                                         TensorInfo(TensorShape( 3U,  8U, 9U), 1, DataType::S16),    // Broadcast Z dimension is not allowed
                                                         TensorInfo(TensorShape(32U, 13U, 2U, 2), 1, DataType::F32), // Batching is allowed
                                                       }),
-               framework::dataset::make("Input2Info",{ TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::F32),
+               framework::dataset::make("RhsInfo",{ TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::F32),
                                                        TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::F16),
                                                        TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::S16),
                                                        TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::S32),
                                                        TensorInfo(TensorShape(48U, 11U, 2U), 1, DataType::F32),
                                                        TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::F32),
+                                                       TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::QASYMM8),    // Unsupported data type QASYMM8
+                                                       TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::QASYMM8_SIGNED),    // Unsupported data type QASYMM8
                                                        TensorInfo(TensorShape(32U,  1U, 1U), 1, DataType::F32),    // Broadcasting allowed for rhs
                                                        TensorInfo(TensorShape(15U,  1U, 3U), 1, DataType::F32),
                                                        TensorInfo(TensorShape( 3U,  8U, 1U), 1, DataType::S16),
                                                        TensorInfo(TensorShape(32U, 13U, 2U, 2), 1, DataType::F32),
                                                       })),
-               framework::dataset::make("Expected", { true, false, true, true, false, true, true, false, false, true})),
+               framework::dataset::make("Expected", { true, false, true, true, false, true, false, false, true, false, false, true})),
                input1_info, input2_info, expected)
 {
     // Create a new workload sketch
@@ -79,7 +90,7 @@ DATA_TEST_CASE(Validate, framework::DatasetMode::ALL, zip(zip(
     auto              gpu_ctx        = GpuWorkloadContext{ &cl_compile_ctx };
     GpuWorkloadSketch sketch{ &gpu_ctx };
 
-    // Fuse Elementwise Add
+    // Validate Elementwise Add
     auto          lhs_info         = sketch.create_tensor_info(input1_info);
     auto          rhs_info         = sketch.create_tensor_info(input2_info);
 
@@ -89,59 +100,73 @@ DATA_TEST_CASE(Validate, framework::DatasetMode::ALL, zip(zip(
 // clang-format on
 // *INDENT-ON*
 
-RelativeTolerance<float>            tolerance_f32(0.01f);                 /**< Tolerance value for comparing reference's output against implementation's output for DataType::F32 */
-RelativeTolerance<half_float::half> tolerance_f16(half_float::half(0.1)); /**< Tolerance value for comparing reference's output against implementation's output for DataType::F16 */
-constexpr float                     tolerance_num = 0.01f;                /**< Tolerance number */
+constexpr AbsoluteTolerance<float> tolerance_f16(0.0001f);  /**< Tolerance value for comparing reference's output against implementation's output for DataType::F16 */
+constexpr AbsoluteTolerance<float> tolerance_f32(0.0001f);  /**< Tolerance value for comparing reference's output against implementation's output for DataType::F32 */
+constexpr float                    tolerance_num = 0.0001f; /**< Tolerance number */
 
 template <typename T>
-using DynamicFusionAddOpFixture = DynamicFusionGpuElementwiseBinaryOneOpValidationFixture<CLTensor, CLAccessor, GpuAdd, T>;
+using DynamicFusionCLAddFixture = DynamicFusionGpuElementwiseBinaryOneOpValidationFixture<CLTensor, CLAccessor, GpuAdd, T>;
 
 template <typename T>
-using DynamicFusionAddOpBroadcastFixture = DynamicFusionGpuElementwiseBinaryBroadcastOneOpValidationFixture<CLTensor, CLAccessor, GpuAdd, T>;
+using DynamicFusionCLAddBroadcastFixture = DynamicFusionGpuElementwiseBinaryBroadcastOneOpValidationFixture<CLTensor, CLAccessor, GpuAdd, T>;
 
 template <typename T>
-using DynamicFusionGpuFuseTwoAddOpsFixture = DynamicFusionGpuElementwiseBinaryTwoOpsValidationFixture<CLTensor, CLAccessor, GpuAdd, T>;
+using DynamicFusionCLAddTwoOpsFixture = DynamicFusionGpuElementwiseBinaryTwoOpsValidationFixture<CLTensor, CLAccessor, GpuAdd, T>;
 
 TEST_SUITE(FP32)
-FIXTURE_DATA_TEST_CASE(RunSmallOneOp, DynamicFusionAddOpFixture<float>, framework::DatasetMode::PRECOMMIT, combine(combine(combine(
-                                                                                                                       framework::dataset::make("ElementwiseOp", { ArithmeticOperation::ADD }),
-                                                                                                                       datasets::SmallShapesNoBatches()),
-                                                                                                                   framework::dataset::make("DataType", { DataType::F32 })),
-                                                                                                                   framework::dataset::make("InPlace", { false, true })))
+FIXTURE_DATA_TEST_CASE(RunSmallOneOp,
+                       DynamicFusionCLAddFixture<float>,
+                       framework::DatasetMode::PRECOMMIT,
+                       combine(combine(combine(framework::dataset::make("ElementwiseOp", { ArithmeticOperation::ADD }),
+                                               datasets::SmallShapes()),
+                                       framework::dataset::make("DataType", { DataType::F32 })),
+                               framework::dataset::make("InPlace", { false })))
 {
     // Validate output
     validate(CLAccessor(_target), _reference, tolerance_f32);
 }
-FIXTURE_DATA_TEST_CASE(RunLargeOneOp, DynamicFusionAddOpFixture<float>, framework::DatasetMode::NIGHTLY, combine(combine(combine(
-                                                                                                                     framework::dataset::make("ElementwiseOp", { ArithmeticOperation::ADD }),
-                                                                                                                     datasets::LargeShapesNoBatches()),
-                                                                                                                 framework::dataset::make("DataType", { DataType::F32 })),
-                                                                                                                 framework::dataset::make("InPlace", { false, true })))
+FIXTURE_DATA_TEST_CASE(RunLargeOneOp,
+                       DynamicFusionCLAddFixture<float>,
+                       framework::DatasetMode::NIGHTLY,
+                       combine(combine(combine(framework::dataset::make("ElementwiseOp", { ArithmeticOperation::ADD }),
+                                               datasets::LargeShapes()),
+                                       framework::dataset::make("DataType", { DataType::F32 })),
+                               framework::dataset::make("InPlace", { false })))
 {
     // Validate output
     validate(CLAccessor(_target), _reference, tolerance_f32);
 }
-FIXTURE_DATA_TEST_CASE(RunSmallBroadcastOneOp, DynamicFusionAddOpBroadcastFixture<float>, framework::DatasetMode::PRECOMMIT, combine(combine(combine(framework::dataset::make("ElementwiseOp", { ArithmeticOperation::ADD }),
-                       datasets::TemporaryLimitedSmallShapesBroadcast()),
-                       framework::dataset::make("DataType", { DataType::F32 })),
-                       framework::dataset::make("InPlace", { false, true })))
+FIXTURE_DATA_TEST_CASE(RunSmallBroadcastOneOp,
+                       DynamicFusionCLAddBroadcastFixture<float>,
+                       framework::DatasetMode::PRECOMMIT,
+                       combine(combine(combine(framework::dataset::make("ElementwiseOp", { ArithmeticOperation::ADD }),
+                                               datasets::TemporaryLimitedSmallShapesBroadcast()),
+                                       framework::dataset::make("DataType", { DataType::F32 })),
+                               framework::dataset::make("InPlace", { false })))
 {
     // Validate output
     validate(CLAccessor(_target), _reference, tolerance_f32);
 }
 
-FIXTURE_DATA_TEST_CASE(RunLargeBroadcastOneOp, DynamicFusionAddOpBroadcastFixture<float>, framework::DatasetMode::PRECOMMIT, combine(combine(combine(framework::dataset::make("ElementwiseOp", { ArithmeticOperation::ADD }),
-                       datasets::TemporaryLimitedLargeShapesBroadcast()),
-                       framework::dataset::make("DataType", { DataType::F32 })),
-                       framework::dataset::make("InPlace", { false, true })))
+FIXTURE_DATA_TEST_CASE(RunLargeBroadcastOneOp,
+                       DynamicFusionCLAddBroadcastFixture<float>,
+                       framework::DatasetMode::NIGHTLY,
+                       combine(combine(combine(framework::dataset::make("ElementwiseOp", { ArithmeticOperation::ADD }),
+                                               datasets::TemporaryLimitedLargeShapesBroadcast()),
+                                       framework::dataset::make("DataType", { DataType::F32 })),
+                               framework::dataset::make("InPlace", { false })))
 {
     // Validate output
     validate(CLAccessor(_target), _reference, tolerance_f32);
 }
-FIXTURE_DATA_TEST_CASE(RunSmallTwoOps, DynamicFusionGpuFuseTwoAddOpsFixture<float>, framework::DatasetMode::PRECOMMIT, combine(combine(combine(framework::dataset::make("ElementwiseOp", { ArithmeticOperation::ADD }),
-                       datasets::DynamicFusionElementwiseBinaryTwoOpsSmallShapes()),
-                       framework::dataset::make("DataType", { DataType::F32 })),
-                       framework::dataset::make("InPlace", { false })))
+FIXTURE_DATA_TEST_CASE(RunSmallTwoOps,
+                       DynamicFusionCLAddTwoOpsFixture<float>,
+                       framework::DatasetMode::PRECOMMIT,
+                       combine(combine(combine(combine(framework::dataset::make("ElementwiseOp", { ArithmeticOperation::ADD }),
+                                                       datasets::DynamicFusionElementwiseBinaryTwoOpsSmallShapes()),
+                                               framework::dataset::make("DataType", { DataType::F32 })),
+                                       framework::dataset::make("InPlace", { false })),
+                               framework::dataset::make("FuseTwoOps", { true })))
 {
     // Validate output
     validate(CLAccessor(_target), _reference, tolerance_f32);
@@ -149,19 +174,25 @@ FIXTURE_DATA_TEST_CASE(RunSmallTwoOps, DynamicFusionGpuFuseTwoAddOpsFixture<floa
 TEST_SUITE_END() // FP32
 
 TEST_SUITE(FP16)
-FIXTURE_DATA_TEST_CASE(RunSmallOneOp, DynamicFusionAddOpFixture<half>, framework::DatasetMode::ALL, combine(combine(combine(framework::dataset::make("ElementwiseOp", { ArithmeticOperation::ADD }),
-                                                                                                                    datasets::SmallShapesNoBatches()),
-                                                                                                                    framework::dataset::make("DataType", { DataType::F16 })),
-                                                                                                            framework::dataset::make("InPlace", { false, true })))
+FIXTURE_DATA_TEST_CASE(RunSmallOneOp,
+                       DynamicFusionCLAddFixture<half>,
+                       framework::DatasetMode::ALL,
+                       combine(combine(combine(framework::dataset::make("ElementwiseOp", { ArithmeticOperation::ADD }),
+                                               datasets::SmallShapes()),
+                                       framework::dataset::make("DataType", { DataType::F16 })),
+                               framework::dataset::make("InPlace", { false })))
 {
     // Validate output
     validate(CLAccessor(_target), _reference, tolerance_f32, tolerance_num);
 }
 
-FIXTURE_DATA_TEST_CASE(RunSmallBroadcastOneOp, DynamicFusionAddOpBroadcastFixture<half>, framework::DatasetMode::ALL, combine(combine(combine(framework::dataset::make("ElementwiseOp", { ArithmeticOperation::ADD }),
-                       datasets::TemporaryLimitedSmallShapesBroadcast()),
-                       framework::dataset::make("DataType", { DataType::F16 })),
-                       framework::dataset::make("InPlace", { false })))
+FIXTURE_DATA_TEST_CASE(RunSmallBroadcastOneOp,
+                       DynamicFusionCLAddBroadcastFixture<half>,
+                       framework::DatasetMode::ALL,
+                       combine(combine(combine(framework::dataset::make("ElementwiseOp", { ArithmeticOperation::ADD }),
+                                               datasets::TemporaryLimitedSmallShapesBroadcast()),
+                                       framework::dataset::make("DataType", { DataType::F16 })),
+                               framework::dataset::make("InPlace", { false })))
 {
     // Validate output
     validate(CLAccessor(_target), _reference, tolerance_f32, tolerance_num);
@@ -170,10 +201,13 @@ FIXTURE_DATA_TEST_CASE(RunSmallBroadcastOneOp, DynamicFusionAddOpBroadcastFixtur
 TEST_SUITE_END() // FP16
 
 TEST_SUITE(S32)
-FIXTURE_DATA_TEST_CASE(RunSmall, DynamicFusionAddOpFixture<int32_t>, framework::DatasetMode::PRECOMMIT, combine(combine(combine(framework::dataset::make("ElementwiseOp", { ArithmeticOperation::ADD }),
-                                                                                                                        datasets::SmallShapesNoBatches()),
-                                                                                                                        framework::dataset::make("DataType", { DataType::S32 })),
-                                                                                                                framework::dataset::make("InPlace", { false })))
+FIXTURE_DATA_TEST_CASE(RunSmall,
+                       DynamicFusionCLAddFixture<int32_t>,
+                       framework::DatasetMode::PRECOMMIT,
+                       combine(combine(combine(framework::dataset::make("ElementwiseOp", { ArithmeticOperation::ADD }),
+                                               datasets::SmallShapes()),
+                                       framework::dataset::make("DataType", { DataType::S32 })),
+                               framework::dataset::make("InPlace", { false })))
 {
     // Validate output
     validate(CLAccessor(_target), _reference);
@@ -181,18 +215,24 @@ FIXTURE_DATA_TEST_CASE(RunSmall, DynamicFusionAddOpFixture<int32_t>, framework::
 TEST_SUITE_END() // S32
 
 TEST_SUITE(S16)
-FIXTURE_DATA_TEST_CASE(RunSmall, DynamicFusionAddOpFixture<int16_t>, framework::DatasetMode::PRECOMMIT, combine(combine(combine(framework::dataset::make("ElementwiseOp", { ArithmeticOperation::ADD }),
-                                                                                                                        datasets::SmallShapesNoBatches()),
-                                                                                                                        framework::dataset::make("DataType", { DataType::S16 })),
-                                                                                                                framework::dataset::make("InPlace", { false })))
+FIXTURE_DATA_TEST_CASE(RunSmall,
+                       DynamicFusionCLAddFixture<int16_t>,
+                       framework::DatasetMode::PRECOMMIT,
+                       combine(combine(combine(framework::dataset::make("ElementwiseOp", { ArithmeticOperation::ADD }),
+                                               datasets::SmallShapes()),
+                                       framework::dataset::make("DataType", { DataType::S16 })),
+                               framework::dataset::make("InPlace", { false })))
 {
     // Validate output
     validate(CLAccessor(_target), _reference);
 }
-FIXTURE_DATA_TEST_CASE(RunLarge, DynamicFusionAddOpFixture<int16_t>, framework::DatasetMode::PRECOMMIT, combine(combine(combine(framework::dataset::make("ElementwiseOp", { ArithmeticOperation::ADD }),
-                                                                                                                        datasets::LargeShapesNoBatches()),
-                                                                                                                        framework::dataset::make("DataType", { DataType::S16 })),
-                                                                                                                framework::dataset::make("InPlace", { false })))
+FIXTURE_DATA_TEST_CASE(RunLarge,
+                       DynamicFusionCLAddFixture<int16_t>,
+                       framework::DatasetMode::NIGHTLY,
+                       combine(combine(combine(framework::dataset::make("ElementwiseOp", { ArithmeticOperation::ADD }),
+                                               datasets::LargeShapes()),
+                                       framework::dataset::make("DataType", { DataType::S16 })),
+                               framework::dataset::make("InPlace", { false })))
 {
     // Validate output
     validate(CLAccessor(_target), _reference);
@@ -200,10 +240,13 @@ FIXTURE_DATA_TEST_CASE(RunLarge, DynamicFusionAddOpFixture<int16_t>, framework::
 TEST_SUITE_END() // S16
 
 TEST_SUITE(U8)
-FIXTURE_DATA_TEST_CASE(RunSmall, DynamicFusionAddOpFixture<uint8_t>, framework::DatasetMode::PRECOMMIT, combine(combine(combine(framework::dataset::make("ElementwiseOp", { ArithmeticOperation::ADD }),
-                                                                                                                        datasets::SmallShapesNoBatches()),
-                                                                                                                        framework::dataset::make("DataType", { DataType::U8 })),
-                                                                                                                framework::dataset::make("InPlace", { false })))
+FIXTURE_DATA_TEST_CASE(RunSmall,
+                       DynamicFusionCLAddFixture<uint8_t>,
+                       framework::DatasetMode::PRECOMMIT,
+                       combine(combine(combine(framework::dataset::make("ElementwiseOp", { ArithmeticOperation::ADD }),
+                                               datasets::SmallShapes()),
+                                       framework::dataset::make("DataType", { DataType::U8 })),
+                               framework::dataset::make("InPlace", { false })))
 {
     // Validate output
     validate(CLAccessor(_target), _reference);
diff --git a/tests/validation/dynamic_fusion/gpu/cl/Mul.cpp b/tests/validation/dynamic_fusion/gpu/cl/Mul.cpp
new file mode 100644
index 0000000000..a9e8f9c15f
--- /dev/null
+++ b/tests/validation/dynamic_fusion/gpu/cl/Mul.cpp
@@ -0,0 +1,223 @@
+/*
+ * 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.
+ */
+
+#include "arm_compute/dynamic_fusion/sketch/gpu/GpuWorkloadSketch.h"
+#include "arm_compute/dynamic_fusion/sketch/gpu/operators/GpuMul.h"
+
+#include "tests/CL/CLAccessor.h"
+#include "tests/framework/Fixture.h"
+#include "tests/framework/Macros.h"
+#include "tests/framework/datasets/Datasets.h"
+#include "tests/validation/Validation.h"
+
+#include "tests/datasets/DynamicFusionDataset.h"
+#include "tests/datasets/ShapeDatasets.h"
+#include "tests/validation/fixtures/dynamic_fusion/operators/MulFixture.h"
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+/* Synced with tests/validation/CL/PixelwiseMultiplication.cpp from the standard interface.
+ *
+ * Difference              | Why the difference
+ * No integer tests        | Not supported yet
+ * No quantized tests      | Not supported yet
+ * No convert policy tests | Not needed as convert policy is ignored by floating types
+ * No scale tests          | Not supported yet
+ * No rounding modes tests | Not supported yet
+ * No in place tests       | Not supported yet
+ * No activation tests     | Not needed in dynamic fusion interface
+ *
+ */
+namespace
+{
+constexpr AbsoluteTolerance<float> tolerance_f16(0.0001f); /**< Tolerance value for comparing reference's output against implementation's output for DataType::F16 */
+constexpr AbsoluteTolerance<float> tolerance_f32(0.0001f); /**< Tolerance value for comparing reference's output against implementation's output for DataType::F32 */
+} // namespace
+TEST_SUITE(CL)
+TEST_SUITE(DYNAMIC_FUSION)
+TEST_SUITE(MUL)
+
+// *INDENT-OFF*
+// clang-format off
+DATA_TEST_CASE(Validate, framework::DatasetMode::ALL, zip(zip(
+               framework::dataset::make("LhsInfo", { TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::F32),
+                                                        TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::F16),
+                                                        TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::F32),    // Invalid data type combination
+                                                        TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::U8),     // Unsupported data type U8
+                                                        TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::S8),     // Unsupported data type S8
+                                                        TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::S16),    // Unsupported data type S16
+                                                        TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::S32),    // Unsupported data type S32
+                                                        TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::QASYMM8),    // Unsupported data type QASYMM8
+                                                        TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::QASYMM8_SIGNED),    // Unsupported data type QASYMM8_SIGNED
+                                                        TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::F32),    // Mismatching shapes
+                                                        TensorInfo(TensorShape(32U,  1U, 1U), 1, DataType::F32),    // Broadcasting allowed for lhs
+                                                        TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::F32),
+                                                        TensorInfo(TensorShape(15U, 23U, 3U), 1, DataType::F32),    // Broadcast Y dimension is not allowed
+                                                        TensorInfo(TensorShape( 3U,  8U, 9U), 1, DataType::F32),    // Broadcast Z dimension is not allowed
+                                                        TensorInfo(TensorShape(32U, 13U, 2U, 2), 1, DataType::F32), // Batching is allowed
+                                                      }),
+               framework::dataset::make("RhsInfo",{ TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::F32),
+                                                       TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::F16),
+                                                       TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::F16),
+                                                       TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::U8),
+                                                       TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::S8),
+                                                       TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::S16),
+                                                       TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::S32),
+                                                       TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::QASYMM8),
+                                                       TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::QASYMM8_SIGNED),
+                                                       TensorInfo(TensorShape(48U, 11U, 2U), 1, DataType::F32),
+                                                       TensorInfo(TensorShape(32U, 13U, 2U), 1, DataType::F32),
+                                                       TensorInfo(TensorShape(32U,  1U, 1U), 1, DataType::F32),    // Broadcasting allowed for rhs
+                                                       TensorInfo(TensorShape(15U,  1U, 3U), 1, DataType::F32),
+                                                       TensorInfo(TensorShape( 3U,  8U, 1U), 1, DataType::F32),
+                                                       TensorInfo(TensorShape(32U, 13U, 2U, 2), 1, DataType::F32),
+                                                      })),
+               framework::dataset::make("Expected", { true, true, false, false, false, false, false, false, false, false, true, true, false, false, true })),
+               input1_info, input2_info, expected)
+{
+    // Create a new workload sketch
+    auto              cl_compile_ctx = CLKernelLibrary::get().get_compile_context();
+    auto              gpu_ctx        = GpuWorkloadContext{ &cl_compile_ctx };
+    GpuWorkloadSketch sketch{ &gpu_ctx };
+
+    // Validate Elementwise Mul
+    auto          lhs_info         = sketch.create_tensor_info(input1_info);
+    auto          rhs_info         = sketch.create_tensor_info(input2_info);
+
+    bool res = bool(GpuMul::validate_op(sketch, &lhs_info, &rhs_info));
+    ARM_COMPUTE_EXPECT(res == expected, framework::LogLevel::ERRORS);
+}
+// clang-format on
+// *INDENT-ON*
+
+template <typename T>
+using DynamicFusionCLMulFixture = DynamicFusionMulOneOpValidationFixture<CLTensor, CLAccessor, GpuMul, T>;
+template <typename T>
+using DynamicFusionCLMulBroadcastFixture = DynamicFusionMulBroadcastValidationFixture<CLTensor, CLAccessor, GpuMul, T>;
+template <typename T>
+using DynamicFusionCLMulTwoOpsFixture = DynamicFusionMulTwoOpsValidationFixture<CLTensor, CLAccessor, GpuMul, T>;
+
+TEST_SUITE(F16)
+FIXTURE_DATA_TEST_CASE(RunSmallOneOp,
+                       DynamicFusionCLMulFixture<half>,
+                       framework::DatasetMode::ALL,
+                       combine(combine(datasets::SmallShapes(),
+                                       framework::dataset::make("DataType", { DataType::F16 })),
+                               framework::dataset::make("InPlace", { false })))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference, tolerance_f16);
+}
+
+FIXTURE_DATA_TEST_CASE(RunSmallBroadcastOneOp,
+                       DynamicFusionCLMulBroadcastFixture<half>,
+                       framework::DatasetMode::PRECOMMIT,
+                       combine(combine(datasets::TemporaryLimitedSmallShapesBroadcast(),
+                                       framework::dataset::make("DataType", { DataType::F16 })),
+                               framework::dataset::make("InPlace", { false })))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference, tolerance_f16);
+}
+
+FIXTURE_DATA_TEST_CASE(RunLargeBroadcastOneOp,
+                       DynamicFusionCLMulBroadcastFixture<half>,
+                       framework::DatasetMode::NIGHTLY,
+                       combine(combine(datasets::TemporaryLimitedLargeShapesBroadcast(),
+                                       framework::dataset::make("DataType", { DataType::F16 })),
+                               framework::dataset::make("InPlace", { false })))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference, tolerance_f16);
+}
+TEST_SUITE_END() // F16
+
+TEST_SUITE(F32)
+FIXTURE_DATA_TEST_CASE(RunSmallOneOp,
+                       DynamicFusionCLMulFixture<float>,
+                       framework::DatasetMode::PRECOMMIT,
+                       combine(combine(datasets::SmallShapes(),
+                                       framework::dataset::make("DataType", { DataType::F32 })),
+                               framework::dataset::make("InPlace", { false })))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference, tolerance_f32);
+}
+
+FIXTURE_DATA_TEST_CASE(RunLargeOneOp,
+                       DynamicFusionCLMulFixture<float>,
+                       framework::DatasetMode::NIGHTLY,
+                       combine(combine(datasets::LargeShapes(),
+                                       framework::dataset::make("DataType", { DataType::F32 })),
+                               framework::dataset::make("InPlace", { false })))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference, tolerance_f32);
+}
+
+FIXTURE_DATA_TEST_CASE(RunSmallBroadcastOneOp,
+                       DynamicFusionCLMulBroadcastFixture<float>,
+                       framework::DatasetMode::PRECOMMIT,
+                       combine(combine(datasets::TemporaryLimitedSmallShapesBroadcast(),
+                                       framework::dataset::make("DataType", { DataType::F32 })),
+                               framework::dataset::make("InPlace", { false })))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference, tolerance_f32);
+}
+
+FIXTURE_DATA_TEST_CASE(RunLargeBroadcastOneOp,
+                       DynamicFusionCLMulBroadcastFixture<float>,
+                       framework::DatasetMode::NIGHTLY,
+                       combine(combine(datasets::TemporaryLimitedLargeShapesBroadcast(),
+                                       framework::dataset::make("DataType", { DataType::F32 })),
+                               framework::dataset::make("InPlace", { false })))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference, tolerance_f32);
+}
+
+FIXTURE_DATA_TEST_CASE(RunSmallTwoOps,
+                       DynamicFusionCLMulTwoOpsFixture<float>,
+                       framework::DatasetMode::PRECOMMIT,
+                       combine(combine(combine(datasets::DynamicFusionElementwiseBinaryTwoOpsSmallShapes(),
+                                               framework::dataset::make("DataType", { DataType::F32 })),
+                                       framework::dataset::make("InPlace", { false })),
+                               framework::dataset::make("FuseTwoOps", { true })))
+{
+    // Validate output
+    validate(CLAccessor(_target), _reference, tolerance_f32);
+}
+TEST_SUITE_END() // F32
+
+TEST_SUITE_END() // MUL
+TEST_SUITE_END() // DYNAMIC_FUSION
+TEST_SUITE_END() // CL
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
diff --git a/tests/validation/fixtures/dynamic_fusion/gpu/cl/ElementwiseBinaryFixture.h b/tests/validation/fixtures/dynamic_fusion/gpu/cl/ElementwiseBinaryFixture.h
index faed610874..b0680c0e4a 100644
--- a/tests/validation/fixtures/dynamic_fusion/gpu/cl/ElementwiseBinaryFixture.h
+++ b/tests/validation/fixtures/dynamic_fusion/gpu/cl/ElementwiseBinaryFixture.h
@@ -31,12 +31,9 @@
 #include "arm_compute/dynamic_fusion/sketch/gpu/GpuWorkloadSketch.h"
 #include "arm_compute/dynamic_fusion/sketch/gpu/operators/GpuOutput.h"
 
-#include "tests/CL/CLAccessor.h"
 #include "tests/framework/Fixture.h"
 #include "tests/framework/Macros.h"
-#include "tests/validation/Validation.h"
 #include "tests/validation/reference/ElementwiseOperations.h"
-#include "tests/validation/reference/Permute.h"
 
 using namespace arm_compute::experimental::dynamic_fusion;
 
@@ -51,12 +48,13 @@ class DynamicFusionGpuElementwiseBinaryValidationGenericFixture : public framewo
 {
 public:
     template <typename...>
-    void setup(ArithmeticOperation op, TensorShape shape0, TensorShape shape1, TensorShape shape2, const DataType data_type, const bool is_inplace)
+    void setup(ArithmeticOperation ref_op, const TensorShape &shape0, const TensorShape &shape1, const TensorShape &shape2, DataType data_type, bool is_inplace, bool fuse_two_ops = false)
     {
-        _op         = op;
+        _ref_op         = ref_op;
         _is_inplace = is_inplace;
         _data_type  = data_type;
-        _fuse       = shape2.total_size() != 0;
+        _fuse       = fuse_two_ops;
+        ARM_COMPUTE_ERROR_ON_MSG(_fuse && shape2.total_size() == 0, "No shape2 provided for fusion of two ops.");
         ARM_COMPUTE_ERROR_ON_MSG(_fuse && _is_inplace, "In place for fusing case not supported yet.");
         _target    = compute_target(shape0, shape1, shape2);
         _reference = compute_reference(shape0, shape1, shape2);
@@ -68,7 +66,7 @@ protected:
     {
         if(is_data_type_float(tensor.data_type()))
         {
-            switch(_op)
+            switch(_ref_op)
             {
                 case ArithmeticOperation::DIV:
                     library->fill_tensor_uniform_ranged(tensor, i, { std::pair<float, float>(-0.001f, 0.001f) });
@@ -82,7 +80,7 @@ protected:
         }
         else if(tensor.data_type() == DataType::S32)
         {
-            switch(_op)
+            switch(_ref_op)
             {
                 case ArithmeticOperation::DIV:
                     library->fill_tensor_uniform_ranged(tensor, i, { std::pair<int32_t, int32_t>(-1U, 1U) });
@@ -97,7 +95,7 @@ protected:
         }
     }
 
-    TensorType compute_target(TensorShape shape0, TensorShape shape1, TensorShape shape2)
+    TensorType compute_target(const TensorShape &shape0, const TensorShape &shape1, const TensorShape &shape2)
     {
         // Create a new workload sketch
         auto              cl_compile_ctx = CLKernelLibrary::get().get_compile_context();
@@ -105,7 +103,7 @@ protected:
         GpuWorkloadSketch sketch{ &gpu_ctx };
 
         // Fuse first element wise binary Op
-        TensorInfo lhs_info = sketch.create_tensor_info(shape0, 1, _data_type);
+        TensorInfo lhs_info = sketch.create_tensor_info(TensorInfo(shape0, 1, _data_type));
         TensorInfo rhs_info = sketch.create_tensor_info(TensorInfo(shape1, 1, _data_type));
         TensorInfo dst_info = sketch.create_tensor_info();
 
@@ -115,7 +113,7 @@ protected:
 
         if(_fuse)
         {
-            rhs_info_fuse          = sketch.create_tensor_info(shape2, 1, _data_type);
+            rhs_info_fuse          = sketch.create_tensor_info(TensorInfo(shape2, 1, _data_type));
             ITensorInfo *ans2_info = FunctionType::create_op(sketch, ans_info, &rhs_info_fuse);
             GpuOutput::create_op(sketch, ans2_info, &dst_info);
         }
@@ -183,7 +181,7 @@ protected:
         return t_dst;
     }
 
-    SimpleTensor<T> compute_reference(TensorShape shape0, TensorShape shape1, TensorShape shape2)
+    SimpleTensor<T> compute_reference(const TensorShape &shape0, const TensorShape &shape1, const TensorShape &shape2)
     {
         const TensorShape out_shape      = TensorShape::broadcast_shape(shape0, shape1);
         const TensorShape out_shape_fuse = TensorShape::broadcast_shape(out_shape, shape1);
@@ -194,21 +192,22 @@ protected:
         SimpleTensor<T> ref_rhs_fuse{ shape2, _data_type, 1, QuantizationInfo() };
         SimpleTensor<T> ref_dst{ out_shape, _data_type, 1, QuantizationInfo() };
         SimpleTensor<T> ref_dst_fuse{ out_shape_fuse, _data_type, 1, QuantizationInfo() };
+
         // Fill reference
         fill(ref_lhs, 0);
         fill(ref_rhs, 1);
 
-        reference::arithmetic_operation<T>(_op, ref_lhs, ref_rhs, ref_dst, ConvertPolicy::WRAP);
+        reference::arithmetic_operation<T>(_ref_op, ref_lhs, ref_rhs, ref_dst, ConvertPolicy::WRAP);
         if(_fuse)
         {
             fill(ref_rhs_fuse, 2);
-            reference::arithmetic_operation<T>(_op, ref_dst, ref_rhs_fuse, ref_dst_fuse, ConvertPolicy::WRAP);
+            reference::arithmetic_operation<T>(_ref_op, ref_dst, ref_rhs_fuse, ref_dst_fuse, ConvertPolicy::WRAP);
         }
         SimpleTensor<T> *ret = _fuse ? &ref_dst_fuse : &ref_dst;
         return *ret;
     }
 
-    ArithmeticOperation _op{ ArithmeticOperation::ADD };
+    ArithmeticOperation _ref_op{ ArithmeticOperation::ADD };
     TensorType          _target{};
     SimpleTensor<T>     _reference{};
     DataType            _data_type{};
@@ -222,9 +221,9 @@ class DynamicFusionGpuElementwiseBinaryOneOpValidationFixture : public DynamicFu
 {
 public:
     template <typename...>
-    void setup(ArithmeticOperation op, TensorShape shape, const DataType data_type, const bool is_inplace)
+    void setup(ArithmeticOperation ref_op, const TensorShape &shape0, DataType data_type, bool is_inplace)
     {
-        DynamicFusionGpuElementwiseBinaryValidationGenericFixture<TensorType, AccessorType, FunctionType, T>::setup(op, shape, shape, TensorShape(), data_type, is_inplace);
+        DynamicFusionGpuElementwiseBinaryValidationGenericFixture<TensorType, AccessorType, FunctionType, T>::setup(ref_op, shape0, shape0, TensorShape(), data_type, is_inplace);
     }
 };
 
@@ -233,9 +232,9 @@ class DynamicFusionGpuElementwiseBinaryBroadcastOneOpValidationFixture : public
 {
 public:
     template <typename...>
-    void setup(ArithmeticOperation op, TensorShape shape0, TensorShape shape1, const DataType data_type, const bool is_inplace)
+    void setup(ArithmeticOperation ref_op, const TensorShape &shape0, const TensorShape &shape1, DataType data_type, bool is_inplace)
     {
-        DynamicFusionGpuElementwiseBinaryValidationGenericFixture<TensorType, AccessorType, FunctionType, T>::setup(op, shape0, shape1, TensorShape(), data_type, is_inplace);
+        DynamicFusionGpuElementwiseBinaryValidationGenericFixture<TensorType, AccessorType, FunctionType, T>::setup(ref_op, shape0, shape1, TensorShape(), data_type, is_inplace);
     }
 };
 
@@ -244,9 +243,9 @@ class DynamicFusionGpuElementwiseBinaryTwoOpsValidationFixture : public DynamicF
 {
 public:
     template <typename...>
-    void setup(ArithmeticOperation op, TensorShape shape0, TensorShape shape1, TensorShape shape2, const DataType data_type, const bool is_inplace)
+    void setup(ArithmeticOperation ref_op, const TensorShape &shape0, const TensorShape &shape1, const TensorShape &shape2, DataType data_type, bool is_inplace, bool fuse_two_ops)
     {
-        DynamicFusionGpuElementwiseBinaryValidationGenericFixture<TensorType, AccessorType, FunctionType, T>::setup(op, shape0, shape1, shape2, data_type, is_inplace);
+        DynamicFusionGpuElementwiseBinaryValidationGenericFixture<TensorType, AccessorType, FunctionType, T>::setup(ref_op, shape0, shape1, shape2, data_type, is_inplace, fuse_two_ops);
     }
 };
 
diff --git a/tests/validation/fixtures/dynamic_fusion/operators/MulFixture.h b/tests/validation/fixtures/dynamic_fusion/operators/MulFixture.h
new file mode 100644
index 0000000000..0530707c38
--- /dev/null
+++ b/tests/validation/fixtures/dynamic_fusion/operators/MulFixture.h
@@ -0,0 +1,235 @@
+/*
+* 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 TESTS_VALIDATION_FIXTURES_DYNAMIC_FUSION_OPERATORS_MULFIXTURE
+#define TESTS_VALIDATION_FIXTURES_DYNAMIC_FUSION_OPERATORS_MULFIXTURE
+
+#include "arm_compute/core/CL/CLKernelLibrary.h"
+#include "arm_compute/core/TensorInfo.h"
+#include "arm_compute/core/Types.h"
+#include "arm_compute/dynamic_fusion/runtime/gpu/cl/ClWorkloadRuntime.h"
+#include "arm_compute/dynamic_fusion/sketch/gpu/GpuWorkloadSketch.h"
+#include "arm_compute/dynamic_fusion/sketch/gpu/operators/GpuOutput.h"
+
+#include "tests/Globals.h"
+#include "tests/framework/Fixture.h"
+#include "tests/framework/Macros.h"
+#include "tests/validation/reference/PixelWiseMultiplication.h"
+
+using namespace arm_compute::experimental::dynamic_fusion;
+
+namespace arm_compute
+{
+namespace test
+{
+namespace validation
+{
+/* We use a separate test fixture for Multiplication op instead of reusing ElementwiseBinaryFixture to avoid exposing
+ * the internal enum ElementwiseOp to the public utils/TypePrinters.h as required by the data test case macros
+ * to print the test data.
+ */
+template <typename TensorType, typename AccessorType, typename FunctionType, typename T>
+class DynamicFusionMulValidationFixture : public framework::Fixture
+{
+public:
+   template <typename...>
+   void setup(const TensorShape &shape0, const TensorShape &shape1, const TensorShape &shape2, DataType data_type, bool is_inplace, bool fuse_two_ops = false)
+   {
+       _data_type  = data_type;
+       _is_inplace = is_inplace;
+       _fuse       = fuse_two_ops;
+       ARM_COMPUTE_ERROR_ON_MSG(_fuse && shape2.total_size() == 0, "No shape2 provided for fusion of two ops.");
+       ARM_COMPUTE_ERROR_ON_MSG(_fuse && _is_inplace, "In place for fusing case not supported yet.");
+       _target    = compute_target(shape0, shape1, shape2);
+       _reference = compute_reference(shape0, shape1, shape2);
+   }
+
+protected:
+   template <typename U>
+   void fill(U &&tensor, int i)
+   {
+       library->fill_tensor_uniform(tensor, i);
+   }
+
+   TensorType compute_target(const TensorShape &shape0, const TensorShape &shape1, const TensorShape &shape2)
+   {
+       // Create a new workload sketch
+       auto              cl_compile_ctx = CLKernelLibrary::get().get_compile_context();
+       auto              gpu_ctx        = GpuWorkloadContext{ &cl_compile_ctx };
+       GpuWorkloadSketch sketch{ &gpu_ctx };
+
+       // Fuse first multiplication op
+       TensorInfo lhs_info = sketch.create_tensor_info(TensorInfo(shape0, 1, _data_type));
+       TensorInfo rhs_info = sketch.create_tensor_info(TensorInfo(shape1, 1, _data_type));
+       TensorInfo dst_info = sketch.create_tensor_info();
+
+       TensorInfo rhs_info_fuse;
+
+       ITensorInfo *ans_info = FunctionType::create_op(sketch, &lhs_info, &rhs_info);
+
+       if(_fuse)
+       {
+           rhs_info_fuse          = sketch.create_tensor_info(TensorInfo(shape2, 1, _data_type));
+           ITensorInfo *ans2_info = FunctionType::create_op(sketch, ans_info, &rhs_info_fuse);
+           GpuOutput::create_op(sketch, ans2_info, &dst_info);
+       }
+       else
+       {
+           GpuOutput::create_op(sketch, ans_info, &dst_info);
+       }
+
+       // Configure runtime
+       ClWorkloadRuntime runtime;
+       runtime.configure(sketch);
+
+       // (Important) Allocate auxiliary tensor memory if there are any
+       for(auto &data : runtime.get_auxiliary_tensors())
+       {
+           CLTensor     *tensor      = std::get<0>(data);
+           TensorInfo    info        = std::get<1>(data);
+           AuxMemoryInfo aux_mem_req = std::get<2>(data);
+           tensor->allocator()->init(info, aux_mem_req.alignment);
+           tensor->allocator()->allocate(); // Use ACL allocated memory
+       }
+
+       // Construct user tensors
+       TensorType t_lhs{};
+       TensorType t_rhs{};
+       TensorType t_rhs_fuse{};
+       TensorType t_dst{};
+
+       // Initialize user tensors
+       t_lhs.allocator()->init(lhs_info);
+       t_rhs.allocator()->init(rhs_info);
+       t_dst.allocator()->init(dst_info);
+       if(_fuse)
+       {
+           t_rhs_fuse.allocator()->init(rhs_info_fuse);
+       }
+
+       // Allocate and fill user tensors
+       // Instead of using ACL allocator, the user can choose to import memory into the tensors
+       t_lhs.allocator()->allocate();
+       t_rhs.allocator()->allocate();
+       t_dst.allocator()->allocate();
+       if(_fuse)
+       {
+           t_rhs_fuse.allocator()->allocate();
+       }
+
+       fill(AccessorType(t_lhs), 0);
+       fill(AccessorType(t_rhs), 1);
+       if(_fuse)
+       {
+           fill(AccessorType(t_rhs_fuse), 2);
+       }
+
+       // Run runtime
+       if(_fuse)
+       {
+           runtime.run({ &t_lhs, &t_rhs, &t_rhs_fuse, &t_dst });
+       }
+       else
+       {
+           runtime.run({ &t_lhs, &t_rhs, &t_dst });
+       }
+
+       return t_dst;
+   }
+
+   SimpleTensor<T> compute_reference(const TensorShape &shape0, const TensorShape &shape1, const TensorShape &shape2)
+   {
+       // Create reference
+       SimpleTensor<T> ref_lhs{ shape0, _data_type, 1, QuantizationInfo() };
+       SimpleTensor<T> ref_rhs{ shape1, _data_type, 1, QuantizationInfo() };
+       SimpleTensor<T> ref_rhs_fuse{ shape2, _data_type, 1, QuantizationInfo() };
+
+       // Fill reference
+       fill(ref_lhs, 0);
+       fill(ref_rhs, 1);
+       SimpleTensor<T> ref_dst = reference::pixel_wise_multiplication<T, T, T>(ref_lhs,
+                                                                               ref_rhs,
+                                                                               1.f,
+                                                                               ConvertPolicy::SATURATE,
+                                                                               RoundingPolicy::TO_NEAREST_UP,
+                                                                               _data_type,
+                                                                               QuantizationInfo());
+       if(_fuse)
+       {
+           fill(ref_rhs_fuse, 2);
+           SimpleTensor<T> ref_dst_fuse = reference::pixel_wise_multiplication<T, T, T>(ref_dst,
+                                                                                        ref_rhs_fuse,
+                                                                                        1.f,
+                                                                                        ConvertPolicy::SATURATE,
+                                                                                        RoundingPolicy::TO_NEAREST_UP,
+                                                                                        _data_type,
+                                                                                        QuantizationInfo());
+           return ref_dst_fuse;
+       }
+       return ref_dst;
+   }
+
+   TensorType      _target{};
+   SimpleTensor<T> _reference{};
+   DataType        _data_type{};
+   bool            _is_inplace{ false };
+   bool            _fuse{ false };
+};
+
+template <typename TensorType, typename AccessorType, typename FunctionType, typename T>
+class DynamicFusionMulOneOpValidationFixture : public DynamicFusionMulValidationFixture<TensorType, AccessorType, FunctionType, T>
+{
+public:
+   template <typename...>
+   void setup(const TensorShape &shape0, DataType data_type, bool is_inplace)
+   {
+       DynamicFusionMulValidationFixture<TensorType, AccessorType, FunctionType, T>::setup(shape0, shape0, TensorShape(), data_type, is_inplace);
+   }
+};
+
+template <typename TensorType, typename AccessorType, typename FunctionType, typename T>
+class DynamicFusionMulBroadcastValidationFixture : public DynamicFusionMulValidationFixture<TensorType, AccessorType, FunctionType, T>
+{
+public:
+   template <typename...>
+   void setup(const TensorShape &shape0, const TensorShape &shape1, DataType data_type, bool is_inplace)
+   {
+       DynamicFusionMulValidationFixture<TensorType, AccessorType, FunctionType, T>::setup(shape0, shape1, TensorShape(), data_type, is_inplace);
+   }
+};
+
+template <typename TensorType, typename AccessorType, typename FunctionType, typename T>
+class DynamicFusionMulTwoOpsValidationFixture : public DynamicFusionMulValidationFixture<TensorType, AccessorType, FunctionType, T>
+{
+public:
+   template <typename...>
+   void setup(const TensorShape &shape0, const TensorShape &shape1, const TensorShape &shape2, DataType data_type, bool is_inplace, bool fuse_two_ops)
+   {
+       DynamicFusionMulValidationFixture<TensorType, AccessorType, FunctionType, T>::setup(shape0, shape1, shape2, data_type, is_inplace, fuse_two_ops);
+   }
+};
+
+} // namespace validation
+} // namespace test
+} // namespace arm_compute
+#endif /* TESTS_VALIDATION_FIXTURES_DYNAMIC_FUSION_OPERATORS_MULFIXTURE */