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>

15 files changed, 843 insertions, 154 deletions

diff --git a/Android.bp b/Android.bp
index 34f722f6fe..78aa64a5f9 100644
--- a/Android.bp
+++ b/Android.bp
@@ -619,6 +619,7 @@ cc_library_static {
         "src/dynamic_fusion/sketch/gpu/operators/GpuClamp.cpp",
         "src/dynamic_fusion/sketch/gpu/operators/GpuConv2d.cpp",
         "src/dynamic_fusion/sketch/gpu/operators/GpuDepthwiseConv2d.cpp",
+        "src/dynamic_fusion/sketch/gpu/operators/GpuMul.cpp",
         "src/dynamic_fusion/sketch/gpu/operators/GpuOutput.cpp",
         "src/dynamic_fusion/sketch/gpu/operators/GpuPool2d.cpp",
         "src/dynamic_fusion/sketch/gpu/operators/GpuReshape.cpp",
diff --git a/arm_compute/dynamic_fusion/sketch/gpu/operators/GpuAdd.h b/arm_compute/dynamic_fusion/sketch/gpu/operators/GpuAdd.h
index 6ac5d4e500..796fd6f83a 100644
--- a/arm_compute/dynamic_fusion/sketch/gpu/operators/GpuAdd.h
+++ b/arm_compute/dynamic_fusion/sketch/gpu/operators/GpuAdd.h
@@ -25,10 +25,12 @@
 #define ARM_COMPUTE_DYNAMIC_FUSION_SKETCH_GPU_OPERATORS_GPUADD
 
 #include "arm_compute/core/Error.h"
-#include "arm_compute/core/ITensorInfo.h"
 
 namespace arm_compute
 {
+/** Forward declaration */
+class ITensorInfo;
+
 namespace experimental
 {
 namespace dynamic_fusion
diff --git a/arm_compute/dynamic_fusion/sketch/gpu/operators/GpuMul.h b/arm_compute/dynamic_fusion/sketch/gpu/operators/GpuMul.h
new file mode 100644
index 0000000000..3e0ebdd96c
--- /dev/null
+++ b/arm_compute/dynamic_fusion/sketch/gpu/operators/GpuMul.h
@@ -0,0 +1,95 @@
+/*
+ * 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_DYNAMIC_FUSION_SKETCH_GPU_OPERATORS_GPUMUL
+#define ARM_COMPUTE_DYNAMIC_FUSION_SKETCH_GPU_OPERATORS_GPUMUL
+
+#include "arm_compute/core/Error.h"
+
+namespace arm_compute
+{
+/** Forward declaration */
+class ITensorInfo;
+
+namespace experimental
+{
+namespace dynamic_fusion
+{
+/** Forward declaration */
+class GpuWorkloadContext;
+class GpuWorkloadSketch;
+
+/** Operator interface. */
+class GpuMul final
+{
+public:
+    /** Create an operator and fuse it into the workload sketch.
+     *    @note If @ref validate_op() fails, the creation also fails and may throw an error.
+     *    @note If @ref validate_op() fails, @p sketch remains unchanged and valid.
+     *
+     * Valid data type configurations:
+     * |lhs            |rhs            |dst           |
+     * |:--------------|:--------------|:-------------|
+     * |F16            |F16            |F16           |
+     * |F32            |F32            |F32           |
+     *
+     * Valid data layouts:
+     * - Any
+     *
+     * @param[in,out] sketch Workload sketch into which the operator will be fused
+     * @param[in]     lhs    Left hand side tensor info. Data types supported: F16/F32.
+     * @param[in]     rhs    Right hand side tensor info. Data types supported: Same as @p lhs.
+     *
+     * @return Pointer for the destination tensor info
+     */
+    static ITensorInfo *create_op(GpuWorkloadSketch &sketch,
+                                  ITensorInfo       *lhs,
+                                  ITensorInfo       *rhs);
+
+    /** Check if the operator configuration is supported, irrespective of fusion
+     *
+     * @param[in] context Workload context within which the operator is running
+     * @param[in] lhs     Left hand side tensor info. Data types supported: F16/F32.
+     * @param[in] rhs     Right hand side tensor info. Data types supported: Same as @p lhs.
+     *
+     * @return Status
+     */
+    static Status is_supported_op(const GpuWorkloadContext &context,
+                                  const ITensorInfo        *lhs,
+                                  const ITensorInfo        *rhs);
+
+    /** Validate the operator and check if the configuration is supported and if it can be fused into the workload sketch.
+     *
+     * Parameters are similar to @ref GpuMul::create_op()
+     *
+     * @return Status
+     */
+    static Status validate_op(const GpuWorkloadSketch &sketch,
+                              const ITensorInfo       *rhs,
+                              const ITensorInfo       *lhs);
+};
+
+} // namespace dynamic_fusion
+} // namespace experimental
+} // namespace arm_compute
+#endif /* ARM_COMPUTE_DYNAMIC_FUSION_SKETCH_GPU_OPERATORS_GPUMUL */
diff --git a/filelist.json b/filelist.json
index 3cb3a7a76f..a66d2a3384 100644
--- a/filelist.json
+++ b/filelist.json
@@ -2234,13 +2234,14 @@
       "src/dynamic_fusion/sketch/gpu/operators/GpuCast.cpp",
       "src/dynamic_fusion/sketch/gpu/operators/GpuClamp.cpp",
       "src/dynamic_fusion/sketch/gpu/operators/GpuConv2d.cpp",
+      "src/dynamic_fusion/sketch/gpu/operators/GpuDepthwiseConv2d.cpp",
+      "src/dynamic_fusion/sketch/gpu/operators/GpuMul.cpp",
       "src/dynamic_fusion/sketch/gpu/operators/GpuReshape.cpp",
       "src/dynamic_fusion/sketch/gpu/operators/GpuPool2d.cpp",
       "src/dynamic_fusion/sketch/gpu/operators/GpuOutput.cpp",
       "src/dynamic_fusion/sketch/gpu/operators/GpuResize.cpp",
       "src/dynamic_fusion/sketch/gpu/operators/GpuSoftmax.cpp",
       "src/dynamic_fusion/sketch/gpu/operators/internal/GpuElementwiseBinaryCommon.cpp",
-      "src/dynamic_fusion/sketch/gpu/operators/GpuDepthwiseConv2d.cpp",
       "src/dynamic_fusion/sketch/gpu/template_writer/cl/ClTemplateActivation.cpp",
       "src/dynamic_fusion/sketch/gpu/template_writer/cl/ClTemplateCast.cpp",
       "src/dynamic_fusion/sketch/gpu/template_writer/cl/ClTemplateDepthwiseConv2d.cpp",
diff --git a/src/core/CL/cl_kernels/tile_helpers.h b/src/core/CL/cl_kernels/tile_helpers.h
index 507e172dfb..baee0d561b 100644
--- a/src/core/CL/cl_kernels/tile_helpers.h
+++ b/src/core/CL/cl_kernels/tile_helpers.h
@@ -1063,6 +1063,7 @@
 
 #define V_ADD(A_VAL, B_VAL) ((A_VAL) + (B_VAL))
 #define V_DIV(A_VAL, B_VAL) ((A_VAL) / (B_VAL))
+#define V_MUL(A_VAL, B_VAL) ((A_VAL) * (B_VAL))
 
 /** Element-wise activation for quantized types
  *
@@ -1130,6 +1131,9 @@
 
 #define T_ELTWISE_BROADCAST_DIV_X(DST_DATA_TYPE, M0, N0, lhs, rhs, dst) T_ELTWISE_BROADCAST_X(V_DIV, DST_DATA_TYPE, M0, N0, lhs, rhs, dst)
 
+#define T_ELTWISE_BROADCAST_LHS_X_MUL(DST_DATA_TYPE, M0, N0, lhs, rhs, dst) T_ELTWISE_BROADCAST_LHS_X(V_MUL, DST_DATA_TYPE, M0, N0, lhs, rhs, dst)
+#define T_ELTWISE_BROADCAST_RHS_X_MUL(DST_DATA_TYPE, M0, N0, lhs, rhs, dst) T_ELTWISE_BROADCAST_X(V_MUL, DST_DATA_TYPE, M0, N0, lhs, rhs, dst)
+
 /** Element-wise scale with a constant value
  *
  * @note Performs: LHS * constant = DST
@@ -1193,6 +1197,7 @@
 
 #define T_ELTWISE_ADD(DST_DATA_TYPE, M0, N0, lhs, rhs, dst) T_ELTWISE(V_ADD, DST_DATA_TYPE, M0, N0, lhs, rhs, dst)
 #define T_ELTWISE_DIV(DST_DATA_TYPE, M0, N0, lhs, rhs, dst) T_ELTWISE(V_DIV, DST_DATA_TYPE, M0, N0, lhs, rhs, dst)
+#define T_ELTWISE_MUL(DST_DATA_TYPE, M0, N0, lhs, rhs, dst) T_ELTWISE(V_MUL, DST_DATA_TYPE, M0, N0, lhs, rhs, dst)
 
 /** Element-wise operation between two tiles (LHS and RHS)
  *
diff --git a/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentElementwiseBinary.cpp b/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentElementwiseBinary.cpp
index 9a218b3e75..2611d6d575 100644
--- a/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentElementwiseBinary.cpp
+++ b/src/dynamic_fusion/sketch/gpu/components/cl/ClComponentElementwiseBinary.cpp
@@ -37,7 +37,8 @@ namespace
 {
 std::set<ElementwiseBinaryCommonAttributes::ElementwiseOp> supported_ops
 {
-    ElementwiseBinaryCommonAttributes::ElementwiseOp::ADD
+    ElementwiseBinaryCommonAttributes::ElementwiseOp::Add,
+    ElementwiseBinaryCommonAttributes::ElementwiseOp::Mul
 };
 }
 
diff --git a/src/dynamic_fusion/sketch/gpu/operators/GpuAdd.cpp b/src/dynamic_fusion/sketch/gpu/operators/GpuAdd.cpp
index a02160cba8..e7ee1c10df 100644
--- a/src/dynamic_fusion/sketch/gpu/operators/GpuAdd.cpp
+++ b/src/dynamic_fusion/sketch/gpu/operators/GpuAdd.cpp
@@ -22,8 +22,8 @@
  * SOFTWARE.
  */
 #include "arm_compute/dynamic_fusion/sketch/gpu/operators/GpuAdd.h"
+#include "arm_compute/dynamic_fusion/sketch/gpu/GpuWorkloadSketch.h"
 
-#include "src/dynamic_fusion/sketch/gpu/GpuWorkloadSketchImpl.h"
 #include "src/dynamic_fusion/sketch/gpu/operators/internal/GpuElementwiseBinaryCommon.h"
 
 namespace arm_compute
@@ -36,9 +36,13 @@ Status GpuAdd::validate_op(const GpuWorkloadSketch &sketch,
                            const ITensorInfo       *lhs,
                            const ITensorInfo       *rhs)
 {
-    // Set the elementwise operation to ADD then call the elementwise common validate_op
+    ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(lhs, rhs);
+    ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(lhs, 1, DataType::F16, DataType::F32, DataType::U8, DataType::S16, DataType::S32);
+    ARM_COMPUTE_RETURN_ERROR_ON_MSG(lhs->data_type() != rhs->data_type(), "Input tensors must be the same data type");
+
+    // Set the elementwise operation to Add then call the elementwise common validate_op
     ElementwiseBinaryCommonAttributes common_attributes{};
-    common_attributes.operation(ElementwiseBinaryCommonAttributes::ElementwiseOp::ADD);
+    common_attributes.operation(ElementwiseBinaryCommonAttributes::ElementwiseOp::Add);
     return GpuElementwiseBinaryCommon::validate_op(sketch, lhs, rhs, common_attributes);
 }
 
@@ -46,9 +50,13 @@ Status GpuAdd::is_supported_op(const GpuWorkloadContext &context,
                                const ITensorInfo        *lhs,
                                const ITensorInfo        *rhs)
 {
-    // Set the elementwise operation to ADD then call the elementwise common is_supported_op
+    ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(lhs, rhs);
+    ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(lhs, 1, DataType::F16, DataType::F32, DataType::U8, DataType::S16, DataType::S32);
+    ARM_COMPUTE_RETURN_ERROR_ON_MSG(lhs->data_type() != rhs->data_type(), "Input tensors must be the same data type");
+
+    // Set the elementwise operation to Add then call the elementwise common is_supported_op
     ElementwiseBinaryCommonAttributes common_attributes{};
-    common_attributes.operation(ElementwiseBinaryCommonAttributes::ElementwiseOp::ADD);
+    common_attributes.operation(ElementwiseBinaryCommonAttributes::ElementwiseOp::Add);
     return GpuElementwiseBinaryCommon::is_supported_op(context, lhs, rhs, common_attributes);
 }
 
@@ -57,9 +65,9 @@ ITensorInfo *GpuAdd::create_op(GpuWorkloadSketch &sketch,
                                ITensorInfo       *rhs)
 {
     // No need to log or validate as they'll be handled inside GpuElementwiseBinaryCommon::create_op()
-    // Set the elementwise operation to ADD then call the elementwise common create_op
+    // Set the elementwise operation to Add then call the elementwise common create_op
     ElementwiseBinaryCommonAttributes common_attributes{};
-    common_attributes.operation(ElementwiseBinaryCommonAttributes::ElementwiseOp::ADD);
+    common_attributes.operation(ElementwiseBinaryCommonAttributes::ElementwiseOp::Add);
     return GpuElementwiseBinaryCommon::create_op(sketch, lhs, rhs, common_attributes);
 }
 
diff --git a/src/dynamic_fusion/sketch/gpu/operators/GpuMul.cpp b/src/dynamic_fusion/sketch/gpu/operators/GpuMul.cpp
new file mode 100644
index 0000000000..464a32cbad
--- /dev/null
+++ b/src/dynamic_fusion/sketch/gpu/operators/GpuMul.cpp
@@ -0,0 +1,75 @@
+/*
+ * 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/operators/GpuMul.h"
+#include "arm_compute/dynamic_fusion/sketch/gpu/GpuWorkloadSketch.h"
+
+#include "src/dynamic_fusion/sketch/gpu/operators/internal/GpuElementwiseBinaryCommon.h"
+
+namespace arm_compute
+{
+namespace experimental
+{
+namespace dynamic_fusion
+{
+Status GpuMul::validate_op(const GpuWorkloadSketch &sketch,
+                           const ITensorInfo       *lhs,
+                           const ITensorInfo       *rhs)
+{
+    ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(lhs, rhs);
+    ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(lhs, 1, DataType::F16, DataType::F32);
+    ARM_COMPUTE_RETURN_ERROR_ON_MSG(lhs->data_type() != rhs->data_type(), "Input tensors must be the same data type");
+
+    // Set the elementwise operation to Mul then call the elementwise common validate_op
+    ElementwiseBinaryCommonAttributes common_attributes{};
+    common_attributes.operation(ElementwiseBinaryCommonAttributes::ElementwiseOp::Mul);
+    return GpuElementwiseBinaryCommon::validate_op(sketch, lhs, rhs, common_attributes);
+}
+
+Status GpuMul::is_supported_op(const GpuWorkloadContext &context,
+                               const ITensorInfo        *lhs,
+                               const ITensorInfo        *rhs)
+{
+    ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(lhs, rhs);
+    ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(lhs, 1, DataType::F16, DataType::F32);
+    ARM_COMPUTE_RETURN_ERROR_ON_MSG(lhs->data_type() != rhs->data_type(), "Input tensors must be the same data type");
+
+    // Set the elementwise operation to Mul then call the elementwise common is_supported_op
+    ElementwiseBinaryCommonAttributes common_attributes{};
+    common_attributes.operation(ElementwiseBinaryCommonAttributes::ElementwiseOp::Mul);
+    return GpuElementwiseBinaryCommon::is_supported_op(context, lhs, rhs, common_attributes);
+}
+
+ITensorInfo *GpuMul::create_op(GpuWorkloadSketch &sketch,
+                               ITensorInfo       *lhs,
+                               ITensorInfo       *rhs)
+{
+    // Set the elementwise operation to Mul then call the elementwise common create_op
+    ElementwiseBinaryCommonAttributes common_attributes{};
+    common_attributes.operation(ElementwiseBinaryCommonAttributes::ElementwiseOp::Mul);
+    return GpuElementwiseBinaryCommon::create_op(sketch, lhs, rhs, common_attributes);
+}
+
+} // namespace dynamic_fusion
+} // namespace experimental
+} // namespace arm_compute
diff --git a/src/dynamic_fusion/sketch/gpu/operators/internal/GpuElementwiseBinaryCommon.h b/src/dynamic_fusion/sketch/gpu/operators/internal/GpuElementwiseBinaryCommon.h
index cbefa379e6..0b58b6eb96 100644
--- a/src/dynamic_fusion/sketch/gpu/operators/internal/GpuElementwiseBinaryCommon.h
+++ b/src/dynamic_fusion/sketch/gpu/operators/internal/GpuElementwiseBinaryCommon.h
@@ -25,11 +25,12 @@
 #define SRC_DYNAMIC_FUSION_SKETCH_GPU_OPERATORS_INTERNAL_GPUELEMENTWISEBINARYCOMMON
 
 #include "arm_compute/core/Error.h"
-#include "arm_compute/core/ITensorInfo.h"
 
 namespace arm_compute
 {
+/** Forward declaration */
 class ITensorInfo;
+
 namespace experimental
 {
 namespace dynamic_fusion
@@ -39,14 +40,15 @@ class ElementwiseBinaryCommonAttributes
 public:
     enum class ElementwiseOp
     {
-        ADD,          /**< (x + y) */
-        SUB,          /**< (x - y) */
-        DIV,          /**< (x / y) */
-        MIN,          /**< Min(x, y) */
-        MAX,          /**< Max(x, y) */
-        SQUARED_DIFF, /**< (x - y)^2 */
-        POWER,        /**< x ^ y */
-        PRELU,        /**< y*x if x < 0, x otherwise */
+        Add,         /**< (x + y) */
+        Sub,         /**< (x - y) */
+        Div,         /**< (x / y) */
+        Mul,         /**< (x * y) */
+        Min,         /**< Min(x, y) */
+        Max,         /**< Max(x, y) */
+        SquaredDiff, /**< (x - y)^2 */
+        Power,       /**< x ^ y */
+        Prelu,       /**< y*x if x < 0, x otherwise */
     };
     /** Set operation*/
     ElementwiseBinaryCommonAttributes &operation(const ElementwiseBinaryCommonAttributes::ElementwiseOp &operation);
diff --git a/src/dynamic_fusion/sketch/gpu/template_writer/cl/ClTemplateElementwiseBinary.cpp b/src/dynamic_fusion/sketch/gpu/template_writer/cl/ClTemplateElementwiseBinary.cpp
index 01017ed909..0dd7ca5e78 100644
--- a/src/dynamic_fusion/sketch/gpu/template_writer/cl/ClTemplateElementwiseBinary.cpp
+++ b/src/dynamic_fusion/sketch/gpu/template_writer/cl/ClTemplateElementwiseBinary.cpp
@@ -68,7 +68,7 @@ std::string ClTemplateElementwiseBinary::get_component_code(const ComponentGroup
 
     code =
 R"_(
-    //------------------ START KERNEL {{meta_kernel_id}} ELTWISE_OP ---------------------
+    //------------------ START KERNEL {{meta_kernel_id}} {{ELTWISE_OP}} ---------------------
 )_";
 
     if(is_root)
@@ -139,7 +139,7 @@ R"_(
     code +=
 R"_(
     }
-    //------------------ END KERNEL {{meta_kernel_id}} ELTWISE_OP ---------------------
+    //------------------ END KERNEL {{meta_kernel_id}} {{ELTWISE_OP}} ---------------------
 )_";
 
     return code;
@@ -168,33 +168,34 @@ void ClTemplateElementwiseBinary::declare_variables(GpuKernelVariableTable &vtab
 
 TagLUT ClTemplateElementwiseBinary::get_tag_lut(const GpuKernelVariableTable &vtable, const ComponentGroup &comp_group) const
 {
-    TagLUT             lut{};
+    TagLUT lut{};
 
     // Local build options
     lut["meta_kernel_id"] = id();
     lut["DATA_TYPE"]      = get_cl_type_from_data_type(_lhs->data_type());
     // Arguments and global shared variables
 
-    lut["lhs"] = vtable.get_variable(_lhs);
-    lut["rhs"] = vtable.get_variable(_rhs);
-    lut["dst"] = vtable.get_variable(_dst);
+    lut["lhs"]     = vtable.get_variable(_lhs);
+    lut["rhs"]     = vtable.get_variable(_rhs);
+    lut["dst"]     = vtable.get_variable(_dst);
     lut["arg_dst"] = vtable.get_variable(comp_group.get_any_dst_tensor());
 
     switch(_attributes.operation())
     {
-        case Attributes::ElementwiseOp::ADD:
+        case Attributes::ElementwiseOp::Add:
             lut["ELTWISE_OP"] = "ADD";
             break;
+        case Attributes::ElementwiseOp::Mul:
+            lut["ELTWISE_OP"] = "MUL";
+            break;
         default:
             ARM_COMPUTE_ERROR("Arithmetic Operation not supported");
     }
 
     ARM_COMPUTE_ERROR_ON(
-        comp_group.is_intermediate_tensor(_lhs) &&
-        detail::have_different_dimensions(_lhs->tensor_shape(), _dst->tensor_shape(), 0));
+        comp_group.is_intermediate_tensor(_lhs) && detail::have_different_dimensions(_lhs->tensor_shape(), _dst->tensor_shape(), 0));
     ARM_COMPUTE_ERROR_ON(
-        comp_group.is_intermediate_tensor(_rhs) &&
-        detail::have_different_dimensions(_rhs->tensor_shape(), _dst->tensor_shape(), 0));
+        comp_group.is_intermediate_tensor(_rhs) && detail::have_different_dimensions(_rhs->tensor_shape(), _dst->tensor_shape(), 0));
 
     // Set broadcast parameters
     // PRE: All tensors are broadcast-compatible
@@ -222,9 +223,9 @@ TagLUT ClTemplateElementwiseBinary::get_tag_lut(const GpuKernelVariableTable &vt
     lut["rhs_m0"]          = (rhs_broadcast_yz) ? "1" : "M0";
     lut["rhs_start_ind_1"] = (rhs_broadcast_yz) ? "0" : "g_ind_1";
 
-    lut["BROADCAST_OP"]    = (lhs_broadcast_yz) ? "BROADCAST_LHS_X_" :
-                             (rhs_broadcast_yz) ? "BROADCAST_RHS_X_" :
-                                                  "";
+    lut["BROADCAST_OP"] = (lhs_broadcast_yz) ? "BROADCAST_LHS_X_" :
+                          (rhs_broadcast_yz) ? "BROADCAST_RHS_X_" :
+                                               "";
 
     return lut;
 }
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 */
diff --git a/utils/TypePrinter.h b/utils/TypePrinter.h
index 61f0eb9cec..448f184432 100644
--- a/utils/TypePrinter.h
+++ b/utils/TypePrinter.h
@@ -421,8 +421,7 @@ inline ::std::ostream &operator<<(::std::ostream &os, const GEMMLHSMatrixInfo &g
  */
 inline ::std::ostream &operator<<(::std::ostream &os, const GEMMRHSMatrixInfo &gemm_info)
 {
-    os << "( n0=" << (unsigned int)gemm_info.n0 << " k0=" << gemm_info.k0 << "  h0=" << gemm_info.h0 << "  trans=" << gemm_info.transpose << "  inter=" << gemm_info.interleave << " exp_img=" <<
-       gemm_info.export_to_cl_image << "})";
+    os << "( n0=" << (unsigned int)gemm_info.n0 << " k0=" << gemm_info.k0 << "  h0=" << gemm_info.h0 << "  trans=" << gemm_info.transpose << "  inter=" << gemm_info.interleave << " exp_img=" << gemm_info.export_to_cl_image << "})";
     return os;
 }
 
@@ -475,8 +474,7 @@ inline std::string to_string(const GEMMKernelInfo &gemm_info)
 inline ::std::ostream &operator<<(::std::ostream &os, const BoundingBoxTransformInfo &bbox_info)
 {
     auto weights = bbox_info.weights();
-    os << "(" << bbox_info.img_width() << "x" << bbox_info.img_height() << ")~" << bbox_info.scale() << "(weights={" << weights[0] << ", " << weights[1] << ", " << weights[2] << ", " << weights[3] <<
-       "})";
+    os << "(" << bbox_info.img_width() << "x" << bbox_info.img_height() << ")~" << bbox_info.scale() << "(weights={" << weights[0] << ", " << weights[1] << ", " << weights[2] << ", " << weights[3] << "})";
     return os;
 }
 
@@ -3305,46 +3303,46 @@ inline std::string to_string(const Conv3dInfo &conv3d_info)
 inline std::string to_string(const WeightFormat wf)
 {
 #define __CASE_WEIGHT_FORMAT(wf) \
-case WeightFormat::wf:       \
-    return #wf;
+    case WeightFormat::wf:       \
+        return #wf;
     switch(wf)
     {
-            __CASE_WEIGHT_FORMAT(UNSPECIFIED)
-            __CASE_WEIGHT_FORMAT(ANY)
-            __CASE_WEIGHT_FORMAT(OHWI)
-            __CASE_WEIGHT_FORMAT(OHWIo2)
-            __CASE_WEIGHT_FORMAT(OHWIo4)
-            __CASE_WEIGHT_FORMAT(OHWIo8)
-            __CASE_WEIGHT_FORMAT(OHWIo16)
-            __CASE_WEIGHT_FORMAT(OHWIo32)
-            __CASE_WEIGHT_FORMAT(OHWIo64)
-            __CASE_WEIGHT_FORMAT(OHWIo128)
-            __CASE_WEIGHT_FORMAT(OHWIo4i2)
-            __CASE_WEIGHT_FORMAT(OHWIo4i2_bf16)
-            __CASE_WEIGHT_FORMAT(OHWIo8i2)
-            __CASE_WEIGHT_FORMAT(OHWIo8i2_bf16)
-            __CASE_WEIGHT_FORMAT(OHWIo16i2)
-            __CASE_WEIGHT_FORMAT(OHWIo16i2_bf16)
-            __CASE_WEIGHT_FORMAT(OHWIo32i2)
-            __CASE_WEIGHT_FORMAT(OHWIo32i2_bf16)
-            __CASE_WEIGHT_FORMAT(OHWIo64i2)
-            __CASE_WEIGHT_FORMAT(OHWIo64i2_bf16)
-            __CASE_WEIGHT_FORMAT(OHWIo4i4)
-            __CASE_WEIGHT_FORMAT(OHWIo4i4_bf16)
-            __CASE_WEIGHT_FORMAT(OHWIo8i4)
-            __CASE_WEIGHT_FORMAT(OHWIo8i4_bf16)
-            __CASE_WEIGHT_FORMAT(OHWIo16i4)
-            __CASE_WEIGHT_FORMAT(OHWIo16i4_bf16)
-            __CASE_WEIGHT_FORMAT(OHWIo32i4)
-            __CASE_WEIGHT_FORMAT(OHWIo32i4_bf16)
-            __CASE_WEIGHT_FORMAT(OHWIo64i4)
-            __CASE_WEIGHT_FORMAT(OHWIo64i4_bf16)
-            __CASE_WEIGHT_FORMAT(OHWIo2i8)
-            __CASE_WEIGHT_FORMAT(OHWIo4i8)
-            __CASE_WEIGHT_FORMAT(OHWIo8i8)
-            __CASE_WEIGHT_FORMAT(OHWIo16i8)
-            __CASE_WEIGHT_FORMAT(OHWIo32i8)
-            __CASE_WEIGHT_FORMAT(OHWIo64i8)
+        __CASE_WEIGHT_FORMAT(UNSPECIFIED)
+        __CASE_WEIGHT_FORMAT(ANY)
+        __CASE_WEIGHT_FORMAT(OHWI)
+        __CASE_WEIGHT_FORMAT(OHWIo2)
+        __CASE_WEIGHT_FORMAT(OHWIo4)
+        __CASE_WEIGHT_FORMAT(OHWIo8)
+        __CASE_WEIGHT_FORMAT(OHWIo16)
+        __CASE_WEIGHT_FORMAT(OHWIo32)
+        __CASE_WEIGHT_FORMAT(OHWIo64)
+        __CASE_WEIGHT_FORMAT(OHWIo128)
+        __CASE_WEIGHT_FORMAT(OHWIo4i2)
+        __CASE_WEIGHT_FORMAT(OHWIo4i2_bf16)
+        __CASE_WEIGHT_FORMAT(OHWIo8i2)
+        __CASE_WEIGHT_FORMAT(OHWIo8i2_bf16)
+        __CASE_WEIGHT_FORMAT(OHWIo16i2)
+        __CASE_WEIGHT_FORMAT(OHWIo16i2_bf16)
+        __CASE_WEIGHT_FORMAT(OHWIo32i2)
+        __CASE_WEIGHT_FORMAT(OHWIo32i2_bf16)
+        __CASE_WEIGHT_FORMAT(OHWIo64i2)
+        __CASE_WEIGHT_FORMAT(OHWIo64i2_bf16)
+        __CASE_WEIGHT_FORMAT(OHWIo4i4)
+        __CASE_WEIGHT_FORMAT(OHWIo4i4_bf16)
+        __CASE_WEIGHT_FORMAT(OHWIo8i4)
+        __CASE_WEIGHT_FORMAT(OHWIo8i4_bf16)
+        __CASE_WEIGHT_FORMAT(OHWIo16i4)
+        __CASE_WEIGHT_FORMAT(OHWIo16i4_bf16)
+        __CASE_WEIGHT_FORMAT(OHWIo32i4)
+        __CASE_WEIGHT_FORMAT(OHWIo32i4_bf16)
+        __CASE_WEIGHT_FORMAT(OHWIo64i4)
+        __CASE_WEIGHT_FORMAT(OHWIo64i4_bf16)
+        __CASE_WEIGHT_FORMAT(OHWIo2i8)
+        __CASE_WEIGHT_FORMAT(OHWIo4i8)
+        __CASE_WEIGHT_FORMAT(OHWIo8i8)
+        __CASE_WEIGHT_FORMAT(OHWIo16i8)
+        __CASE_WEIGHT_FORMAT(OHWIo32i8)
+        __CASE_WEIGHT_FORMAT(OHWIo64i8)
         default:
             return "invalid value";
     }
@@ -3629,7 +3627,7 @@ inline std::string to_string(const experimental::dynamic_fusion::ResizeAttribute
  */
 inline ::std::ostream &operator<<(::std::ostream &os, const experimental::dynamic_fusion::SoftmaxAttributes &softmax_attr)
 {
-    os << "SofmtaxAttributes="
+    os << "SoftmaxAttributes="
        << "["
        << "Beta=" << softmax_attr.beta() << ", "
        << "Is Log Softmax=" << softmax_attr.is_log_softmax() << ", "