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>

5 files changed, 118 insertions, 31 deletions

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;
 }