Add elementwise with dynamic vector dim (#1198)

* Add elementwise with dynamic vector dim

* Reduce number of instaces

* Fixes

* Fixes

[ROCm/composable_kernel commit: 9c052804a7]

library/src/tensor_operation_instance/gpu/permute_scale/CMakeLists.txt

@@ -1,7 +1,13 @@
 add_instance_library(device_permute_scale_instance 
-	device_permute_scale_1d_instances.cpp
-	device_permute_scale_2d_instances.cpp
-	device_permute_scale_3d_instances.cpp
-	device_permute_scale_4d_instances.cpp
-	device_permute_scale_5d_instances.cpp
-	device_permute_scale_6d_instances.cpp)
+	device_permute_scale_1d_fp16_instances.cpp
+	device_permute_scale_2d_fp16_instances.cpp
+	device_permute_scale_3d_fp16_instances.cpp
+	device_permute_scale_4d_fp16_instances.cpp
+	device_permute_scale_5d_fp16_instances.cpp
+	device_permute_scale_6d_fp16_instances.cpp
+	device_permute_scale_1d_fp32_instances.cpp
+	device_permute_scale_2d_fp32_instances.cpp
+	device_permute_scale_3d_fp32_instances.cpp
+	device_permute_scale_4d_fp32_instances.cpp
+	device_permute_scale_5d_fp32_instances.cpp
+	device_permute_scale_6d_fp32_instances.cpp)

library/src/tensor_operation_instance/gpu/permute_scale/device_permute_scale_1d_fp16_instances.cpp

@@ -9,18 +9,13 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-void add_device_permute_scale_1d_f16_instances(
-    std::vector<std::unique_ptr<
-        DeviceElementwise<ck::Tuple<F16>, ck::Tuple<F16>, Pass, UnaryOp, Scale, 1>>>& instances)
-{
-    add_device_operation_instances(instances, device_permute_scale_f16_instances<1>{});
-}
+using Scale = element_wise::Scale;
 
-void add_device_permute_scale_1d_f32_instances(
-    std::vector<std::unique_ptr<
-        DeviceElementwise<ck::Tuple<F32>, ck::Tuple<F32>, Pass, UnaryOp, Scale, 1>>>& instances)
+void add_device_permute_scale_1d_f16_instances(
+    std::vector<std::unique_ptr<DeviceElementwise<ck::Tuple<F16>, ck::Tuple<F16>, Scale, 1>>>&
+        instances)
 {
-    add_device_operation_instances(instances, device_permute_scale_f32_instances<1>{});
+    add_device_operation_instances(instances, device_permute_scale_f16_instances<1, Scale>{});
 }
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/permute_scale/device_permute_scale_1d_fp32_instances.cpp

@@ -0,0 +1,24 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/gpu/permute_scale/device_permute_scale_instances.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using Scale = element_wise::Scale;
+
+void add_device_permute_scale_1d_f32_instances(
+    std::vector<std::unique_ptr<DeviceElementwise<ck::Tuple<F32>, ck::Tuple<F32>, Scale, 1>>>&
+        instances)
+{
+    add_device_operation_instances(instances, device_permute_scale_f32_instances<1, Scale>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/permute_scale/device_permute_scale_2d_fp16_instances.cpp

@@ -9,18 +9,13 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-void add_device_permute_scale_2d_f16_instances(
-    std::vector<std::unique_ptr<
-        DeviceElementwise<ck::Tuple<F16>, ck::Tuple<F16>, Pass, UnaryOp, Scale, 2>>>& instances)
-{
-    add_device_operation_instances(instances, device_permute_scale_f16_instances<2>{});
-}
+using Scale = element_wise::Scale;
 
-void add_device_permute_scale_2d_f32_instances(
-    std::vector<std::unique_ptr<
-        DeviceElementwise<ck::Tuple<F32>, ck::Tuple<F32>, Pass, UnaryOp, Scale, 2>>>& instances)
+void add_device_permute_scale_2d_f16_instances(
+    std::vector<std::unique_ptr<DeviceElementwise<ck::Tuple<F16>, ck::Tuple<F16>, Scale, 2>>>&
+        instances)
 {
-    add_device_operation_instances(instances, device_permute_scale_f32_instances<2>{});
+    add_device_operation_instances(instances, device_permute_scale_f16_instances<2, Scale>{});
 }
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/permute_scale/device_permute_scale_2d_fp32_instances.cpp

@@ -0,0 +1,24 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/gpu/permute_scale/device_permute_scale_instances.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using Scale = element_wise::Scale;
+
+void add_device_permute_scale_2d_f32_instances(
+    std::vector<std::unique_ptr<DeviceElementwise<ck::Tuple<F32>, ck::Tuple<F32>, Scale, 2>>>&
+        instances)
+{
+    add_device_operation_instances(instances, device_permute_scale_f32_instances<2, Scale>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/permute_scale/device_permute_scale_3d_fp16_instances.cpp

@@ -9,18 +9,13 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-void add_device_permute_scale_3d_f16_instances(
-    std::vector<std::unique_ptr<
-        DeviceElementwise<ck::Tuple<F16>, ck::Tuple<F16>, Pass, UnaryOp, Scale, 3>>>& instances)
-{
-    add_device_operation_instances(instances, device_permute_scale_f16_instances<3>{});
-}
+using Scale = element_wise::Scale;
 
-void add_device_permute_scale_3d_f32_instances(
-    std::vector<std::unique_ptr<
-        DeviceElementwise<ck::Tuple<F32>, ck::Tuple<F32>, Pass, UnaryOp, Scale, 3>>>& instances)
+void add_device_permute_scale_3d_f16_instances(
+    std::vector<std::unique_ptr<DeviceElementwise<ck::Tuple<F16>, ck::Tuple<F16>, Scale, 3>>>&
+        instances)
 {
-    add_device_operation_instances(instances, device_permute_scale_f32_instances<3>{});
+    add_device_operation_instances(instances, device_permute_scale_f16_instances<3, Scale>{});
 }
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/permute_scale/device_permute_scale_3d_fp32_instances.cpp

@@ -0,0 +1,24 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/gpu/permute_scale/device_permute_scale_instances.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using Scale = element_wise::Scale;
+
+void add_device_permute_scale_3d_f32_instances(
+    std::vector<std::unique_ptr<DeviceElementwise<ck::Tuple<F32>, ck::Tuple<F32>, Scale, 3>>>&
+        instances)
+{
+    add_device_operation_instances(instances, device_permute_scale_f32_instances<3, Scale>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/permute_scale/device_permute_scale_4d_fp16_instances.cpp

@@ -9,18 +9,13 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-void add_device_permute_scale_4d_f16_instances(
-    std::vector<std::unique_ptr<
-        DeviceElementwise<ck::Tuple<F16>, ck::Tuple<F16>, Pass, UnaryOp, Scale, 4>>>& instances)
-{
-    add_device_operation_instances(instances, device_permute_scale_f16_instances<4>{});
-}
+using Scale = element_wise::Scale;
 
-void add_device_permute_scale_4d_f32_instances(
-    std::vector<std::unique_ptr<
-        DeviceElementwise<ck::Tuple<F32>, ck::Tuple<F32>, Pass, UnaryOp, Scale, 4>>>& instances)
+void add_device_permute_scale_4d_f16_instances(
+    std::vector<std::unique_ptr<DeviceElementwise<ck::Tuple<F16>, ck::Tuple<F16>, Scale, 4>>>&
+        instances)
 {
-    add_device_operation_instances(instances, device_permute_scale_f32_instances<4>{});
+    add_device_operation_instances(instances, device_permute_scale_f16_instances<4, Scale>{});
 }
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/permute_scale/device_permute_scale_4d_fp32_instances.cpp

@@ -0,0 +1,24 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/gpu/permute_scale/device_permute_scale_instances.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using Scale = element_wise::Scale;
+
+void add_device_permute_scale_4d_f32_instances(
+    std::vector<std::unique_ptr<DeviceElementwise<ck::Tuple<F32>, ck::Tuple<F32>, Scale, 4>>>&
+        instances)
+{
+    add_device_operation_instances(instances, device_permute_scale_f32_instances<4, Scale>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/permute_scale/device_permute_scale_5d_fp16_instances.cpp

@@ -9,18 +9,13 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-void add_device_permute_scale_5d_f16_instances(
-    std::vector<std::unique_ptr<
-        DeviceElementwise<ck::Tuple<F16>, ck::Tuple<F16>, Pass, UnaryOp, Scale, 5>>>& instances)
-{
-    add_device_operation_instances(instances, device_permute_scale_f16_instances<5>{});
-}
+using Scale = element_wise::Scale;
 
-void add_device_permute_scale_5d_f32_instances(
-    std::vector<std::unique_ptr<
-        DeviceElementwise<ck::Tuple<F32>, ck::Tuple<F32>, Pass, UnaryOp, Scale, 5>>>& instances)
+void add_device_permute_scale_5d_f16_instances(
+    std::vector<std::unique_ptr<DeviceElementwise<ck::Tuple<F16>, ck::Tuple<F16>, Scale, 5>>>&
+        instances)
 {
-    add_device_operation_instances(instances, device_permute_scale_f32_instances<5>{});
+    add_device_operation_instances(instances, device_permute_scale_f16_instances<5, Scale>{});
 }
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/permute_scale/device_permute_scale_5d_fp32_instances.cpp

@@ -0,0 +1,24 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/gpu/permute_scale/device_permute_scale_instances.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using Scale = element_wise::Scale;
+
+void add_device_permute_scale_5d_f32_instances(
+    std::vector<std::unique_ptr<DeviceElementwise<ck::Tuple<F32>, ck::Tuple<F32>, Scale, 5>>>&
+        instances)
+{
+    add_device_operation_instances(instances, device_permute_scale_f32_instances<5, Scale>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/permute_scale/device_permute_scale_6d_fp16_instances.cpp

@@ -9,18 +9,13 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-void add_device_permute_scale_6d_f16_instances(
-    std::vector<std::unique_ptr<
-        DeviceElementwise<ck::Tuple<F16>, ck::Tuple<F16>, Pass, UnaryOp, Scale, 6>>>& instances)
-{
-    add_device_operation_instances(instances, device_permute_scale_f16_instances<6>{});
-}
+using Scale = element_wise::Scale;
 
-void add_device_permute_scale_6d_f32_instances(
-    std::vector<std::unique_ptr<
-        DeviceElementwise<ck::Tuple<F32>, ck::Tuple<F32>, Pass, UnaryOp, Scale, 6>>>& instances)
+void add_device_permute_scale_6d_f16_instances(
+    std::vector<std::unique_ptr<DeviceElementwise<ck::Tuple<F16>, ck::Tuple<F16>, Scale, 6>>>&
+        instances)
 {
-    add_device_operation_instances(instances, device_permute_scale_f32_instances<6>{});
+    add_device_operation_instances(instances, device_permute_scale_f16_instances<6, Scale>{});
 }
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/permute_scale/device_permute_scale_6d_fp32_instances.cpp

@@ -0,0 +1,24 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/gpu/permute_scale/device_permute_scale_instances.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using Scale = element_wise::Scale;
+
+void add_device_permute_scale_6d_f32_instances(
+    std::vector<std::unique_ptr<DeviceElementwise<ck::Tuple<F32>, ck::Tuple<F32>, Scale, 6>>>&
+        instances)
+{
+    add_device_operation_instances(instances, device_permute_scale_f32_instances<6, Scale>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck