Reduction external API and client examples (#493)

* Change to the DeviceReduce base class template to include all problem description information

* Add external api for reduction

* Add client example to test the reduction external api

* Spelling correction

* Re-implement the host_reduction to follow the DeviceReduce base API format

* Change the reduce profiler to call the external API for collecting device instances

* Rename reduce client example directory from 08_reduce to 12_reduce

* Remove (void) before the functional call

* Tiny update in reduce client example

* Tiny update in profile_reduce_impl.hpp

* Rename the reduce client example directory

Co-authored-by: Po Yen Chen <PoYen.Chen@amd.com>

[ROCm/composable_kernel commit: 80e0526741]

library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_add.cpp

@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&);
 // clang-format on
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_amax.cpp

@@ -11,14 +11,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, UnaryAbs, PassThrough>>&);
+template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<BF16, F32, BF16, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_avg.cpp

@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, UnaryDivide>>&);
+template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&);
 // clang-format on
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_max.cpp

@@ -11,14 +11,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<BF16, F32, BF16, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_min.cpp

@@ -11,14 +11,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<BF16, F32, BF16, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_norm2.cpp

@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 3, UnarySquare, UnarySqrt>>&); 
-template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 4, UnarySquare, UnarySqrt>>&); 
-template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 1, UnarySquare, UnarySqrt>>&); 
-template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<2, 1, UnarySquare, UnarySqrt>>&);
+template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+template void add_device_reduce_instance_blockwise<BF16, F32, BF16, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&);
 // clang-format on
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f16_f16_amax.cpp

@@ -11,14 +11,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F16, F16, F16, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F16, F16, F16, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, UnaryAbs, PassThrough>>&);
+template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<F16, F16, F16, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F16, F16, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<F16, F16, F16, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F16, F16, F16, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f16_f16_max.cpp

@@ -11,14 +11,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F16, F16, F16, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F16, F16, F16, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<F16, F16, F16, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F16, F16, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<F16, F16, F16, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F16, F16, F16, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f16_f16_min.cpp

@@ -11,14 +11,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F16, F16, F16, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F16, F16, F16, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<F16, F16, F16, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F16, F16, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<F16, F16, F16, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<F16, F16, F16, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F16, F16, F16, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f32_f16_add.cpp

@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_blockwise<F16, F32, F16, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_blockwise<F16, F32, F16, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_blockwise<F16, F32, F16, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_blockwise<F16, F32, F16, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+template void add_device_reduce_instance_blockwise<F16, F32, F16, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F32, F16, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_blockwise<F16, F32, F16, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F32, F16, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_blockwise<F16, F32, F16, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F32, F16, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_blockwise<F16, F32, F16, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F32, F16, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&);
 // clang-format on
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f32_f16_avg.cpp

@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_blockwise<F16, F32, F16, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_blockwise<F16, F32, F16, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_blockwise<F16, F32, F16, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_blockwise<F16, F32, F16, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, UnaryDivide>>&);
+template void add_device_reduce_instance_blockwise<F16, F32, F16, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F16, F32, F16, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_blockwise<F16, F32, F16, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F16, F32, F16, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_blockwise<F16, F32, F16, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F16, F32, F16, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_blockwise<F16, F32, F16, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F16, F32, F16, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&);
 // clang-format on
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f32_f16_norm2.cpp

@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_blockwise<F16, F32, F16, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 3, UnarySquare, UnarySqrt>>&); 
-template void add_device_reduce_instance_blockwise<F16, F32, F16, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 4, UnarySquare, UnarySqrt>>&); 
-template void add_device_reduce_instance_blockwise<F16, F32, F16, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 1, UnarySquare, UnarySqrt>>&); 
-template void add_device_reduce_instance_blockwise<F16, F32, F16, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<2, 1, UnarySquare, UnarySqrt>>&);
+template void add_device_reduce_instance_blockwise<F16, F32, F16, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F16, F32, F16, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+template void add_device_reduce_instance_blockwise<F16, F32, F16, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F16, F32, F16, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+template void add_device_reduce_instance_blockwise<F16, F32, F16, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F16, F32, F16, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+template void add_device_reduce_instance_blockwise<F16, F32, F16, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F16, F32, F16, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&);
 // clang-format on
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_add.cpp

@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_blockwise<F32, F32, F32, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&);
 // clang-format on
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_amax.cpp

@@ -11,14 +11,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, UnaryAbs, PassThrough>>&);
+template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<F32, F32, F32, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<F32, F32, F32, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_avg.cpp

@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, UnaryDivide>>&);
+template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_blockwise<F32, F32, F32, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&);
 // clang-format on
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_max.cpp

@@ -11,14 +11,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<F32, F32, F32, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<F32, F32, F32, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_min.cpp

@@ -11,14 +11,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<F32, F32, F32, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<F32, F32, F32, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_norm2.cpp

@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 3, UnarySquare, UnarySqrt>>&); 
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 4, UnarySquare, UnarySqrt>>&); 
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 1, UnarySquare, UnarySqrt>>&); 
-template void add_device_reduce_instance_blockwise<F32, F32, F32, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<2, 1, UnarySquare, UnarySqrt>>&);
+template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+template void add_device_reduce_instance_blockwise<F32, F32, F32, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+template void add_device_reduce_instance_blockwise<F32, F32, F32, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&);
 // clang-format on
 // clang-format on
 

library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f64_f32_add.cpp

@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_blockwise<F32, F64, F32, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_blockwise<F32, F64, F32, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_blockwise<F32, F64, F32, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_blockwise<F32, F64, F32, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+template void add_device_reduce_instance_blockwise<F32, F64, F32, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_blockwise<F32, F64, F32, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_blockwise<F32, F64, F32, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_blockwise<F32, F64, F32, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&);
 // clang-format on
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f64_f32_avg.cpp

@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_blockwise<F32, F64, F32, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_blockwise<F32, F64, F32, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_blockwise<F32, F64, F32, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_blockwise<F32, F64, F32, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, UnaryDivide>>&);
+template void add_device_reduce_instance_blockwise<F32, F64, F32, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_blockwise<F32, F64, F32, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_blockwise<F32, F64, F32, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_blockwise<F32, F64, F32, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&);
 // clang-format on
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f64_f32_norm2.cpp

@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_blockwise<F32, F64, F32, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 3, UnarySquare, UnarySqrt>>&); 
-template void add_device_reduce_instance_blockwise<F32, F64, F32, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 4, UnarySquare, UnarySqrt>>&); 
-template void add_device_reduce_instance_blockwise<F32, F64, F32, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 1, UnarySquare, UnarySqrt>>&); 
-template void add_device_reduce_instance_blockwise<F32, F64, F32, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<2, 1, UnarySquare, UnarySqrt>>&);
+template void add_device_reduce_instance_blockwise<F32, F64, F32, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+template void add_device_reduce_instance_blockwise<F32, F64, F32, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+template void add_device_reduce_instance_blockwise<F32, F64, F32, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+template void add_device_reduce_instance_blockwise<F32, F64, F32, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&);
 // clang-format on
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_add.cpp

@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_blockwise<F64, F64, F64, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_blockwise<F64, F64, F64, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&);
 // clang-format on
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_amax.cpp

@@ -11,14 +11,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F64, F64, F64, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F64, F64, F64, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, UnaryAbs, PassThrough>>&);
+template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<F64, F64, F64, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<F64, F64, F64, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F64, F64, F64, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_avg.cpp

@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_blockwise<F64, F64, F64, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, UnaryDivide>>&);
+template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_blockwise<F64, F64, F64, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&);
 // clang-format on
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_max.cpp

@@ -11,14 +11,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F64, F64, F64, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F64, F64, F64, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<F64, F64, F64, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<F64, F64, F64, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F64, F64, F64, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_min.cpp

@@ -11,14 +11,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F64, F64, F64, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<F64, F64, F64, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<F64, F64, F64, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<F64, F64, F64, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F64, F64, F64, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_norm2.cpp

@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 3, UnarySquare, UnarySqrt>>&); 
-template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 4, UnarySquare, UnarySqrt>>&); 
-template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 1, UnarySquare, UnarySqrt>>&); 
-template void add_device_reduce_instance_blockwise<F64, F64, F64, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<2, 1, UnarySquare, UnarySqrt>>&);
+template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+template void add_device_reduce_instance_blockwise<F64, F64, F64, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+template void add_device_reduce_instance_blockwise<F64, F64, F64, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&);
 // clang-format on
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i32_i8_add.cpp

@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_blockwise<I8, I32, I8, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_blockwise<I8, I32, I8, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_blockwise<I8, I32, I8, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_blockwise<I8, I32, I8, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+template void add_device_reduce_instance_blockwise<I8, I32, I8, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I32, I8, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_blockwise<I8, I32, I8, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I32, I8, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_blockwise<I8, I32, I8, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I32, I8, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_blockwise<I8, I32, I8, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I32, I8, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&);
 // clang-format on
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i32_i8_avg.cpp

@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_blockwise<I8, I32, I8, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_blockwise<I8, I32, I8, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_blockwise<I8, I32, I8, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_blockwise<I8, I32, I8, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, UnaryDivide>>&);
+template void add_device_reduce_instance_blockwise<I8, I32, I8, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<I8, I32, I8, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_blockwise<I8, I32, I8, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<I8, I32, I8, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_blockwise<I8, I32, I8, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<I8, I32, I8, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_blockwise<I8, I32, I8, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<I8, I32, I8, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&);
 // clang-format on
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i8_i8_amax.cpp

@@ -11,14 +11,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<I8, I8, I8, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<I8, I8, I8, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, UnaryAbs, PassThrough>>&);
+template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<I8, I8, I8, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I8, I8, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<I8, I8, I8, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<I8, I8, I8, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i8_i8_max.cpp

@@ -11,14 +11,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<I8, I8, I8, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<I8, I8, I8, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<I8, I8, I8, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I8, I8, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<I8, I8, I8, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<I8, I8, I8, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i8_i8_min.cpp

@@ -11,14 +11,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<I8, I8, I8, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_blockwise<I8, I8, I8, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<I8, I8, I8, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I8, I8, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<I8, I8, I8, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_blockwise<I8, I8, I8, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<I8, I8, I8, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_b16_f32_f32_add.cpp

@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_multiblock_atomic_add<BF16, F32, F32, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_multiblock_atomic_add<BF16, F32, F32, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_multiblock_atomic_add<BF16, F32, F32, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_multiblock_atomic_add<BF16, F32, F32, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+template void add_device_reduce_instance_multiblock_atomic_add<BF16, F32, F32, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, F32, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_multiblock_atomic_add<BF16, F32, F32, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, F32, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_multiblock_atomic_add<BF16, F32, F32, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, F32, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_multiblock_atomic_add<BF16, F32, F32, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, F32, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&);
 // clang-format on
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_b16_f32_f32_avg.cpp

@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_multiblock_atomic_add<BF16, F32, F32, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_multiblock_atomic_add<BF16, F32, F32, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_multiblock_atomic_add<BF16, F32, F32, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_multiblock_atomic_add<BF16, F32, F32, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, UnaryDivide>>&);
+template void add_device_reduce_instance_multiblock_atomic_add<BF16, F32, F32, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<BF16, F32, F32, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_multiblock_atomic_add<BF16, F32, F32, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<BF16, F32, F32, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_multiblock_atomic_add<BF16, F32, F32, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<BF16, F32, F32, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_multiblock_atomic_add<BF16, F32, F32, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<BF16, F32, F32, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&);
 // clang-format on
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f16_f32_f32_add.cpp

@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_multiblock_atomic_add<F16, F32, F32, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_multiblock_atomic_add<F16, F32, F32, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_multiblock_atomic_add<F16, F32, F32, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_multiblock_atomic_add<F16, F32, F32, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+template void add_device_reduce_instance_multiblock_atomic_add<F16, F32, F32, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F32, F32, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_multiblock_atomic_add<F16, F32, F32, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F32, F32, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_multiblock_atomic_add<F16, F32, F32, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F32, F32, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_multiblock_atomic_add<F16, F32, F32, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F32, F32, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&);
 // clang-format on
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f16_f32_f32_avg.cpp

@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_multiblock_atomic_add<F16, F32, F32, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_multiblock_atomic_add<F16, F32, F32, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_multiblock_atomic_add<F16, F32, F32, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_multiblock_atomic_add<F16, F32, F32, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, UnaryDivide>>&);
+template void add_device_reduce_instance_multiblock_atomic_add<F16, F32, F32, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F16, F32, F32, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_multiblock_atomic_add<F16, F32, F32, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F16, F32, F32, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_multiblock_atomic_add<F16, F32, F32, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F16, F32, F32, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_multiblock_atomic_add<F16, F32, F32, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F16, F32, F32, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&);
 // clang-format on
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f32_f32_f32_add.cpp

@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_multiblock_atomic_add<F32, F32, F32, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_multiblock_atomic_add<F32, F32, F32, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_multiblock_atomic_add<F32, F32, F32, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_multiblock_atomic_add<F32, F32, F32, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+template void add_device_reduce_instance_multiblock_atomic_add<F32, F32, F32, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_multiblock_atomic_add<F32, F32, F32, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_multiblock_atomic_add<F32, F32, F32, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_multiblock_atomic_add<F32, F32, F32, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&);
 // clang-format on
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f32_f32_f32_avg.cpp

@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_multiblock_atomic_add<F32, F32, F32, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_multiblock_atomic_add<F32, F32, F32, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_multiblock_atomic_add<F32, F32, F32, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_multiblock_atomic_add<F32, F32, F32, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, UnaryDivide>>&);
+template void add_device_reduce_instance_multiblock_atomic_add<F32, F32, F32, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_multiblock_atomic_add<F32, F32, F32, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_multiblock_atomic_add<F32, F32, F32, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_multiblock_atomic_add<F32, F32, F32, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&);
 // clang-format on
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f32_f64_f32_add.cpp

@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_multiblock_atomic_add<F32, F64, F32, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_multiblock_atomic_add<F32, F64, F32, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_multiblock_atomic_add<F32, F64, F32, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_multiblock_atomic_add<F32, F64, F32, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+template void add_device_reduce_instance_multiblock_atomic_add<F32, F64, F32, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_multiblock_atomic_add<F32, F64, F32, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_multiblock_atomic_add<F32, F64, F32, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_multiblock_atomic_add<F32, F64, F32, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&);
 // clang-format on
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f32_f64_f32_avg.cpp

@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_multiblock_atomic_add<F32, F64, F32, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_multiblock_atomic_add<F32, F64, F32, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_multiblock_atomic_add<F32, F64, F32, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_multiblock_atomic_add<F32, F64, F32, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, UnaryDivide>>&);
+template void add_device_reduce_instance_multiblock_atomic_add<F32, F64, F32, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_multiblock_atomic_add<F32, F64, F32, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_multiblock_atomic_add<F32, F64, F32, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_multiblock_atomic_add<F32, F64, F32, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&);
 // clang-format on
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f64_f64_f64_add.cpp

@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_multiblock_atomic_add<F64, F64, F64, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_multiblock_atomic_add<F64, F64, F64, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_multiblock_atomic_add<F64, F64, F64, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_multiblock_atomic_add<F64, F64, F64, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+template void add_device_reduce_instance_multiblock_atomic_add<F64, F64, F64, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_multiblock_atomic_add<F64, F64, F64, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_multiblock_atomic_add<F64, F64, F64, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_multiblock_atomic_add<F64, F64, F64, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&);
 // clang-format on
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f64_f64_f64_avg.cpp

@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_multiblock_atomic_add<F64, F64, F64, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_multiblock_atomic_add<F64, F64, F64, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_multiblock_atomic_add<F64, F64, F64, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_multiblock_atomic_add<F64, F64, F64, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, UnaryDivide>>&);
+template void add_device_reduce_instance_multiblock_atomic_add<F64, F64, F64, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_multiblock_atomic_add<F64, F64, F64, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_multiblock_atomic_add<F64, F64, F64, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_multiblock_atomic_add<F64, F64, F64, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&);
 // clang-format on
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_add.cpp

@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&);
 // clang-format on
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_amax.cpp

@@ -11,14 +11,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, UnaryAbs, PassThrough>>&);
+template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<BF16, F32, BF16, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_avg.cpp

@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, UnaryDivide>>&);
+template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&);
 // clang-format on
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_max.cpp

@@ -11,14 +11,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<BF16, F32, BF16, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_min.cpp

@@ -11,14 +11,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<BF16, F32, BF16, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_norm2.cpp

@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 3, UnarySquare, UnarySqrt>>&); 
-template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 4, UnarySquare, UnarySqrt>>&); 
-template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 1, UnarySquare, UnarySqrt>>&); 
-template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<2, 1, UnarySquare, UnarySqrt>>&);
+template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+template void add_device_reduce_instance_threadwise<BF16, F32, BF16, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<BF16, F32, BF16, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&);
 // clang-format on
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f16_f16_amax.cpp

@@ -11,14 +11,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F16, F16, F16, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F16, F16, F16, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, UnaryAbs, PassThrough>>&);
+template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<F16, F16, F16, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F16, F16, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<F16, F16, F16, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F16, F16, F16, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f16_f16_max.cpp

@@ -11,14 +11,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F16, F16, F16, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F16, F16, F16, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<F16, F16, F16, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F16, F16, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<F16, F16, F16, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F16, F16, F16, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f16_f16_min.cpp

@@ -11,14 +11,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F16, F16, F16, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F16, F16, F16, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<F16, F16, F16, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F16, F16, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<F16, F16, F16, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F16, F16, F16, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<F16, F16, F16, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F16, F16, F16, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f32_f16_add.cpp

@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_threadwise<F16, F32, F16, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_threadwise<F16, F32, F16, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_threadwise<F16, F32, F16, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_threadwise<F16, F32, F16, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+template void add_device_reduce_instance_threadwise<F16, F32, F16, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F32, F16, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_threadwise<F16, F32, F16, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F32, F16, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_threadwise<F16, F32, F16, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F32, F16, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_threadwise<F16, F32, F16, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F16, F32, F16, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&);
 // clang-format on
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f32_f16_avg.cpp

@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_threadwise<F16, F32, F16, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_threadwise<F16, F32, F16, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_threadwise<F16, F32, F16, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_threadwise<F16, F32, F16, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, UnaryDivide>>&);
+template void add_device_reduce_instance_threadwise<F16, F32, F16, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F16, F32, F16, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_threadwise<F16, F32, F16, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F16, F32, F16, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_threadwise<F16, F32, F16, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F16, F32, F16, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_threadwise<F16, F32, F16, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F16, F32, F16, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&);
 // clang-format on
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f32_f16_norm2.cpp

@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_threadwise<F16, F32, F16, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 3, UnarySquare, UnarySqrt>>&); 
-template void add_device_reduce_instance_threadwise<F16, F32, F16, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 4, UnarySquare, UnarySqrt>>&); 
-template void add_device_reduce_instance_threadwise<F16, F32, F16, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 1, UnarySquare, UnarySqrt>>&); 
-template void add_device_reduce_instance_threadwise<F16, F32, F16, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<2, 1, UnarySquare, UnarySqrt>>&);
+template void add_device_reduce_instance_threadwise<F16, F32, F16, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F16, F32, F16, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+template void add_device_reduce_instance_threadwise<F16, F32, F16, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F16, F32, F16, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+template void add_device_reduce_instance_threadwise<F16, F32, F16, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F16, F32, F16, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+template void add_device_reduce_instance_threadwise<F16, F32, F16, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F16, F32, F16, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&);
 // clang-format on
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_add.cpp

@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_threadwise<F32, F32, F32, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_threadwise<F32, F32, F32, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&);
 // clang-format on
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_amax.cpp

@@ -11,14 +11,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F32, F32, F32, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F32, F32, F32, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, UnaryAbs, PassThrough>>&);
+template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<F32, F32, F32, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<F32, F32, F32, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_avg.cpp

@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_threadwise<F32, F32, F32, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, UnaryDivide>>&);
+template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_threadwise<F32, F32, F32, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&);
 // clang-format on
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_max.cpp

@@ -11,14 +11,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F32, F32, F32, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F32, F32, F32, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<F32, F32, F32, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<F32, F32, F32, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_min.cpp

@@ -11,14 +11,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F32, F32, F32, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F32, F32, F32, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<F32, F32, F32, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<F32, F32, F32, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F32, F32, F32, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_norm2.cpp

@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 3, UnarySquare, UnarySqrt>>&); 
-template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 4, UnarySquare, UnarySqrt>>&); 
-template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 1, UnarySquare, UnarySqrt>>&); 
-template void add_device_reduce_instance_threadwise<F32, F32, F32, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<2, 1, UnarySquare, UnarySqrt>>&);
+template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+template void add_device_reduce_instance_threadwise<F32, F32, F32, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+template void add_device_reduce_instance_threadwise<F32, F32, F32, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F32, F32, F32, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&);
 // clang-format on
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f64_f32_add.cpp

@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_threadwise<F32, F64, F32, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_threadwise<F32, F64, F32, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_threadwise<F32, F64, F32, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_threadwise<F32, F64, F32, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+template void add_device_reduce_instance_threadwise<F32, F64, F32, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_threadwise<F32, F64, F32, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_threadwise<F32, F64, F32, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_threadwise<F32, F64, F32, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&);
 // clang-format on
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f64_f32_avg.cpp

@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_threadwise<F32, F64, F32, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_threadwise<F32, F64, F32, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_threadwise<F32, F64, F32, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_threadwise<F32, F64, F32, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, UnaryDivide>>&);
+template void add_device_reduce_instance_threadwise<F32, F64, F32, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_threadwise<F32, F64, F32, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_threadwise<F32, F64, F32, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_threadwise<F32, F64, F32, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&);
 // clang-format on
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f64_f32_norm2.cpp

@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_threadwise<F32, F64, F32, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 3, UnarySquare, UnarySqrt>>&); 
-template void add_device_reduce_instance_threadwise<F32, F64, F32, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 4, UnarySquare, UnarySqrt>>&); 
-template void add_device_reduce_instance_threadwise<F32, F64, F32, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 1, UnarySquare, UnarySqrt>>&); 
-template void add_device_reduce_instance_threadwise<F32, F64, F32, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<2, 1, UnarySquare, UnarySqrt>>&);
+template void add_device_reduce_instance_threadwise<F32, F64, F32, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+template void add_device_reduce_instance_threadwise<F32, F64, F32, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+template void add_device_reduce_instance_threadwise<F32, F64, F32, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+template void add_device_reduce_instance_threadwise<F32, F64, F32, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F32, F64, F32, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&);
 // clang-format on
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_add.cpp

@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_threadwise<F64, F64, F64, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_threadwise<F64, F64, F64, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&);
 // clang-format on
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_amax.cpp

@@ -11,14 +11,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F64, F64, F64, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F64, F64, F64, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, UnaryAbs, PassThrough>>&);
+template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<F64, F64, F64, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<F64, F64, F64, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<F64, F64, F64, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_avg.cpp

@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_threadwise<F64, F64, F64, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, UnaryDivide>>&);
+template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_threadwise<F64, F64, F64, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&);
 // clang-format on
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_max.cpp

@@ -11,14 +11,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F64, F64, F64, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F64, F64, F64, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<F64, F64, F64, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<F64, F64, F64, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F64, F64, F64, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_min.cpp

@@ -11,14 +11,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F64, F64, F64, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<F64, F64, F64, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<F64, F64, F64, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<F64, F64, F64, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<F64, F64, F64, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_norm2.cpp

@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 3, UnarySquare, UnarySqrt>>&); 
-template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 4, UnarySquare, UnarySqrt>>&); 
-template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<4, 1, UnarySquare, UnarySqrt>>&); 
-template void add_device_reduce_instance_threadwise<F64, F64, F64, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<2, 1, UnarySquare, UnarySqrt>>&);
+template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 3, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 4, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+template void add_device_reduce_instance_threadwise<F64, F64, F64, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 4, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&); 
+template void add_device_reduce_instance_threadwise<F64, F64, F64, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>(std::vector<DeviceReducePtr<F64, F64, F64, 2, 1, ReduceAdd, UnarySquare, UnarySqrt, false, false>>&);
 // clang-format on
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i32_i8_add.cpp

@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_threadwise<I8, I32, I8, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_threadwise<I8, I32, I8, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_threadwise<I8, I32, I8, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&); 
-template void add_device_reduce_instance_threadwise<I8, I32, I8, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+template void add_device_reduce_instance_threadwise<I8, I32, I8, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I32, I8, 4, 3, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_threadwise<I8, I32, I8, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I32, I8, 4, 4, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_threadwise<I8, I32, I8, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I32, I8, 4, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&); 
+template void add_device_reduce_instance_threadwise<I8, I32, I8, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I32, I8, 2, 1, ReduceAdd, PassThrough, PassThrough, false, false>>&);
 // clang-format on
 // clang-format on
 

library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i32_i8_avg.cpp

@@ -11,10 +11,10 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_threadwise<I8, I32, I8, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_threadwise<I8, I32, I8, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_threadwise<I8, I32, I8, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, UnaryDivide>>&); 
-template void add_device_reduce_instance_threadwise<I8, I32, I8, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, UnaryDivide>>&);
+template void add_device_reduce_instance_threadwise<I8, I32, I8, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<I8, I32, I8, 4, 3, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_threadwise<I8, I32, I8, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<I8, I32, I8, 4, 4, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_threadwise<I8, I32, I8, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<I8, I32, I8, 4, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&); 
+template void add_device_reduce_instance_threadwise<I8, I32, I8, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>(std::vector<DeviceReducePtr<I8, I32, I8, 2, 1, ReduceAdd, PassThrough, UnaryDivide, false, false>>&);
 // clang-format on
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i8_i8_amax.cpp

@@ -11,14 +11,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<I8, I8, I8, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, UnaryAbs, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<I8, I8, I8, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, UnaryAbs, PassThrough>>&);
+template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<I8, I8, I8, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I8, I8, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 3, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 4, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<I8, I8, I8, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>(std::vector<DeviceReducePtr<I8, I8, I8, 2, 1, ReduceAMax, UnaryAbs, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i8_i8_max.cpp

@@ -11,14 +11,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<I8, I8, I8, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<I8, I8, I8, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 3, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 4, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 1, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<I8, I8, I8, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I8, I8, 2, 1, ReduceMax, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 3, ReduceMax, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 4, ReduceMax, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 1, ReduceMax, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<I8, I8, I8, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<I8, I8, I8, 2, 1, ReduceMax, PassThrough, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance

library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i8_i8_min.cpp

@@ -11,14 +11,14 @@ namespace instance {
 
 // clang-format off
 // InDataType | AccDataType | OutDataType | Rank | NumReduceDim | ReduceOperation | InElementwiseOp | AccElementwiseOp | PropagateNan | UseIndex 
-template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<I8, I8, I8, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 3, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 4, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<4, 1, PassThrough, PassThrough>>&);
-template void add_device_reduce_instance_threadwise<I8, I8, I8, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<2, 1, PassThrough, PassThrough>>&);
+template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 3, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 4, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 1, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<I8, I8, I8, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>(std::vector<DeviceReducePtr<I8, I8, I8, 2, 1, ReduceMin, PassThrough, PassThrough, false, false>>&);
+template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 3, ReduceMin, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 4, ReduceMin, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<I8, I8, I8, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<I8, I8, I8, 4, 1, ReduceMin, PassThrough, PassThrough, false, true>>&);
+template void add_device_reduce_instance_threadwise<I8, I8, I8, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>(std::vector<DeviceReducePtr<I8, I8, I8, 2, 1, ReduceMin, PassThrough, PassThrough, false, true>>&);
 // clang-format on
 
 } // namespace instance