Merge branch 'develop' into jd/dev_pkg

include/ck/tensor_operation/gpu/block/blockwise_tensor_slice_transfer_v5r1.hpp

@@ -14,7 +14,7 @@ namespace ck {
 // 2. ThreadwiseTensorSliceTransfer_v3 does not keep reference to tensor descriptor
 // 3. ThreadwiseTensorSliceTransfer_v3::Run() does not construct new tensor coordinate
 template <index_t BlockSize,
-          InMemoryDataOperationEnum_t DstInMemOp,
+          InMemoryDataOperationEnum DstInMemOp,
           typename BlockSliceLengths,
           typename ThreadSliceLengths,
           typename ThreadClusterLengths,

include/ck/tensor_operation/gpu/block/blockwise_tensor_slice_transfer_v6r1.hpp

@@ -15,7 +15,7 @@ namespace ck {
 // 3. ThreadwiseTensorSliceTransfer_v3::Run() does not construct new tensor coordinate
 template <index_t BlockSize,
           typename ElementwiseOperation,
-          InMemoryDataOperationEnum_t DstInMemOp,
+          InMemoryDataOperationEnum DstInMemOp,
           typename BlockSliceLengths,
           typename ThreadClusterLengths,
           typename ThreadClusterArrangeOrder,

include/ck/tensor_operation/gpu/block/blockwise_tensor_slice_transfer_v6r2.hpp

@@ -15,7 +15,7 @@ namespace ck {
 // 3. Run() does not construct new tensor coordinate
 template <index_t BlockSize,
           typename ElementwiseOperation,
-          InMemoryDataOperationEnum_t DstInMemOp,
+          InMemoryDataOperationEnum DstInMemOp,
           typename BlockSliceLengths,
           typename ThreadClusterLengths,
           typename ThreadClusterArrangeOrder,

include/ck/tensor_operation/gpu/block/blockwise_tensor_slice_transfer_v6r3.hpp

@@ -15,7 +15,7 @@ namespace ck {
 // 3. ThreadwiseTensorSliceTransfer_v3::Run() does not construct new tensor coordinate
 template <index_t BlockSize,
           typename ElementwiseOperation,
-          InMemoryDataOperationEnum_t DstInMemOp,
+          InMemoryDataOperationEnum DstInMemOp,
           typename BlockSliceLengths,
           typename ThreadClusterLengths,
           typename ThreadClusterArrangeOrder,

include/ck/tensor_operation/gpu/device/convolution_backward_data_specialization.hpp

@@ -5,7 +5,7 @@ namespace ck {
 namespace tensor_operation {
 namespace device {
 
-enum ConvolutionBackwardDataSpecialization_t
+enum struct ConvolutionBackwardDataSpecialization
 {
     Default,
     Filter1x1Stride1Pad0,

include/ck/tensor_operation/gpu/device/convolution_forward_specialization.hpp

 #ifndef CONVOLUTION_FORWARD_SPECIALIZATION
 #define CONVOLUTION_FORWARD_SPECIALIZATION
 
+#include <string>
+
 namespace ck {
 namespace tensor_operation {
 namespace device {
 
-enum ConvolutionForwardSpecialization_t
+enum struct ConvolutionForwardSpecialization
 {
     Default,
     Filter1x1Pad0,
@@ -13,6 +15,18 @@ enum ConvolutionForwardSpecialization_t
     OddC,
 };
 
+inline std::string getConvFwdSpecializationStr(const ConvolutionForwardSpecialization& s)
+{
+    switch(s)
+    {
+    case ConvolutionForwardSpecialization::Default: return "Default";
+    case ConvolutionForwardSpecialization::Filter1x1Pad0: return "Filter1x1Pad0";
+    case ConvolutionForwardSpecialization::Filter1x1Stride1Pad0: return "Filter1x1Stride1Pad0";
+    case ConvolutionForwardSpecialization::OddC: return "OddC";
+    default: return "Unrecognized specialization!";
+    }
+}
+
 } // namespace device
 } // namespace tensor_operation
 } // namespace ck

include/ck/tensor_operation/gpu/device/device_conv_backward_weight.hpp

@@ -11,7 +11,7 @@ namespace device {
 template <typename InElementwiseOperation,
           typename WeiElementwiseOperation,
           typename OutElementwiseOperation>
-struct DeviceConvWrw : public BaseOperator
+struct DeviceConvBwdWeight : public BaseOperator
 {
     virtual std::unique_ptr<BaseArgument>
     MakeArgumentPointer(const void* p_in,
@@ -38,8 +38,8 @@ struct DeviceConvWrw : public BaseOperator
 template <typename InElementwiseOperation,
           typename WeiElementwiseOperation,
           typename OutElementwiseOperation>
-using DeviceConvWrwPtr = std::unique_ptr<
-    DeviceConvWrw<InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation>>;
+using DeviceConvBwdWeightPtr = std::unique_ptr<
+    DeviceConvBwdWeight<InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation>>;
 
 } // namespace device
 } // namespace tensor_operation