TransData算子识别

输出推荐消除的TransData算子列表，如：

Total time of TransData task is 292.109000(us), accounted for 22.658301% of the total Tasks.

# TransData算子总时长 292.109000(us)，占所有任务时间占比 22.658301%。

Identifications of TransData operators: Classification of transdata operators

# TransData算子识别：算子分类如下

Type of TransData[Ops_Not_Support_Heavy_Format] accounted for 2.682611% of the total time.

# TransData类型[Ops_Not_Support_Heavy_Format] 占总时间比例 2.682611%。

Op Name: trans_TransData_128; Task Duration(us): 2.604000; Input Formats: NC1HWC0; Output Formats: NHWC;

# 算子名: trans_TransData_128; 持续时间(us): 2.604000; 输入格式: NC1HWC0; 输出格式: NHWC;

Op Name: trans_TransData_178; Task Duration(us): 2.291000; Input Formats: NC1HWC0;

# 算子名: trans_TransData_178; 持续时间(us): 2.291000; 输入格式: NC1HWC0;

Recommendations of Reshape_Ops_Interrupts_Format

# Reshape_Ops_Interrupts_Format的优化建议：

1. Attempt to modify the model to avoid discontinuous operations.

# 在不影响精度的情况下尽量避免非连续的操作。

2. Modify the model and use clone and continuous operations to break the combination of multiple non-consecutive operations.

# 使用clone、contiguous将多个非连续操作的组合断开。

Type of TransData[Transform_Between_5HD_And_NZ] accounted for 0.556567% of the total time.

# TransData类型[Transform_Between_5HD_And_NZ] 占总时间比例 0.556567%。

Op Name: trans_TransData_274; Task Duration(us): 247.130000; Input Formats: NC1HWC0; Output Formats: NHWC;

# 算子名: trans_TransData_274; Task Duration(us): 247.130000; Input Formats: NC1HWC0; Output Formats: NHWC;

Op Name: trans_TransData_440; Task Duration(us): 228.376000; Input Formats: NDC1HWC0; Output Formats: NDHWC;

# 算子名: trans_TransData_440; Task Duration(us): 228.376000; Input Formats: NDC1HWC0; Output Formats: NDHWC;

Recommendations of Transform_Between_5HD_And_NZ

# Transform_Between_5HD_And_NZ的优化建议

1. Attempt to reduce the continuous combination of different classes of heavy operators, for example, Conv2d and BatchMatMul

# 减少不同类型重型算子连续排布，比如，相邻的Conv2d和BatchMatMul。

Recommendations of Other_Transform

# Other_Transform的优化建议

1. It is a reasonable scenario that transdata operation exists, for example, 4D to 5D before Conv2D.

# 这是TransData合理存在的场景，比如在Conv2D算子之前，需要把格式由4D转为5D。