TorchAir C++层日志

功能简介

TorchAir C++层日志开启主要通过环境变量TNG_LOG_LEVEL实现，方便进行C++层功能调试和问题定位。

支持的日志级别如下：

• TNG_LOG_LEVEL：0，日志级别DEBUG，开启后输出DEBUG、INFO、WARNING、ERROR日志。
• TNG_LOG_LEVEL：1，日志级别INFO，开启后输出INFO、WARNING、ERROR日志。
• TNG_LOG_LEVEL：2，日志级别WARNING，开启后输出WARNING、ERROR日志。
• TNG_LOG_LEVEL：3，日志级别ERROR，开启后输出ERROR日志。

环境变量TNG_LOG_LEVEL的默认值为“3”。

使用方法

在shell或python脚本中，提前配置环境变量TNG_LOG_LEVEL。

• 在shell环境或者shell脚本

  export TNG_LOG_LEVEL=0

• python脚本
  

  该方式设置环境变量时，需早于import torchair，否则影响日志显示。

  import os
  os.environ['TNG_LOG_LEVEL'] = '0'

C++侧debug日志样例如下

[DEBUG] TORCHAIR [static_npu_graph_executor.cpp:37] Assemble aten device input 0 at::Tensor(shape=[4, 1], dtype='float', device=npu:1, addr=0x124180000000) to ge::Tensor(shape=[4, 1], format=ND, dtype=DT_FLOAT, device=NPU, addr=0x124180000000)
[DEBUG] TORCHAIR [static_npu_graph_executor.cpp:37] Assemble aten device input 1 at::Tensor(shape=[4, 4], dtype='float', device=npu:1, addr=0x124180000200) to ge::Tensor(shape=[4, 4], format=ND, dtype=DT_FLOAT, device=NPU, addr=0x124180000200)
[DEBUG] TORCHAIR [static_npu_graph_executor.cpp:37] Assemble aten device input 2 at::Tensor(shape=[4, 4], dtype='int', device=npu:1, addr=0x124180000400) to ge::Tensor(shape=[4, 4], format=ND, dtype=DT_INT32, device=NPU, addr=0x124180000400)
[DEBUG] TORCHAIR [static_npu_graph_executor.cpp:96] Create empty output 0 at::Tensor(shape=[4, 4], dtype='float', device=npu:1, addr=0x124180000800)
[DEBUG] TORCHAIR [static_npu_graph_executor.cpp:104] Assemble torch output 0 at::Tensor(shape=[4, 4], dtype='float', device=npu:1, addr=0x124180000800) to ge::Tensor(shape=[4, 4], format=ND, dtype=DT_FLOAT, device=NPU, addr=0x124180000800)
[INFO] TORCHAIR [AllocatorManager.cpp:11] Start to EnsureAllocatorRegistered according to stream = 0x228dd940
[INFO] TORCHAIR [AllocatorManager.cpp:19] External allocator did not registered, register allocator = 0x251a6ec0 while stream = 0x228dd940
[DEBUG] TORCHAIR [memory/Allocator.cpp:15] [MemoryTrace] Malloc memory from NPUCachingAllocator success, block = 0x234855e0
[DEBUG] TORCHAIR [memory/Allocator.cpp:25] [MemoryTrace] Malloc the mem_block success, mem_block = 0x23282c80, device_ptr = 0x124180000a00, size = 1536
[INFO] TORCHAIR [AllocatorManager.cpp:11] Start to EnsureAllocatorRegistered according to stream = 0x228dd940
[INFO] TORCHAIR [AllocatorManager.cpp:15] External allocator has registered, stream = 0x228dd940 , allocator = 0x251a6ec0
[INFO] TORCHAIR [static_npu_graph_executor.cpp:145] Try to alloc and update feature map memory, graph id = 0 , size = 512
[INFO] TORCHAIR [memory/Allocator.cpp:80] [MemoryTrace] MallocFeatureMemory: Try Malloc size = 512
[INFO] TORCHAIR [memory/Allocator.cpp:65] [MemoryTrace] MallocPoolMemory: Try Malloc size = 512
[DEBUG] TORCHAIR [memory/Allocator.cpp:15] [MemoryTrace] Malloc memory from NPUCachingAllocator success, block = 0x234856f0
[DEBUG] TORCHAIR [memory/Allocator.cpp:25] [MemoryTrace] Malloc the mem_block success, mem_block = 0x23283a80, device_ptr = 0x124180001000, size = 1024
[INFO] TORCHAIR [memory/Allocator.cpp:71] [MemoryTrace] MallocPoolMemory: Malloc memory success, size = 1024, and addr = 0x124180001000, use count = 1
[INFO] TORCHAIR [static_npu_graph_executor.cpp:160] AllocAndUpdateFeatureMemory success, feature map addr = 0x124180001000 , size = 1024
[INFO] TORCHAIR [static_npu_graph_executor.cpp:212] StaticNpuGraphExecutor::Run graph 0 on stream 0x228dd940 successfully