nvtx Python package

Installation

Install using conda:

conda install -c conda-forge nvtx

...or using pip:

python -m pip install nvtx

Quick start

1. Annotate the parts of your Python code you wish to profile/analyze:

   # demo.py
   
   import time
   import nvtx
   
   @nvtx.annotate(color="blue")
   def my_function():
       for i in range(5):
           with nvtx.annotate("my_loop", color="red"):
               time.sleep(i)
   
   my_function()
   

2. Use NVIDIA Nsight Systems to collect profile data:

   nsys profile -t nvtx python demo.py
   

3. Visualize the results using the Nsight systems GUI:

   

Documentation

See here for detailed docs.

Release Notes

0.2.16

• Add support for counter annotations through the NVTX Counters extension.
• Add nvtx.writer for writing pre-collected events and counter samples (NVTXW).
• Drop support for Python 3.8.
• Fix several payload-handling edge cases.

0.2.15

• Extend Domain API to accept keyword arguments for event attributes.
• Implement testing improvements.

0.2.14

• Add support for NumPy arrays as payloads.
• Set up support for free-threaded Python.
• Apply NVTX_DISABLE environment variable to Profile class.

0.2.13

• Fix tracing with tools that do not implement nvtxDomainIsEnabled.
• Fix decorator ranges not ending when an exception is thrown.
• Fix automatic function name annotation for decorator ranges.

0.2.12

• Expose API via domain object for better performance.
• No-OP when domain is disabled.
• Improved documentation.