warp.ScopedTimer#
- class warp.ScopedTimer(
- name,
- active=True,
- print=True,
- detailed=False,
- dict=None,
- use_nvtx=False,
- color='rapids',
- synchronize=False,
- cuda_filter=0,
- report_func=None,
- skip_tape=False,
Context manager for timing code blocks.
This context manager measures the execution time of the enclosed code block and can optionally print the elapsed time, collect detailed profiling data, or record NVTX ranges.
- Parameters:
- __init__(
- name,
- active=True,
- print=True,
- detailed=False,
- dict=None,
- use_nvtx=False,
- color='rapids',
- synchronize=False,
- cuda_filter=0,
- report_func=None,
- skip_tape=False,
Context manager object for a timer
- Parameters:
name (str) – Name of timer
active (bool) – Enables this timer
print (bool) – At context manager exit, print elapsed time to
sys.stdoutdetailed (bool) – Collects additional profiling data using cProfile and calls
print_stats()at context exitdict (dict[str, list[float]] | None) – A dictionary of lists to which the elapsed time will be appended using
nameas a keyuse_nvtx (bool) – If true, timing functionality is replaced by an NVTX range
color (int | str) – ARGB value (e.g. 0x00FFFF) or color name (e.g. ‘cyan’) associated with the NVTX range
synchronize (bool) – Synchronize the CPU thread with any outstanding CUDA work to return accurate GPU timings
cuda_filter (int) – Filter flags for CUDA activity timing, e.g.
warp.TIMING_KERNELorwarp.TIMING_ALLreport_func (Callable[[list[TimingResult], str], None] | None) – A callback function to print the activity report. If
None,warp.timing_print()will be used.skip_tape (bool) – If true, the timer will not be recorded in the tape
- timing_results#
The list of activity timing results, if collection was requested using
cuda_filter- Type:
Methods
__init__(name[, active, print, detailed, ...])Context manager object for a timer
Attributes
- enabled = True#