warp.CaptureMode

class warp.CaptureMode(*values)

CUDA stream capture mode; mirrors cudaStreamCaptureMode.

Controls how strictly CUDA rejects capture-unsafe runtime APIs while a capture is active. RELAXED is typically required when composing with libraries that perform lazy / capture-unsafe runtime calls (e.g. context initialization) during the capture.

The default mode used by capture_begin() and ScopedCapture is THREAD_LOCAL, which preserves the historical Warp behavior.

__init__(*args, **kwds)

Methods

conjugate	Returns self, the complex conjugate of any int.
bit_length()	Number of bits necessary to represent self in binary.
bit_count()	Number of ones in the binary representation of the absolute value of self.
to_bytes([length, byteorder, signed])	Return an array of bytes representing an integer.
from_bytes(bytes[, byteorder, signed])
as_integer_ratio()	Return a pair of integers, whose ratio is equal to the original int.
is_integer()	Returns True.
__init__(*args, **kwds)

Attributes

real	the real part of a complex number
imag	the imaginary part of a complex number
numerator	the numerator of a rational number in lowest terms
denominator	the denominator of a rational number in lowest terms
GLOBAL	Capture-unsafe runtime APIs called from any thread invalidate the capture.
THREAD_LOCAL	Capture-unsafe runtime APIs called from the capturing thread invalidate the capture, but other threads are unaffected.
RELAXED	Capture-unsafe runtime APIs are tolerated and do not invalidate the capture.

GLOBAL = 0

Capture-unsafe runtime APIs called from any thread invalidate the capture. This is the strictest mode and matches cudaStreamCaptureModeGlobal.

THREAD_LOCAL = 1

Capture-unsafe runtime APIs called from the capturing thread invalidate the capture, but other threads are unaffected. Matches cudaStreamCaptureModeThreadLocal. This is the default Warp uses.

RELAXED = 2

Capture-unsafe runtime APIs are tolerated and do not invalidate the capture. Matches cudaStreamCaptureModeRelaxed. Useful when composing with libraries that still perform lazy / capture-unsafe CUDA runtime calls (e.g. cudaFree(0) during lazy context / allocator init) during the capture.