warp.RegisteredGLBuffer#

class warp.RegisteredGLBuffer(*args, **kwargs)[source]#

Helper class to register a GL buffer with CUDA so that it can be mapped to a Warp array.

Example usage:

import warp as wp
import numpy as np
from pyglet.gl import *

wp.init()

# create a GL buffer
gl_buffer_id = GLuint()
glGenBuffers(1, gl_buffer_id)

# copy some data to the GL buffer
glBindBuffer(GL_ARRAY_BUFFER, gl_buffer_id)
gl_data = np.arange(1024, dtype=np.float32)
glBufferData(GL_ARRAY_BUFFER, gl_data.nbytes, gl_data.ctypes.data, GL_DYNAMIC_DRAW)
glBindBuffer(GL_ARRAY_BUFFER, 0)

# register the GL buffer with CUDA
cuda_gl_buffer = wp.RegisteredGLBuffer(gl_buffer_id)

# map the GL buffer to a Warp array
arr = cuda_gl_buffer.map(dtype=wp.float32, shape=(1024,))
# launch a Warp kernel to manipulate or read the array
wp.launch(my_kernel, dim=1024, inputs=[arr])
# unmap the GL buffer
cuda_gl_buffer.unmap()

__init__( gl_buffer_id, device=None, flags=NONE, fallback_to_copy=True, )[source]#

Parameters:

gl_buffer_id (int) – The OpenGL buffer id (GLuint).
device (Device | str | None) – The device to register the buffer with. If None, the current device will be used.
flags (int) – A combination of the flags constants NONE, READ_ONLY, and WRITE_DISCARD.
fallback_to_copy (bool) – If True and CUDA/OpenGL interop is not available, fall back to copy operations between the Warp array and the OpenGL buffer. Otherwise, a RuntimeError will be raised.

Note

The fallback_to_copy option (to use copy operations if CUDA graphics interop functionality is not available) requires pyglet version 2.0 or later. Install via pip install pyglet==2.*.

Methods

`__init__`(gl_buffer_id[, device, flags, ...])
`map`(dtype, shape)	Map the OpenGL buffer to a Warp array.
`unmap`()	Unmap the OpenGL buffer.

Attributes

`NONE`	Flag that specifies no hints about how this resource will be used.
`READ_ONLY`	Flag that specifies that CUDA will not write to this resource.
`WRITE_DISCARD`	Flag that specifies that CUDA will not read from this resource and will write over the entire contents of the resource, so none of the data previously stored in the resource will be preserved.

NONE = 0#: Flag that specifies no hints about how this resource will be used. It is therefore assumed that this resource will be read from and written to by CUDA. This is the default value.

READ_ONLY = 1#: Flag that specifies that CUDA will not write to this resource.

WRITE_DISCARD = 2#: Flag that specifies that CUDA will not read from this resource and will write over the entire contents of the resource, so none of the data previously stored in the resource will be preserved.

map(dtype, shape)[source]#

Map the OpenGL buffer to a Warp array.

Parameters:

dtype – The type of each element in the array.
shape – The shape of the array.

Returns:

A Warp array object representing the mapped OpenGL buffer.

Return type:

array

unmap()[source]#: Unmap the OpenGL buffer.