Skip to content

Ensure NUMBA can manipulate memory from CUDA graphs before the graph is launched #437

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
gmarkall merged 7 commits into from
Oct 8, 2025
Merged

Ensure NUMBA can manipulate memory from CUDA graphs before the graph is launched #437

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
7 commits
Select commit Hold shift + click to select a range
f3af490
Compute alloc_size from info when data is already allocated
caugonnet Aug 28, 2025
66b0af1
Workaround to avoid checking if a pointer is valid
caugonnet Aug 28, 2025
a770da2
take a raw pointer instead of using get_devptr_for_active_ctx
caugonnet Aug 28, 2025
00d629c
revert a previous change
caugonnet Aug 28, 2025
1a7a287
Fix for ctypes binding
gmarkall Aug 28, 2025
327b139
Merge branch 'main' into cuda_graph_future_memory
caugonnet Sep 23, 2025
0ea9004
Merge branch 'NVIDIA:main' into cuda_graph_future_memory
caugonnet Oct 7, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
7 changes: 6 additions & 1 deletion numba_cuda/numba/cuda/api.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -8,6 +8,7 @@
import contextlib
import os

from numba.cuda.cudadrv import drvapi
import numpy as np

from .cudadrv import devicearray, devices, driver
Expand Down Expand Up @@ -47,7 +48,11 @@ def from_cuda_array_interface(desc, owner=None, sync=True):
)
size = driver.memory_size_from_info(shape, strides, dtype.itemsize)

devptr = driver.get_devptr_for_active_ctx(desc["data"][0])
if config.CUDA_USE_NVIDIA_BINDING:
cudevptr_class = driver.binding.CUdeviceptr
else:
cudevptr_class = drvapi.cu_device_ptr
devptr = cudevptr_class(desc["data"][0])
data = driver.MemoryPointer(
current_context(), devptr, size=size, owner=owner
)
Expand Down
8 changes: 3 additions & 5 deletions numba_cuda/numba/cuda/cudadrv/devicearray.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -105,13 +105,11 @@ def __init__(self, shape, strides, dtype, stream=0, gpu_data=None):
self.size = int(functools.reduce(operator.mul, self.shape, 1))
# prepare gpu memory
if self.size > 0:
self.alloc_size = _driver.memory_size_from_info(
self.shape, self.strides, self.dtype.itemsize
)
if gpu_data is None:
self.alloc_size = _driver.memory_size_from_info(
self.shape, self.strides, self.dtype.itemsize
)
gpu_data = devices.get_context().memalloc(self.alloc_size)
else:
self.alloc_size = _driver.device_memory_size(gpu_data)
else:
# Make NULL pointer for empty allocation
if _driver.USE_NV_BINDING:
Expand Down