Don't assume host pointers are directly usable on the device. #1342
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Since CUDA.jl 3.4, we now always reconstruct a buffer object when creating an array from a pointer. In the case of a HostBuffer, that means we want the host pointer in the buffer object, which is converted to a device pointer on request. However, unsafe_wrap is invoked with a device pointer. Generally, those pointers are identical, but on GPUs where CU_DEVICE_ATTRIBUTE_CAN_USE_HOST_POINTER_FOR_REGISTERED_MEM==0, that is not the case. Instead, recover the host pointer from the device pointer passed to unsafe_wrap. This is going back and forth between both pointer representations, but those calls are reasonably cheap.
maleadt
force-pushed
the
tb/unsafe_wrap_hostbuffer
branch
from
122f017
to
1236a12
Compare
Codecov Report
@@ Coverage Diff @@
## master #1342 +/- ##
==========================================
- Coverage 78.97% 78.96% -0.01%
==========================================
Files 119 119
Lines 8657 8659 +2
==========================================
+ Hits 6837 6838 +1
- Misses 1820 1821 +1
Continue to review full report at Codecov.
|
maleadt
added a commit
that referenced
this pull request
Jan 27, 2022
Since CUDA.jl 3.4, we now always reconstruct a buffer object when creating an array from a pointer. In the case of a HostBuffer, that means we want the host pointer in the buffer object, which is converted to a device pointer on request. However, unsafe_wrap is invoked with a device pointer. Generally, those pointers are identical, but on GPUs where CU_DEVICE_ATTRIBUTE_CAN_USE_HOST_POINTER_FOR_REGISTERED_MEM==0, that is not the case. Instead, recover the host pointer from the device pointer passed to unsafe_wrap. This is going back and forth between both pointer representations, but those calls are reasonably cheap.
luraess
added a commit
to eth-cscs/ImplicitGlobalGrid.jl
that referenced
this pull request
Jan 27, 2022
Modify CUDA compat to v3 excluding v3.4 - v3.7.0 because of #25, now that bug JuliaGPU/CUDA.jl#1342 is fixed.
simonbyrne
pushed a commit
to simonbyrne/CUDA.jl
that referenced
this pull request
Nov 13, 2023
…PU#1342) Since CUDA.jl 3.4, we now always reconstruct a buffer object when creating an array from a pointer. In the case of a HostBuffer, that means we want the host pointer in the buffer object, which is converted to a device pointer on request. However, unsafe_wrap is invoked with a device pointer. Generally, those pointers are identical, but on GPUs where CU_DEVICE_ATTRIBUTE_CAN_USE_HOST_POINTER_FOR_REGISTERED_MEM==0, that is not the case. Instead, recover the host pointer from the device pointer passed to unsafe_wrap. This is going back and forth between both pointer representations, but those calls are reasonably cheap.
marinlauber
pushed a commit
to marinlauber/ImplicitGlobalGrid.jl
that referenced
this pull request
Jul 10, 2024
Modify CUDA compat to v3 excluding v3.4 - v3.7.0 because of eth-cscs#25, now that bug JuliaGPU/CUDA.jl#1342 is fixed.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Since CUDA.jl 3.4, we now always reconstruct a buffer object when
creating an array from a pointer. In the case of a HostBuffer,
that means we want the host pointer in the buffer object,
which is converted to a device pointer on request.
However, unsafe_wrap is invoked with a device pointer. Generally,
those pointers are identical, but on GPUs where
CU_DEVICE_ATTRIBUTE_CAN_USE_HOST_POINTER_FOR_REGISTERED_MEM==0,
that is not the case.
Instead, recover the host pointer from the device pointer passed to
unsafe_wrap. This is going back and forth between both pointer
representations, but those calls are reasonably cheap.