Add buffer location enum #34
Merged
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
sherry-yuan
force-pushed
the
buffer_location
branch
3 times, most recently
from
630349d
to
6b013eb
Compare
|
This change is almost ready to go, the enum made into public khronos repo. I will make a separate PR to sync upstream headers, and then polish this a little. PR that sync Khronos header is here |
sherry-yuan
force-pushed
the
buffer_location
branch
from
da69878
to
0dbe561
Compare
|
This change is ready to go. |
pcolberg
requested changes
Jan 24, 2022
There was a problem hiding this comment.
Thanks @sherry-yuan, great work!
Could you amend the commit message with links to the corresponding PRs in the OpenCL specification and header repositories?
sherry-yuan
force-pushed
the
buffer_location
branch
from
bf9ef40
to
5f6aec7
Compare
pcolberg
requested changes
Jan 24, 2022
sherry-yuan
force-pushed
the
buffer_location
branch
from
5f6aec7
to
3c2e402
Compare
Sycl runtime calls clEnqueueWriteBuffer before clSetKernelArgs, this cause the buffer to be allocated on the device before knowing the right place of allocating the memory. As a result, later when kernel get invoked, the memory has to be copied from device's default global memory to the buffer location specified in kernel. This is an additional memory copy operation. This extension does not interfer with the other way of setting buffer location (i.e through clSetKernelArgs). This property exist for integration with sycl runtime, not for pure opencl user to use. If opencl user wish to use this property, they have to make sure the buffer location passed into clCreateBufferWithPropertyINTEL has to match the one defined in kernel function interface, otherwise the extra memory copy issue will remain. When resizing reserved allocation, we now have the information to allocate minimum amount of space required according to the property passed in. This is done to align with opencl docs and header in; KhronosGroup/OpenCL-Headers#193 KhronosGroup/OpenCL-Docs#746
sherry-yuan
force-pushed
the
buffer_location
branch
from
3c2e402
to
06324e2
Compare
pcolberg
approved these changes
Jan 24, 2022
# for free
to join this conversation on GitHub.
Already have an account?
# 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.
Sycl runtime calls clEnqueueWriteBuffer before clSetKernelArgs, this cause the buffer to be allocated on the device before knowing the right place of allocating the memory. As a result, later when kernel get invoked, the memory has to be copied from device's default global memory to the buffer location specified in kernel. This is an additional memory copy operation.
This extension does not interfer with the other way of setting buffer location (i.e through clSetKernelArgs). This property exist for integration with sycl runtime, not for pure opencl user to use. If opencl user wish to use this property, they have to make sure the buffer location passed into clCreateBufferWithPropertyINTEL has to match the one defined in kernel function interface, otherwise the extra memory copy issue will remain.
When resizing reserved allocation, we now have the information to allocate minimum amount of space required according to the property passed in.