[SYCL][NVPTX] Split max_work_group_size into 3 NVVM annotations #14420
Conversation
This file contains hidden or 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
NVVM IR supports separated maxntidx, maxntidy, and maxntidz annotations.
The backend will print them individually as three dimensions. This
better preserves programmer intent than prematurely flattening them
together.
Note that the semantics are in fact identical; the CUDA implementation
internally multiplies all dimensions together and only guarantees that
the total is never exceeded, but not that any individual dimension is
not exceeded. Thus 64,1,1 is identical to 4,4,4.
We try and preserve a logical mapping of dimensions by index flipping
between SYCL (z,y,x) and NVVM (x,y,z) in CUDA terminology despite, as
mentioned above, it being largely irrelevant.
Also this patch simplifies the attribute's getter functions as all
dimensions are mandatory, and the getters seemed copied from the
reqd_work_group_size attribute where some are optional.
We could probably improve the code further by making the operands
"unsigned" and not "Expr", and renaming them from X,Y,Z to Dim{0,1,2} as
per the SYCL spec. This has been left for future work, however, as
there's a non-trivial amount of code that expects to be able to treat
the max_work_group_size and reqd_work_group_size attributes identically
through templates and identical helper methods.
frasercrmck
had a problem deploying
to
WindowsCILock
GitHub Actions
Failure
steffenlarsen
approved these changes
Jul 3, 2024
frasercrmck
had a problem deploying
to
WindowsCILock
GitHub Actions
Failure
|
@intel/llvm-gatekeepers this is good to merge, thank you! |
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Outdated suggestions cannot be applied.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.
NVVM IR supports separated maxntidx, maxntidy, and maxntidz annotations. The backend will print them individually as three dimensions. This better preserves programmer intent than prematurely flattening them together.
Note that the semantics are in fact identical; the CUDA implementation internally multiplies all dimensions together and only guarantees that the total is never exceeded, but not that any individual dimension is not exceeded. Thus 64,1,1 is identical to 4,4,4.
We try and preserve a logical mapping of dimensions by index flipping between SYCL (z,y,x) and NVVM (x,y,z) in CUDA terminology despite, as mentioned above, it being largely irrelevant.
Also this patch simplifies the attribute's getter functions as all dimensions are mandatory, and the getters seemed copied from the reqd_work_group_size attribute where some are optional.
We could probably improve the code further by making the operands "unsigned" and not "Expr", and renaming them from X,Y,Z to Dim{0,1,2} as per the SYCL spec. This has been left for future work, however, as there's a non-trivial amount of code that expects to be able to treat the max_work_group_size and reqd_work_group_size attributes identically through templates and identical helper methods.