[SYCL] Allow free function kernel args be templated on integer expressions #20187
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
…ions `constexpr` variables are not forward-declarable so if one is used as a template parameter of a free function kernel argument, we cannot reference the variable, but must inline the value into the integration header.
AlexeySachkov
temporarily deployed
to
WindowsCILock
GitHub Actions
Inactive
AlexeySachkov
commented
Sep 24, 2025
| [[__sycl_detail__::add_ir_attributes_function("sycl-nd-range-kernel", 2)]] | ||
| void constexpr_v(Arg<A>) {} | ||
|
|
||
| // CHECK: void constexpr_v(Arg<2> ); |
There was a problem hiding this comment.
Note for reviewers: we used to print Arg<A> here where A isn't forward-declared in the integration header (because we can't do for constexpr variables) which caused compilation errors.
There was a problem hiding this comment.
What about putting constexpr function calls there?
There was a problem hiding this comment.
Added such test in e502a1b
AlexeySachkov
temporarily deployed
to
WindowsCILock
GitHub Actions
Inactive
AlexeySachkov
temporarily deployed
to
WindowsCILock
GitHub Actions
Inactive
Fznamznon
reviewed
Sep 24, 2025
|
|
||
| auto TemplateArgPrinter = [&](const TemplateArgument &Arg) { | ||
| if (Arg.getKind() != TemplateArgument::ArgKind::Expression || | ||
| Arg.isInstantiationDependent()) { |
There was a problem hiding this comment.
Can Arg.isInstantiationDependent() ever be true in SemaSYCL?
There was a problem hiding this comment.
Yes, it can. This extra check was prompted by
template <int ArrSize>
[[__sycl_detail__::add_ir_attributes_function("sycl-single-task-kernel", 0)]]
void ff_7(KArgWithPtrArray<ArrSize> KArg) {
for (int j = 0; j < ArrSize; j++)
for (int i = KArg.start[j]; i <= KArg.end[j]; i++)
KArg.data[j][i] = KArg.start[j] + KArg.end[j];
} (snippet from clang/test/CodeGenSYCL/free_function_int_header.cpp).
ArrSize template argument is instantiation dependent here
| return; | ||
| } | ||
|
|
||
| Expr *E = Arg.getAsExpr(); |
There was a problem hiding this comment.
Can E end up being null?
There was a problem hiding this comment.
It shouldn't be because I check the argument kind above, but I added an assertion just in case (and to help static analyzers)
| [[__sycl_detail__::add_ir_attributes_function("sycl-nd-range-kernel", 2)]] | ||
| void constexpr_v(Arg<A>) {} | ||
|
|
||
| // CHECK: void constexpr_v(Arg<2> ); |
There was a problem hiding this comment.
What about putting constexpr function calls there?
AlexeySachkov
temporarily deployed
to
WindowsCILock
GitHub Actions
Inactive
AlexeySachkov
temporarily deployed
to
WindowsCILock
GitHub Actions
Inactive
AlexeySachkov
temporarily deployed
to
WindowsCILock
GitHub Actions
Inactive
Fznamznon
approved these changes
Sep 24, 2025
|
@intel/llvm-gatekeepers please consider merging |
AlexeySachkov
deleted the
private/asachkov/one-more-free-function-kernels-fix
branch
AlexeySachkov
added a commit
to AlexeySachkov/llvm
that referenced
this pull request
Sep 29, 2025
…sions (intel#20187) `constexpr` variables are not forward-declarable so if one is used as a template parameter of a free function kernel argument, we cannot reference the variable, but must inline the value into the integration header.
AlexeySachkov
added a commit
that referenced
this pull request
Sep 30, 2025
This is joined cherry-pick of #20187 and #20123 --- [SYCL] Allow free function kernel args be templated on integer expressions (#20187) `constexpr` variables are not forward-declarable so if one is used as a template parameter of a free function kernel argument, we cannot reference the variable, but must inline the value into the integration header. --- [SYCL] Fix error with type aliases used as free function kernel args (#20123) This PR fixes type name that is being printed as free function kernel argument type in its forward-declaration in the integration header. Before the change, we used the original argument type name, which could be an alias - this patch makes use of the canonical type's name to make sure that all type aliases are "unwrapped".
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.
constexprvariables are not forward-declarable so if one is used as a template parameter of a free function kernel argument, we cannot reference the variable, but must inline the value into the integration header.