Skip to content

Code review for tryGetSourceBasedKernel #2

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.

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

Conversation

aelovikov-intel
Copy link

No description provided.

Self, std::move(OwnerBundle),
ArgMask, UrProgram, CacheMutex);
Self, OwnerBundle, ArgMask,
UrProgram, CacheMutex);
}
return nullptr;
Copy link
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

If this return is ever returned in future, then we might have double move from the same object. The version on the right avoids that possibility entirely.

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

And what's wrong with double move from the same object? std::move() is just a cast, and OwnerBundle became empty only when move ctor for MKernelBundleImpl is executed. And if the move ctor is executed, we must leave the loop.

Copy link
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

And if the move ctor is executed, we must leave the loop.

Only now. If that logic changes, the moves must be updated and it's very easy to miss that. With this approach this mistake won't be possible.

It's also one less thing to focus on if there is any bug (hypothetically) in that area.

@@ -629,7 +627,7 @@ class kernel_bundle_impl
getSyclObjImpl(DevImg);
if (std::shared_ptr<kernel_impl> PotentialKernelImpl =
// move is performed only when SourceBasedKernel is not null
DevImgImpl->tryGetSourceBasedKernel(Name, MContext, std::move(Self),
DevImgImpl->tryGetSourceBasedKernel(Name, MContext, *this,
Copy link
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

No comment is necessary to explain that there is no double-move here anymore. Also, no "optimization" of moving Self out of the loop anymore (one less thing to explain to the reader).

@@ -964,7 +960,7 @@ class kernel_bundle_impl
getSyclObjImpl(DevImg);
if (std::shared_ptr<kernel_impl> SourceBasedKernel =
// move is performed only when SourceBasedKernel is not null
DevImgImpl->tryGetSourceBasedKernel(Name, MContext, std::move(Self),
DevImgImpl->tryGetSourceBasedKernel(Name, MContext, *this,
Copy link
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Likewise.

@aelovikov-intel aelovikov-intel deleted the code-review branch June 16, 2025 14:25
Alexandr-Konovalov pushed a commit that referenced this pull request Jun 27, 2025
Fixes #123300

What is seen 
```
clang-repl> int x = 42;
clang-repl> auto capture = [&]() { return x * 2; };
In file included from <<< inputs >>>:1:
input_line_4:1:17: error: non-local lambda expression cannot have a capture-default
    1 | auto capture = [&]() { return x * 2; };
      |                 ^
zsh: segmentation fault  clang-repl --Xcc="-v"

(lldb) bt
* thread #1, queue = 'com.apple.main-thread', stop reason = EXC_BAD_ACCESS (code=1, address=0x8)
  * frame #0: 0x0000000107b4f8b8 libclang-cpp.19.1.dylib`clang::IncrementalParser::CleanUpPTU(clang::PartialTranslationUnit&) + 988
    frame #1: 0x0000000107b4f1b4 libclang-cpp.19.1.dylib`clang::IncrementalParser::ParseOrWrapTopLevelDecl() + 416
    frame #2: 0x0000000107b4fb94 libclang-cpp.19.1.dylib`clang::IncrementalParser::Parse(llvm::StringRef) + 612
    frame intel#3: 0x0000000107b52fec libclang-cpp.19.1.dylib`clang::Interpreter::ParseAndExecute(llvm::StringRef, clang::Value*) + 180
    frame intel#4: 0x0000000100003498 clang-repl`main + 3560
    frame intel#5: 0x000000018d39a0e0 dyld`start + 2360
```

Though the error is justified, we shouldn't be interested in exiting
through a segfault in such cases.

The issue is that empty named decls weren't being taken care of
resulting into this assert


https://github.com/llvm/llvm-project/blob/c1a229252617ed58f943bf3f4698bd8204ee0f04/clang/include/clang/AST/DeclarationName.h#L503

Can also be seen when the example is attempted through xeus-cpp-lite.


![image](https://github.com/user-attachments/assets/9b0e6ead-138e-4b06-9ad9-fcb9f8d5bf6e)
Alexandr-Konovalov pushed a commit that referenced this pull request Jun 27, 2025
With non -O0, the call stack is not preserved, like malloc_shared will
be inlined, the call stack would be like
```
 #0 in int* sycl::_V1::malloc_host<int>(unsigned long, sycl::_V1::context const&, sycl::_V1::property_list const&, sycl::_V1::detail::code_location const&) /tmp/syclws/include/sycl/usm.hpp:215:27
 #1 in ?? (/lib/x86_64-linux-gnu/libc.so.6+0x757867a2a1c9)
 #2 in __libc_start_main (/lib/x86_64-linux-gnu/libc.so.6+0x757867a2a28a)
```
instead of
```
 #0 in int* sycl::_V1::malloc_host<int>(unsigned long, sycl::_V1::context const&, sycl::_V1::property_list const&, sycl::_V1::detail::code_location const&) /tmp/syclws/include/sycl/usm.hpp:215:27
 #1 in int* sycl::_V1::malloc_host<int>(unsigned long, sycl::_V1::queue const&, sycl::_V1::property_list const&, sycl::_V1::detail::code_location const&) /tmp/syclws/include/sycl/usm.hpp:223:10
 #2 in main /tmp/syclws/llvm/sycl/test-e2e/MemorySanitizer/track-origins/check_host_usm_initialized_on_host.cpp:15:17
 intel#3 in ?? (/lib/x86_64-linux-gnu/libc.so.6+0x7a67f842a1c9)
 intel#4 in __libc_start_main (/lib/x86_64-linux-gnu/libc.so.6+0x7a67f842a28a)
```

Also, add env to every %{run} directive to make sure they are not
affected by system env.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

2 participants