Atomics backend refactor #1631
Merged
Atomics backend refactor #1631
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
wmaxey
force-pushed
the
fea/atomic_refactor_simplify
branch
from
e085e18 to
a780c26
Compare
* This matches other implementations.
miscco
reviewed
Apr 29, 2024
* We previously defined or *clobbered* the existing STL definitions. * See: `ATOMIC_BOOL_LOCK_FREE`->`LIBCUDACXX_ATOMIC_BOOL_LOCK_FREE`
Co-authored-by: Michael Schellenberger Costa <[email protected]>
wmaxey
commented
May 9, 2024
Comment on lines
+41
to
+102
| template <typename _Tp> | ||
| __atomic_alignment_wrapper<__remove_cv_t<_Tp>>& __atomic_auto_align(_Tp* __a) | ||
| { | ||
| using __aligned_t = __atomic_alignment_wrapper<__remove_cv_t<_Tp>>; | ||
| return *reinterpret_cast<__aligned_t*>(__a); | ||
| }; | ||
| template <typename _Tp> | ||
| const __atomic_alignment_wrapper<__remove_cv_t<_Tp>>& __atomic_auto_align(const _Tp* __a) | ||
| { | ||
| using __aligned_t = const __atomic_alignment_wrapper<__remove_cv_t<_Tp>>; | ||
| return *reinterpret_cast<__aligned_t*>(__a); | ||
| }; | ||
| template <typename _Tp> | ||
| volatile __atomic_alignment_wrapper<__remove_cv_t<_Tp>>& __atomic_auto_align(volatile _Tp* __a) | ||
| { | ||
| using __aligned_t = volatile __atomic_alignment_wrapper<__remove_cv_t<_Tp>>; | ||
| return *reinterpret_cast<__aligned_t*>(__a); | ||
| }; | ||
| template <typename _Tp> | ||
| const volatile __atomic_alignment_wrapper<__remove_cv_t<_Tp>>& __atomic_auto_align(const volatile _Tp* __a) | ||
| { | ||
| using __aligned_t = const volatile __atomic_alignment_wrapper<__remove_cv_t<_Tp>>; | ||
| return *reinterpret_cast<__aligned_t*>(__a); | ||
| }; | ||
|
|
||
| // Guard ifdef for lock free query in case it is assigned elsewhere (MSVC/CUDA) | ||
| inline void __atomic_thread_fence_host(memory_order __order) | ||
| { | ||
| __atomic_thread_fence(__atomic_order_to_int(__order)); | ||
| } | ||
|
|
||
| inline void __atomic_signal_fence_host(memory_order __order) | ||
| { | ||
| __atomic_signal_fence(__atomic_order_to_int(__order)); | ||
| } | ||
|
|
||
| template <typename _Tp, typename _Up> | ||
| inline void __atomic_store_host(_Tp* __a, _Up __val, memory_order __order) | ||
| { | ||
| __atomic_store( | ||
| &__atomic_auto_align<_Tp>(__a), &__atomic_auto_align<__remove_cv_t<_Tp>>(&__val), __atomic_order_to_int(__order)); | ||
| } | ||
|
|
||
| template <typename _Tp> | ||
| inline auto __atomic_load_host(_Tp* __a, memory_order __order) -> __remove_cv_t<_Tp> | ||
| { | ||
| __remove_cv_t<_Tp> __ret; | ||
| __atomic_load( | ||
| &__atomic_auto_align<_Tp>(__a), &__atomic_auto_align<__remove_cv_t<_Tp>>(&__ret), __atomic_order_to_int(__order)); | ||
| return __ret; | ||
| } | ||
|
|
||
| template <typename _Tp, typename _Up> | ||
| inline auto __atomic_exchange_host(_Tp* __a, _Up __val, memory_order __order) -> __remove_cv_t<_Tp> | ||
| { | ||
| __remove_cv_t<_Tp> __ret; | ||
| __atomic_exchange(&__atomic_auto_align<_Tp>(__a), | ||
| &__atomic_auto_align<__remove_cv_t<_Tp>>(&__val), | ||
| &__atomic_auto_align<__remove_cv_t<_Tp>>(&__ret), | ||
| __atomic_order_to_int(__order)); | ||
| return __ret; | ||
| } |
There was a problem hiding this comment.
@miscco @griwes
Time for excuses: __atomic_auto_align why did I write this?
Because alignment information is lost as we descend into the abyss. And if we continued to link we'd end up requiring libatomic. There could be a better way to write this. Any ideas?
https://gcc.godbolt.org/z/vdxffdnKc
This is where all that wrap/unwrap nastyness came from as well. It was just more or less poorly written when a first pass was made.
🟨 CI Results [ Failed: 21 | Passed: 281 | Total: 302 ]
|
| # | Runner |
|---|---|
| 232 | linux-amd64-cpu16 |
| 28 | linux-amd64-gpu-v100-latest-1 |
| 24 | linux-arm64-cpu16 |
| 18 | windows-amd64-cpu16 |
👃 Inspect Changes
Modifications in project?
| Project | |
|---|---|
| CCCL Infrastructure | |
| +/- | libcu++ |
| CUB | |
| Thrust |
Modifications in project or dependencies?
| Project | |
|---|---|
| CCCL Infrastructure | |
| +/- | libcu++ |
| +/- | CUB |
| +/- | Thrust |
🟨 CI Results [ Failed: 20 | Passed: 282 | Total: 302 ]
|
| # | Runner |
|---|---|
| 232 | linux-amd64-cpu16 |
| 28 | linux-amd64-gpu-v100-latest-1 |
| 24 | linux-arm64-cpu16 |
| 18 | windows-amd64-cpu16 |
👃 Inspect Changes
Modifications in project?
| Project | |
|---|---|
| CCCL Infrastructure | |
| +/- | libcu++ |
| CUB | |
| Thrust |
Modifications in project or dependencies?
| Project | |
|---|---|
| CCCL Infrastructure | |
| +/- | libcu++ |
| +/- | CUB |
| +/- | Thrust |
🟨 CI Results [ Failed: 2 | Passed: 300 | Total: 302 ]
|
| # | Runner |
|---|---|
| 232 | linux-amd64-cpu16 |
| 28 | linux-amd64-gpu-v100-latest-1 |
| 24 | linux-arm64-cpu16 |
| 18 | windows-amd64-cpu16 |
👃 Inspect Changes
Modifications in project?
| Project | |
|---|---|
| CCCL Infrastructure | |
| +/- | libcu++ |
| CUB | |
| Thrust |
Modifications in project or dependencies?
| Project | |
|---|---|
| CCCL Infrastructure | |
| +/- | libcu++ |
| +/- | CUB |
| +/- | Thrust |
|
pre-commit.ci autofix |
|
/ok to test |
🟩 CI Results [ Failed: 0 | Passed: 302 | Total: 302 ]
|
| # | Runner |
|---|---|
| 232 | linux-amd64-cpu16 |
| 28 | linux-amd64-gpu-v100-latest-1 |
| 24 | linux-arm64-cpu16 |
| 18 | windows-amd64-cpu16 |
👃 Inspect Changes
Modifications in project?
| Project | |
|---|---|
| CCCL Infrastructure | |
| +/- | libcu++ |
| CUB | |
| Thrust |
Modifications in project or dependencies?
| Project | |
|---|---|
| CCCL Infrastructure | |
| +/- | libcu++ |
| +/- | CUB |
| +/- | Thrust |
j3soon
added a commit
to j3soon/cccl
that referenced
this pull request
Dec 3, 2024
The `trie_mt.cpp` example doesn't use libcu++, so `LIBCUDACXX_ATOMIC_VAR_INIT` is undeclared:
/root/cccl/libcudacxx/examples/trie_mt.cpp:39:30: error: 'LIBCUDACXX_ATOMIC_VAR_INIT' was not declared in this scope
39 | std::atomic<trie*> ptr = LIBCUDACXX_ATOMIC_VAR_INIT(nullptr);
| ^~~~~~~~~~~~~~~~~~~~~~~~~~
/root/cccl/libcudacxx/examples/trie_mt.cpp:41:29: error: 'LIBCUDACXX_ATOMIC_VAR_INIT' was not declared in this scope
41 | std::atomic_flag flag = LIBCUDACXX_ATOMIC_VAR_INIT(0);
| ^~~~~~~~~~~~~~~~~~~~~~~~~~
/root/cccl/libcudacxx/examples/trie_mt.cpp:43:28: error: 'LIBCUDACXX_ATOMIC_VAR_INIT' was not declared in this scope
43 | std::atomic<int> count = LIBCUDACXX_ATOMIC_VAR_INIT(0);
| ^~~~~~~~~~~~~~~~~~~~~~~~~~
gmake[2]: *** [CMakeFiles/trie_mt.dir/build.make:76: CMakeFiles/trie_mt.dir/trie_mt.cpp.o] Error 1
gmake[1]: *** [CMakeFiles/Makefile2:117: CMakeFiles/trie_mt.dir/all] Error 2
gmake: *** [Makefile:91: all] Error 2
This issue is introduced in PR NVIDIA#1631
and commit 12c2892
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.
Description
closes: #1615
Codegen before and after changes is identical. Though the next step is to compare with additional FileCheck tests.
Checklist