Extend address stability detection for thrust::transform
#2404
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
* Introduces address stability detection and opt-in in libcu++ * Mark lambdas in Thrust BabelStream benchmark address oblivious Fixes: NVIDIA#2263
miscco
reviewed
Sep 16, 2024
| It iterator; | ||
|
|
||
| _CCCL_HOST_DEVICE kernel_arg() {} // in case It is not default-constructible | ||
| _CCCL_HOST_DEVICE kernel_arg() {} // needed in case It is not default-constructible |
There was a problem hiding this comment.
Should this explicitly construct aligned_ptr?
| // since we switch the active member of the union, we have to use placement new. This also uses the copy constructor | ||
| // of It, which works in more cases than assignment (e.g. thrust::transform_iterator with non-copy-assignable functor, | ||
| // e.g. in merge sort tests) | ||
| new (&arg.iterator) It(it); |
There was a problem hiding this comment.
This might want to use cuda::std::__construct_at
Comment on lines
+10
to
+11
| #ifndef _LIBCUDACXX___TYPE_TRAITS_ADDRESS_STABILITY_H | ||
| #define _LIBCUDACXX___TYPE_TRAITS_ADDRESS_STABILITY_H |
There was a problem hiding this comment.
This is not in cuda::std so that should rather be:
Suggested change
| #ifndef _LIBCUDACXX___TYPE_TRAITS_ADDRESS_STABILITY_H | |
| #define _LIBCUDACXX___TYPE_TRAITS_ADDRESS_STABILITY_H | |
| #ifndef _CUDA___FUNCTIONAL_ADDRESS_STABILITY_H | |
| #define _CUDA___FUNCTIONAL_ADDRESS_STABILITY_H |
Comment on lines
+42
to
+45
| template <typename R, typename... Args> | ||
| struct __all_parameters_by_value_fptr<R (*)(Args...)> | ||
| : _CUDA_VSTD::conjunction<_CUDA_VSTD::_Not<_CUDA_VSTD::is_reference<Args>>...> | ||
| {}; |
There was a problem hiding this comment.
This could be implemented more efficiently with
Suggested change
| template <typename R, typename... Args> | |
| struct __all_parameters_by_value_fptr<R (*)(Args...)> | |
| : _CUDA_VSTD::conjunction<_CUDA_VSTD::_Not<_CUDA_VSTD::is_reference<Args>>...> | |
| {}; | |
| template <typename R, typename... Args> | |
| struct __all_parameters_by_value_fptr<R (*)(Args...)> | |
| : _CUDA_VSTD::__all_of<!_CCCL_TRAIT(is_reference, _Args)...> | |
| {}; |
There was a problem hiding this comment.
I see that that is C++14 only so we could use the equivalent thing from sfinae_helpers
template <bool... _Preds>
struct __all_dummy;
template <bool... _Pred>
using __all = is_same<__all_dummy<_Pred...>, __all_dummy<((void) _Pred, true)...>>;
Comment on lines
+36
to
+37
| #include <cuda/__functional/address_stability.h> | ||
| #include <cuda/functional> |
There was a problem hiding this comment.
In benchmaks we conventionally only include the public header
Suggested change
| #include <cuda/__functional/address_stability.h> | |
| #include <cuda/functional> | |
| #include <cuda/functional> |
| @@ -0,0 +1,198 @@ | |||
| #include <cuda/__functional/address_stability.h> | |||
There was a problem hiding this comment.
Suggested change
| #include <cuda/__functional/address_stability.h> | |
| #include <cuda/functional> |
Comment on lines
+426
to
+429
| // The following test cannot be compiled because of a bug in the conversion of thrust::tuple on MSVC 2017 | ||
| #ifndef _CCCL_COMPILER_MSVC_2017 | ||
| // we specialize zip_function for sum_five, but do nothing in the call operator so the test below would fail if the | ||
| // zip_function is actually called (and not unwrapped) |
There was a problem hiding this comment.
Please recheck now that we completely replaced that with cuda::std::tuple
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.
This PR extendes the automatic detection of function objects for which arguments can savely be copied before passing them to the function object.
Fixes: #2403
Merge before:
thrust::transformto usecub::DeviceTransform#2263