Skip to content

Cuda Memcpy Async Tail Chunk Overrun #9602

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.

Sign up for GitHub

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

Open
fbusato wants to merge 1 commit into
base: main
Choose a base branch
from
Open

Cuda Memcpy Async Tail Chunk Overrun #9602

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
36 changes: 18 additions & 18 deletions libcudacxx/include/cuda/__memcpy_async/check_preconditions.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -22,9 +22,11 @@
#endif // no system header

#include <cuda/__memory/aligned_size.h>
#include <cuda/__memory/ranges_overlap.h>
#include <cuda/std/__algorithm/max.h>
#include <cuda/std/__algorithm/min.h>
#include <cuda/std/__cstddef/types.h>
#include <cuda/std/cstdint>
#include <cuda/std/__memory/is_sufficiently_aligned.h>

#include <cuda/std/__cccl/prologue.h>

Expand All @@ -45,31 +47,29 @@ _CCCL_BEGIN_NAMESPACE_CUDA

// Check the memcpy_async preconditions, return value is intended for testing purposes exclusively
template <class _Tp, class _Size>
_CCCL_HOST_DEVICE_API inline bool __memcpy_async_check_pre(_Tp* __dst, const _Tp* __src, _Size __size)
_CCCL_HOST_DEVICE_API inline bool __memcpy_async_check_pre(_Tp* __dst, const _Tp* __src, _Size __size_bytes)
{
constexpr auto __align = ::cuda::std::max(alignof(_Tp), __get_size_align_v<_Size>);

const auto __dst_val = reinterpret_cast<uintptr_t>(__dst);
const auto __src_val = reinterpret_cast<uintptr_t>(__src);

// check src and dst alignment
_LIBCUDACXX_MEMCPY_ASYNC_PRE_ASSERT(
__dst_val % __align == 0, "destination pointer must be aligned to the specified alignment");
_LIBCUDACXX_MEMCPY_ASYNC_PRE_ASSERT(
__src_val % __align == 0, "source pointer must be aligned to the specified alignment");

// check src and dst overlap
constexpr auto __align = ::cuda::std::max(alignof(_Tp), __get_size_align_v<_Size>);
constexpr auto __copy_size = ::cuda::std::min(::cuda::std::size_t{16}, __align);
_LIBCUDACXX_MEMCPY_ASYNC_PRE_ASSERT(__size_bytes % __copy_size == 0, //
"size must be a multiple of the memcpy_async copy chunk size");
_LIBCUDACXX_MEMCPY_ASYNC_PRE_ASSERT(::cuda::std::is_sufficiently_aligned<__align>(__dst),
"destination pointer must be aligned to the specified alignment");
_LIBCUDACXX_MEMCPY_ASYNC_PRE_ASSERT(::cuda::std::is_sufficiently_aligned<__align>(__src), //
"source pointer must be aligned to the specified alignment");
const auto __dst_char = reinterpret_cast<const char*>(__dst);
const auto __src_char = reinterpret_cast<const char*>(__src);
_LIBCUDACXX_MEMCPY_ASYNC_PRE_ASSERT(
!((__dst_val <= __src_val && __src_val < __dst_val + __size)
|| (__src_val <= __dst_val && __dst_val < __src_val + __size)),
!::cuda::ranges_overlap(__dst_char, __dst_char + __size_bytes, __src_char, __src_char + __size_bytes),
"destination and source buffers must not overlap");
return true;
}

template <class _Size>
_CCCL_HOST_DEVICE_API inline bool __memcpy_async_check_pre(void* __dst, const void* __src, _Size __size)
_CCCL_HOST_DEVICE_API inline bool __memcpy_async_check_pre(void* __dst, const void* __src, _Size __size_bytes)
{
return ::cuda::__memcpy_async_check_pre(reinterpret_cast<char*>(__dst), reinterpret_cast<const char*>(__src), __size);
return ::cuda::__memcpy_async_check_pre(
reinterpret_cast<char*>(__dst), reinterpret_cast<const char*>(__src), __size_bytes);
}

_CCCL_END_NAMESPACE_CUDA
Expand Down
Loading