Skip to content

Another OpenMP fix. #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.

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

Merged
merged 4 commits into from
Dec 21, 2021
Merged

Another OpenMP fix. #2

Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
b13f384
Revert "Avoid function-local static variables. (#1)"
olupton Dec 21, 2021
70ea71e
Use static cache sizes if OpenMP is enabled.
olupton Dec 21, 2021
d64df86
Restrict OpenMP fix to NVIDIA compilers.
olupton Dec 21, 2021
68ac743
Enable workaround for OpenACC too.
olupton Dec 21, 2021
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
33 changes: 9 additions & 24 deletions Eigen/src/Core/products/GeneralBlockPanelKernel.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -90,29 +90,14 @@ struct CacheSizes {
std::ptrdiff_t m_l3;
};

// In C++17 this could be an inline variable, see for example
// https://stackoverflow.com/questions/38043442/how-do-inline-variables-work
template <typename>
struct CacheSizeGlobalHelper {
static CacheSizes s_cacheSizes;
};
#ifdef _OPENMP
#pragma omp declare target
#endif
template <typename T>
CacheSizes CacheSizeGlobalHelper<T>::s_cacheSizes;
#ifdef _OPENMP
#pragma omp end declare target
#endif

/** \internal */
EIGEN_DEVICE_FUNC
inline void manage_caching_sizes(Action action, std::ptrdiff_t* l1, std::ptrdiff_t* l2, std::ptrdiff_t* l3)
{
#ifdef EIGEN_CUDA_ARCH
#if defined(EIGEN_CUDA_ARCH) || (defined(__NVCOMPILER) && (defined(_OPENMP) || defined(_OPENACC)))
if (action==GetAction)
{
#if EIGEN_CUDA_ARCH >= 700
#if EIGEN_CUDA_ARCH >= 700 || (defined(__NVCOMPILER) && (defined(_OPENMP) || defined(_OPENACC)))
// Volta, Turing, or newer
// - the L1 cache is configurable at runtime, with a minimum of 32 KB/SM
// - the L2 cache depends on the actual card, with a minimum of 64 KB/SM
Expand All @@ -133,22 +118,22 @@ inline void manage_caching_sizes(Action action, std::ptrdiff_t* l1, std::ptrdiff
eigen_internal_assert(false);
}
#else // EIGEN_CUDA_ARCH
static CacheSizes m_cacheSizes;

auto& cacheSizes = CacheSizeGlobalHelper<void>::s_cacheSizes;
if(action==SetAction)
{
// set the cpu cache size and cache all block sizes from a global cache size in byte
eigen_internal_assert(l1!=0 && l2!=0);
cacheSizes.m_l1 = *l1;
cacheSizes.m_l2 = *l2;
cacheSizes.m_l3 = *l3;
m_cacheSizes.m_l1 = *l1;
m_cacheSizes.m_l2 = *l2;
m_cacheSizes.m_l3 = *l3;
}
else if(action==GetAction)
{
eigen_internal_assert(l1!=0 && l2!=0);
*l1 = cacheSizes.m_l1;
*l2 = cacheSizes.m_l2;
*l3 = cacheSizes.m_l3;
*l1 = m_cacheSizes.m_l1;
*l2 = m_cacheSizes.m_l2;
*l3 = m_cacheSizes.m_l3;
}
else
{
Expand Down