Skip to content

[SYCL] Optimize kernel cache lock #17613

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 2 commits into from
Mar 25, 2025
Merged

[SYCL] Optimize kernel cache lock #17613

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
d4718d7
Add try_lock method to the sycl::_V1::detail::SpinLock
vinser52 Mar 21, 2025
60eba4a
Use spinlock in the KernelProgramCache instead of std::mutex
vinser52 Mar 21, 2025
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
2 changes: 2 additions & 0 deletions sycl/include/sycl/detail/spinlock.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -26,6 +26,8 @@ namespace detail {
/// std::mutex, that doesn't provide such guarantees).
class SpinLock {
public:
bool try_lock() { return !MLock.test_and_set(std::memory_order_acquire); }

void lock() {
while (MLock.test_and_set(std::memory_order_acquire))
std::this_thread::yield();
Expand Down
14 changes: 9 additions & 5 deletions sycl/source/detail/kernel_program_cache.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -15,6 +15,7 @@
#include <sycl/detail/common.hpp>
#include <sycl/detail/locked.hpp>
#include <sycl/detail/os_util.hpp>
#include <sycl/detail/spinlock.hpp>
#include <sycl/detail/ur.hpp>
#include <sycl/detail/util.hpp>

Expand Down Expand Up @@ -421,7 +422,7 @@ class KernelProgramCache {

template <typename KeyT>
KernelFastCacheValT tryToGetKernelFast(KeyT &&CacheKey) {
std::unique_lock<std::mutex> Lock(MKernelFastCacheMutex);
KernelFastCacheReadLockT Lock(MKernelFastCacheMutex);
auto It = MKernelFastCache.find(CacheKey);
if (It != MKernelFastCache.end()) {
traceKernel("Kernel fetched.", CacheKey.second, true);
Expand All @@ -445,7 +446,7 @@ class KernelProgramCache {
return;
}
// Save reference between the program and the fast cache key.
std::unique_lock<std::mutex> Lock(MKernelFastCacheMutex);
KernelFastCacheWriteLockT Lock(MKernelFastCacheMutex);
MProgramToKernelFastCacheKeyMap[Program].emplace_back(CacheKey);

// if no insertion took place, thus some other thread has already inserted
Expand Down Expand Up @@ -483,7 +484,7 @@ class KernelProgramCache {

{
// Remove corresponding entries from KernelFastCache.
std::unique_lock<std::mutex> Lock(MKernelFastCacheMutex);
KernelFastCacheWriteLockT Lock(MKernelFastCacheMutex);
if (auto FastCacheKeyItr =
MProgramToKernelFastCacheKeyMap.find(NativePrg);
FastCacheKeyItr != MProgramToKernelFastCacheKeyMap.end()) {
Expand Down Expand Up @@ -630,7 +631,7 @@ class KernelProgramCache {
std::lock_guard<std::mutex> EvictionListLock(MProgramEvictionListMutex);
std::lock_guard<std::mutex> L1(MProgramCacheMutex);
std::lock_guard<std::mutex> L2(MKernelsPerProgramCacheMutex);
std::lock_guard<std::mutex> L3(MKernelFastCacheMutex);
KernelFastCacheWriteLockT L3(MKernelFastCacheMutex);
MCachedPrograms = ProgramCache{};
MKernelsPerProgramCache = KernelCacheT{};
MKernelFastCache = KernelFastCacheT{};
Expand Down Expand Up @@ -758,7 +759,10 @@ class KernelProgramCache {
KernelCacheT MKernelsPerProgramCache;
ContextPtr MParentContext;

std::mutex MKernelFastCacheMutex;
using KernelFastCacheMutexT = SpinLock;
using KernelFastCacheReadLockT = std::lock_guard<KernelFastCacheMutexT>;
using KernelFastCacheWriteLockT = std::lock_guard<KernelFastCacheMutexT>;
KernelFastCacheMutexT MKernelFastCacheMutex;
KernelFastCacheT MKernelFastCache;

// Map between fast kernel cache keys and program handle.
Expand Down