Skip to content

[SYCL] Remove build options from fast kernel cache key #16101

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 1 commit into from
Nov 18, 2024
Merged

[SYCL] Remove build options from fast kernel cache key #16101

Show file tree
Hide file tree
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
19 changes: 14 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 @@ -163,10 +163,19 @@ class KernelProgramCache {
::boost::unordered_map<ur_program_handle_t, KernelByNameT>;

using KernelFastCacheKeyT =
std::tuple<SerializedObj, ur_device_handle_t, std::string, std::string>;
std::tuple<SerializedObj, /* Serialized spec constants. */
ur_device_handle_t, /* UR device handle pointer */
std::string /* Kernel Name */
>;

using KernelFastCacheValT =
std::tuple<ur_kernel_handle_t, std::mutex *, const KernelArgMask *,
ur_program_handle_t>;
std::tuple<ur_kernel_handle_t, /* UR kernel handle pointer. */
std::mutex *, /* Mutex guarding this kernel. */
const KernelArgMask *, /* Eliminated kernel argument mask. */
ur_program_handle_t /* UR program handle corresponding to this
kernel. */
>;

// This container is used as a fast path for retrieving cached kernels.
// unordered_flat_map is used here to reduce lookup overhead.
// The slow path is used only once for each newly created kernel, so the
Expand Down Expand Up @@ -283,7 +292,7 @@ class KernelProgramCache {
std::unique_lock<std::mutex> Lock(MKernelFastCacheMutex);
auto It = MKernelFastCache.find(CacheKey);
if (It != MKernelFastCache.end()) {
traceKernel("Kernel fetched.", std::get<3>(CacheKey), true);
traceKernel("Kernel fetched.", std::get<2>(CacheKey), true);
return It->second;
}
return std::make_tuple(nullptr, nullptr, nullptr, nullptr);
Expand All @@ -294,7 +303,7 @@ class KernelProgramCache {
std::unique_lock<std::mutex> Lock(MKernelFastCacheMutex);
// if no insertion took place, thus some other thread has already inserted
// smth in the cache
traceKernel("Kernel inserted.", std::get<3>(CacheKey), true);
traceKernel("Kernel inserted.", std::get<2>(CacheKey), true);
MKernelFastCache.emplace(CacheKey, CacheVal);
}

Expand Down
10 changes: 2 additions & 8 deletions sycl/source/detail/program_manager/program_manager.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -964,17 +964,11 @@ ProgramManager::getOrCreateKernel(const ContextImplPtr &ContextImpl,
using KernelArgMaskPairT = KernelProgramCache::KernelArgMaskPairT;

KernelProgramCache &Cache = ContextImpl->getKernelProgramCache();

std::string CompileOpts, LinkOpts;
SerializedObj SpecConsts;
applyOptionsFromEnvironment(CompileOpts, LinkOpts);
// Should always come last!
appendCompileEnvironmentVariablesThatAppend(CompileOpts);
appendLinkEnvironmentVariablesThatAppend(LinkOpts);

ur_device_handle_t UrDevice = DeviceImpl->getHandleRef();

auto key = std::make_tuple(std::move(SpecConsts), UrDevice,
CompileOpts + LinkOpts, KernelName);
auto key = std::make_tuple(std::move(SpecConsts), UrDevice, KernelName);
if (SYCLConfig<SYCL_CACHE_IN_MEM>::get()) {
auto ret_tuple = Cache.tryToGetKernelFast(key);
constexpr size_t Kernel = 0; // see KernelFastCacheValT tuple
Expand Down
Loading