Skip to content

[SYCL][UR][L0] Replace memory type look-up with UMF tracking #10807

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 6 commits into from
Sep 14, 2023
Merged

[SYCL][UR][L0] Replace memory type look-up with UMF tracking #10807

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
6594009
[SYCL][UR][L0] Replace memory type look-up with UMF tracking
kswiecicki Aug 10, 2023
cd4c466
[SYCL][UR][L0] Route direct mem alloc calls through proxy pool
kswiecicki Aug 17, 2023
0f279ec
[SYCL][UR][L0] Ignore UsmHostFlags
kswiecicki Aug 21, 2023
7c295df
[SYCL][UR] Set UMF pool tracking var before FetchContent_Declare
kswiecicki Aug 21, 2023
69fb9eb
[SYCL][UR][L0] Refactor USMMemoryProvider
kswiecicki Aug 22, 2023
7fafe41
[SYCL][UR][L0] Change expected API calls in usm_pooling.cpp test
kswiecicki Aug 30, 2023
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: 1 addition & 1 deletion sycl/plugins/unified_runtime/CMakeLists.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -6,6 +6,7 @@ if (NOT DEFINED UNIFIED_RUNTIME_LIBRARY OR NOT DEFINED UNIFIED_RUNTIME_INCLUDE_D
set(UNIFIED_RUNTIME_REPO "https://github.com/oneapi-src/unified-runtime.git")
set(UNIFIED_RUNTIME_TAG v0.7.1)

set(UMF_ENABLE_POOL_TRACKING ON)
message(STATUS "Will fetch Unified Runtime from ${UNIFIED_RUNTIME_REPO}")
FetchContent_Declare(unified-runtime
GIT_REPOSITORY ${UNIFIED_RUNTIME_REPO}
Expand Down Expand Up @@ -46,7 +47,6 @@ if (NOT DEFINED UNIFIED_RUNTIME_LIBRARY OR NOT DEFINED UNIFIED_RUNTIME_INCLUDE_D
set(UNIFIED_RUNTIME_INCLUDE_DIR "${UNIFIED_RUNTIME_SOURCE_DIR}/include")
endif()


add_library(UnifiedRuntime-Headers INTERFACE)

target_include_directories(UnifiedRuntime-Headers
Expand Down
46 changes: 39 additions & 7 deletions sycl/plugins/unified_runtime/ur/adapters/level_zero/context.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -182,7 +182,7 @@ ur_result_t ur_context_handle_t_::initialize() {
// Note that the CCS devices and their respective subdevices share a
// common ze_device_handle and therefore, also share USM allocators.
auto createUSMAllocators = [this](ur_device_handle_t Device) {
auto MemProvider = umf::memoryProviderMakeUnique<USMDeviceMemoryProvider>(
auto MemProvider = umf::memoryProviderMakeUnique<L0DeviceMemoryProvider>(
reinterpret_cast<ur_context_handle_t>(this), Device)
.second;
DeviceMemPools.emplace(
Expand All @@ -193,7 +193,7 @@ ur_result_t ur_context_handle_t_::initialize() {
.Configs[usm::DisjointPoolMemType::Device])
.second));

MemProvider = umf::memoryProviderMakeUnique<USMSharedMemoryProvider>(
MemProvider = umf::memoryProviderMakeUnique<L0SharedMemoryProvider>(
reinterpret_cast<ur_context_handle_t>(this), Device)
.second;
SharedMemPools.emplace(
Expand All @@ -204,10 +204,9 @@ ur_result_t ur_context_handle_t_::initialize() {
.Configs[usm::DisjointPoolMemType::Shared])
.second));

MemProvider =
umf::memoryProviderMakeUnique<USMSharedReadOnlyMemoryProvider>(
reinterpret_cast<ur_context_handle_t>(this), Device)
.second;
MemProvider = umf::memoryProviderMakeUnique<L0SharedReadOnlyMemoryProvider>(
reinterpret_cast<ur_context_handle_t>(this), Device)
.second;
SharedReadOnlyMemPools.emplace(
std::piecewise_construct, std::make_tuple(Device->ZeDevice),
std::make_tuple(
Expand All @@ -216,6 +215,33 @@ ur_result_t ur_context_handle_t_::initialize() {
DisjointPoolConfigInstance
.Configs[usm::DisjointPoolMemType::SharedReadOnly])
.second));

MemProvider = umf::memoryProviderMakeUnique<L0DeviceMemoryProvider>(
reinterpret_cast<ur_context_handle_t>(this), Device)
.second;
DeviceMemProxyPools.emplace(
std::piecewise_construct, std::make_tuple(Device->ZeDevice),
std::make_tuple(
umf::poolMakeUnique<USMProxyPool, 1>({std::move(MemProvider)})
.second));

MemProvider = umf::memoryProviderMakeUnique<L0SharedMemoryProvider>(
reinterpret_cast<ur_context_handle_t>(this), Device)
.second;
SharedMemProxyPools.emplace(
std::piecewise_construct, std::make_tuple(Device->ZeDevice),
std::make_tuple(
umf::poolMakeUnique<USMProxyPool, 1>({std::move(MemProvider)})
.second));

MemProvider = umf::memoryProviderMakeUnique<L0SharedReadOnlyMemoryProvider>(
reinterpret_cast<ur_context_handle_t>(this), Device)
.second;
SharedReadOnlyMemProxyPools.emplace(
std::piecewise_construct, std::make_tuple(Device->ZeDevice),
std::make_tuple(
umf::poolMakeUnique<USMProxyPool, 1>({std::move(MemProvider)})
.second));
};

// Recursive helper to call createUSMAllocators for all sub-devices
Expand All @@ -236,7 +262,7 @@ ur_result_t ur_context_handle_t_::initialize() {
// Create USM pool for host. Device and Shared USM allocations
// are device-specific. Host allocations are not device-dependent therefore
// we don't need a map with device as key.
auto MemProvider = umf::memoryProviderMakeUnique<USMHostMemoryProvider>(
auto MemProvider = umf::memoryProviderMakeUnique<L0HostMemoryProvider>(
reinterpret_cast<ur_context_handle_t>(this), nullptr)
.second;
HostMemPool =
Expand All @@ -245,6 +271,12 @@ ur_result_t ur_context_handle_t_::initialize() {
DisjointPoolConfigInstance.Configs[usm::DisjointPoolMemType::Host])
.second;

MemProvider = umf::memoryProviderMakeUnique<L0HostMemoryProvider>(
reinterpret_cast<ur_context_handle_t>(this), nullptr)
.second;
HostMemProxyPool =
umf::poolMakeUnique<USMProxyPool, 1>({std::move(MemProvider)}).second;

// We may allocate memory to this root device so create allocators.
if (SingleRootDevice &&
DeviceMemPools.find(SingleRootDevice->ZeDevice) == DeviceMemPools.end()) {
Expand Down
14 changes: 9 additions & 5 deletions sycl/plugins/unified_runtime/ur/adapters/level_zero/context.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -101,14 +101,18 @@ struct ur_context_handle_t_ : _ur_object {
std::unordered_map<ze_device_handle_t, umf::pool_unique_handle_t>
SharedReadOnlyMemPools;

// Since L0 native runtime does not distinguisg "shared device_read_only"
// vs regular "shared" allocations, we have keep track of it to use
// proper memory pool when freeing allocations.
std::unordered_set<void *> SharedReadOnlyAllocs;

// Store the host memory pool. It does not depend on any device.
umf::pool_unique_handle_t HostMemPool;

// Allocation-tracking proxy pools for direct allocations. No pooling used.
std::unordered_map<ze_device_handle_t, umf::pool_unique_handle_t>
DeviceMemProxyPools;
std::unordered_map<ze_device_handle_t, umf::pool_unique_handle_t>
SharedMemProxyPools;
std::unordered_map<ze_device_handle_t, umf::pool_unique_handle_t>
SharedReadOnlyMemProxyPools;
umf::pool_unique_handle_t HostMemProxyPool;

// We need to store all memory allocations in the context because there could
// be kernels with indirect access. Kernels with indirect access start to
// reference all existing memory allocations at the time when they are
Expand Down
Loading