Skip to content
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

[SYCL] Optimize context passing in the ProgramManager #17835

Open
wants to merge 2 commits into
base: sycl
Choose a base branch
from
Open

[SYCL] Optimize context passing in the ProgramManager #17835

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
d323267
[SYCL] Optimize context passing in the ProgramManager
slawekptak Apr 3, 2025
bad0f12
Pass context as reference in build()
slawekptak Apr 4, 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
8 changes: 3 additions & 5 deletions sycl/source/detail/helpers.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -61,12 +61,11 @@ retrieveKernelBinary(const QueueImplPtr &Queue, const char *KernelName,
return {nullptr, nullptr};
}
auto ContextImpl = Queue->getContextImplPtr();
auto Context = detail::createSyclObjFromImpl<context>(ContextImpl);
auto DeviceImpl = Queue->getDeviceImplPtr();
auto Device = detail::createSyclObjFromImpl<device>(DeviceImpl);
ur_program_handle_t Program =
detail::ProgramManager::getInstance().createURProgram(
**DeviceImage, Context, {std::move(Device)});
**DeviceImage, ContextImpl, {std::move(Device)});
return {*DeviceImage, Program};
}

Expand All @@ -88,13 +87,12 @@ retrieveKernelBinary(const QueueImplPtr &Queue, const char *KernelName,
Program = KernelCG->MSyclKernel->getDeviceImage()->get_ur_program_ref();
} else {
auto ContextImpl = Queue->getContextImplPtr();
auto Context = detail::createSyclObjFromImpl<context>(ContextImpl);
auto DeviceImpl = Queue->getDeviceImplPtr();
auto Device = detail::createSyclObjFromImpl<device>(DeviceImpl);
DeviceImage = &detail::ProgramManager::getInstance().getDeviceImage(
KernelName, Context, Device);
KernelName, ContextImpl, Device);
Program = detail::ProgramManager::getInstance().createURProgram(
*DeviceImage, Context, {std::move(Device)});
*DeviceImage, ContextImpl, {std::move(Device)});
}
return {DeviceImage, Program};
}
Expand Down
2 changes: 1 addition & 1 deletion sycl/source/detail/memory_manager.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1224,7 +1224,7 @@ getOrBuildProgramForDeviceGlobal(QueueImplPtr Queue,
auto Context = createSyclObjFromImpl<context>(ContextImpl);
ProgramManager &PM = ProgramManager::getInstance();
RTDeviceBinaryImage &Img =
PM.getDeviceImage(DeviceGlobalEntry->MImages, Context, Device);
PM.getDeviceImage(DeviceGlobalEntry->MImages, ContextImpl, Device);
device_image_plain DeviceImage =
PM.getDeviceImageFromBinaryImage(&Img, Context, Device);
device_image_plain BuiltImage =
Expand Down
96 changes: 49 additions & 47 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 @@ -115,19 +115,19 @@ static ur_program_handle_t createSpirvProgram(const ContextImplPtr &Context,
}

// TODO replace this with a new UR API function
static bool isDeviceBinaryTypeSupported(const context &C,
static bool isDeviceBinaryTypeSupported(const ContextImplPtr &ContextImpl,
ur::DeviceBinaryType Format) {
// All formats except SYCL_DEVICE_BINARY_TYPE_SPIRV are supported.
if (Format != SYCL_DEVICE_BINARY_TYPE_SPIRV)
return true;

const backend ContextBackend = detail::getSyclObjImpl(C)->getBackend();
const backend ContextBackend = ContextImpl->getBackend();

// The CUDA backend cannot use SPIR-V
if (ContextBackend == backend::ext_oneapi_cuda)
return false;

std::vector<device> Devices = C.get_devices();
const std::vector<device> &Devices = ContextImpl->getDevices();

// Program type is SPIR-V, so we need a device compiler to do JIT.
for (const device &D : Devices) {
Expand All @@ -137,7 +137,8 @@ static bool isDeviceBinaryTypeSupported(const context &C,

// OpenCL 2.1 and greater require clCreateProgramWithIL
if (ContextBackend == backend::opencl) {
std::string ver = C.get_platform().get_info<info::platform::version>();
std::string ver = ContextImpl->get_info<info::context::platform>()
.get_info<info::platform::version>();
if (ver.find("OpenCL 1.0") == std::string::npos &&
ver.find("OpenCL 1.1") == std::string::npos &&
ver.find("OpenCL 1.2") == std::string::npos &&
Expand Down Expand Up @@ -187,16 +188,15 @@ static bool isDeviceBinaryTypeSupported(const context &C,

ur_program_handle_t
ProgramManager::createURProgram(const RTDeviceBinaryImage &Img,
const context &Context,
const ContextImplPtr &ContextImpl,
const std::vector<device> &Devices) {
if constexpr (DbgProgMgr > 0) {
std::vector<ur_device_handle_t> URDevices;
std::transform(
Devices.begin(), Devices.end(), std::back_inserter(URDevices),
[](const device &Dev) { return getSyclObjImpl(Dev)->getHandleRef(); });
std::cerr << ">>> ProgramManager::createPIProgram(" << &Img << ", "
<< getSyclObjImpl(Context).get() << ", " << VecToString(URDevices)
<< ")\n";
<< ContextImpl.get() << ", " << VecToString(URDevices) << ")\n";
}
const sycl_device_binary_struct &RawImg = Img.getRawData();

Expand Down Expand Up @@ -224,7 +224,7 @@ ProgramManager::createURProgram(const RTDeviceBinaryImage &Img,
// sycl::detail::pi::PiDeviceBinaryType Format = Img->Format;
// assert(Format != SYCL_DEVICE_BINARY_TYPE_NONE && "Image format not set");

if (!isDeviceBinaryTypeSupported(Context, Format))
if (!isDeviceBinaryTypeSupported(ContextImpl, Format))
throw sycl::exception(
sycl::errc::feature_not_supported,
"SPIR-V online compilation is not supported in this context");
Expand All @@ -233,23 +233,22 @@ ProgramManager::createURProgram(const RTDeviceBinaryImage &Img,
const auto &ProgMetadata = Img.getProgramMetadataUR();

// Load the image
const ContextImplPtr &Ctx = getSyclObjImpl(Context);
std::vector<const uint8_t *> Binaries(
Devices.size(), const_cast<uint8_t *>(RawImg.BinaryStart));
std::vector<size_t> Lengths(Devices.size(), ImgSize);
ur_program_handle_t Res =
Format == SYCL_DEVICE_BINARY_TYPE_SPIRV
? createSpirvProgram(Ctx, RawImg.BinaryStart, ImgSize)
: createBinaryProgram(Ctx, Devices, Binaries.data(), Lengths.data(),
ProgMetadata);
? createSpirvProgram(ContextImpl, RawImg.BinaryStart, ImgSize)
: createBinaryProgram(ContextImpl, Devices, Binaries.data(),
Lengths.data(), ProgMetadata);

{
std::lock_guard<std::mutex> Lock(MNativeProgramsMutex);
// associate the UR program with the image it was created for
NativePrograms.insert({Res, {Ctx, &Img}});
NativePrograms.insert({Res, {ContextImpl, &Img}});
}

Ctx->addDeviceGlobalInitializer(Res, Devices, &Img);
ContextImpl->addDeviceGlobalInitializer(Res, Devices, &Img);

if constexpr (DbgProgMgr > 1)
std::cerr << "created program: " << Res
Expand Down Expand Up @@ -518,7 +517,7 @@ static void applyOptionsFromEnvironment(std::string &CompileOpts,
std::pair<ur_program_handle_t, bool> ProgramManager::getOrCreateURProgram(
const RTDeviceBinaryImage &MainImg,
const std::vector<const RTDeviceBinaryImage *> &AllImages,
const context &Context, const std::vector<device> &Devices,
const ContextImplPtr &ContextImpl, const std::vector<device> &Devices,
const std::string &CompileAndLinkOptions, SerializedObj SpecConsts) {
ur_program_handle_t NativePrg;

Expand All @@ -540,11 +539,10 @@ std::pair<ur_program_handle_t, bool> ProgramManager::getOrCreateURProgram(
ProgMetadataVector.insert(ProgMetadataVector.end(),
ImgProgMetadata.begin(), ImgProgMetadata.end());
}
NativePrg =
createBinaryProgram(getSyclObjImpl(Context), Devices, BinPtrs.data(),
Lengths.data(), ProgMetadataVector);
NativePrg = createBinaryProgram(ContextImpl, Devices, BinPtrs.data(),
Lengths.data(), ProgMetadataVector);
} else {
NativePrg = createURProgram(MainImg, Context, Devices);
NativePrg = createURProgram(MainImg, ContextImpl, Devices);
}
return {NativePrg, Binaries.size()};
}
Expand Down Expand Up @@ -853,10 +851,10 @@ ur_program_handle_t ProgramManager::getBuiltURProgram(
RootDevImpl->getHandleRef(), UR_DEVICE_INFO_BUILD_ON_SUBDEVICE,
sizeof(ur_bool_t), &MustBuildOnSubdevice, nullptr);

auto Context = createSyclObjFromImpl<context>(ContextImpl);
auto Device = createSyclObjFromImpl<device>(
MustBuildOnSubdevice == true ? DeviceImpl : RootDevImpl);
const RTDeviceBinaryImage &Img = getDeviceImage(KernelName, Context, Device);
const RTDeviceBinaryImage &Img =
getDeviceImage(KernelName, ContextImpl, Device);

// Check that device supports all aspects used by the kernel
if (auto exception = checkDevSupportDeviceRequirements(Device, Img, NDRDesc))
Expand All @@ -875,19 +873,19 @@ ur_program_handle_t ProgramManager::getBuiltURProgram(
std::copy(DeviceImagesToLink.begin(), DeviceImagesToLink.end(),
std::back_inserter(AllImages));

return getBuiltURProgram(std::move(AllImages), Context, {std::move(Device)});
return getBuiltURProgram(std::move(AllImages), ContextImpl,
{std::move(Device)});
}

ur_program_handle_t ProgramManager::getBuiltURProgram(
const BinImgWithDeps &ImgWithDeps, const context &Context,
const BinImgWithDeps &ImgWithDeps, const ContextImplPtr &ContextImpl,
const std::vector<device> &Devs, const DevImgPlainWithDeps *DevImgWithDeps,
const SerializedObj &SpecConsts) {
std::string CompileOpts;
std::string LinkOpts;
applyOptionsFromEnvironment(CompileOpts, LinkOpts);
auto BuildF = [this, &ImgWithDeps, &DevImgWithDeps, &Context, &Devs,
auto BuildF = [this, &ImgWithDeps, &DevImgWithDeps, &ContextImpl, &Devs,
&CompileOpts, &LinkOpts, &SpecConsts] {
const ContextImplPtr &ContextImpl = getSyclObjImpl(Context);
const AdapterPtr &Adapter = ContextImpl->getAdapter();
const RTDeviceBinaryImage &MainImg = *ImgWithDeps.getMain();
applyOptionsFromImage(CompileOpts, LinkOpts, MainImg, Devs, Adapter);
Expand All @@ -896,7 +894,7 @@ ur_program_handle_t ProgramManager::getBuiltURProgram(
appendLinkEnvironmentVariablesThatAppend(LinkOpts);

auto [NativePrg, DeviceCodeWasInCache] =
getOrCreateURProgram(MainImg, ImgWithDeps.getAll(), Context, Devs,
getOrCreateURProgram(MainImg, ImgWithDeps.getAll(), ContextImpl, Devs,
CompileOpts + LinkOpts, SpecConsts);

if (!DeviceCodeWasInCache && MainImg.supportsSpecConstants()) {
Expand Down Expand Up @@ -936,7 +934,8 @@ ur_program_handle_t ProgramManager::getBuiltURProgram(
if (UseDeviceLibs)
DeviceLibReqMask |= getDeviceLibReqMask(*BinImg);

ur_program_handle_t NativePrg = createURProgram(*BinImg, Context, Devs);
ur_program_handle_t NativePrg =
createURProgram(*BinImg, ContextImpl, Devs);

if (BinImg->supportsSpecConstants()) {
enableITTAnnotationsIfNeeded(NativePrg, Adapter);
Expand Down Expand Up @@ -1001,7 +1000,6 @@ ur_program_handle_t ProgramManager::getBuiltURProgram(
auto CacheKey =
std::make_pair(std::make_pair(SpecConsts, ImgId), URDevicesSet);

const ContextImplPtr &ContextImpl = getSyclObjImpl(Context);
KernelProgramCache &Cache = ContextImpl->getKernelProgramCache();
auto GetCachedBuildF = [&Cache, &CacheKey]() {
return Cache.getOrInsertProgram(CacheKey);
Expand Down Expand Up @@ -1476,7 +1474,8 @@ sycl_device_binary getRawImg(RTDeviceBinaryImage *Img) {
template <typename StorageKey>
RTDeviceBinaryImage *getBinImageFromMultiMap(
const std::unordered_multimap<StorageKey, RTDeviceBinaryImage *> &ImagesSet,
const StorageKey &Key, const context &Context, const device &Device) {
const StorageKey &Key, const ContextImplPtr &ContextImpl,
const device &Device) {
auto [ItBegin, ItEnd] = ImagesSet.equal_range(Key);
if (ItBegin == ItEnd)
return nullptr;
Expand Down Expand Up @@ -1506,19 +1505,20 @@ RTDeviceBinaryImage *getBinImageFromMultiMap(
uint32_t ImgInd = 0;
// Ask the native runtime under the given context to choose the device image
// it prefers.
getSyclObjImpl(Context)->getAdapter()->call<UrApiKind::urDeviceSelectBinary>(
ContextImpl->getAdapter()->call<UrApiKind::urDeviceSelectBinary>(
getSyclObjImpl(Device)->getHandleRef(), UrBinaries.data(),
UrBinaries.size(), &ImgInd);
return DeviceFilteredImgs[ImgInd];
}

RTDeviceBinaryImage &
ProgramManager::getDeviceImage(const std::string &KernelName,
const context &Context, const device &Device) {
const ContextImplPtr &ContextImpl,
const device &Device) {
if constexpr (DbgProgMgr > 0) {
std::cerr << ">>> ProgramManager::getDeviceImage(\"" << KernelName << "\", "
<< getSyclObjImpl(Context).get() << ", "
<< getSyclObjImpl(Device).get() << ")\n";
<< ContextImpl.get() << ", " << getSyclObjImpl(Device).get()
<< ")\n";

std::cerr << "available device images:\n";
debugPrintBinaryImages();
Expand All @@ -1528,7 +1528,7 @@ ProgramManager::getDeviceImage(const std::string &KernelName,
assert(m_SpvFileImage);
return getDeviceImage(
std::unordered_set<RTDeviceBinaryImage *>({m_SpvFileImage.get()}),
Context, Device);
ContextImpl, Device);
}

RTDeviceBinaryImage *Img = nullptr;
Expand All @@ -1537,9 +1537,9 @@ ProgramManager::getDeviceImage(const std::string &KernelName,
if (auto KernelId = m_KernelName2KernelIDs.find(KernelName);
KernelId != m_KernelName2KernelIDs.end()) {
Img = getBinImageFromMultiMap(m_KernelIDs2BinImage, KernelId->second,
Context, Device);
ContextImpl, Device);
} else {
Img = getBinImageFromMultiMap(m_ServiceKernels, KernelName, Context,
Img = getBinImageFromMultiMap(m_ServiceKernels, KernelName, ContextImpl,
Device);
}
}
Expand All @@ -1561,13 +1561,13 @@ ProgramManager::getDeviceImage(const std::string &KernelName,

RTDeviceBinaryImage &ProgramManager::getDeviceImage(
const std::unordered_set<RTDeviceBinaryImage *> &ImageSet,
const context &Context, const device &Device) {
const ContextImplPtr &ContextImpl, const device &Device) {
assert(ImageSet.size() > 0);

if constexpr (DbgProgMgr > 0) {
std::cerr << ">>> ProgramManager::getDeviceImage(Custom SPV file "
<< getSyclObjImpl(Context).get() << ", "
<< getSyclObjImpl(Device).get() << ")\n";
<< ContextImpl.get() << ", " << getSyclObjImpl(Device).get()
<< ")\n";

std::cerr << "available device images:\n";
debugPrintBinaryImages();
Expand All @@ -1589,7 +1589,7 @@ RTDeviceBinaryImage &ProgramManager::getDeviceImage(
getUrDeviceTarget(RawImgs[BinaryCount]->DeviceTargetSpec);
}

getSyclObjImpl(Context)->getAdapter()->call<UrApiKind::urDeviceSelectBinary>(
ContextImpl->getAdapter()->call<UrApiKind::urDeviceSelectBinary>(
getSyclObjImpl(Device)->getHandleRef(), UrBinaries.data(),
UrBinaries.size(), &ImgInd);

Expand Down Expand Up @@ -2845,8 +2845,9 @@ ProgramManager::compile(const DevImgPlainWithDeps &ImgWithDeps,
const AdapterPtr &Adapter =
getSyclObjImpl(InputImpl->get_context())->getAdapter();

ur_program_handle_t Prog = createURProgram(*InputImpl->get_bin_image_ref(),
InputImpl->get_context(), Devs);
ur_program_handle_t Prog =
createURProgram(*InputImpl->get_bin_image_ref(),
getSyclObjImpl(InputImpl->get_context()), Devs);

if (InputImpl->get_bin_image_ref()->supportsSpecConstants())
setSpecializationConstants(InputImpl, Prog, Adapter);
Expand Down Expand Up @@ -3036,7 +3037,8 @@ ProgramManager::build(const DevImgPlainWithDeps &DevImgWithDeps,
const std::shared_ptr<device_image_impl> &MainInputImpl =
getSyclObjImpl(DevImgWithDeps.getMain());

const context Context = MainInputImpl->get_context();
const context &Context = MainInputImpl->get_context();
const ContextImplPtr &ContextImpl = detail::getSyclObjImpl(Context);

std::vector<const RTDeviceBinaryImage *> BinImgs;
BinImgs.reserve(DevImgWithDeps.size());
Expand Down Expand Up @@ -3065,7 +3067,7 @@ ProgramManager::build(const DevImgPlainWithDeps &DevImgWithDeps,
}

ur_program_handle_t ResProgram = getBuiltURProgram(
std::move(BinImgs), Context, Devs, &DevImgWithDeps, SpecConstBlob);
std::move(BinImgs), ContextImpl, Devs, &DevImgWithDeps, SpecConstBlob);

DeviceImageImplPtr ExecImpl = std::make_shared<detail::device_image_impl>(
MainInputImpl->get_bin_image_ref(), Context, Devs,
Expand Down Expand Up @@ -3185,7 +3187,8 @@ ur_kernel_handle_t ProgramManager::getOrCreateMaterializedKernel(

if constexpr (DbgProgMgr > 0)
std::cerr << ">>> Adding the kernel to the cache.\n";
auto Program = createURProgram(Img, Context, {Device});
const ContextImplPtr &ContextImpl = detail::getSyclObjImpl(Context);
auto Program = createURProgram(Img, ContextImpl, {Device});
auto DeviceImpl = detail::getSyclObjImpl(Device);
auto &Adapter = DeviceImpl->getAdapter();
UrFuncInfo<UrApiKind::urProgramRelease> programReleaseInfo;
Expand All @@ -3200,8 +3203,7 @@ ur_kernel_handle_t ProgramManager::getOrCreateMaterializedKernel(
std::vector<ur_program_handle_t> ExtraProgramsToLink;
std::vector<ur_device_handle_t> Devs = {DeviceImpl->getHandleRef()};
auto BuildProgram =
build(std::move(ProgramManaged), detail::getSyclObjImpl(Context),
CompileOpts, LinkOpts, Devs,
build(std::move(ProgramManaged), ContextImpl, CompileOpts, LinkOpts, Devs,
/*For non SPIR-V devices DeviceLibReqdMask is always 0*/ 0,
ExtraProgramsToLink);
ur_kernel_handle_t UrKernel{nullptr};
Expand Down
11 changes: 6 additions & 5 deletions sycl/source/detail/program_manager/program_manager.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -135,15 +135,15 @@ class ProgramManager {
static ProgramManager &getInstance();

RTDeviceBinaryImage &getDeviceImage(const std::string &KernelName,
const context &Context,
const ContextImplPtr &ContextImpl,
const device &Device);

RTDeviceBinaryImage &getDeviceImage(
const std::unordered_set<RTDeviceBinaryImage *> &ImagesToVerify,
const context &Context, const device &Device);
const ContextImplPtr &ContextImpl, const device &Device);

ur_program_handle_t createURProgram(const RTDeviceBinaryImage &Img,
const context &Context,
const ContextImplPtr &ContextImpl,
const std::vector<device> &Devices);
/// Creates a UR program using either a cached device code binary if present
/// in the persistent cache or from the supplied device image otherwise.
Expand All @@ -167,7 +167,7 @@ class ProgramManager {
std::pair<ur_program_handle_t, bool> getOrCreateURProgram(
const RTDeviceBinaryImage &Img,
const std::vector<const RTDeviceBinaryImage *> &AllImages,
const context &Context, const std::vector<device> &Devices,
const ContextImplPtr &ContextImpl, const std::vector<device> &Devices,
const std::string &CompileAndLinkOptions, SerializedObj SpecConsts);
/// Builds or retrieves from cache a program defining the kernel with given
/// name.
Expand All @@ -192,7 +192,8 @@ class ProgramManager {
/// \param SpecConsts is an optional parameter containing spec constant values
/// the program should be built with.
ur_program_handle_t
getBuiltURProgram(const BinImgWithDeps &ImgWithDeps, const context &Context,
getBuiltURProgram(const BinImgWithDeps &ImgWithDeps,
const ContextImplPtr &ContextImpl,
const std::vector<device> &Devs,
const DevImgPlainWithDeps *DevImgWithDeps = nullptr,
const SerializedObj &SpecConsts = {});
Expand Down