intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2026-01-03 06:49:52 +08:00
Code Issues Packages Projects Releases Wiki Activity

refactor: split kernel residency into internal and argument containers

Browse Source 
Related-To: NEO-11719

Signed-off-by: Zbigniew Zdanowicz <zbigniew.zdanowicz@intel.com>
This commit is contained in:
Zbigniew Zdanowicz
2024-07-23 12:14:04 +00:00
committed by Compute-Runtime-Automation
parent ae68df3832
commit 1c1e437d4b
16 changed files with 93 additions and 73 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
level_zero/core/source/cmdlist/cmdlist_hw_skl_to_tgllp.inl
View File

@@ -254,8 +254,12 @@ ze_result_t CommandListCoreFamily<gfxCoreFamily>::appendLaunchKernelWithParams(K
appendSignalEventPostWalker(event, nullptr, nullptr, false, false, false);
commandContainer.addToResidencyContainer(kernelImmutableData->getIsaGraphicsAllocation());
auto &residencyContainer = kernel->getResidencyContainer();
for (auto resource : residencyContainer) {
auto &argumentsResidencyContainer = kernel->getArgumentsResidencyContainer();
for (auto resource : argumentsResidencyContainer) {
commandContainer.addToResidencyContainer(resource);
}
auto &internalResidencyContainer = kernel->getInternalResidencyContainer();
for (auto resource : internalResidencyContainer) {
commandContainer.addToResidencyContainer(resource);
}

29
level_zero/core/source/cmdlist/cmdlist_hw_xehp_and_later.inl
View File

@@ -225,15 +225,20 @@ ze_result_t CommandListCoreFamily<gfxCoreFamily>::appendLaunchKernelWithParams(K
bool isKernelUsingSystemAllocation = false;
if (!launchParams.isBuiltInKernel) {
auto &kernelAllocations = kernel->getResidencyContainer();
for (auto &allocation : kernelAllocations) {
if (allocation == nullptr) {
continue;
auto verifyKernelUsingSystemAllocations = [&isKernelUsingSystemAllocation](const NEO::ResidencyContainer &kernelResidencyContainer) {
for (const auto &allocation : kernelResidencyContainer) {
if (allocation == nullptr) {
continue;
}
if (allocation->getAllocationType() == NEO::AllocationType::bufferHostMemory) {
isKernelUsingSystemAllocation = true;
}
}
if (allocation->getAllocationType() == NEO::AllocationType::bufferHostMemory) {
isKernelUsingSystemAllocation = true;
}
}
};
verifyKernelUsingSystemAllocations(kernel->getArgumentsResidencyContainer());
verifyKernelUsingSystemAllocations(kernel->getInternalResidencyContainer());
} else {
isKernelUsingSystemAllocation = launchParams.isDestinationAllocationInSystemMemory;
}
@@ -437,9 +442,13 @@ ze_result_t CommandListCoreFamily<gfxCoreFamily>::appendLaunchKernelWithParams(K
// Attach kernel residency to our CommandList residency
{
commandContainer.addToResidencyContainer(kernelImmutableData->getIsaGraphicsAllocation());
auto &internalResidencyContainer = kernel->getInternalResidencyContainer();
for (auto resource : internalResidencyContainer) {
commandContainer.addToResidencyContainer(resource);
}
if (!launchParams.omitAddingKernelResidency) {
auto &residencyContainer = kernel->getResidencyContainer();
for (auto resource : residencyContainer) {
auto &argumentsResidencyContainer = kernel->getArgumentsResidencyContainer();
for (auto resource : argumentsResidencyContainer) {
commandContainer.addToResidencyContainer(resource);
}
}

3
level_zero/core/source/kernel/kernel.h
View File

@@ -148,7 +148,8 @@ struct Kernel : _ze_kernel_handle_t, virtual NEO::DispatchKernelEncoderI {
virtual const KernelImmutableData *getImmutableData() const = 0;
virtual const std::vector<NEO::GraphicsAllocation *> &getResidencyContainer() const = 0;
virtual const std::vector<NEO::GraphicsAllocation *> &getArgumentsResidencyContainer() const = 0;
virtual const std::vector<NEO::GraphicsAllocation *> &getInternalResidencyContainer() const = 0;
virtual UnifiedMemoryControls getUnifiedMemoryControls() const = 0;
virtual bool hasIndirectAllocationsAllowed() const = 0;

36
level_zero/core/source/kernel/kernel_imp.cpp
View File

@@ -586,7 +586,7 @@ ze_result_t KernelImp::setArgImmediate(uint32_t argIndex, size_t argSize, const
ze_result_t KernelImp::setArgRedescribedImage(uint32_t argIndex, ze_image_handle_t argVal) {
const auto &arg = kernelImmData->getDescriptor().payloadMappings.explicitArgs[argIndex].as<NEO::ArgDescImage>();
if (argVal == nullptr) {
residencyContainer[argIndex] = nullptr;
argumentsResidencyContainer[argIndex] = nullptr;
return ZE_RESULT_SUCCESS;
}
@@ -620,7 +620,7 @@ ze_result_t KernelImp::setArgRedescribedImage(uint32_t argIndex, ze_image_handle
} else {
image->copyRedescribedSurfaceStateToSSH(surfaceStateHeapData.get(), arg.bindful);
}
residencyContainer[argIndex] = image->getAllocation();
argumentsResidencyContainer[argIndex] = image->getAllocation();
return ZE_RESULT_SUCCESS;
}
@@ -656,7 +656,7 @@ ze_result_t KernelImp::setArgBufferWithAlloc(uint32_t argIndex, uintptr_t argVal
}
this->setKernelArgUncached(argIndex, argIsUncacheable);
}
residencyContainer[argIndex] = allocation;
argumentsResidencyContainer[argIndex] = allocation;
return ZE_RESULT_SUCCESS;
}
@@ -727,7 +727,7 @@ ze_result_t KernelImp::setArgBuffer(uint32_t argIndex, size_t argSize, const voi
}
if (nullptr == argVal) {
residencyContainer[argIndex] = nullptr;
argumentsResidencyContainer[argIndex] = nullptr;
const auto &arg = kernelImmData->getDescriptor().payloadMappings.explicitArgs[argIndex].as<NEO::ArgDescPointer>();
uintptr_t nullBufferValue = 0;
NEO::patchPointer(ArrayRef<uint8_t>(crossThreadData.get(), crossThreadDataSize), arg, nullBufferValue);
@@ -774,7 +774,7 @@ ze_result_t KernelImp::setArgBuffer(uint32_t argIndex, size_t argSize, const voi
for (const auto &mappedAllocationData : allocData->virtualReservationData->mappedAllocations) {
// Add additional allocations to the residency container if the virtual reservation spans multiple allocations.
if (requestedAddress != mappedAllocationData.second->ptr) {
this->residencyContainer.push_back(mappedAllocationData.second->mappedAllocation->allocation);
this->argumentsResidencyContainer.push_back(mappedAllocationData.second->mappedAllocation->allocation);
}
}
}
@@ -784,7 +784,7 @@ ze_result_t KernelImp::setArgBuffer(uint32_t argIndex, size_t argSize, const voi
ze_result_t KernelImp::setArgImage(uint32_t argIndex, size_t argSize, const void *argVal) {
if (argVal == nullptr) {
residencyContainer[argIndex] = nullptr;
argumentsResidencyContainer[argIndex] = nullptr;
return ZE_RESULT_SUCCESS;
}
@@ -824,10 +824,10 @@ ze_result_t KernelImp::setArgImage(uint32_t argIndex, size_t argSize, const void
image->copySurfaceStateToSSH(surfaceStateHeapData.get(), arg.bindful, isMediaBlockImage);
}
residencyContainer[argIndex] = image->getAllocation();
argumentsResidencyContainer[argIndex] = image->getAllocation();
if (image->getImplicitArgsAllocation()) {
this->residencyContainer.push_back(image->getImplicitArgsAllocation());
this->argumentsResidencyContainer.push_back(image->getImplicitArgsAllocation());
}
auto imageInfo = image->getImageInfo();
@@ -1091,13 +1091,13 @@ ze_result_t KernelImp::initialize(const ze_kernel_desc_t *desc) {
}
}
residencyContainer.resize(this->kernelArgHandlers.size(), nullptr);
argumentsResidencyContainer.resize(this->kernelArgHandlers.size(), nullptr);
auto &kernelAttributes = kernelDescriptor.kernelAttributes;
if ((kernelAttributes.perHwThreadPrivateMemorySize != 0U) && (false == module->shouldAllocatePrivateMemoryPerDispatch())) {
this->privateMemoryGraphicsAllocation = allocatePrivateMemoryGraphicsAllocation();
this->patchCrossthreadDataWithPrivateAllocation(this->privateMemoryGraphicsAllocation);
this->residencyContainer.push_back(this->privateMemoryGraphicsAllocation);
this->internalResidencyContainer.push_back(this->privateMemoryGraphicsAllocation);
}
this->createPrintfBuffer();
@@ -1106,8 +1106,8 @@ ze_result_t KernelImp::initialize(const ze_kernel_desc_t *desc) {
this->setAssertBuffer();
residencyContainer.insert(residencyContainer.end(), kernelImmData->getResidencyContainer().begin(),
kernelImmData->getResidencyContainer().end());
internalResidencyContainer.insert(internalResidencyContainer.end(), kernelImmData->getResidencyContainer().begin(),
kernelImmData->getResidencyContainer().end());
ModuleImp *moduleImp = reinterpret_cast<ModuleImp *>(this->module);
const auto indirectDetectionVersion = moduleImp->getTranslationUnit()->programInfo.indirectDetectionVersion;
@@ -1138,7 +1138,7 @@ ze_result_t KernelImp::initialize(const ze_kernel_desc_t *desc) {
}
for (auto rtStack : rtDispatchGlobalsInfo->rtStacks) {
this->residencyContainer.push_back(rtStack);
this->internalResidencyContainer.push_back(rtStack);
}
auto address = rtDispatchGlobalsInfo->rtDispatchGlobalsArray->getGpuAddressToPatch();
@@ -1151,7 +1151,7 @@ ze_result_t KernelImp::initialize(const ze_kernel_desc_t *desc) {
pImplicitArgs->rtGlobalBufferPtr = address;
}
this->residencyContainer.push_back(rtDispatchGlobalsInfo->rtDispatchGlobalsArray);
this->internalResidencyContainer.push_back(rtDispatchGlobalsInfo->rtDispatchGlobalsArray);
}
this->midThreadPreemptionDisallowedForRayTracingKernels = productHelper.isMidThreadPreemptionDisallowedForRayTracingKernels();
return ZE_RESULT_SUCCESS;
@@ -1160,7 +1160,7 @@ ze_result_t KernelImp::initialize(const ze_kernel_desc_t *desc) {
void KernelImp::createPrintfBuffer() {
if (this->kernelImmData->getDescriptor().kernelAttributes.flags.usesPrintf || pImplicitArgs) {
this->printfBuffer = PrintfHandler::createPrintfBuffer(this->module->getDevice());
this->residencyContainer.push_back(printfBuffer);
this->internalResidencyContainer.push_back(printfBuffer);
if (this->kernelImmData->getDescriptor().kernelAttributes.flags.usesPrintf) {
NEO::patchPointer(ArrayRef<uint8_t>(crossThreadData.get(), crossThreadDataSize),
this->getImmutableData()->getDescriptor().payloadMappings.implicitArgs.printfSurfaceAddress,
@@ -1186,14 +1186,14 @@ bool KernelImp::usesRegionGroupBarrier() const {
}
void KernelImp::patchSyncBuffer(NEO::GraphicsAllocation *gfxAllocation, size_t bufferOffset) {
this->residencyContainer.push_back(gfxAllocation);
this->internalResidencyContainer.push_back(gfxAllocation);
NEO::patchPointer(ArrayRef<uint8_t>(crossThreadData.get(), crossThreadDataSize),
this->getImmutableData()->getDescriptor().payloadMappings.implicitArgs.syncBufferAddress,
static_cast<uintptr_t>(ptrOffset(gfxAllocation->getGpuAddressToPatch(), bufferOffset)));
}
void KernelImp::patchRegionGroupBarrier(NEO::GraphicsAllocation *gfxAllocation, size_t bufferOffset) {
this->residencyContainer.push_back(gfxAllocation);
this->internalResidencyContainer.push_back(gfxAllocation);
NEO::patchPointer(ArrayRef<uint8_t>(crossThreadData.get(), crossThreadDataSize),
this->getImmutableData()->getDescriptor().payloadMappings.implicitArgs.regionGroupBarrierBuffer,
@@ -1335,7 +1335,7 @@ void KernelImp::setAssertBuffer() {
NEO::patchPointer(ArrayRef<uint8_t>(crossThreadData.get(), crossThreadDataSize),
this->getImmutableData()->getDescriptor().payloadMappings.implicitArgs.assertBufferAddress,
static_cast<uintptr_t>(assertHandler->getAssertBuffer()->getGpuAddressToPatch()));
this->residencyContainer.push_back(assertHandler->getAssertBuffer());
this->internalResidencyContainer.push_back(assertHandler->getAssertBuffer());
if (pImplicitArgs) {
pImplicitArgs->assertBufferPtr = static_cast<uintptr_t>(assertHandler->getAssertBuffer()->getGpuAddressToPatch());

11
level_zero/core/source/kernel/kernel_imp.h
View File

@@ -75,8 +75,12 @@ struct KernelImp : Kernel {
const uint8_t *getCrossThreadData() const override { return crossThreadData.get(); }
uint32_t getCrossThreadDataSize() const override { return crossThreadDataSize; }
const std::vector<NEO::GraphicsAllocation *> &getResidencyContainer() const override {
return residencyContainer;
const std::vector<NEO::GraphicsAllocation *> &getArgumentsResidencyContainer() const override {
return argumentsResidencyContainer;
}
const std::vector<NEO::GraphicsAllocation *> &getInternalResidencyContainer() const override {
return internalResidencyContainer;
}
ze_result_t setArgImmediate(uint32_t argIndex, size_t argSize, const void *argVal);
@@ -210,7 +214,8 @@ struct KernelImp : Kernel {
typedef ze_result_t (KernelImp::*KernelArgHandler)(uint32_t argIndex, size_t argSize, const void *argVal);
std::vector<KernelArgInfo> kernelArgInfos;
std::vector<KernelImp::KernelArgHandler> kernelArgHandlers;
std::vector<NEO::GraphicsAllocation *> residencyContainer;
std::vector<NEO::GraphicsAllocation *> argumentsResidencyContainer;
std::vector<NEO::GraphicsAllocation *> internalResidencyContainer;
std::mutex *devicePrintfKernelMutex = nullptr;
NEO::GraphicsAllocation *printfBuffer = nullptr;