intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2026-01-08 05:56:36 +08:00
Code Issues Packages Projects Releases Wiki Activity

refactor: do not add all mapped physical allocations to kernel residency

Browse Source 
Related-To: NEO-11719

Signed-off-by: Zbigniew Zdanowicz <zbigniew.zdanowicz@intel.com>
This commit is contained in:
Zbigniew Zdanowicz
2024-07-24 12:26:41 +00:00
committed by Compute-Runtime-Automation
parent 9632f91ee8
commit 42bf58a800
2 changed files with 37 additions and 16 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
level_zero/core/source/kernel/kernel_imp.cpp
View File

@@ -770,15 +770,6 @@ ze_result_t KernelImp::setArgBuffer(uint32_t argIndex, size_t argSize, const voi
const uint32_t allocId = allocData->getAllocId();
kernelArgInfos[argIndex] = KernelArgInfo{requestedAddress, allocId, allocationsCounter, false};
if (allocData->virtualReservationData) {
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->argumentsResidencyContainer.push_back(mappedAllocationData.second->mappedAllocation->allocation);
}
}
}
return setArgBufferWithAlloc(argIndex, gpuAddress, alloc, peerAllocData);
}

44
level_zero/core/test/unit_tests/sources/module/test_module.cpp
View File

@@ -2640,14 +2640,20 @@ class MultiDeviceModuleSetArgBufferTest : public MultiDeviceModuleFixture, publi
};
HWTEST_F(MultiDeviceModuleSetArgBufferTest,
givenCallsToSetArgBufferWithReservedMemoryThenResidencyContainerHasAllMappedAllocations) {
givenCallsToSetArgBufferWithReservedMemoryThenKernelResidencyContainerHasKernelArgMappedAllocationAndMemoryInterfaceHasAllMappedAllocations) {
using RENDER_SURFACE_STATE = typename FamilyType::RENDER_SURFACE_STATE;
for (uint32_t rootDeviceIndex = 0; rootDeviceIndex < numRootDevices; rootDeviceIndex++) {
createModuleFromMockBinary(rootDeviceIndex);
auto device = driverHandle->devices[rootDeviceIndex];
driverHandle->devices[rootDeviceIndex]->getNEODevice()->getExecutionEnvironment()->rootDeviceEnvironments[rootDeviceIndex]->memoryOperationsInterface =
auto neoDevice = device->getNEODevice();
neoDevice->getExecutionEnvironment()->rootDeviceEnvironments[rootDeviceIndex]->memoryOperationsInterface =
std::make_unique<NEO::MockMemoryOperations>();
NEO::MockMemoryOperations *mockMemoryInterface = static_cast<NEO::MockMemoryOperations *>(
neoDevice->getExecutionEnvironment()->rootDeviceEnvironments[rootDeviceIndex]->memoryOperationsInterface.get());
mockMemoryInterface->captureGfxAllocationsForMakeResident = true;
ze_kernel_handle_t kernelHandle;
void *ptr = nullptr;
size_t size = MemoryConstants::pageSize64k;
@@ -2694,7 +2700,16 @@ HWTEST_F(MultiDeviceModuleSetArgBufferTest,
EXPECT_EQ(surfaceStateAddress->getHeight(), static_cast<uint32_t>(length.surfaceState.height + 1));
EXPECT_EQ(surfaceStateAddress->getDepth(), static_cast<uint32_t>(length.surfaceState.depth + 1));
EXPECT_TRUE(phys1Resident);
EXPECT_TRUE(phys2Resident);
EXPECT_FALSE(phys2Resident);
auto physicalIt = driverHandle->getMemoryManager()->getPhysicalMemoryAllocationMap().find(static_cast<void *>(phPhysicalMemory));
auto physical1Allocation = physicalIt->second->allocation;
EXPECT_EQ(NEO::MemoryOperationsStatus::success, mockMemoryInterface->isResident(neoDevice, *physical1Allocation));
physicalIt = driverHandle->getMemoryManager()->getPhysicalMemoryAllocationMap().find(static_cast<void *>(phPhysicalMemory2));
auto physical2Allocation = physicalIt->second->allocation;
EXPECT_EQ(NEO::MemoryOperationsStatus::success, mockMemoryInterface->isResident(neoDevice, *physical2Allocation));
res = context->unMapVirtualMem(ptr, size);
EXPECT_EQ(ZE_RESULT_SUCCESS, res);
res = context->unMapVirtualMem(offsetAddress, size);
@@ -2710,14 +2725,20 @@ HWTEST_F(MultiDeviceModuleSetArgBufferTest,
}
HWTEST_F(MultiDeviceModuleSetArgBufferTest,
givenCallsToSetArgBufferWithOffsetReservedMemoryThenResidencyContainerHasAllMappedAllocations) {
givenCallsToSetArgBufferWithOffsetReservedMemoryThenKernelResidencyHasArgMappedAllocationAndMemoryInterfaceHasAllMappedAllocations) {
using RENDER_SURFACE_STATE = typename FamilyType::RENDER_SURFACE_STATE;
for (uint32_t rootDeviceIndex = 0; rootDeviceIndex < numRootDevices; rootDeviceIndex++) {
createModuleFromMockBinary(rootDeviceIndex);
auto device = driverHandle->devices[rootDeviceIndex];
driverHandle->devices[rootDeviceIndex]->getNEODevice()->getExecutionEnvironment()->rootDeviceEnvironments[rootDeviceIndex]->memoryOperationsInterface =
auto neoDevice = device->getNEODevice();
neoDevice->getExecutionEnvironment()->rootDeviceEnvironments[rootDeviceIndex]->memoryOperationsInterface =
std::make_unique<NEO::MockMemoryOperations>();
NEO::MockMemoryOperations *mockMemoryInterface = static_cast<NEO::MockMemoryOperations *>(
neoDevice->getExecutionEnvironment()->rootDeviceEnvironments[rootDeviceIndex]->memoryOperationsInterface.get());
mockMemoryInterface->captureGfxAllocationsForMakeResident = true;
ze_kernel_handle_t kernelHandle;
void *ptr = nullptr;
size_t size = MemoryConstants::pageSize64k;
@@ -2763,8 +2784,17 @@ HWTEST_F(MultiDeviceModuleSetArgBufferTest,
EXPECT_EQ(surfaceStateAddress->getWidth(), static_cast<uint32_t>(length.surfaceState.width));
EXPECT_EQ(surfaceStateAddress->getHeight(), static_cast<uint32_t>(length.surfaceState.height + 1));
EXPECT_EQ(surfaceStateAddress->getDepth(), static_cast<uint32_t>(length.surfaceState.depth + 1));
EXPECT_TRUE(phys1Resident);
EXPECT_FALSE(phys1Resident);
EXPECT_TRUE(phys2Resident);
auto physicalIt = driverHandle->getMemoryManager()->getPhysicalMemoryAllocationMap().find(static_cast<void *>(phPhysicalMemory));
auto physical1Allocation = physicalIt->second->allocation;
EXPECT_EQ(NEO::MemoryOperationsStatus::success, mockMemoryInterface->isResident(neoDevice, *physical1Allocation));
physicalIt = driverHandle->getMemoryManager()->getPhysicalMemoryAllocationMap().find(static_cast<void *>(phPhysicalMemory2));
auto physical2Allocation = physicalIt->second->allocation;
EXPECT_EQ(NEO::MemoryOperationsStatus::success, mockMemoryInterface->isResident(neoDevice, *physical2Allocation));
res = context->unMapVirtualMem(ptr, size);
EXPECT_EQ(ZE_RESULT_SUCCESS, res);
res = context->unMapVirtualMem(offsetAddress, size);
@@ -2901,7 +2931,7 @@ HWTEST_F(MultiDeviceModuleSetArgBufferTest,
EXPECT_EQ(surfaceStateAddress->getHeight(), static_cast<uint32_t>(length.surfaceState.height + 1));
EXPECT_EQ(surfaceStateAddress->getDepth(), static_cast<uint32_t>(length.surfaceState.depth + 1));
EXPECT_TRUE(phys1Resident);
EXPECT_TRUE(phys2Resident);
EXPECT_FALSE(phys2Resident);
res = context->unMapVirtualMem(ptr, size);
EXPECT_EQ(ZE_RESULT_SUCCESS, res);
res = context->unMapVirtualMem(offsetAddress, size);