intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2025-12-19 06:24:51 +08:00
Code Issues Packages Projects Releases Wiki Activity

refactor: Use timestamp types for events and in order nodes

Browse Source 
Switching tagAllocation into UC causes regressions in non event
scenarios. It is no longer used as dependant type for semaphores.
Restoring previous GMM_USAGE settings for tag.
Marking events and in order nodes using only timestamp types as they
have proper GMM_USAGE settings already and can be both in smem and lmem.

Resolves: NEO-13847

Signed-off-by: Lukasz Jobczyk <lukasz.jobczyk@intel.com>
This commit is contained in:
Lukasz Jobczyk
2025-01-30 10:23:55 +00:00
committed by Compute-Runtime-Automation
parent 9305652446
commit 09e23804dc
11 changed files with 12 additions and 24 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
level_zero/core/source/event/event.cpp
View File

@@ -100,8 +100,7 @@ ze_result_t EventPool::initialize(DriverHandle *driver, Context *context, uint32
initializeSizeParameters(numDevices, deviceHandles, *driverHandleImp, rootDeviceEnvironment);
NEO::AllocationType allocationType = isEventPoolTimestampFlagSet() ? NEO::AllocationType::timestampPacketTagBuffer
: NEO::AllocationType::tagBuffer;
NEO::AllocationType allocationType = NEO::AllocationType::timestampPacketTagBuffer;
if (this->devices.size() > 1) {
this->isDeviceEventPoolAllocation = false;
}
@@ -448,7 +447,7 @@ ze_result_t EventPool::openEventPoolIpcHandle(const ze_ipc_event_pool_handle_t &
return ZE_RESULT_ERROR_INVALID_ARGUMENT;
}
NEO::AllocationType allocationType = NEO::AllocationType::tagBuffer;
NEO::AllocationType allocationType = NEO::AllocationType::timestampPacketTagBuffer;
if (eventPool->isDeviceEventPoolAllocation) {
allocationType = NEO::AllocationType::gpuTimestampDeviceBuffer;
}

6
level_zero/core/test/unit_tests/sources/cmdlist/test_in_order_cmdlist_1.cpp
View File

@@ -2412,7 +2412,7 @@ HWTEST2_F(InOrderCmdListTests, givenDebugFlagSetWhenDispatchingThenEnsureHostAll
EXPECT_NE(nullptr, immCmdList2->inOrderExecInfo->getHostCounterAllocation());
EXPECT_NE(immCmdList2->inOrderExecInfo->getDeviceCounterAllocation(), immCmdList2->inOrderExecInfo->getHostCounterAllocation());
EXPECT_EQ(AllocationType::tagBuffer, immCmdList1->inOrderExecInfo->getHostCounterAllocation()->getAllocationType());
EXPECT_EQ(AllocationType::timestampPacketTagBuffer, immCmdList1->inOrderExecInfo->getHostCounterAllocation()->getAllocationType());
EXPECT_EQ(immCmdList1->inOrderExecInfo->getBaseHostAddress(), immCmdList1->inOrderExecInfo->getHostCounterAllocation()->getUnderlyingBuffer());
EXPECT_FALSE(immCmdList1->inOrderExecInfo->getHostCounterAllocation()->isAllocatedInLocalMemoryPool());
@@ -2421,7 +2421,7 @@ HWTEST2_F(InOrderCmdListTests, givenDebugFlagSetWhenDispatchingThenEnsureHostAll
auto hostAllocOffset = ptrDiff(immCmdList2->inOrderExecInfo->getBaseHostAddress(), immCmdList1->inOrderExecInfo->getBaseHostAddress());
EXPECT_NE(0u, hostAllocOffset);
EXPECT_EQ(AllocationType::tagBuffer, immCmdList2->inOrderExecInfo->getHostCounterAllocation()->getAllocationType());
EXPECT_EQ(AllocationType::timestampPacketTagBuffer, immCmdList2->inOrderExecInfo->getHostCounterAllocation()->getAllocationType());
EXPECT_EQ(immCmdList2->inOrderExecInfo->getBaseHostAddress(), ptrOffset(immCmdList2->inOrderExecInfo->getHostCounterAllocation()->getUnderlyingBuffer(), hostAllocOffset));
EXPECT_FALSE(immCmdList2->inOrderExecInfo->getHostCounterAllocation()->isAllocatedInLocalMemoryPool());
@@ -2454,7 +2454,7 @@ HWTEST2_F(InOrderCmdListTests, givenInOrderModeWhenDispatchingThenHandleDependen
auto immCmdList = createImmCmdList<gfxCoreFamily>();
EXPECT_NE(nullptr, immCmdList->inOrderExecInfo.get());
EXPECT_EQ(AllocationType::timestampPacketTagBuffer, immCmdList->inOrderExecInfo->getDeviceCounterAllocation()->getAllocationType());
EXPECT_EQ(AllocationType::gpuTimestampDeviceBuffer, immCmdList->inOrderExecInfo->getDeviceCounterAllocation()->getAllocationType());
EXPECT_EQ(0u, immCmdList->inOrderExecInfo->getCounterValue());

2
level_zero/core/test/unit_tests/sources/cmdlist/test_in_order_cmdlist_2.cpp
View File

@@ -382,7 +382,7 @@ HWTEST2_F(CopyOffloadInOrderTests, givenProfilingEventWithRelaxedOrderingWhenApp
}
}
HWTEST2_F(CopyOffloadInOrderTests, givenAtomicSignalingModeWhenUpdatingCounterThenUseCorrectHwCommands, IsAtLeastXeHpCore) {
HWTEST2_F(CopyOffloadInOrderTests, givenAtomicSignalingModeWhenUpdatingCounterThenUseCorrectHwCommands, IsAtLeastXe2HpgCore) {
using MI_ATOMIC = typename FamilyType::MI_ATOMIC;
using ATOMIC_OPCODES = typename FamilyType::MI_ATOMIC::ATOMIC_OPCODES;
using DATA_SIZE = typename FamilyType::MI_ATOMIC::DATA_SIZE;

2
level_zero/core/test/unit_tests/sources/event/test_event.cpp
View File

@@ -369,7 +369,7 @@ TEST_F(EventPoolCreate, GivenDeviceThenEventPoolIsCreated) {
if (l0GfxCoreHelper.alwaysAllocateEventInLocalMem()) {
EXPECT_EQ(NEO::AllocationType::gpuTimestampDeviceBuffer, eventPool->getAllocation().getAllocationType());
} else {
EXPECT_EQ(NEO::AllocationType::tagBuffer, eventPool->getAllocation().getAllocationType());
EXPECT_EQ(NEO::AllocationType::timestampPacketTagBuffer, eventPool->getAllocation().getAllocationType());
}
eventPool->destroy();
}

7
shared/source/gmm_helper/cache_settings_helper.cpp
View File

@@ -12,7 +12,6 @@
#include "shared/source/helpers/gfx_core_helper.h"
#include "shared/source/helpers/hw_info.h"
#include "shared/source/memory_manager/allocation_type.h"
#include "shared/source/os_interface/os_interface.h"
#include "shared/source/os_interface/product_helper.h"
#include "shared/source/release_helper/release_helper.h"
@@ -88,12 +87,8 @@ GMM_RESOURCE_USAGE_TYPE_ENUM CacheSettingsHelper::getDefaultUsageTypeWithCaching
case AllocationType::fillPattern:
case AllocationType::svmCpu:
case AllocationType::svmZeroCopy:
if (debugManager.flags.DisableCachingForStatefulBufferAccess.get()) {
return getDefaultUsageTypeWithCachingDisabled(allocationType, productHelper);
}
return GMM_RESOURCE_USAGE_OCL_SYSTEM_MEMORY_BUFFER;
case AllocationType::tagBuffer:
if (productHelper.isDcFlushAllowed() && OSInterface::isSemaphoreDependantResourceUCRequired) {
if (debugManager.flags.DisableCachingForStatefulBufferAccess.get()) {
return getDefaultUsageTypeWithCachingDisabled(allocationType, productHelper);
}
return GMM_RESOURCE_USAGE_OCL_SYSTEM_MEMORY_BUFFER;

2
shared/source/helpers/in_order_cmd_helpers.h
View File

@@ -30,7 +30,7 @@ class DeviceAllocNodeType {
static constexpr size_t defaultAllocatorTagCount = 128;
static constexpr AllocationType getAllocationType() { return deviceAlloc ? NEO::AllocationType::timestampPacketTagBuffer : NEO::AllocationType::tagBuffer; }
static constexpr AllocationType getAllocationType() { return deviceAlloc ? NEO::AllocationType::gpuTimestampDeviceBuffer : NEO::AllocationType::timestampPacketTagBuffer; }
static constexpr TagNodeType getTagNodeType() { return TagNodeType::counter64b; }

1
shared/source/os_interface/linux/os_interface_linux.cpp
View File

@@ -22,7 +22,6 @@ bool OSInterface::osEnabled64kbPages = false;
bool OSInterface::newResourceImplicitFlush = true;
bool OSInterface::gpuIdleImplicitFlush = true;
bool OSInterface::requiresSupportForWddmTrimNotification = false;
bool OSInterface::isSemaphoreDependantResourceUCRequired = false;
bool OSInterface::isDebugAttachAvailable() const {
if (driverModel && driverModel->getDriverModelType() == DriverModelType::drm) {

1
shared/source/os_interface/os_interface.h
View File

@@ -123,7 +123,6 @@ class OSInterface : public NonCopyableClass {
static bool newResourceImplicitFlush;
static bool gpuIdleImplicitFlush;
static bool requiresSupportForWddmTrimNotification;
static bool isSemaphoreDependantResourceUCRequired;
static std::vector<std::unique_ptr<HwDeviceId>> discoverDevices(ExecutionEnvironment &executionEnvironment);
static std::vector<std::unique_ptr<HwDeviceId>> discoverDevice(ExecutionEnvironment &executionEnvironment, std::string &osPciPath);

1
shared/source/os_interface/windows/os_interface_win.cpp
View File

@@ -14,7 +14,6 @@ bool OSInterface::osEnabled64kbPages = true;
bool OSInterface::newResourceImplicitFlush = true;
bool OSInterface::gpuIdleImplicitFlush = false;
bool OSInterface::requiresSupportForWddmTrimNotification = true;
bool OSInterface::isSemaphoreDependantResourceUCRequired = true;
bool OSInterface::isDebugAttachAvailable() const {
if (driverModel) {

5
shared/test/unit_test/gmm_helper/gmm_helper_tests.cpp
View File

@@ -9,7 +9,6 @@
#include "shared/source/gmm_helper/gmm_helper.h"
#include "shared/source/helpers/gfx_core_helper.h"
#include "shared/source/os_interface/device_factory.h"
#include "shared/source/os_interface/os_interface.h"
#include "shared/source/sku_info/operations/sku_info_transfer.h"
#include "shared/test/common/fixtures/mock_execution_environment_gmm_fixture.h"
#include "shared/test/common/helpers/debug_manager_state_restore.h"
@@ -719,10 +718,8 @@ TEST(GmmTest, givenAllocationTypeWhenGettingUsageTypeThenReturnCorrectValue) {
case AllocationType::fillPattern:
case AllocationType::svmCpu:
case AllocationType::svmZeroCopy:
expectedUsage = forceUncached ? uncachedGmmUsageType : GMM_RESOURCE_USAGE_OCL_SYSTEM_MEMORY_BUFFER;
break;
case AllocationType::tagBuffer:
expectedUsage = (forceUncached || (productHelper.isDcFlushAllowed() && OSInterface::isSemaphoreDependantResourceUCRequired)) ? uncachedGmmUsageType : GMM_RESOURCE_USAGE_OCL_SYSTEM_MEMORY_BUFFER;
expectedUsage = forceUncached ? uncachedGmmUsageType : GMM_RESOURCE_USAGE_OCL_SYSTEM_MEMORY_BUFFER;
break;
default:
expectedUsage = forceUncached ? uncachedGmmUsageType : GMM_RESOURCE_USAGE_OCL_BUFFER;

4
shared/test/unit_test/utilities/tag_allocator_tests.cpp
View File

@@ -481,8 +481,8 @@ TEST_F(TagAllocatorTest, givenTagAllocatorWhenGraphicsAllocationIsCreatedThenSet
EXPECT_EQ(AllocationType::timestampPacketTagBuffer, timestampPacketTag->getBaseGraphicsAllocation()->getAllocationType());
EXPECT_EQ(AllocationType::profilingTagBuffer, hwTimeStampsTag->getBaseGraphicsAllocation()->getAllocationType());
EXPECT_EQ(AllocationType::profilingTagBuffer, hwPerfCounterTag->getBaseGraphicsAllocation()->getAllocationType());
EXPECT_EQ(AllocationType::timestampPacketTagBuffer, inOrderDeviceTag->getBaseGraphicsAllocation()->getAllocationType());
EXPECT_EQ(AllocationType::tagBuffer, inOrderHostTag->getBaseGraphicsAllocation()->getAllocationType());
EXPECT_EQ(AllocationType::gpuTimestampDeviceBuffer, inOrderDeviceTag->getBaseGraphicsAllocation()->getAllocationType());
EXPECT_EQ(AllocationType::timestampPacketTagBuffer, inOrderHostTag->getBaseGraphicsAllocation()->getAllocationType());
}
TEST_F(TagAllocatorTest, givenMultipleRootDevicesWhenPopulatingTagsThenCreateMultiGraphicsAllocation) {