intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2026-01-06 02:18:05 +08:00
Code Issues Packages Projects Releases Wiki Activity

Reduce the size of masks to 4.

Browse Source 
32 is not required.

Signed-off-by: Michal Mrozek <michal.mrozek@intel.com>
This commit is contained in:
Michal Mrozek
2022-09-19 16:38:59 +00:00
committed by Compute-Runtime-Automation
parent afc5596a0a
commit 3d5e34f727
13 changed files with 29 additions and 32 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
opencl/test/unit_test/command_queue/enqueue_kernel_1_tests.cpp
View File

@@ -1548,10 +1548,10 @@ HWTEST_F(EnqueueKernelTest, givenContextWithSeveralDevicesWhenEnqueueKernelThenD
clEnqueueNDRangeKernel(this->pCmdQ, mockKernel.mockMultiDeviceKernel, 1, nullptr, gws, nullptr, 0, nullptr, nullptr); clEnqueueNDRangeKernel(this->pCmdQ, mockKernel.mockMultiDeviceKernel, 1, nullptr, gws, nullptr, 0, nullptr, nullptr);
EXPECT_FALSE(mockCsr->passedDispatchFlags.areMultipleSubDevicesInContext); EXPECT_FALSE(mockCsr->passedDispatchFlags.areMultipleSubDevicesInContext);
context->deviceBitfields[rootDeviceIndex].set(7, true); context->deviceBitfields[rootDeviceIndex].set(3, true);
clEnqueueNDRangeKernel(this->pCmdQ, mockKernel.mockMultiDeviceKernel, 1, nullptr, gws, nullptr, 0, nullptr, nullptr); clEnqueueNDRangeKernel(this->pCmdQ, mockKernel.mockMultiDeviceKernel, 1, nullptr, gws, nullptr, 0, nullptr, nullptr);
EXPECT_TRUE(mockCsr->passedDispatchFlags.areMultipleSubDevicesInContext); EXPECT_TRUE(mockCsr->passedDispatchFlags.areMultipleSubDevicesInContext);
context->deviceBitfields[rootDeviceIndex].set(7, false); context->deviceBitfields[rootDeviceIndex].set(3, false);
} }
HWTEST_F(EnqueueKernelTest, givenNonVMEKernelWhenEnqueueKernelThenDispatchFlagsDoesntHaveMediaSamplerRequired) { HWTEST_F(EnqueueKernelTest, givenNonVMEKernelWhenEnqueueKernelThenDispatchFlagsDoesntHaveMediaSamplerRequired) {

4
opencl/test/unit_test/device/sub_device_tests.cpp
View File

@@ -158,11 +158,11 @@ TEST(SubDevicesTest, givenClDeviceWithSubDevicesWhenSubDeviceInternalRefCountsAr
TEST(SubDevicesTest, givenDeviceWithSubDevicesWhenSubDeviceCreationFailThenWholeDeviceIsDestroyed) { TEST(SubDevicesTest, givenDeviceWithSubDevicesWhenSubDeviceCreationFailThenWholeDeviceIsDestroyed) {
DebugManagerStateRestore restorer; DebugManagerStateRestore restorer;
DebugManager.flags.CreateMultipleSubDevices.set(10); DebugManager.flags.CreateMultipleSubDevices.set(4);
MockExecutionEnvironment executionEnvironment; MockExecutionEnvironment executionEnvironment;
executionEnvironment.prepareRootDeviceEnvironments(1); executionEnvironment.prepareRootDeviceEnvironments(1);
executionEnvironment.incRefInternal(); executionEnvironment.incRefInternal();
executionEnvironment.memoryManager.reset(new FailMemoryManager(10, executionEnvironment)); executionEnvironment.memoryManager.reset(new FailMemoryManager(4, executionEnvironment));
auto device = Device::create<RootDevice>(&executionEnvironment, 0u); auto device = Device::create<RootDevice>(&executionEnvironment, 0u);
EXPECT_EQ(nullptr, device); EXPECT_EQ(nullptr, device);
} }

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

@@ -17,7 +17,7 @@ namespace NEO {
class AffinityMaskHelper { class AffinityMaskHelper {
public: public:
using AffinityMaskContainer = std::vector<std::bitset<32>>; using AffinityMaskContainer = std::vector<std::bitset<4>>;
AffinityMaskHelper(bool allSubdevicesActive) { AffinityMaskHelper(bool allSubdevicesActive) {
if (!allSubdevicesActive) { if (!allSubdevicesActive) {

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

@@ -14,7 +14,7 @@
namespace NEO { namespace NEO {
struct EngineControl; struct EngineControl;
using EngineControlContainer = std::vector<EngineControl>; using EngineControlContainer = std::vector<EngineControl>;
using DeviceBitfield = std::bitset<32>; using DeviceBitfield = std::bitset<4>;
class Device; class Device;
using DeviceVector = std::vector<std::unique_ptr<Device>>; using DeviceVector = std::vector<std::unique_ptr<Device>>;

2
shared/source/memory_manager/local_memory_usage.cpp
View File

@@ -56,7 +56,7 @@ void LocalMemoryUsageBankSelector::reserveOnBank(uint32_t bankIndex, uint64_t al
} }
void LocalMemoryUsageBankSelector::updateUsageInfo(uint32_t memoryBanks, uint64_t allocationSize, bool reserve) { void LocalMemoryUsageBankSelector::updateUsageInfo(uint32_t memoryBanks, uint64_t allocationSize, bool reserve) {
auto banks = std::bitset<32>(memoryBanks); auto banks = std::bitset<4>(memoryBanks);
for (uint32_t bankIndex = 0; bankIndex < banks.size() && bankIndex < banksCount; bankIndex++) { for (uint32_t bankIndex = 0; bankIndex < banks.size() && bankIndex < banksCount; bankIndex++) {
if (banks.test(bankIndex)) { if (banks.test(bankIndex)) {
if (reserve) { if (reserve) {

2
shared/source/os_interface/hw_info_config.inl
View File

@@ -120,7 +120,7 @@ bool HwInfoConfigHw<gfxProduct>::getConcurrentAccessMemCapabilitiesSupported(Usm
auto supported = false; auto supported = false;
if (DebugManager.flags.EnableUsmConcurrentAccessSupport.get() > 0) { if (DebugManager.flags.EnableUsmConcurrentAccessSupport.get() > 0) {
auto capabilityBitset = std::bitset<32>(DebugManager.flags.EnableUsmConcurrentAccessSupport.get()); auto capabilityBitset = std::bitset<4>(DebugManager.flags.EnableUsmConcurrentAccessSupport.get());
supported = capabilityBitset.test(static_cast<uint32_t>(capability)); supported = capabilityBitset.test(static_cast<uint32_t>(capability));
} }

2
shared/source/os_interface/linux/drm_memory_manager.cpp
View File

@@ -1495,7 +1495,7 @@ BufferObject *DrmMemoryManager::createBufferObjectInMemoryRegion(Drm *drm, Gmm *
uint32_t handle = 0; uint32_t handle = 0;
uint32_t ret = 0; uint32_t ret = 0;
auto banks = std::bitset<32>(memoryBanks); auto banks = std::bitset<4>(memoryBanks);
if (banks.count() > 1) { if (banks.count() > 1) {
ret = memoryInfo->createGemExtWithMultipleRegions(memoryBanks, size, handle); ret = memoryInfo->createGemExtWithMultipleRegions(memoryBanks, size, handle);
} else { } else {

2
shared/source/os_interface/linux/memory_info.cpp
View File

@@ -125,7 +125,7 @@ uint32_t MemoryInfo::createGemExtWithSingleRegion(uint32_t memoryBanks, size_t a
uint32_t MemoryInfo::createGemExtWithMultipleRegions(uint32_t memoryBanks, size_t allocSize, uint32_t &handle) { uint32_t MemoryInfo::createGemExtWithMultipleRegions(uint32_t memoryBanks, size_t allocSize, uint32_t &handle) {
auto pHwInfo = this->drm.getRootDeviceEnvironment().getHardwareInfo(); auto pHwInfo = this->drm.getRootDeviceEnvironment().getHardwareInfo();
auto banks = std::bitset<32>(memoryBanks); auto banks = std::bitset<4>(memoryBanks);
MemRegionsVec memRegions{}; MemRegionsVec memRegions{};
size_t currentBank = 0; size_t currentBank = 0;
size_t i = 0; size_t i = 0;

4
shared/test/common/libult/create_command_stream.h
View File

@@ -1,5 +1,5 @@
/* /*
* Copyright (C) 2018-2021 Intel Corporation * Copyright (C) 2018-2022 Intel Corporation
* *
* SPDX-License-Identifier: MIT * SPDX-License-Identifier: MIT
* *
@@ -15,7 +15,7 @@ namespace NEO {
class CommandStreamReceiver; class CommandStreamReceiver;
class ExecutionEnvironment; class ExecutionEnvironment;
using DeviceBitfield = std::bitset<32>; using DeviceBitfield = std::bitset<4>;
extern CommandStreamReceiver *createCommandStream(ExecutionEnvironment &executionEnvironment, extern CommandStreamReceiver *createCommandStream(ExecutionEnvironment &executionEnvironment,
uint32_t rootDeviceIndex, uint32_t rootDeviceIndex,

6
shared/test/unit_test/memory_manager/local_memory_usage_tests.cpp
View File

@@ -1,5 +1,5 @@
/* /*
* Copyright (C) 2019-2021 Intel Corporation * Copyright (C) 2019-2022 Intel Corporation
* *
* SPDX-License-Identifier: MIT * SPDX-License-Identifier: MIT
* *
@@ -76,7 +76,7 @@ TEST(localMemoryUsageTest, givenOverrideLeastOccupiedBankDebugFlagWhenGetLeastOc
} }
TEST(localMemoryUsageTest, givenLocalMemoryUsageBankSelectorWhenMemoryAllocatedSeveralTimesThenItIsStoredOnDifferentBanks) { TEST(localMemoryUsageTest, givenLocalMemoryUsageBankSelectorWhenMemoryAllocatedSeveralTimesThenItIsStoredOnDifferentBanks) {
MockLocalMemoryUsageBankSelector selector(5u); MockLocalMemoryUsageBankSelector selector(4u);
uint64_t allocationSize = 1024u; uint64_t allocationSize = 1024u;
@@ -88,8 +88,6 @@ TEST(localMemoryUsageTest, givenLocalMemoryUsageBankSelectorWhenMemoryAllocatedS
selector.reserveOnBank(bankIndex, allocationSize); selector.reserveOnBank(bankIndex, allocationSize);
bankIndex = selector.getLeastOccupiedBank(selector.bitfield); bankIndex = selector.getLeastOccupiedBank(selector.bitfield);
selector.reserveOnBank(bankIndex, allocationSize); selector.reserveOnBank(bankIndex, allocationSize);
bankIndex = selector.getLeastOccupiedBank(selector.bitfield);
selector.reserveOnBank(bankIndex, allocationSize);
for (uint32_t i = 0; i < selector.banksCount; i++) { for (uint32_t i = 0; i < selector.banksCount; i++) {
EXPECT_EQ(allocationSize, selector.getOccupiedMemorySizeForBank(i)); EXPECT_EQ(allocationSize, selector.getOccupiedMemorySizeForBank(i));

4
shared/test/unit_test/os_interface/hw_info_config_tests.cpp
View File

@@ -62,9 +62,9 @@ HWTEST_F(HwInfoConfigTest, givenHwInfoConfigWhenGettingMemoryCapabilitiesThenCor
auto hwInfoConfig = HwInfoConfig::get(pInHwInfo.platform.eProductFamily); auto hwInfoConfig = HwInfoConfig::get(pInHwInfo.platform.eProductFamily);
for (auto capabilityBitmask : {0, 0b0001, 0b0010, 0b0100, 0b1000, 0b1111, 0b10000}) { for (auto capabilityBitmask : {0, 0b0001, 0b0010, 0b0100, 0b1000, 0b1111}) {
DebugManager.flags.EnableUsmConcurrentAccessSupport.set(capabilityBitmask); DebugManager.flags.EnableUsmConcurrentAccessSupport.set(capabilityBitmask);
std::bitset<32> capabilityBitset(capabilityBitmask); std::bitset<4> capabilityBitset(capabilityBitmask);
auto hostMemCapabilities = hwInfoConfig->getHostMemCapabilities(&pInHwInfo); auto hostMemCapabilities = hwInfoConfig->getHostMemCapabilities(&pInHwInfo);
if (hostMemCapabilities > 0) { if (hostMemCapabilities > 0) {

17
shared/test/unit_test/os_interface/linux/drm_tests.cpp
View File

@@ -705,7 +705,7 @@ TEST(DrmTest, givenPerContextVMRequiredWhenCreatingOsContextsThenImplicitVmIdPer
EXPECT_EQ(2u, drmMock.receivedContextParamRequestCount); EXPECT_EQ(2u, drmMock.receivedContextParamRequestCount);
auto &drmVmIds = osContext.getDrmVmIds(); auto &drmVmIds = osContext.getDrmVmIds();
EXPECT_EQ(32u, drmVmIds.size()); EXPECT_EQ(4u, drmVmIds.size());
EXPECT_EQ(20u, drmVmIds[0]); EXPECT_EQ(20u, drmVmIds[0]);
} }
@@ -718,7 +718,7 @@ TEST(DrmTest, givenPerContextVMRequiredWhenCreatingOsContextForSubDeviceThenImpl
DrmMock drmMock(rootEnv); DrmMock drmMock(rootEnv);
EXPECT_TRUE(drmMock.requirePerContextVM); EXPECT_TRUE(drmMock.requirePerContextVM);
drmMock.storedRetValForVmId = 20; drmMock.storedRetValForVmId = 4;
DeviceBitfield deviceBitfield(1 << 3); DeviceBitfield deviceBitfield(1 << 3);
OsContextLinux osContext(drmMock, 0u, EngineDescriptorHelper::getDefaultDescriptor(deviceBitfield)); OsContextLinux osContext(drmMock, 0u, EngineDescriptorHelper::getDefaultDescriptor(deviceBitfield));
@@ -726,9 +726,9 @@ TEST(DrmTest, givenPerContextVMRequiredWhenCreatingOsContextForSubDeviceThenImpl
EXPECT_EQ(2u, drmMock.receivedContextParamRequestCount); EXPECT_EQ(2u, drmMock.receivedContextParamRequestCount);
auto &drmVmIds = osContext.getDrmVmIds(); auto &drmVmIds = osContext.getDrmVmIds();
EXPECT_EQ(32u, drmVmIds.size()); EXPECT_EQ(4u, drmVmIds.size());
EXPECT_EQ(20u, drmVmIds[3]); EXPECT_EQ(4u, drmVmIds[3]);
EXPECT_EQ(0u, drmVmIds[0]); EXPECT_EQ(0u, drmVmIds[0]);
EXPECT_EQ(0u, drmVmIds[2]); EXPECT_EQ(0u, drmVmIds[2]);
@@ -742,7 +742,7 @@ TEST(DrmTest, givenPerContextVMRequiredWhenCreatingOsContextsForRootDeviceThenIm
DrmMock drmMock(rootEnv); DrmMock drmMock(rootEnv);
EXPECT_TRUE(drmMock.requirePerContextVM); EXPECT_TRUE(drmMock.requirePerContextVM);
drmMock.storedRetValForVmId = 20; drmMock.storedRetValForVmId = 4;
DeviceBitfield deviceBitfield(1 | 1 << 1); DeviceBitfield deviceBitfield(1 | 1 << 1);
OsContextLinux osContext(drmMock, 0u, EngineDescriptorHelper::getDefaultDescriptor(deviceBitfield)); OsContextLinux osContext(drmMock, 0u, EngineDescriptorHelper::getDefaultDescriptor(deviceBitfield));
@@ -750,13 +750,12 @@ TEST(DrmTest, givenPerContextVMRequiredWhenCreatingOsContextsForRootDeviceThenIm
EXPECT_EQ(2 * 2u, drmMock.receivedContextParamRequestCount); EXPECT_EQ(2 * 2u, drmMock.receivedContextParamRequestCount);
auto &drmVmIds = osContext.getDrmVmIds(); auto &drmVmIds = osContext.getDrmVmIds();
EXPECT_EQ(32u, drmVmIds.size()); EXPECT_EQ(4u, drmVmIds.size());
EXPECT_EQ(20u, drmVmIds[0]); EXPECT_EQ(4u, drmVmIds[0]);
EXPECT_EQ(20u, drmVmIds[1]); EXPECT_EQ(4u, drmVmIds[1]);
EXPECT_EQ(0u, drmVmIds[2]); EXPECT_EQ(0u, drmVmIds[2]);
EXPECT_EQ(0u, drmVmIds[31]);
} }
TEST(DrmTest, givenNoPerContextVmsDrmWhenCreatingOsContextsThenVmIdIsNotQueriedAndStored) { TEST(DrmTest, givenNoPerContextVmsDrmWhenCreatingOsContextsThenVmIdIsNotQueriedAndStored) {

10
shared/test/unit_test/unified_memory/unified_memory_tests.cpp
View File

@@ -1,5 +1,5 @@
/* /*
* Copyright (C) 2019-2020 Intel Corporation * Copyright (C) 2019-2022 Intel Corporation
* *
* SPDX-License-Identifier: MIT * SPDX-License-Identifier: MIT
* *
@@ -12,10 +12,10 @@
#include <bitset> #include <bitset>
TEST(UnifiedMemoryTests, givenInternalMemoryTypesThenAllHaveOnlyOneBitSet) { TEST(UnifiedMemoryTests, givenInternalMemoryTypesThenAllHaveOnlyOneBitSet) {
EXPECT_EQ(1u, std::bitset<32>(InternalMemoryType::DEVICE_UNIFIED_MEMORY).count()); EXPECT_EQ(1u, std::bitset<4>(InternalMemoryType::DEVICE_UNIFIED_MEMORY).count());
EXPECT_EQ(1u, std::bitset<32>(InternalMemoryType::HOST_UNIFIED_MEMORY).count()); EXPECT_EQ(1u, std::bitset<4>(InternalMemoryType::HOST_UNIFIED_MEMORY).count());
EXPECT_EQ(1u, std::bitset<32>(InternalMemoryType::SHARED_UNIFIED_MEMORY).count()); EXPECT_EQ(1u, std::bitset<4>(InternalMemoryType::SHARED_UNIFIED_MEMORY).count());
EXPECT_EQ(1u, std::bitset<32>(InternalMemoryType::SVM).count()); EXPECT_EQ(1u, std::bitset<4>(InternalMemoryType::SVM).count());
} }
TEST(UnifiedMemoryTests, givenUnifiedMemoryControlsWhenDedicatedFieldsAreSetThenMaskIsProperlyGenerated) { TEST(UnifiedMemoryTests, givenUnifiedMemoryControlsWhenDedicatedFieldsAreSetThenMaskIsProperlyGenerated) {