intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2025-12-21 09:14:47 +08:00
Code Issues Packages Projects Releases Wiki Activity

Use correct Heap32 index for local memory allocations

Browse Source 
Change-Id: I068f712ab2b05ee3d5a9716b21de685a7fee3a02
Signed-off-by: Bartosz Dunajski <bartosz.dunajski@intel.com>
This commit is contained in:
Bartosz Dunajski
2020-07-01 14:03:46 +02:00
committed by sys_ocldev
parent 4076d3e294
commit 6dd07bd3bc
40 changed files with 99 additions and 77 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
level_zero/core/source/cmdqueue/cmdqueue_hw_base.inl
View File

@@ -53,7 +53,7 @@ void CommandQueueHw<gfxCoreFamily>::programGeneralStateBaseAddress(uint64_t gsba
gsba,
true,
(device->getMOCS(true, false) >> 1),
neoDevice->getMemoryManager()->getInternalHeapBaseAddress(device->getRootDeviceIndex()),
neoDevice->getMemoryManager()->getInternalHeapBaseAddress(device->getRootDeviceIndex(), false),
true,
neoDevice->getGmmHelper(),
false);

4
level_zero/core/test/unit_tests/sources/cmdqueue/test_cmdqueue.cpp
View File

@@ -124,7 +124,7 @@ using CommandQueueSBASupport = IsWithinProducts<IGFX_SKYLAKE, IGFX_TIGERLAKE_LP>
struct MockMemoryManagerCommandQueueSBA : public MemoryManagerMock {
MockMemoryManagerCommandQueueSBA(NEO::ExecutionEnvironment &executionEnvironment) : MemoryManagerMock(const_cast<NEO::ExecutionEnvironment &>(executionEnvironment)) {}
MOCK_METHOD1(getInternalHeapBaseAddress, uint64_t(uint32_t rootDeviceIndex));
MOCK_METHOD2(getInternalHeapBaseAddress, uint64_t(uint32_t rootDeviceIndex, bool useLocalMemory));
};
struct CommandQueueProgramSBATest : public ::testing::Test {
@@ -169,7 +169,7 @@ HWTEST2_F(CommandQueueProgramSBATest, whenCreatingCommandQueueThenItIsInitialize
uint32_t alignedSize = 4096u;
NEO::LinearStream child(commandQueue->commandStream->getSpace(alignedSize), alignedSize);
EXPECT_CALL(*memoryManager, getInternalHeapBaseAddress(rootDeviceIndex))
EXPECT_CALL(*memoryManager, getInternalHeapBaseAddress(rootDeviceIndex, false))
.Times(1);
commandQueue->programGeneralStateBaseAddress(0u, child);

5
opencl/source/memory_manager/os_agnostic_memory_manager.cpp
View File

@@ -102,8 +102,9 @@ GraphicsAllocation *OsAgnosticMemoryManager::allocateGraphicsMemory64kb(const Al
return memoryAllocation;
}
GraphicsAllocation *OsAgnosticMemoryManager::allocate32BitGraphicsMemoryImpl(const AllocationData &allocationData) {
auto heap = useInternal32BitAllocator(allocationData.type) ? HeapIndex::HEAP_INTERNAL_DEVICE_MEMORY : HeapIndex::HEAP_EXTERNAL;
GraphicsAllocation *OsAgnosticMemoryManager::allocate32BitGraphicsMemoryImpl(const AllocationData &allocationData, bool useLocalMemory) {
auto heap = useInternal32BitAllocator(allocationData.type) ? selectInternalHeap(useLocalMemory)
: selectExternalHeap(useLocalMemory);
auto gfxPartition = getGfxPartition(allocationData.rootDeviceIndex);
if (allocationData.hostPtr) {
auto allocationSize = alignSizeWholePage(allocationData.hostPtr, allocationData.size);

2
opencl/source/memory_manager/os_agnostic_memory_manager.h
View File

@@ -89,7 +89,7 @@ class OsAgnosticMemoryManager : public MemoryManager {
void *lockResourceImpl(GraphicsAllocation &graphicsAllocation) override { return graphicsAllocation.getUnderlyingBuffer(); }
void unlockResourceImpl(GraphicsAllocation &graphicsAllocation) override {}
GraphicsAllocation *allocate32BitGraphicsMemoryImpl(const AllocationData &allocationData) override;
GraphicsAllocation *allocate32BitGraphicsMemoryImpl(const AllocationData &allocationData, bool useLocalMemory) override;
GraphicsAllocation *allocateGraphicsMemoryInDevicePool(const AllocationData &allocationData, AllocationStatus &status) override;
MemoryAllocation *createMemoryAllocation(GraphicsAllocation::AllocationType allocationType, void *driverAllocatedCpuPointer, void *pMem, uint64_t gpuAddress, size_t memSize,
uint64_t count, MemoryPool::Type pool, uint32_t rootDeviceIndex, bool uncacheable, bool flushL3Required, bool requireSpecificBitness);

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

@@ -222,7 +222,7 @@ HWTEST_F(EnqueueCopyBufferRectTest, WhenCopyingBufferRect2DThenL3ProgrammingIsCo
HWCMDTEST_F(IGFX_GEN8_CORE, EnqueueCopyBufferRectTest, When2DEnqueueIsDoneThenStateBaseAddressIsProperlyProgrammed) {
enqueueCopyBufferRect2D<FamilyType>();
auto &ultCsr = this->pDevice->getUltCommandStreamReceiver<FamilyType>();
validateStateBaseAddress<FamilyType>(ultCsr.getMemoryManager()->getInternalHeapBaseAddress(ultCsr.rootDeviceIndex),
validateStateBaseAddress<FamilyType>(ultCsr.getMemoryManager()->getInternalHeapBaseAddress(ultCsr.rootDeviceIndex, pIOH->getGraphicsAllocation()->isAllocatedInLocalMemoryPool()),
pDSH, pIOH, pSSH, itorPipelineSelect, itorWalker, cmdList, 0llu);
}
@@ -342,7 +342,7 @@ HWTEST_F(EnqueueCopyBufferRectTest, WhenCopyingBufferRect3DThenL3ProgrammingIsCo
HWCMDTEST_F(IGFX_GEN8_CORE, EnqueueCopyBufferRectTest, When3DEnqueueIsDoneThenStateBaseAddressIsProperlyProgrammed) {
enqueueCopyBufferRect3D<FamilyType>();
auto &ultCsr = this->pDevice->getUltCommandStreamReceiver<FamilyType>();
validateStateBaseAddress<FamilyType>(ultCsr.getMemoryManager()->getInternalHeapBaseAddress(ultCsr.rootDeviceIndex),
validateStateBaseAddress<FamilyType>(ultCsr.getMemoryManager()->getInternalHeapBaseAddress(ultCsr.rootDeviceIndex, pIOH->getGraphicsAllocation()->isAllocatedInLocalMemoryPool()),
pDSH, pIOH, pSSH, itorPipelineSelect, itorWalker, cmdList, 0llu);
}

2
opencl/test/unit_test/command_queue/enqueue_copy_buffer_tests.cpp
View File

@@ -190,7 +190,7 @@ HWTEST_F(EnqueueCopyBufferTest, WhenCopyingBufferThenL3ProgrammingIsCorrect) {
HWCMDTEST_F(IGFX_GEN8_CORE, EnqueueCopyBufferTest, WhenEnqueueIsDoneThenStateBaseAddressIsProperlyProgrammed) {
enqueueCopyBufferAndParse<FamilyType>();
auto &ultCsr = this->pDevice->getUltCommandStreamReceiver<FamilyType>();
validateStateBaseAddress<FamilyType>(ultCsr.getMemoryManager()->getInternalHeapBaseAddress(ultCsr.rootDeviceIndex),
validateStateBaseAddress<FamilyType>(ultCsr.getMemoryManager()->getInternalHeapBaseAddress(ultCsr.rootDeviceIndex, pIOH->getGraphicsAllocation()->isAllocatedInLocalMemoryPool()),
pDSH, pIOH, pSSH, itorPipelineSelect, itorWalker, cmdList, 0llu);
}

2
opencl/test/unit_test/command_queue/enqueue_copy_buffer_to_image_tests.cpp
View File

@@ -96,7 +96,7 @@ HWTEST_F(EnqueueCopyBufferToImageTest, WhenCopyingBufferToImageThenL3Programming
HWCMDTEST_F(IGFX_GEN8_CORE, EnqueueCopyBufferToImageTest, WhenEnqueueIsDoneThenStateBaseAddressIsProperlyProgrammed) {
enqueueCopyBufferToImage<FamilyType>();
auto &ultCsr = this->pDevice->getUltCommandStreamReceiver<FamilyType>();
validateStateBaseAddress<FamilyType>(ultCsr.getMemoryManager()->getInternalHeapBaseAddress(ultCsr.rootDeviceIndex),
validateStateBaseAddress<FamilyType>(ultCsr.getMemoryManager()->getInternalHeapBaseAddress(ultCsr.rootDeviceIndex, pIOH->getGraphicsAllocation()->isAllocatedInLocalMemoryPool()),
pDSH, pIOH, pSSH, itorPipelineSelect, itorWalker, cmdList, 0llu);
}

2
opencl/test/unit_test/command_queue/enqueue_copy_image_tests.cpp
View File

@@ -98,7 +98,7 @@ HWTEST_F(EnqueueCopyImageTest, WhenCopyingImageThenL3ProgrammingIsCorrect) {
HWCMDTEST_F(IGFX_GEN8_CORE, EnqueueCopyImageTest, WhenEnqueueIsDoneThenStateBaseAddressIsProperlyProgrammed) {
enqueueCopyImage<FamilyType>();
auto &ultCsr = this->pDevice->getUltCommandStreamReceiver<FamilyType>();
validateStateBaseAddress<FamilyType>(ultCsr.getMemoryManager()->getInternalHeapBaseAddress(ultCsr.rootDeviceIndex),
validateStateBaseAddress<FamilyType>(ultCsr.getMemoryManager()->getInternalHeapBaseAddress(ultCsr.rootDeviceIndex, pIOH->getGraphicsAllocation()->isAllocatedInLocalMemoryPool()),
pDSH, pIOH, pSSH, itorPipelineSelect, itorWalker, cmdList, 0llu);
}

2
opencl/test/unit_test/command_queue/enqueue_copy_image_to_buffer_tests.cpp
View File

@@ -97,7 +97,7 @@ HWTEST_F(EnqueueCopyImageToBufferTest, WhenCopyingImageToBufferThenL3Programming
HWCMDTEST_F(IGFX_GEN8_CORE, EnqueueCopyImageToBufferTest, WhenEnqueueIsDoneThenStateBaseAddressIsProperlyProgrammed) {
enqueueCopyImageToBuffer<FamilyType>();
auto &ultCsr = this->pDevice->getUltCommandStreamReceiver<FamilyType>();
validateStateBaseAddress<FamilyType>(ultCsr.getMemoryManager()->getInternalHeapBaseAddress(ultCsr.rootDeviceIndex),
validateStateBaseAddress<FamilyType>(ultCsr.getMemoryManager()->getInternalHeapBaseAddress(ultCsr.rootDeviceIndex, pIOH->getGraphicsAllocation()->isAllocatedInLocalMemoryPool()),
pDSH, pIOH, pSSH, itorPipelineSelect, itorWalker, cmdList, 0llu);
}

2
opencl/test/unit_test/command_queue/enqueue_fill_buffer_tests.cpp
View File

@@ -209,7 +209,7 @@ HWTEST_F(EnqueueFillBufferCmdTests, WhenFillingBufferThenL3ProgrammingIsCorrect)
HWCMDTEST_F(IGFX_GEN8_CORE, EnqueueFillBufferCmdTests, WhenEnqueueIsDoneThenStateBaseAddressIsProperlyProgrammed) {
enqueueFillBuffer<FamilyType>();
auto &ultCsr = this->pDevice->getUltCommandStreamReceiver<FamilyType>();
validateStateBaseAddress<FamilyType>(ultCsr.getMemoryManager()->getInternalHeapBaseAddress(ultCsr.rootDeviceIndex),
validateStateBaseAddress<FamilyType>(ultCsr.getMemoryManager()->getInternalHeapBaseAddress(ultCsr.rootDeviceIndex, pIOH->getGraphicsAllocation()->isAllocatedInLocalMemoryPool()),
pDSH, pIOH, pSSH, itorPipelineSelect, itorWalker, cmdList, 0llu);
}

2
opencl/test/unit_test/command_queue/enqueue_fill_image_tests.cpp
View File

@@ -105,7 +105,7 @@ HWTEST_F(EnqueueFillImageTest, WhenFillingImageThenL3ProgrammingIsCorrect) {
HWCMDTEST_F(IGFX_GEN8_CORE, EnqueueFillImageTest, WhenEnqueueIsDoneThenStateBaseAddressIsProperlyProgrammed) {
enqueueFillImage<FamilyType>();
auto &ultCsr = this->pDevice->getUltCommandStreamReceiver<FamilyType>();
validateStateBaseAddress<FamilyType>(ultCsr.getMemoryManager()->getInternalHeapBaseAddress(ultCsr.rootDeviceIndex),
validateStateBaseAddress<FamilyType>(ultCsr.getMemoryManager()->getInternalHeapBaseAddress(ultCsr.rootDeviceIndex, pIOH->getGraphicsAllocation()->isAllocatedInLocalMemoryPool()),
pDSH, pIOH, pSSH, itorPipelineSelect, itorWalker, cmdList, 0llu);
}

6
opencl/test/unit_test/command_queue/enqueue_kernel_2_tests.cpp
View File

@@ -186,9 +186,9 @@ HWCMDTEST_P(IGFX_GEN8_CORE, EnqueueWorkItemTestsWithLimitedParamSet, LoadRegiste
HWCMDTEST_P(IGFX_GEN8_CORE, EnqueueWorkItemTestsWithLimitedParamSet, WhenEnqueueIsDoneThenStateBaseAddressIsProperlyProgrammed) {
enqueueKernel<FamilyType>();
auto &ultCsr = this->pDevice->getUltCommandStreamReceiver<FamilyType>();
validateStateBaseAddress<FamilyType>(ultCsr.getMemoryManager()->getInternalHeapBaseAddress(ultCsr.rootDeviceIndex),
validateStateBaseAddress<FamilyType>(ultCsr.getMemoryManager()->getInternalHeapBaseAddress(ultCsr.rootDeviceIndex, pIOH->getGraphicsAllocation()->isAllocatedInLocalMemoryPool()),
pDSH, pIOH, pSSH, itorPipelineSelect, itorWalker, cmdList,
context->getMemoryManager()->peekForce32BitAllocations() ? context->getMemoryManager()->getExternalHeapBaseAddress(ultCsr.rootDeviceIndex) : 0llu);
context->getMemoryManager()->peekForce32BitAllocations() ? context->getMemoryManager()->getExternalHeapBaseAddress(ultCsr.rootDeviceIndex, false) : 0llu);
}
HWCMDTEST_P(IGFX_GEN8_CORE, EnqueueWorkItemTestsWithLimitedParamSet, MediaInterfaceDescriptorLoad) {
@@ -510,7 +510,7 @@ HWCMDTEST_P(IGFX_GEN8_CORE, EnqueueKernelWithScratch, givenDeviceForcing32bitAll
auto GSHaddress = sba->getGeneralStateBaseAddress();
EXPECT_EQ(memoryManager->getExternalHeapBaseAddress(graphicsAllocation->getRootDeviceIndex()), GSHaddress);
EXPECT_EQ(memoryManager->getExternalHeapBaseAddress(graphicsAllocation->getRootDeviceIndex(), graphicsAllocation->isAllocatedInLocalMemoryPool()), GSHaddress);
//now re-try to see if SBA is not programmed

2
opencl/test/unit_test/command_queue/enqueue_read_buffer_rect_tests.cpp
View File

@@ -208,7 +208,7 @@ HWTEST_F(EnqueueReadBufferRectTest, WhenReadingBufferThenL3ProgrammingIsCorrect)
HWCMDTEST_F(IGFX_GEN8_CORE, EnqueueReadBufferRectTest, When2DEnqueueIsDoneThenStateBaseAddressIsProperlyProgrammed) {
enqueueReadBufferRect2D<FamilyType>();
auto &ultCsr = this->pDevice->getUltCommandStreamReceiver<FamilyType>();
validateStateBaseAddress<FamilyType>(ultCsr.getMemoryManager()->getInternalHeapBaseAddress(ultCsr.rootDeviceIndex),
validateStateBaseAddress<FamilyType>(ultCsr.getMemoryManager()->getInternalHeapBaseAddress(ultCsr.rootDeviceIndex, pIOH->getGraphicsAllocation()->isAllocatedInLocalMemoryPool()),
pDSH, pIOH, pSSH, itorPipelineSelect, itorWalker, cmdList, 0llu);
}

2
opencl/test/unit_test/command_queue/enqueue_read_buffer_tests.cpp
View File

@@ -172,7 +172,7 @@ HWCMDTEST_F(IGFX_GEN8_CORE, EnqueueReadBufferTypeTest, WhenEnqueueIsDoneThenStat
srcBuffer->forceDisallowCPUCopy = true;
enqueueReadBuffer<FamilyType>();
auto &ultCsr = this->pDevice->getUltCommandStreamReceiver<FamilyType>();
validateStateBaseAddress<FamilyType>(ultCsr.getMemoryManager()->getInternalHeapBaseAddress(ultCsr.rootDeviceIndex),
validateStateBaseAddress<FamilyType>(ultCsr.getMemoryManager()->getInternalHeapBaseAddress(ultCsr.rootDeviceIndex, pIOH->getGraphicsAllocation()->isAllocatedInLocalMemoryPool()),
pDSH, pIOH, pSSH, itorPipelineSelect, itorWalker, cmdList, 0llu);
}

2
opencl/test/unit_test/command_queue/enqueue_read_image_tests.cpp
View File

@@ -108,7 +108,7 @@ HWTEST_F(EnqueueReadImageTest, WhenReadingImageThenL3ProgrammingIsCorrect) {
HWCMDTEST_F(IGFX_GEN8_CORE, EnqueueReadImageTest, WhenEnqueueIsDoneThenStateBaseAddressIsProperlyProgrammed) {
enqueueReadImage<FamilyType>();
auto &ultCsr = this->pDevice->getUltCommandStreamReceiver<FamilyType>();
validateStateBaseAddress<FamilyType>(ultCsr.getMemoryManager()->getInternalHeapBaseAddress(ultCsr.rootDeviceIndex),
validateStateBaseAddress<FamilyType>(ultCsr.getMemoryManager()->getInternalHeapBaseAddress(ultCsr.rootDeviceIndex, pIOH->getGraphicsAllocation()->isAllocatedInLocalMemoryPool()),
pDSH, pIOH, pSSH, itorPipelineSelect, itorWalker, cmdList, 0llu);
}

2
opencl/test/unit_test/command_queue/enqueue_write_buffer_rect_tests.cpp
View File

@@ -182,7 +182,7 @@ HWTEST_F(EnqueueWriteBufferRectTest, WhenWritingBufferThenL3ProgrammingIsCorrect
HWCMDTEST_F(IGFX_GEN8_CORE, EnqueueWriteBufferRectTest, When2DEnqueueIsDoneThenStateBaseAddressIsProperlyProgrammed) {
enqueueWriteBufferRect2D<FamilyType>();
auto &ultCsr = this->pDevice->getUltCommandStreamReceiver<FamilyType>();
validateStateBaseAddress<FamilyType>(ultCsr.getMemoryManager()->getInternalHeapBaseAddress(ultCsr.rootDeviceIndex),
validateStateBaseAddress<FamilyType>(ultCsr.getMemoryManager()->getInternalHeapBaseAddress(ultCsr.rootDeviceIndex, pIOH->getGraphicsAllocation()->isAllocatedInLocalMemoryPool()),
pDSH, pIOH, pSSH, itorPipelineSelect, itorWalker, cmdList, 0llu);
}

2
opencl/test/unit_test/command_queue/enqueue_write_buffer_tests.cpp
View File

@@ -170,7 +170,7 @@ HWCMDTEST_F(IGFX_GEN8_CORE, EnqueueWriteBufferTypeTest, WhenEnqueueIsDoneThenSta
srcBuffer->forceDisallowCPUCopy = true;
enqueueWriteBuffer<FamilyType>();
auto &ultCsr = this->pDevice->getUltCommandStreamReceiver<FamilyType>();
validateStateBaseAddress<FamilyType>(ultCsr.getMemoryManager()->getInternalHeapBaseAddress(ultCsr.rootDeviceIndex),
validateStateBaseAddress<FamilyType>(ultCsr.getMemoryManager()->getInternalHeapBaseAddress(ultCsr.rootDeviceIndex, pIOH->getGraphicsAllocation()->isAllocatedInLocalMemoryPool()),
pDSH, pIOH, pSSH, itorPipelineSelect, itorWalker, cmdList, 0llu);
}

2
opencl/test/unit_test/command_queue/enqueue_write_image_tests.cpp
View File

@@ -108,7 +108,7 @@ HWTEST_F(EnqueueWriteImageTest, WhenWritingImageThenL3ProgrammingIsCorrect) {
HWCMDTEST_F(IGFX_GEN8_CORE, EnqueueWriteImageTest, WhenEnqueueIsDoneThenStateBaseAddressIsProperlyProgrammed) {
enqueueWriteImage<FamilyType>();
auto &ultCsr = this->pDevice->getUltCommandStreamReceiver<FamilyType>();
validateStateBaseAddress<FamilyType>(ultCsr.getMemoryManager()->getInternalHeapBaseAddress(ultCsr.rootDeviceIndex),
validateStateBaseAddress<FamilyType>(ultCsr.getMemoryManager()->getInternalHeapBaseAddress(ultCsr.rootDeviceIndex, pIOH->getGraphicsAllocation()->isAllocatedInLocalMemoryPool()),
pDSH, pIOH, pSSH, itorPipelineSelect, itorWalker, cmdList, 0llu);
}

2
opencl/test/unit_test/command_stream/command_stream_receiver_flush_task_1_tests.cpp
View File

@@ -457,7 +457,7 @@ HWTEST_F(CommandStreamReceiverFlushTaskTests, stateBaseAddressProgrammingShouldM
auto &cmd = *reinterpret_cast<STATE_BASE_ADDRESS *>(cmdStateBaseAddress);
EXPECT_EQ(dsh.getCpuBase(), reinterpret_cast<void *>(cmd.getDynamicStateBaseAddress()));
EXPECT_EQ(commandStreamReceiver.getMemoryManager()->getInternalHeapBaseAddress(commandStreamReceiver.rootDeviceIndex), cmd.getInstructionBaseAddress());
EXPECT_EQ(commandStreamReceiver.getMemoryManager()->getInternalHeapBaseAddress(commandStreamReceiver.rootDeviceIndex, ioh.getGraphicsAllocation()->isAllocatedInLocalMemoryPool()), cmd.getInstructionBaseAddress());
EXPECT_EQ(ioh.getCpuBase(), reinterpret_cast<void *>(cmd.getIndirectObjectBaseAddress()));
EXPECT_EQ(ssh.getCpuBase(), reinterpret_cast<void *>(cmd.getSurfaceStateBaseAddress()));

4
opencl/test/unit_test/command_stream/command_stream_receiver_flush_task_2_tests.cpp
View File

@@ -638,7 +638,7 @@ HWCMDTEST_F(IGFX_GEN8_CORE, CommandStreamReceiverFlushTaskTests, givenTwoConsecu
}
if (sharedDeviceInfo.force32BitAddressess) {
EXPECT_EQ(pDevice->getMemoryManager()->getExternalHeapBaseAddress(graphicsAllocationScratch->getRootDeviceIndex()), GSHaddress);
EXPECT_EQ(pDevice->getMemoryManager()->getExternalHeapBaseAddress(graphicsAllocationScratch->getRootDeviceIndex(), false), GSHaddress);
} else {
if (is64bit) {
EXPECT_EQ(graphicsAddress - ScratchSpaceConstants::scratchSpaceOffsetFor64Bit, GSHaddress);
@@ -752,7 +752,7 @@ HWCMDTEST_F(IGFX_GEN8_CORE, CommandStreamReceiverFlushTaskTests, givenNDRangeKer
EXPECT_EQ(highPartGraphicsAddress, scratchBaseHighPart);
if (sharedDeviceInfo.force32BitAddressess) {
EXPECT_EQ(pDevice->getMemoryManager()->getExternalHeapBaseAddress(graphicsAllocationScratch->getRootDeviceIndex()), GSHaddress);
EXPECT_EQ(pDevice->getMemoryManager()->getExternalHeapBaseAddress(graphicsAllocationScratch->getRootDeviceIndex(), false), GSHaddress);
} else {
if (is64bit) {
EXPECT_EQ(graphicsAddress - ScratchSpaceConstants::scratchSpaceOffsetFor64Bit, GSHaddress);

16
opencl/test/unit_test/memory_manager/memory_manager_tests.cpp
View File

@@ -76,6 +76,14 @@ TEST(MemoryManagerTest, givenDebugVariableWhenCreatingMemoryManagerThenSetSuppor
}
}
TEST(MemoryManagerTest, givenLocalMemoryRequiredWhenSelectingHeapThenPickDeviceHeapIndex) {
EXPECT_EQ(HeapIndex::HEAP_INTERNAL_DEVICE_MEMORY, MemoryManager::selectInternalHeap(true));
EXPECT_EQ(HeapIndex::HEAP_INTERNAL, MemoryManager::selectInternalHeap(false));
EXPECT_EQ(HeapIndex::HEAP_EXTERNAL_DEVICE_MEMORY, MemoryManager::selectExternalHeap(true));
EXPECT_EQ(HeapIndex::HEAP_EXTERNAL, MemoryManager::selectExternalHeap(false));
}
TEST(MemoryManagerTest, whenCreatingAllocPropertiesForMultiStorageResourceThenMultiStorageResourcesFlagIsSetToTrue) {
AllocationProperties properties{0, false, 0u, GraphicsAllocation::AllocationType::SCRATCH_SURFACE, false, true, 0};
EXPECT_TRUE(properties.multiStorageResource);
@@ -649,7 +657,7 @@ TEST(OsAgnosticMemoryManager, givenMemoryManagerWhenAskedFor32BitAllocationWhenL
auto gfxAllocation = memoryManager.allocateGraphicsMemoryWithAlignment(allocationData);
ASSERT_NE(gfxAllocation, nullptr);
EXPECT_NE(gfxAllocation->getGpuBaseAddress(), 0ull);
EXPECT_EQ(gfxAllocation->getGpuBaseAddress(), memoryManager.getExternalHeapBaseAddress(gfxAllocation->getRootDeviceIndex()));
EXPECT_EQ(gfxAllocation->getGpuBaseAddress(), memoryManager.getExternalHeapBaseAddress(gfxAllocation->getRootDeviceIndex(), gfxAllocation->isAllocatedInLocalMemoryPool()));
memoryManager.freeGraphicsMemory(gfxAllocation);
}
@@ -2029,19 +2037,19 @@ TEST(MemoryManagerTest, givenForcedLinearImages3DImageAndProperDescriptorValuesW
TEST(HeapSelectorTest, given32bitInternalAllocationWhenSelectingHeapThenInternalHeapIsUsed) {
GraphicsAllocation allocation{0, GraphicsAllocation::AllocationType::KERNEL_ISA, nullptr, 0, 0, 0, MemoryPool::MemoryNull};
allocation.set32BitAllocation(true);
EXPECT_EQ(HeapIndex::HEAP_INTERNAL_DEVICE_MEMORY, MemoryManager::selectHeap(&allocation, false, false));
EXPECT_EQ(MemoryManager::selectInternalHeap(allocation.isAllocatedInLocalMemoryPool()), MemoryManager::selectHeap(&allocation, false, false));
}
TEST(HeapSelectorTest, givenNon32bitInternalAllocationWhenSelectingHeapThenInternalHeapIsUsed) {
GraphicsAllocation allocation{0, GraphicsAllocation::AllocationType::KERNEL_ISA, nullptr, 0, 0, 0, MemoryPool::MemoryNull};
allocation.set32BitAllocation(false);
EXPECT_EQ(HeapIndex::HEAP_INTERNAL_DEVICE_MEMORY, MemoryManager::selectHeap(&allocation, false, false));
EXPECT_EQ(MemoryManager::selectInternalHeap(allocation.isAllocatedInLocalMemoryPool()), MemoryManager::selectHeap(&allocation, false, false));
}
TEST(HeapSelectorTest, given32bitExternalAllocationWhenSelectingHeapThenExternalHeapIsUsed) {
GraphicsAllocation allocation{0, GraphicsAllocation::AllocationType::UNKNOWN, nullptr, 0, 0, 0, MemoryPool::MemoryNull};
allocation.set32BitAllocation(true);
EXPECT_EQ(HeapIndex::HEAP_EXTERNAL, MemoryManager::selectHeap(&allocation, false, false));
EXPECT_EQ(MemoryManager::selectExternalHeap(allocation.isAllocatedInLocalMemoryPool()), MemoryManager::selectHeap(&allocation, false, false));
}
TEST(HeapSelectorTest, givenLimitedAddressSpaceWhenSelectingHeapForExternalAllocationThenStandardHeapIsUsed) {

3
opencl/test/unit_test/mocks/linux/mock_drm_memory_manager.cpp
View File

@@ -69,7 +69,8 @@ DrmAllocation *TestedDrmMemoryManager::allocate32BitGraphicsMemory(uint32_t root
AllocationData allocationData;
MockAllocationProperties properties(rootDeviceIndex, allocateMemory, size, allocationType);
getAllocationData(allocationData, properties, ptr, createStorageInfoFromProperties(properties));
return allocate32BitGraphicsMemoryImpl(allocationData);
bool useLocalMemory = !allocationData.flags.useSystemMemory && this->localMemorySupported[rootDeviceIndex];
return allocate32BitGraphicsMemoryImpl(allocationData, useLocalMemory);
}
TestedDrmMemoryManager::~TestedDrmMemoryManager() {
DrmMemoryManager::commonCleanup();

7
opencl/test/unit_test/mocks/mock_memory_manager.cpp
View File

@@ -102,14 +102,15 @@ GraphicsAllocation *MockMemoryManager::allocate32BitGraphicsMemory(uint32_t root
AllocationData allocationData{};
MockAllocationProperties properties(rootDeviceIndex, allocateMemory, size, allocationType);
getAllocationData(allocationData, properties, ptr, createStorageInfoFromProperties(properties));
return allocate32BitGraphicsMemoryImpl(allocationData);
bool useLocalMemory = !allocationData.flags.useSystemMemory && this->localMemorySupported[rootDeviceIndex];
return allocate32BitGraphicsMemoryImpl(allocationData, useLocalMemory);
}
GraphicsAllocation *MockMemoryManager::allocate32BitGraphicsMemoryImpl(const AllocationData &allocationData) {
GraphicsAllocation *MockMemoryManager::allocate32BitGraphicsMemoryImpl(const AllocationData &allocationData, bool useLocalMemory) {
if (failAllocate32Bit) {
return nullptr;
}
return OsAgnosticMemoryManager::allocate32BitGraphicsMemoryImpl(allocationData);
return OsAgnosticMemoryManager::allocate32BitGraphicsMemoryImpl(allocationData, useLocalMemory);
}
FailMemoryManager::FailMemoryManager(int32_t failedAllocationsCount, ExecutionEnvironment &executionEnvironment) : MockMemoryManager(executionEnvironment) {

4
opencl/test/unit_test/mocks/mock_memory_manager.h
View File

@@ -107,7 +107,7 @@ class MockMemoryManager : public MemoryManagerCreate<OsAgnosticMemoryManager> {
bool isCpuCopyRequired(const void *ptr) override { return cpuCopyRequired; }
GraphicsAllocation *allocate32BitGraphicsMemory(uint32_t rootDeviceIndex, size_t size, const void *ptr, GraphicsAllocation::AllocationType allocationType);
GraphicsAllocation *allocate32BitGraphicsMemoryImpl(const AllocationData &allocationData) override;
GraphicsAllocation *allocate32BitGraphicsMemoryImpl(const AllocationData &allocationData, bool useLocalMemory) override;
void forceLimitedRangeAllocator(uint32_t rootDeviceIndex, uint64_t range) { getGfxPartition(rootDeviceIndex)->init(range, 0, 0, gfxPartitions.size()); }
@@ -186,7 +186,7 @@ class FailMemoryManager : public MockMemoryManager {
GraphicsAllocation *allocateGraphicsMemoryWithProperties(const AllocationProperties &properties, const void *ptr) override {
return nullptr;
}
GraphicsAllocation *allocate32BitGraphicsMemoryImpl(const AllocationData &allocationData) override {
GraphicsAllocation *allocate32BitGraphicsMemoryImpl(const AllocationData &allocationData, bool useLocalMemory) override {
return nullptr;
}
GraphicsAllocation *createGraphicsAllocationFromSharedHandle(osHandle handle, const AllocationProperties &properties, bool requireSpecificBitness) override {

12
opencl/test/unit_test/os_interface/linux/drm_memory_manager_tests.cpp
View File

@@ -926,7 +926,7 @@ TEST_F(DrmMemoryManagerTest, givenMemoryManagerWhenAskedFor32BitAllocationThen32
EXPECT_LT(address64bit, MemoryConstants::max32BitAddress);
EXPECT_TRUE(allocation->is32BitAllocation());
EXPECT_EQ(GmmHelper::canonize(memoryManager->getExternalHeapBaseAddress(allocation->getRootDeviceIndex())), allocation->getGpuBaseAddress());
EXPECT_EQ(GmmHelper::canonize(memoryManager->getExternalHeapBaseAddress(allocation->getRootDeviceIndex(), allocation->isAllocatedInLocalMemoryPool())), allocation->getGpuBaseAddress());
memoryManager->freeGraphicsMemory(allocation);
}
@@ -2029,7 +2029,7 @@ TEST_F(DrmMemoryManagerTest, given32BitAddressingWhenBufferFromSharedHandleAndBi
auto drmAllocation = static_cast<DrmAllocation *>(graphicsAllocation);
EXPECT_TRUE(graphicsAllocation->is32BitAllocation());
EXPECT_EQ(1, lseekCalledCount);
EXPECT_EQ(GmmHelper::canonize(memoryManager->getExternalHeapBaseAddress(graphicsAllocation->getRootDeviceIndex())), drmAllocation->getGpuBaseAddress());
EXPECT_EQ(GmmHelper::canonize(memoryManager->getExternalHeapBaseAddress(graphicsAllocation->getRootDeviceIndex(), drmAllocation->isAllocatedInLocalMemoryPool())), drmAllocation->getGpuBaseAddress());
memoryManager->freeGraphicsMemory(graphicsAllocation);
}
@@ -2482,7 +2482,7 @@ TEST_F(DrmMemoryManagerTest, givenMemoryManagerWhenAskedForInternalAllocationWit
auto gpuPtr = drmAllocation->getGpuAddress();
auto heapBase = GmmHelper::canonize(memoryManager->getInternalHeapBaseAddress(drmAllocation->getRootDeviceIndex()));
auto heapBase = GmmHelper::canonize(memoryManager->getInternalHeapBaseAddress(drmAllocation->getRootDeviceIndex(), drmAllocation->isAllocatedInLocalMemoryPool()));
auto heapSize = 4 * GB;
EXPECT_GE(gpuPtr, heapBase);
@@ -2515,7 +2515,7 @@ TEST_F(DrmMemoryManagerTest, givenLimitedRangeAllocatorWhenAskedForInternalAlloc
auto gpuPtr = drmAllocation->getGpuAddress();
auto heapBase = GmmHelper::canonize(memoryManager->getInternalHeapBaseAddress(drmAllocation->getRootDeviceIndex()));
auto heapBase = GmmHelper::canonize(memoryManager->getInternalHeapBaseAddress(drmAllocation->getRootDeviceIndex(), drmAllocation->isAllocatedInLocalMemoryPool()));
auto heapSize = 4 * GB;
EXPECT_GE(gpuPtr, heapBase);
@@ -2572,7 +2572,7 @@ TEST_F(DrmMemoryManagerTest, givenMemoryManagerWhenAskedForInternalAllocationWit
auto gpuPtr = drmAllocation->getGpuAddress();
auto heapBase = GmmHelper::canonize(memoryManager->getInternalHeapBaseAddress(drmAllocation->getRootDeviceIndex()));
auto heapBase = GmmHelper::canonize(memoryManager->getInternalHeapBaseAddress(drmAllocation->getRootDeviceIndex(), drmAllocation->isAllocatedInLocalMemoryPool()));
auto heapSize = 4 * GB;
EXPECT_GE(gpuPtr, heapBase);
@@ -2689,7 +2689,7 @@ TEST_F(DrmMemoryManagerBasic, givenDefaultDrmMemoryManagerWhenItIsQueriedForInte
true,
executionEnvironment));
auto heapBase = memoryManager->getGfxPartition(rootDeviceIndex)->getHeapBase(HeapIndex::HEAP_INTERNAL_DEVICE_MEMORY);
EXPECT_EQ(heapBase, memoryManager->getInternalHeapBaseAddress(rootDeviceIndex));
EXPECT_EQ(heapBase, memoryManager->getInternalHeapBaseAddress(rootDeviceIndex, true));
}
TEST_F(DrmMemoryManagerBasic, givenMemoryManagerWithEnabledHostMemoryValidationWhenFeatureIsQueriedThenTrueIsReturned) {

3
opencl/test/unit_test/os_interface/windows/mock_wddm_memory_manager.h
View File

@@ -52,7 +52,8 @@ class MockWddmMemoryManager : public MemoryManagerCreate<WddmMemoryManager> {
AllocationData allocationData;
MockAllocationProperties properties(rootDeviceIndex, allocateMemory, size, allocationType);
getAllocationData(allocationData, properties, ptr, createStorageInfoFromProperties(properties));
return allocate32BitGraphicsMemoryImpl(allocationData);
bool useLocalMemory = !allocationData.flags.useSystemMemory && this->localMemorySupported[rootDeviceIndex];
return allocate32BitGraphicsMemoryImpl(allocationData, useLocalMemory);
}
void freeGraphicsMemoryImpl(GraphicsAllocation *gfxAllocation) override {

20
opencl/test/unit_test/os_interface/windows/wddm_memory_manager_tests.cpp
View File

@@ -122,7 +122,7 @@ TEST(WddmMemoryManagerExternalHeapTest, WhenExternalHeapIsCreatedThenItHasCorrec
std::unique_ptr<WddmMemoryManager> memoryManager = std::unique_ptr<WddmMemoryManager>(new WddmMemoryManager(*executionEnvironment));
EXPECT_EQ(base, memoryManager->getExternalHeapBaseAddress(0));
EXPECT_EQ(base, memoryManager->getExternalHeapBaseAddress(0, false));
}
TEST(WddmMemoryManagerWithDeferredDeleterTest, givenWmmWhenAsyncDeleterIsEnabledAndWaitForDeletionsIsCalledThenDeleterInWddmIsSetToNullptr) {
@@ -566,7 +566,7 @@ TEST_F(WddmMemoryManagerTest, GivenForce32bitAddressingAndRequireSpecificBitness
if (is64bit) {
EXPECT_TRUE(gpuAllocation->is32BitAllocation());
uint64_t base = memoryManager->getExternalHeapBaseAddress(gpuAllocation->getRootDeviceIndex());
uint64_t base = memoryManager->getExternalHeapBaseAddress(gpuAllocation->getRootDeviceIndex(), gpuAllocation->isAllocatedInLocalMemoryPool());
EXPECT_EQ(GmmHelper::canonize(base), gpuAllocation->getGpuBaseAddress());
}
@@ -970,7 +970,7 @@ TEST_F(WddmMemoryManagerTest, WhenAllocating32BitMemoryThenGpuBaseAddressIsCanno
ASSERT_NE(nullptr, gpuAllocation);
uint64_t cannonizedAddress = GmmHelper::canonize(memoryManager->getGfxPartition(0)->getHeapBase(HeapIndex::HEAP_EXTERNAL));
uint64_t cannonizedAddress = GmmHelper::canonize(memoryManager->getGfxPartition(0)->getHeapBase(MemoryManager::selectExternalHeap(gpuAllocation->isAllocatedInLocalMemoryPool())));
EXPECT_EQ(cannonizedAddress, gpuAllocation->getGpuBaseAddress());
memoryManager->freeGraphicsMemory(gpuAllocation);
@@ -1087,12 +1087,12 @@ TEST_F(WddmMemoryManagerTest, givenManagerWithEnabledDeferredDeleterWhenFirstAnd
TEST_F(WddmMemoryManagerTest, givenNullPtrAndSizePassedToCreateInternalAllocationWhenCallIsMadeThenAllocationIsCreatedIn32BitHeapInternal) {
auto wddmAllocation = static_cast<WddmAllocation *>(memoryManager->allocate32BitGraphicsMemory(rootDeviceIndex, MemoryConstants::pageSize, nullptr, GraphicsAllocation::AllocationType::INTERNAL_HEAP));
ASSERT_NE(nullptr, wddmAllocation);
EXPECT_EQ(wddmAllocation->getGpuBaseAddress(), GmmHelper::canonize(memoryManager->getInternalHeapBaseAddress(wddmAllocation->getRootDeviceIndex())));
EXPECT_EQ(wddmAllocation->getGpuBaseAddress(), GmmHelper::canonize(memoryManager->getInternalHeapBaseAddress(wddmAllocation->getRootDeviceIndex(), wddmAllocation->isAllocatedInLocalMemoryPool())));
EXPECT_NE(nullptr, wddmAllocation->getUnderlyingBuffer());
EXPECT_EQ(4096u, wddmAllocation->getUnderlyingBufferSize());
EXPECT_NE((uint64_t)wddmAllocation->getUnderlyingBuffer(), wddmAllocation->getGpuAddress());
auto cannonizedHeapBase = GmmHelper::canonize(memoryManager->getInternalHeapBaseAddress(rootDeviceIndex));
auto cannonizedHeapEnd = GmmHelper::canonize(memoryManager->getGfxPartition(rootDeviceIndex)->getHeapLimit(HeapIndex::HEAP_INTERNAL_DEVICE_MEMORY));
auto cannonizedHeapBase = GmmHelper::canonize(memoryManager->getInternalHeapBaseAddress(rootDeviceIndex, wddmAllocation->isAllocatedInLocalMemoryPool()));
auto cannonizedHeapEnd = GmmHelper::canonize(memoryManager->getGfxPartition(rootDeviceIndex)->getHeapLimit(MemoryManager::selectInternalHeap(wddmAllocation->isAllocatedInLocalMemoryPool())));
EXPECT_GT(wddmAllocation->getGpuAddress(), cannonizedHeapBase);
EXPECT_LT(wddmAllocation->getGpuAddress() + wddmAllocation->getUnderlyingBufferSize(), cannonizedHeapEnd);
@@ -1106,12 +1106,12 @@ TEST_F(WddmMemoryManagerTest, givenPtrAndSizePassedToCreateInternalAllocationWhe
auto ptr = reinterpret_cast<void *>(0x1000000);
auto wddmAllocation = static_cast<WddmAllocation *>(memoryManager->allocate32BitGraphicsMemory(rootDeviceIndex, MemoryConstants::pageSize, ptr, GraphicsAllocation::AllocationType::INTERNAL_HEAP));
ASSERT_NE(nullptr, wddmAllocation);
EXPECT_EQ(wddmAllocation->getGpuBaseAddress(), GmmHelper::canonize(memoryManager->getInternalHeapBaseAddress(rootDeviceIndex)));
EXPECT_EQ(wddmAllocation->getGpuBaseAddress(), GmmHelper::canonize(memoryManager->getInternalHeapBaseAddress(rootDeviceIndex, wddmAllocation->isAllocatedInLocalMemoryPool())));
EXPECT_EQ(ptr, wddmAllocation->getUnderlyingBuffer());
EXPECT_EQ(4096u, wddmAllocation->getUnderlyingBufferSize());
EXPECT_NE((uint64_t)wddmAllocation->getUnderlyingBuffer(), wddmAllocation->getGpuAddress());
auto cannonizedHeapBase = GmmHelper::canonize(memoryManager->getInternalHeapBaseAddress(rootDeviceIndex));
auto cannonizedHeapEnd = GmmHelper::canonize(memoryManager->getGfxPartition(rootDeviceIndex)->getHeapLimit(HeapIndex::HEAP_INTERNAL_DEVICE_MEMORY));
auto cannonizedHeapBase = GmmHelper::canonize(memoryManager->getInternalHeapBaseAddress(rootDeviceIndex, wddmAllocation->isAllocatedInLocalMemoryPool()));
auto cannonizedHeapEnd = GmmHelper::canonize(memoryManager->getGfxPartition(rootDeviceIndex)->getHeapLimit(MemoryManager::selectInternalHeap(wddmAllocation->isAllocatedInLocalMemoryPool())));
EXPECT_GT(wddmAllocation->getGpuAddress(), cannonizedHeapBase);
EXPECT_LT(wddmAllocation->getGpuAddress() + wddmAllocation->getUnderlyingBufferSize(), cannonizedHeapEnd);
@@ -1386,7 +1386,7 @@ TEST(WddmMemoryManagerDefaults, givenDefaultWddmMemoryManagerWhenItIsQueriedForI
wddm->init();
MockWddmMemoryManager memoryManager(*executionEnvironment);
auto heapBase = wddm->getGfxPartition().Heap32[static_cast<uint32_t>(HeapIndex::HEAP_INTERNAL_DEVICE_MEMORY)].Base;
EXPECT_EQ(heapBase, memoryManager.getInternalHeapBaseAddress(0));
EXPECT_EQ(heapBase, memoryManager.getInternalHeapBaseAddress(0, true));
}
TEST_F(MockWddmMemoryManagerTest, givenValidateAllocationFunctionWhenItIsCalledWithTripleAllocationThenSuccessIsReturned) {

2
opencl/test/unit_test/program/kernel_info_tests.cpp
View File

@@ -128,7 +128,7 @@ TEST(KernelInfoTest, givenKernelInfoWhenCreateKernelAllocationThenCopyWholeKerne
class MyMemoryManager : public OsAgnosticMemoryManager {
public:
using OsAgnosticMemoryManager::OsAgnosticMemoryManager;
GraphicsAllocation *allocate32BitGraphicsMemoryImpl(const AllocationData &allocationData) override { return nullptr; }
GraphicsAllocation *allocate32BitGraphicsMemoryImpl(const AllocationData &allocationData, bool useLocalMemory) override { return nullptr; }
};
TEST(KernelInfoTest, givenKernelInfoWhenCreateKernelAllocationAndCannotAllocateMemoryThenReturnsFalse) {

2
opencl/test/unit_test/program/program_tests.cpp
View File

@@ -652,7 +652,7 @@ TEST_P(ProgramFromBinaryTest, givenProgramWhenItIsBeingBuildThenItContainsGraphi
EXPECT_NE(kernelInfo->heapInfo.pKernelHeap, kernelIsa);
EXPECT_EQ(0, memcmp(kernelIsa, kernelInfo->heapInfo.pKernelHeap, kernelInfo->heapInfo.KernelHeapSize));
auto rootDeviceIndex = graphicsAllocation->getRootDeviceIndex();
EXPECT_EQ(GmmHelper::decanonize(graphicsAllocation->getGpuBaseAddress()), pProgram->getDevice().getMemoryManager()->getInternalHeapBaseAddress(rootDeviceIndex));
EXPECT_EQ(GmmHelper::decanonize(graphicsAllocation->getGpuBaseAddress()), pProgram->getDevice().getMemoryManager()->getInternalHeapBaseAddress(rootDeviceIndex, graphicsAllocation->isAllocatedInLocalMemoryPool()));
}
TEST_P(ProgramFromBinaryTest, whenProgramIsBeingRebuildThenOutdatedGlobalBuffersAreFreed) {

2
shared/source/command_container/cmdcontainer.cpp
View File

@@ -82,7 +82,7 @@ bool CommandContainer::initialize(Device *device) {
indirectHeaps[i] = std::make_unique<IndirectHeap>(allocationIndirectHeaps[i], requireInternalHeap);
}
instructionHeapBaseAddress = device->getMemoryManager()->getInternalHeapBaseAddress(device->getRootDeviceIndex());
instructionHeapBaseAddress = device->getMemoryManager()->getInternalHeapBaseAddress(device->getRootDeviceIndex(), allocationIndirectHeaps[IndirectHeap::Type::INDIRECT_OBJECT]->isAllocatedInLocalMemoryPool());
iddBlock = nullptr;
nextIddInBlock = this->getNumIddPerBlock();

5
shared/source/command_stream/command_stream_receiver_hw_base.inl
View File

@@ -350,7 +350,8 @@ CompletionStamp CommandStreamReceiverHw<GfxFamily>::flushTask(
if (is64bit && scratchSpaceController->getScratchSpaceAllocation() && !force32BitAllocations) {
newGSHbase = scratchSpaceController->calculateNewGSH();
} else if (is64bit && force32BitAllocations && dispatchFlags.gsba32BitRequired) {
newGSHbase = getMemoryManager()->getExternalHeapBaseAddress(rootDeviceIndex);
bool useLocalMemory = scratchSpaceController->getScratchSpaceAllocation() ? scratchSpaceController->getScratchSpaceAllocation()->isAllocatedInLocalMemoryPool() : false;
newGSHbase = getMemoryManager()->getExternalHeapBaseAddress(rootDeviceIndex, useLocalMemory);
GSBAFor32BitProgrammed = true;
}
@@ -364,7 +365,7 @@ CompletionStamp CommandStreamReceiverHw<GfxFamily>::flushTask(
newGSHbase,
true,
mocsIndex,
getMemoryManager()->getInternalHeapBaseAddress(rootDeviceIndex),
getMemoryManager()->getInternalHeapBaseAddress(rootDeviceIndex, ioh.getGraphicsAllocation()->isAllocatedInLocalMemoryPool()),
true,
device.getGmmHelper(),
isMultiOsContextCapable());

2
shared/source/memory_manager/graphics_allocation.h
View File

@@ -233,6 +233,8 @@ class GraphicsAllocation : public IDNode<GraphicsAllocation> {
uint32_t getUsedPageSize() const;
bool isAllocatedInLocalMemoryPool() const { return (this->memoryPool == MemoryPool::LocalMemory); }
OsHandleStorage fragmentsStorage;
StorageInfo storageInfo = {};

6
shared/source/memory_manager/memory_manager.cpp
View File

@@ -401,7 +401,7 @@ GraphicsAllocation *MemoryManager::allocateGraphicsMemory(const AllocationData &
}
if (useInternal32BitAllocator(allocationData.type) ||
(force32bitAllocations && allocationData.flags.allow32Bit && is64bit)) {
return allocate32BitGraphicsMemoryImpl(allocationData);
return allocate32BitGraphicsMemoryImpl(allocationData, false);
}
if (allocationData.hostPtr) {
return allocateGraphicsMemoryWithHostPtr(allocationData);
@@ -480,10 +480,10 @@ void MemoryManager::unlockResource(GraphicsAllocation *graphicsAllocation) {
HeapIndex MemoryManager::selectHeap(const GraphicsAllocation *allocation, bool hasPointer, bool isFullRangeSVM) {
if (allocation) {
if (useInternal32BitAllocator(allocation->getAllocationType())) {
return HeapIndex::HEAP_INTERNAL_DEVICE_MEMORY;
return selectInternalHeap(allocation->isAllocatedInLocalMemoryPool());
}
if (allocation->is32BitAllocation()) {
return HeapIndex::HEAP_EXTERNAL;
return selectExternalHeap(allocation->isAllocatedInLocalMemoryPool());
}
}
if (isFullRangeSVM) {

9
shared/source/memory_manager/memory_manager.h
View File

@@ -98,8 +98,8 @@ class MemoryManager {
virtual uint64_t getLocalMemorySize(uint32_t rootDeviceIndex) = 0;
uint64_t getMaxApplicationAddress() { return is64bit ? MemoryConstants::max64BitAppAddress : MemoryConstants::max32BitAppAddress; };
MOCKABLE_VIRTUAL uint64_t getInternalHeapBaseAddress(uint32_t rootDeviceIndex) { return getGfxPartition(rootDeviceIndex)->getHeapBase(HeapIndex::HEAP_INTERNAL_DEVICE_MEMORY); }
uint64_t getExternalHeapBaseAddress(uint32_t rootDeviceIndex) { return getGfxPartition(rootDeviceIndex)->getHeapBase(HeapIndex::HEAP_EXTERNAL); }
MOCKABLE_VIRTUAL uint64_t getInternalHeapBaseAddress(uint32_t rootDeviceIndex, bool useLocalMemory) { return getGfxPartition(rootDeviceIndex)->getHeapBase(selectInternalHeap(useLocalMemory)); }
uint64_t getExternalHeapBaseAddress(uint32_t rootDeviceIndex, bool useLocalMemory) { return getGfxPartition(rootDeviceIndex)->getHeapBase(selectExternalHeap(useLocalMemory)); }
bool isLimitedRange(uint32_t rootDeviceIndex) { return getGfxPartition(rootDeviceIndex)->isLimitedRange(); }
@@ -162,6 +162,8 @@ class MemoryManager {
virtual void releaseReservedCpuAddressRange(void *reserved, size_t size, uint32_t rootDeviceIndex){};
void *getReservedMemory(size_t size, size_t alignment);
GfxPartition *getGfxPartition(uint32_t rootDeviceIndex) { return gfxPartitions.at(rootDeviceIndex).get(); }
static HeapIndex selectInternalHeap(bool useLocalMemory) { return useLocalMemory ? HeapIndex::HEAP_INTERNAL_DEVICE_MEMORY : HeapIndex::HEAP_INTERNAL; }
static HeapIndex selectExternalHeap(bool useLocalMemory) { return useLocalMemory ? HeapIndex::HEAP_EXTERNAL_DEVICE_MEMORY : HeapIndex::HEAP_EXTERNAL; }
static uint32_t maxOsContextCount;
virtual void commonCleanup(){};
@@ -174,6 +176,7 @@ class MemoryManager {
return allocationType == GraphicsAllocation::AllocationType::KERNEL_ISA ||
allocationType == GraphicsAllocation::AllocationType::INTERNAL_HEAP;
}
static bool isCopyRequired(ImageInfo &imgInfo, const void *hostPtr);
bool useNonSvmHostPtrAlloc(GraphicsAllocation::AllocationType allocationType, uint32_t rootDeviceIndex);
@@ -185,7 +188,7 @@ class MemoryManager {
virtual GraphicsAllocation *allocateGraphicsMemoryWithHostPtr(const AllocationData &allocationData);
virtual GraphicsAllocation *allocateGraphicsMemoryWithAlignment(const AllocationData &allocationData) = 0;
virtual GraphicsAllocation *allocateGraphicsMemory64kb(const AllocationData &allocationData) = 0;
virtual GraphicsAllocation *allocate32BitGraphicsMemoryImpl(const AllocationData &allocationData) = 0;
virtual GraphicsAllocation *allocate32BitGraphicsMemoryImpl(const AllocationData &allocationData, bool useLocalMemory) = 0;
virtual GraphicsAllocation *allocateGraphicsMemoryInDevicePool(const AllocationData &allocationData, AllocationStatus &status) = 0;
GraphicsAllocation *allocateGraphicsMemoryForImageFromHostPtr(const AllocationData &allocationData);
MOCKABLE_VIRTUAL GraphicsAllocation *allocateGraphicsMemoryForImage(const AllocationData &allocationData);

6
shared/source/os_interface/linux/drm_memory_manager.cpp
View File

@@ -357,9 +357,9 @@ GraphicsAllocation *DrmMemoryManager::allocateGraphicsMemoryForImageImpl(const A
return allocation;
}
DrmAllocation *DrmMemoryManager::allocate32BitGraphicsMemoryImpl(const AllocationData &allocationData) {
DrmAllocation *DrmMemoryManager::allocate32BitGraphicsMemoryImpl(const AllocationData &allocationData, bool useLocalMemory) {
auto internal = useInternal32BitAllocator(allocationData.type);
auto allocatorToUse = internal ? HeapIndex::HEAP_INTERNAL_DEVICE_MEMORY : HeapIndex::HEAP_EXTERNAL;
auto allocatorToUse = internal ? selectInternalHeap(useLocalMemory) : selectExternalHeap(useLocalMemory);
if (allocationData.hostPtr) {
uintptr_t inputPtr = reinterpret_cast<uintptr_t>(allocationData.hostPtr);
@@ -490,7 +490,7 @@ GraphicsAllocation *DrmMemoryManager::createGraphicsAllocationFromSharedHandle(o
if (requireSpecificBitness && this->force32bitAllocations) {
drmAllocation->set32BitAllocation(true);
drmAllocation->setGpuBaseAddress(GmmHelper::canonize(getExternalHeapBaseAddress(properties.rootDeviceIndex)));
drmAllocation->setGpuBaseAddress(GmmHelper::canonize(getExternalHeapBaseAddress(properties.rootDeviceIndex, drmAllocation->isAllocatedInLocalMemoryPool())));
}
if (properties.imgInfo) {

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

@@ -81,7 +81,7 @@ class DrmMemoryManager : public MemoryManager {
MOCKABLE_VIRTUAL void *lockResourceInLocalMemoryImpl(BufferObject *bo);
MOCKABLE_VIRTUAL void unlockResourceInLocalMemoryImpl(BufferObject *bo);
void unlockResourceImpl(GraphicsAllocation &graphicsAllocation) override;
DrmAllocation *allocate32BitGraphicsMemoryImpl(const AllocationData &allocationData) override;
DrmAllocation *allocate32BitGraphicsMemoryImpl(const AllocationData &allocationData, bool useLocalMemory) override;
GraphicsAllocation *allocateGraphicsMemoryInDevicePool(const AllocationData &allocationData, AllocationStatus &status) override;
Drm &getDrm(uint32_t rootDeviceIndex) const;

2
shared/source/os_interface/linux/drm_memory_manager_allocate_in_device_pool_dg1.cpp
View File

@@ -170,7 +170,7 @@ GraphicsAllocation *DrmMemoryManager::allocateGraphicsMemoryInDevicePool(const A
allocation->setCpuPtrAndGpuAddress(cpuAddress, gpuAddress);
}
if (useInternal32BitAllocator(allocationData.type)) {
allocation->setGpuBaseAddress(GmmHelper::canonize(getInternalHeapBaseAddress(allocationData.rootDeviceIndex)));
allocation->setGpuBaseAddress(GmmHelper::canonize(getInternalHeapBaseAddress(allocationData.rootDeviceIndex, true)));
}
status = AllocationStatus::Success;

7
shared/source/os_interface/windows/wddm_memory_manager.cpp
View File

@@ -250,7 +250,7 @@ GraphicsAllocation *WddmMemoryManager::allocateGraphicsMemoryWithHostPtr(const A
return MemoryManager::allocateGraphicsMemoryWithHostPtr(allocationData);
}
GraphicsAllocation *WddmMemoryManager::allocate32BitGraphicsMemoryImpl(const AllocationData &allocationData) {
GraphicsAllocation *WddmMemoryManager::allocate32BitGraphicsMemoryImpl(const AllocationData &allocationData, bool useLocalMemory) {
Gmm *gmm = nullptr;
const void *ptrAligned = nullptr;
size_t sizeAligned = allocationData.size;
@@ -285,7 +285,8 @@ GraphicsAllocation *WddmMemoryManager::allocate32BitGraphicsMemoryImpl(const All
return nullptr;
}
auto baseAddress = useInternal32BitAllocator(allocationData.type) ? getInternalHeapBaseAddress(allocationData.rootDeviceIndex) : getExternalHeapBaseAddress(allocationData.rootDeviceIndex);
auto baseAddress = useInternal32BitAllocator(allocationData.type) ? getInternalHeapBaseAddress(allocationData.rootDeviceIndex, useLocalMemory)
: getExternalHeapBaseAddress(allocationData.rootDeviceIndex, useLocalMemory);
wddmAllocation->setGpuBaseAddress(GmmHelper::canonize(baseAddress));
return wddmAllocation.release();
@@ -313,7 +314,7 @@ GraphicsAllocation *WddmMemoryManager::createAllocationFromHandle(osHandle handl
allocation->setReservedAddressRange(ptr, size);
} else if (requireSpecificBitness && this->force32bitAllocations) {
allocation->set32BitAllocation(true);
allocation->setGpuBaseAddress(GmmHelper::canonize(getExternalHeapBaseAddress(allocation->getRootDeviceIndex())));
allocation->setGpuBaseAddress(GmmHelper::canonize(getExternalHeapBaseAddress(allocation->getRootDeviceIndex(), false)));
}
status = mapGpuVirtualAddress(allocation.get(), allocation->getReservedAddressPtr());
DEBUG_BREAK_IF(!status);

2
shared/source/os_interface/windows/wddm_memory_manager.h
View File

@@ -73,7 +73,7 @@ class WddmMemoryManager : public MemoryManager {
void *lockResourceImpl(GraphicsAllocation &graphicsAllocation) override;
void unlockResourceImpl(GraphicsAllocation &graphicsAllocation) override;
void freeAssociatedResourceImpl(GraphicsAllocation &graphicsAllocation) override;
GraphicsAllocation *allocate32BitGraphicsMemoryImpl(const AllocationData &allocationData) override;
GraphicsAllocation *allocate32BitGraphicsMemoryImpl(const AllocationData &allocationData, bool useLocalMemory) override;
GraphicsAllocation *allocateGraphicsMemoryInDevicePool(const AllocationData &allocationData, AllocationStatus &status) override;
MOCKABLE_VIRTUAL size_t getHugeGfxMemoryChunkSize() const { return 4 * MemoryConstants::gigaByte - MemoryConstants::pageSize64k; }

9
shared/test/unit_test/command_container/command_container_tests.cpp
View File

@@ -114,7 +114,8 @@ TEST_F(CommandContainerTest, givenCommandContainerWhenInitializeThenEverythingIs
auto heapAllocation = cmdContainer.getIndirectHeapAllocation(static_cast<HeapType>(i));
EXPECT_EQ(indirectHeap->getGraphicsAllocation(), heapAllocation);
}
EXPECT_EQ(cmdContainer.getInstructionHeapBaseAddress(), pDevice->getMemoryManager()->getInternalHeapBaseAddress(0));
EXPECT_EQ(cmdContainer.getInstructionHeapBaseAddress(),
pDevice->getMemoryManager()->getInternalHeapBaseAddress(0, cmdContainer.getIndirectHeap(HeapType::INDIRECT_OBJECT)->getGraphicsAllocation()->isAllocatedInLocalMemoryPool()));
}
TEST_F(CommandContainerTest, givenCommandContainerWhenInitializeWithoutDeviceThenReturnedFalse) {
@@ -445,12 +446,14 @@ TEST_F(CommandContainerHeaps, givenCommandContainerForDifferentRootDevicesThenIn
CommandContainer cmdContainer0;
cmdContainer0.initialize(device0.get());
uint64_t baseAddressHeapDevice0 = device0.get()->getMemoryManager()->getInternalHeapBaseAddress(device0->getRootDeviceIndex());
bool useLocalMemory0 = cmdContainer0.getIndirectHeap(HeapType::INDIRECT_OBJECT)->getGraphicsAllocation()->isAllocatedInLocalMemoryPool();
uint64_t baseAddressHeapDevice0 = device0.get()->getMemoryManager()->getInternalHeapBaseAddress(device0->getRootDeviceIndex(), useLocalMemory0);
EXPECT_EQ(cmdContainer0.getInstructionHeapBaseAddress(), baseAddressHeapDevice0);
CommandContainer cmdContainer1;
cmdContainer1.initialize(device1.get());
uint64_t baseAddressHeapDevice1 = device1.get()->getMemoryManager()->getInternalHeapBaseAddress(device1->getRootDeviceIndex());
bool useLocalMemory1 = cmdContainer0.getIndirectHeap(HeapType::INDIRECT_OBJECT)->getGraphicsAllocation()->isAllocatedInLocalMemoryPool();
uint64_t baseAddressHeapDevice1 = device1.get()->getMemoryManager()->getInternalHeapBaseAddress(device1->getRootDeviceIndex(), useLocalMemory1);
EXPECT_EQ(cmdContainer1.getInstructionHeapBaseAddress(), baseAddressHeapDevice1);
}