intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2025-12-24 21:18:24 +08:00
Code Issues Packages Projects Releases Wiki Activity

performance(ocl): set pool allocator threshold 1MB

Browse Source 
Increase pool allocator threshold to 1MB
Remove stack allocations based on threshold in tests.

Related-To: NEO-9690

Signed-off-by: Dominik Dabek <dominik.dabek@intel.com>
This commit is contained in:
Dominik Dabek
2023-12-06 17:10:58 +00:00
committed by Compute-Runtime-Automation
parent 9ce3713ace
commit fc1d93af8e
5 changed files with 13 additions and 12 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
opencl/source/context/context.cpp
View File

@@ -541,6 +541,7 @@ Buffer *Context::BufferPool::allocate(const MemoryProperties &memoryProperties,
return nullptr;
}
bufferRegion.origin -= BufferPool::startingOffset;
DEBUG_BREAK_IF(bufferRegion.origin >= BufferPool::aggregatedSmallBuffersPoolSize);
bufferRegion.size = requestedSize;
auto bufferFromPool = this->mainStorage->createSubBuffer(flags, flagsIntel, &bufferRegion, errcodeRet);
bufferFromPool->createFunction = this->mainStorage->createFunction;

5
opencl/test/unit_test/aub_tests/command_queue/aub_multicontext_tests_xehp_and_later.cpp
View File

@@ -79,9 +79,10 @@ struct MultitileMulticontextTests : public MulticontextAubFixture, public ::test
for (uint32_t tile = 0; tile < tileDevices.size(); tile++) {
for (uint32_t tileEngine = 0; tileEngine < commandQueues[tile].size(); tileEngine++) {
getSimulatedCsr<FamilyType>(tile, tileEngine)->pollForCompletion();
auto regularBufferGpuAddress = static_cast<uintptr_t>(ptrOffset(regularBuffers[tile][tileEngine]->getGraphicsAllocation(rootDeviceIndex)->getGpuAddress(), regularBuffers[tile][tileEngine]->getOffset()));
auto tileOnlyBufferGpuAddress = static_cast<uintptr_t>(ptrOffset(tileOnlyBuffers[tile][tileEngine]->getGraphicsAllocation(rootDeviceIndex)->getGpuAddress(), tileOnlyBuffers[tile][tileEngine]->getOffset()));
auto regularBufferGpuAddress = static_cast<uintptr_t>(regularBuffers[tile][tileEngine]->getGraphicsAllocation(rootDeviceIndex)->getGpuAddress());
auto tileOnlyBufferGpuAddress = static_cast<uintptr_t>(tileOnlyBuffers[tile][tileEngine]->getGraphicsAllocation(rootDeviceIndex)->getGpuAddress());
expectMemory<FamilyType>(reinterpret_cast<void *>(regularBufferGpuAddress), writePattern, bufferSize, tile, tileEngine);
expectMemory<FamilyType>(reinterpret_cast<void *>(tileOnlyBufferGpuAddress), writePattern, bufferSize, tile, tileEngine);
}

3
opencl/test/unit_test/aub_tests/command_stream/copy_engine_aub_tests_xehp_and_later.h
View File

@@ -52,6 +52,7 @@ struct CopyEngineXeHPAndLater : public MulticontextAubFixture, public ::testing:
debugManager.flags.RenderCompressedBuffersEnabled.set(true);
debugManager.flags.RenderCompressedImagesEnabled.set(true);
debugManager.flags.EnableFreeMemory.set(false);
debugManager.flags.ExperimentalSmallBufferPoolAllocator.set(0);
MulticontextAubFixture::setUp(numTiles, EnabledCommandStreamers::Single, true);
@@ -100,7 +101,7 @@ struct CopyEngineXeHPAndLater : public MulticontextAubFixture, public ::testing:
EXPECT_EQ(CL_SUCCESS, retVal);
if (compressed) {
EXPECT_TRUE(graphicsAllocation->getDefaultGmm()->isCompressionEnabled);
EXPECT_TRUE(graphicsAllocation->isCompressionEnabled());
}
EXPECT_EQ(!inLocalMemory, MemoryPoolHelper::isSystemMemoryPool(graphicsAllocation->getMemoryPool()));

14
opencl/test/unit_test/mem_obj/buffer_pool_alloc_tests.cpp
View File

@@ -362,8 +362,8 @@ TEST_F(AggregatedSmallBuffersEnabledTest, givenCopyHostPointerWhenCreatingBuffer
context->setSpecialQueue(commandQueue, rootDeviceIndex);
flags = CL_MEM_COPY_HOST_PTR;
unsigned char dataToCopy[PoolAllocator::smallBufferThreshold];
hostPtr = dataToCopy;
auto dataToCopy = std::unique_ptr<unsigned char[]>(new unsigned char[PoolAllocator::smallBufferThreshold]());
hostPtr = dataToCopy.get();
EXPECT_TRUE(poolAllocator->isAggregatedSmallBuffersEnabled(context.get()));
EXPECT_EQ(1u, poolAllocator->bufferPools.size());
@@ -430,7 +430,6 @@ TEST_F(AggregatedSmallBuffersKernelTest, givenBufferFromPoolWhenOffsetSubbufferI
std::unique_ptr<Buffer> buffer(Buffer::create(context.get(), flags, size, hostPtr, retVal));
EXPECT_EQ(retVal, CL_SUCCESS);
EXPECT_NE(buffer, nullptr);
EXPECT_GT(buffer->getOffset(), 0u);
cl_buffer_region region;
region.origin = 0xc0;
region.size = 32;
@@ -547,8 +546,8 @@ TEST_F(AggregatedSmallBuffersEnabledApiTest, givenSmallBufferWhenCreatingBufferT
TEST_F(AggregatedSmallBuffersEnabledApiTest, givenUseHostPointerWhenCreatingBufferThenDoNotUsePool) {
flags |= CL_MEM_USE_HOST_PTR;
unsigned char hostData[PoolAllocator::smallBufferThreshold];
hostPtr = hostData;
auto hostData = std::unique_ptr<unsigned char[]>(new unsigned char[PoolAllocator::smallBufferThreshold]());
hostPtr = hostData.get();
cl_mem smallBuffer = clCreateBuffer(clContext, flags, size, hostPtr, &retVal);
EXPECT_EQ(retVal, CL_SUCCESS);
EXPECT_NE(smallBuffer, nullptr);
@@ -634,9 +633,9 @@ TEST_F(AggregatedSmallBuffersEnabledApiTest, givenSubBufferNotFromPoolAndAggrega
TEST_F(AggregatedSmallBuffersEnabledApiTest, givenCopyHostPointerWhenCreatingBufferThenUsePoolAndCopyHostPointer) {
flags |= CL_MEM_COPY_HOST_PTR;
unsigned char dataToCopy[PoolAllocator::smallBufferThreshold];
auto dataToCopy = std::unique_ptr<unsigned char[]>(new unsigned char[PoolAllocator::smallBufferThreshold]());
dataToCopy[0] = 123;
hostPtr = dataToCopy;
hostPtr = dataToCopy.get();
auto contextRefCountBefore = context->getRefInternalCount();
cl_mem smallBuffer = clCreateBuffer(clContext, flags, size, hostPtr, &retVal);
EXPECT_EQ(context->getRefInternalCount(), contextRefCountBefore + 1);
@@ -672,7 +671,6 @@ TEST_F(AggregatedSmallBuffersSubBufferApiTest, givenBufferFromPoolWhenCreateSubB
EXPECT_EQ(retVal, CL_SUCCESS);
EXPECT_NE(buffer, nullptr);
MockBuffer *mockBuffer = static_cast<MockBuffer *>(buffer);
EXPECT_GT(mockBuffer->offset, 0u);
EXPECT_EQ(ptrOffset(poolAllocator->bufferPools[0].mainStorage->getCpuAddress(), mockBuffer->getOffset()), mockBuffer->getCpuAddress());
cl_buffer_region region{};

2
shared/source/utilities/buffer_pool_allocator.h
View File

@@ -26,7 +26,7 @@ template <typename PoolT>
struct SmallBuffersParams {
protected:
static constexpr auto aggregatedSmallBuffersPoolSize = 2 * MemoryConstants::megaByte;
static constexpr auto smallBufferThreshold = 4 * MemoryConstants::kiloByte;
static constexpr auto smallBufferThreshold = 1 * MemoryConstants::megaByte;
static constexpr auto chunkAlignment = 512u;
static constexpr auto startingOffset = chunkAlignment;
};