intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2025-12-22 10:17:01 +08:00
Code Issues Packages Projects Releases Wiki Activity

feature: add pooling of USM global/constant surface

Browse Source 
Related-To: NEO-12287
Signed-off-by: Fabian Zwoliński <fabian.zwolinski@intel.com>
This commit is contained in:
Fabian Zwoliński
2025-09-12 13:52:33 +00:00
committed by Compute-Runtime-Automation
parent ca2b26e87b
commit 68698c9a74
15 changed files with 601 additions and 38 deletions
Download Patch File Download Diff File Expand all files Collapse all files

47
opencl/test/unit_test/program/program_tests.cpp
View File

@@ -27,7 +27,9 @@
#include "shared/source/memory_manager/graphics_allocation.h"
#include "shared/source/memory_manager/surface.h"
#include "shared/source/os_interface/os_context.h"
#include "shared/source/program/program_initialization.h"
#include "shared/source/utilities/arrayref.h"
#include "shared/test/common/compiler_interface/linker_mock.h"
#include "shared/test/common/device_binary_format/patchtokens_tests.h"
#include "shared/test/common/helpers/debug_manager_state_restore.h"
#include "shared/test/common/helpers/gtest_helpers.h"
@@ -42,6 +44,8 @@
#include "shared/test/common/mocks/mock_elf.h"
#include "shared/test/common/mocks/mock_graphics_allocation.h"
#include "shared/test/common/mocks/mock_modules_zebin.h"
#include "shared/test/common/mocks/mock_product_helper.h"
#include "shared/test/common/mocks/mock_usm_memory_pool.h"
#include "shared/test/common/test_macros/hw_test.h"
#include "shared/test/common/utilities/base_object_utils.h"
@@ -536,6 +540,49 @@ TEST_F(ProgramFromBinaryTest, whenProgramIsBeingRebuildThenOutdatedGlobalBuffers
EXPECT_EQ(nullptr, pProgram->buildInfos[pClDevice->getRootDeviceIndex()].globalSurface.get());
}
TEST_F(ProgramFromBinaryTest, GivenUsmPoolAnd2MBAlignmentEnabledWhenProgramIsBeingRebuildThenOutdatedGlobalBuffersAreFreedFromUsmPool) {
ASSERT_NE(nullptr, this->pContext->getSVMAllocsManager());
pProgram->build(pProgram->getDevices(), nullptr);
EXPECT_EQ(nullptr, pProgram->buildInfos[pClDevice->getRootDeviceIndex()].constantSurface);
EXPECT_EQ(nullptr, pProgram->buildInfos[pClDevice->getRootDeviceIndex()].globalSurface);
auto usmConstantSurfaceAllocPool = new MockUsmMemAllocPool;
auto usmGlobalSurfaceAllocPool = new MockUsmMemAllocPool;
pClDevice->getDevice().resetUsmConstantSurfaceAllocPool(usmConstantSurfaceAllocPool);
pClDevice->getDevice().resetUsmGlobalSurfaceAllocPool(usmGlobalSurfaceAllocPool);
auto mockProductHelper = new MockProductHelper;
pClDevice->getDevice().getRootDeviceEnvironmentRef().productHelper.reset(mockProductHelper);
mockProductHelper->is2MBLocalMemAlignmentEnabledResult = true;
std::vector<unsigned char> initData(1024, 0x5B);
WhiteBox<NEO::LinkerInput> linkerInput;
linkerInput.traits.exportsGlobalConstants = true;
linkerInput.traits.exportsGlobalVariables = true;
pProgram->buildInfos[pClDevice->getRootDeviceIndex()].constantSurface.reset(allocateGlobalsSurface(pContext->getSVMAllocsManager(), pClDevice->getDevice(), initData.size(), 0u, true, &linkerInput, initData.data()));
auto &constantSurface = pProgram->buildInfos[pClDevice->getRootDeviceIndex()].constantSurface;
EXPECT_TRUE(pClDevice->getDevice().getUsmConstantSurfaceAllocPool()->isInPool(reinterpret_cast<void *>(constantSurface->getGpuAddress())));
pProgram->processGenBinary(*pClDevice);
EXPECT_EQ(nullptr, pProgram->buildInfos[pClDevice->getRootDeviceIndex()].constantSurface.get());
EXPECT_EQ(nullptr, pProgram->buildInfos[pClDevice->getRootDeviceIndex()].globalSurface.get());
EXPECT_EQ(1u, usmConstantSurfaceAllocPool->freeSVMAllocCalled);
EXPECT_EQ(0u, usmGlobalSurfaceAllocPool->freeSVMAllocCalled);
pProgram->buildInfos[pClDevice->getRootDeviceIndex()].globalSurface.reset(allocateGlobalsSurface(pContext->getSVMAllocsManager(), pClDevice->getDevice(), initData.size(), 0u, false, &linkerInput, initData.data()));
auto &globalSurface = pProgram->buildInfos[pClDevice->getRootDeviceIndex()].globalSurface;
EXPECT_TRUE(pClDevice->getDevice().getUsmGlobalSurfaceAllocPool()->isInPool(reinterpret_cast<void *>(globalSurface->getGpuAddress())));
pProgram->processGenBinary(*pClDevice);
EXPECT_EQ(nullptr, pProgram->buildInfos[pClDevice->getRootDeviceIndex()].constantSurface.get());
EXPECT_EQ(nullptr, pProgram->buildInfos[pClDevice->getRootDeviceIndex()].globalSurface.get());
EXPECT_EQ(1u, usmConstantSurfaceAllocPool->freeSVMAllocCalled);
EXPECT_EQ(1u, usmGlobalSurfaceAllocPool->freeSVMAllocCalled);
}
TEST_F(ProgramFromBinaryTest, givenProgramWhenCleanKernelInfoIsCalledThenKernelAllocationIsFreed) {
pProgram->build(pProgram->getDevices(), nullptr);
EXPECT_EQ(1u, pProgram->getNumKernels());