intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2026-01-03 06:49:52 +08:00
Code Issues Packages Projects Releases Wiki Activity

Zebin: set kernel barriers based on ext funcs

Browse Source 
This change allows for modifying kernel's barrier count
based on called external functions metadata passed
via zeInfo section in zebin.

Added parsing external functions metadata.
Added resolving external functions call graph.
Added updating kernel barriers based on called external functions.
Added support for L0 dynamic link.

Signed-off-by: Krystian Chmielewski <krystian.chmielewski@intel.com>
This commit is contained in:
Krystian Chmielewski
2022-02-28 17:44:06 +00:00
committed by Compute-Runtime-Automation
parent bae9e6f5b5
commit 0ccce5a6d7
24 changed files with 1143 additions and 98 deletions
Download Patch File Download Diff File Expand all files Collapse all files

22
opencl/test/unit_test/program/program_data_tests.cpp
View File

@@ -1,5 +1,5 @@
/*
* Copyright (C) 2018-2021 Intel Corporation
* Copyright (C) 2018-2022 Intel Corporation
*
* SPDX-License-Identifier: MIT
*
@@ -488,7 +488,7 @@ TEST_F(ProgramDataTest, GivenProgramWith32bitPointerOptWhenProgramScopeConstantB
programInfo.globalConstants.initData = constantSurface.mockGfxAllocation.getUnderlyingBuffer();
pProgram->setLinkerInput(pClDevice->getRootDeviceIndex(), std::move(programInfo.linkerInput));
pProgram->linkBinary(&pClDevice->getDevice(), programInfo.globalConstants.initData, programInfo.globalVariables.initData, {});
pProgram->linkBinary(&pClDevice->getDevice(), programInfo.globalConstants.initData, programInfo.globalVariables.initData, {}, prog->externalFunctions);
uint32_t expectedAddr = static_cast<uint32_t>(constantSurface.getGraphicsAllocation(pClDevice->getRootDeviceIndex())->getGpuAddressToPatch());
EXPECT_EQ(expectedAddr, constantSurfaceStorage[0]);
EXPECT_EQ(sentinel, constantSurfaceStorage[1]);
@@ -534,7 +534,7 @@ TEST_F(ProgramDataTest, GivenProgramWith32bitPointerOptWhenProgramScopeGlobalPoi
programInfo.globalVariables.initData = globalSurface.mockGfxAllocation.getUnderlyingBuffer();
pProgram->setLinkerInput(pClDevice->getRootDeviceIndex(), std::move(programInfo.linkerInput));
pProgram->linkBinary(&pClDevice->getDevice(), programInfo.globalConstants.initData, programInfo.globalVariables.initData, {});
pProgram->linkBinary(&pClDevice->getDevice(), programInfo.globalConstants.initData, programInfo.globalVariables.initData, {}, prog->externalFunctions);
uint32_t expectedAddr = static_cast<uint32_t>(globalSurface.getGraphicsAllocation(pClDevice->getRootDeviceIndex())->getGpuAddressToPatch());
EXPECT_EQ(expectedAddr, globalSurfaceStorage[0]);
EXPECT_EQ(sentinel, globalSurfaceStorage[1]);
@@ -561,7 +561,7 @@ TEST(ProgramLinkBinaryTest, whenLinkerInputEmptyThenLinkSuccessful) {
auto device = std::make_unique<MockClDevice>(MockDevice::createWithNewExecutionEnvironment<MockDevice>(defaultHwInfo.get()));
MockProgram program{nullptr, false, toClDeviceVector(*device)};
program.setLinkerInput(device->getRootDeviceIndex(), std::move(linkerInput));
auto ret = program.linkBinary(&device->getDevice(), nullptr, nullptr, {});
auto ret = program.linkBinary(&device->getDevice(), nullptr, nullptr, {}, program.externalFunctions);
EXPECT_EQ(CL_SUCCESS, ret);
}
@@ -586,7 +586,7 @@ TEST(ProgramLinkBinaryTest, whenLinkerUnresolvedExternalThenLinkFailedAndBuildLo
std::string buildLog = program.getBuildLog(device->getRootDeviceIndex());
EXPECT_TRUE(buildLog.empty());
auto ret = program.linkBinary(&device->getDevice(), nullptr, nullptr, {});
auto ret = program.linkBinary(&device->getDevice(), nullptr, nullptr, {}, program.externalFunctions);
EXPECT_NE(CL_SUCCESS, ret);
program.getKernelInfoArray(rootDeviceIndex).clear();
buildLog = program.getBuildLog(rootDeviceIndex);
@@ -633,7 +633,7 @@ TEST_F(ProgramDataTest, whenLinkerInputValidThenIsaIsProperlyPatched) {
buildInfo.globalSurface = new MockGraphicsAllocation(globalVariablesBuffer.data(), globalVariablesBuffer.size());
buildInfo.constantSurface = new MockGraphicsAllocation(globalConstantsBuffer.data(), globalConstantsBuffer.size());
auto ret = program.linkBinary(&pClDevice->getDevice(), globalConstantsInitData.data(), globalVariablesInitData.data(), {});
auto ret = program.linkBinary(&pClDevice->getDevice(), globalConstantsInitData.data(), globalVariablesInitData.data(), {}, program.externalFunctions);
EXPECT_EQ(CL_SUCCESS, ret);
linkerInput.reset(static_cast<WhiteBox<LinkerInput> *>(buildInfo.linkerInput.release()));
@@ -681,7 +681,7 @@ TEST_F(ProgramDataTest, whenRelocationsAreNotNeededThenIsaIsPreserved) {
buildInfo.globalSurface = new MockGraphicsAllocation(globalVariablesBuffer.data(), globalVariablesBuffer.size());
buildInfo.constantSurface = new MockGraphicsAllocation(globalConstantsBuffer.data(), globalConstantsBuffer.size());
auto ret = program.linkBinary(&pClDevice->getDevice(), globalConstantsInitData.data(), globalVariablesInitData.data(), {});
auto ret = program.linkBinary(&pClDevice->getDevice(), globalConstantsInitData.data(), globalVariablesInitData.data(), {}, program.externalFunctions);
EXPECT_EQ(CL_SUCCESS, ret);
EXPECT_EQ(kernelHeapData, kernelHeap);
@@ -722,7 +722,7 @@ TEST(ProgramStringSectionTest, WhenConstStringBufferIsPresentThenUseItForLinking
const char constStringData[] = "Hello World!\n";
auto stringsAddr = reinterpret_cast<uintptr_t>(constStringData);
auto ret = program.linkBinary(&device->getDevice(), nullptr, nullptr, {constStringData, sizeof(constStringData)});
auto ret = program.linkBinary(&device->getDevice(), nullptr, nullptr, {constStringData, sizeof(constStringData)}, program.externalFunctions);
EXPECT_EQ(CL_SUCCESS, ret);
EXPECT_EQ(static_cast<size_t>(stringsAddr), *reinterpret_cast<size_t *>(patchAddr));
@@ -747,7 +747,7 @@ TEST(ProgramImplicitArgsTest, givenImplicitRelocationAndStackCallsThenKernelRequ
linkerInput->relocations.push_back({{implicitArgsRelocationSymbolName, 0x8, LinkerInput::RelocationInfo::Type::AddressLow, SegmentType::Instructions}});
linkerInput->traits.requiresPatchingOfInstructionSegments = true;
program.setLinkerInput(rootDeviceIndex, std::move(linkerInput));
auto ret = program.linkBinary(&device->getDevice(), nullptr, nullptr, {});
auto ret = program.linkBinary(&device->getDevice(), nullptr, nullptr, {}, program.externalFunctions);
EXPECT_EQ(CL_SUCCESS, ret);
EXPECT_TRUE(kernelInfo.kernelDescriptor.kernelAttributes.flags.requiresImplicitArgs);
@@ -779,7 +779,7 @@ TEST(ProgramImplicitArgsTest, givenImplicitRelocationAndEnabledDebuggerThenKerne
linkerInput->relocations.push_back({{implicitArgsRelocationSymbolName, 0x8, LinkerInput::RelocationInfo::Type::AddressLow, SegmentType::Instructions}});
linkerInput->traits.requiresPatchingOfInstructionSegments = true;
program.setLinkerInput(rootDeviceIndex, std::move(linkerInput));
auto ret = program.linkBinary(&device->getDevice(), nullptr, nullptr, {});
auto ret = program.linkBinary(&device->getDevice(), nullptr, nullptr, {}, program.externalFunctions);
EXPECT_EQ(CL_SUCCESS, ret);
EXPECT_TRUE(kernelInfo.kernelDescriptor.kernelAttributes.flags.requiresImplicitArgs);
@@ -805,7 +805,7 @@ TEST(ProgramImplicitArgsTest, givenImplicitRelocationAndNoStackCallsAndDisabledD
linkerInput->relocations.push_back({{implicitArgsRelocationSymbolName, 0x8, LinkerInput::RelocationInfo::Type::AddressLow, SegmentType::Instructions}});
linkerInput->traits.requiresPatchingOfInstructionSegments = true;
program.setLinkerInput(rootDeviceIndex, std::move(linkerInput));
auto ret = program.linkBinary(&device->getDevice(), nullptr, nullptr, {});
auto ret = program.linkBinary(&device->getDevice(), nullptr, nullptr, {}, program.externalFunctions);
EXPECT_EQ(CL_SUCCESS, ret);
EXPECT_FALSE(kernelInfo.kernelDescriptor.kernelAttributes.flags.requiresImplicitArgs);

20
opencl/test/unit_test/program/program_tests.cpp
View File

@@ -1201,13 +1201,13 @@ TEST_F(ProgramFromSourceTest, GivenAdvancedOptionsWhenCreatingProgramThenSuccess
const char *sources[1] = {pSourceBuffer.get()};
EXPECT_NE(nullptr, pSourceBuffer);
//According to spec: If lengths is NULL, all strings in the strings argument are considered null-terminated.
// According to spec: If lengths is NULL, all strings in the strings argument are considered null-terminated.
p = Program::create(pContext, 1, sources, nullptr, retVal);
EXPECT_EQ(CL_SUCCESS, retVal);
EXPECT_NE(nullptr, p);
delete p;
//According to spec: If an element in lengths is zero, its accompanying string is null-terminated.
// According to spec: If an element in lengths is zero, its accompanying string is null-terminated.
p = Program::create(pContext, 1, sources, &sourceSize, retVal);
EXPECT_EQ(CL_SUCCESS, retVal);
EXPECT_NE(nullptr, p);
@@ -1406,7 +1406,7 @@ HWTEST_F(PatchTokenTests, givenKernelRequiringConstantAllocationWhenMakeResident
auto &residencyVector = pCommandStreamReceiver->getResidencyAllocations();
//we expect kernel ISA here and constant allocation
// we expect kernel ISA here and constant allocation
auto kernelIsa = pKernel->getKernelInfo().getGraphicsAllocation();
auto constantAllocation = pProgram->getConstantSurface(pDevice->getRootDeviceIndex());
@@ -3216,3 +3216,17 @@ TEST(ProgramTest, givenLockedProgramWhenReleasingForKernelIsCalledForEachRetainT
program.releaseForKernel();
EXPECT_FALSE(program.isLocked());
}
TEST_F(ProgramTests, givenValidZebinWithKernelCallingExternalFunctionThenUpdateKernelsBarrierCount) {
ZebinTestData::ZebinWithExternalFunctionsInfo zebin;
auto program = std::make_unique<MockProgram>(nullptr, false, toClDeviceVector(*pClDevice));
program->buildInfos[rootDeviceIndex].unpackedDeviceBinary = makeCopy(zebin.storage.data(), zebin.storage.size());
program->buildInfos[rootDeviceIndex].unpackedDeviceBinarySize = zebin.storage.size();
auto retVal = program->processGenBinary(*pClDevice);
EXPECT_EQ(CL_SUCCESS, retVal);
ASSERT_EQ(2U, program->buildInfos[rootDeviceIndex].kernelInfoArray.size());
auto &kernelInfo = program->buildInfos[rootDeviceIndex].kernelInfoArray[0];
EXPECT_EQ(zebin.barrierCount, kernelInfo->kernelDescriptor.kernelAttributes.barrierCount);
}