intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2025-12-24 20:39:56 +08:00
Code Issues Packages Projects Releases Wiki Activity

Add support for a new relocation name for implicit args

Browse Source 
Related-To: NEO-5081
Signed-off-by: Mateusz Jablonski <mateusz.jablonski@intel.com>
This commit is contained in:
Mateusz Jablonski
2022-03-15 08:58:41 +00:00
committed by Compute-Runtime-Automation
parent c14dc6d9ed
commit c7e76ce109
5 changed files with 212 additions and 198 deletions
Download Patch File Download Diff File Expand all files Collapse all files

14
shared/source/compiler_interface/linker.cpp
View File

@@ -342,11 +342,17 @@ void Linker::patchInstructionsSegments(const std::vector<PatchableSegment> &inst
}
UNRECOVERABLE_IF(nullptr == instSeg.hostPointer);
auto relocAddress = ptrOffset(instSeg.hostPointer, static_cast<uintptr_t>(relocation.offset));
if (relocation.symbolName == implicitArgsRelocationSymbolName) {
if (pImplicitArgsRelocationAddresses.find(segId) == pImplicitArgsRelocationAddresses.end()) {
pImplicitArgsRelocationAddresses.insert({segId, {}});
bool isImplicitArgsRelocation = false;
for (const auto &implicitArgsRelocationSymbolName : implicitArgsRelocationSymbolNames) {
if (relocation.symbolName == implicitArgsRelocationSymbolName) {
if (pImplicitArgsRelocationAddresses.find(segId) == pImplicitArgsRelocationAddresses.end()) {
pImplicitArgsRelocationAddresses.insert({segId, {}});
}
pImplicitArgsRelocationAddresses[segId].push_back(reinterpret_cast<uint32_t *>(relocAddress));
isImplicitArgsRelocation = true;
}
pImplicitArgsRelocationAddresses[segId].push_back(reinterpret_cast<uint32_t *>(relocAddress));
}
if (isImplicitArgsRelocation) {
continue;
}
auto symbolIt = relocatedSymbols.find(relocation.symbolName);

2
shared/source/kernel/implicit_args.h
View File

@@ -41,7 +41,7 @@ struct ImplicitArgs {
static_assert((sizeof(ImplicitArgs) & 31) == 0, "Implicit args size need to be aligned to 32");
static_assert(std::is_pod<ImplicitArgs>::value);
constexpr const char *implicitArgsRelocationSymbolName = "INTEL_PATCH_CROSS_THREAD_OFFSET_OFF_R0";
constexpr const char *implicitArgsRelocationSymbolNames[] = {"INTEL_PATCH_CROSS_THREAD_OFFSET_OFF_R0", "__INTEL_PATCH_CROSS_THREAD_OFFSET_OFF_R0"};
namespace ImplicitArgsHelper {
std::array<uint8_t, 3> getDimensionOrderForLocalIds(const uint8_t *workgroupDimensionsOrder, std::optional<std::pair<bool /* localIdsGeneratedByRuntime */, uint32_t /* walkOrderForHwGenerationOfLocalIds */>> hwGenerationOfLocalIdsParams);

382
shared/test/unit_test/compiler_interface/linker_tests.cpp
View File

@@ -2009,177 +2009,183 @@ TEST(LinkerTests, GivenDebugDataWhenApplyingDebugDataRelocationsThenRelocationsA
}
TEST(LinkerTests, givenImplicitArgRelocationAndStackCallsThenPatchRelocationWithSizeOfImplicitArgStructAndUpdateKernelDescriptor) {
NEO::LinkerInput linkerInput;
for (const auto &implicitArgsRelocationSymbolName : implicitArgsRelocationSymbolNames) {
NEO::LinkerInput linkerInput;
vISA::GenRelocEntry reloc = {};
std::string relocationName = implicitArgsRelocationSymbolName;
memcpy_s(reloc.r_symbol, 1024, relocationName.c_str(), relocationName.size());
reloc.r_offset = 8;
reloc.r_type = vISA::GenRelocType::R_SYM_ADDR_32;
vISA::GenRelocEntry reloc = {};
std::string relocationName = implicitArgsRelocationSymbolName;
memcpy_s(reloc.r_symbol, 1024, relocationName.c_str(), relocationName.size());
reloc.r_offset = 8;
reloc.r_type = vISA::GenRelocType::R_SYM_ADDR_32;
vISA::GenRelocEntry relocs[] = {reloc};
constexpr uint32_t numRelocations = 1;
bool decodeRelocSuccess = linkerInput.decodeRelocationTable(&relocs, numRelocations, 0);
EXPECT_TRUE(decodeRelocSuccess);
vISA::GenRelocEntry relocs[] = {reloc};
constexpr uint32_t numRelocations = 1;
bool decodeRelocSuccess = linkerInput.decodeRelocationTable(&relocs, numRelocations, 0);
EXPECT_TRUE(decodeRelocSuccess);
NEO::Linker linker(linkerInput);
NEO::Linker::SegmentInfo globalVarSegment, globalConstSegment, exportedFuncSegment;
globalVarSegment.gpuAddress = 8;
globalVarSegment.segmentSize = 64;
globalConstSegment.gpuAddress = 128;
globalConstSegment.segmentSize = 256;
exportedFuncSegment.gpuAddress = 4096;
exportedFuncSegment.segmentSize = 1024;
NEO::Linker::UnresolvedExternals unresolvedExternals;
NEO::Linker::KernelDescriptorsT kernelDescriptors;
NEO::Linker::ExternalFunctionsT externalFunctions;
KernelDescriptor kernelDescriptor;
kernelDescriptors.push_back(&kernelDescriptor);
kernelDescriptor.kernelAttributes.flags.requiresImplicitArgs = false;
kernelDescriptor.kernelAttributes.flags.useStackCalls = true;
NEO::Linker linker(linkerInput);
NEO::Linker::SegmentInfo globalVarSegment, globalConstSegment, exportedFuncSegment;
globalVarSegment.gpuAddress = 8;
globalVarSegment.segmentSize = 64;
globalConstSegment.gpuAddress = 128;
globalConstSegment.segmentSize = 256;
exportedFuncSegment.gpuAddress = 4096;
exportedFuncSegment.segmentSize = 1024;
NEO::Linker::UnresolvedExternals unresolvedExternals;
NEO::Linker::KernelDescriptorsT kernelDescriptors;
NEO::Linker::ExternalFunctionsT externalFunctions;
KernelDescriptor kernelDescriptor;
kernelDescriptors.push_back(&kernelDescriptor);
kernelDescriptor.kernelAttributes.flags.requiresImplicitArgs = false;
kernelDescriptor.kernelAttributes.flags.useStackCalls = true;
UltDeviceFactory deviceFactory{1, 0};
UltDeviceFactory deviceFactory{1, 0};
std::vector<char> instructionSegment;
uint32_t initData = 0x77777777;
instructionSegment.resize(32, static_cast<char>(initData));
NEO::Linker::PatchableSegment seg0;
seg0.hostPointer = instructionSegment.data();
seg0.segmentSize = instructionSegment.size();
NEO::Linker::PatchableSegments patchableInstructionSegments{seg0};
std::vector<char> instructionSegment;
uint32_t initData = 0x77777777;
instructionSegment.resize(32, static_cast<char>(initData));
NEO::Linker::PatchableSegment seg0;
seg0.hostPointer = instructionSegment.data();
seg0.segmentSize = instructionSegment.size();
NEO::Linker::PatchableSegments patchableInstructionSegments{seg0};
auto linkResult = linker.link(globalVarSegment, globalConstSegment, exportedFuncSegment, {},
nullptr, nullptr, patchableInstructionSegments, unresolvedExternals,
deviceFactory.rootDevices[0], nullptr, nullptr, kernelDescriptors, externalFunctions);
EXPECT_EQ(NEO::LinkingStatus::LinkedFully, linkResult);
auto relocatedSymbols = linker.extractRelocatedSymbols();
EXPECT_EQ(0U, unresolvedExternals.size());
EXPECT_EQ(0U, relocatedSymbols.size());
auto linkResult = linker.link(globalVarSegment, globalConstSegment, exportedFuncSegment, {},
nullptr, nullptr, patchableInstructionSegments, unresolvedExternals,
deviceFactory.rootDevices[0], nullptr, nullptr, kernelDescriptors, externalFunctions);
EXPECT_EQ(NEO::LinkingStatus::LinkedFully, linkResult);
auto relocatedSymbols = linker.extractRelocatedSymbols();
EXPECT_EQ(0U, unresolvedExternals.size());
EXPECT_EQ(0U, relocatedSymbols.size());
auto addressToPatch = reinterpret_cast<const uint32_t *>(instructionSegment.data() + reloc.r_offset);
EXPECT_EQ(sizeof(ImplicitArgs), *addressToPatch);
EXPECT_EQ(initData, *(addressToPatch - 1));
EXPECT_EQ(initData, *(addressToPatch + 1));
EXPECT_TRUE(kernelDescriptor.kernelAttributes.flags.requiresImplicitArgs);
auto addressToPatch = reinterpret_cast<const uint32_t *>(instructionSegment.data() + reloc.r_offset);
EXPECT_EQ(sizeof(ImplicitArgs), *addressToPatch);
EXPECT_EQ(initData, *(addressToPatch - 1));
EXPECT_EQ(initData, *(addressToPatch + 1));
EXPECT_TRUE(kernelDescriptor.kernelAttributes.flags.requiresImplicitArgs);
}
}
TEST(LinkerTests, givenImplicitArgRelocationAndEnabledDebuggerThenPatchRelocationWithSizeOfImplicitArgStructAndUpdateKernelDescriptor) {
if (!defaultHwInfo->capabilityTable.debuggerSupported) {
GTEST_SKIP();
}
NEO::LinkerInput linkerInput;
for (const auto &implicitArgsRelocationSymbolName : implicitArgsRelocationSymbolNames) {
NEO::LinkerInput linkerInput;
vISA::GenRelocEntry reloc = {};
std::string relocationName = implicitArgsRelocationSymbolName;
memcpy_s(reloc.r_symbol, 1024, relocationName.c_str(), relocationName.size());
reloc.r_offset = 8;
reloc.r_type = vISA::GenRelocType::R_SYM_ADDR_32;
vISA::GenRelocEntry reloc = {};
std::string relocationName = implicitArgsRelocationSymbolName;
memcpy_s(reloc.r_symbol, 1024, relocationName.c_str(), relocationName.size());
reloc.r_offset = 8;
reloc.r_type = vISA::GenRelocType::R_SYM_ADDR_32;
vISA::GenRelocEntry relocs[] = {reloc};
constexpr uint32_t numRelocations = 1;
bool decodeRelocSuccess = linkerInput.decodeRelocationTable(&relocs, numRelocations, 0);
EXPECT_TRUE(decodeRelocSuccess);
vISA::GenRelocEntry relocs[] = {reloc};
constexpr uint32_t numRelocations = 1;
bool decodeRelocSuccess = linkerInput.decodeRelocationTable(&relocs, numRelocations, 0);
EXPECT_TRUE(decodeRelocSuccess);
NEO::Linker linker(linkerInput);
NEO::Linker::SegmentInfo globalVarSegment, globalConstSegment, exportedFuncSegment;
globalVarSegment.gpuAddress = 8;
globalVarSegment.segmentSize = 64;
globalConstSegment.gpuAddress = 128;
globalConstSegment.segmentSize = 256;
exportedFuncSegment.gpuAddress = 4096;
exportedFuncSegment.segmentSize = 1024;
NEO::Linker::UnresolvedExternals unresolvedExternals;
NEO::Linker::KernelDescriptorsT kernelDescriptors;
NEO::Linker::ExternalFunctionsT externalFunctions;
KernelDescriptor kernelDescriptor;
kernelDescriptors.push_back(&kernelDescriptor);
kernelDescriptor.kernelAttributes.flags.requiresImplicitArgs = false;
kernelDescriptor.kernelAttributes.flags.useStackCalls = false;
NEO::Linker linker(linkerInput);
NEO::Linker::SegmentInfo globalVarSegment, globalConstSegment, exportedFuncSegment;
globalVarSegment.gpuAddress = 8;
globalVarSegment.segmentSize = 64;
globalConstSegment.gpuAddress = 128;
globalConstSegment.segmentSize = 256;
exportedFuncSegment.gpuAddress = 4096;
exportedFuncSegment.segmentSize = 1024;
NEO::Linker::UnresolvedExternals unresolvedExternals;
NEO::Linker::KernelDescriptorsT kernelDescriptors;
NEO::Linker::ExternalFunctionsT externalFunctions;
KernelDescriptor kernelDescriptor;
kernelDescriptors.push_back(&kernelDescriptor);
kernelDescriptor.kernelAttributes.flags.requiresImplicitArgs = false;
kernelDescriptor.kernelAttributes.flags.useStackCalls = false;
DebugManagerStateRestore restorer;
DebugManager.flags.EnableMockSourceLevelDebugger.set(1);
UltDeviceFactory deviceFactory{1, 0};
auto device = deviceFactory.rootDevices[0];
EXPECT_NE(nullptr, device->getDebugger());
DebugManagerStateRestore restorer;
DebugManager.flags.EnableMockSourceLevelDebugger.set(1);
UltDeviceFactory deviceFactory{1, 0};
auto device = deviceFactory.rootDevices[0];
EXPECT_NE(nullptr, device->getDebugger());
std::vector<char> instructionSegment;
uint32_t initData = 0x77777777;
instructionSegment.resize(32, static_cast<char>(initData));
NEO::Linker::PatchableSegment seg0;
seg0.hostPointer = instructionSegment.data();
seg0.segmentSize = instructionSegment.size();
NEO::Linker::PatchableSegments patchableInstructionSegments{seg0};
std::vector<char> instructionSegment;
uint32_t initData = 0x77777777;
instructionSegment.resize(32, static_cast<char>(initData));
NEO::Linker::PatchableSegment seg0;
seg0.hostPointer = instructionSegment.data();
seg0.segmentSize = instructionSegment.size();
NEO::Linker::PatchableSegments patchableInstructionSegments{seg0};
auto linkResult = linker.link(globalVarSegment, globalConstSegment, exportedFuncSegment, {},
nullptr, nullptr, patchableInstructionSegments, unresolvedExternals,
device, nullptr, nullptr, kernelDescriptors, externalFunctions);
EXPECT_EQ(NEO::LinkingStatus::LinkedFully, linkResult);
auto relocatedSymbols = linker.extractRelocatedSymbols();
EXPECT_EQ(0U, unresolvedExternals.size());
EXPECT_EQ(0U, relocatedSymbols.size());
auto linkResult = linker.link(globalVarSegment, globalConstSegment, exportedFuncSegment, {},
nullptr, nullptr, patchableInstructionSegments, unresolvedExternals,
device, nullptr, nullptr, kernelDescriptors, externalFunctions);
EXPECT_EQ(NEO::LinkingStatus::LinkedFully, linkResult);
auto relocatedSymbols = linker.extractRelocatedSymbols();
EXPECT_EQ(0U, unresolvedExternals.size());
EXPECT_EQ(0U, relocatedSymbols.size());
auto addressToPatch = reinterpret_cast<const uint32_t *>(instructionSegment.data() + reloc.r_offset);
EXPECT_EQ(sizeof(ImplicitArgs), *addressToPatch);
EXPECT_EQ(initData, *(addressToPatch - 1));
EXPECT_EQ(initData, *(addressToPatch + 1));
EXPECT_TRUE(kernelDescriptor.kernelAttributes.flags.requiresImplicitArgs);
auto addressToPatch = reinterpret_cast<const uint32_t *>(instructionSegment.data() + reloc.r_offset);
EXPECT_EQ(sizeof(ImplicitArgs), *addressToPatch);
EXPECT_EQ(initData, *(addressToPatch - 1));
EXPECT_EQ(initData, *(addressToPatch + 1));
EXPECT_TRUE(kernelDescriptor.kernelAttributes.flags.requiresImplicitArgs);
}
}
TEST(LinkerTests, givenImplicitArgRelocationWithoutStackCallsAndDisabledDebuggerThenPatchRelocationWithZeroAndUpdateKernelDescriptor) {
NEO::LinkerInput linkerInput;
for (const auto &implicitArgsRelocationSymbolName : implicitArgsRelocationSymbolNames) {
NEO::LinkerInput linkerInput;
vISA::GenRelocEntry reloc = {};
std::string relocationName = implicitArgsRelocationSymbolName;
memcpy_s(reloc.r_symbol, 1024, relocationName.c_str(), relocationName.size());
reloc.r_offset = 8;
reloc.r_type = vISA::GenRelocType::R_SYM_ADDR_32;
vISA::GenRelocEntry reloc = {};
std::string relocationName = implicitArgsRelocationSymbolName;
memcpy_s(reloc.r_symbol, 1024, relocationName.c_str(), relocationName.size());
reloc.r_offset = 8;
reloc.r_type = vISA::GenRelocType::R_SYM_ADDR_32;
vISA::GenRelocEntry relocs[] = {reloc};
constexpr uint32_t numRelocations = 1;
bool decodeRelocSuccess = linkerInput.decodeRelocationTable(&relocs, numRelocations, 0);
EXPECT_TRUE(decodeRelocSuccess);
vISA::GenRelocEntry relocs[] = {reloc};
constexpr uint32_t numRelocations = 1;
bool decodeRelocSuccess = linkerInput.decodeRelocationTable(&relocs, numRelocations, 0);
EXPECT_TRUE(decodeRelocSuccess);
NEO::Linker linker(linkerInput);
NEO::Linker::SegmentInfo globalVarSegment, globalConstSegment, exportedFuncSegment;
globalVarSegment.gpuAddress = 8;
globalVarSegment.segmentSize = 64;
globalConstSegment.gpuAddress = 128;
globalConstSegment.segmentSize = 256;
exportedFuncSegment.gpuAddress = 4096;
exportedFuncSegment.segmentSize = 1024;
NEO::Linker::UnresolvedExternals unresolvedExternals;
NEO::Linker::KernelDescriptorsT kernelDescriptors;
NEO::Linker::ExternalFunctionsT externalFunctions;
KernelDescriptor kernelDescriptor;
kernelDescriptors.push_back(&kernelDescriptor);
kernelDescriptor.kernelAttributes.flags.requiresImplicitArgs = false;
kernelDescriptor.kernelAttributes.flags.useStackCalls = false;
NEO::Linker linker(linkerInput);
NEO::Linker::SegmentInfo globalVarSegment, globalConstSegment, exportedFuncSegment;
globalVarSegment.gpuAddress = 8;
globalVarSegment.segmentSize = 64;
globalConstSegment.gpuAddress = 128;
globalConstSegment.segmentSize = 256;
exportedFuncSegment.gpuAddress = 4096;
exportedFuncSegment.segmentSize = 1024;
NEO::Linker::UnresolvedExternals unresolvedExternals;
NEO::Linker::KernelDescriptorsT kernelDescriptors;
NEO::Linker::ExternalFunctionsT externalFunctions;
KernelDescriptor kernelDescriptor;
kernelDescriptors.push_back(&kernelDescriptor);
kernelDescriptor.kernelAttributes.flags.requiresImplicitArgs = false;
kernelDescriptor.kernelAttributes.flags.useStackCalls = false;
UltDeviceFactory deviceFactory{1, 0};
auto device = deviceFactory.rootDevices[0];
EXPECT_EQ(nullptr, device->getDebugger());
UltDeviceFactory deviceFactory{1, 0};
auto device = deviceFactory.rootDevices[0];
EXPECT_EQ(nullptr, device->getDebugger());
std::vector<char> instructionSegment;
uint32_t initData = 0x77777777;
instructionSegment.resize(32, static_cast<char>(initData));
NEO::Linker::PatchableSegment seg0;
seg0.hostPointer = instructionSegment.data();
seg0.segmentSize = instructionSegment.size();
NEO::Linker::PatchableSegments patchableInstructionSegments{seg0};
std::vector<char> instructionSegment;
uint32_t initData = 0x77777777;
instructionSegment.resize(32, static_cast<char>(initData));
NEO::Linker::PatchableSegment seg0;
seg0.hostPointer = instructionSegment.data();
seg0.segmentSize = instructionSegment.size();
NEO::Linker::PatchableSegments patchableInstructionSegments{seg0};
auto linkResult = linker.link(globalVarSegment, globalConstSegment, exportedFuncSegment, {},
nullptr, nullptr, patchableInstructionSegments, unresolvedExternals,
device, nullptr, nullptr, kernelDescriptors, externalFunctions);
EXPECT_EQ(NEO::LinkingStatus::LinkedFully, linkResult);
auto relocatedSymbols = linker.extractRelocatedSymbols();
EXPECT_EQ(0U, unresolvedExternals.size());
EXPECT_EQ(0U, relocatedSymbols.size());
auto linkResult = linker.link(globalVarSegment, globalConstSegment, exportedFuncSegment, {},
nullptr, nullptr, patchableInstructionSegments, unresolvedExternals,
device, nullptr, nullptr, kernelDescriptors, externalFunctions);
EXPECT_EQ(NEO::LinkingStatus::LinkedFully, linkResult);
auto relocatedSymbols = linker.extractRelocatedSymbols();
EXPECT_EQ(0U, unresolvedExternals.size());
EXPECT_EQ(0U, relocatedSymbols.size());
auto addressToPatch = reinterpret_cast<const uint32_t *>(instructionSegment.data() + reloc.r_offset);
EXPECT_EQ(0u, *addressToPatch);
EXPECT_EQ(initData, *(addressToPatch - 1));
EXPECT_EQ(initData, *(addressToPatch + 1));
EXPECT_FALSE(kernelDescriptor.kernelAttributes.flags.requiresImplicitArgs);
auto addressToPatch = reinterpret_cast<const uint32_t *>(instructionSegment.data() + reloc.r_offset);
EXPECT_EQ(0u, *addressToPatch);
EXPECT_EQ(initData, *(addressToPatch - 1));
EXPECT_EQ(initData, *(addressToPatch + 1));
EXPECT_FALSE(kernelDescriptor.kernelAttributes.flags.requiresImplicitArgs);
}
}
TEST(LinkerTests, givenNoImplicitArgRelocationAndStackCallsThenImplicitArgsAreNotRequired) {
@@ -2276,60 +2282,62 @@ TEST(LinkerTests, givenNoImplicitArgRelocationAndEnabledDebuggerThenImplicitArgs
}
TEST(LinkerTests, givenMultipleImplicitArgsRelocationsWithinSingleKernelWhenLinkingThenPatchAllOfThem) {
NEO::LinkerInput linkerInput;
for (const auto &implicitArgsRelocationSymbolName : implicitArgsRelocationSymbolNames) {
NEO::LinkerInput linkerInput;
vISA::GenRelocEntry reloc0 = {};
std::string relocationName = implicitArgsRelocationSymbolName;
memcpy_s(reloc0.r_symbol, 1024, relocationName.c_str(), relocationName.size());
reloc0.r_offset = 8;
reloc0.r_type = vISA::GenRelocType::R_SYM_ADDR_32;
vISA::GenRelocEntry reloc0 = {};
std::string relocationName = implicitArgsRelocationSymbolName;
memcpy_s(reloc0.r_symbol, 1024, relocationName.c_str(), relocationName.size());
reloc0.r_offset = 8;
reloc0.r_type = vISA::GenRelocType::R_SYM_ADDR_32;
vISA::GenRelocEntry reloc1 = reloc0;
reloc1.r_offset = 24;
vISA::GenRelocEntry reloc1 = reloc0;
reloc1.r_offset = 24;
vISA::GenRelocEntry relocs[] = {reloc0, reloc1};
constexpr uint32_t numRelocations = 2;
bool decodeRelocSuccess = linkerInput.decodeRelocationTable(&relocs, numRelocations, 0);
EXPECT_TRUE(decodeRelocSuccess);
vISA::GenRelocEntry relocs[] = {reloc0, reloc1};
constexpr uint32_t numRelocations = 2;
bool decodeRelocSuccess = linkerInput.decodeRelocationTable(&relocs, numRelocations, 0);
EXPECT_TRUE(decodeRelocSuccess);
NEO::Linker linker(linkerInput);
NEO::Linker::SegmentInfo globalVarSegment, globalConstSegment, exportedFuncSegment;
globalVarSegment.gpuAddress = 8;
globalVarSegment.segmentSize = 64;
globalConstSegment.gpuAddress = 128;
globalConstSegment.segmentSize = 256;
exportedFuncSegment.gpuAddress = 4096;
exportedFuncSegment.segmentSize = 1024;
NEO::Linker::UnresolvedExternals unresolvedExternals;
NEO::Linker::ExternalFunctionsT externalFunctions;
NEO::Linker::KernelDescriptorsT kernelDescriptors;
KernelDescriptor kernelDescriptor;
kernelDescriptors.push_back(&kernelDescriptor);
kernelDescriptor.kernelAttributes.flags.requiresImplicitArgs = false;
kernelDescriptor.kernelAttributes.flags.useStackCalls = true;
NEO::Linker linker(linkerInput);
NEO::Linker::SegmentInfo globalVarSegment, globalConstSegment, exportedFuncSegment;
globalVarSegment.gpuAddress = 8;
globalVarSegment.segmentSize = 64;
globalConstSegment.gpuAddress = 128;
globalConstSegment.segmentSize = 256;
exportedFuncSegment.gpuAddress = 4096;
exportedFuncSegment.segmentSize = 1024;
NEO::Linker::UnresolvedExternals unresolvedExternals;
NEO::Linker::ExternalFunctionsT externalFunctions;
NEO::Linker::KernelDescriptorsT kernelDescriptors;
KernelDescriptor kernelDescriptor;
kernelDescriptors.push_back(&kernelDescriptor);
kernelDescriptor.kernelAttributes.flags.requiresImplicitArgs = false;
kernelDescriptor.kernelAttributes.flags.useStackCalls = true;
UltDeviceFactory deviceFactory{1, 0};
UltDeviceFactory deviceFactory{1, 0};
std::vector<char> instructionSegment;
char initData = 0x77;
instructionSegment.resize(32, initData);
NEO::Linker::PatchableSegment seg0;
seg0.hostPointer = instructionSegment.data();
seg0.segmentSize = instructionSegment.size();
NEO::Linker::PatchableSegments patchableInstructionSegments{seg0};
std::vector<char> instructionSegment;
char initData = 0x77;
instructionSegment.resize(32, initData);
NEO::Linker::PatchableSegment seg0;
seg0.hostPointer = instructionSegment.data();
seg0.segmentSize = instructionSegment.size();
NEO::Linker::PatchableSegments patchableInstructionSegments{seg0};
auto linkResult = linker.link(globalVarSegment, globalConstSegment, exportedFuncSegment, {},
nullptr, nullptr, patchableInstructionSegments, unresolvedExternals,
deviceFactory.rootDevices[0], nullptr, nullptr, kernelDescriptors, externalFunctions);
EXPECT_EQ(NEO::LinkingStatus::LinkedFully, linkResult);
auto relocatedSymbols = linker.extractRelocatedSymbols();
EXPECT_EQ(0U, unresolvedExternals.size());
EXPECT_EQ(0U, relocatedSymbols.size());
auto linkResult = linker.link(globalVarSegment, globalConstSegment, exportedFuncSegment, {},
nullptr, nullptr, patchableInstructionSegments, unresolvedExternals,
deviceFactory.rootDevices[0], nullptr, nullptr, kernelDescriptors, externalFunctions);
EXPECT_EQ(NEO::LinkingStatus::LinkedFully, linkResult);
auto relocatedSymbols = linker.extractRelocatedSymbols();
EXPECT_EQ(0U, unresolvedExternals.size());
EXPECT_EQ(0U, relocatedSymbols.size());
for (const auto &reloc : relocs) {
auto addressToPatch = reinterpret_cast<const uint32_t *>(instructionSegment.data() + reloc.r_offset);
EXPECT_EQ(sizeof(ImplicitArgs), *addressToPatch);
EXPECT_TRUE(kernelDescriptor.kernelAttributes.flags.requiresImplicitArgs);
for (const auto &reloc : relocs) {
auto addressToPatch = reinterpret_cast<const uint32_t *>(instructionSegment.data() + reloc.r_offset);
EXPECT_EQ(sizeof(ImplicitArgs), *addressToPatch);
EXPECT_TRUE(kernelDescriptor.kernelAttributes.flags.requiresImplicitArgs);
}
}
}