mirror of
https://github.com/intel/compute-runtime.git
synced 2026-01-05 09:09:04 +08:00
Set arg immediate for all Kernel devices
convert tests to multi root device scenario Related-To: NEO-5001 Signed-off-by: Mateusz Jablonski <mateusz.jablonski@intel.com>
This commit is contained in:
Mateusz Jablonski
committed by
Compute-Runtime-Automation
Compute-Runtime-Automation
parent
e665111013
commit
c3083592ad
@@ -6,7 +6,7 @@
|
||||
*/
|
||||
|
||||
#include "opencl/source/kernel/kernel.h"
|
||||
#include "opencl/test/unit_test/fixtures/cl_device_fixture.h"
|
||||
#include "opencl/test/unit_test/fixtures/multi_root_device_fixture.h"
|
||||
#include "opencl/test/unit_test/mocks/mock_context.h"
|
||||
#include "opencl/test/unit_test/mocks/mock_kernel.h"
|
||||
#include "opencl/test/unit_test/mocks/mock_program.h"
|
||||
@@ -18,64 +18,68 @@
|
||||
using namespace NEO;
|
||||
|
||||
template <typename T>
|
||||
class KernelArgImmediateTest : public Test<ClDeviceFixture> {
|
||||
class KernelArgImmediateTest : public MultiRootDeviceWithSubDevicesFixture {
|
||||
public:
|
||||
protected:
|
||||
void SetUp() override {
|
||||
ClDeviceFixture::SetUp();
|
||||
memset(pCrossThreadData, 0xfe, sizeof(pCrossThreadData));
|
||||
program = std::make_unique<MockProgram>(toClDeviceVector(*pClDevice));
|
||||
MultiRootDeviceWithSubDevicesFixture::SetUp();
|
||||
program = std::make_unique<MockProgram>(context.get(), false, context->getDevices());
|
||||
|
||||
// define kernel info
|
||||
pKernelInfo = std::make_unique<KernelInfo>();
|
||||
pKernelInfo->kernelDescriptor.kernelAttributes.simdSize = 1;
|
||||
KernelInfoContainer kernelInfos;
|
||||
kernelInfos.resize(3);
|
||||
for (auto &rootDeviceIndex : this->context->getRootDeviceIndices()) {
|
||||
memset(&pCrossThreadData[rootDeviceIndex], 0xfe, sizeof(pCrossThreadData[rootDeviceIndex]));
|
||||
|
||||
// setup kernel arg offsets
|
||||
KernelArgPatchInfo kernelArgPatchInfo;
|
||||
// define kernel info
|
||||
pKernelInfo[rootDeviceIndex] = std::make_unique<KernelInfo>();
|
||||
pKernelInfo[rootDeviceIndex]->kernelDescriptor.kernelAttributes.simdSize = 1;
|
||||
|
||||
pKernelInfo->kernelArgInfo.resize(4);
|
||||
pKernelInfo->kernelArgInfo[3].kernelArgPatchInfoVector.push_back(kernelArgPatchInfo);
|
||||
pKernelInfo->kernelArgInfo[3].kernelArgPatchInfoVector.push_back(kernelArgPatchInfo);
|
||||
pKernelInfo->kernelArgInfo[3].kernelArgPatchInfoVector.push_back(kernelArgPatchInfo);
|
||||
pKernelInfo->kernelArgInfo[2].kernelArgPatchInfoVector.push_back(kernelArgPatchInfo);
|
||||
pKernelInfo->kernelArgInfo[1].kernelArgPatchInfoVector.push_back(kernelArgPatchInfo);
|
||||
pKernelInfo->kernelArgInfo[0].kernelArgPatchInfoVector.push_back(kernelArgPatchInfo);
|
||||
// setup kernel arg offsets
|
||||
KernelArgPatchInfo kernelArgPatchInfo;
|
||||
|
||||
pKernelInfo->kernelArgInfo[3].kernelArgPatchInfoVector[2].crossthreadOffset = 0x38;
|
||||
pKernelInfo->kernelArgInfo[3].kernelArgPatchInfoVector[1].crossthreadOffset = 0x28;
|
||||
pKernelInfo->kernelArgInfo[3].kernelArgPatchInfoVector[0].crossthreadOffset = 0x20;
|
||||
pKernelInfo->kernelArgInfo[2].kernelArgPatchInfoVector[0].crossthreadOffset = 0x30;
|
||||
pKernelInfo->kernelArgInfo[1].kernelArgPatchInfoVector[0].crossthreadOffset = 0x40;
|
||||
pKernelInfo->kernelArgInfo[0].kernelArgPatchInfoVector[0].crossthreadOffset = 0x50;
|
||||
pKernelInfo->kernelArgInfo[3].kernelArgPatchInfoVector[2].size = sizeof(T);
|
||||
pKernelInfo->kernelArgInfo[3].kernelArgPatchInfoVector[1].size = sizeof(T);
|
||||
pKernelInfo->kernelArgInfo[3].kernelArgPatchInfoVector[0].size = sizeof(T);
|
||||
pKernelInfo->kernelArgInfo[2].kernelArgPatchInfoVector[0].size = sizeof(T);
|
||||
pKernelInfo->kernelArgInfo[1].kernelArgPatchInfoVector[0].size = sizeof(T);
|
||||
pKernelInfo->kernelArgInfo[0].kernelArgPatchInfoVector[0].size = sizeof(T);
|
||||
pKernelInfo[rootDeviceIndex]->kernelArgInfo.resize(4);
|
||||
pKernelInfo[rootDeviceIndex]->kernelArgInfo[3].kernelArgPatchInfoVector.push_back(kernelArgPatchInfo);
|
||||
pKernelInfo[rootDeviceIndex]->kernelArgInfo[3].kernelArgPatchInfoVector.push_back(kernelArgPatchInfo);
|
||||
pKernelInfo[rootDeviceIndex]->kernelArgInfo[3].kernelArgPatchInfoVector.push_back(kernelArgPatchInfo);
|
||||
pKernelInfo[rootDeviceIndex]->kernelArgInfo[2].kernelArgPatchInfoVector.push_back(kernelArgPatchInfo);
|
||||
pKernelInfo[rootDeviceIndex]->kernelArgInfo[1].kernelArgPatchInfoVector.push_back(kernelArgPatchInfo);
|
||||
pKernelInfo[rootDeviceIndex]->kernelArgInfo[0].kernelArgPatchInfoVector.push_back(kernelArgPatchInfo);
|
||||
|
||||
program = std::make_unique<MockProgram>(toClDeviceVector(*pClDevice));
|
||||
pKernel = new MockKernel(program.get(), MockKernel::toKernelInfoContainer(*pKernelInfo, rootDeviceIndex));
|
||||
pKernelInfo[rootDeviceIndex]->kernelArgInfo[3].kernelArgPatchInfoVector[2].crossthreadOffset = 0x38;
|
||||
pKernelInfo[rootDeviceIndex]->kernelArgInfo[3].kernelArgPatchInfoVector[1].crossthreadOffset = 0x28;
|
||||
pKernelInfo[rootDeviceIndex]->kernelArgInfo[3].kernelArgPatchInfoVector[0].crossthreadOffset = 0x20;
|
||||
pKernelInfo[rootDeviceIndex]->kernelArgInfo[2].kernelArgPatchInfoVector[0].crossthreadOffset = 0x30;
|
||||
pKernelInfo[rootDeviceIndex]->kernelArgInfo[1].kernelArgPatchInfoVector[0].crossthreadOffset = 0x40;
|
||||
pKernelInfo[rootDeviceIndex]->kernelArgInfo[0].kernelArgPatchInfoVector[0].crossthreadOffset = 0x50;
|
||||
pKernelInfo[rootDeviceIndex]->kernelArgInfo[3].kernelArgPatchInfoVector[2].size = sizeof(T);
|
||||
pKernelInfo[rootDeviceIndex]->kernelArgInfo[3].kernelArgPatchInfoVector[1].size = sizeof(T);
|
||||
pKernelInfo[rootDeviceIndex]->kernelArgInfo[3].kernelArgPatchInfoVector[0].size = sizeof(T);
|
||||
pKernelInfo[rootDeviceIndex]->kernelArgInfo[2].kernelArgPatchInfoVector[0].size = sizeof(T);
|
||||
pKernelInfo[rootDeviceIndex]->kernelArgInfo[1].kernelArgPatchInfoVector[0].size = sizeof(T);
|
||||
pKernelInfo[rootDeviceIndex]->kernelArgInfo[0].kernelArgPatchInfoVector[0].size = sizeof(T);
|
||||
|
||||
kernelInfos[rootDeviceIndex] = pKernelInfo[rootDeviceIndex].get();
|
||||
}
|
||||
|
||||
pKernel = new MockKernel(program.get(), kernelInfos);
|
||||
ASSERT_EQ(CL_SUCCESS, pKernel->initialize());
|
||||
pKernel->setCrossThreadData(pCrossThreadData, sizeof(pCrossThreadData));
|
||||
|
||||
pKernel->setKernelArgHandler(0, &Kernel::setArgImmediate);
|
||||
pKernel->setKernelArgHandler(1, &Kernel::setArgImmediate);
|
||||
pKernel->setKernelArgHandler(2, &Kernel::setArgImmediate);
|
||||
pKernel->setKernelArgHandler(3, &Kernel::setArgImmediate);
|
||||
for (auto &rootDeviceIndex : this->context->getRootDeviceIndices()) {
|
||||
pKernel->setCrossThreadDataForRootDeviceIndex(rootDeviceIndex, &pCrossThreadData[rootDeviceIndex], sizeof(pCrossThreadData[rootDeviceIndex]));
|
||||
}
|
||||
}
|
||||
|
||||
void TearDown() override {
|
||||
delete pKernel;
|
||||
|
||||
ClDeviceFixture::TearDown();
|
||||
MultiRootDeviceWithSubDevicesFixture::TearDown();
|
||||
}
|
||||
|
||||
cl_int retVal = CL_SUCCESS;
|
||||
std::unique_ptr<MockProgram> program;
|
||||
MockKernel *pKernel = nullptr;
|
||||
std::unique_ptr<KernelInfo> pKernelInfo;
|
||||
char pCrossThreadData[0x60];
|
||||
std::unique_ptr<KernelInfo> pKernelInfo[3];
|
||||
char pCrossThreadData[3][0x60];
|
||||
};
|
||||
|
||||
typedef ::testing::Types<
|
||||
@@ -97,10 +101,12 @@ TYPED_TEST(KernelArgImmediateTest, WhenSettingKernelArgThenArgIsSetCorrectly) {
|
||||
auto pVal = &val;
|
||||
this->pKernel->setArg(0, sizeof(TypeParam), pVal);
|
||||
|
||||
auto pKernelArg = (TypeParam *)(this->pKernel->getCrossThreadData(this->rootDeviceIndex) +
|
||||
this->pKernelInfo->kernelArgInfo[0].kernelArgPatchInfoVector[0].crossthreadOffset);
|
||||
for (auto &rootDeviceIndex : this->context->getRootDeviceIndices()) {
|
||||
auto pKernelArg = (TypeParam *)(this->pKernel->getCrossThreadData(rootDeviceIndex) +
|
||||
this->pKernelInfo[rootDeviceIndex]->kernelArgInfo[0].kernelArgPatchInfoVector[0].crossthreadOffset);
|
||||
|
||||
EXPECT_EQ(val, *pKernelArg);
|
||||
EXPECT_EQ(val, *pKernelArg);
|
||||
}
|
||||
}
|
||||
|
||||
TYPED_TEST(KernelArgImmediateTest, GivenInvalidIndexWhenSettingKernelArgThenInvalidArgIndexErrorIsReturned) {
|
||||
@@ -114,55 +120,65 @@ TYPED_TEST(KernelArgImmediateTest, GivenInvalidIndexWhenSettingKernelArgThenInva
|
||||
TYPED_TEST(KernelArgImmediateTest, GivenMultipleArgumentsWhenSettingKernelArgThenEachArgIsSetCorrectly) {
|
||||
auto val = (TypeParam)0xaaaaaaaaULL;
|
||||
auto pVal = &val;
|
||||
|
||||
this->pKernel->setArg(0, sizeof(TypeParam), pVal);
|
||||
for (auto &rootDeviceIndex : this->context->getRootDeviceIndices()) {
|
||||
|
||||
auto pKernelArg = (TypeParam *)(this->pKernel->getCrossThreadData(this->rootDeviceIndex) +
|
||||
this->pKernelInfo->kernelArgInfo[0].kernelArgPatchInfoVector[0].crossthreadOffset);
|
||||
|
||||
EXPECT_EQ(val, *pKernelArg);
|
||||
auto pKernelArg = (TypeParam *)(this->pKernel->getCrossThreadData(rootDeviceIndex) +
|
||||
this->pKernelInfo[rootDeviceIndex]->kernelArgInfo[0].kernelArgPatchInfoVector[0].crossthreadOffset);
|
||||
|
||||
EXPECT_EQ(val, *pKernelArg);
|
||||
}
|
||||
val = (TypeParam)0xbbbbbbbbULL;
|
||||
this->pKernel->setArg(1, sizeof(TypeParam), &val);
|
||||
|
||||
pKernelArg = (TypeParam *)(this->pKernel->getCrossThreadData(this->rootDeviceIndex) +
|
||||
this->pKernelInfo->kernelArgInfo[1].kernelArgPatchInfoVector[0].crossthreadOffset);
|
||||
|
||||
EXPECT_EQ(val, *pKernelArg);
|
||||
for (auto &rootDeviceIndex : this->context->getRootDeviceIndices()) {
|
||||
auto pKernelArg = (TypeParam *)(this->pKernel->getCrossThreadData(rootDeviceIndex) +
|
||||
this->pKernelInfo[rootDeviceIndex]->kernelArgInfo[1].kernelArgPatchInfoVector[0].crossthreadOffset);
|
||||
|
||||
EXPECT_EQ(val, *pKernelArg);
|
||||
}
|
||||
val = (TypeParam)0xccccccccULL;
|
||||
this->pKernel->setArg(2, sizeof(TypeParam), &val);
|
||||
|
||||
pKernelArg = (TypeParam *)(this->pKernel->getCrossThreadData(this->rootDeviceIndex) +
|
||||
this->pKernelInfo->kernelArgInfo[2].kernelArgPatchInfoVector[0].crossthreadOffset);
|
||||
for (auto &rootDeviceIndex : this->context->getRootDeviceIndices()) {
|
||||
auto pKernelArg = (TypeParam *)(this->pKernel->getCrossThreadData(rootDeviceIndex) +
|
||||
this->pKernelInfo[rootDeviceIndex]->kernelArgInfo[2].kernelArgPatchInfoVector[0].crossthreadOffset);
|
||||
|
||||
EXPECT_EQ(val, *pKernelArg);
|
||||
EXPECT_EQ(val, *pKernelArg);
|
||||
}
|
||||
}
|
||||
|
||||
TYPED_TEST(KernelArgImmediateTest, GivenCrossThreadDataOverwritesWhenSettingKernelArgThenArgsAreSetCorrectly) {
|
||||
TypeParam val = (TypeParam)0xaaaaaaaaULL;
|
||||
TypeParam *pVal = &val;
|
||||
this->pKernel->setArg(0, sizeof(TypeParam), pVal);
|
||||
for (auto &rootDeviceIndex : this->context->getRootDeviceIndices()) {
|
||||
|
||||
TypeParam *pKernelArg = (TypeParam *)(this->pKernel->getCrossThreadData(this->rootDeviceIndex) +
|
||||
this->pKernelInfo->kernelArgInfo[0].kernelArgPatchInfoVector[0].crossthreadOffset);
|
||||
TypeParam *pKernelArg = (TypeParam *)(this->pKernel->getCrossThreadData(rootDeviceIndex) +
|
||||
this->pKernelInfo[rootDeviceIndex]->kernelArgInfo[0].kernelArgPatchInfoVector[0].crossthreadOffset);
|
||||
|
||||
EXPECT_EQ(val, *pKernelArg);
|
||||
EXPECT_EQ(val, *pKernelArg);
|
||||
}
|
||||
|
||||
val = (TypeParam)0xbbbbbbbbULL;
|
||||
this->pKernel->setArg(1, sizeof(TypeParam), &val);
|
||||
|
||||
pKernelArg = (TypeParam *)(this->pKernel->getCrossThreadData(this->rootDeviceIndex) +
|
||||
this->pKernelInfo->kernelArgInfo[1].kernelArgPatchInfoVector[0].crossthreadOffset);
|
||||
|
||||
EXPECT_EQ(val, *pKernelArg);
|
||||
for (auto &rootDeviceIndex : this->context->getRootDeviceIndices()) {
|
||||
auto pKernelArg = (TypeParam *)(this->pKernel->getCrossThreadData(rootDeviceIndex) +
|
||||
this->pKernelInfo[rootDeviceIndex]->kernelArgInfo[1].kernelArgPatchInfoVector[0].crossthreadOffset);
|
||||
|
||||
EXPECT_EQ(val, *pKernelArg);
|
||||
}
|
||||
val = (TypeParam)0xccccccccULL;
|
||||
this->pKernel->setArg(0, sizeof(TypeParam), &val);
|
||||
|
||||
pKernelArg = (TypeParam *)(this->pKernel->getCrossThreadData(this->rootDeviceIndex) +
|
||||
this->pKernelInfo->kernelArgInfo[0].kernelArgPatchInfoVector[0].crossthreadOffset);
|
||||
for (auto &rootDeviceIndex : this->context->getRootDeviceIndices()) {
|
||||
auto pKernelArg = (TypeParam *)(this->pKernel->getCrossThreadData(rootDeviceIndex) +
|
||||
this->pKernelInfo[rootDeviceIndex]->kernelArgInfo[0].kernelArgPatchInfoVector[0].crossthreadOffset);
|
||||
|
||||
EXPECT_EQ(val, *pKernelArg);
|
||||
EXPECT_EQ(val, *pKernelArg);
|
||||
}
|
||||
}
|
||||
|
||||
TYPED_TEST(KernelArgImmediateTest, GivenMultipleStructElementsWhenSettingKernelArgThenArgsAreSetCorrectly) {
|
||||
@@ -178,155 +194,170 @@ TYPED_TEST(KernelArgImmediateTest, GivenMultipleStructElementsWhenSettingKernelA
|
||||
immediateStruct.unused[1] = 0xfe;
|
||||
immediateStruct.unused[2] = 0xfe;
|
||||
|
||||
this->pKernelInfo->kernelArgInfo[3].kernelArgPatchInfoVector[0].sourceOffset = offsetof(struct ImmediateStruct, a);
|
||||
this->pKernelInfo->kernelArgInfo[3].kernelArgPatchInfoVector[1].sourceOffset = offsetof(struct ImmediateStruct, b);
|
||||
|
||||
for (auto &rootDeviceIndex : this->context->getRootDeviceIndices()) {
|
||||
this->pKernelInfo[rootDeviceIndex]->kernelArgInfo[3].kernelArgPatchInfoVector[0].sourceOffset = offsetof(struct ImmediateStruct, a);
|
||||
this->pKernelInfo[rootDeviceIndex]->kernelArgInfo[3].kernelArgPatchInfoVector[1].sourceOffset = offsetof(struct ImmediateStruct, b);
|
||||
}
|
||||
this->pKernel->setArg(3, sizeof(immediateStruct), &immediateStruct);
|
||||
|
||||
auto pCrossthreadA = (TypeParam *)(this->pKernel->getCrossThreadData(this->rootDeviceIndex) +
|
||||
this->pKernelInfo->kernelArgInfo[3].kernelArgPatchInfoVector[0].crossthreadOffset);
|
||||
EXPECT_EQ(immediateStruct.a, *pCrossthreadA);
|
||||
for (auto &rootDeviceIndex : this->context->getRootDeviceIndices()) {
|
||||
auto pCrossthreadA = (TypeParam *)(this->pKernel->getCrossThreadData(rootDeviceIndex) +
|
||||
this->pKernelInfo[rootDeviceIndex]->kernelArgInfo[3].kernelArgPatchInfoVector[0].crossthreadOffset);
|
||||
EXPECT_EQ(immediateStruct.a, *pCrossthreadA);
|
||||
|
||||
auto pCrossthreadB = (TypeParam *)(this->pKernel->getCrossThreadData(this->rootDeviceIndex) +
|
||||
this->pKernelInfo->kernelArgInfo[3].kernelArgPatchInfoVector[1].crossthreadOffset);
|
||||
EXPECT_EQ(immediateStruct.b, *pCrossthreadB);
|
||||
auto pCrossthreadB = (TypeParam *)(this->pKernel->getCrossThreadData(rootDeviceIndex) +
|
||||
this->pKernelInfo[rootDeviceIndex]->kernelArgInfo[3].kernelArgPatchInfoVector[1].crossthreadOffset);
|
||||
EXPECT_EQ(immediateStruct.b, *pCrossthreadB);
|
||||
}
|
||||
}
|
||||
|
||||
TYPED_TEST(KernelArgImmediateTest, givenTooLargePatchSizeWhenSettingArgThenDontReadMemoryBeyondLimit) {
|
||||
TypeParam memory[2];
|
||||
std::memset(&memory[0], 0xaa, sizeof(TypeParam));
|
||||
std::memset(&memory[1], 0xbb, sizeof(TypeParam));
|
||||
|
||||
const auto destinationMemoryAddress = this->pKernel->getCrossThreadData(this->rootDeviceIndex) +
|
||||
this->pKernelInfo->kernelArgInfo[0].kernelArgPatchInfoVector[0].crossthreadOffset;
|
||||
const auto memoryBeyondLimitAddress = destinationMemoryAddress + sizeof(TypeParam);
|
||||
for (auto &rootDeviceIndex : this->context->getRootDeviceIndices()) {
|
||||
TypeParam memory[2];
|
||||
std::memset(&memory[0], 0xaa, sizeof(TypeParam));
|
||||
std::memset(&memory[1], 0xbb, sizeof(TypeParam));
|
||||
|
||||
const auto memoryBeyondLimitBefore = *reinterpret_cast<TypeParam *>(memoryBeyondLimitAddress);
|
||||
const auto destinationMemoryAddress = this->pKernel->getCrossThreadData(rootDeviceIndex) +
|
||||
this->pKernelInfo[rootDeviceIndex]->kernelArgInfo[0].kernelArgPatchInfoVector[0].crossthreadOffset;
|
||||
const auto memoryBeyondLimitAddress = destinationMemoryAddress + sizeof(TypeParam);
|
||||
|
||||
this->pKernelInfo->kernelArgInfo[0].kernelArgPatchInfoVector[0].size = sizeof(TypeParam) + 1;
|
||||
auto retVal = this->pKernel->setArg(0, sizeof(TypeParam), &memory[0]);
|
||||
const auto memoryBeyondLimitBefore = *reinterpret_cast<TypeParam *>(memoryBeyondLimitAddress);
|
||||
|
||||
const auto memoryBeyondLimitAfter = *reinterpret_cast<TypeParam *>(memoryBeyondLimitAddress);
|
||||
EXPECT_EQ(memoryBeyondLimitBefore, memoryBeyondLimitAfter);
|
||||
EXPECT_EQ(memory[0], *reinterpret_cast<TypeParam *>(destinationMemoryAddress));
|
||||
this->pKernelInfo[rootDeviceIndex]->kernelArgInfo[0].kernelArgPatchInfoVector[0].size = sizeof(TypeParam) + 1;
|
||||
auto retVal = this->pKernel->setArg(0, sizeof(TypeParam), &memory[0]);
|
||||
|
||||
EXPECT_EQ(CL_SUCCESS, retVal);
|
||||
const auto memoryBeyondLimitAfter = *reinterpret_cast<TypeParam *>(memoryBeyondLimitAddress);
|
||||
EXPECT_EQ(memoryBeyondLimitBefore, memoryBeyondLimitAfter);
|
||||
EXPECT_EQ(memory[0], *reinterpret_cast<TypeParam *>(destinationMemoryAddress));
|
||||
|
||||
EXPECT_EQ(CL_SUCCESS, retVal);
|
||||
}
|
||||
}
|
||||
|
||||
TYPED_TEST(KernelArgImmediateTest, givenNotTooLargePatchSizeWhenSettingArgThenDontReadMemoryBeyondLimit) {
|
||||
TypeParam memory[2];
|
||||
std::memset(&memory[0], 0xaa, sizeof(TypeParam));
|
||||
std::memset(&memory[1], 0xbb, sizeof(TypeParam));
|
||||
|
||||
const auto destinationMemoryAddress = this->pKernel->getCrossThreadData(this->rootDeviceIndex) +
|
||||
this->pKernelInfo->kernelArgInfo[0].kernelArgPatchInfoVector[0].crossthreadOffset;
|
||||
const auto memoryBeyondLimitAddress = destinationMemoryAddress + sizeof(TypeParam);
|
||||
for (auto &rootDeviceIndex : this->context->getRootDeviceIndices()) {
|
||||
TypeParam memory[2];
|
||||
std::memset(&memory[0], 0xaa, sizeof(TypeParam));
|
||||
std::memset(&memory[1], 0xbb, sizeof(TypeParam));
|
||||
|
||||
const auto memoryBeyondLimitBefore = *reinterpret_cast<TypeParam *>(memoryBeyondLimitAddress);
|
||||
const auto destinationMemoryAddress = this->pKernel->getCrossThreadData(rootDeviceIndex) +
|
||||
this->pKernelInfo[rootDeviceIndex]->kernelArgInfo[0].kernelArgPatchInfoVector[0].crossthreadOffset;
|
||||
const auto memoryBeyondLimitAddress = destinationMemoryAddress + sizeof(TypeParam);
|
||||
|
||||
this->pKernelInfo->kernelArgInfo[0].kernelArgPatchInfoVector[0].size = sizeof(TypeParam);
|
||||
auto retVal = this->pKernel->setArg(0, sizeof(TypeParam), &memory[0]);
|
||||
const auto memoryBeyondLimitBefore = *reinterpret_cast<TypeParam *>(memoryBeyondLimitAddress);
|
||||
|
||||
const auto memoryBeyondLimitAfter = *reinterpret_cast<TypeParam *>(memoryBeyondLimitAddress);
|
||||
EXPECT_EQ(memoryBeyondLimitBefore, memoryBeyondLimitAfter);
|
||||
EXPECT_EQ(memory[0], *reinterpret_cast<TypeParam *>(destinationMemoryAddress));
|
||||
this->pKernelInfo[rootDeviceIndex]->kernelArgInfo[0].kernelArgPatchInfoVector[0].size = sizeof(TypeParam);
|
||||
auto retVal = this->pKernel->setArg(0, sizeof(TypeParam), &memory[0]);
|
||||
|
||||
EXPECT_EQ(CL_SUCCESS, retVal);
|
||||
const auto memoryBeyondLimitAfter = *reinterpret_cast<TypeParam *>(memoryBeyondLimitAddress);
|
||||
EXPECT_EQ(memoryBeyondLimitBefore, memoryBeyondLimitAfter);
|
||||
EXPECT_EQ(memory[0], *reinterpret_cast<TypeParam *>(destinationMemoryAddress));
|
||||
|
||||
EXPECT_EQ(CL_SUCCESS, retVal);
|
||||
}
|
||||
}
|
||||
|
||||
TYPED_TEST(KernelArgImmediateTest, givenMulitplePatchesAndFirstPatchSizeTooLargeWhenSettingArgThenDontReadMemoryBeyondLimit) {
|
||||
if (sizeof(TypeParam) == 1)
|
||||
return; // multiple patch chars don't make sense
|
||||
|
||||
TypeParam memory[2];
|
||||
std::memset(&memory[0], 0xaa, sizeof(TypeParam));
|
||||
std::memset(&memory[1], 0xbb, sizeof(TypeParam));
|
||||
for (auto &rootDeviceIndex : this->context->getRootDeviceIndices()) {
|
||||
TypeParam memory[2];
|
||||
std::memset(&memory[0], 0xaa, sizeof(TypeParam));
|
||||
std::memset(&memory[1], 0xbb, sizeof(TypeParam));
|
||||
|
||||
const auto destinationMemoryAddress1 = this->pKernel->getCrossThreadData(this->rootDeviceIndex) +
|
||||
this->pKernelInfo->kernelArgInfo[3].kernelArgPatchInfoVector[2].crossthreadOffset;
|
||||
const auto destinationMemoryAddress2 = this->pKernel->getCrossThreadData(this->rootDeviceIndex) +
|
||||
this->pKernelInfo->kernelArgInfo[3].kernelArgPatchInfoVector[1].crossthreadOffset;
|
||||
const auto memoryBeyondLimitAddress1 = destinationMemoryAddress1 + sizeof(TypeParam);
|
||||
const auto memoryBeyondLimitAddress2 = destinationMemoryAddress2 + sizeof(TypeParam) / 2;
|
||||
const auto destinationMemoryAddress1 = this->pKernel->getCrossThreadData(rootDeviceIndex) +
|
||||
this->pKernelInfo[rootDeviceIndex]->kernelArgInfo[3].kernelArgPatchInfoVector[2].crossthreadOffset;
|
||||
const auto destinationMemoryAddress2 = this->pKernel->getCrossThreadData(rootDeviceIndex) +
|
||||
this->pKernelInfo[rootDeviceIndex]->kernelArgInfo[3].kernelArgPatchInfoVector[1].crossthreadOffset;
|
||||
const auto memoryBeyondLimitAddress1 = destinationMemoryAddress1 + sizeof(TypeParam);
|
||||
const auto memoryBeyondLimitAddress2 = destinationMemoryAddress2 + sizeof(TypeParam) / 2;
|
||||
|
||||
const std::vector<unsigned char> memoryBeyondLimitBefore1(memoryBeyondLimitAddress1, memoryBeyondLimitAddress1 + sizeof(TypeParam));
|
||||
const std::vector<unsigned char> memoryBeyondLimitBefore2(memoryBeyondLimitAddress2, memoryBeyondLimitAddress2 + sizeof(TypeParam) / 2);
|
||||
const std::vector<unsigned char> memoryBeyondLimitBefore1(memoryBeyondLimitAddress1, memoryBeyondLimitAddress1 + sizeof(TypeParam));
|
||||
const std::vector<unsigned char> memoryBeyondLimitBefore2(memoryBeyondLimitAddress2, memoryBeyondLimitAddress2 + sizeof(TypeParam) / 2);
|
||||
|
||||
this->pKernelInfo->kernelArgInfo[3].kernelArgPatchInfoVector[2].sourceOffset = 0;
|
||||
this->pKernelInfo->kernelArgInfo[3].kernelArgPatchInfoVector[1].sourceOffset = sizeof(TypeParam) / 2;
|
||||
this->pKernelInfo->kernelArgInfo[3].kernelArgPatchInfoVector[2].size = sizeof(TypeParam);
|
||||
this->pKernelInfo->kernelArgInfo[3].kernelArgPatchInfoVector[1].size = sizeof(TypeParam) / 2;
|
||||
auto retVal = this->pKernel->setArg(3, sizeof(TypeParam), &memory[0]);
|
||||
this->pKernelInfo[rootDeviceIndex]->kernelArgInfo[3].kernelArgPatchInfoVector[2].sourceOffset = 0;
|
||||
this->pKernelInfo[rootDeviceIndex]->kernelArgInfo[3].kernelArgPatchInfoVector[1].sourceOffset = sizeof(TypeParam) / 2;
|
||||
this->pKernelInfo[rootDeviceIndex]->kernelArgInfo[3].kernelArgPatchInfoVector[2].size = sizeof(TypeParam);
|
||||
this->pKernelInfo[rootDeviceIndex]->kernelArgInfo[3].kernelArgPatchInfoVector[1].size = sizeof(TypeParam) / 2;
|
||||
auto retVal = this->pKernel->setArg(3, sizeof(TypeParam), &memory[0]);
|
||||
|
||||
EXPECT_EQ(0, std::memcmp(memoryBeyondLimitBefore1.data(), memoryBeyondLimitAddress1, sizeof(TypeParam)));
|
||||
EXPECT_EQ(0, std::memcmp(memoryBeyondLimitBefore2.data(), memoryBeyondLimitAddress2, sizeof(TypeParam) / 2));
|
||||
EXPECT_EQ(0, std::memcmp(memoryBeyondLimitBefore1.data(), memoryBeyondLimitAddress1, sizeof(TypeParam)));
|
||||
EXPECT_EQ(0, std::memcmp(memoryBeyondLimitBefore2.data(), memoryBeyondLimitAddress2, sizeof(TypeParam) / 2));
|
||||
|
||||
EXPECT_EQ(0, std::memcmp(&memory[0], destinationMemoryAddress1, sizeof(TypeParam)));
|
||||
EXPECT_EQ(0, std::memcmp(&memory[0], destinationMemoryAddress2, sizeof(TypeParam) / 2));
|
||||
EXPECT_EQ(0, std::memcmp(&memory[0], destinationMemoryAddress1, sizeof(TypeParam)));
|
||||
EXPECT_EQ(0, std::memcmp(&memory[0], destinationMemoryAddress2, sizeof(TypeParam) / 2));
|
||||
|
||||
EXPECT_EQ(CL_SUCCESS, retVal);
|
||||
EXPECT_EQ(CL_SUCCESS, retVal);
|
||||
}
|
||||
}
|
||||
|
||||
TYPED_TEST(KernelArgImmediateTest, givenMulitplePatchesAndSecondPatchSizeTooLargeWhenSettingArgThenDontReadMemoryBeyondLimit) {
|
||||
if (sizeof(TypeParam) == 1)
|
||||
return; // multiple patch chars don't make sense
|
||||
|
||||
TypeParam memory[2];
|
||||
std::memset(&memory[0], 0xaa, sizeof(TypeParam));
|
||||
std::memset(&memory[1], 0xbb, sizeof(TypeParam));
|
||||
for (auto &rootDeviceIndex : this->context->getRootDeviceIndices()) {
|
||||
TypeParam memory[2];
|
||||
std::memset(&memory[0], 0xaa, sizeof(TypeParam));
|
||||
std::memset(&memory[1], 0xbb, sizeof(TypeParam));
|
||||
|
||||
const auto destinationMemoryAddress1 = this->pKernel->getCrossThreadData(this->rootDeviceIndex) +
|
||||
this->pKernelInfo->kernelArgInfo[3].kernelArgPatchInfoVector[2].crossthreadOffset;
|
||||
const auto destinationMemoryAddress2 = this->pKernel->getCrossThreadData(this->rootDeviceIndex) +
|
||||
this->pKernelInfo->kernelArgInfo[3].kernelArgPatchInfoVector[1].crossthreadOffset;
|
||||
const auto memoryBeyondLimitAddress1 = destinationMemoryAddress1 + sizeof(TypeParam) / 2;
|
||||
const auto memoryBeyondLimitAddress2 = destinationMemoryAddress2 + sizeof(TypeParam) / 2;
|
||||
const auto destinationMemoryAddress1 = this->pKernel->getCrossThreadData(rootDeviceIndex) +
|
||||
this->pKernelInfo[rootDeviceIndex]->kernelArgInfo[3].kernelArgPatchInfoVector[2].crossthreadOffset;
|
||||
const auto destinationMemoryAddress2 = this->pKernel->getCrossThreadData(rootDeviceIndex) +
|
||||
this->pKernelInfo[rootDeviceIndex]->kernelArgInfo[3].kernelArgPatchInfoVector[1].crossthreadOffset;
|
||||
const auto memoryBeyondLimitAddress1 = destinationMemoryAddress1 + sizeof(TypeParam) / 2;
|
||||
const auto memoryBeyondLimitAddress2 = destinationMemoryAddress2 + sizeof(TypeParam) / 2;
|
||||
|
||||
const std::vector<unsigned char> memoryBeyondLimitBefore1(memoryBeyondLimitAddress1, memoryBeyondLimitAddress1 + sizeof(TypeParam) / 2);
|
||||
const std::vector<unsigned char> memoryBeyondLimitBefore2(memoryBeyondLimitAddress2, memoryBeyondLimitAddress2 + sizeof(TypeParam) / 2);
|
||||
const std::vector<unsigned char> memoryBeyondLimitBefore1(memoryBeyondLimitAddress1, memoryBeyondLimitAddress1 + sizeof(TypeParam) / 2);
|
||||
const std::vector<unsigned char> memoryBeyondLimitBefore2(memoryBeyondLimitAddress2, memoryBeyondLimitAddress2 + sizeof(TypeParam) / 2);
|
||||
|
||||
this->pKernelInfo->kernelArgInfo[3].kernelArgPatchInfoVector[0].size = 0;
|
||||
this->pKernelInfo->kernelArgInfo[3].kernelArgPatchInfoVector[2].sourceOffset = 0;
|
||||
this->pKernelInfo->kernelArgInfo[3].kernelArgPatchInfoVector[1].sourceOffset = sizeof(TypeParam) / 2;
|
||||
this->pKernelInfo->kernelArgInfo[3].kernelArgPatchInfoVector[2].size = sizeof(TypeParam) / 2;
|
||||
this->pKernelInfo->kernelArgInfo[3].kernelArgPatchInfoVector[1].size = sizeof(TypeParam);
|
||||
auto retVal = this->pKernel->setArg(3, sizeof(TypeParam), &memory[0]);
|
||||
this->pKernelInfo[rootDeviceIndex]->kernelArgInfo[3].kernelArgPatchInfoVector[0].size = 0;
|
||||
this->pKernelInfo[rootDeviceIndex]->kernelArgInfo[3].kernelArgPatchInfoVector[2].sourceOffset = 0;
|
||||
this->pKernelInfo[rootDeviceIndex]->kernelArgInfo[3].kernelArgPatchInfoVector[1].sourceOffset = sizeof(TypeParam) / 2;
|
||||
this->pKernelInfo[rootDeviceIndex]->kernelArgInfo[3].kernelArgPatchInfoVector[2].size = sizeof(TypeParam) / 2;
|
||||
this->pKernelInfo[rootDeviceIndex]->kernelArgInfo[3].kernelArgPatchInfoVector[1].size = sizeof(TypeParam);
|
||||
auto retVal = this->pKernel->setArg(3, sizeof(TypeParam), &memory[0]);
|
||||
|
||||
EXPECT_EQ(0, std::memcmp(memoryBeyondLimitBefore1.data(), memoryBeyondLimitAddress1, sizeof(TypeParam) / 2));
|
||||
EXPECT_EQ(0, std::memcmp(memoryBeyondLimitBefore2.data(), memoryBeyondLimitAddress2, sizeof(TypeParam) / 2));
|
||||
EXPECT_EQ(0, std::memcmp(memoryBeyondLimitBefore1.data(), memoryBeyondLimitAddress1, sizeof(TypeParam) / 2));
|
||||
EXPECT_EQ(0, std::memcmp(memoryBeyondLimitBefore2.data(), memoryBeyondLimitAddress2, sizeof(TypeParam) / 2));
|
||||
|
||||
EXPECT_EQ(0, std::memcmp(&memory[0], destinationMemoryAddress1, sizeof(TypeParam) / 2));
|
||||
EXPECT_EQ(0, std::memcmp(&memory[0], destinationMemoryAddress2, sizeof(TypeParam) / 2));
|
||||
EXPECT_EQ(0, std::memcmp(&memory[0], destinationMemoryAddress1, sizeof(TypeParam) / 2));
|
||||
EXPECT_EQ(0, std::memcmp(&memory[0], destinationMemoryAddress2, sizeof(TypeParam) / 2));
|
||||
|
||||
EXPECT_EQ(CL_SUCCESS, retVal);
|
||||
EXPECT_EQ(CL_SUCCESS, retVal);
|
||||
}
|
||||
}
|
||||
|
||||
TYPED_TEST(KernelArgImmediateTest, givenMultiplePatchesAndOneSourceOffsetBeyondArgumentWhenSettingArgThenDontCopyThisPatch) {
|
||||
TypeParam memory[2];
|
||||
std::memset(&memory[0], 0xaa, sizeof(TypeParam));
|
||||
std::memset(&memory[1], 0xbb, sizeof(TypeParam));
|
||||
for (auto &rootDeviceIndex : this->context->getRootDeviceIndices()) {
|
||||
TypeParam memory[2];
|
||||
std::memset(&memory[0], 0xaa, sizeof(TypeParam));
|
||||
std::memset(&memory[1], 0xbb, sizeof(TypeParam));
|
||||
|
||||
const auto destinationMemoryAddress1 = this->pKernel->getCrossThreadData(this->rootDeviceIndex) +
|
||||
this->pKernelInfo->kernelArgInfo[3].kernelArgPatchInfoVector[1].crossthreadOffset;
|
||||
const auto destinationMemoryAddress2 = this->pKernel->getCrossThreadData(this->rootDeviceIndex) +
|
||||
this->pKernelInfo->kernelArgInfo[3].kernelArgPatchInfoVector[2].crossthreadOffset;
|
||||
const auto memoryBeyondLimitAddress1 = destinationMemoryAddress1 + sizeof(TypeParam);
|
||||
const auto memoryBeyondLimitAddress2 = destinationMemoryAddress2;
|
||||
const auto destinationMemoryAddress1 = this->pKernel->getCrossThreadData(rootDeviceIndex) +
|
||||
this->pKernelInfo[rootDeviceIndex]->kernelArgInfo[3].kernelArgPatchInfoVector[1].crossthreadOffset;
|
||||
const auto destinationMemoryAddress2 = this->pKernel->getCrossThreadData(rootDeviceIndex) +
|
||||
this->pKernelInfo[rootDeviceIndex]->kernelArgInfo[3].kernelArgPatchInfoVector[2].crossthreadOffset;
|
||||
const auto memoryBeyondLimitAddress1 = destinationMemoryAddress1 + sizeof(TypeParam);
|
||||
const auto memoryBeyondLimitAddress2 = destinationMemoryAddress2;
|
||||
|
||||
const std::vector<unsigned char> memoryBeyondLimitBefore1(memoryBeyondLimitAddress1, memoryBeyondLimitAddress1 + sizeof(TypeParam));
|
||||
const std::vector<unsigned char> memoryBeyondLimitBefore2(memoryBeyondLimitAddress2, memoryBeyondLimitAddress2 + sizeof(TypeParam));
|
||||
const std::vector<unsigned char> memoryBeyondLimitBefore1(memoryBeyondLimitAddress1, memoryBeyondLimitAddress1 + sizeof(TypeParam));
|
||||
const std::vector<unsigned char> memoryBeyondLimitBefore2(memoryBeyondLimitAddress2, memoryBeyondLimitAddress2 + sizeof(TypeParam));
|
||||
|
||||
this->pKernelInfo->kernelArgInfo[3].kernelArgPatchInfoVector[0].size = 0;
|
||||
this->pKernelInfo->kernelArgInfo[3].kernelArgPatchInfoVector[1].sourceOffset = 0;
|
||||
this->pKernelInfo->kernelArgInfo[3].kernelArgPatchInfoVector[1].size = sizeof(TypeParam);
|
||||
this->pKernelInfo->kernelArgInfo[3].kernelArgPatchInfoVector[2].sourceOffset = sizeof(TypeParam);
|
||||
this->pKernelInfo->kernelArgInfo[3].kernelArgPatchInfoVector[2].size = 1;
|
||||
auto retVal = this->pKernel->setArg(3, sizeof(TypeParam), &memory[0]);
|
||||
this->pKernelInfo[rootDeviceIndex]->kernelArgInfo[3].kernelArgPatchInfoVector[0].size = 0;
|
||||
this->pKernelInfo[rootDeviceIndex]->kernelArgInfo[3].kernelArgPatchInfoVector[1].sourceOffset = 0;
|
||||
this->pKernelInfo[rootDeviceIndex]->kernelArgInfo[3].kernelArgPatchInfoVector[1].size = sizeof(TypeParam);
|
||||
this->pKernelInfo[rootDeviceIndex]->kernelArgInfo[3].kernelArgPatchInfoVector[2].sourceOffset = sizeof(TypeParam);
|
||||
this->pKernelInfo[rootDeviceIndex]->kernelArgInfo[3].kernelArgPatchInfoVector[2].size = 1;
|
||||
auto retVal = this->pKernel->setArg(3, sizeof(TypeParam), &memory[0]);
|
||||
|
||||
EXPECT_EQ(0, std::memcmp(memoryBeyondLimitBefore1.data(), memoryBeyondLimitAddress1, memoryBeyondLimitBefore1.size()));
|
||||
EXPECT_EQ(0, std::memcmp(memoryBeyondLimitBefore2.data(), memoryBeyondLimitAddress2, memoryBeyondLimitBefore2.size()));
|
||||
EXPECT_EQ(0, std::memcmp(memoryBeyondLimitBefore1.data(), memoryBeyondLimitAddress1, memoryBeyondLimitBefore1.size()));
|
||||
EXPECT_EQ(0, std::memcmp(memoryBeyondLimitBefore2.data(), memoryBeyondLimitAddress2, memoryBeyondLimitBefore2.size()));
|
||||
|
||||
EXPECT_EQ(0, std::memcmp(&memory[0], destinationMemoryAddress1, sizeof(TypeParam)));
|
||||
EXPECT_EQ(0, std::memcmp(&memory[0], destinationMemoryAddress1, sizeof(TypeParam)));
|
||||
|
||||
EXPECT_EQ(CL_SUCCESS, retVal);
|
||||
EXPECT_EQ(CL_SUCCESS, retVal);
|
||||
}
|
||||
}
|
||||
|
||||
Reference in New Issue
Block a user