intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2026-01-08 05:56:36 +08:00
Code Issues Packages Projects Releases Wiki Activity

Correct storing KernelInfos in multi device programs

Browse Source 
Related-To: NEO-5001
Signed-off-by: Mateusz Jablonski <mateusz.jablonski@intel.com>
This commit is contained in:
Mateusz Jablonski
2020-12-01 13:20:58 +00:00
committed by Compute-Runtime-Automation
parent 2e2d5584b8
commit 077ae2726e
5 changed files with 84 additions and 21 deletions
Download Patch File Download Diff File Expand all files Collapse all files

68
opencl/test/unit_test/program/program_tests.cpp
View File

@@ -621,7 +621,9 @@ TEST_P(ProgramFromBinaryTest, whenProgramIsBeingRebuildThenOutdatedGlobalBuffers
TEST_P(ProgramFromBinaryTest, givenProgramWhenCleanKernelInfoIsCalledThenKernelAllocationIsFreed) {
pProgram->build(pProgram->getDevices(), nullptr, true);
EXPECT_EQ(1u, pProgram->getNumKernels());
pProgram->cleanCurrentKernelInfo();
for (auto i = 0u; i < pProgram->buildInfos.size(); i++) {
pProgram->cleanCurrentKernelInfo(i);
}
EXPECT_EQ(0u, pProgram->getNumKernels());
}
@@ -632,7 +634,7 @@ HWTEST_P(ProgramFromBinaryTest, givenProgramWhenCleanCurrentKernelInfoIsCalledBu
auto kernelAllocation = pProgram->getKernelInfo(static_cast<size_t>(0u), rootDeviceIndex)->getGraphicsAllocation();
kernelAllocation->updateTaskCount(100, csr.getOsContext().getContextId());
*csr.getTagAddress() = 0;
pProgram->cleanCurrentKernelInfo();
pProgram->cleanCurrentKernelInfo(rootDeviceIndex);
EXPECT_FALSE(csr.getTemporaryAllocations().peekIsEmpty());
EXPECT_EQ(csr.getTemporaryAllocations().peekHead(), kernelAllocation);
EXPECT_TRUE(this->pDevice->getUltCommandStreamReceiver<FamilyType>().requiresInstructionCacheFlush);
@@ -658,7 +660,7 @@ HWTEST_P(ProgramFromBinaryTest, givenIsaAllocationUsedByMultipleCsrsWhenItIsDele
EXPECT_FALSE(csr0.requiresInstructionCacheFlush);
EXPECT_FALSE(csr1.requiresInstructionCacheFlush);
pProgram->cleanCurrentKernelInfo();
pProgram->cleanCurrentKernelInfo(rootDeviceIndex);
EXPECT_TRUE(csr0.requiresInstructionCacheFlush);
EXPECT_TRUE(csr1.requiresInstructionCacheFlush);
}
@@ -3063,6 +3065,66 @@ TEST(BuildProgramTest, givenMultiDeviceProgramWhenBuildingThenStoreAndProcessBin
EXPECT_EQ(CL_SUCCESS, retVal);
}
TEST(BuildProgramTest, givenMultiDeviceProgramWhenBuildingThenStoreKernelInfoPerEachRootDevice) {
MockProgram *pProgram = nullptr;
std::unique_ptr<char[]> pSource = nullptr;
size_t sourceSize = 0;
std::string testFile;
KernelBinaryHelper kbHelper("CopyBuffer_simd16");
testFile.append(clFiles);
testFile.append("CopyBuffer_simd16.cl");
pSource = loadDataFromFile(
testFile.c_str(),
sourceSize);
ASSERT_NE(0u, sourceSize);
ASSERT_NE(nullptr, pSource);
const char *sources[1] = {pSource.get()};
MockUnrestrictiveContextMultiGPU context;
cl_int retVal = CL_INVALID_PROGRAM;
pProgram = Program::create<MockProgram>(
&context,
1,
sources,
&sourceSize,
retVal);
EXPECT_NE(nullptr, pProgram);
ASSERT_EQ(CL_SUCCESS, retVal);
cl_build_status buildStatus;
for (const auto &device : context.getDevices()) {
retVal = clGetProgramBuildInfo(pProgram, device, CL_PROGRAM_BUILD_STATUS, sizeof(buildStatus), &buildStatus, NULL);
EXPECT_EQ(CL_SUCCESS, retVal);
EXPECT_EQ(CL_BUILD_NONE, buildStatus);
}
retVal = clBuildProgram(
pProgram,
0,
nullptr,
nullptr,
nullptr,
nullptr);
ASSERT_EQ(CL_SUCCESS, retVal);
for (auto &rootDeviceIndex : context.getRootDeviceIndices()) {
EXPECT_LT(0u, pProgram->getNumKernels());
for (auto i = 0u; i < pProgram->getNumKernels(); i++) {
EXPECT_NE(nullptr, pProgram->getKernelInfo(i, rootDeviceIndex));
}
}
retVal = clReleaseProgram(pProgram);
EXPECT_EQ(CL_SUCCESS, retVal);
}
TEST(ProgramTest, whenProgramIsBuiltAsAnExecutableForAtLeastOneDeviceThenIsBuiltMethodReturnsTrue) {
MockSpecializedContext context;
MockProgram program(&context, false, context.getDevices());