intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2026-01-10 07:08:04 +08:00
Code Issues Packages Projects Releases Wiki Activity

Add support for allocating RT globals

Browse Source 
Signed-off-by: Bartosz Dunajski <bartosz.dunajski@intel.com>
This commit is contained in:
Bartosz Dunajski
2022-04-25 14:54:13 +00:00
committed by Compute-Runtime-Automation
parent 4a5f1b9358
commit 192ffa4c26
14 changed files with 72 additions and 83 deletions
Download Patch File Download Diff File Expand all files Collapse all files

17
level_zero/core/test/unit_tests/sources/kernel/test_kernel.cpp
View File

@@ -693,6 +693,8 @@ TEST_F(KernelImmutableDataTests, givenKernelWithPrivateMemoryBiggerThanGlobalMem
}
TEST_F(KernelImmutableDataTests, whenHasRTCallsIsTrueThenRayTracingIsInitialized) {
static_cast<OsAgnosticMemoryManager *>(device->getNEODevice()->getMemoryManager())->turnOnFakingBigAllocations();
KernelDescriptor mockDescriptor = {};
mockDescriptor.kernelAttributes.flags.hasRTCalls = true;
mockDescriptor.kernelMetadata.kernelName = "rt_test";
@@ -724,8 +726,6 @@ TEST_F(KernelImmutableDataTests, whenHasRTCallsIsTrueThenRayTracingIsInitialized
immDataVector->push_back(std::move(mockKernelImmutableData));
neoDevice->setRTDispatchGlobalsForceAllocation();
auto result = kernel->initialize(&kernelDesc);
EXPECT_EQ(ZE_RESULT_SUCCESS, result);
EXPECT_NE(nullptr, module.get()->getDevice()->getNEODevice()->getRTMemoryBackedBuffer());
@@ -853,6 +853,8 @@ TEST_F(KernelImmutableDataTests, whenHasRTCallsIsFalseThenRayTracingIsNotInitial
}
TEST_F(KernelImmutableDataTests, whenHasRTCallsIsTrueThenCrossThreadDataIsPatched) {
static_cast<OsAgnosticMemoryManager *>(device->getNEODevice()->getMemoryManager())->turnOnFakingBigAllocations();
KernelDescriptor mockDescriptor = {};
mockDescriptor.kernelAttributes.flags.hasRTCalls = true;
mockDescriptor.kernelMetadata.kernelName = "rt_test";
@@ -860,10 +862,13 @@ TEST_F(KernelImmutableDataTests, whenHasRTCallsIsTrueThenCrossThreadDataIsPatche
mockDescriptor.kernelAttributes.requiredWorkgroupSize[i] = 0;
}
constexpr uint16_t rtGlobalPointerPatchOffset = 8;
std::unique_ptr<MockImmutableData> mockKernelImmutableData =
std::make_unique<MockImmutableData>(32u);
mockKernelImmutableData->kernelDescriptor = &mockDescriptor;
mockDescriptor.payloadMappings.implicitArgs.rtDispatchGlobals.pointerSize = 4;
mockDescriptor.payloadMappings.implicitArgs.rtDispatchGlobals.pointerSize = 8;
mockDescriptor.payloadMappings.implicitArgs.rtDispatchGlobals.stateless = rtGlobalPointerPatchOffset;
ModuleBuildLog *moduleBuildLog = nullptr;
module = std::make_unique<MockModule>(device,
@@ -888,16 +893,14 @@ TEST_F(KernelImmutableDataTests, whenHasRTCallsIsTrueThenCrossThreadDataIsPatche
kernel->crossThreadData.reset(reinterpret_cast<uint8_t *>(crossThreadData.get()));
kernel->crossThreadDataSize = sizeof(uint32_t[4]);
neoDevice->setRTDispatchGlobalsForceAllocation();
auto result = kernel->initialize(&kernelDesc);
EXPECT_EQ(ZE_RESULT_SUCCESS, result);
auto rtDispatchGlobals = neoDevice->getRTDispatchGlobals(NEO::RayTracingHelper::maxBvhLevels);
EXPECT_NE(nullptr, rtDispatchGlobals);
auto dispatchGlobalsAddressPatched = *reinterpret_cast<uintptr_t *>(crossThreadData.get());
auto dispatchGlobalsGpuAddressOffset = static_cast<uintptr_t>(rtDispatchGlobals->getGpuAddressToPatch());
auto dispatchGlobalsAddressPatched = *reinterpret_cast<uint64_t *>(ptrOffset(crossThreadData.get(), rtGlobalPointerPatchOffset));
auto dispatchGlobalsGpuAddressOffset = static_cast<uint64_t>(rtDispatchGlobals->getGpuAddressToPatch());
EXPECT_EQ(dispatchGlobalsGpuAddressOffset, dispatchGlobalsAddressPatched);
kernel->crossThreadData.release();