intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2025-12-26 23:33:20 +08:00
Code Issues Packages Projects Releases Wiki Activity

feature: add kernel getters for sync buffer and region barrier allocations

Browse Source 
Related-To: NEO-13350

Signed-off-by: Zbigniew Zdanowicz <zbigniew.zdanowicz@intel.com>
This commit is contained in:
Zbigniew Zdanowicz
2024-11-27 22:44:25 +00:00
committed by Compute-Runtime-Automation
parent 81ba52aac4
commit 92ebbe75c0
2 changed files with 35 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files

22
level_zero/core/source/kernel/kernel_imp.h
View File

@@ -203,6 +203,28 @@ struct KernelImp : Kernel {
bool checkKernelContainsStatefulAccess();
size_t getSyncBufferIndex() const {
return syncBufferIndex;
}
NEO::GraphicsAllocation *getSyncBufferAllocation() const {
if (std::numeric_limits<size_t>::max() == syncBufferIndex) {
return nullptr;
}
return internalResidencyContainer[syncBufferIndex];
}
size_t getRegionGroupBarrierIndex() const {
return regionGroupBarrierIndex;
}
NEO::GraphicsAllocation *getRegionGroupBarrierAllocation() const {
if (std::numeric_limits<size_t>::max() == regionGroupBarrierIndex) {
return nullptr;
}
return internalResidencyContainer[regionGroupBarrierIndex];
}
protected:
KernelImp() = default;

20
level_zero/core/test/unit_tests/sources/cmdlist/test_cmdlist_append_launch_kernel_3.cpp
View File

@@ -440,7 +440,8 @@ HWTEST2_F(CommandListAppendLaunchKernel, givenKernelUsingSyncBufferWhenAppendLau
Mock<::L0::KernelImp> kernel;
auto pMockModule = std::unique_ptr<Module>(new Mock<Module>(device, nullptr));
kernel.module = pMockModule.get();
EXPECT_EQ(std::numeric_limits<size_t>::max(), kernel.syncBufferIndex);
EXPECT_EQ(std::numeric_limits<size_t>::max(), kernel.getSyncBufferIndex());
EXPECT_EQ(nullptr, kernel.getSyncBufferAllocation());
kernel.setGroupSize(4, 1, 1);
ze_group_count_t groupCount{8, 1, 1};
@@ -467,11 +468,13 @@ HWTEST2_F(CommandListAppendLaunchKernel, givenKernelUsingSyncBufferWhenAppendLau
auto mockSyncBufferHandler = reinterpret_cast<MockSyncBufferHandler *>(device->getNEODevice()->syncBufferHandler.get());
auto syncBufferAllocation = mockSyncBufferHandler->graphicsAllocation;
EXPECT_NE(std::numeric_limits<size_t>::max(), kernel.syncBufferIndex);
EXPECT_NE(std::numeric_limits<size_t>::max(), kernel.getSyncBufferIndex());
auto syncBufferAllocationIt = std::find(kernel.internalResidencyContainer.begin(), kernel.internalResidencyContainer.end(), syncBufferAllocation);
ASSERT_NE(kernel.internalResidencyContainer.end(), syncBufferAllocationIt);
auto expectedIndex = static_cast<size_t>(std::distance(kernel.internalResidencyContainer.begin(), syncBufferAllocationIt));
EXPECT_EQ(expectedIndex, kernel.syncBufferIndex);
EXPECT_EQ(expectedIndex, kernel.getSyncBufferIndex());
EXPECT_EQ(syncBufferAllocation, kernel.getSyncBufferAllocation());
pCommandList = std::make_unique<WhiteBox<::L0::CommandListCoreFamily<gfxCoreFamily>>>();
pCommandList->initialize(device, engineGroupType, 0u);
@@ -481,7 +484,7 @@ HWTEST2_F(CommandListAppendLaunchKernel, givenKernelUsingSyncBufferWhenAppendLau
EXPECT_EQ(ZE_RESULT_SUCCESS, result);
// sync buffer index once set should not change
EXPECT_EQ(expectedIndex, kernel.syncBufferIndex);
EXPECT_EQ(expectedIndex, kernel.getSyncBufferIndex());
syncBufferAllocationIt = std::find(kernel.internalResidencyContainer.begin(), kernel.internalResidencyContainer.end(), syncBufferAllocation);
ASSERT_NE(kernel.internalResidencyContainer.end(), syncBufferAllocationIt);
// verify syncBufferAllocation is added only once
@@ -520,7 +523,8 @@ HWTEST2_F(CommandListAppendLaunchKernel, givenKernelUsingRegionGroupBarrierWhenA
Mock<::L0::KernelImp> kernel;
auto pMockModule = std::unique_ptr<Module>(new Mock<Module>(device, nullptr));
kernel.module = pMockModule.get();
EXPECT_EQ(std::numeric_limits<size_t>::max(), kernel.regionGroupBarrierIndex);
EXPECT_EQ(std::numeric_limits<size_t>::max(), kernel.getRegionGroupBarrierIndex());
EXPECT_EQ(nullptr, kernel.getRegionGroupBarrierAllocation());
kernel.crossThreadData = std::make_unique<uint8_t[]>(64);
kernel.crossThreadDataSize = 64;
@@ -556,12 +560,14 @@ HWTEST2_F(CommandListAppendLaunchKernel, givenKernelUsingRegionGroupBarrierWhenA
auto regionGroupBarrierAllocIt = std::find(kernel.internalResidencyContainer.begin(), kernel.internalResidencyContainer.end(), regionGroupBarrierAllocation);
ASSERT_NE(kernel.internalResidencyContainer.end(), regionGroupBarrierAllocIt);
auto expectedIndex = static_cast<size_t>(std::distance(kernel.internalResidencyContainer.begin(), regionGroupBarrierAllocIt));
EXPECT_EQ(expectedIndex, kernel.regionGroupBarrierIndex);
EXPECT_EQ(expectedIndex, kernel.getRegionGroupBarrierIndex());
EXPECT_EQ(regionGroupBarrierAllocation, kernel.getRegionGroupBarrierAllocation());
EXPECT_EQ(ZE_RESULT_SUCCESS, cmdList->appendLaunchKernel(kernel.toHandle(), groupCount, nullptr, 0, nullptr, launchParams, false));
// region group barrier index once set should not change
EXPECT_EQ(expectedIndex, kernel.regionGroupBarrierIndex);
EXPECT_EQ(expectedIndex, kernel.getRegionGroupBarrierIndex());
regionGroupBarrierAllocIt = std::find(kernel.internalResidencyContainer.begin(), kernel.internalResidencyContainer.end(), regionGroupBarrierAllocation);
ASSERT_NE(kernel.internalResidencyContainer.end(), regionGroupBarrierAllocIt);
// verify regionGroupBarrierAllocation is added only once