feature: Optimize intra-module kernel ISA allocations

So far, there is a separate page allocated for each kernel's ISA within
`KernelImmutableData::initialize()`. Apparently the ISA blocks are often
much smaller than a 64k page, which leads to poor memory utilization and
was even observed to cause the device OOM error if a single module has
several keys.

Improve the situation by reusing the parent allocation (owned by the
module instance) for modules, which kernel ISAs can fit together within
a single 64k page. This improves the memory utilization on a single
module level.

Related-To: NEO-7788
Signed-off-by: Maciej Bielski <maciej.bielski@intel.com>
This commit is contained in:
Maciej Bielski
2023-08-09 13:39:39 +00:00
committed by Compute-Runtime-Automation
parent 1b7e178b25
commit c348831470
22 changed files with 649 additions and 177 deletions

View File

@@ -997,6 +997,45 @@ HWTEST2_F(EncodeDispatchKernelTest, givenBindlessKernelWhenDispatchingKernelThen
EXPECT_NE(usedAfter, usedBefore);
}
HWTEST2_F(EncodeDispatchKernelTest, givenKernelsSharingISAParentAllocationsWhenProgrammingWalkerThenKernelStartPointerHasProperOffset, IsBeforeXeHpCore) {
using INTERFACE_DESCRIPTOR_DATA = typename FamilyType::INTERFACE_DESCRIPTOR_DATA;
auto dispatchInterface = std::make_unique<MockDispatchKernelEncoder>();
dispatchInterface->getIsaOffsetInParentAllocationResult = 8 << INTERFACE_DESCRIPTOR_DATA::KERNELSTARTPOINTER_BIT_SHIFT;
uint32_t dims[] = {2, 1, 1};
bool requiresUncachedMocs = false;
EncodeDispatchKernelArgs dispatchArgs = createDefaultDispatchKernelArgs(pDevice, dispatchInterface.get(), dims, requiresUncachedMocs);
EncodeDispatchKernel<FamilyType>::encode(*cmdContainer.get(), dispatchArgs);
auto idd = static_cast<INTERFACE_DESCRIPTOR_DATA *>(cmdContainer->getIddBlock());
EXPECT_EQ(idd->getKernelStartPointer(), dispatchInterface->getIsaAllocation()->getGpuAddressToPatch() + dispatchInterface->getIsaOffsetInParentAllocation());
}
HWTEST_F(EncodeDispatchKernelTest, givenKernelStartPointerAlignmentInInterfaceDescriptorWhenHelperGetterUsedThenCorrectValueReturned) {
using INTERFACE_DESCRIPTOR_DATA = typename FamilyType::INTERFACE_DESCRIPTOR_DATA;
EXPECT_EQ(INTERFACE_DESCRIPTOR_DATA::KERNELSTARTPOINTER_ALIGN_SIZE, pDevice->getGfxCoreHelper().getKernelIsaPointerAlignment());
}
HWTEST2_F(EncodeDispatchKernelTest, givenKernelsSharingISAParentAllocationsWhenProgrammingWalkerThenKernelStartPointerHasProperOffset, IsAtLeastXeHpCore) {
using WALKER_TYPE = typename FamilyType::WALKER_TYPE;
using INTERFACE_DESCRIPTOR_DATA = typename FamilyType::INTERFACE_DESCRIPTOR_DATA;
auto dispatchInterface = std::make_unique<MockDispatchKernelEncoder>();
dispatchInterface->getIsaOffsetInParentAllocationResult = 8 << INTERFACE_DESCRIPTOR_DATA::KERNELSTARTPOINTER_BIT_SHIFT;
uint32_t dims[] = {2, 1, 1};
bool requiresUncachedMocs = false;
EncodeDispatchKernelArgs dispatchArgs = createDefaultDispatchKernelArgs(pDevice, dispatchInterface.get(), dims, requiresUncachedMocs);
EncodeDispatchKernel<FamilyType>::encode(*cmdContainer.get(), dispatchArgs);
GenCmdList commands;
CmdParse<FamilyType>::parseCommandBuffer(commands, ptrOffset(cmdContainer->getCommandStream()->getCpuBase(), 0), cmdContainer->getCommandStream()->getUsed());
auto itor = find<WALKER_TYPE *>(commands.begin(), commands.end());
ASSERT_NE(itor, commands.end());
auto walkerCmd = genCmdCast<WALKER_TYPE *>(*itor);
EXPECT_EQ(walkerCmd->getInterfaceDescriptor().getKernelStartPointer(), dispatchInterface->getIsaAllocation()->getGpuAddressToPatch() + dispatchInterface->getIsaOffsetInParentAllocation());
}
HWTEST_F(EncodeDispatchKernelTest, givenNonBindlessOrStatelessArgWhenDispatchingKernelThenSurfaceStateOffsetInCrossThreadDataIsNotPatched) {
using BINDING_TABLE_STATE = typename FamilyType::BINDING_TABLE_STATE;
using DataPortBindlessSurfaceExtendedMessageDescriptor = typename FamilyType::DataPortBindlessSurfaceExtendedMessageDescriptor;