intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2026-01-05 09:09:04 +08:00
Code Issues Packages Projects Releases Wiki Activity

Move event timestamp function to cmdlist_hw.inl

Browse Source 
Change-Id: I0fdff3a70172a81a32e416d62b00bc57420640f3
Signed-off-by: Aravind Gopalakrishnan <Aravind.Gopalakrishnan@intel.com>
This commit is contained in:
Aravind Gopalakrishnan
2020-06-23 17:27:35 -07:00
committed by sys_ocldev
parent c22a117e07
commit d635eb6300
3 changed files with 100 additions and 43 deletions
Download Patch File Download Diff File Expand all files Collapse all files

47
level_zero/core/source/cmdlist/cmdlist_hw.inl
View File

@@ -1306,6 +1306,53 @@ ze_result_t CommandListCoreFamily<gfxCoreFamily>::appendWaitOnEvents(uint32_t nu
return ZE_RESULT_SUCCESS;
}
template <GFXCORE_FAMILY gfxCoreFamily>
void CommandListCoreFamily<gfxCoreFamily>::appendEventForProfiling(ze_event_handle_t hEvent, bool beforeWalker) {
if (!hEvent) {
return;
}
if (isCopyOnly()) {
appendEventForProfilingCopyCommand(hEvent, beforeWalker);
} else {
auto event = Event::fromHandle(hEvent);
if (!event->isTimestampEvent) {
return;
}
commandContainer.addToResidencyContainer(&event->getAllocation());
auto baseAddr = event->getGpuAddress();
if (beforeWalker) {
auto contextStartAddr = baseAddr;
auto globalStartAddr = baseAddr + offsetof(KernelTimestampEvent, globalStart);
NEO::EncodeStoreMMIO<GfxFamily>::encode(*commandContainer.getCommandStream(),
REG_GLOBAL_TIMESTAMP_LDW, globalStartAddr);
NEO::EncodeStoreMMIO<GfxFamily>::encode(*commandContainer.getCommandStream(),
GP_THREAD_TIME_REG_ADDRESS_OFFSET_LOW, contextStartAddr);
} else {
auto contextEndAddr = baseAddr + offsetof(KernelTimestampEvent, contextEnd);
auto globalEndAddr = baseAddr + offsetof(KernelTimestampEvent, globalEnd);
NEO::PipeControlArgs args;
args.dcFlushEnable = false;
NEO::MemorySynchronizationCommands<GfxFamily>::addPipeControl(*commandContainer.getCommandStream(), args);
NEO::EncodeStoreMMIO<GfxFamily>::encode(*commandContainer.getCommandStream(),
REG_GLOBAL_TIMESTAMP_LDW, globalEndAddr);
NEO::EncodeStoreMMIO<GfxFamily>::encode(*commandContainer.getCommandStream(),
GP_THREAD_TIME_REG_ADDRESS_OFFSET_LOW, contextEndAddr);
args.dcFlushEnable = (event->signalScope == ZE_EVENT_SCOPE_FLAG_NONE) ? false : true;
if (args.dcFlushEnable) {
NEO::MemorySynchronizationCommands<GfxFamily>::addPipeControl(*commandContainer.getCommandStream(), args);
}
}
}
}
template <GFXCORE_FAMILY gfxCoreFamily>
ze_result_t CommandListCoreFamily<gfxCoreFamily>::reserveSpace(size_t size, void **ptr) {
auto availableSpace = commandContainer.getCommandStream()->getAvailableSpace();

43
level_zero/core/source/cmdlist/cmdlist_hw_base.inl
View File

@@ -75,49 +75,6 @@ ze_result_t CommandListCoreFamily<gfxCoreFamily>::appendLaunchKernelWithParams(z
return ZE_RESULT_SUCCESS;
}
template <GFXCORE_FAMILY gfxCoreFamily>
void CommandListCoreFamily<gfxCoreFamily>::appendEventForProfiling(ze_event_handle_t hEvent, bool beforeWalker) {
if (!hEvent) {
return;
}
if (isCopyOnly()) {
appendEventForProfilingCopyCommand(hEvent, beforeWalker);
} else {
auto event = Event::fromHandle(hEvent);
if (!event->isTimestampEvent) {
return;
}
commandContainer.addToResidencyContainer(&event->getAllocation());
auto baseAddr = event->getGpuAddress();
if (beforeWalker) {
auto contextStartAddr = baseAddr;
auto globalStartAddr = baseAddr + offsetof(KernelTimestampEvent, globalStart);
NEO::EncodeStoreMMIO<GfxFamily>::encode(*commandContainer.getCommandStream(), REG_GLOBAL_TIMESTAMP_LDW, globalStartAddr);
NEO::EncodeStoreMMIO<GfxFamily>::encode(*commandContainer.getCommandStream(), GP_THREAD_TIME_REG_ADDRESS_OFFSET_LOW, contextStartAddr);
} else {
auto contextEndAddr = baseAddr + offsetof(KernelTimestampEvent, contextEnd);
auto globalEndAddr = baseAddr + offsetof(KernelTimestampEvent, globalEnd);
NEO::PipeControlArgs args;
args.dcFlushEnable = false;
NEO::MemorySynchronizationCommands<GfxFamily>::addPipeControl(*commandContainer.getCommandStream(), args);
NEO::EncodeStoreMMIO<GfxFamily>::encode(*commandContainer.getCommandStream(), REG_GLOBAL_TIMESTAMP_LDW, globalEndAddr);
NEO::EncodeStoreMMIO<GfxFamily>::encode(*commandContainer.getCommandStream(), GP_THREAD_TIME_REG_ADDRESS_OFFSET_LOW, contextEndAddr);
args.dcFlushEnable = (event->signalScope == ZE_EVENT_SCOPE_FLAG_NONE) ? false : true;
if (args.dcFlushEnable) {
NEO::MemorySynchronizationCommands<GfxFamily>::addPipeControl(*commandContainer.getCommandStream(), args);
}
}
}
}
template <GFXCORE_FAMILY gfxCoreFamily>
bool CommandListCoreFamily<gfxCoreFamily>::useMemCopyToBlitFill(size_t patternSize) {
return patternSize > sizeof(uint32_t);

53
level_zero/core/test/unit_tests/sources/cmdlist/test_cmdlist.cpp
View File

@@ -584,6 +584,7 @@ HWTEST_F(CommandListCreate, givenCommandListyWhenAppendSignalEventThePipeControl
EXPECT_NE(cmdList.end(), itor);
}
HWTEST_F(CommandListCreate, givenCommandListWithCopyOnlyWhenAppendWaitEventsWithDcFlushThenMiFlushDWIsProgrammed) {
using MI_FLUSH_DW = typename FamilyType::MI_FLUSH_DW;
std::unique_ptr<L0::CommandList> commandList(CommandList::create(productFamily, device, true));
@@ -618,6 +619,58 @@ HWTEST_F(CommandListCreate, givenCommandListyWhenAppendWaitEventsWithDcFlushTheP
EXPECT_NE(cmdList.end(), itor);
}
HWTEST2_F(CommandListCreate, givenCommandListWhenTimestampPassedToMemoryCopyThenAppendProfilingCalledOnceBeforeAndAfterCommand, Platforms) {
using GfxFamily = typename NEO::GfxFamilyMapper<gfxCoreFamily>::GfxFamily;
using MI_STORE_REGISTER_MEM = typename GfxFamily::MI_STORE_REGISTER_MEM;
using PIPE_CONTROL = typename FamilyType::PIPE_CONTROL;
MockAppendMemoryCopy<gfxCoreFamily> commandList;
commandList.initialize(device, false);
void *srcPtr = reinterpret_cast<void *>(0x1234);
void *dstPtr = reinterpret_cast<void *>(0x2345);
ze_event_pool_desc_t eventPoolDesc = {
ZE_EVENT_POOL_DESC_VERSION_CURRENT,
ZE_EVENT_POOL_FLAG_TIMESTAMP,
1};
ze_event_desc_t eventDesc = {
ZE_EVENT_DESC_VERSION_CURRENT,
0,
ZE_EVENT_SCOPE_FLAG_NONE,
ZE_EVENT_SCOPE_FLAG_NONE};
auto eventPool = std::unique_ptr<L0::EventPool>(L0::EventPool::create(driverHandle.get(), 0, nullptr, &eventPoolDesc));
auto event = std::unique_ptr<L0::Event>(L0::Event::create(eventPool.get(), &eventDesc, device));
commandList.appendMemoryCopy(dstPtr, srcPtr, 0x100, event->toHandle(), 0, nullptr);
EXPECT_GT(commandList.appendMemoryCopyKernelWithGACalledTimes, 0u);
EXPECT_EQ(commandList.appendMemoryCopyBlitCalledTimes, 0u);
GenCmdList cmdList;
ASSERT_TRUE(FamilyType::PARSE::parseCommandBuffer(
cmdList, ptrOffset(commandList.commandContainer.getCommandStream()->getCpuBase(), 0),
commandList.commandContainer.getCommandStream()->getUsed()));
auto itor = find<MI_STORE_REGISTER_MEM *>(cmdList.begin(), cmdList.end());
EXPECT_NE(cmdList.end(), itor);
auto cmd = genCmdCast<MI_STORE_REGISTER_MEM *>(*itor);
EXPECT_EQ(cmd->getRegisterAddress(), REG_GLOBAL_TIMESTAMP_LDW);
itor++;
EXPECT_NE(cmdList.end(), itor);
cmd = genCmdCast<MI_STORE_REGISTER_MEM *>(*itor);
EXPECT_EQ(cmd->getRegisterAddress(), GP_THREAD_TIME_REG_ADDRESS_OFFSET_LOW);
itor = find<PIPE_CONTROL *>(itor, cmdList.end());
EXPECT_NE(cmdList.end(), itor);
itor = find<MI_STORE_REGISTER_MEM *>(itor, cmdList.end());
EXPECT_NE(cmdList.end(), itor);
cmd = genCmdCast<MI_STORE_REGISTER_MEM *>(*itor);
EXPECT_EQ(cmd->getRegisterAddress(), REG_GLOBAL_TIMESTAMP_LDW);
itor++;
EXPECT_NE(cmdList.end(), itor);
cmd = genCmdCast<MI_STORE_REGISTER_MEM *>(*itor);
EXPECT_EQ(cmd->getRegisterAddress(), GP_THREAD_TIME_REG_ADDRESS_OFFSET_LOW);
}
template <GFXCORE_FAMILY gfxCoreFamily>
class MockCommandListForMemFill : public WhiteBox<::L0::CommandListCoreFamily<gfxCoreFamily>> {
public: