Media/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2025-09-15 13:01:45 +08:00
Code Issues Packages Projects Releases Wiki Activity

Add multi tile support for OCL post sync barrier

Browse Source 
Related-To: NEO-6262

Signed-off-by: Zbigniew Zdanowicz <zbigniew.zdanowicz@intel.com>
This commit is contained in:
Zbigniew Zdanowicz
2021-12-09 19:31:27 +00:00
committed by Compute-Runtime-Automation
parent 4b589bffd3
commit 56bef79733
7 changed files with 274 additions and 12 deletions
Download Patch File Download Diff File Expand all files Collapse all files

44
opencl/test/unit_test/command_stream/command_stream_receiver_hw_1_tests.cpp
View File

@ -1424,3 +1424,47 @@ HWCMDTEST_F(IGFX_GEN8_CORE, UltCommandStreamReceiverTest, WhenProgrammingActiveP
size_t usedAfter = commandStreamReceiver.commandStream.getUsed();
EXPECT_EQ(usedBefore, usedAfter);
}
HWCMDTEST_F(IGFX_GEN8_CORE, UltCommandStreamReceiverTest, givenBarrierNodeSetWhenProgrammingBarrierCommandThenExpectPostSyncPipeControl) {
using PIPE_CONTROL = typename FamilyType::PIPE_CONTROL;
auto &hwInfo = pDevice->getHardwareInfo();
auto commandStreamReceiver = &pDevice->getUltCommandStreamReceiver<FamilyType>();
auto &commandStreamCSR = commandStreamReceiver->getCS();
TagNodeBase *tagNode = commandStreamReceiver->getTimestampPacketAllocator()->getTag();
uint64_t gpuAddress = TimestampPacketHelper::getContextEndGpuAddress(*tagNode);
TimestampPacketDependencies timestampPacketDependencies;
timestampPacketDependencies.barrierNodes.add(tagNode);
DispatchFlags dispatchFlags = DispatchFlagsHelper::createDefaultDispatchFlags();
dispatchFlags.barrierTimestampPacketNodes = &timestampPacketDependencies.barrierNodes;
size_t expectedCmdSize = MemorySynchronizationCommands<FamilyType>::getSizeForPipeControlWithPostSyncOperation(hwInfo);
size_t estimatedCmdSize = commandStreamReceiver->getCmdSizeForStallingCommands(dispatchFlags);
EXPECT_EQ(expectedCmdSize, estimatedCmdSize);
commandStreamReceiver->programStallingCommandsForBarrier(commandStreamCSR, dispatchFlags);
EXPECT_EQ(estimatedCmdSize, commandStreamCSR.getUsed());
parseCommands<FamilyType>(commandStreamCSR, 0);
findHardwareCommands<FamilyType>();
auto cmdItor = cmdList.begin();
if (MemorySynchronizationCommands<FamilyType>::isPipeControlWArequired(hwInfo)) {
PIPE_CONTROL *pipeControl = genCmdCast<PIPE_CONTROL *>(*cmdItor);
ASSERT_NE(nullptr, pipeControl);
cmdItor++;
if (MemorySynchronizationCommands<FamilyType>::getSizeForSingleAdditionalSynchronization(hwInfo) > 0) {
cmdItor++;
}
}
PIPE_CONTROL *pipeControl = genCmdCast<PIPE_CONTROL *>(*cmdItor);
ASSERT_NE(nullptr, pipeControl);
EXPECT_EQ(PIPE_CONTROL::POST_SYNC_OPERATION::POST_SYNC_OPERATION_WRITE_IMMEDIATE_DATA, pipeControl->getPostSyncOperation());
EXPECT_TRUE(pipeControl->getCommandStreamerStallEnable());
EXPECT_EQ(0u, pipeControl->getImmediateData());
EXPECT_EQ(gpuAddress, UnitTestHelper<FamilyType>::getPipeControlPostSyncAddress(*pipeControl));
}

168
opencl/test/unit_test/command_stream/command_stream_receiver_hw_tests_xehp_and_later.cpp
View File

@ -950,3 +950,171 @@ HWTEST2_F(CommandStreamReceiverHwTestXeHPAndLater, givenStaticPartitionEnabledWh
EXPECT_EQ(estimatedCmdSize, offset);
}
HWTEST2_F(CommandStreamReceiverHwTestXeHPAndLater, givenStaticPartitionEnabledWhenSinglePartitionUsedForPostSyncBarrierThenExpectOnlyPostSyncCommands, IsAtLeastXeHpCore) {
using PIPE_CONTROL = typename FamilyType::PIPE_CONTROL;
auto &hwInfo = pDevice->getHardwareInfo();
auto commandStreamReceiver = new MockCsrHw<FamilyType>(*pDevice->executionEnvironment, pDevice->getRootDeviceIndex(), pDevice->getDeviceBitfield());
pDevice->resetCommandStreamReceiver(commandStreamReceiver);
auto &commandStreamCSR = commandStreamReceiver->getCS();
TagNodeBase *tagNode = commandStreamReceiver->getTimestampPacketAllocator()->getTag();
uint64_t gpuAddress = TimestampPacketHelper::getContextEndGpuAddress(*tagNode);
TimestampPacketDependencies timestampPacketDependencies;
timestampPacketDependencies.barrierNodes.add(tagNode);
DispatchFlags dispatchFlags = DispatchFlagsHelper::createDefaultDispatchFlags();
dispatchFlags.barrierTimestampPacketNodes = &timestampPacketDependencies.barrierNodes;
commandStreamReceiver->staticWorkPartitioningEnabled = true;
commandStreamReceiver->activePartitions = 1;
size_t expectedCmdSize = MemorySynchronizationCommands<FamilyType>::getSizeForPipeControlWithPostSyncOperation(hwInfo);
size_t estimatedCmdSize = commandStreamReceiver->getCmdSizeForStallingCommands(dispatchFlags);
EXPECT_EQ(expectedCmdSize, estimatedCmdSize);
commandStreamReceiver->programStallingCommandsForBarrier(commandStreamCSR, dispatchFlags);
EXPECT_EQ(estimatedCmdSize, commandStreamCSR.getUsed());
parseCommands<FamilyType>(commandStreamCSR, 0);
findHardwareCommands<FamilyType>();
auto cmdItor = cmdList.begin();
if (MemorySynchronizationCommands<FamilyType>::isPipeControlWArequired(hwInfo)) {
PIPE_CONTROL *pipeControl = genCmdCast<PIPE_CONTROL *>(*cmdItor);
ASSERT_NE(nullptr, pipeControl);
cmdItor++;
if (MemorySynchronizationCommands<FamilyType>::getSizeForSingleAdditionalSynchronization(hwInfo) > 0) {
cmdItor++;
}
}
PIPE_CONTROL *pipeControl = genCmdCast<PIPE_CONTROL *>(*cmdItor);
ASSERT_NE(nullptr, pipeControl);
EXPECT_EQ(PIPE_CONTROL::POST_SYNC_OPERATION::POST_SYNC_OPERATION_WRITE_IMMEDIATE_DATA, pipeControl->getPostSyncOperation());
EXPECT_TRUE(pipeControl->getCommandStreamerStallEnable());
EXPECT_EQ(0u, pipeControl->getImmediateData());
EXPECT_EQ(gpuAddress, UnitTestHelper<FamilyType>::getPipeControlPostSyncAddress(*pipeControl));
}
HWTEST2_F(CommandStreamReceiverHwTestXeHPAndLater, givenStaticPartitionDisabledWhenMultiplePartitionsUsedForPostSyncBarrierThenExpectOnlyPostSyncCommands, IsAtLeastXeHpCore) {
using PIPE_CONTROL = typename FamilyType::PIPE_CONTROL;
auto &hwInfo = pDevice->getHardwareInfo();
auto commandStreamReceiver = new MockCsrHw<FamilyType>(*pDevice->executionEnvironment, pDevice->getRootDeviceIndex(), pDevice->getDeviceBitfield());
pDevice->resetCommandStreamReceiver(commandStreamReceiver);
auto &commandStreamCSR = commandStreamReceiver->getCS();
TagNodeBase *tagNode = commandStreamReceiver->getTimestampPacketAllocator()->getTag();
uint64_t gpuAddress = TimestampPacketHelper::getContextEndGpuAddress(*tagNode);
TimestampPacketDependencies timestampPacketDependencies;
timestampPacketDependencies.barrierNodes.add(tagNode);
DispatchFlags dispatchFlags = DispatchFlagsHelper::createDefaultDispatchFlags();
dispatchFlags.barrierTimestampPacketNodes = &timestampPacketDependencies.barrierNodes;
commandStreamReceiver->staticWorkPartitioningEnabled = false;
commandStreamReceiver->activePartitions = 2;
size_t expectedCmdSize = MemorySynchronizationCommands<FamilyType>::getSizeForPipeControlWithPostSyncOperation(hwInfo);
size_t estimatedCmdSize = commandStreamReceiver->getCmdSizeForStallingCommands(dispatchFlags);
EXPECT_EQ(expectedCmdSize, estimatedCmdSize);
commandStreamReceiver->programStallingCommandsForBarrier(commandStreamCSR, dispatchFlags);
EXPECT_EQ(estimatedCmdSize, commandStreamCSR.getUsed());
parseCommands<FamilyType>(commandStreamCSR, 0);
findHardwareCommands<FamilyType>();
auto cmdItor = cmdList.begin();
if (MemorySynchronizationCommands<FamilyType>::isPipeControlWArequired(hwInfo)) {
PIPE_CONTROL *pipeControl = genCmdCast<PIPE_CONTROL *>(*cmdItor);
ASSERT_NE(nullptr, pipeControl);
cmdItor++;
if (MemorySynchronizationCommands<FamilyType>::getSizeForSingleAdditionalSynchronization(hwInfo) > 0) {
cmdItor++;
}
}
PIPE_CONTROL *pipeControl = genCmdCast<PIPE_CONTROL *>(*cmdItor);
ASSERT_NE(nullptr, pipeControl);
EXPECT_EQ(PIPE_CONTROL::POST_SYNC_OPERATION::POST_SYNC_OPERATION_WRITE_IMMEDIATE_DATA, pipeControl->getPostSyncOperation());
EXPECT_TRUE(pipeControl->getCommandStreamerStallEnable());
EXPECT_EQ(0u, pipeControl->getImmediateData());
EXPECT_EQ(gpuAddress, UnitTestHelper<FamilyType>::getPipeControlPostSyncAddress(*pipeControl));
}
HWTEST2_F(CommandStreamReceiverHwTestXeHPAndLater, givenStaticPartitionEnabledWhenMultiplePartitionsUsedThenExpectImplicitScalingPostSyncBarrierWithoutSelfCleanup, IsAtLeastXeHpCore) {
using PIPE_CONTROL = typename FamilyType::PIPE_CONTROL;
using MI_BATCH_BUFFER_START = typename FamilyType::MI_BATCH_BUFFER_START;
using MI_ATOMIC = typename FamilyType::MI_ATOMIC;
using MI_SEMAPHORE_WAIT = typename FamilyType::MI_SEMAPHORE_WAIT;
auto &hwInfo = pDevice->getHardwareInfo();
auto commandStreamReceiver = new MockCsrHw<FamilyType>(*pDevice->executionEnvironment, pDevice->getRootDeviceIndex(), pDevice->getDeviceBitfield());
pDevice->resetCommandStreamReceiver(commandStreamReceiver);
auto &commandStreamCSR = commandStreamReceiver->getCS();
TagNodeBase *tagNode = commandStreamReceiver->getTimestampPacketAllocator()->getTag();
uint64_t gpuAddress = TimestampPacketHelper::getContextEndGpuAddress(*tagNode);
TimestampPacketDependencies timestampPacketDependencies;
timestampPacketDependencies.barrierNodes.add(tagNode);
DispatchFlags dispatchFlags = DispatchFlagsHelper::createDefaultDispatchFlags();
dispatchFlags.barrierTimestampPacketNodes = &timestampPacketDependencies.barrierNodes;
commandStreamReceiver->staticWorkPartitioningEnabled = true;
commandStreamReceiver->activePartitions = 2;
size_t expectedSize = MemorySynchronizationCommands<FamilyType>::getSizeForPipeControlWithPostSyncOperation(hwInfo) +
sizeof(MI_ATOMIC) + sizeof(MI_SEMAPHORE_WAIT) +
sizeof(MI_BATCH_BUFFER_START) +
2 * sizeof(uint32_t);
size_t estimatedCmdSize = commandStreamReceiver->getCmdSizeForStallingCommands(dispatchFlags);
EXPECT_EQ(expectedSize, estimatedCmdSize);
commandStreamReceiver->programStallingCommandsForBarrier(commandStreamCSR, dispatchFlags);
EXPECT_EQ(estimatedCmdSize, commandStreamCSR.getUsed());
EXPECT_EQ(2u, tagNode->getPacketsUsed());
parseCommands<FamilyType>(commandStreamCSR, 0);
findHardwareCommands<FamilyType>();
auto cmdItor = cmdList.begin();
if (MemorySynchronizationCommands<FamilyType>::isPipeControlWArequired(hwInfo)) {
PIPE_CONTROL *pipeControl = genCmdCast<PIPE_CONTROL *>(*cmdItor);
ASSERT_NE(nullptr, pipeControl);
cmdItor++;
if (MemorySynchronizationCommands<FamilyType>::getSizeForSingleAdditionalSynchronization(hwInfo) > 0) {
cmdItor++;
}
}
PIPE_CONTROL *pipeControl = genCmdCast<PIPE_CONTROL *>(*cmdItor);
ASSERT_NE(nullptr, pipeControl);
EXPECT_EQ(PIPE_CONTROL::POST_SYNC_OPERATION::POST_SYNC_OPERATION_WRITE_IMMEDIATE_DATA, pipeControl->getPostSyncOperation());
EXPECT_TRUE(pipeControl->getCommandStreamerStallEnable());
EXPECT_EQ(0u, pipeControl->getImmediateData());
EXPECT_EQ(gpuAddress, UnitTestHelper<FamilyType>::getPipeControlPostSyncAddress(*pipeControl));
EXPECT_TRUE(pipeControl->getWorkloadPartitionIdOffsetEnable());
cmdItor++;
if (MemorySynchronizationCommands<FamilyType>::getSizeForSingleAdditionalSynchronization(hwInfo) > 0) {
cmdItor++;
}
MI_ATOMIC *miAtomic = genCmdCast<MI_ATOMIC *>(*cmdItor);
ASSERT_NE(nullptr, miAtomic);
cmdItor++;
MI_SEMAPHORE_WAIT *miSemaphore = genCmdCast<MI_SEMAPHORE_WAIT *>(*cmdItor);
ASSERT_NE(nullptr, miSemaphore);
cmdItor++;
MI_BATCH_BUFFER_START *bbStart = genCmdCast<MI_BATCH_BUFFER_START *>(*cmdItor);
ASSERT_NE(nullptr, bbStart);
}