intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2026-01-14 02:32:59 +08:00
Code Issues Packages Projects Releases Wiki Activity

Add cross regular and intermediate command lists base address state transitions

Browse Source 
- updates coming from regular list are updated in csr last sent variables
- all per context and per kernel transitions kept in single place
- state updates from intermediate to regular are set in csr properties
- global atomics support duplicates removed

Related-To: NEO-5055

Signed-off-by: Zbigniew Zdanowicz <zbigniew.zdanowicz@intel.com>
This commit is contained in:
Zbigniew Zdanowicz
2023-02-02 18:57:24 +00:00
committed by Compute-Runtime-Automation
parent 140e59810f
commit bf2072c3ea
16 changed files with 363 additions and 337 deletions
Download Patch File Download Diff File Expand all files Collapse all files

209
opencl/test/unit_test/command_stream/command_stream_receiver_hw_1_tests.cpp
View File

@@ -1552,212 +1552,3 @@ HWCMDTEST_F(IGFX_GEN8_CORE, UltCommandStreamReceiverTest, givenBarrierNodeSetWhe
EXPECT_EQ(0u, pipeControl->getImmediateData());
EXPECT_EQ(gpuAddress, UnitTestHelper<FamilyType>::getPipeControlPostSyncAddress(*pipeControl));
}
HWTEST_F(UltCommandStreamReceiverTest, givenFrontEndStateNotInitedWhenTransitionFrontEndPropertiesThenExpectCorrectValuesStored) {
auto dispatchFlags = DispatchFlagsHelper::createDefaultDispatchFlags();
auto &commandStreamReceiver = pDevice->getUltCommandStreamReceiver<FamilyType>();
commandStreamReceiver.feSupportFlags.computeDispatchAllWalker = false;
commandStreamReceiver.feSupportFlags.disableEuFusion = false;
commandStreamReceiver.setMediaVFEStateDirty(false);
commandStreamReceiver.feSupportFlags.disableOverdispatch = true;
dispatchFlags.additionalKernelExecInfo = AdditionalKernelExecInfo::NotApplicable;
commandStreamReceiver.handleFrontEndStateTransition(dispatchFlags);
EXPECT_FALSE(commandStreamReceiver.getMediaVFEStateDirty());
dispatchFlags.additionalKernelExecInfo = AdditionalKernelExecInfo::NotSet;
commandStreamReceiver.handleFrontEndStateTransition(dispatchFlags);
EXPECT_FALSE(commandStreamReceiver.getMediaVFEStateDirty());
dispatchFlags.additionalKernelExecInfo = AdditionalKernelExecInfo::DisableOverdispatch;
commandStreamReceiver.handleFrontEndStateTransition(dispatchFlags);
EXPECT_TRUE(commandStreamReceiver.getMediaVFEStateDirty());
commandStreamReceiver.setMediaVFEStateDirty(false);
commandStreamReceiver.feSupportFlags.disableOverdispatch = false;
commandStreamReceiver.lastAdditionalKernelExecInfo = AdditionalKernelExecInfo::NotSet;
commandStreamReceiver.handleFrontEndStateTransition(dispatchFlags);
EXPECT_FALSE(commandStreamReceiver.getMediaVFEStateDirty());
commandStreamReceiver.feSupportFlags.computeDispatchAllWalker = true;
dispatchFlags.kernelExecutionType = KernelExecutionType::NotApplicable;
commandStreamReceiver.handleFrontEndStateTransition(dispatchFlags);
EXPECT_FALSE(commandStreamReceiver.getMediaVFEStateDirty());
dispatchFlags.kernelExecutionType = KernelExecutionType::Default;
commandStreamReceiver.handleFrontEndStateTransition(dispatchFlags);
EXPECT_FALSE(commandStreamReceiver.getMediaVFEStateDirty());
dispatchFlags.kernelExecutionType = KernelExecutionType::Concurrent;
commandStreamReceiver.handleFrontEndStateTransition(dispatchFlags);
EXPECT_TRUE(commandStreamReceiver.getMediaVFEStateDirty());
commandStreamReceiver.setMediaVFEStateDirty(false);
commandStreamReceiver.feSupportFlags.computeDispatchAllWalker = false;
commandStreamReceiver.lastKernelExecutionType = KernelExecutionType::Default;
commandStreamReceiver.handleFrontEndStateTransition(dispatchFlags);
EXPECT_FALSE(commandStreamReceiver.getMediaVFEStateDirty());
commandStreamReceiver.feSupportFlags.disableEuFusion = true;
dispatchFlags.disableEUFusion = false;
commandStreamReceiver.handleFrontEndStateTransition(dispatchFlags);
EXPECT_TRUE(commandStreamReceiver.getMediaVFEStateDirty());
commandStreamReceiver.setMediaVFEStateDirty(false);
commandStreamReceiver.streamProperties.frontEndState.disableEUFusion.value = 0;
dispatchFlags.disableEUFusion = true;
commandStreamReceiver.handleFrontEndStateTransition(dispatchFlags);
EXPECT_TRUE(commandStreamReceiver.getMediaVFEStateDirty());
commandStreamReceiver.setMediaVFEStateDirty(false);
dispatchFlags.disableEUFusion = false;
commandStreamReceiver.handleFrontEndStateTransition(dispatchFlags);
EXPECT_FALSE(commandStreamReceiver.getMediaVFEStateDirty());
commandStreamReceiver.feSupportFlags.disableEuFusion = false;
commandStreamReceiver.streamProperties.frontEndState.disableEUFusion.value = -1;
dispatchFlags.disableEUFusion = false;
commandStreamReceiver.handleFrontEndStateTransition(dispatchFlags);
EXPECT_FALSE(commandStreamReceiver.getMediaVFEStateDirty());
}
HWTEST_F(UltCommandStreamReceiverTest, givenFrontEndStateInitedWhenTransitionFrontEndPropertiesThenExpectCorrectValuesStored) {
auto dispatchFlags = DispatchFlagsHelper::createDefaultDispatchFlags();
auto &commandStreamReceiver = pDevice->getUltCommandStreamReceiver<FamilyType>();
commandStreamReceiver.feSupportFlags.computeDispatchAllWalker = false;
commandStreamReceiver.feSupportFlags.disableEuFusion = false;
commandStreamReceiver.setMediaVFEStateDirty(false);
commandStreamReceiver.feSupportFlags.disableOverdispatch = true;
commandStreamReceiver.streamProperties.frontEndState.disableOverdispatch.value = 0;
dispatchFlags.additionalKernelExecInfo = AdditionalKernelExecInfo::NotSet;
commandStreamReceiver.handleFrontEndStateTransition(dispatchFlags);
EXPECT_FALSE(commandStreamReceiver.getMediaVFEStateDirty());
dispatchFlags.additionalKernelExecInfo = AdditionalKernelExecInfo::DisableOverdispatch;
commandStreamReceiver.handleFrontEndStateTransition(dispatchFlags);
EXPECT_TRUE(commandStreamReceiver.getMediaVFEStateDirty());
commandStreamReceiver.setMediaVFEStateDirty(false);
commandStreamReceiver.streamProperties.frontEndState.disableOverdispatch.value = 1;
dispatchFlags.additionalKernelExecInfo = AdditionalKernelExecInfo::NotSet;
commandStreamReceiver.handleFrontEndStateTransition(dispatchFlags);
EXPECT_TRUE(commandStreamReceiver.getMediaVFEStateDirty());
commandStreamReceiver.setMediaVFEStateDirty(false);
commandStreamReceiver.feSupportFlags.disableOverdispatch = false;
commandStreamReceiver.feSupportFlags.computeDispatchAllWalker = true;
commandStreamReceiver.streamProperties.frontEndState.computeDispatchAllWalkerEnable.value = 0;
dispatchFlags.kernelExecutionType = KernelExecutionType::Default;
commandStreamReceiver.handleFrontEndStateTransition(dispatchFlags);
EXPECT_FALSE(commandStreamReceiver.getMediaVFEStateDirty());
dispatchFlags.kernelExecutionType = KernelExecutionType::Concurrent;
commandStreamReceiver.handleFrontEndStateTransition(dispatchFlags);
EXPECT_TRUE(commandStreamReceiver.getMediaVFEStateDirty());
commandStreamReceiver.setMediaVFEStateDirty(false);
commandStreamReceiver.streamProperties.frontEndState.computeDispatchAllWalkerEnable.value = 1;
dispatchFlags.kernelExecutionType = KernelExecutionType::Default;
commandStreamReceiver.handleFrontEndStateTransition(dispatchFlags);
EXPECT_TRUE(commandStreamReceiver.getMediaVFEStateDirty());
commandStreamReceiver.setMediaVFEStateDirty(false);
}
HWTEST_F(UltCommandStreamReceiverTest, givenPipelineSelectStateNotInitedWhenTransitionPipelineSelectPropertiesThenExpectCorrectValuesStored) {
auto dispatchFlags = DispatchFlagsHelper::createDefaultDispatchFlags();
auto &commandStreamReceiver = pDevice->getUltCommandStreamReceiver<FamilyType>();
commandStreamReceiver.pipelineSupportFlags.systolicMode = false;
commandStreamReceiver.pipelineSupportFlags.mediaSamplerDopClockGate = true;
dispatchFlags.pipelineSelectArgs.mediaSamplerRequired = false;
commandStreamReceiver.handlePipelineSelectStateTransition(dispatchFlags);
EXPECT_TRUE(commandStreamReceiver.csrSizeRequestFlags.mediaSamplerConfigChanged);
commandStreamReceiver.pipelineSupportFlags.mediaSamplerDopClockGate = false;
commandStreamReceiver.lastMediaSamplerConfig = -1;
commandStreamReceiver.handlePipelineSelectStateTransition(dispatchFlags);
EXPECT_FALSE(commandStreamReceiver.csrSizeRequestFlags.mediaSamplerConfigChanged);
commandStreamReceiver.pipelineSupportFlags.mediaSamplerDopClockGate = true;
commandStreamReceiver.lastMediaSamplerConfig = 0;
commandStreamReceiver.handlePipelineSelectStateTransition(dispatchFlags);
EXPECT_FALSE(commandStreamReceiver.csrSizeRequestFlags.mediaSamplerConfigChanged);
dispatchFlags.pipelineSelectArgs.mediaSamplerRequired = true;
commandStreamReceiver.handlePipelineSelectStateTransition(dispatchFlags);
EXPECT_TRUE(commandStreamReceiver.csrSizeRequestFlags.mediaSamplerConfigChanged);
commandStreamReceiver.pipelineSupportFlags.mediaSamplerDopClockGate = false;
commandStreamReceiver.pipelineSupportFlags.systolicMode = true;
commandStreamReceiver.lastSystolicPipelineSelectMode = false;
dispatchFlags.pipelineSelectArgs.systolicPipelineSelectMode = true;
commandStreamReceiver.handlePipelineSelectStateTransition(dispatchFlags);
EXPECT_TRUE(commandStreamReceiver.csrSizeRequestFlags.systolicPipelineSelectMode);
commandStreamReceiver.pipelineSupportFlags.systolicMode = false;
commandStreamReceiver.lastSystolicPipelineSelectMode = false;
dispatchFlags.pipelineSelectArgs.systolicPipelineSelectMode = true;
commandStreamReceiver.handlePipelineSelectStateTransition(dispatchFlags);
EXPECT_FALSE(commandStreamReceiver.csrSizeRequestFlags.systolicPipelineSelectMode);
commandStreamReceiver.pipelineSupportFlags.systolicMode = true;
commandStreamReceiver.lastSystolicPipelineSelectMode = false;
dispatchFlags.pipelineSelectArgs.systolicPipelineSelectMode = false;
commandStreamReceiver.handlePipelineSelectStateTransition(dispatchFlags);
EXPECT_FALSE(commandStreamReceiver.csrSizeRequestFlags.systolicPipelineSelectMode);
}
HWTEST_F(UltCommandStreamReceiverTest,
givenPipelineSelectStateInitedWhenTransitionPipelineSelectPropertiesThenExpectCorrectValuesStored) {
auto dispatchFlags = DispatchFlagsHelper::createDefaultDispatchFlags();
auto &commandStreamReceiver = pDevice->getUltCommandStreamReceiver<FamilyType>();
commandStreamReceiver.pipelineSupportFlags.systolicMode = false;
commandStreamReceiver.pipelineSupportFlags.mediaSamplerDopClockGate = true;
commandStreamReceiver.streamProperties.pipelineSelect.mediaSamplerDopClockGate.value = 1;
commandStreamReceiver.lastMediaSamplerConfig = -1;
dispatchFlags.pipelineSelectArgs.mediaSamplerRequired = false;
commandStreamReceiver.handlePipelineSelectStateTransition(dispatchFlags);
EXPECT_TRUE(commandStreamReceiver.csrSizeRequestFlags.mediaSamplerConfigChanged);
commandStreamReceiver.streamProperties.pipelineSelect.mediaSamplerDopClockGate.value = 0;
dispatchFlags.pipelineSelectArgs.mediaSamplerRequired = true;
commandStreamReceiver.handlePipelineSelectStateTransition(dispatchFlags);
EXPECT_TRUE(commandStreamReceiver.csrSizeRequestFlags.mediaSamplerConfigChanged);
commandStreamReceiver.streamProperties.pipelineSelect.mediaSamplerDopClockGate.value = 0;
commandStreamReceiver.lastMediaSamplerConfig = 1;
dispatchFlags.pipelineSelectArgs.mediaSamplerRequired = false;
commandStreamReceiver.handlePipelineSelectStateTransition(dispatchFlags);
EXPECT_FALSE(commandStreamReceiver.csrSizeRequestFlags.mediaSamplerConfigChanged);
commandStreamReceiver.pipelineSupportFlags.mediaSamplerDopClockGate = false;
commandStreamReceiver.pipelineSupportFlags.systolicMode = true;
commandStreamReceiver.streamProperties.pipelineSelect.systolicMode.value = 1;
commandStreamReceiver.lastSystolicPipelineSelectMode = false;
dispatchFlags.pipelineSelectArgs.systolicPipelineSelectMode = false;
commandStreamReceiver.handlePipelineSelectStateTransition(dispatchFlags);
EXPECT_TRUE(commandStreamReceiver.csrSizeRequestFlags.systolicPipelineSelectMode);
commandStreamReceiver.streamProperties.pipelineSelect.systolicMode.value = 0;
dispatchFlags.pipelineSelectArgs.systolicPipelineSelectMode = true;
commandStreamReceiver.handlePipelineSelectStateTransition(dispatchFlags);
EXPECT_TRUE(commandStreamReceiver.csrSizeRequestFlags.systolicPipelineSelectMode);
commandStreamReceiver.streamProperties.pipelineSelect.systolicMode.value = 0;
commandStreamReceiver.lastSystolicPipelineSelectMode = true;
dispatchFlags.pipelineSelectArgs.systolicPipelineSelectMode = false;
commandStreamReceiver.handlePipelineSelectStateTransition(dispatchFlags);
EXPECT_FALSE(commandStreamReceiver.csrSizeRequestFlags.systolicPipelineSelectMode);
}

40
opencl/test/unit_test/helpers/test_preamble_xehp_and_later.cpp
View File

@@ -374,46 +374,6 @@ HWCMDTEST_F(IGFX_XE_HP_CORE, XeHpCommandStreamReceiverFlushTaskTests, givenMultE
EXPECT_TRUE(stateBaseAddress->getDisableSupportForMultiGpuPartialWritesForStatelessMessages());
}
struct MultiGpuGlobalAtomicsTest : public XeHpCommandStreamReceiverFlushTaskTests,
public ::testing::WithParamInterface<std::tuple<bool, bool, bool, bool>> {
};
HWCMDTEST_P(IGFX_XE_HP_CORE, MultiGpuGlobalAtomicsTest, givenFlushingCommandStreamReceiverThenDisableSupportForMultiGpuAtomicsForStatelessAccessesIsSetCorrectly) {
bool isMultiOsContextCapable, useGlobalAtomics, areMultipleSubDevicesInContext, enableMultiGpuAtomicsOptimization;
std::tie(isMultiOsContextCapable, useGlobalAtomics, areMultipleSubDevicesInContext, enableMultiGpuAtomicsOptimization) = GetParam();
DebugManagerStateRestore stateRestore;
DebugManager.flags.EnableMultiGpuAtomicsOptimization.set(enableMultiGpuAtomicsOptimization);
using STATE_BASE_ADDRESS = typename FamilyType::STATE_BASE_ADDRESS;
auto &commandStreamReceiver = pDevice->getUltCommandStreamReceiver<FamilyType>();
commandStreamReceiver.multiOsContextCapable = isMultiOsContextCapable;
flushTaskFlags.useGlobalAtomics = useGlobalAtomics;
flushTaskFlags.areMultipleSubDevicesInContext = areMultipleSubDevicesInContext;
flushTask(commandStreamReceiver, false, 0, false, false);
HardwareParse hwParserCsr;
hwParserCsr.parseCommands<FamilyType>(commandStreamReceiver.commandStream, 0);
hwParserCsr.findHardwareCommands<FamilyType>();
ASSERT_NE(nullptr, hwParserCsr.cmdStateBaseAddress);
auto stateBaseAddress = static_cast<STATE_BASE_ADDRESS *>(hwParserCsr.cmdStateBaseAddress);
auto enabled = isMultiOsContextCapable;
if (enableMultiGpuAtomicsOptimization) {
enabled = useGlobalAtomics && (enabled || areMultipleSubDevicesInContext);
}
EXPECT_EQ(!enabled, stateBaseAddress->getDisableSupportForMultiGpuAtomicsForStatelessAccesses());
}
INSTANTIATE_TEST_CASE_P(MultiGpuGlobalAtomics,
MultiGpuGlobalAtomicsTest,
::testing::Combine(
::testing::Bool(),
::testing::Bool(),
::testing::Bool(),
::testing::Bool()));
HWCMDTEST_F(IGFX_XE_HP_CORE, XeHpCommandStreamReceiverFlushTaskTests, givenDebugKeysThatOverrideMultiGpuSettingWhenStateBaseAddressIsProgrammedThenValuesMatch) {
DebugManagerStateRestore restorer;
using STATE_BASE_ADDRESS = typename FamilyType::STATE_BASE_ADDRESS;