feature: improve waiting and signaling Events via KMD calls

Related-To: NEO-8179

Signed-off-by: Dunajski, Bartosz <bartosz.dunajski@intel.com>

level_zero/core/source/cmdlist/cmdlist_hw.h

@@ -172,7 +172,7 @@ struct CommandListCoreFamily : CommandListImp {
     ze_result_t appendSignalEvent(ze_event_handle_t hEvent) override;
     ze_result_t appendWaitOnEvents(uint32_t numEvents, ze_event_handle_t *phEvent, bool relaxedOrderingAllowed, bool trackDependencies, bool signalInOrderCompletion) override;
     void appendWaitOnInOrderDependency(std::shared_ptr<InOrderExecInfo> &inOrderExecInfo, uint64_t waitValue, uint32_t offset, bool relaxedOrderingAllowed, bool implicitDependency);
-    void appendSignalInOrderDependencyCounter();
+    void appendSignalInOrderDependencyCounter(Event *signalEvent);
     void handleInOrderDependencyCounter(Event *signalEvent, bool nonWalkerInOrderCmdsChaining);
 
     ze_result_t appendWriteGlobalTimestamp(uint64_t *dstptr, ze_event_handle_t hSignalEvent,

level_zero/core/source/cmdlist/cmdlist_hw.inl

@@ -6,7 +6,6 @@
  */
 
 #include "shared/source/built_ins/built_ins.h"
-#include "shared/source/command_container/encode_interrupt_helper.h"
 #include "shared/source/command_container/encode_surface_state.h"
 #include "shared/source/command_stream/command_stream_receiver.h"
 #include "shared/source/device/device.h"
@@ -169,7 +168,7 @@ void CommandListCoreFamily<gfxCoreFamily>::handleInOrderDependencyCounter(Event
 
         UNRECOVERABLE_IF(inOrderAllocationOffset + offset >= inOrderExecInfo->inOrderDependencyCounterAllocation.getUnderlyingBufferSize());
 
-        CommandListCoreFamily<gfxCoreFamily>::appendSignalInOrderDependencyCounter(); // write 1 on new offset
+        CommandListCoreFamily<gfxCoreFamily>::appendSignalInOrderDependencyCounter(nullptr); // write 1 on new offset
     }
 
     inOrderExecInfo->inOrderDependencyCounter++;
@@ -545,7 +544,7 @@ ze_result_t CommandListCoreFamily<gfxCoreFamily>::appendEventReset(ze_event_hand
     }
 
     if (this->isInOrderExecutionEnabled()) {
-        appendSignalInOrderDependencyCounter();
+        appendSignalInOrderDependencyCounter(event);
         handleInOrderDependencyCounter(event, false);
     }
 
@@ -586,7 +585,7 @@ ze_result_t CommandListCoreFamily<gfxCoreFamily>::appendMemoryRangesBarrier(uint
     addToMappedEventList(signalEvent);
 
     if (this->isInOrderExecutionEnabled()) {
-        appendSignalInOrderDependencyCounter();
+        appendSignalInOrderDependencyCounter(signalEvent);
         handleInOrderDependencyCounter(signalEvent, false);
     }
 
@@ -1508,7 +1507,7 @@ ze_result_t CommandListCoreFamily<gfxCoreFamily>::appendMemoryCopy(void *dstptr,
                 NEO::PipeControlArgs args;
                 NEO::MemorySynchronizationCommands<GfxFamily>::addSingleBarrier(*commandContainer.getCommandStream(), args);
             }
-            appendSignalInOrderDependencyCounter();
+            appendSignalInOrderDependencyCounter(signalEvent);
         }
 
         if (!isCopyOnly() || inOrderCopyOnlySignalingAllowed) {
@@ -1604,7 +1603,7 @@ ze_result_t CommandListCoreFamily<gfxCoreFamily>::appendMemoryCopyRegion(void *d
 
     if (this->isInOrderExecutionEnabled()) {
         if (inOrderCopyOnlySignalingAllowed) {
-            appendSignalInOrderDependencyCounter();
+            appendSignalInOrderDependencyCounter(signalEvent);
         }
 
         if (!isCopyOnly() || inOrderCopyOnlySignalingAllowed) {
@@ -2058,7 +2057,7 @@ ze_result_t CommandListCoreFamily<gfxCoreFamily>::appendMemoryFill(void *ptr,
                 NEO::PipeControlArgs args;
                 NEO::MemorySynchronizationCommands<GfxFamily>::addSingleBarrier(*commandContainer.getCommandStream(), args);
             }
-            appendSignalInOrderDependencyCounter();
+            appendSignalInOrderDependencyCounter(signalEvent);
         } else {
             nonWalkerInOrderCmdChaining = isInOrderNonWalkerSignalingRequired(signalEvent);
         }
@@ -2127,7 +2126,7 @@ ze_result_t CommandListCoreFamily<gfxCoreFamily>::appendBlitFill(void *ptr,
         appendSignalEventPostWalker(signalEvent, false);
 
         if (isInOrderExecutionEnabled()) {
-            appendSignalInOrderDependencyCounter();
+            appendSignalInOrderDependencyCounter(signalEvent);
             handleInOrderDependencyCounter(signalEvent, false);
         }
     }
@@ -2350,7 +2349,7 @@ ze_result_t CommandListCoreFamily<gfxCoreFamily>::appendSignalEvent(ze_event_han
     dispatchEventPostSyncOperation(event, Event::STATE_SIGNALED, false, false, appendPipeControlWithPostSync, false);
 
     if (this->isInOrderExecutionEnabled()) {
-        appendSignalInOrderDependencyCounter();
+        appendSignalInOrderDependencyCounter(event);
         handleInOrderDependencyCounter(event, false);
     }
 
@@ -2409,10 +2408,6 @@ void CommandListCoreFamily<gfxCoreFamily>::appendWaitOnInOrderDependency(std::sh
 
         gpuAddress += sizeof(uint64_t);
     }
-
-    if (NEO::EncodeUserInterruptHelper::isOperationAllowed(NEO::EncodeUserInterruptHelper::afterSemaphoreMask)) {
-        NEO::EnodeUserInterrupt<GfxFamily>::encode(*commandContainer.getCommandStream());
-    }
 }
 
 template <GFXCORE_FAMILY gfxCoreFamily>
@@ -2498,7 +2493,7 @@ ze_result_t CommandListCoreFamily<gfxCoreFamily>::appendWaitOnEvents(uint32_t nu
     }
 
     if (signalInOrderCompletion) {
-        appendSignalInOrderDependencyCounter();
+        appendSignalInOrderDependencyCounter(nullptr);
         handleInOrderDependencyCounter(nullptr, false);
     }
 
@@ -2516,7 +2511,7 @@ ze_result_t CommandListCoreFamily<gfxCoreFamily>::appendWaitOnEvents(uint32_t nu
 }
 
 template <GFXCORE_FAMILY gfxCoreFamily>
-void CommandListCoreFamily<gfxCoreFamily>::appendSignalInOrderDependencyCounter() {
+void CommandListCoreFamily<gfxCoreFamily>::appendSignalInOrderDependencyCounter(Event *signalEvent) {
     using MI_STORE_DATA_IMM = typename GfxFamily::MI_STORE_DATA_IMM;
 
     uint64_t signalValue = inOrderExecInfo->inOrderDependencyCounter + 1;
@@ -2530,7 +2525,7 @@ void CommandListCoreFamily<gfxCoreFamily>::appendSignalInOrderDependencyCounter(
 
     addCmdForPatching(nullptr, miStoreCmd, nullptr, signalValue, InOrderPatchCommandHelpers::PatchCmdType::Sdi);
 
-    if (NEO::EncodeUserInterruptHelper::isOperationAllowed(NEO::EncodeUserInterruptHelper::onSignalingFenceMask)) {
+    if ((NEO::DebugManager.flags.ProgramUserInterruptOnResolvedDependency.get() == 1) && signalEvent && signalEvent->isKmdWaitModeEnabled()) {
         NEO::EnodeUserInterrupt<GfxFamily>::encode(*commandContainer.getCommandStream());
     }
 }
@@ -2670,7 +2665,7 @@ ze_result_t CommandListCoreFamily<gfxCoreFamily>::appendWriteGlobalTimestamp(
     appendSignalEventPostWalker(signalEvent, false);
 
     if (this->isInOrderExecutionEnabled()) {
-        appendSignalInOrderDependencyCounter();
+        appendSignalInOrderDependencyCounter(signalEvent);
         handleInOrderDependencyCounter(signalEvent, false);
     }
 
@@ -3181,7 +3176,7 @@ ze_result_t CommandListCoreFamily<gfxCoreFamily>::appendBarrier(ze_event_handle_
     appendSignalEventPostWalker(signalEvent, this->isInOrderExecutionEnabled());
 
     if (isInOrderExecutionEnabled()) {
-        appendSignalInOrderDependencyCounter();
+        appendSignalInOrderDependencyCounter(signalEvent);
         handleInOrderDependencyCounter(signalEvent, false);
     }
 
@@ -3344,7 +3339,7 @@ ze_result_t CommandListCoreFamily<gfxCoreFamily>::appendWaitOnMemory(void *desc,
     appendSignalEventPostWalker(signalEvent, false);
 
     if (this->isInOrderExecutionEnabled()) {
-        appendSignalInOrderDependencyCounter();
+        appendSignalInOrderDependencyCounter(signalEvent);
         handleInOrderDependencyCounter(signalEvent, false);
     }
 
@@ -3392,7 +3387,7 @@ ze_result_t CommandListCoreFamily<gfxCoreFamily>::appendWriteToMemory(void *desc
     }
 
     if (this->isInOrderExecutionEnabled()) {
-        appendSignalInOrderDependencyCounter();
+        appendSignalInOrderDependencyCounter(nullptr);
         handleInOrderDependencyCounter(nullptr, false);
     }
 

level_zero/core/source/cmdlist/cmdlist_hw_immediate.inl

@@ -471,7 +471,7 @@ void CommandListCoreFamilyImmediate<gfxCoreFamily>::handleInOrderNonWalkerSignal
     }
 
     CommandListCoreFamily<gfxCoreFamily>::appendWaitOnSingleEvent(event, nonWalkerSignalingHasRelaxedOrdering);
-    CommandListCoreFamily<gfxCoreFamily>::appendSignalInOrderDependencyCounter();
+    CommandListCoreFamily<gfxCoreFamily>::appendSignalInOrderDependencyCounter(event);
 }
 
 template <GFXCORE_FAMILY gfxCoreFamily>

level_zero/core/source/cmdlist/cmdlist_hw_skl_to_tgllp.inl

@@ -278,7 +278,7 @@ ze_result_t CommandListCoreFamily<gfxCoreFamily>::appendLaunchKernelWithParams(K
         NEO::PipeControlArgs args;
 
         NEO::MemorySynchronizationCommands<GfxFamily>::addSingleBarrier(*commandContainer.getCommandStream(), args);
-        appendSignalInOrderDependencyCounter();
+        appendSignalInOrderDependencyCounter(event);
     }
 
     return ZE_RESULT_SUCCESS;

level_zero/core/source/cmdlist/cmdlist_hw_xehp_and_later.inl

@@ -331,7 +331,7 @@ ze_result_t CommandListCoreFamily<gfxCoreFamily>::appendLaunchKernelWithParams(K
         if (inOrderNonWalkerSignalling) {
             if (!launchParams.skipInOrderNonWalkerSignaling) {
                 appendWaitOnSingleEvent(eventForInOrderExec, false);
-                appendSignalInOrderDependencyCounter();
+                appendSignalInOrderDependencyCounter(eventForInOrderExec);
             }
         } else {
             UNRECOVERABLE_IF(!dispatchKernelArgs.outWalkerPtr);

level_zero/core/source/device/bcs_split.h

@@ -134,7 +134,7 @@ struct BcsSplit {
         cmdList->appendEventForProfilingAllWalkers(this->events.marker[markerEventIndex], false, true);
 
         if (cmdList->isInOrderExecutionEnabled()) {
-            cmdList->appendSignalInOrderDependencyCounter();
+            cmdList->appendSignalInOrderDependencyCounter(signalEvent);
             cmdList->handleInOrderDependencyCounter(signalEvent, false);
         }
 

level_zero/core/source/event/event.h

@@ -242,6 +242,8 @@ struct Event : _ze_event_handle_t {
     const CommandQueue *getLatestUsedCmdQueue() const { return latestUsedCmdQueue; }
     bool hasKerneMappedTsCapability = false;
     std::shared_ptr<InOrderExecInfo> &getInOrderExecInfo() { return inOrderExecInfo; }
+    void enableKmdWaitMode() { kmdWaitMode = true; }
+    bool isKmdWaitModeEnabled() const { return kmdWaitMode; }
 
   protected:
     Event(EventPool *eventPool, int index, Device *device) : device(device), eventPool(eventPool), index(index) {}
@@ -299,6 +301,7 @@ struct Event : _ze_event_handle_t {
     bool usingContextEndOffset = false;
     bool signalAllEventPackets = false;
     bool isFromIpcPool = false;
+    bool kmdWaitMode = false;
     uint64_t timestampRefreshIntervalInNanoSec = 0;
 };
 

level_zero/core/source/event/event_impl.inl

@@ -87,6 +87,10 @@ Event *Event::create(EventPool *eventPool, const ze_event_desc_t *desc, Device *
         event->enableCounterBasedMode(true);
     }
 
+    if (NEO::DebugManager.flags.WaitForUserFenceOnEventHostSynchronize.get() == 1) {
+        event->enableKmdWaitMode();
+    }
+
     return event;
 }
 
@@ -457,7 +461,7 @@ ze_result_t EventImp<TagSizeT>::hostSynchronize(uint64_t timeout) {
     waitStartTime = std::chrono::high_resolution_clock::now();
     lastHangCheckTime = waitStartTime;
     do {
-        if (NEO::DebugManager.flags.WaitForUserFenceOnEventHostSynchronize.get() == 1 && isCounterBased()) {
+        if (isKmdWaitModeEnabled() && isCounterBased()) {
             ret = waitForUserFence(timeout);
         } else {
             ret = queryStatus();

level_zero/core/test/unit_tests/sources/cmdlist/test_cmdlist_append_launch_kernel_3.cpp

@@ -6,7 +6,6 @@
  */
 
 #include "shared/source/command_container/command_encoder.h"
-#include "shared/source/command_container/encode_interrupt_helper.h"
 #include "shared/source/command_container/encode_surface_state.h"
 #include "shared/source/command_container/implicit_scaling.h"
 #include "shared/source/helpers/api_specific_config.h"
@@ -1031,6 +1030,8 @@ HWTEST2_F(InOrderCmdListTests, givenDebugFlagSetWhenEventHostSyncCalledThenCallW
     auto immCmdList = createImmCmdList<gfxCoreFamily>();
 
     auto eventPool = createEvents<FamilyType>(2, false);
+    EXPECT_TRUE(events[0]->isKmdWaitModeEnabled());
+    EXPECT_TRUE(events[1]->isKmdWaitModeEnabled());
 
     EXPECT_EQ(ZE_RESULT_NOT_READY, events[0]->hostSynchronize(2));
 
@@ -1150,43 +1151,6 @@ HWTEST2_F(InOrderCmdListTests, givenInOrderModeWhenSubmittingThenProgramSemaphor
     ASSERT_TRUE(verifyInOrderDependency<FamilyType>(itor, 1, immCmdList->inOrderExecInfo->inOrderDependencyCounterAllocation.getGpuAddress() + counterOffset, immCmdList->isQwordInOrderCounter()));
 }
 
-HWTEST2_F(InOrderCmdListTests, givenDebugFlagSetWhenDispatchingSemaphoreThenProgramUserInterrupt, IsAtLeastSkl) {
-    using MI_USER_INTERRUPT = typename FamilyType::MI_USER_INTERRUPT;
-    using MI_SEMAPHORE_WAIT = typename FamilyType::MI_SEMAPHORE_WAIT;
-
-    DebugManager.flags.ProgramUserInterruptOnResolvedDependency.set(NEO::EncodeUserInterruptHelper::afterSemaphoreMask);
-
-    auto eventPool = createEvents<FamilyType>(1, false);
-    auto eventHandle = events[0]->toHandle();
-    events[0]->makeCounterBasedInitiallyDisabled();
-
-    auto immCmdList = createImmCmdList<gfxCoreFamily>();
-
-    auto cmdStream = immCmdList->getCmdContainer().getCommandStream();
-
-    immCmdList->appendLaunchKernel(kernel->toHandle(), groupCount, nullptr, 0, nullptr, launchParams, false);
-
-    auto offset = cmdStream->getUsed();
-
-    immCmdList->appendBarrier(nullptr, 1, &eventHandle, false);
-
-    GenCmdList cmdList;
-    ASSERT_TRUE(FamilyType::PARSE::parseCommandBuffer(
-        cmdList,
-        ptrOffset(cmdStream->getCpuBase(), offset),
-        cmdStream->getUsed() - offset));
-
-    auto itor = find<MI_SEMAPHORE_WAIT *>(cmdList.begin(), cmdList.end());
-
-    ASSERT_NE(cmdList.end(), itor);
-
-    auto userInterruptCmd = genCmdCast<MI_USER_INTERRUPT *>(*(++itor));
-    ASSERT_NE(nullptr, userInterruptCmd);
-
-    auto allCmds = findAll<MI_USER_INTERRUPT *>(cmdList.begin(), cmdList.end());
-    EXPECT_EQ(1u, allCmds.size());
-}
-
 HWTEST2_F(InOrderCmdListTests, givenTimestmapEventWhenProgrammingBarrierThenDontAddPipeControl, IsAtLeastSkl) {
     using PIPE_CONTROL = typename FamilyType::PIPE_CONTROL;
 
@@ -1218,22 +1182,22 @@ HWTEST2_F(InOrderCmdListTests, givenDebugFlagSetWhenDispatchingStoreDataImmThenP
     using MI_USER_INTERRUPT = typename FamilyType::MI_USER_INTERRUPT;
     using MI_STORE_DATA_IMM = typename FamilyType::MI_STORE_DATA_IMM;
 
-    DebugManager.flags.ProgramUserInterruptOnResolvedDependency.set(NEO::EncodeUserInterruptHelper::onSignalingFenceMask);
+    DebugManager.flags.ProgramUserInterruptOnResolvedDependency.set(1);
 
-    auto eventPool = createEvents<FamilyType>(1, false);
+    auto eventPool = createEvents<FamilyType>(2, false);
     auto eventHandle = events[0]->toHandle();
     events[0]->makeCounterBasedInitiallyDisabled();
 
+    EXPECT_FALSE(events[1]->isKmdWaitModeEnabled());
+
     auto immCmdList = createImmCmdList<gfxCoreFamily>();
 
     auto cmdStream = immCmdList->getCmdContainer().getCommandStream();
 
     immCmdList->appendLaunchKernel(kernel->toHandle(), groupCount, nullptr, 0, nullptr, launchParams, false);
-
     auto offset = cmdStream->getUsed();
 
-    immCmdList->appendBarrier(nullptr, 1, &eventHandle, false);
-
+    auto validateInterrupt = [&](bool interruptExpected) {
         GenCmdList cmdList;
         ASSERT_TRUE(FamilyType::PARSE::parseCommandBuffer(
             cmdList,
@@ -1250,61 +1214,26 @@ HWTEST2_F(InOrderCmdListTests, givenDebugFlagSetWhenDispatchingStoreDataImmThenP
         EXPECT_EQ(immCmdList->inOrderExecInfo->inOrderDependencyCounterAllocation.getGpuAddress(), sdiCmd->getAddress());
 
         auto userInterruptCmd = genCmdCast<MI_USER_INTERRUPT *>(*(++itor));
-    ASSERT_NE(nullptr, userInterruptCmd);
+        ASSERT_EQ(interruptExpected, nullptr != userInterruptCmd);
 
         auto allCmds = findAll<MI_USER_INTERRUPT *>(cmdList.begin(), cmdList.end());
-    EXPECT_EQ(1u, allCmds.size());
-}
-
-HWTEST2_F(InOrderCmdListTests, givenDebugFlagSetAsMaskWhenDispatchingStoreDataImmAndSemaphoreThenProgramUserInterrupt, IsAtLeastSkl) {
-    using MI_USER_INTERRUPT = typename FamilyType::MI_USER_INTERRUPT;
-    using MI_STORE_DATA_IMM = typename FamilyType::MI_STORE_DATA_IMM;
-
-    constexpr int32_t invalidMask = 0b100;
-
-    DebugManager.flags.ProgramUserInterruptOnResolvedDependency.set(invalidMask);
-
-    auto eventPool = createEvents<FamilyType>(1, false);
-    auto eventHandle = events[0]->toHandle();
-    events[0]->makeCounterBasedInitiallyDisabled();
-
-    auto immCmdList = createImmCmdList<gfxCoreFamily>();
-
-    auto cmdStream = immCmdList->getCmdContainer().getCommandStream();
-
-    immCmdList->appendLaunchKernel(kernel->toHandle(), groupCount, nullptr, 0, nullptr, launchParams, false);
-
-    auto offset = cmdStream->getUsed();
+        EXPECT_EQ(interruptExpected ? 1u : 0u, allCmds.size());
+    };
 
+    // no signal Event
     immCmdList->appendBarrier(nullptr, 1, &eventHandle, false);
+    validateInterrupt(false);
 
-    {
-        GenCmdList cmdList;
-        ASSERT_TRUE(FamilyType::PARSE::parseCommandBuffer(
-            cmdList,
-            ptrOffset(cmdStream->getCpuBase(), offset),
-            cmdStream->getUsed() - offset));
-
-        auto allCmds = findAll<MI_USER_INTERRUPT *>(cmdList.begin(), cmdList.end());
-        EXPECT_EQ(0u, allCmds.size());
-    }
-
-    DebugManager.flags.ProgramUserInterruptOnResolvedDependency.set(NEO::EncodeUserInterruptHelper::onSignalingFenceMask | NEO::EncodeUserInterruptHelper::afterSemaphoreMask);
-
+    // regular signal Event
     offset = cmdStream->getUsed();
+    immCmdList->appendBarrier(events[1]->toHandle(), 1, &eventHandle, false);
+    validateInterrupt(false);
 
-    immCmdList->appendBarrier(nullptr, 1, &eventHandle, false);
-
-    {
-        GenCmdList cmdList;
-        ASSERT_TRUE(FamilyType::PARSE::parseCommandBuffer(
-            cmdList,
-            ptrOffset(cmdStream->getCpuBase(), offset),
-            cmdStream->getUsed() - offset));
-
-        auto allCmds = findAll<MI_USER_INTERRUPT *>(cmdList.begin(), cmdList.end());
-        EXPECT_EQ(2u, allCmds.size());
-    }
+    // signal Event with kmd wait mode
+    offset = cmdStream->getUsed();
+    events[1]->enableKmdWaitMode();
+    immCmdList->appendBarrier(events[1]->toHandle(), 1, &eventHandle, false);
+    validateInterrupt(true);
 }
 
 HWTEST2_F(InOrderCmdListTests, givenInOrderModeWhenWaitingForEventFromPreviousAppendThenSkip, IsAtLeastXeHpCore) {
@@ -4067,7 +3996,7 @@ HWTEST2_F(MultiTileInOrderCmdListTests, givenMultiTileInOrderModeWhenSignalingSy
 
     auto cmdStream = immCmdList->getCmdContainer().getCommandStream();
 
-    immCmdList->appendSignalInOrderDependencyCounter();
+    immCmdList->appendSignalInOrderDependencyCounter(nullptr);
 
     GenCmdList cmdList;
     ASSERT_TRUE(FamilyType::PARSE::parseCommandBuffer(cmdList, cmdStream->getCpuBase(), cmdStream->getUsed()));

shared/source/command_container/CMakeLists.txt

@@ -15,7 +15,6 @@ set(NEO_CORE_COMMAND_CONTAINER
     ${CMAKE_CURRENT_SOURCE_DIR}/encode_alu_helper.h
     ${CMAKE_CURRENT_SOURCE_DIR}/encode_compute_mode_bdw_and_later.inl
     ${CMAKE_CURRENT_SOURCE_DIR}/encode_compute_mode_tgllp_and_later.inl
-    ${CMAKE_CURRENT_SOURCE_DIR}/encode_interrupt_helper.h
     ${CMAKE_CURRENT_SOURCE_DIR}/encode_surface_state.h
     ${CMAKE_CURRENT_SOURCE_DIR}/implicit_scaling.cpp
     ${CMAKE_CURRENT_SOURCE_DIR}/implicit_scaling.h

shared/source/command_container/encode_interrupt_helper.h

@@ -1,30 +0,0 @@
-/*
- * Copyright (C) 2023 Intel Corporation
- *
- * SPDX-License-Identifier: MIT
- *
- */
-
-/*
- * Copyright (C)2023 Intel Corporation
- *
- * SPDX-License-Identifier: MIT
- *
- */
-
-#pragma once
-
-#include "shared/source/debug_settings/debug_settings_manager.h"
-
-namespace NEO {
-struct EncodeUserInterruptHelper {
-    static constexpr int32_t afterSemaphoreMask = 0b01;
-    static constexpr int32_t onSignalingFenceMask = 0b10;
-
-    static bool isOperationAllowed(int32_t mode) {
-        const int32_t flagValue = NEO::DebugManager.flags.ProgramUserInterruptOnResolvedDependency.get();
-
-        return (flagValue != -1 && (flagValue & mode));
-    }
-};
-} // namespace NEO

shared/source/debug_settings/debug_variables_base.inl

@@ -258,7 +258,7 @@ DECLARE_DEBUG_VARIABLE(int32_t, TrackNumCsrClientsOnSyncPoints, -1, "-1: default
 DECLARE_DEBUG_VARIABLE(int32_t, OverrideDriverVersion, -1, "-1: default, >=0: Use value as reported driver version")
 DECLARE_DEBUG_VARIABLE(int32_t, WaitForUserFenceOnEventHostSynchronize, -1, "-1: default, 0: Disabled, 1: Enabled. If enabled, use WaitUserFence KMD call for in-order Events instead of active polling on host.")
 DECLARE_DEBUG_VARIABLE(int32_t, DisableSystemPointerKernelArgument, -1, "-1: default, 0: Disabled, 1: using a system pointer for kernel argument returns an error.")
-DECLARE_DEBUG_VARIABLE(int32_t, ProgramUserInterruptOnResolvedDependency, -1, "-1: default, 0: Disabled, >=1: bitfield. 01b: program after semaphore, 10b: on signaling fence (non-walker append).")
+DECLARE_DEBUG_VARIABLE(int32_t, ProgramUserInterruptOnResolvedDependency, -1, "-1: default, 0: Disabled, 1: On signaling append completion (if possible) - for example in-order counter update")
 DECLARE_DEBUG_VARIABLE(int32_t, EnableInOrderRegularCmdListPatching, -1, "-1: default, 0: Disabled, 1: If set, patch counter value on execute call")
 DECLARE_DEBUG_VARIABLE(int32_t, EnableInOrderRelaxedOrderingForEventsChaining, -1, "-1: default, 0: Disabled, 1: If set, send 2 immediate flushes to avoid stalling RelaxedOrdering Scheduler.")