From fe3f32eb688674bf83d0bc3ee9fcd409cb0763dc Mon Sep 17 00:00:00 2001
From: Zbigniew Zdanowicz <zbigniew.zdanowicz@intel.com>
Date: Fri, 20 Jan 2023 11:30:38 +0000
Subject: [PATCH] Fix relax ordering check for event remaining packets

Related-To: NEO-7490

Signed-off-by: Zbigniew Zdanowicz <zbigniew.zdanowicz@intel.com>
---
 level_zero/core/source/cmdlist/cmdlist_hw.h   |  1 -
 level_zero/core/source/cmdlist/cmdlist_hw.inl | 30 ++-----------------
 2 files changed, 3 insertions(+), 28 deletions(-)

diff --git a/level_zero/core/source/cmdlist/cmdlist_hw.h b/level_zero/core/source/cmdlist/cmdlist_hw.h
index ec0741715a..480b83e944 100644
--- a/level_zero/core/source/cmdlist/cmdlist_hw.h
+++ b/level_zero/core/source/cmdlist/cmdlist_hw.h
@@ -300,7 +300,6 @@ struct CommandListCoreFamily : CommandListImp {
                compactL3FlushEvent(dcFlush);
     }
     void allocateKernelPrivateMemoryIfNeeded(Kernel *kernel, uint32_t sizePerHwThread);
-    void waitOnRemainingEventPackets(Event *event);
     CmdListEventOperation estimateEventPostSync(Event *event, uint32_t operations);
     void dispatchPostSyncCopy(uint64_t gpuAddress, uint32_t value, bool workloadPartition);
     void dispatchPostSyncCompute(uint64_t gpuAddress, uint32_t value, bool workloadPartition);
diff --git a/level_zero/core/source/cmdlist/cmdlist_hw.inl b/level_zero/core/source/cmdlist/cmdlist_hw.inl
index c4b3e2560b..c418860804 100644
--- a/level_zero/core/source/cmdlist/cmdlist_hw.inl
+++ b/level_zero/core/source/cmdlist/cmdlist_hw.inl
@@ -2026,7 +2026,9 @@ ze_result_t CommandListCoreFamily<gfxCoreFamily>::appendWaitOnEvents(uint32_t nu
         commandContainer.addToResidencyContainer(&event->getAllocation(this->device));
         gpuAddr = event->getCompletionFieldGpuAddress(this->device);
         uint32_t packetsToWait = event->getPacketsInUse();
-
+        if (this->signalAllEventPackets) {
+            packetsToWait = event->getMaxPacketsCount();
+        }
         for (uint32_t i = 0u; i < packetsToWait; i++) {
             if (relaxedOrdering) {
                 NEO::EncodeBatchBufferStartOrEnd<GfxFamily>::programConditionalDataMemBatchBufferStart(*commandContainer.getCommandStream(), 0, gpuAddr, eventStateClear,
@@ -2040,9 +2042,6 @@ ze_result_t CommandListCoreFamily<gfxCoreFamily>::appendWaitOnEvents(uint32_t nu
 
             gpuAddr += event->getSinglePacketSize();
         }
-        if (this->signalAllEventPackets) {
-            waitOnRemainingEventPackets(event);
-        }
     }
 
     if (NEO::DebugManager.flags.EnableSWTags.get()) {
@@ -2714,29 +2713,6 @@ void CommandListCoreFamily<gfxCoreFamily>::allocateKernelPrivateMemoryIfNeeded(K
     }
 }
 
-template <GFXCORE_FAMILY gfxCoreFamily>
-void CommandListCoreFamily<gfxCoreFamily>::waitOnRemainingEventPackets(Event *event) {
-    using COMPARE_OPERATION = typename GfxFamily::MI_SEMAPHORE_WAIT::COMPARE_OPERATION;
-
-    uint32_t packetUsed = event->getPacketsInUse();
-    uint32_t packetsRemaining = event->getMaxPacketsCount() - packetUsed;
-    if (packetsRemaining == 0) {
-        return;
-    }
-
-    uint64_t gpuAddress = event->getCompletionFieldGpuAddress(this->device);
-    size_t packetSize = event->getSinglePacketSize();
-    gpuAddress += packetSize * packetUsed;
-
-    for (uint32_t i = 0; i < packetsRemaining; i++) {
-        NEO::EncodeSempahore<GfxFamily>::addMiSemaphoreWaitCommand(*commandContainer.getCommandStream(),
-                                                                   gpuAddress,
-                                                                   Event::STATE_CLEARED,
-                                                                   COMPARE_OPERATION::COMPARE_OPERATION_SAD_NOT_EQUAL_SDD);
-        gpuAddress += packetSize;
-    }
-}
-
 template <GFXCORE_FAMILY gfxCoreFamily>
 CmdListEventOperation CommandListCoreFamily<gfxCoreFamily>::estimateEventPostSync(Event *event, uint32_t operations) {
     CmdListEventOperation ret;