Add sfence when stop ring buffer

Signed-off-by: Lukasz Jobczyk <[email protected]> Source: 5011046
intel · Nov 29, 2022 · 4798c8a · 4798c8a
1 parent 9a8d4ee
commit 4798c8a
Show file tree

Hide file tree

Showing 3 changed files with 39 additions and 12 deletions.
diff --git a/shared/source/direct_submission/direct_submission_hw.h b/shared/source/direct_submission/direct_submission_hw.h
@@ -107,6 +107,7 @@ class DirectSubmissionHw {
     GraphicsAllocation *switchRingBuffersAllocations();
     virtual uint64_t updateTagValue() = 0;
     virtual void getTagAddressValue(TagData &tagData) = 0;
+    void unblockGpu();
 
     void cpuCachelineFlush(void *ptr, size_t size);
 

diff --git a/shared/source/direct_submission/direct_submission_hw.inl b/shared/source/direct_submission/direct_submission_hw.inl
@@ -159,6 +159,19 @@ bool DirectSubmissionHw<GfxFamily, Dispatcher>::makeResourcesResident(DirectSubm
     return ret;
 }
 
+template <typename GfxFamily, typename Dispatcher>
+inline void DirectSubmissionHw<GfxFamily, Dispatcher>::unblockGpu() {
+    if (sfenceMode >= DirectSubmissionSfenceMode::BeforeSemaphoreOnly) {
+        CpuIntrinsics::sfence();
+    }
+
+    semaphoreData->QueueWorkCount = currentQueueWorkCount;
+
+    if (sfenceMode == DirectSubmissionSfenceMode::BeforeAndAfterSemaphore) {
+        CpuIntrinsics::sfence();
+    }
+}
+
 template <typename GfxFamily, typename Dispatcher>
 inline void DirectSubmissionHw<GfxFamily, Dispatcher>::cpuCachelineFlush(void *ptr, size_t size) {
     if (disableCpuCacheFlush) {
@@ -273,8 +286,7 @@ bool DirectSubmissionHw<GfxFamily, Dispatcher>::stopRingBuffer() {
     EncodeNoop<GfxFamily>::alignToCacheLine(ringCommandStream);
 
     cpuCachelineFlush(flushPtr, getSizeEnd());
-
-    semaphoreData->QueueWorkCount = currentQueueWorkCount;
+    this->unblockGpu();
     cpuCachelineFlush(semaphorePtr, MemoryConstants::cacheLineSize);
 
     this->handleStopRingBuffer();
@@ -478,16 +490,7 @@ bool DirectSubmissionHw<GfxFamily, Dispatcher>::dispatchCommandBuffer(BatchBuffe
         reserved = *ringBufferStart;
     }
 
-    if (sfenceMode >= DirectSubmissionSfenceMode::BeforeSemaphoreOnly) {
-        CpuIntrinsics::sfence();
-    }
-
-    //unblock GPU
-    semaphoreData->QueueWorkCount = currentQueueWorkCount;
-
-    if (sfenceMode == DirectSubmissionSfenceMode::BeforeAndAfterSemaphore) {
-        CpuIntrinsics::sfence();
-    }
+    this->unblockGpu();
 
     cpuCachelineFlush(semaphorePtr, MemoryConstants::cacheLineSize);
     currentQueueWorkCount++;

diff --git a/shared/test/unit_test/direct_submission/direct_submission_tests_2.cpp b/shared/test/unit_test/direct_submission/direct_submission_tests_2.cpp
@@ -879,3 +879,26 @@ HWTEST_F(DirectSubmissionDispatchBufferTest, givenDebugFlagSetWhenDispatchingWor
         EXPECT_EQ(initialCounterValue + expectedCount, CpuIntrinsicsTests::sfenceCounter);
     }
 }
+
+HWTEST_F(DirectSubmissionDispatchBufferTest, givenDebugFlagSetWhenStoppingRingbufferThenProgramSfenceInstruction) {
+    DebugManagerStateRestore restorer{};
+
+    using Dispatcher = BlitterDispatcher<FamilyType>;
+
+    FlushStampTracker flushStamp(true);
+
+    for (int32_t debugFlag : {-1, 0, 1, 2}) {
+        DebugManager.flags.DirectSubmissionInsertSfenceInstructionPriorToSubmission.set(debugFlag);
+
+        MockDirectSubmissionHw<FamilyType, Dispatcher> directSubmission(*pDevice->getDefaultEngine().commandStreamReceiver);
+        EXPECT_TRUE(directSubmission.initialize(true, true));
+
+        auto initialCounterValue = CpuIntrinsicsTests::sfenceCounter.load();
+
+        EXPECT_TRUE(directSubmission.stopRingBuffer());
+
+        uint32_t expectedCount = (debugFlag == -1) ? 2 : static_cast<uint32_t>(debugFlag);
+
+        EXPECT_EQ(initialCounterValue + expectedCount, CpuIntrinsicsTests::sfenceCounter);
+    }
+}