Optimize setKernelArgSVMPointer

If same pointer is already set, we don't need to set it again.

Related-To: NEO-6737

Signed-off-by: Dominik Dabek <dominik.dabek@intel.com>

opencl/test/unit_test/api/cl_set_kernel_arg_svm_pointer_tests.inl

@@ -1,5 +1,5 @@
 /*
- * Copyright (C) 2018-2021 Intel Corporation
+ * Copyright (C) 2018-2022 Intel Corporation
  *
  * SPDX-License-Identifier: MIT
  *
@@ -206,4 +206,31 @@ TEST_F(clSetKernelArgSVMPointerTests, GivenSvmAndPointerWithInvalidOffsetWhenSet
         clSVMFree(pContext, ptrSvm);
     }
 }
+
+TEST_F(clSetKernelArgSVMPointerTests, GivenSvmAndValidArgValueWhenSettingSameKernelArgTwiceThenSetArgSvmAllocCalledOnlyFirstTime) {
+    const ClDeviceInfo &devInfo = pDevice->getDeviceInfo();
+    if (devInfo.svmCapabilities != 0) {
+        EXPECT_EQ(0u, pMockKernel->setArgSvmAllocCalls);
+        void *ptrSvm = clSVMAlloc(pContext, CL_MEM_READ_WRITE, 256, 4);
+        EXPECT_NE(nullptr, ptrSvm);
+
+        auto retVal = clSetKernelArgSVMPointer(
+            pMockMultiDeviceKernel, // cl_kernel kernel
+            0,                      // cl_uint arg_index
+            ptrSvm                  // const void *arg_value
+        );
+        EXPECT_EQ(CL_SUCCESS, retVal);
+        EXPECT_EQ(1u, pMockKernel->setArgSvmAllocCalls);
+
+        retVal = clSetKernelArgSVMPointer(
+            pMockMultiDeviceKernel, // cl_kernel kernel
+            0,                      // cl_uint arg_index
+            ptrSvm                  // const void *arg_value
+        );
+        EXPECT_EQ(CL_SUCCESS, retVal);
+        EXPECT_EQ(1u, pMockKernel->setArgSvmAllocCalls);
+
+        clSVMFree(pContext, ptrSvm);
+    }
+}
 } // namespace ULT

opencl/test/unit_test/built_ins/built_in_tests.cpp

@@ -532,7 +532,7 @@ HWTEST2_P(AuxBuiltInTests, givenKernelWithAuxTranslationRequiredWhenEnqueueCalle
         MockBuffer::setAllocationType(gfxAllocation, pDevice->getRootDeviceEnvironment().getGmmClientContext(), true);
 
         auto ptr = reinterpret_cast<void *>(gfxAllocation->getGpuAddressToPatch());
-        mockKernel.mockKernel->setArgSvmAlloc(0, ptr, gfxAllocation);
+        mockKernel.mockKernel->setArgSvmAlloc(0, ptr, gfxAllocation, 0u);
 
         gmm.reset(gfxAllocation->getDefaultGmm());
     }

opencl/test/unit_test/command_queue/blit_enqueue_tests.cpp

@@ -140,7 +140,7 @@ struct BlitEnqueueTests : public ::testing::Test {
         for (uint32_t i = 0; i < allocs.size(); i++) {
             auto alloc = allocs[i];
             auto ptr = reinterpret_cast<void *>(alloc->getGpuAddressToPatch());
-            mockKernel->mockKernel->setArgSvmAlloc(i, ptr, alloc);
+            mockKernel->mockKernel->setArgSvmAlloc(i, ptr, alloc, 0u);
         }
     }
 

opencl/test/unit_test/context/driver_diagnostics_tests.cpp

@@ -476,7 +476,7 @@ TEST_F(PerformanceHintTest, givenPrintDriverDiagnosticsDebugModeEnabledWhenCallF
 
     mockKernel.mockKernel->initialize();
     mockKernel.mockKernel->auxTranslationRequired = true;
-    mockKernel.mockKernel->setArgSvmAlloc(0, ptr, &gfxAllocation);
+    mockKernel.mockKernel->setArgSvmAlloc(0, ptr, &gfxAllocation, 0u);
 
     testing::internal::CaptureStdout();
     KernelObjsForAuxTranslation kernelObjects;
@@ -579,7 +579,7 @@ TEST_F(PerformanceHintTest, givenPrintDriverDiagnosticsDebugModeEnabledWhenKerne
 
     mockKernel.mockKernel->initialize();
     mockKernel.mockKernel->auxTranslationRequired = true;
-    mockKernel.mockKernel->setArgSvmAlloc(0, ptr, &gfxAllocation);
+    mockKernel.mockKernel->setArgSvmAlloc(0, ptr, &gfxAllocation, 0u);
 
     testing::internal::CaptureStdout();
 
@@ -606,7 +606,7 @@ TEST_F(PerformanceHintTest, givenPrintDriverDiagnosticsDebugModeDisabledWhenCall
 
     mockKernel.mockKernel->initialize();
     mockKernel.mockKernel->auxTranslationRequired = true;
-    mockKernel.mockKernel->setArgSvmAlloc(0, ptr, &gfxAllocation);
+    mockKernel.mockKernel->setArgSvmAlloc(0, ptr, &gfxAllocation, 0u);
 
     testing::internal::CaptureStdout();
 
@@ -627,7 +627,7 @@ TEST_F(PerformanceHintTest, whenCallingFillWithKernelObjsForAuxTranslationOnNull
     mockKernel.kernelInfo.addArgBuffer(0, 0, 0, 0);
 
     mockKernel.mockKernel->initialize();
-    mockKernel.mockKernel->setArgSvmAlloc(0, nullptr, nullptr);
+    mockKernel.mockKernel->setArgSvmAlloc(0, nullptr, nullptr, 0u);
 
     testing::internal::CaptureStdout();
 

opencl/test/unit_test/kernel/cache_flush_tests.inl

@@ -240,7 +240,7 @@ class GivenCacheFlushAfterWalkerEnabledAndProperSteppingIsSetWhenAllocationRequi
 
         mockKernel.kernelInfo.kernelAllocation = memoryManager->allocateGraphicsMemoryWithProperties(MockAllocationProperties(pDevice->getRootDeviceIndex(), true, MemoryConstants::pageSize, AllocationType::INTERNAL_HEAP));
         mockKernel.mockKernel->kernelArgRequiresCacheFlush.resize(1);
-        mockKernel.mockKernel->setArgSvmAlloc(0, svm, svmAllocation);
+        mockKernel.mockKernel->setArgSvmAlloc(0, svm, svmAllocation, 0u);
 
         cmdQ->getUltCommandStreamReceiver().timestampPacketWriteEnabled = false;
 
@@ -295,7 +295,7 @@ class GivenCacheFlushAfterWalkerAndTimestampPacketsEnabledWhenAllocationRequires
 
         mockKernel.kernelInfo.kernelAllocation = memoryManager->allocateGraphicsMemoryWithProperties(MockAllocationProperties(pDevice->getRootDeviceIndex(), true, MemoryConstants::pageSize, AllocationType::INTERNAL_HEAP));
         mockKernel.mockKernel->kernelArgRequiresCacheFlush.resize(1);
-        mockKernel.mockKernel->setArgSvmAlloc(0, svm, svmAllocation);
+        mockKernel.mockKernel->setArgSvmAlloc(0, svm, svmAllocation, 0);
 
         cmdQ->enqueueKernel(mockKernel, 1, nullptr, StatickSize3<16, 1, 1>(), StatickSize3<16, 1, 1>(), 0, nullptr, nullptr);
 
@@ -353,7 +353,7 @@ class GivenCacheFlushAfterWalkerDisabledAndProperSteppingIsSetWhenAllocationRequ
 
         mockKernel.kernelInfo.kernelAllocation = memoryManager->allocateGraphicsMemoryWithProperties(MockAllocationProperties(pDevice->getRootDeviceIndex(), true, MemoryConstants::pageSize, AllocationType::INTERNAL_HEAP));
         mockKernel.mockKernel->kernelArgRequiresCacheFlush.resize(1);
-        mockKernel.mockKernel->setArgSvmAlloc(0, svm, svmAllocation);
+        mockKernel.mockKernel->setArgSvmAlloc(0, svm, svmAllocation, 0u);
 
         cmdQ->enqueueKernel(mockKernel, 1, nullptr, StatickSize3<16, 1, 1>(), StatickSize3<16, 1, 1>(), 0, nullptr, nullptr);
 

opencl/test/unit_test/kernel/cache_flush_xehp_and_later_tests.inl

@@ -266,7 +266,7 @@ class GivenCacheFlushAfterWalkerEnabledWhenAllocationRequiresCacheFlushThenFlush
 
         mockKernel.kernelInfo.kernelAllocation = memoryManager->allocateGraphicsMemoryWithProperties(MockAllocationProperties(pDevice->getRootDeviceIndex(), true, MemoryConstants::pageSize, AllocationType::INTERNAL_HEAP, pDevice->getDeviceBitfield()));
         mockKernel.mockKernel->kernelArgRequiresCacheFlush.resize(1);
-        mockKernel.mockKernel->setArgSvmAlloc(0, svm, svmAllocation);
+        mockKernel.mockKernel->setArgSvmAlloc(0, svm, svmAllocation, 0u);
 
         cmdQ->enqueueKernel(mockKernel, 1, nullptr, StatickSize3<16, 1, 1>(), StatickSize3<16, 1, 1>(), 0, nullptr, nullptr);
 
@@ -319,7 +319,7 @@ class GivenCacheFlushAfterWalkerAndTimestampPacketsEnabledWhenAllocationRequires
 
         mockKernel.kernelInfo.kernelAllocation = memoryManager->allocateGraphicsMemoryWithProperties(MockAllocationProperties(pDevice->getRootDeviceIndex(), true, MemoryConstants::pageSize, AllocationType::INTERNAL_HEAP, pDevice->getDeviceBitfield()));
         mockKernel.mockKernel->kernelArgRequiresCacheFlush.resize(1);
-        mockKernel.mockKernel->setArgSvmAlloc(0, svm, svmAllocation);
+        mockKernel.mockKernel->setArgSvmAlloc(0, svm, svmAllocation, 0u);
 
         cmdQ->enqueueKernel(mockKernel, 1, nullptr, StatickSize3<16, 1, 1>(), StatickSize3<16, 1, 1>(), 0, nullptr, nullptr);
 
@@ -375,7 +375,7 @@ class GivenCacheFlushAfterWalkerDisabledWhenAllocationRequiresCacheFlushThenFlus
 
         mockKernel.kernelInfo.kernelAllocation = memoryManager->allocateGraphicsMemoryWithProperties(MockAllocationProperties(pDevice->getRootDeviceIndex(), true, MemoryConstants::pageSize, AllocationType::INTERNAL_HEAP, pDevice->getDeviceBitfield()));
         mockKernel.mockKernel->kernelArgRequiresCacheFlush.resize(1);
-        mockKernel.mockKernel->setArgSvmAlloc(0, svm, svmAllocation);
+        mockKernel.mockKernel->setArgSvmAlloc(0, svm, svmAllocation, 0u);
 
         cmdQ->enqueueKernel(mockKernel, 1, nullptr, StatickSize3<16, 1, 1>(), StatickSize3<16, 1, 1>(), 0, nullptr, nullptr);
 

opencl/test/unit_test/kernel/clone_kernel_tests.cpp

@@ -129,6 +129,7 @@ TEST_F(CloneKernelTest, GivenUnsetArgWhenCloningKernelThenKernelInfoIsCorrect) {
         EXPECT_EQ(0u, pSourceKernel[rootDeviceIndex]->getKernelArgInfo(0).size);
         EXPECT_EQ(0u, pSourceKernel[rootDeviceIndex]->getPatchedArgumentsNum());
         EXPECT_FALSE(pSourceKernel[rootDeviceIndex]->getKernelArgInfo(0).isPatched);
+        EXPECT_EQ(0u, pSourceKernel[rootDeviceIndex]->getKernelArgInfo(0).allocId);
     }
 
     retVal = pClonedMultiDeviceKernel->cloneKernel(pSourceMultiDeviceKernel.get());
@@ -142,6 +143,7 @@ TEST_F(CloneKernelTest, GivenUnsetArgWhenCloningKernelThenKernelInfoIsCorrect) {
         EXPECT_EQ(pSourceKernel[rootDeviceIndex]->getKernelArgInfo(0).size, pClonedKernel[rootDeviceIndex]->getKernelArgInfo(0).size);
         EXPECT_EQ(pSourceKernel[rootDeviceIndex]->getPatchedArgumentsNum(), pClonedKernel[rootDeviceIndex]->getPatchedArgumentsNum());
         EXPECT_EQ(pSourceKernel[rootDeviceIndex]->getKernelArgInfo(0).isPatched, pClonedKernel[rootDeviceIndex]->getKernelArgInfo(0).isPatched);
+        EXPECT_EQ(pSourceKernel[rootDeviceIndex]->getKernelArgInfo(0).allocId, pClonedKernel[rootDeviceIndex]->getKernelArgInfo(0).allocId);
     }
 }
 
@@ -162,6 +164,7 @@ TEST_F(CloneKernelTest, GivenArgLocalWhenCloningKernelThenKernelInfoIsCorrect) {
         EXPECT_NE(0u, pSourceKernel[rootDeviceIndex]->getKernelArgInfo(0).size);
         EXPECT_EQ(1u, pSourceKernel[rootDeviceIndex]->getPatchedArgumentsNum());
         EXPECT_TRUE(pSourceKernel[rootDeviceIndex]->getKernelArgInfo(0).isPatched);
+        EXPECT_EQ(0u, pSourceKernel[rootDeviceIndex]->getKernelArgInfo(0).allocId);
     }
 
     retVal = pClonedMultiDeviceKernel->cloneKernel(pSourceMultiDeviceKernel.get());
@@ -175,6 +178,7 @@ TEST_F(CloneKernelTest, GivenArgLocalWhenCloningKernelThenKernelInfoIsCorrect) {
         EXPECT_EQ(pSourceKernel[rootDeviceIndex]->getKernelArgInfo(0).size, pClonedKernel[rootDeviceIndex]->getKernelArgInfo(0).size);
         EXPECT_EQ(pSourceKernel[rootDeviceIndex]->getPatchedArgumentsNum(), pClonedKernel[rootDeviceIndex]->getPatchedArgumentsNum());
         EXPECT_EQ(pSourceKernel[rootDeviceIndex]->getKernelArgInfo(0).isPatched, pClonedKernel[rootDeviceIndex]->getKernelArgInfo(0).isPatched);
+        EXPECT_EQ(pSourceKernel[rootDeviceIndex]->getKernelArgInfo(0).allocId, pClonedKernel[rootDeviceIndex]->getKernelArgInfo(0).allocId);
 
         EXPECT_EQ(alignUp(slmSize, 1024), pClonedKernel[rootDeviceIndex]->slmTotalSize);
     }
@@ -200,6 +204,7 @@ TEST_F(CloneKernelTest, GivenArgBufferWhenCloningKernelThenKernelInfoIsCorrect)
         EXPECT_NE(0u, pSourceKernel[rootDeviceIndex]->getKernelArgInfo(0).size);
         EXPECT_EQ(1u, pSourceKernel[rootDeviceIndex]->getPatchedArgumentsNum());
         EXPECT_TRUE(pSourceKernel[rootDeviceIndex]->getKernelArgInfo(0).isPatched);
+        EXPECT_EQ(0u, pSourceKernel[rootDeviceIndex]->getKernelArgInfo(0).allocId);
     }
 
     retVal = pClonedMultiDeviceKernel->cloneKernel(pSourceMultiDeviceKernel.get());
@@ -213,6 +218,7 @@ TEST_F(CloneKernelTest, GivenArgBufferWhenCloningKernelThenKernelInfoIsCorrect)
         EXPECT_EQ(pSourceKernel[rootDeviceIndex]->getKernelArgInfo(0).size, pClonedKernel[rootDeviceIndex]->getKernelArgInfo(0).size);
         EXPECT_EQ(pSourceKernel[rootDeviceIndex]->getPatchedArgumentsNum(), pClonedKernel[rootDeviceIndex]->getPatchedArgumentsNum());
         EXPECT_EQ(pSourceKernel[rootDeviceIndex]->getKernelArgInfo(0).isPatched, pClonedKernel[rootDeviceIndex]->getKernelArgInfo(0).isPatched);
+        EXPECT_EQ(pSourceKernel[rootDeviceIndex]->getKernelArgInfo(0).allocId, pClonedKernel[rootDeviceIndex]->getKernelArgInfo(0).allocId);
 
         auto pKernelArg = (cl_mem *)(pClonedKernel[rootDeviceIndex]->getCrossThreadData() +
                                      pClonedKernel[rootDeviceIndex]->getKernelInfo().getArgDescriptorAt(0).as<ArgDescPointer>().stateless);
@@ -240,6 +246,7 @@ TEST_F(CloneKernelTest, GivenArgPipeWhenCloningKernelThenKernelInfoIsCorrect) {
     EXPECT_NE(0u, pSourceKernel[rootDeviceIndex]->getKernelArgInfo(0).size);
     EXPECT_EQ(1u, pSourceKernel[rootDeviceIndex]->getPatchedArgumentsNum());
     EXPECT_TRUE(pSourceKernel[rootDeviceIndex]->getKernelArgInfo(0).isPatched);
+    EXPECT_EQ(0u, pSourceKernel[rootDeviceIndex]->getKernelArgInfo(0).allocId);
 
     retVal = pClonedMultiDeviceKernel->cloneKernel(pSourceMultiDeviceKernel.get());
     EXPECT_EQ(CL_SUCCESS, retVal);
@@ -251,6 +258,7 @@ TEST_F(CloneKernelTest, GivenArgPipeWhenCloningKernelThenKernelInfoIsCorrect) {
     EXPECT_EQ(pSourceKernel[rootDeviceIndex]->getKernelArgInfo(0).size, pClonedKernel[rootDeviceIndex]->getKernelArgInfo(0).size);
     EXPECT_EQ(pSourceKernel[rootDeviceIndex]->getPatchedArgumentsNum(), pClonedKernel[rootDeviceIndex]->getPatchedArgumentsNum());
     EXPECT_EQ(pSourceKernel[rootDeviceIndex]->getKernelArgInfo(0).isPatched, pClonedKernel[rootDeviceIndex]->getKernelArgInfo(0).isPatched);
+    EXPECT_EQ(pSourceKernel[rootDeviceIndex]->getKernelArgInfo(0).allocId, pClonedKernel[rootDeviceIndex]->getKernelArgInfo(0).allocId);
 
     auto pKernelArg = (cl_mem *)(pClonedKernel[rootDeviceIndex]->getCrossThreadData() +
                                  pClonedKernel[rootDeviceIndex]->getKernelInfo().getArgDescriptorAt(0).as<ArgDescPointer>().stateless);
@@ -286,6 +294,7 @@ TEST_F(CloneKernelTest, GivenArgImageWhenCloningKernelThenKernelInfoIsCorrect) {
     EXPECT_NE(0u, pSourceKernel[rootDeviceIndex]->getKernelArgInfo(0).size);
     EXPECT_EQ(1u, pSourceKernel[rootDeviceIndex]->getPatchedArgumentsNum());
     EXPECT_TRUE(pSourceKernel[rootDeviceIndex]->getKernelArgInfo(0).isPatched);
+    EXPECT_EQ(0u, pSourceKernel[rootDeviceIndex]->getKernelArgInfo(0).allocId);
 
     retVal = pClonedMultiDeviceKernel->cloneKernel(pSourceMultiDeviceKernel.get());
     EXPECT_EQ(CL_SUCCESS, retVal);
@@ -297,6 +306,7 @@ TEST_F(CloneKernelTest, GivenArgImageWhenCloningKernelThenKernelInfoIsCorrect) {
     EXPECT_EQ(pSourceKernel[rootDeviceIndex]->getKernelArgInfo(0).size, pClonedKernel[rootDeviceIndex]->getKernelArgInfo(0).size);
     EXPECT_EQ(pSourceKernel[rootDeviceIndex]->getPatchedArgumentsNum(), pClonedKernel[rootDeviceIndex]->getPatchedArgumentsNum());
     EXPECT_EQ(pSourceKernel[rootDeviceIndex]->getKernelArgInfo(0).isPatched, pClonedKernel[rootDeviceIndex]->getKernelArgInfo(0).isPatched);
+    EXPECT_EQ(pSourceKernel[rootDeviceIndex]->getKernelArgInfo(0).allocId, pClonedKernel[rootDeviceIndex]->getKernelArgInfo(0).allocId);
 
     auto crossThreadData = reinterpret_cast<uint32_t *>(pClonedKernel[rootDeviceIndex]->getCrossThreadData());
     auto &clonedArg = pClonedKernel[rootDeviceIndex]->getKernelInfo().getArgDescriptorAt(0).as<ArgDescImage>();
@@ -340,6 +350,7 @@ TEST_F(CloneKernelTest, GivenArgAcceleratorWhenCloningKernelThenKernelInfoIsCorr
     EXPECT_NE(0u, pSourceKernel[rootDeviceIndex]->getKernelArgInfo(0).size);
     EXPECT_EQ(1u, pSourceKernel[rootDeviceIndex]->getPatchedArgumentsNum());
     EXPECT_TRUE(pSourceKernel[rootDeviceIndex]->getKernelArgInfo(0).isPatched);
+    EXPECT_EQ(0u, pSourceKernel[rootDeviceIndex]->getKernelArgInfo(0).allocId);
 
     retVal = pClonedMultiDeviceKernel->cloneKernel(pSourceMultiDeviceKernel.get());
     EXPECT_EQ(CL_SUCCESS, retVal);
@@ -351,6 +362,7 @@ TEST_F(CloneKernelTest, GivenArgAcceleratorWhenCloningKernelThenKernelInfoIsCorr
     EXPECT_EQ(pSourceKernel[rootDeviceIndex]->getKernelArgInfo(0).size, pClonedKernel[rootDeviceIndex]->getKernelArgInfo(0).size);
     EXPECT_EQ(pSourceKernel[rootDeviceIndex]->getPatchedArgumentsNum(), pClonedKernel[rootDeviceIndex]->getPatchedArgumentsNum());
     EXPECT_EQ(pSourceKernel[rootDeviceIndex]->getKernelArgInfo(0).isPatched, pClonedKernel[rootDeviceIndex]->getKernelArgInfo(0).isPatched);
+    EXPECT_EQ(pSourceKernel[rootDeviceIndex]->getKernelArgInfo(0).allocId, pClonedKernel[rootDeviceIndex]->getKernelArgInfo(0).allocId);
 
     auto crossThreadData = reinterpret_cast<uint32_t *>(pClonedKernel[rootDeviceIndex]->getCrossThreadData());
     ASSERT_TRUE(pClonedKernel[rootDeviceIndex]->getKernelInfo().getArgDescriptorAt(0).getExtendedTypeInfo().hasVmeExtendedDescriptor);
@@ -395,6 +407,7 @@ TEST_F(CloneKernelTest, GivenArgSamplerWhenCloningKernelThenKernelInfoIsCorrect)
     EXPECT_NE(0u, pSourceKernel[rootDeviceIndex]->getKernelArgInfo(0).size);
     EXPECT_EQ(1u, pSourceKernel[rootDeviceIndex]->getPatchedArgumentsNum());
     EXPECT_TRUE(pSourceKernel[rootDeviceIndex]->getKernelArgInfo(0).isPatched);
+    EXPECT_EQ(0u, pSourceKernel[rootDeviceIndex]->getKernelArgInfo(0).allocId);
 
     retVal = pClonedMultiDeviceKernel->cloneKernel(pSourceMultiDeviceKernel.get());
     EXPECT_EQ(CL_SUCCESS, retVal);
@@ -406,6 +419,7 @@ TEST_F(CloneKernelTest, GivenArgSamplerWhenCloningKernelThenKernelInfoIsCorrect)
     EXPECT_EQ(pSourceKernel[rootDeviceIndex]->getKernelArgInfo(0).size, pClonedKernel[rootDeviceIndex]->getKernelArgInfo(0).size);
     EXPECT_EQ(pSourceKernel[rootDeviceIndex]->getPatchedArgumentsNum(), pClonedKernel[rootDeviceIndex]->getPatchedArgumentsNum());
     EXPECT_EQ(pSourceKernel[rootDeviceIndex]->getKernelArgInfo(0).isPatched, pClonedKernel[rootDeviceIndex]->getKernelArgInfo(0).isPatched);
+    EXPECT_EQ(pSourceKernel[rootDeviceIndex]->getKernelArgInfo(0).allocId, pClonedKernel[rootDeviceIndex]->getKernelArgInfo(0).allocId);
 
     auto crossThreadData = reinterpret_cast<uint32_t *>(pClonedKernel[rootDeviceIndex]->getCrossThreadData());
     const auto &clonedArg = pClonedKernel[rootDeviceIndex]->getKernelInfo().getArgDescriptorAt(0).as<ArgDescSampler>();
@@ -436,6 +450,7 @@ TEST_F(CloneKernelTest, GivenArgSvmWhenCloningKernelThenKernelInfoIsCorrect) {
         EXPECT_NE(0u, pSourceKernel[rootDeviceIndex]->getKernelArgInfo(0).size);
         EXPECT_EQ(1u, pSourceKernel[rootDeviceIndex]->getPatchedArgumentsNum());
         EXPECT_TRUE(pSourceKernel[rootDeviceIndex]->getKernelArgInfo(0).isPatched);
+        EXPECT_EQ(0u, pSourceKernel[rootDeviceIndex]->getKernelArgInfo(0).allocId);
     }
 
     retVal = pClonedMultiDeviceKernel->cloneKernel(pSourceMultiDeviceKernel.get());
@@ -449,6 +464,7 @@ TEST_F(CloneKernelTest, GivenArgSvmWhenCloningKernelThenKernelInfoIsCorrect) {
         EXPECT_EQ(pSourceKernel[rootDeviceIndex]->getKernelArgInfo(0).size, pClonedKernel[rootDeviceIndex]->getKernelArgInfo(0).size);
         EXPECT_EQ(pSourceKernel[rootDeviceIndex]->getPatchedArgumentsNum(), pClonedKernel[rootDeviceIndex]->getPatchedArgumentsNum());
         EXPECT_EQ(pSourceKernel[rootDeviceIndex]->getKernelArgInfo(0).isPatched, pClonedKernel[rootDeviceIndex]->getKernelArgInfo(0).isPatched);
+        EXPECT_EQ(pSourceKernel[rootDeviceIndex]->getKernelArgInfo(0).allocId, pClonedKernel[rootDeviceIndex]->getKernelArgInfo(0).allocId);
 
         auto pKernelArg = (void **)(pClonedKernel[rootDeviceIndex]->getCrossThreadData() +
                                     pClonedKernel[rootDeviceIndex]->getKernelInfo().getArgDescriptorAt(0).as<ArgDescPointer>().stateless);
@@ -468,7 +484,7 @@ TEST_F(CloneKernelTest, GivenArgSvmAllocWhenCloningKernelThenKernelInfoIsCorrect
         multiGraphicsAllocation.addAllocation(svmAlloc);
     }
 
-    retVal = pSourceMultiDeviceKernel->setArgSvmAlloc(0, memory, &multiGraphicsAllocation);
+    retVal = pSourceMultiDeviceKernel->setArgSvmAlloc(0, memory, &multiGraphicsAllocation, 1u);
     ASSERT_EQ(CL_SUCCESS, retVal);
 
     for (auto &rootDeviceIndex : this->context->getRootDeviceIndices()) {
@@ -478,6 +494,7 @@ TEST_F(CloneKernelTest, GivenArgSvmAllocWhenCloningKernelThenKernelInfoIsCorrect
         EXPECT_NE(0u, pSourceKernel[rootDeviceIndex]->getKernelArgInfo(0).size);
         EXPECT_EQ(1u, pSourceKernel[rootDeviceIndex]->getPatchedArgumentsNum());
         EXPECT_TRUE(pSourceKernel[rootDeviceIndex]->getKernelArgInfo(0).isPatched);
+        EXPECT_EQ(1u, pSourceKernel[rootDeviceIndex]->getKernelArgInfo(0).allocId);
     }
 
     retVal = pClonedMultiDeviceKernel->cloneKernel(pSourceMultiDeviceKernel.get());
@@ -491,6 +508,7 @@ TEST_F(CloneKernelTest, GivenArgSvmAllocWhenCloningKernelThenKernelInfoIsCorrect
         EXPECT_EQ(pSourceKernel[rootDeviceIndex]->getKernelArgInfo(0).size, pClonedKernel[rootDeviceIndex]->getKernelArgInfo(0).size);
         EXPECT_EQ(pSourceKernel[rootDeviceIndex]->getPatchedArgumentsNum(), pClonedKernel[rootDeviceIndex]->getPatchedArgumentsNum());
         EXPECT_EQ(pSourceKernel[rootDeviceIndex]->getKernelArgInfo(0).isPatched, pClonedKernel[rootDeviceIndex]->getKernelArgInfo(0).isPatched);
+        EXPECT_EQ(pSourceKernel[rootDeviceIndex]->getKernelArgInfo(0).allocId, pClonedKernel[rootDeviceIndex]->getKernelArgInfo(0).allocId);
 
         auto pKernelArg = (void **)(pClonedKernel[rootDeviceIndex]->getCrossThreadData() +
                                     pClonedKernel[rootDeviceIndex]->getKernelInfo().getArgDescriptorAt(0).as<ArgDescPointer>().stateless);
@@ -514,6 +532,7 @@ TEST_F(CloneKernelTest, GivenArgImmediateWhenCloningKernelThenKernelInfoIsCorrec
         EXPECT_NE(0u, pSourceKernel[rootDeviceIndex]->getKernelArgInfo(0).size);
         EXPECT_EQ(1u, pSourceKernel[rootDeviceIndex]->getPatchedArgumentsNum());
         EXPECT_TRUE(pSourceKernel[rootDeviceIndex]->getKernelArgInfo(0).isPatched);
+        EXPECT_EQ(0u, pSourceKernel[rootDeviceIndex]->getKernelArgInfo(0).allocId);
     }
 
     retVal = pClonedMultiDeviceKernel->cloneKernel(pSourceMultiDeviceKernel.get());
@@ -527,6 +546,7 @@ TEST_F(CloneKernelTest, GivenArgImmediateWhenCloningKernelThenKernelInfoIsCorrec
         EXPECT_EQ(pSourceKernel[rootDeviceIndex]->getKernelArgInfo(0).size, pClonedKernel[rootDeviceIndex]->getKernelArgInfo(0).size);
         EXPECT_EQ(pSourceKernel[rootDeviceIndex]->getPatchedArgumentsNum(), pClonedKernel[rootDeviceIndex]->getPatchedArgumentsNum());
         EXPECT_EQ(pSourceKernel[rootDeviceIndex]->getKernelArgInfo(0).isPatched, pClonedKernel[rootDeviceIndex]->getKernelArgInfo(0).isPatched);
+        EXPECT_EQ(pSourceKernel[rootDeviceIndex]->getKernelArgInfo(0).allocId, pClonedKernel[rootDeviceIndex]->getKernelArgInfo(0).allocId);
 
         auto pKernelArg = (TypeParam *)(pClonedKernel[rootDeviceIndex]->getCrossThreadData() +
                                         pClonedKernel[rootDeviceIndex]->getKernelInfo().getArgDescriptorAt(0).as<ArgDescValue>().elements[0].offset);

opencl/test/unit_test/kernel/kernel_arg_buffer_tests.cpp

@@ -381,7 +381,7 @@ TEST_F(KernelArgBufferTest, givenGfxAllocationWhenHasDirectStatelessAccessToHost
     for (auto pureStatefulBufferAccess : {false, true}) {
         pKernelInfo->setBufferStateful(0, pureStatefulBufferAccess);
 
-        auto retVal = pKernel->setArgSvmAlloc(0, ptr, &gfxAllocation);
+        auto retVal = pKernel->setArgSvmAlloc(0, ptr, &gfxAllocation, 0u);
         EXPECT_EQ(CL_SUCCESS, retVal);
 
         EXPECT_FALSE(pKernel->hasDirectStatelessAccessToHostMemory());
@@ -397,7 +397,7 @@ TEST_F(KernelArgBufferTest, givenGfxAllocationInHostMemoryWhenHasDirectStateless
     for (auto pureStatefulBufferAccess : {false, true}) {
         pKernelInfo->setBufferStateful(0, pureStatefulBufferAccess);
 
-        auto retVal = pKernel->setArgSvmAlloc(0, ptr, &gfxAllocation);
+        auto retVal = pKernel->setArgSvmAlloc(0, ptr, &gfxAllocation, 0u);
         EXPECT_EQ(CL_SUCCESS, retVal);
 
         EXPECT_EQ(!pureStatefulBufferAccess, pKernel->hasDirectStatelessAccessToHostMemory());
@@ -557,7 +557,7 @@ TEST_F(KernelArgBufferTest, givenSetArgSvmAllocOnKernelWithDirectStatelessAccess
     MockGraphicsAllocation gfxAllocation(ptr, 128);
     gfxAllocation.setAllocationType(AllocationType::BUFFER_HOST_MEMORY);
 
-    auto retVal = pKernel->setArgSvmAlloc(0, ptr, &gfxAllocation);
+    auto retVal = pKernel->setArgSvmAlloc(0, ptr, &gfxAllocation, 0u);
     EXPECT_EQ(CL_SUCCESS, retVal);
 
     EXPECT_TRUE(pKernel->hasDirectStatelessAccessToHostMemory());
@@ -577,7 +577,7 @@ TEST_F(KernelArgBufferTest, givenSetArgSvmAllocOnKernelWithNoDirectStatelessAcce
     void *ptr = &data;
     MockGraphicsAllocation gfxAllocation(ptr, 128);
 
-    auto retVal = pKernel->setArgSvmAlloc(0, ptr, &gfxAllocation);
+    auto retVal = pKernel->setArgSvmAlloc(0, ptr, &gfxAllocation, 0u);
     EXPECT_EQ(CL_SUCCESS, retVal);
 
     EXPECT_FALSE(pKernel->hasDirectStatelessAccessToHostMemory());

opencl/test/unit_test/kernel/kernel_arg_svm_tests.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (C) 2018-2021 Intel Corporation
+ * Copyright (C) 2018-2022 Intel Corporation
  *
  * SPDX-License-Identifier: MIT
  *
@@ -110,7 +110,7 @@ TEST_F(KernelArgSvmTest, GivenValidSvmAllocWhenSettingKernelArgThenArgumentsAreS
 
     MockGraphicsAllocation svmAlloc(svmPtr, 256);
 
-    auto retVal = pKernel->setArgSvmAlloc(0, svmPtr, &svmAlloc);
+    auto retVal = pKernel->setArgSvmAlloc(0, svmPtr, &svmAlloc, 0u);
     EXPECT_EQ(CL_SUCCESS, retVal);
 
     auto pKernelArg = (void **)(pKernel->getCrossThreadData() +
@@ -126,7 +126,7 @@ TEST_F(KernelArgSvmTest, GivenSvmAllocWithUncacheableWhenSettingKernelArgThenKer
     MockGraphicsAllocation svmAlloc(svmPtr.get(), 256);
     svmAlloc.setUncacheable(true);
 
-    auto retVal = pKernel->setArgSvmAlloc(0, svmPtr.get(), &svmAlloc);
+    auto retVal = pKernel->setArgSvmAlloc(0, svmPtr.get(), &svmAlloc, 0u);
     EXPECT_EQ(CL_SUCCESS, retVal);
 
     EXPECT_TRUE(pKernel->hasUncacheableStatelessArgs());
@@ -138,13 +138,13 @@ TEST_F(KernelArgSvmTest, GivenSvmAllocWithoutUncacheableAndKenelWithUncachebleAr
     MockGraphicsAllocation svmAlloc(svmPtr.get(), 256);
     svmAlloc.setUncacheable(true);
 
-    auto retVal = pKernel->setArgSvmAlloc(0, svmPtr.get(), &svmAlloc);
+    auto retVal = pKernel->setArgSvmAlloc(0, svmPtr.get(), &svmAlloc, 0u);
     EXPECT_EQ(CL_SUCCESS, retVal);
     EXPECT_TRUE(pKernel->hasUncacheableStatelessArgs());
 
     svmAlloc.setUncacheable(false);
     pKernel->kernelArguments[0].isStatelessUncacheable = true;
-    retVal = pKernel->setArgSvmAlloc(0, svmPtr.get(), &svmAlloc);
+    retVal = pKernel->setArgSvmAlloc(0, svmPtr.get(), &svmAlloc, 0u);
     EXPECT_EQ(CL_SUCCESS, retVal);
     EXPECT_FALSE(pKernel->hasUncacheableStatelessArgs());
 }
@@ -155,7 +155,7 @@ HWTEST_F(KernelArgSvmTest, GivenValidSvmAllocStatefulWhenSettingKernelArgThenArg
     MockGraphicsAllocation svmAlloc(svmPtr, 256);
 
     pKernelInfo->argAsPtr(0).bindful = 0;
-    auto retVal = pKernel->setArgSvmAlloc(0, svmPtr, &svmAlloc);
+    auto retVal = pKernel->setArgSvmAlloc(0, svmPtr, &svmAlloc, 0u);
     EXPECT_EQ(CL_SUCCESS, retVal);
 
     EXPECT_NE(0u, pKernel->getSurfaceStateHeapSize());
@@ -179,7 +179,7 @@ HWTEST_F(KernelArgSvmTest, givenOffsetedSvmPointerWhenSetArgSvmAllocIsCalledThen
     MockGraphicsAllocation svmAlloc(svmPtr.get(), 256);
 
     pKernelInfo->argAsPtr(0).bindful = 0;
-    pKernel->setArgSvmAlloc(0, offsetedPtr, &svmAlloc);
+    pKernel->setArgSvmAlloc(0, offsetedPtr, &svmAlloc, 0u);
 
     typedef typename FamilyType::RENDER_SURFACE_STATE RENDER_SURFACE_STATE;
     auto surfaceState = reinterpret_cast<const RENDER_SURFACE_STATE *>(
@@ -196,7 +196,7 @@ HWTEST_F(KernelArgSvmTest, givenDeviceSupportingSharedSystemAllocationsWhenSetAr
     auto systemPointer = reinterpret_cast<void *>(0xfeedbac);
 
     pKernelInfo->argAsPtr(0).bindful = 0;
-    pKernel->setArgSvmAlloc(0, systemPointer, nullptr);
+    pKernel->setArgSvmAlloc(0, systemPointer, nullptr, 0u);
 
     typedef typename FamilyType::RENDER_SURFACE_STATE RENDER_SURFACE_STATE;
     auto surfaceState = reinterpret_cast<const RENDER_SURFACE_STATE *>(
@@ -329,7 +329,7 @@ struct SetArgHandlerSetArgSvm {
 
 struct SetArgHandlerSetArgSvmAlloc {
     static void setArg(Kernel &kernel, uint32_t argNum, void *ptrToPatch, size_t allocSize, GraphicsAllocation &alloc) {
-        kernel.setArgSvmAlloc(argNum, ptrToPatch, &alloc);
+        kernel.setArgSvmAlloc(argNum, ptrToPatch, &alloc, 0u);
     }
 
     static constexpr bool supportsOffsets() {
@@ -420,7 +420,7 @@ TEST_F(KernelArgSvmTest, givenWritableSvmAllocationWhenSettingAsArgThenDoNotExpe
     svmAlloc.setMemObjectsAllocationWithWritableFlags(true);
     svmAlloc.setFlushL3Required(false);
 
-    auto retVal = pKernel->setArgSvmAlloc(0, svmPtr, &svmAlloc);
+    auto retVal = pKernel->setArgSvmAlloc(0, svmPtr, &svmAlloc, 0u);
     EXPECT_EQ(CL_SUCCESS, retVal);
     EXPECT_EQ(nullptr, pKernel->kernelArgRequiresCacheFlush[0]);
 
@@ -435,7 +435,7 @@ TEST_F(KernelArgSvmTest, givenCacheFlushSvmAllocationWhenSettingAsArgThenExpectA
     svmAlloc.setMemObjectsAllocationWithWritableFlags(false);
     svmAlloc.setFlushL3Required(true);
 
-    auto retVal = pKernel->setArgSvmAlloc(0, svmPtr, &svmAlloc);
+    auto retVal = pKernel->setArgSvmAlloc(0, svmPtr, &svmAlloc, 0u);
     EXPECT_EQ(CL_SUCCESS, retVal);
     EXPECT_EQ(&svmAlloc, pKernel->kernelArgRequiresCacheFlush[0]);
 
@@ -450,7 +450,7 @@ TEST_F(KernelArgSvmTest, givenNoCacheFlushSvmAllocationWhenSettingAsArgThenNotEx
     svmAlloc.setMemObjectsAllocationWithWritableFlags(false);
     svmAlloc.setFlushL3Required(false);
 
-    auto retVal = pKernel->setArgSvmAlloc(0, svmPtr, &svmAlloc);
+    auto retVal = pKernel->setArgSvmAlloc(0, svmPtr, &svmAlloc, 0u);
     EXPECT_EQ(CL_SUCCESS, retVal);
     EXPECT_EQ(nullptr, pKernel->kernelArgRequiresCacheFlush[0]);
 
@@ -506,7 +506,7 @@ TEST_F(KernelArgSvmTest, givenCpuAddressIsNullWhenGpuAddressIsValidThenExpectSvm
 
     MockGraphicsAllocation svmAlloc(nullptr, reinterpret_cast<uint64_t>(svmPtr), 256);
 
-    auto retVal = pKernel->setArgSvmAlloc(0, svmPtr, &svmAlloc);
+    auto retVal = pKernel->setArgSvmAlloc(0, svmPtr, &svmAlloc, 0u);
     EXPECT_EQ(CL_SUCCESS, retVal);
 
     auto pKernelArg = (void **)(pKernel->getCrossThreadData() +

opencl/test/unit_test/kernel/kernel_is_patched_tests.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (C) 2018-2021 Intel Corporation
+ * Copyright (C) 2018-2022 Intel Corporation
  *
  * SPDX-License-Identifier: MIT
  *
@@ -66,7 +66,7 @@ TEST_F(PatchedKernelTest, givenKernelWithoutAllArgsSetWhenIsPatchedIsCalledThenR
 
 TEST_F(PatchedKernelTest, givenArgSvmAllocWhenArgIsSetThenArgIsPatched) {
     EXPECT_FALSE(kernel->getKernelArguments()[0].isPatched);
-    kernel->setArgSvmAlloc(0, nullptr, nullptr);
+    kernel->setArgSvmAlloc(0, nullptr, nullptr, 0u);
     EXPECT_TRUE(kernel->getKernelArguments()[0].isPatched);
 }
 

opencl/test/unit_test/mem_obj/buffer_set_arg_tests.cpp

@@ -273,7 +273,8 @@ TEST_F(BufferSetArgTest, GivenSvmPointerWhenSettingKernelArgThenAddressToPatchIs
     retVal = pKernel->setArgSvmAlloc(
         0,
         ptrSVM,
-        pSvmAlloc);
+        pSvmAlloc,
+        0u);
     ASSERT_EQ(CL_SUCCESS, retVal);
 
     auto pKernelArg = (void **)(pKernel->getCrossThreadData() +

opencl/test/unit_test/mocks/mock_kernel.cpp

@@ -44,4 +44,9 @@ bool MockKernel::requiresCacheFlushCommand(const CommandQueue &commandQueue) con
 
     return false;
 }
+
+cl_int MockKernel::setArgSvmAlloc(uint32_t argIndex, void *svmPtr, GraphicsAllocation *svmAlloc, uint32_t allocId) {
+    ++setArgSvmAllocCalls;
+    return Kernel::setArgSvmAlloc(argIndex, svmPtr, svmAlloc, allocId);
+}
 } // namespace NEO

opencl/test/unit_test/mocks/mock_kernel.h

@@ -245,8 +245,11 @@ class MockKernel : public Kernel {
 
     bool requiresCacheFlushCommand(const CommandQueue &commandQueue) const override;
 
+    cl_int setArgSvmAlloc(uint32_t argIndex, void *svmPtr, GraphicsAllocation *svmAlloc, uint32_t allocId) override;
+
     uint32_t makeResidentCalls = 0;
     uint32_t getResidencyCalls = 0;
+    uint32_t setArgSvmAllocCalls = 0;
 
     bool canKernelTransformImages = true;
     bool isPatchedOverride = true;