Merge remote-tracking branch 'intel_llvm/sycl' into esimd_L1_L2_rewor…

…k_L3_assert
intel · Mar 5, 2024 · f44dd13 · f44dd13
2 parents 5fc647c + f8abcb7
commit f44dd13
Show file tree

Hide file tree

Showing 11 changed files with 36 additions and 27 deletions.
diff --git a/clang/lib/Driver/ToolChains/SYCL.cpp b/clang/lib/Driver/ToolChains/SYCL.cpp
@@ -371,6 +371,8 @@ static llvm::SmallVector<StringRef, 16> SYCLDeviceLibList {
   "bfloat16", "crt", "cmath", "cmath-fp64", "complex", "complex-fp64",
 #if defined(_WIN32)
       "msvc-math",
+#else
+      "sanitizer",
 #endif
       "imf", "imf-fp64", "itt-compiler-wrappers", "itt-stubs",
       "itt-user-wrappers", "fallback-cassert", "fallback-cstring",

diff --git a/clang/test/Driver/sycl-device-lib.cpp b/clang/test/Driver/sycl-device-lib.cpp
@@ -36,6 +36,7 @@
 // SYCL_DEVICE_LIB_UNBUNDLE_DEFAULT-NEXT: clang-offload-bundler{{.*}} "-type=o" "-targets=sycl-spir64-unknown-unknown" "-input={{.*}}libsycl-fallback-imf.o" "-output={{.*}}libsycl-fallback-imf-{{.*}}.o" "-unbundle"
 // SYCL_DEVICE_LIB_UNBUNDLE_DEFAULT-NEXT: clang-offload-bundler{{.*}} "-type=o" "-targets=sycl-spir64-unknown-unknown" "-input={{.*}}libsycl-fallback-imf-fp64.o" "-output={{.*}}libsycl-fallback-imf-fp64-{{.*}}.o" "-unbundle"
 // SYCL_DEVICE_LIB_UNBUNDLE_DEFAULT-NEXT: clang-offload-bundler{{.*}} "-type=o" "-targets=sycl-spir64-unknown-unknown" "-input={{.*}}libsycl-fallback-imf-bf16.o" "-output={{.*}}libsycl-fallback-imf-bf16-{{.*}}.o" "-unbundle"
+
 /// ###########################################################################
 /// test sycl fallback device libraries are not linked by default
 // RUN: %clangxx -fsycl -fsycl-device-lib-jit-link %s --sysroot=%S/Inputs/SYCL -### 2>&1 \
@@ -251,3 +252,4 @@
 // SYCL_DEVICE_ASAN_MACRO-SAME: "USE_SYCL_DEVICE_ASAN"
 // SYCL_DEVICE_ASAN_MACRO-NEXT: llvm-link{{.*}}
 // SYCL_DEVICE_ASAN_MACRO:  clang-offload-bundler{{.*}} "-type=o" "-targets=sycl-spir64-unknown-unknown" "-input={{.*}}libsycl-sanitizer.o" "-output={{.*}}libsycl-sanitizer-{{.*}}.o" "-unbundle"
+// SYCL_DEVICE_ASAN_MACRO: llvm-link{{.*}} "-only-needed" "{{.*}}" "-o" "{{.*}}.bc" "--suppress-warnings"
diff --git a/libdevice/include/spir_global_var.hpp b/libdevice/include/spir_global_var.hpp
@@ -11,15 +11,6 @@
 // Treat this header as system one to workaround frontend's restriction
 #pragma clang system_header
 
-#ifndef SPIR_GLOBAL_VAR
-#ifdef __SYCL_DEVICE_ONLY__
-#define SPIR_GLOBAL_VAR __attribute__((sycl_global_var))
-#else
-#warning "SPIR_GLOBAL_VAR not defined in host mode. Defining as empty macro."
-#define SPIR_GLOBAL_VAR
-#endif
-#endif
-
 #define __SYCL_GLOBAL__ __attribute__((opencl_global))
 #define __SYCL_LOCAL__ __attribute__((opencl_local))
 #define __SYCL_PRIVATE__ __attribute__((opencl_private))

diff --git a/sycl/include/sycl/ext/intel/experimental/grf_size_properties.hpp b/sycl/include/sycl/ext/intel/experimental/grf_size_properties.hpp
@@ -36,7 +36,7 @@ inline constexpr grf_size_key::value_t<Size> grf_size;
 inline constexpr grf_size_automatic_key::value_t grf_size_automatic;
 
 } // namespace ext::intel::experimental
-namespace ext::oneapi::experimental {
+namespace ext::oneapi::experimental::detail {
 template <unsigned int Size>
 struct PropertyMetaInfo<
     sycl::ext::intel::experimental::grf_size_key::value_t<Size>> {
@@ -79,7 +79,6 @@ struct ConflictingProperties<sycl::detail::register_alloc_mode_key, Properties>
               sycl::ext::intel::experimental::grf_size_automatic_key,
               Properties>::value> {};
 
-} // namespace detail
-} // namespace ext::oneapi::experimental
+} // namespace ext::oneapi::experimental::detail
 } // namespace _V1
 } // namespace sycl
diff --git a/sycl/source/detail/event_impl.hpp b/sycl/source/detail/event_impl.hpp
@@ -289,6 +289,10 @@ class event_impl {
     return MEventFromSubmittedExecCommandBuffer;
   }
 
+  const std::vector<EventImplPtr> &getPostCompleteEvents() const {
+    return MPostCompleteEvents;
+  }
+
 protected:
   // When instrumentation is enabled emits trace event for event wait begin and
   // returns the telemetry event generated for the wait

diff --git a/sycl/source/detail/graph_impl.cpp b/sycl/source/detail/graph_impl.cpp
@@ -764,7 +764,9 @@ exec_graph_impl::enqueue(const std::shared_ptr<sycl::detail::queue_impl> &Queue,
                          sycl::detail::CG::StorageInitHelper CGData) {
   WriteLock Lock(MMutex);
 
-  std::vector<sycl::detail::EventImplPtr> PartitionEvents;
+  // Map of the partitions to their execution events
+  std::unordered_map<std::shared_ptr<partition>, sycl::detail::EventImplPtr>
+      PartitionsExecutionEvents;
 
   auto CreateNewEvent([&]() {
     auto NewEvent = std::make_shared<sycl::detail::event_impl>(Queue);
@@ -787,7 +789,7 @@ exec_graph_impl::enqueue(const std::shared_ptr<sycl::detail::queue_impl> &Queue,
     }
 
     for (auto const &DepPartition : CurrentPartition->MPredecessors) {
-      CGData.MEvents.push_back(MPartitionsExecutionEvents[DepPartition]);
+      CGData.MEvents.push_back(PartitionsExecutionEvents[DepPartition]);
     }
 
     auto CommandBuffer =
@@ -819,7 +821,13 @@ exec_graph_impl::enqueue(const std::shared_ptr<sycl::detail::queue_impl> &Queue,
               sycl::backend::ext_oneapi_level_zero) {
             Event->wait(Event);
           } else {
+            auto &AttachedEventsList = Event->getPostCompleteEvents();
+            CGData.MEvents.reserve(AttachedEventsList.size() + 1);
             CGData.MEvents.push_back(Event);
+            // Add events of the previous execution of all graph partitions.
+            for (auto &AttachedEvent : AttachedEventsList) {
+              CGData.MEvents.push_back(AttachedEvent);
+            }
           }
           ++It;
         } else {
@@ -929,15 +937,15 @@ exec_graph_impl::enqueue(const std::shared_ptr<sycl::detail::queue_impl> &Queue,
       NewEvent->setStateIncomplete();
       NewEvent->getPreparedDepsEvents() = ScheduledEvents;
     }
-    MPartitionsExecutionEvents[CurrentPartition] = NewEvent;
+    PartitionsExecutionEvents[CurrentPartition] = NewEvent;
   }
 
   // Keep track of this execution event so we can make sure it's completed in
   // the destructor.
   MExecutionEvents.push_back(NewEvent);
   // Attach events of previous partitions to ensure that when the returned event
   // is complete all execution associated with the graph have been completed.
-  for (auto const &Elem : MPartitionsExecutionEvents) {
+  for (auto const &Elem : PartitionsExecutionEvents) {
     if (Elem.second != NewEvent) {
       NewEvent->attachEventToComplete(Elem.second);
     }

diff --git a/sycl/source/detail/graph_impl.hpp b/sycl/source/detail/graph_impl.hpp
@@ -1190,9 +1190,6 @@ class exec_graph_impl {
   std::vector<sycl::detail::EventImplPtr> MExecutionEvents;
   /// List of the partitions that compose the exec graph.
   std::vector<std::shared_ptr<partition>> MPartitions;
-  /// Map of the partitions to their execution events
-  std::unordered_map<std::shared_ptr<partition>, sycl::detail::EventImplPtr>
-      MPartitionsExecutionEvents;
   /// Storage for copies of nodes from the original modifiable graph.
   std::vector<std::shared_ptr<node_impl>> MNodeStorage;
 };

diff --git a/sycl/test-e2e/Matrix/SG32/joint_matrix_bf16_fill_k_cache.cpp b/sycl/test-e2e/Matrix/SG32/joint_matrix_bf16_fill_k_cache.cpp
@@ -8,8 +8,11 @@
 // REQUIRES: matrix
 // REQUIRES-INTEL-DRIVER: lin: 27501, win: 101.4943
 
-// RUN: %{build} -o %t.out -ffp-model=precise
-// RUN: %{run} %t.out
+// RUN: %{build} -o %t_gpu.out -ffp-model=precise
+// RUN: %if gpu %{ %{run} %t_gpu.out %}
+
+// RUN: %{build}  -ffp-model=precise -o %t_cpu.out -DtM=16 -DtK=32 -DNCACHE1=32 -DKCACHE1=32
+// RUN: %if cpu %{ %{run} %t_cpu.out %}
 
 // -ffp-model=precise is added to not depend on compiler defaults.
 

diff --git a/sycl/test-e2e/Matrix/joint_matrix_bf16_fill_k_cache.cpp b/sycl/test-e2e/Matrix/joint_matrix_bf16_fill_k_cache.cpp
@@ -7,8 +7,11 @@
 //===----------------------------------------------------------------------===//
 // REQUIRES: matrix
 
-// RUN: %{build} -o %t.out -ffp-model=precise
-// RUN: %{run} %t.out
+// RUN: %{build} -o %t_gpu.out -ffp-model=precise
+// RUN: %if gpu %{ %{run} %t_gpu.out %}
+
+// RUN: %{build}  -ffp-model=precise -o %t_cpu.out -DtM=16 -DtK=32 -DNCACHE1=32 -DKCACHE1=32
+// RUN: %if cpu %{ %{run} %t_cpu.out %}
 
 // -ffp-model=precise is added to not depend on compiler defaults.
 

diff --git a/sycl/test-e2e/Matrix/joint_matrix_prefetch_impl.hpp b/sycl/test-e2e/Matrix/joint_matrix_prefetch_impl.hpp
@@ -69,15 +69,16 @@ int main() {
           .get_info<sycl::ext::oneapi::experimental::info::device::
                         matrix_combinations>();
   bool support_p = false;
-  for (int i = 0; i < combinations.size(); i++) {
-    if (combinations[i].atype == matrix_type::tf32) {
+  // joint_matrix_prefetch is not supported on DG2
+  for (unsigned int i = 0; i < combinations.size(); i++) {
+    if (combinations[i].nsize == 0 || combinations[i].nsize == 16) {
       support_p = true;
       break;
     }
   }
   if (!support_p) {
     std::cout << "Prefetch not supported on this device" << std::endl;
-    // Once the test is not marke as XFAIL, this should change to return 0;
+    // Once the test is not marked as XFAIL, this should change to return 0;
     return 1;
   }
   static constexpr size_t M = TM * 2;

diff --git a/sycl/test-e2e/Plugin/level_zero_usm_residency.cpp b/sycl/test-e2e/Plugin/level_zero_usm_residency.cpp
@@ -23,7 +23,6 @@ int main(int argc, char *argv[]) {
   // DEVICE: ---> piextUSMDeviceAlloc
   // DEVICE: ZE ---> zeMemAllocDevice
   // DEVICE: ZE ---> zeContextMakeMemoryResident
-  // DEVICE-NOT: ZE ---> zeContextMakeMemoryResident
 
   auto ptr2 = malloc_shared<int>(1, Q);
   // SHARED: ---> piextUSMSharedAlloc