Merge branch 'llvm' into review/yang/dsan_nullpointer

Author: AllanZyne

.github/workflows/benchmarks_compute.yml

@@ -143,8 +143,8 @@ jobs:
         --ci-defaults ${{matrix.adapter.sycl_config}}
         --cmake-opt="-DLLVM_INSTALL_UTILS=ON"
         --cmake-opt="-DSYCL_PI_TESTS=OFF"
-        --cmake-opt="-DSYCL_PI_UR_USE_FETCH_CONTENT=OFF"
-        --cmake-opt="-DSYCL_PI_UR_SOURCE_DIR=${{github.workspace}}/ur-repo/"
+        --cmake-opt="-DSYCL_UR_USE_FETCH_CONTENT=OFF"
+        --cmake-opt="-DSYCL_UR_SOURCE_DIR=${{github.workspace}}/ur-repo/"
         --cmake-opt=-DCMAKE_C_COMPILER_LAUNCHER=ccache
         --cmake-opt=-DCMAKE_CXX_COMPILER_LAUNCHER=ccache
 

.github/workflows/e2e_core.yml

@@ -130,8 +130,8 @@ jobs:
         --ci-defaults ${{matrix.adapter.config}}
         --cmake-opt="-DLLVM_INSTALL_UTILS=ON"
         --cmake-opt="-DSYCL_PI_TESTS=OFF"
-        --cmake-opt="-DSYCL_PI_UR_USE_FETCH_CONTENT=OFF"
-        --cmake-opt="-DSYCL_PI_UR_SOURCE_DIR=${{github.workspace}}/ur-repo/"
+        --cmake-opt="-DSYCL_UR_USE_FETCH_CONTENT=OFF"
+        --cmake-opt="-DSYCL_UR_SOURCE_DIR=${{github.workspace}}/ur-repo/"
         --cmake-opt=-DCMAKE_C_COMPILER_LAUNCHER=ccache
         --cmake-opt=-DCMAKE_CXX_COMPILER_LAUNCHER=ccache
 
@@ -157,7 +157,7 @@ jobs:
     - name: Setup SYCL variables
       run: |
         which clang++ sycl-ls
-        SYCL_PI_TRACE=-1 sycl-ls
+        SYCL_UR_TRACE=-1 sycl-ls
 
     - name: Build e2e tests
       run: >

.github/workflows/e2e_level_zero.yml

@@ -21,9 +21,9 @@ jobs:
       config: ""
       unit: "gpu"
       # Failing tests
-      xfail: "Config/select_device.cpp;DeviceCodeSplit/grf.cpp;ESIMD/grf.cpp;KernelAndProgram/target_register_alloc_mode.cpp;Matrix/SG32/get_coord_int8_matB.cpp;Matrix/get_coord_int8_matB.cpp;Matrix/joint_matrix_prefetch.cpp;Matrix/joint_matrix_rowmajorA_rowmajorB.cpp;Plugin/level_zero_barrier_optimization.cpp"
+      xfail: "DeviceCodeSplit/grf.cpp;ESIMD/grf.cpp;ESIMD/mask_expand_load.cpp;KernelAndProgram/target_register_alloc_mode.cpp;Matrix/SG32/get_coord_int8_matB.cpp;Matrix/get_coord_int8_matB.cpp;Matrix/joint_matrix_prefetch.cpp;Matrix/joint_matrix_rowmajorA_rowmajorB.cpp;  SYCL :: ESIMD/mask_expand_load.cpp;Matrix/SPVCooperativeMatrix/joint_matrix_bf16_fill_k_cache_OOB.cpp;Matrix/SPVCooperativeMatrix/joint_matrix_out_bounds.cpp;Matrix/SPVCooperativeMatrix/joint_matrix_prefetch.cpp;Matrix/SPVCooperativeMatrix/joint_matrix_rowmajorA_rowmajorB.cpp;Matrix/element_wise_all_ops_1d.cpp;Matrix/element_wise_all_ops_1d_cont.cpp;Matrix/element_wise_all_ops_scalar.cpp;Matrix/joint_matrix_bf16_fill_k_cache_OOB.cpp;Matrix/joint_matrix_bf16_fill_k_cache_prefetch.cpp;Matrix/joint_matrix_out_bounds.cpp;Matrix/joint_matrix_unaligned_k.cpp"
       # Flaky tests
-      filter_out: "ESIMD/named_barriers/loop_extended.cpp"
+      filter_out: "ESIMD/named_barriers/loop_extended.cpp;ESIMD/local_accessor_copy_to_from.cpp;"
       # These runners by default spawn upwards of 260 workers.
       # We also add a time out just in case some test hangs
       extra_lit_flags: "--param gpu-intel-pvc=True --param gpu-intel-pvc-1T=True -sv -j 100 --max-time=3600"

source/adapters/level_zero/CMakeLists.txt

@@ -179,11 +179,15 @@ if(UR_BUILD_ADAPTER_L0_V2)
         ${CMAKE_CURRENT_SOURCE_DIR}/common.hpp
         ${CMAKE_CURRENT_SOURCE_DIR}/device.hpp
         ${CMAKE_CURRENT_SOURCE_DIR}/platform.hpp
+        ${CMAKE_CURRENT_SOURCE_DIR}/program.hpp
+        ${CMAKE_CURRENT_SOURCE_DIR}/helpers/kernel_helpers.hpp
         ${CMAKE_CURRENT_SOURCE_DIR}/adapter.cpp
         ${CMAKE_CURRENT_SOURCE_DIR}/common.cpp
         ${CMAKE_CURRENT_SOURCE_DIR}/device.cpp
         ${CMAKE_CURRENT_SOURCE_DIR}/ur_interface_loader.cpp
         ${CMAKE_CURRENT_SOURCE_DIR}/platform.cpp
+        ${CMAKE_CURRENT_SOURCE_DIR}/program.cpp
+        ${CMAKE_CURRENT_SOURCE_DIR}/helpers/kernel_helpers.cpp
         ${CMAKE_CURRENT_SOURCE_DIR}/../../ur/ur.cpp
         # v2-only sources
         ${CMAKE_CURRENT_SOURCE_DIR}/v2/command_list_cache.hpp
@@ -194,8 +198,10 @@ if(UR_BUILD_ADAPTER_L0_V2)
         ${CMAKE_CURRENT_SOURCE_DIR}/v2/event_provider_normal.hpp
         ${CMAKE_CURRENT_SOURCE_DIR}/v2/event_provider.hpp
         ${CMAKE_CURRENT_SOURCE_DIR}/v2/event.hpp
+        ${CMAKE_CURRENT_SOURCE_DIR}/v2/kernel.hpp
         ${CMAKE_CURRENT_SOURCE_DIR}/v2/queue_api.hpp
         ${CMAKE_CURRENT_SOURCE_DIR}/v2/queue_immediate_in_order.hpp
+        ${CMAKE_CURRENT_SOURCE_DIR}/v2/usm.hpp
         ${CMAKE_CURRENT_SOURCE_DIR}/v2/api.cpp
         ${CMAKE_CURRENT_SOURCE_DIR}/v2/command_list_cache.cpp
         ${CMAKE_CURRENT_SOURCE_DIR}/v2/context.cpp
@@ -204,9 +210,11 @@ if(UR_BUILD_ADAPTER_L0_V2)
         ${CMAKE_CURRENT_SOURCE_DIR}/v2/event_provider_counter.cpp
         ${CMAKE_CURRENT_SOURCE_DIR}/v2/event_provider_normal.cpp
         ${CMAKE_CURRENT_SOURCE_DIR}/v2/event.cpp
+        ${CMAKE_CURRENT_SOURCE_DIR}/v2/kernel.cpp
         ${CMAKE_CURRENT_SOURCE_DIR}/v2/queue_api.cpp
         ${CMAKE_CURRENT_SOURCE_DIR}/v2/queue_create.cpp
         ${CMAKE_CURRENT_SOURCE_DIR}/v2/queue_immediate_in_order.cpp
+        ${CMAKE_CURRENT_SOURCE_DIR}/v2/usm.cpp
     )
 
     if(NOT WIN32)
@@ -220,6 +228,8 @@ if(UR_BUILD_ADAPTER_L0_V2)
         )
     endif()
 
+    target_compile_definitions(ur_adapter_level_zero_v2 PUBLIC UR_ADAPTER_LEVEL_ZERO_V2)
+
     # TODO: fix level_zero adapter conversion warnings
     target_compile_options(ur_adapter_level_zero_v2 PRIVATE
         $<$<CXX_COMPILER_ID:MSVC>:/wd4805 /wd4244>
@@ -245,6 +255,7 @@ if(UR_BUILD_ADAPTER_L0_V2)
 
     target_include_directories(ur_adapter_level_zero_v2 PRIVATE
         "${CMAKE_CURRENT_SOURCE_DIR}/../.."
+        "${CMAKE_CURRENT_SOURCE_DIR}/../../ur"
         LevelZeroLoader-Headers
     )
 endif()

source/adapters/level_zero/command_buffer.hpp

@@ -17,6 +17,7 @@
 #include "common.hpp"
 
 #include "context.hpp"
+#include "kernel.hpp"
 #include "queue.hpp"
 
 struct command_buffer_profiling_t {

source/adapters/level_zero/common.hpp

@@ -25,6 +25,8 @@
 
 #include <umf_pools/disjoint_pool_config_parser.hpp>
 
+#include "logger/ur_logger.hpp"
+
 struct _ur_platform_handle_t;
 
 static auto getUrResultString = [](ur_result_t Result) {
@@ -168,7 +170,7 @@ static auto getUrResultString = [](ur_result_t Result) {
   }
 };
 
-// Trace an internal PI call; returns in case of an error.
+// Trace an internal UR call; returns in case of an error.
 #define UR_CALL(Call)                                                          \
   {                                                                            \
     if (PrintTrace)                                                            \
@@ -180,6 +182,18 @@ static auto getUrResultString = [](ur_result_t Result) {
       return Result;                                                           \
   }
 
+// Trace an internal UR call; throw in case of an error.
+#define UR_CALL_THROWS(Call)                                                   \
+  {                                                                            \
+    if (PrintTrace)                                                            \
+      logger::always("UR ---> {}", #Call);                                     \
+    ur_result_t Result = (Call);                                               \
+    if (PrintTrace)                                                            \
+      logger::always("UR <--- {}({})", #Call, getUrResultString(Result));      \
+    if (Result != UR_RESULT_SUCCESS)                                           \
+      throw Result;                                                            \
+  }
+
 // Controls UR L0 calls tracing.
 enum UrDebugLevel {
   UR_L0_DEBUG_NONE = 0x0,

source/adapters/level_zero/context.cpp

@@ -829,3 +829,12 @@ bool ur_context_handle_t_::isValidDevice(ur_device_handle_t Device) const {
   }
   return false;
 }
+
+const std::vector<ur_device_handle_t> &
+ur_context_handle_t_::getDevices() const {
+  return Devices;
+}
+
+ze_context_handle_t ur_context_handle_t_::getZeHandle() const {
+  return ZeContext;
+}

source/adapters/level_zero/context.hpp

@@ -193,6 +193,9 @@ struct ur_context_handle_t_ : _ur_object {
   // Return the Platform, which is the same for all devices in the context
   ur_platform_handle_t getPlatform() const;
 
+  // Get vector of devices from this context
+  const std::vector<ur_device_handle_t> &getDevices() const;
+
   // Get index of the free slot in the available pool. If there is no available
   // pool then create new one. The HostVisible parameter tells if we need a
   // slot for a host-visible event. The ProfilingEnabled tells is we need a
@@ -303,6 +306,9 @@ struct ur_context_handle_t_ : _ur_object {
   // For that the Device or its root devices need to be in the context.
   bool isValidDevice(ur_device_handle_t Device) const;
 
+  // Get handle to the L0 context
+  ze_context_handle_t getZeHandle() const;
+
 private:
   // Get the cache of events for a provided scope and profiling mode.
   auto getEventCache(bool HostVisible, bool WithProfiling,

source/adapters/level_zero/helpers/kernel_helpers.cpp

@@ -12,9 +12,14 @@
 #include "logger/ur_logger.hpp"
 
 #include "../common.hpp"
-#include "../context.hpp"
 #include "../device.hpp"
 
+#ifdef UR_ADAPTER_LEVEL_ZERO_V2
+#include "../context.hpp"
+#else
+#include "../v2/context.hpp"
+#endif
+
 ur_result_t getSuggestedLocalWorkSize(ur_device_handle_t hDevice,
                                       ze_kernel_handle_t hZeKernel,
                                       size_t GlobalWorkSize3D[3],

source/adapters/level_zero/platform.cpp

@@ -503,6 +503,8 @@ ur_result_t ur_platform_handle_t_::populateDeviceCacheIfNeeded() {
   return UR_RESULT_SUCCESS;
 }
 
+size_t ur_platform_handle_t_::getNumDevices() { return URDevicesCache.size(); }
+
 ur_device_handle_t ur_platform_handle_t_::getDeviceById(DeviceId id) {
   for (auto &dev : URDevicesCache) {
     if (dev->Id == id) {