[L0] Use zesInit for SysMan API usage

- Change to using zesInit and zes data structures for accessing L0
  SysMan functionality.
- Updated Platform & Devices to store zes handles if sysman support is
  available.
- Given Legacy Environment Variable from user, then fallback to
  old functionality.
- Fixed Return code on error to be consistently unsupported enumeration.

Signed-off-by: Neil R. Spruit <[email protected]>

Author: nrspruit

source/adapters/level_zero/adapter.cpp

@@ -52,7 +52,30 @@ class ur_legacy_sink : public logger::Sink {
   };
 };
 
-ur_result_t initPlatforms(PlatformVec &platforms) noexcept try {
+// Find the corresponding ZesDevice Handle for a given ZeDevice
+ur_result_t getZesDeviceHandle(zes_uuid_t coreDeviceUuid,
+                               zes_device_handle_t *ZesDevice,
+                               uint32_t *SubDeviceId, ze_bool_t *SubDevice) {
+  uint32_t ZesDriverCount = 0;
+  std::vector<zes_driver_handle_t> ZesDrivers;
+  std::vector<zes_device_handle_t> ZesDevices;
+  ze_result_t ZesResult = ZE_RESULT_ERROR_INVALID_ARGUMENT;
+  ZE2UR_CALL(zesDriverGet, (&ZesDriverCount, nullptr));
+  ZesDrivers.resize(ZesDriverCount);
+  ZE2UR_CALL(zesDriverGet, (&ZesDriverCount, ZesDrivers.data()));
+  for (uint32_t I = 0; I < ZesDriverCount; ++I) {
+    ZesResult = ZE_CALL_NOCHECK(
+        zesDriverGetDeviceByUuidExp,
+        (ZesDrivers[I], coreDeviceUuid, ZesDevice, SubDevice, SubDeviceId));
+    if (ZesResult == ZE_RESULT_SUCCESS) {
+      return UR_RESULT_SUCCESS;
+    }
+  }
+  return UR_RESULT_ERROR_INVALID_ARGUMENT;
+}
+
+ur_result_t initPlatforms(PlatformVec &platforms,
+                          ze_result_t ZesResult) noexcept try {
   uint32_t ZeDriverCount = 0;
   ZE2UR_CALL(zeDriverGet, (&ZeDriverCount, nullptr));
   if (ZeDriverCount == 0) {
@@ -65,24 +88,37 @@ ur_result_t initPlatforms(PlatformVec &platforms) noexcept try {
 
   ZE2UR_CALL(zeDriverGet, (&ZeDriverCount, ZeDrivers.data()));
   for (uint32_t I = 0; I < ZeDriverCount; ++I) {
+    bool DriverInit = false;
     ze_device_properties_t device_properties{};
     device_properties.stype = ZE_STRUCTURE_TYPE_DEVICE_PROPERTIES;
     uint32_t ZeDeviceCount = 0;
     ZE2UR_CALL(zeDeviceGet, (ZeDrivers[I], &ZeDeviceCount, nullptr));
     ZeDevices.resize(ZeDeviceCount);
     ZE2UR_CALL(zeDeviceGet, (ZeDrivers[I], &ZeDeviceCount, ZeDevices.data()));
+    auto platform = std::make_unique<ur_platform_handle_t_>(ZeDrivers[I]);
     // Check if this driver has GPU Devices
     for (uint32_t D = 0; D < ZeDeviceCount; ++D) {
       ZE2UR_CALL(zeDeviceGetProperties, (ZeDevices[D], &device_properties));
-
       if (ZE_DEVICE_TYPE_GPU == device_properties.type) {
-        // If this Driver is a GPU, save it as a usable platform.
-        auto platform = std::make_unique<ur_platform_handle_t_>(ZeDrivers[I]);
-        UR_CALL(platform->initialize());
+        if (!DriverInit) {
+          // If this Driver is a GPU, save it as a usable platform.
+          UR_CALL(platform->initialize());
 
-        // Save a copy in the cache for future uses.
-        platforms.push_back(std::move(platform));
-        break;
+          // Save a copy in the cache for future uses.
+          platforms.push_back(std::move(platform));
+          DriverInit = true;
+        }
+        if (ZesResult == ZE_RESULT_SUCCESS) {
+          ur_zes_device_handle_data_t ZesDeviceData;
+          zes_uuid_t ZesUUID;
+          std::memcpy(&ZesUUID, &device_properties.uuid, sizeof(zes_uuid_t));
+          if (getZesDeviceHandle(
+                  ZesUUID, &ZesDeviceData.ZesDevice, &ZesDeviceData.SubDeviceId,
+                  &ZesDeviceData.SubDevice) == UR_RESULT_SUCCESS) {
+            platforms.back()->ZedeviceToZesDeviceMap.insert(
+                std::make_pair(ZeDevices[D], std::move(ZesDeviceData)));
+          }
+        }
       }
     }
   }
@@ -172,7 +208,9 @@ ur_adapter_handle_t_::ur_adapter_handle_t_()
       return;
     }
 
-    ur_result_t err = initPlatforms(platforms);
+    GlobalAdapter->ZesResult = ZE_CALL_NOCHECK(zesInit, (0));
+
+    ur_result_t err = initPlatforms(platforms, *GlobalAdapter->ZesResult);
     if (err == UR_RESULT_SUCCESS) {
       result = std::move(platforms);
     } else {

source/adapters/level_zero/adapter.hpp

@@ -27,6 +27,7 @@ struct ur_adapter_handle_t_ {
   std::mutex Mutex;
 
   std::optional<ze_result_t> ZeResult;
+  std::optional<ze_result_t> ZesResult;
   ZeCache<Result<PlatformVec>> PlatformCache;
   logger::Logger &logger;
 };

source/adapters/level_zero/device.cpp

@@ -747,11 +747,13 @@ ur_result_t urDeviceGetInfo(
   }
 
   case UR_DEVICE_INFO_GLOBAL_MEM_FREE: {
-    if (getenv("ZES_ENABLE_SYSMAN") == nullptr) {
-      setErrorMessage("Set ZES_ENABLE_SYSMAN=1 to obtain free memory",
-                      UR_RESULT_ERROR_UNINITIALIZED,
+    bool SysManEnv = getenv_tobool("ZES_ENABLE_SYSMAN", false);
+    if ((Device->Platform->ZedeviceToZesDeviceMap.size() == 0) && !SysManEnv) {
+      setErrorMessage("SysMan support is unavailable on this system. Please "
+                      "check your level zero driver installation.",
+                      UR_RESULT_ERROR_UNSUPPORTED_ENUMERATION,
                       static_cast<int32_t>(ZE_RESULT_ERROR_UNINITIALIZED));
-      return UR_RESULT_ERROR_ADAPTER_SPECIFIC;
+      return UR_RESULT_ERROR_UNSUPPORTED_ENUMERATION;
     }
     // Calculate the global memory size as the max limit that can be reported as
     // "free" memory for the user to allocate.
@@ -760,30 +762,57 @@ ur_result_t urDeviceGetInfo(
     // Currently this is only the one enumerated with ordinal 0.
     uint64_t FreeMemory = 0;
     uint32_t MemCount = 0;
-    ZE2UR_CALL(zesDeviceEnumMemoryModules, (ZeDevice, &MemCount, nullptr));
+
+    zes_device_handle_t ZesDevice = Device->ZeDevice;
+    struct ur_zes_device_handle_data_t ZesDeviceData = {};
+    // If legacy sysman is enabled thru the environment variable, then zesInit
+    // will fail, but sysman is still usable so go the legacy route.
+    if (!SysManEnv) {
+      auto It = Device->Platform->ZedeviceToZesDeviceMap.find(Device->ZeDevice);
+      if (It == Device->Platform->ZedeviceToZesDeviceMap.end()) {
+        // no matching device
+        return UR_RESULT_ERROR_UNSUPPORTED_ENUMERATION;
+      } else {
+        ZesDeviceData =
+            Device->Platform->ZedeviceToZesDeviceMap[Device->ZeDevice];
+        ZesDevice = ZesDeviceData.ZesDevice;
+      }
+    }
+
+    ZE2UR_CALL(zesDeviceEnumMemoryModules, (ZesDevice, &MemCount, nullptr));
     if (MemCount != 0) {
       std::vector<zes_mem_handle_t> ZesMemHandles(MemCount);
       ZE2UR_CALL(zesDeviceEnumMemoryModules,
-                 (ZeDevice, &MemCount, ZesMemHandles.data()));
+                 (ZesDevice, &MemCount, ZesMemHandles.data()));
       for (auto &ZesMemHandle : ZesMemHandles) {
         ZesStruct<zes_mem_properties_t> ZesMemProperties;
         ZE2UR_CALL(zesMemoryGetProperties, (ZesMemHandle, &ZesMemProperties));
         // For root-device report memory from all memory modules since that
         // is what totally available in the default implicit scaling mode.
         // For sub-devices only report memory local to them.
-        if (!Device->isSubDevice() || Device->ZeDeviceProperties->subdeviceId ==
-                                          ZesMemProperties.subdeviceId) {
-
-          ZesStruct<zes_mem_state_t> ZesMemState;
-          ZE2UR_CALL(zesMemoryGetState, (ZesMemHandle, &ZesMemState));
-          FreeMemory += ZesMemState.free;
+        if (SysManEnv) {
+          if (!Device->isSubDevice() ||
+              Device->ZeDeviceProperties->subdeviceId ==
+                  ZesMemProperties.subdeviceId) {
+
+            ZesStruct<zes_mem_state_t> ZesMemState;
+            ZE2UR_CALL(zesMemoryGetState, (ZesMemHandle, &ZesMemState));
+            FreeMemory += ZesMemState.free;
+          }
+        } else {
+          if (ZesDeviceData.SubDeviceId == ZesMemProperties.subdeviceId ||
+              !ZesDeviceData.SubDevice) {
+            ZesStruct<zes_mem_state_t> ZesMemState;
+            ZE2UR_CALL(zesMemoryGetState, (ZesMemHandle, &ZesMemState));
+            FreeMemory += ZesMemState.free;
+          }
         }
       }
     }
     if (MemCount > 0) {
       return ReturnValue(std::min(GlobalMemSize, FreeMemory));
     } else {
-      return UR_RESULT_ERROR_UNSUPPORTED_FEATURE;
+      return UR_RESULT_ERROR_UNSUPPORTED_ENUMERATION;
     }
   }
   case UR_DEVICE_INFO_MEMORY_CLOCK_RATE: {

source/adapters/level_zero/platform.hpp

@@ -12,11 +12,18 @@
 #include "common.hpp"
 #include "ur_api.h"
 #include "ze_api.h"
+#include "zes_api.h"
 
 struct ur_device_handle_t_;
 
 typedef size_t DeviceId;
 
+struct ur_zes_device_handle_data_t {
+  zes_device_handle_t ZesDevice;
+  uint32_t SubDeviceId;
+  ze_bool_t SubDevice = false;
+};
+
 struct ur_platform_handle_t_ : public _ur_platform {
   ur_platform_handle_t_(ze_driver_handle_t Driver)
       : ZeDriver{Driver}, ZeApiVersion{ZE_API_VERSION_CURRENT} {}
@@ -27,6 +34,11 @@ struct ur_platform_handle_t_ : public _ur_platform {
   // a pretty good fit to keep here.
   ze_driver_handle_t ZeDriver;
 
+  // Cache of the ZesDevices mapped to the ZeDevices for use in zes apis calls
+  // based on a ze device handle.
+  std::unordered_map<ze_device_handle_t, ur_zes_device_handle_data_t>
+      ZedeviceToZesDeviceMap;
+
   // Given a multi driver scenario, the driver handle must be translated to the
   // internal driver handle to allow calls to driver experimental apis.
   ze_driver_handle_t ZeDriverHandleExpTranslated;