{UR][L0] Return device version based on DeviceIpVersion

When being queried a device version:
- CUDA adapter returns the compute capability version of the device,
- HIP adapter returns the architecture name.

Both of these are what define a GPU architecture to compile binaries for and can be used with GPU compilers, this is useful information for applications and developers.

On L0/Intel side, prior to this change, it returns the L0 API version.
This information is already returned in BACKEND_RUNTIME_VERSION and DRIVER_VERSION queries.

With this change, we align the adapters behavior with CUDA and HIP, and return the Intel device architecture version that can be used with ocloc compiler.

source/adapters/level_zero/device.cpp

@@ -338,8 +338,26 @@ UR_APIEXPORT ur_result_t UR_APICALL urDeviceGetInfo(
   case UR_DEVICE_INFO_DRIVER_VERSION:
   case UR_DEVICE_INFO_BACKEND_RUNTIME_VERSION:
     return ReturnValue(Device->Platform->ZeDriverVersion.c_str());
-  case UR_DEVICE_INFO_VERSION:
-    return ReturnValue(Device->Platform->ZeDriverApiVersion.c_str());
+  case UR_DEVICE_INFO_VERSION: {
+    // from compute-runtime/shared/source/helpers/hw_ip_version.h
+    typedef struct {
+      union {
+        uint32_t value;
+        struct {
+          uint32_t revision : 6;
+          uint32_t reserved : 8;
+          uint32_t release : 8;
+          uint32_t architecture : 10;
+        } components;
+      };
+    } ipVersion_t;
+    ipVersion_t IpVersion;
+    IpVersion.value = Device->ZeDeviceIpVersionExt->ipVersion;
+    std::stringstream S;
+    S << IpVersion.components.architecture << "."
+      << IpVersion.components.release << "." << IpVersion.components.revision;
+    return ReturnValue(S.str().c_str());
+  }
   case UR_DEVICE_INFO_PARTITION_MAX_SUB_DEVICES: {
     auto Res = Device->Platform->populateDeviceCacheIfNeeded();
     if (Res != UR_RESULT_SUCCESS) {