[SYCL] Update the sycl_ext_intel_device_info Extension to Include XE Device Props

[HPS-1]

Level Zero is adding an extension to zeDeviceGetProperties that returns the following information (link to source code: https://github.com/intel/compute-runtime/blob/a852cba41fda2c0cf140e8a3e264f322e9e38911/level_zero/include/level_zero/ze_intel_gpu.h#L594) :

typedef struct _ze_intel_xe_device_exp_properties_t {
    ze_structure_type_ext_t stype;              ///< [in] type of this structure
    void *pNext;                                ///< [in][optional] must be null or a pointer to extension-specific structure
    uint32_t numXeStacks;                       ///< [out] number of stack/chiplets/tiles
    uint32_t numXeRegionsPerStack;              ///< [out] number of Regions sharing local L2/L3 (XE_CU) per stack
    uint32_t numXeClustersPerRegion;            ///< [out] number of Clusters (slices) per Region
    uint32_t numXeCorePerCluster;               ///< [out] number of XE Cores per Cluster
    uint32_t numExecutionEnginesPerXeCore;      ///< [out] number of Execution Engines (EUs) per XE Core
    uint32_t maxNumHwThreadsPerExecutionEngine; ///< [out] maximal number of HW threads per Execution Engine
    uint32_t maxNumLanesPerHwThread;            ///< [out] maximal number of lanes (virtual SIMD size) per HW thread
} ze_intel_xe_device_exp_properties_t;

Therefore, updating the sycl_ext_intel_device_info extension to add support to query these XE device properties.

[sarnex]

lgtm but please have this reviewed by a second member of the runtime team, after merge is fine

[gmlueck]

Suggested change

	struct xe_stacks_count {
	struct xe_stack_count {

I think you normally use the singular when describing the count of something.

[gmlueck]

It seems like this should be named either eus_per_xe_core or execution_engines_per_xe_core to have the same naming pattern as the others. For example, you have xe_cores_per_cluster not xe_core_count_per_cluster.

Alternatively, you could name them all with the word "count" like xe_core_count_per_cluster.

[gmlueck]

It seems like you are missing an equivalent to maxNumHwThreadsPerExecutionEngine:

maximal number of HW threads per Execution Engine

[HPS-1]

AFAIK this is a duplicate of the existing gpu_hw_threads_per_eu info descriptor (https://github.com/intel/llvm/blob/9d89f0cdbb6c4762a77b2a15e7538e44213973da/sycl/doc/extensions/supported/sycl_ext_intel_device_info.asciidoc#intel-gpu-number-of-hardware-threads-per-eu), so I think it's unnecessary to add it again?