[SYCL] Add number HW threads per EU query

The appropriate spec update: intel#4876
Test: intel/llvm-test-suite#550

Signed-off-by: Dmitry Sidorov <[email protected]>

Author: MrSidims

sycl/include/CL/sycl/aspects.hpp

@@ -48,6 +48,7 @@ enum class aspect {
   ext_oneapi_srgb = 30,
   ext_oneapi_native_assert = 31,
   host_debuggable = 32,
+  ext_intel_gpu_hw_threads_per_eu = 33,
 };
 
 } // namespace sycl

sycl/include/CL/sycl/detail/pi.h

@@ -301,6 +301,7 @@ typedef enum {
   PI_DEVICE_INFO_IMAGE_SRGB = 0x10027,
   PI_DEVICE_INFO_ATOMIC_64 = 0x10110,
   PI_DEVICE_INFO_ATOMIC_MEMORY_ORDER_CAPABILITIES = 0x10111,
+  PI_DEVICE_INFO_GPU_HW_THREADS_PER_EU = 0x10120,
   PI_EXT_ONEAPI_DEVICE_INFO_MAX_GLOBAL_WORK_GROUPS = 0x20000,
   PI_EXT_ONEAPI_DEVICE_INFO_MAX_WORK_GROUPS_1D = 0x20001,
   PI_EXT_ONEAPI_DEVICE_INFO_MAX_WORK_GROUPS_2D = 0x20002,

sycl/include/CL/sycl/info/device_traits.def

@@ -95,6 +95,7 @@ __SYCL_PARAM_TRAITS_SPEC(device, ext_intel_gpu_eu_simd_width, pi_uint32)
 __SYCL_PARAM_TRAITS_SPEC(device, ext_intel_gpu_slices, pi_uint32)
 __SYCL_PARAM_TRAITS_SPEC(device, ext_intel_gpu_subslices_per_slice, pi_uint32)
 __SYCL_PARAM_TRAITS_SPEC(device, ext_intel_gpu_eu_count_per_subslice, pi_uint32)
+__SYCL_PARAM_TRAITS_SPEC(device, ext_intel_gpu_hw_threads_per_eu, pi_uint32)
 __SYCL_PARAM_TRAITS_SPEC(device, ext_intel_max_mem_bandwidth, pi_uint64)
 __SYCL_PARAM_TRAITS_SPEC(device, ext_intel_mem_channel, bool)
 __SYCL_PARAM_TRAITS_SPEC(device, ext_oneapi_srgb, bool)

sycl/include/CL/sycl/info/info_desc.hpp

@@ -156,6 +156,7 @@ enum class device : cl_device_info {
   ext_intel_gpu_subslices_per_slice = PI_DEVICE_INFO_GPU_SUBSLICES_PER_SLICE,
   ext_intel_gpu_eu_count_per_subslice =
       PI_DEVICE_INFO_GPU_EU_COUNT_PER_SUBSLICE,
+  ext_intel_gpu_hw_threads_per_eu = PI_DEVICE_INFO_GPU_HW_THREADS_PER_EU,
   ext_intel_max_mem_bandwidth = PI_DEVICE_INFO_MAX_MEM_BANDWIDTH,
   ext_intel_mem_channel = PI_MEM_PROPERTIES_CHANNEL,
   ext_oneapi_srgb = PI_DEVICE_INFO_IMAGE_SRGB,

sycl/plugins/cuda/pi_cuda.cpp

@@ -1675,6 +1675,7 @@ pi_result cuda_piDeviceGetInfo(pi_device device, pi_device_info param_name,
   case PI_DEVICE_INFO_GPU_SLICES:
   case PI_DEVICE_INFO_GPU_SUBSLICES_PER_SLICE:
   case PI_DEVICE_INFO_GPU_EU_COUNT_PER_SUBSLICE:
+  case PI_DEVICE_INFO_GPU_HW_THREADS_PER_EU:
   case PI_DEVICE_INFO_MAX_MEM_BANDWIDTH:
     // TODO: Check if Intel device UUID extension is utilized for CUDA.
     // For details about this extension, see

sycl/plugins/level_zero/pi_level_zero.cpp

@@ -2463,6 +2463,8 @@ pi_result piDeviceGetInfo(pi_device Device, pi_device_info ParamName,
   case PI_DEVICE_INFO_GPU_EU_COUNT_PER_SUBSLICE:
     return ReturnValue(
         pi_uint32{Device->ZeDeviceProperties->numEUsPerSubslice});
+  case PI_DEVICE_INFO_GPU_HW_THREADS_PER_EU:
+    return ReturnValue(pi_uint32{Device->ZeDeviceProperties->numThreadsPerEU});
   case PI_DEVICE_INFO_MAX_MEM_BANDWIDTH:
     // currently not supported in level zero runtime
     return PI_INVALID_VALUE;

sycl/source/detail/device_impl.cpp

@@ -303,6 +303,11 @@ bool device_impl::has(aspect Aspect) const {
                MDevice, PI_DEVICE_INFO_GPU_EU_COUNT_PER_SUBSLICE,
                sizeof(pi_device_type), &device_type,
                &return_size) == PI_SUCCESS;
+  case aspect::ext_intel_gpu_hw_threads_per_eu:
+    return getPlugin().call_nocheck<detail::PiApiKind::piDeviceGetInfo>(
+               MDevice, PI_DEVICE_INFO_GPU_HW_THREADS_PER_EU,
+               sizeof(pi_device_type), &device_type,
+               &return_size) == PI_SUCCESS;
   case aspect::ext_intel_device_info_uuid: {
     auto Result = getPlugin().call_nocheck<detail::PiApiKind::piDeviceGetInfo>(
         MDevice, PI_DEVICE_INFO_UUID, 0, nullptr, &return_size);

sycl/source/detail/device_info.hpp

@@ -1283,6 +1283,13 @@ get_device_info_host<info::device::ext_intel_gpu_eu_count_per_subslice>() {
       PI_INVALID_DEVICE);
 }
 template <>
+inline cl_uint
+get_device_info_host<info::device::ext_intel_gpu_hw_threads_per_eu>() {
+  throw runtime_error(
+      "Obtaining the HW threads count per EU is not supported on HOST device",
+      PI_INVALID_DEVICE);
+}
+template <>
 inline cl_ulong
 get_device_info_host<info::device::ext_intel_max_mem_bandwidth>() {
   throw runtime_error(

sycl/test/abi/sycl_symbols_linux.dump

@@ -4382,3 +4382,8 @@ _ZNK2cl4sycl9exception8categoryEv
 _ZNK2cl4sycl9kernel_id8get_nameEv
 __sycl_register_lib
 __sycl_unregister_lib
+_ZNK2cl4sycl6device8get_infoILNS0_4info6deviceE131072EEENS3_12param_traitsIS4_XT_EE11return_typeEv
+_ZNK2cl4sycl6device8get_infoILNS0_4info6deviceE131075EEENS3_12param_traitsIS4_XT_EE11return_typeEv
+_ZNK2cl4sycl6device8get_infoILNS0_4info6deviceE131074EEENS3_12param_traitsIS4_XT_EE11return_typeEv
+_ZNK2cl4sycl6device8get_infoILNS0_4info6deviceE131073EEENS3_12param_traitsIS4_XT_EE11return_typeEv
+_ZNK2cl4sycl6device8get_infoILNS0_4info6deviceE65824EEENS3_12param_traitsIS4_XT_EE11return_typeEv