Add support for getting device node mask

DBDuncan · DBDuncan · commit 707a1240d1a3 · 2025-07-10T16:21:34.000+01:00
diff --git a/llvm/include/llvm/SYCLLowerIR/DeviceConfigFile.td b/llvm/include/llvm/SYCLLowerIR/DeviceConfigFile.td
@@ -93,6 +93,7 @@ def AspectExt_intel_fan_speed : Aspect<"ext_intel_fan_speed">;
 def AspectExt_intel_power_limits : Aspect<"ext_intel_power_limits">;
 def AspectExt_oneapi_async_memory_alloc : Aspect<"ext_oneapi_async_memory_alloc">;
 def AspectExt_intel_device_info_luid : Aspect<"ext_intel_device_info_luid">;
+def AspectExt_intel_device_info_node_mask : Aspect<"ext_intel_device_info_node_mask">;
 
 // Deprecated aspects
 def AspectInt64_base_atomics : Aspect<"int64_base_atomics">;
@@ -165,7 +166,8 @@ def : TargetInfo<"__TestAspectList",
     AspectExt_intel_fan_speed,
     AspectExt_intel_power_limits,
     AspectExt_oneapi_async_memory_alloc,
-    AspectExt_intel_device_info_luid],
+    AspectExt_intel_device_info_luid,
+    AspectExt_intel_device_info_node_mask],
     []>;
 // This definition serves the only purpose of testing whether the deprecated aspect list defined in here and in SYCL RT
 // match.
diff --git a/sycl/doc/extensions/supported/sycl_ext_intel_device_info.md b/sycl/doc/extensions/supported/sycl_ext_intel_device_info.md
@@ -20,7 +20,7 @@ The Feature Test Macro SYCL\_EXT\_INTEL\_DEVICE\_INFO will be defined as one of
 | 5     | Device ID is supported |
 | 6     | Memory clock rate and bus width queries are supported |
 | 7     | Throttle reasons, fan speed and power limits queries are supported |
-| 8     | Device LUID is supported |
+| 8     | Device LUID and device node mask is supported |
 
 
 
@@ -657,6 +657,37 @@ The LUID can be obtained using the standard `get_info()` interface.
     }
 ```
 
+
+# Device Node Mask #
+
+A new device descriptor is added which will provide the device node mask.
+
+## Version ##
+
+The extension supports this query in version 8 and later.
+
+| Device Descriptors | Return Type | Description |
+| ------------------ | ----------- | ----------- |
+|`ext::intel::info::device::node_mask` |`unsigned int` | Returns the device node mask. |
+
+## Aspects ##
+
+A new aspect, `ext_intel_device_info_node_mask`, is added.
+
+## Error Condition ##
+
+Throws a synchronous `exception` with the `errc::feature_not_supported` error code if the device does not have `aspect::ext_intel_device_info_node_mask`.
+
+## Example Usage ##
+
+The device node mask can be obtained using the standard `get_info()` interface.
+
+```
+    if (dev.has(aspect::ext_intel_device_info_node_mask)) {
+      auto node_mask = dev.get_info<ext::intel::info::device::node_mask>();
+    }
+```
+
 # Deprecated queries #
 
 The table below lists deprecated, that would soon be removed and their replacements:
diff --git a/sycl/include/sycl/info/aspects.def b/sycl/include/sycl/info/aspects.def
@@ -79,4 +79,5 @@ __SYCL_ASPECT(ext_intel_fan_speed, 85)
 __SYCL_ASPECT(ext_intel_power_limits, 86)
 __SYCL_ASPECT(ext_oneapi_async_memory_alloc, 87)
 __SYCL_ASPECT(ext_intel_device_info_luid, 88)
+__SYCL_ASPECT(ext_intel_device_info_node_mask, 89)
 
diff --git a/sycl/include/sycl/info/ext_intel_device_traits.def b/sycl/include/sycl/info/ext_intel_device_traits.def
@@ -22,6 +22,7 @@ __SYCL_PARAM_TRAITS_SPEC(ext::intel, device, fan_speed, int32_t, UR_DEVICE_INFO_
 __SYCL_PARAM_TRAITS_SPEC(ext::intel, device, min_power_limit, int32_t, UR_DEVICE_INFO_MIN_POWER_LIMIT)
 __SYCL_PARAM_TRAITS_SPEC(ext::intel, device, max_power_limit, int32_t, UR_DEVICE_INFO_MAX_POWER_LIMIT)
 __SYCL_PARAM_TRAITS_SPEC(ext::intel, device, luid, detail::luid_type, __SYCL_TRAIT_HANDLED_IN_RT)
+__SYCL_PARAM_TRAITS_SPEC(ext::intel, device, node_mask, uint32_t, __SYCL_TRAIT_HANDLED_IN_RT)
 #ifdef __SYCL_PARAM_TRAITS_TEMPLATE_SPEC_NEEDS_UNDEF
 #undef __SYCL_PARAM_TRAITS_TEMPLATE_SPEC
 #undef __SYCL_PARAM_TRAITS_TEMPLATE_SPEC_NEEDS_UNDEF
diff --git a/sycl/source/detail/device_impl.hpp b/sycl/source/detail/device_impl.hpp
@@ -1202,6 +1202,13 @@ class device_impl : public std::enable_shared_from_this<device_impl> {
             "The device does not have the ext_intel_device_info_luid aspect");
       return get_info_impl<UR_DEVICE_INFO_LUID>();
     }
+    CASE(ext::intel::info::device::node_mask) {
+      if (!has(aspect::ext_intel_device_info_node_mask))
+        throw exception(
+            make_error_code(errc::feature_not_supported),
+            "The device does not have the ext_intel_device_info_node_mask aspect");
+      return get_info_impl<UR_DEVICE_INFO_NODE_MASK>();
+    }
     else {
       constexpr auto Desc = UrInfoCode<Param>::value;
       return static_cast<typename Param::return_type>(get_info_impl<Desc>());
@@ -1315,6 +1322,9 @@ class device_impl : public std::enable_shared_from_this<device_impl> {
     CASE(ext_intel_device_info_luid) {
       return has_info_desc(UR_DEVICE_INFO_LUID);
     }
+    CASE(ext_intel_device_info_node_mask) {
+      return has_info_desc(UR_DEVICE_INFO_NODE_MASK);
+    }
     CASE(ext_intel_max_mem_bandwidth) {
       // currently not supported
       return false;
diff --git a/sycl/source/detail/ur_device_info_ret_types.inc b/sycl/source/detail/ur_device_info_ret_types.inc
@@ -157,6 +157,7 @@ MAP(UR_DEVICE_INFO_BFLOAT16_CONVERSIONS_NATIVE, ur_bool_t)
 MAP(UR_DEVICE_INFO_KERNEL_LAUNCH_CAPABILITIES, ur_kernel_launch_properties_flags_t)
 // Manually changed std::vector<uint8_t> -> std::array<uint8_t, 8>
 MAP(UR_DEVICE_INFO_LUID, std::array<uint8_t, 8>)
+MAP(UR_DEVICE_INFO_NODE_MASK, uint32_t)
 
 // These aren't present in the specification, extracted from ur_api.h
 // instead.
diff --git a/sycl/test-e2e/Adapters/level_zero/node_mask.cpp b/sycl/test-e2e/Adapters/level_zero/node_mask.cpp
@@ -0,0 +1,47 @@
+// REQUIRES: aspect-ext_intel_device_info_luid
+// REQUIRES: gpu, level_zero, level_zero_dev_kit, windows
+
+// RUN: %{build} %level_zero_options -o %t.out
+// RUN: %{run} %t.out
+
+// Test that the node mask is read correctly from Level Zero.
+
+#include <iomanip>
+#include <iostream>
+#include <level_zero/ze_api.h>
+#include <sstream>
+#include <sycl/backend.hpp>
+#include <sycl/detail/core.hpp>
+
+int main() {
+  sycl::device dev;
+  auto nodeMaskSYCL = dev.get_info<sycl::ext::intel::info::device::node_mask>();
+
+  std::cout << "SYCL: " << nodeMaskSYCL << std::endl;
+
+  auto zedev = sycl::get_native<sycl::backend::ext_oneapi_level_zero>(dev);
+  ze_device_properties_t device_properties{};
+  device_properties.stype = ZE_STRUCTURE_TYPE_DEVICE_PROPERTIES;
+
+  ze_device_luid_ext_properties_t luid_device_properties{};
+  luid_device_properties.stype = ZE_STRUCTURE_TYPE_DEVICE_LUID_EXT_PROPERTIES;
+
+  device_properties.pNext = &luid_device_properties;
+
+  zeDeviceGetProperties(zedev, &device_properties);
+
+  ze_device_luid_ext_properties_t *luid_dev_prop =
+      static_cast<ze_device_luid_ext_properties_t *>(device_properties.pNext);
+
+  uint32_t nodeMaskL0 = luid_dev_prop->nodeMask;
+
+  std::cout << "L0  : " << nodeMaskL0 << std::endl;
+
+  if (nodeMaskSYCL != nodeMaskL0) {
+    std::cout << "FAILED" << std::endl;
+    return -1;
+  }
+
+  std::cout << "PASSED" << std::endl;
+  return 0;
+}
diff --git a/unified-runtime/include/ur_api.h b/unified-runtime/include/ur_api.h
diff --git a/unified-runtime/include/ur_print.hpp b/unified-runtime/include/ur_print.hpp
diff --git a/unified-runtime/scripts/core/device.yml b/unified-runtime/scripts/core/device.yml
@@ -466,6 +466,8 @@ etors:
       desc: "[$x_kernel_launch_properties_flags_t] Bitfield of supported kernel launch properties."
     - name: LUID
       desc: "[uint8_t[]][optional-query] return device LUID"
+    - name: NODE_MASK
+      desc: "[uint32_t][optional-query] return device node mask"
 --- #--------------------------------------------------------------------------
 type: function
 desc: "Retrieves various information about device"
diff --git a/unified-runtime/source/adapters/cuda/device.cpp b/unified-runtime/source/adapters/cuda/device.cpp
@@ -1203,6 +1203,20 @@ UR_APIEXPORT ur_result_t UR_APICALL urDeviceGetInfo(ur_device_handle_t hDevice,
     std::copy(LUID, LUID + 8, Name.begin());
     return ReturnValue(Name.data(), 8);
   }
+  case UR_DEVICE_INFO_NODE_MASK: {
+    // Device node mask is only available on Windows.
+    // Intel extension for device node mask. This returns the node mask as
+    // uint32_t. For details about this extension,
+    // see sycl/doc/extensions/supported/sycl_ext_intel_device_info.md.
+    uint32_t *nodeMask = nullptr;
+    cuDeviceGetLuid(nullptr, nodeMask, hDevice->get());
+
+    if (nodeMask == nullptr) {
+      return UR_RESULT_ERROR_UNSUPPORTED_FEATURE;
+    }
+
+    return ReturnValue(*nodeMask);
+  }
   default:
     break;
   }
diff --git a/unified-runtime/source/adapters/level_zero/device.cpp b/unified-runtime/source/adapters/level_zero/device.cpp
@@ -1366,6 +1366,29 @@ ur_result_t urDeviceGetInfo(
       return UR_RESULT_ERROR_UNSUPPORTED_FEATURE;
     }
   }
+  case UR_DEVICE_INFO_NODE_MASK: {
+    // Device node mask is only available on Windows.
+    // Intel extension for device node mask. This returns the node mask as
+    // uint32_t. For details about this extension,
+    // see sycl/doc/extensions/supported/sycl_ext_intel_device_info.md.
+    
+    // Node mask is provided through the L0 LUID extension so support for this
+    // extension must be checked.
+    if (Device->Platform->ZeLUIDSupported) {
+      ze_device_properties_t DeviceProp = {};
+      DeviceProp.stype = ZE_STRUCTURE_TYPE_DEVICE_PROPERTIES;
+      ze_device_luid_ext_properties_t LuidDesc = {};
+      LuidDesc.stype = ZE_STRUCTURE_TYPE_DEVICE_LUID_EXT_PROPERTIES;
+      DeviceProp.pNext = (void *)&LuidDesc;
+
+      ZE2UR_CALL(zeDeviceGetProperties, (ZeDevice, &DeviceProp));
+
+      const auto &nodeMask = LuidDesc.nodeMask;
+      return ReturnValue(nodeMask, sizeof(nodeMask));
+    } else {
+      return UR_RESULT_ERROR_UNSUPPORTED_FEATURE;
+    }
+  }
   default:
     UR_LOG(ERR, "Unsupported ParamName in urGetDeviceInfo");
     UR_LOG(ERR, "ParamNameParamName={}(0x{})", ParamName,
diff --git a/unified-runtime/test/conformance/testing/include/uur/optional_queries.h b/unified-runtime/test/conformance/testing/include/uur/optional_queries.h
diff --git a/unified-runtime/tools/urinfo/urinfo.hpp b/unified-runtime/tools/urinfo/urinfo.hpp
