Use extension version of clGetKernelSubGroupInfo when necessary.

We do support at least one device (fpga emulator) that reports a cl
version of < 2.1 and supports subgroups, in that scenario we need query
out and use the KHR version of the entry point.

Author: aarongreig

source/adapters/opencl/device.cpp

@@ -61,7 +61,8 @@ ur_result_t cl_adapter::checkDeviceExtensions(
       // doesn't report them.
       if (isIntelFPGAEmuDevice(Dev) &&
           (Ext == "cl_intel_device_attribute_query" ||
-           Ext == "cl_intel_required_subgroup_size")) {
+           Ext == "cl_intel_required_subgroup_size" ||
+           Ext == "cl_khr_subgroups")) {
         Supported = true;
         continue;
       }

source/adapters/opencl/kernel.cpp

@@ -8,6 +8,7 @@
 //
 //===----------------------------------------------------------------------===//
 #include "common.hpp"
+#include "device.hpp"
 
 #include <algorithm>
 #include <cstddef>
@@ -189,11 +190,45 @@ urKernelGetSubGroupInfo(ur_kernel_handle_t hKernel, ur_device_handle_t hDevice,
     InputValueSize = MaxDims * sizeof(size_t);
   }
 
-  cl_int Ret = clGetKernelSubGroupInfo(cl_adapter::cast<cl_kernel>(hKernel),
-                                       cl_adapter::cast<cl_device_id>(hDevice),
-                                       mapURKernelSubGroupInfoToCL(propName),
-                                       InputValueSize, InputValue.get(),
-                                       sizeof(size_t), &RetVal, pPropSizeRet);
+  // We need to allow for the possibility that this device runs an older CL and
+  // supports the original khr subgroup extension.
+  using ApiFuncT =
+      cl_int(CL_API_CALL *)(cl_kernel, cl_device_id, cl_kernel_sub_group_info,
+                            size_t, const void *, size_t, void *, size_t *);
+  ApiFuncT GetKernelSubGroupInfo = nullptr;
+
+  oclv::OpenCLVersion DevVer;
+  CL_RETURN_ON_FAILURE(cl_adapter::getDeviceVersion(
+      cl_adapter::cast<cl_device_id>(hDevice), DevVer));
+
+  if (DevVer < oclv::V2_1) {
+    bool SubgroupExtSupported = false;
+
+    UR_RETURN_ON_FAILURE(cl_adapter::checkDeviceExtensions(
+        cl_adapter::cast<cl_device_id>(hDevice), {"cl_khr_subgroups"},
+        SubgroupExtSupported));
+    if (!SubgroupExtSupported) {
+      return UR_RESULT_ERROR_UNSUPPORTED_FEATURE;
+    }
+    cl_platform_id Platform;
+    CL_RETURN_ON_FAILURE(clGetDeviceInfo(
+        cl_adapter::cast<cl_device_id>(hDevice), CL_DEVICE_PLATFORM,
+        sizeof(Platform), &Platform, nullptr));
+    GetKernelSubGroupInfo =
+        reinterpret_cast<ApiFuncT>(clGetExtensionFunctionAddressForPlatform(
+            Platform, "clGetKernelSubGroupInfoKHR"));
+    if (!GetKernelSubGroupInfo) {
+      return UR_RESULT_ERROR_UNSUPPORTED_FEATURE;
+    }
+  } else {
+    GetKernelSubGroupInfo = clGetKernelSubGroupInfo;
+  }
+
+  cl_int Ret = GetKernelSubGroupInfo(cl_adapter::cast<cl_kernel>(hKernel),
+                                     cl_adapter::cast<cl_device_id>(hDevice),
+                                     mapURKernelSubGroupInfoToCL(propName),
+                                     InputValueSize, InputValue.get(),
+                                     sizeof(size_t), &RetVal, pPropSizeRet);
 
   if (Ret == CL_INVALID_OPERATION) {
     // clGetKernelSubGroupInfo returns CL_INVALID_OPERATION if the device does