Report device fp support via config rather than extension string.

We're trying to move the UR adapters away from returning hard coded OpenCL
extension strings to report device capabilities, this is the first change
in that direction.

Author: aarongreig

source/adapters/cuda/device.cpp

@@ -437,7 +437,18 @@ UR_APIEXPORT ur_result_t UR_APICALL urDeviceGetInfo(ur_device_handle_t hDevice,
   }
   case UR_DEVICE_INFO_HALF_FP_CONFIG: {
     // TODO: is this config consistent across all NVIDIA GPUs?
-    return ReturnValue(0u);
+    if ((Major >= 6) || ((Major == 5) && (Minor >= 3))) {
+      ur_device_fp_capability_flags_t Config =
+          UR_DEVICE_FP_CAPABILITY_FLAG_DENORM |
+          UR_DEVICE_FP_CAPABILITY_FLAG_INF_NAN |
+          UR_DEVICE_FP_CAPABILITY_FLAG_ROUND_TO_NEAREST |
+          UR_DEVICE_FP_CAPABILITY_FLAG_ROUND_TO_ZERO |
+          UR_DEVICE_FP_CAPABILITY_FLAG_ROUND_TO_INF |
+          UR_DEVICE_FP_CAPABILITY_FLAG_FMA;
+      return ReturnValue(Config);
+    } else {
+      return ReturnValue(0u);
+    }
   }
   case UR_DEVICE_INFO_SINGLE_FP_CONFIG: {
     // TODO: is this config consistent across all NVIDIA GPUs?
@@ -615,7 +626,7 @@ UR_APIEXPORT ur_result_t UR_APICALL urDeviceGetInfo(ur_device_handle_t hDevice,
   }
   case UR_DEVICE_INFO_EXTENSIONS: {
 
-    std::string SupportedExtensions = "cl_khr_fp64 cl_khr_subgroups ";
+    std::string SupportedExtensions = "cl_khr_subgroups ";
     SupportedExtensions += "cl_intel_devicelib_assert ";
     // Return supported for the UR command-buffer experimental feature
     SupportedExtensions += "ur_exp_command_buffer ";
@@ -631,10 +642,6 @@ UR_APIEXPORT ur_result_t UR_APICALL urDeviceGetInfo(ur_device_handle_t hDevice,
     UR_CHECK_ERROR(cuDeviceGetAttribute(
         &Minor, CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MINOR, hDevice->get()));
 
-    if ((Major >= 6) || ((Major == 5) && (Minor >= 3))) {
-      SupportedExtensions += "cl_khr_fp16 ";
-    }
-
     return ReturnValue(SupportedExtensions.c_str());
   }
   case UR_DEVICE_INFO_PRINTF_BUFFER_SIZE: {

source/adapters/hip/device.cpp

@@ -355,29 +355,40 @@ UR_APIEXPORT ur_result_t UR_APICALL urDeviceGetInfo(ur_device_handle_t hDevice,
     return ReturnValue(MemBaseAddrAlign);
   }
   case UR_DEVICE_INFO_HALF_FP_CONFIG: {
-    return ReturnValue(0u);
-  }
-  case UR_DEVICE_INFO_SINGLE_FP_CONFIG: {
     ur_device_fp_capability_flags_t Config =
         UR_DEVICE_FP_CAPABILITY_FLAG_DENORM |
         UR_DEVICE_FP_CAPABILITY_FLAG_INF_NAN |
         UR_DEVICE_FP_CAPABILITY_FLAG_ROUND_TO_NEAREST |
         UR_DEVICE_FP_CAPABILITY_FLAG_ROUND_TO_ZERO |
         UR_DEVICE_FP_CAPABILITY_FLAG_ROUND_TO_INF |
-        UR_DEVICE_FP_CAPABILITY_FLAG_FMA |
-        UR_DEVICE_FP_CAPABILITY_FLAG_CORRECTLY_ROUNDED_DIVIDE_SQRT;
+        UR_DEVICE_FP_CAPABILITY_FLAG_FMA;
     return ReturnValue(Config);
   }
-  case UR_DEVICE_INFO_DOUBLE_FP_CONFIG: {
+  case UR_DEVICE_INFO_SINGLE_FP_CONFIG: {
     ur_device_fp_capability_flags_t Config =
         UR_DEVICE_FP_CAPABILITY_FLAG_DENORM |
         UR_DEVICE_FP_CAPABILITY_FLAG_INF_NAN |
         UR_DEVICE_FP_CAPABILITY_FLAG_ROUND_TO_NEAREST |
         UR_DEVICE_FP_CAPABILITY_FLAG_ROUND_TO_ZERO |
         UR_DEVICE_FP_CAPABILITY_FLAG_ROUND_TO_INF |
-        UR_DEVICE_FP_CAPABILITY_FLAG_FMA;
+        UR_DEVICE_FP_CAPABILITY_FLAG_FMA |
+        UR_DEVICE_FP_CAPABILITY_FLAG_CORRECTLY_ROUNDED_DIVIDE_SQRT;
     return ReturnValue(Config);
   }
+  case UR_DEVICE_INFO_DOUBLE_FP_CONFIG: {
+    if (Props.arch.hasDoubles) {
+      ur_device_fp_capability_flags_t Config =
+          UR_DEVICE_FP_CAPABILITY_FLAG_DENORM |
+          UR_DEVICE_FP_CAPABILITY_FLAG_INF_NAN |
+          UR_DEVICE_FP_CAPABILITY_FLAG_ROUND_TO_NEAREST |
+          UR_DEVICE_FP_CAPABILITY_FLAG_ROUND_TO_ZERO |
+          UR_DEVICE_FP_CAPABILITY_FLAG_ROUND_TO_INF |
+          UR_DEVICE_FP_CAPABILITY_FLAG_FMA;
+      return ReturnValue(Config);
+    } else {
+      return ReturnValue(0u);
+    }
+  }
   case UR_DEVICE_INFO_GLOBAL_MEM_CACHE_TYPE: {
     return ReturnValue(UR_DEVICE_MEM_CACHE_TYPE_READ_WRITE_CACHE);
   }
@@ -570,12 +581,6 @@ UR_APIEXPORT ur_result_t UR_APICALL urDeviceGetInfo(ur_device_handle_t hDevice,
     detail::ur::assertion(hipGetDeviceProperties(&Props, hDevice->get()) ==
                           hipSuccess);
 
-    if (Props.arch.hasDoubles) {
-      SupportedExtensions += "cl_khr_fp64 ";
-    }
-
-    SupportedExtensions += "cl_khr_fp16 ";
-
     return ReturnValue(SupportedExtensions.c_str());
   }
   case UR_DEVICE_INFO_PRINTF_BUFFER_SIZE: {

source/adapters/level_zero/device.cpp

@@ -258,8 +258,6 @@ ur_result_t urDeviceGetInfo(
     //   for performance.
     // cl_intel_required_subgroup_size - Extension to allow programmers to
     //   optionally specify the required subgroup size for a kernel function.
-    // cl_khr_fp16 - Optional half floating-point support.
-    // cl_khr_fp64 - Support for double floating-point precision.
     // cl_khr_int64_base_atomics, cl_khr_int64_extended_atomics - Optional
     //   extensions that implement atomic operations on 64-bit signed and
     //   unsigned integers to locations in __global and __local memory.
@@ -269,10 +267,6 @@ ur_result_t urDeviceGetInfo(
     // Hardcoding some extensions we know are supported by all Level Zero
     // devices.
     SupportedExtensions += (ZE_SUPPORTED_EXTENSIONS);
-    if (Device->ZeDeviceModuleProperties->flags & ZE_DEVICE_MODULE_FLAG_FP16)
-      SupportedExtensions += ("cl_khr_fp16 ");
-    if (Device->ZeDeviceModuleProperties->flags & ZE_DEVICE_MODULE_FLAG_FP64)
-      SupportedExtensions += ("cl_khr_fp64 ");
     if (Device->ZeDeviceModuleProperties->flags &
         ZE_DEVICE_MODULE_FLAG_INT64_ATOMICS)
       // int64AtomicsSupported indicates support for both.

source/adapters/opencl/device.cpp

@@ -527,7 +527,8 @@ UR_APIEXPORT ur_result_t UR_APICALL urDeviceGetInfo(ur_device_handle_t hDevice,
           cl_adapter::cast<cl_device_id>(hDevice), {"cl_khr_fp16"}, Supported));
 
       if (!Supported) {
-        return UR_RESULT_ERROR_UNSUPPORTED_ENUMERATION;
+        // Report no support with an empty bitfield
+        return ReturnValue(ur_device_fp_capability_flags_t(0u));
       }
     }