Remove UR_DEVICE_INFO_BFLOAT16

Resolves #2385

include/ur_api.h

@@ -2142,57 +2142,55 @@ typedef enum ur_device_info_t {
   /// [::ur_memory_scope_capability_flags_t] return a bit-field of atomic
   /// memory fence scope capabilities
   UR_DEVICE_INFO_ATOMIC_FENCE_SCOPE_CAPABILITIES = 103,
-  /// [::ur_bool_t] support for bfloat16
-  UR_DEVICE_INFO_BFLOAT16 = 104,
   /// [uint32_t] Returns 1 if the device doesn't have a notion of a
   /// queue index. Otherwise, returns the number of queue indices that are
   /// available for this device.
-  UR_DEVICE_INFO_MAX_COMPUTE_QUEUE_INDICES = 105,
+  UR_DEVICE_INFO_MAX_COMPUTE_QUEUE_INDICES = 104,
   /// [::ur_bool_t] support the ::urKernelSetSpecializationConstants entry
   /// point
-  UR_DEVICE_INFO_KERNEL_SET_SPECIALIZATION_CONSTANTS = 106,
+  UR_DEVICE_INFO_KERNEL_SET_SPECIALIZATION_CONSTANTS = 105,
   /// [uint32_t][optional-query] return the width in bits of the memory bus
   /// interface of the device.
-  UR_DEVICE_INFO_MEMORY_BUS_WIDTH = 107,
+  UR_DEVICE_INFO_MEMORY_BUS_WIDTH = 106,
   /// [size_t[3]] return max 3D work groups
-  UR_DEVICE_INFO_MAX_WORK_GROUPS_3D = 108,
+  UR_DEVICE_INFO_MAX_WORK_GROUPS_3D = 107,
   /// [::ur_bool_t] return true if Async Barrier is supported
-  UR_DEVICE_INFO_ASYNC_BARRIER = 109,
+  UR_DEVICE_INFO_ASYNC_BARRIER = 108,
   /// [::ur_bool_t] return true if specifying memory channels is supported
-  UR_DEVICE_INFO_MEM_CHANNEL_SUPPORT = 110,
+  UR_DEVICE_INFO_MEM_CHANNEL_SUPPORT = 109,
   /// [::ur_bool_t] Return true if the device supports enqueueing commands
   /// to read and write pipes from the host.
-  UR_DEVICE_INFO_HOST_PIPE_READ_WRITE_SUPPORTED = 111,
+  UR_DEVICE_INFO_HOST_PIPE_READ_WRITE_SUPPORTED = 110,
   /// [uint32_t][optional-query] The maximum number of registers available
   /// per block.
-  UR_DEVICE_INFO_MAX_REGISTERS_PER_WORK_GROUP = 112,
+  UR_DEVICE_INFO_MAX_REGISTERS_PER_WORK_GROUP = 111,
   /// [uint32_t][optional-query] The device IP version. The meaning of the
   /// device IP version is implementation-defined, but newer devices should
   /// have a higher version than older devices.
-  UR_DEVICE_INFO_IP_VERSION = 113,
+  UR_DEVICE_INFO_IP_VERSION = 112,
   /// [::ur_bool_t] return true if the device supports virtual memory.
-  UR_DEVICE_INFO_VIRTUAL_MEMORY_SUPPORT = 114,
+  UR_DEVICE_INFO_VIRTUAL_MEMORY_SUPPORT = 113,
   /// [::ur_bool_t] return true if the device supports ESIMD.
-  UR_DEVICE_INFO_ESIMD_SUPPORT = 115,
+  UR_DEVICE_INFO_ESIMD_SUPPORT = 114,
   /// [::ur_device_handle_t[]][optional-query] The set of component devices
   /// contained by this composite device.
-  UR_DEVICE_INFO_COMPONENT_DEVICES = 116,
+  UR_DEVICE_INFO_COMPONENT_DEVICES = 115,
   /// [::ur_device_handle_t][optional-query] The composite device containing
   /// this component device.
-  UR_DEVICE_INFO_COMPOSITE_DEVICE = 117,
+  UR_DEVICE_INFO_COMPOSITE_DEVICE = 116,
   /// [::ur_bool_t] return true if the device supports the
   /// `EnqueueDeviceGlobalVariableWrite` and
   /// `EnqueueDeviceGlobalVariableRead` entry points.
-  UR_DEVICE_INFO_GLOBAL_VARIABLE_SUPPORT = 118,
+  UR_DEVICE_INFO_GLOBAL_VARIABLE_SUPPORT = 117,
   /// [::ur_bool_t] return true if the device supports USM pooling. Pertains
   /// to the `USMPool` entry points and usage of the `pool` parameter of the
   /// USM alloc entry points.
-  UR_DEVICE_INFO_USM_POOL_SUPPORT = 119,
+  UR_DEVICE_INFO_USM_POOL_SUPPORT = 118,
   /// [uint32_t] the number of compute units for specific backend.
-  UR_DEVICE_INFO_NUM_COMPUTE_UNITS = 120,
+  UR_DEVICE_INFO_NUM_COMPUTE_UNITS = 119,
   /// [::ur_bool_t] support the ::urProgramSetSpecializationConstants entry
   /// point
-  UR_DEVICE_INFO_PROGRAM_SET_SPECIALIZATION_CONSTANTS = 121,
+  UR_DEVICE_INFO_PROGRAM_SET_SPECIALIZATION_CONSTANTS = 120,
   /// [::ur_bool_t] Returns true if the device supports the use of
   /// command-buffers.
   UR_DEVICE_INFO_COMMAND_BUFFER_SUPPORT_EXP = 0x1000,

include/ur_print.hpp

@@ -2822,9 +2822,6 @@ inline std::ostream &operator<<(std::ostream &os, enum ur_device_info_t value) {
   case UR_DEVICE_INFO_ATOMIC_FENCE_SCOPE_CAPABILITIES:
     os << "UR_DEVICE_INFO_ATOMIC_FENCE_SCOPE_CAPABILITIES";
     break;
-  case UR_DEVICE_INFO_BFLOAT16:
-    os << "UR_DEVICE_INFO_BFLOAT16";
-    break;
   case UR_DEVICE_INFO_MAX_COMPUTE_QUEUE_INDICES:
     os << "UR_DEVICE_INFO_MAX_COMPUTE_QUEUE_INDICES";
     break;
@@ -4297,19 +4294,6 @@ inline ur_result_t printTagged(std::ostream &os, const void *ptr,
 
     os << ")";
   } break;
-  case UR_DEVICE_INFO_BFLOAT16: {
-    const ur_bool_t *tptr = (const ur_bool_t *)ptr;
-    if (sizeof(ur_bool_t) > size) {
-      os << "invalid size (is: " << size
-         << ", expected: >=" << sizeof(ur_bool_t) << ")";
-      return UR_RESULT_ERROR_INVALID_SIZE;
-    }
-    os << (const void *)(tptr) << " (";
-
-    os << *tptr;
-
-    os << ")";
-  } break;
   case UR_DEVICE_INFO_MAX_COMPUTE_QUEUE_INDICES: {
     const uint32_t *tptr = (const uint32_t *)ptr;
     if (sizeof(uint32_t) > size) {

scripts/core/device.yml

@@ -408,8 +408,6 @@ etors:
       desc: "[$x_memory_order_capability_flags_t] return a bit-field of atomic memory fence order capabilities"
     - name: ATOMIC_FENCE_SCOPE_CAPABILITIES
       desc: "[$x_memory_scope_capability_flags_t] return a bit-field of atomic memory fence scope capabilities"
-    - name: BFLOAT16
-      desc: "[$x_bool_t] support for bfloat16"
     - name: MAX_COMPUTE_QUEUE_INDICES
       desc: |
             [uint32_t] Returns 1 if the device doesn't have a notion of a

source/adapters/cuda/device.cpp

@@ -253,14 +253,6 @@ UR_APIEXPORT ur_result_t UR_APICALL urDeviceGetInfo(ur_device_handle_t hDevice,
         UR_MEMORY_SCOPE_CAPABILITY_FLAG_WORK_GROUP;
     return ReturnValue(Capabilities);
   }
-  case UR_DEVICE_INFO_BFLOAT16: {
-    int Major = 0;
-    UR_CHECK_ERROR(cuDeviceGetAttribute(
-        &Major, CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MAJOR, hDevice->get()));
-
-    bool BFloat16 = (Major >= 8) ? true : false;
-    return ReturnValue(BFloat16);
-  }
   case UR_DEVICE_INFO_SUB_GROUP_SIZES_INTEL: {
     // NVIDIA devices only support one sub-group size (the warp size)
     int WarpSize = 0;

source/adapters/hip/device.cpp

@@ -1052,8 +1052,6 @@ UR_APIEXPORT ur_result_t UR_APICALL urDeviceGetInfo(ur_device_handle_t hDevice,
 #else
     return ReturnValue(ur_bool_t{false});
 #endif
-  case UR_DEVICE_INFO_BFLOAT16:
-    return ReturnValue(true);
   case UR_DEVICE_INFO_ASYNC_BARRIER:
     return ReturnValue(false);
   case UR_DEVICE_INFO_IL_VERSION:

source/adapters/level_zero/device.cpp

@@ -903,10 +903,6 @@ ur_result_t urDeviceGetInfo(
     return ReturnValue(uint32_t{Device->ZeDeviceProperties->numEUsPerSubslice});
   case UR_DEVICE_INFO_GPU_HW_THREADS_PER_EU:
     return ReturnValue(uint32_t{Device->ZeDeviceProperties->numThreadsPerEU});
-  case UR_DEVICE_INFO_BFLOAT16: {
-    // bfloat16 math functions are not yet supported on Intel GPUs.
-    return ReturnValue(ur_bool_t{false});
-  }
   case UR_DEVICE_INFO_ATOMIC_MEMORY_SCOPE_CAPABILITIES: {
     // There are no explicit restrictions in L0 programming guide, so assume all
     // are supported

source/adapters/native_cpu/device.cpp

@@ -344,8 +344,6 @@ UR_APIEXPORT ur_result_t UR_APICALL urDeviceGetInfo(ur_device_handle_t hDevice,
     return ReturnValue(bool{0});
   case UR_DEVICE_INFO_ATOMIC_64:
     return ReturnValue(bool{1});
-  case UR_DEVICE_INFO_BFLOAT16:
-    return ReturnValue(bool{0});
   case UR_DEVICE_INFO_MEM_CHANNEL_SUPPORT:
     return ReturnValue(bool{0});
   case UR_DEVICE_INFO_IMAGE_SRGB:

source/adapters/opencl/device.cpp

@@ -1166,7 +1166,6 @@ UR_APIEXPORT ur_result_t UR_APICALL urDeviceGetInfo(ur_device_handle_t hDevice,
   // We can't query to check if these are supported, they will need to be
   // manually updated if support is ever implemented.
   case UR_DEVICE_INFO_KERNEL_SET_SPECIALIZATION_CONSTANTS:
-  case UR_DEVICE_INFO_BFLOAT16:
   case UR_DEVICE_INFO_ASYNC_BARRIER: {
     return ReturnValue(false);
   }

test/conformance/device/urDeviceGetInfo.cpp

@@ -107,7 +107,6 @@ static std::unordered_map<ur_device_info_t, size_t> device_info_size_map = {
      sizeof(ur_memory_order_capability_flags_t)},
     {UR_DEVICE_INFO_ATOMIC_MEMORY_SCOPE_CAPABILITIES,
      sizeof(ur_memory_scope_capability_flags_t)},
-    {UR_DEVICE_INFO_BFLOAT16, sizeof(ur_bool_t)},
     {UR_DEVICE_INFO_MAX_COMPUTE_QUEUE_INDICES, sizeof(uint32_t)},
     {UR_DEVICE_INFO_KERNEL_SET_SPECIALIZATION_CONSTANTS, sizeof(ur_bool_t)},
     {UR_DEVICE_INFO_PROGRAM_SET_SPECIALIZATION_CONSTANTS, sizeof(ur_bool_t)},
@@ -235,7 +234,6 @@ UUR_DEVICE_TEST_SUITE_WITH_PARAM(
         UR_DEVICE_INFO_BUILD_ON_SUBDEVICE,                     //
         UR_DEVICE_INFO_ATOMIC_64,                              //
         UR_DEVICE_INFO_ATOMIC_MEMORY_ORDER_CAPABILITIES,       //
-        UR_DEVICE_INFO_BFLOAT16,                               //
         UR_DEVICE_INFO_MAX_COMPUTE_QUEUE_INDICES,              //
         UR_DEVICE_INFO_MEMORY_BUS_WIDTH,                       //
         UR_DEVICE_INFO_MAX_WORK_GROUPS_3D,                     //

test/conformance/testing/include/uur/utils.h

@@ -405,7 +405,6 @@ GetDeviceMemoryOrderCapabilities(ur_device_handle_t device,
 ur_result_t
 GetDeviceMemoryScopeCapabilities(ur_device_handle_t device,
                                  ur_memory_scope_capability_flags_t &flags);
-ur_result_t GetDeviceBFloat16Support(ur_device_handle_t device, bool &support);
 ur_result_t GetDeviceMaxComputeQueueIndices(ur_device_handle_t device,
                                             uint32_t &max_indices);
 ur_result_t GetDeviceHostPipeRWSupported(ur_device_handle_t device,

test/conformance/testing/source/utils.cpp

@@ -625,10 +625,6 @@ GetDeviceMemoryScopeCapabilities(ur_device_handle_t device,
       device, UR_DEVICE_INFO_ATOMIC_MEMORY_SCOPE_CAPABILITIES, flags);
 }
 
-ur_result_t GetDeviceBFloat16Support(ur_device_handle_t device, bool &support) {
-  return GetDeviceInfo<bool>(device, UR_DEVICE_INFO_BFLOAT16, support);
-}
-
 ur_result_t GetDeviceMaxComputeQueueIndices(ur_device_handle_t device,
                                             uint32_t &max_indices) {
   return GetDeviceInfo<uint32_t>(

tools/urinfo/urinfo.hpp

@@ -291,8 +291,6 @@ inline void printDeviceInfos(ur_device_handle_t hDevice,
   printDeviceInfo<ur_memory_scope_capability_flags_t>(
       hDevice, UR_DEVICE_INFO_ATOMIC_FENCE_SCOPE_CAPABILITIES);
   std::cout << prefix;
-  printDeviceInfo<ur_bool_t>(hDevice, UR_DEVICE_INFO_BFLOAT16);
-  std::cout << prefix;
   printDeviceInfo<uint32_t>(hDevice, UR_DEVICE_INFO_MAX_COMPUTE_QUEUE_INDICES);
   std::cout << prefix;
   printDeviceInfo<ur_bool_t>(