[UR] Deprecate UR_DEVICE_INFO_BFLOAT16

* Add support for deprecation of enum values in UR spec. * Deprecate UR_DEVICE_INFO_BFLOAT16. * Remove testing for UR_DEVICE_INFO_BFLOAT16. * Remove UR_DEVICE_INFO_BFLOAT16 support from OpenCL adapter.
intel · Feb 20, 2025 · a03e964 · a03e964
1 parent 7b9779e
commit a03e964
Show file tree

Hide file tree

Showing 10 changed files with 64 additions and 74 deletions.
diff --git a/unified-runtime/include/ur_api.h b/unified-runtime/include/ur_api.h
@@ -2143,8 +2143,8 @@ 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,
+  /// [::ur_bool_t][deprecated-value] support for bfloat16
+  UR_DEVICE_INFO_BFLOAT16 [[deprecated]] = 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.

diff --git a/unified-runtime/include/ur_print.hpp b/unified-runtime/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) {

diff --git a/unified-runtime/scripts/YaML.md b/unified-runtime/scripts/YaML.md
@@ -337,6 +337,7 @@ plural form *enumerators* is abbreviated to `etors`.
     + `desc` will be used as the etors's description comment
     + If the enum has `typed_etors`, `desc` must begin with type identifier: {`"[type]"`}
     + `desc` may contain the [optional-query] annotation. This denotes the etor as an info query which is optional for adapters to implement, and may legally result in a non-success error code.
+    + `desc` may contain the [deprecated-value] annotation. This marks the etor with the `[[deprecated]]` attribute specifier.
     + `name` must be a unique ISO-C standard identifier, and be all caps
   - An etor may take the following optional scalar field: {`value`, `version`}
     + `value` must be an ISO-C standard identifier

diff --git a/unified-runtime/scripts/core/device.yml b/unified-runtime/scripts/core/device.yml
@@ -409,7 +409,7 @@ etors:
     - 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"
+      desc: "[$x_bool_t][deprecated-value] support for bfloat16"
     - name: MAX_COMPUTE_QUEUE_INDICES
       desc: |
             [uint32_t] Returns 1 if the device doesn't have a notion of a

diff --git a/unified-runtime/scripts/templates/helper.py b/unified-runtime/scripts/templates/helper.py
@@ -565,6 +565,7 @@ def is_global(item, tags):
 
 class etor_traits:
     RE_OPTIONAL_QUERY = r".*\[optional-query\].*"
+    RE_DEPRECATED = r".*\[deprecated-value\].*"
 
     @classmethod
     def is_optional_query(cls, item):
@@ -573,6 +574,13 @@ def is_optional_query(cls, item):
         except:
             return False
 
+    @classmethod
+    def is_deprecated_etor(cls, item):
+        try:
+            return True if re.match(cls.RE_DEPRECATED, item["desc"]) else False
+        except:
+            return False
+
 
 """
 Public:
@@ -903,13 +911,14 @@ def make_etor_lines(namespace, tags, obj, meta=None):
     lines = []
     for item in obj["etors"]:
         name = make_etor_name(namespace, tags, obj["name"], item["name"], meta)
+        deprecated = " [[deprecated]]" if etor_traits.is_deprecated_etor(item) else ""
 
         if "value" in item:
             delim = ","
             value = _get_value_name(namespace, tags, item["value"])
-            prologue = "%s = %s%s" % (name, value, delim)
+            prologue = "%s%s = %s%s" % (name, deprecated, value, delim)
         else:
-            prologue = "%s," % (name)
+            prologue = "%s%s," % (name, deprecated)
 
         for line in split_line(subt(namespace, tags, item["desc"], True), 70):
             lines.append(" /// %s" % line)

diff --git a/unified-runtime/scripts/templates/print.hpp.mako b/unified-runtime/scripts/templates/print.hpp.mako
@@ -192,11 +192,13 @@ template <typename T> inline ${x}_result_t printTagged(std::ostream &os, const v
     inline std::ostream &operator<<(std::ostream &os, enum ${th.make_enum_name(n, tags, obj)} value) {
         switch (value) {
             %for n, item in enumerate(obj['etors']):
-                <%
-                ename = th.make_etor_name(n, tags, obj['name'], item['name'])
-                %>case ${ename}:
-                    os << "${ename}";
-                    break;
+                %if not th.etor_traits.is_deprecated_etor(item):
+                    <%
+                    ename = th.make_etor_name(n, tags, obj['name'], item['name'])
+                    %>case ${ename}:
+                        os << "${ename}";
+                        break;
+                %endif
             %endfor
                 default:
                     os << "unknown enumerator";
@@ -217,49 +219,51 @@ template <typename T> inline ${x}_result_t printTagged(std::ostream &os, const v
 
         switch (value) {
             %for n, item in enumerate(obj['etors']):
-                <%
-                ename = th.make_etor_name(n, tags, obj['name'], item['name'])
-                vtype = th.etor_get_associated_type(n, tags, item)
-                %>case ${ename}: {
-                    %if th.value_traits.is_array(vtype):
-                    <% atype = th.value_traits.get_array_name(vtype) %>
-                    %if 'void' in atype:
-                    const ${atype} const *tptr = (const ${atype} const*)ptr;
-                    %else:
-                    const ${atype} *tptr = (const ${atype} *)ptr;
-                    %endif
-                        %if "char" in atype: ## print char* arrays as simple NULL-terminated strings
-                            printPtr(os, tptr);
+                %if not th.etor_traits.is_deprecated_etor(item):
+                    <%
+                    ename = th.make_etor_name(n, tags, obj['name'], item['name'])
+                    vtype = th.etor_get_associated_type(n, tags, item)
+                    %>case ${ename}: {
+                        %if th.value_traits.is_array(vtype):
+                        <% atype = th.value_traits.get_array_name(vtype) %>
+                        %if 'void' in atype:
+                        const ${atype} const *tptr = (const ${atype} const*)ptr;
                         %else:
-                            os << "{";
-                            size_t nelems = size / sizeof(${atype});
-                            for (size_t i = 0; i < nelems; ++i) {
-                                if (i != 0) {
-                                    os << ", ";
+                        const ${atype} *tptr = (const ${atype} *)ptr;
+                        %endif
+                            %if "char" in atype: ## print char* arrays as simple NULL-terminated strings
+                                printPtr(os, tptr);
+                            %else:
+                                os << "{";
+                                size_t nelems = size / sizeof(${atype});
+                                for (size_t i = 0; i < nelems; ++i) {
+                                    if (i != 0) {
+                                        os << ", ";
+                                    }
+                                    <%call expr="member(tptr, atype, True)">
+                                        tptr[i]
+                                    </%call>
                                 }
-                                <%call expr="member(tptr, atype, True)">
-                                    tptr[i]
-                                </%call>
-                            }
-                            os << "}";
+                                os << "}";
+                            %endif
+                        %else:
+                        %if 'void' in vtype:
+                        const ${vtype} const *tptr = (const ${vtype} const *)ptr;
+                        %else:
+                        const ${vtype} *tptr = (const ${vtype} *)ptr;
                         %endif
-                    %else:
-                    %if 'void' in vtype:
-                    const ${vtype} const *tptr = (const ${vtype} const *)ptr;
-                    %else:
-                    const ${vtype} *tptr = (const ${vtype} *)ptr;
-                    %endif
-                    if (sizeof(${vtype}) > size) {
-                        os << "invalid size (is: " << size << ", expected: >=" << sizeof(${vtype}) << ")";
-                        return ${X}_RESULT_ERROR_INVALID_SIZE;
-                    }
-                    os << (const void *)(tptr) << " (";
-                    <%call expr="member(tptr, vtype, False)">
-                        *tptr
-                    </%call>
-                    os << ")";
-                    %endif
-                } break;
+                        if (sizeof(${vtype}) > size) {
+                            os << "invalid size (is: " << size << ", expected: >=" << sizeof(${vtype}) << ")";
+                            return ${X}_RESULT_ERROR_INVALID_SIZE;
+                        }
+                        os << (const void *)(tptr) << " (";
+                        <%call expr="member(tptr, vtype, False)">
+                            *tptr
+                        </%call>
+                        os << ")";
+                        %endif
+                    } break;
+                %endif
             %endfor
                 default:
                     os << "unknown enumerator";

diff --git a/unified-runtime/source/adapters/opencl/device.cpp b/unified-runtime/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);
   }

diff --git a/unified-runtime/test/conformance/device/urDeviceGetInfo.cpp b/unified-runtime/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,                     //

diff --git a/unified-runtime/test/conformance/testing/include/uur/utils.h b/unified-runtime/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,

diff --git a/unified-runtime/test/conformance/testing/source/utils.cpp b/unified-runtime/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>(