Merge branch 'main' into cuda-malloc-managed-optimize

Author: JackAKirk

CMakeLists.txt

@@ -25,6 +25,7 @@ set(CMAKE_CXX_STANDARD 17)
 set(CMAKE_CXX_STANDARD_REQUIRED YES)
 
 # Build Options
+option(UR_BUILD_EXAMPLES "Build example applications." ON)
 option(UR_BUILD_TESTS "Build unit tests." ON)
 option(UR_BUILD_TOOLS "build ur tools" ON)
 option(UR_FORMAT_CPP_STYLE "format code style of C++ sources" OFF)
@@ -258,7 +259,9 @@ install(
     EXPORT ${PROJECT_NAME}-targets)
 
 add_subdirectory(source)
-add_subdirectory(examples)
+if(UR_BUILD_EXAMPLES)
+    add_subdirectory(examples)
+endif()
 if(UR_BUILD_TESTS)
     add_subdirectory(test)
 endif()

README.md

@@ -118,6 +118,7 @@ List of options provided by CMake:
 
 | Name | Description | Values | Default |
 | - | - | - | - |
+| UR_BUILD_EXAMPLES | Build example applications | ON/OFF | ON |
 | UR_BUILD_TESTS | Build the tests | ON/OFF | ON |
 | UR_BUILD_TOOLS | Build tools | ON/OFF | ON |
 | UR_FORMAT_CPP_STYLE | Format code style | ON/OFF | OFF |

include/ur_api.h

@@ -5302,7 +5302,9 @@ typedef enum ur_queue_info_t {
                                        ///< The reference count returned should be considered immediately stale.
                                        ///< It is unsuitable for general use in applications. This feature is
                                        ///< provided for identifying memory leaks.
-    UR_QUEUE_INFO_SIZE = 5,            ///< [uint32_t] The size of the queue
+    UR_QUEUE_INFO_SIZE = 5,            ///< [uint32_t] The size of the queue on the device. Only a valid query
+                                       ///< if the queue was created with the `ON_DEVICE` queue flag, otherwise
+                                       ///< `::urQueueGetInfo` will return `::UR_RESULT_ERROR_INVALID_QUEUE`.
     UR_QUEUE_INFO_EMPTY = 6,           ///< [::ur_bool_t] return true if the queue was empty at the time of the
                                        ///< query
     /// @cond
@@ -5317,7 +5319,8 @@ typedef uint32_t ur_queue_flags_t;
 typedef enum ur_queue_flag_t {
     UR_QUEUE_FLAG_OUT_OF_ORDER_EXEC_MODE_ENABLE = UR_BIT(0), ///< Enable/disable out of order execution
     UR_QUEUE_FLAG_PROFILING_ENABLE = UR_BIT(1),              ///< Enable/disable profiling
-    UR_QUEUE_FLAG_ON_DEVICE = UR_BIT(2),                     ///< Is a device queue
+    UR_QUEUE_FLAG_ON_DEVICE = UR_BIT(2),                     ///< Is a device queue. If this is enabled `OUT_OF_ORDER_EXEC_MODE_ENABLE`
+                                                             ///< must also be enabled.
     UR_QUEUE_FLAG_ON_DEVICE_DEFAULT = UR_BIT(3),             ///< Is the default queue for a device
     UR_QUEUE_FLAG_DISCARD_EVENTS = UR_BIT(4),                ///< Events will be discarded
     UR_QUEUE_FLAG_PRIORITY_LOW = UR_BIT(5),                  ///< Low priority queue
@@ -5362,7 +5365,7 @@ typedef enum ur_queue_flag_t {
 ///     - ::UR_RESULT_ERROR_INVALID_NULL_POINTER
 ///         + `propSize != 0 && pPropValue == NULL`
 ///         + `pPropValue == NULL && pPropSizeRet == NULL`
-///     - ::UR_RESULT_ERROR_INVALID_QUEUE
+///     - ::UR_RESULT_ERROR_INVALID_QUEUE - "If `hQueue` isn't a valid queue handle or if `propName` isn't supported by `hQueue`."
 ///     - ::UR_RESULT_ERROR_OUT_OF_HOST_MEMORY
 ///     - ::UR_RESULT_ERROR_OUT_OF_RESOURCES
 UR_APIEXPORT ur_result_t UR_APICALL

scripts/core/queue.yml

@@ -32,7 +32,10 @@ etors:
             The reference count returned should be considered immediately stale. 
             It is unsuitable for general use in applications. This feature is provided for identifying memory leaks.
     - name: SIZE
-      desc: "[uint32_t] The size of the queue"
+      desc: |
+            [uint32_t] The size of the queue on the device. Only a valid query
+            if the queue was created with the `ON_DEVICE` queue flag, otherwise
+            `$xQueueGetInfo` will return `$X_RESULT_ERROR_INVALID_QUEUE`.
     - name: EMPTY
       desc: "[$x_bool_t] return true if the queue was empty at the time of the query"
 --- #--------------------------------------------------------------------------
@@ -49,7 +52,7 @@ etors:
       desc: "Enable/disable profiling"
     - name: ON_DEVICE
       value: "$X_BIT(2)"
-      desc: "Is a device queue"
+      desc: "Is a device queue. If this is enabled `OUT_OF_ORDER_EXEC_MODE_ENABLE` must also be enabled."
     - name: ON_DEVICE_DEFAULT
       value: "$X_BIT(3)"
       desc: "Is the default queue for a device"
@@ -108,6 +111,7 @@ returns:
         - "`propSize != 0 && pPropValue == NULL`"
         - "`pPropValue == NULL && pPropSizeRet == NULL`"
     - $X_RESULT_ERROR_INVALID_QUEUE
+        - "If `hQueue` isn't a valid queue handle or if `propName` isn't supported by `hQueue`."
     - $X_RESULT_ERROR_OUT_OF_HOST_MEMORY
     - $X_RESULT_ERROR_OUT_OF_RESOURCES
 --- #--------------------------------------------------------------------------

source/adapters/cuda/device.cpp

@@ -569,10 +569,7 @@ UR_APIEXPORT ur_result_t UR_APICALL urDeviceGetInfo(ur_device_handle_t hDevice,
         ur_queue_flag_t(UR_QUEUE_FLAG_OUT_OF_ORDER_EXEC_MODE_ENABLE |
                         UR_QUEUE_FLAG_PROFILING_ENABLE));
   case UR_DEVICE_INFO_QUEUE_ON_DEVICE_PROPERTIES: {
-    // The mandated minimum capability:
-    ur_queue_flags_t Capability = UR_QUEUE_FLAG_PROFILING_ENABLE |
-                                  UR_QUEUE_FLAG_OUT_OF_ORDER_EXEC_MODE_ENABLE;
-    return ReturnValue(Capability);
+    return ReturnValue(0);
   }
   case UR_DEVICE_INFO_QUEUE_ON_HOST_PROPERTIES: {
     // The mandated minimum capability: