Refactor RDC documentation

Change-Id: Ieaba84992a8cbd185f4c2d1dc36a175c0429b754

docs/reference/api_intro.rst

@@ -1,96 +1,34 @@
 .. meta::
-  :description: documentation of the installation, configuration, and use of the ROCm Data Center tool 
-  :keywords: ROCm Data Center tool, RDC, ROCm, API, reference, data type, support
+  :description: The ROCm Data Center tool (RDC) addresses key infrastructure challenges regarding AMD GPUs in cluster and data center environments and simplifies their administration
+  :keywords: ROCm Data Center tool API, RDC API
 
 .. _api-intro:
 
-******************************************
+*************************
 Introduction to RDC API
-******************************************
+*************************
 
 .. note::
   This is the alpha version of RDC API and is subject to change without notice. The primary purpose of this API is to solicit feedback. AMD accepts no responsibility for any software breakage caused by API changes.
 
 RDC API
-===========
+========
 
-RDC API is the core library that provides all the RDC features. This section focuses on how RDC API can be used by third-party software.
+RDC API is the core library that provides all the RDC features.
 
-RDC includes the following libraries:
+RDC API includes the following libraries:
+
+* ``librdc_bootstrap.so``: Loads one of the following two libraries during runtime, depending on the mode.
+
+  - ``rdci`` mode: Loads ``librdc_client.so``
+  - ``rdcd`` mode: Loads ``librdc.so``
 
-* ``librdc_bootstrap.so``: Loads during runtime one of the two libraries by detecting the mode.
 * ``librdc_client.so``: Exposes RDC functionality using ``gRPC`` client.
-* ``librdc.so``: RDC API. This depends on ``libamd_smi.so``.
-* ``libamd_smi.so``: Stateless low overhead access to GPU data.
 
+* ``librdc.so``: RDC API. This depends on ``libamd_smi.so``.
+
+* ``libamd_smi.so``: Stateless low overhead access to GPU data.
 
 .. figure:: ../data/api_libs.png
 
     Different libraries and how they are linked.
-
-.. note:: 
-  ``librdc_bootstrap.so`` loads different libraries based on the modes.
-
-Example:
-
-* ``rdci``: ``librdc_bootstrap.so`` loads ``librdc_client.so``
-* ``rdcd``: ``librdc_bootstrap.so`` loads ``librdc.so``
-
-For more information see the :ref:`rdc-ref`.
-
-Job stats use case
-==================
-
-The following pseudocode shows how RDC API can be directly used to record GPU statistics associated with any job or workload. Refer to the example code provided with RDC on how to build it.
-
-For more information, see *Job Stats* in :ref:`rdc-features`.
-
-.. code-block:: shell
-  
-    //Initialize the RDC
-    rdc_handle_t rdc_handle;
-    rdc_status_t result=rdc_init(0);
-    
-    //Dynamically choose to run in standalone or embedded mode
-    bool standalone = false;
-    std::cin>> standalone;
-    if (standalone)
-      result = rdc_connect("127.0.0.1:50051", &rdc_handle, nullptr, nullptr, nullptr); //It will connect to the daemon
-    else
-      result = rdc_start_embedded(RDC_OPERATION_MODE_MANUAL, &rdc_handle); //call library directly, here we run embedded in manual mode
-    
-    //Now we can use the same API for both standalone and embedded
-    //(1) create group
-    rdc_gpu_group_t groupId;
-    result = rdc_group_gpu_create(rdc_handle, RDC_GROUP_EMPTY, "MyGroup1", &groupId);
-    
-    //(2) Add the GPUs to the group
-    result = rdc_group_gpu_add(rdc_handle, groupId, 0); //Add GPU 0
-    result = rdc_group_gpu_add(rdc_handle, groupId, 1); //Add GPU 1
-    
-    //(3) start the recording the Slurm job 123. Set the sample frequency to once per second
-    result = rdc_job_start_stats(rdc_handle, group_id,
-      "123", 1000000);
-    
-    //For standalone mode, the daemon will update and cache the samples
-    //In manual mode, we must call rdc_field_update_all periodically to take samples
-    if (!standalone) { //embedded manual mode
-      for (int i=5; i>0; i--) { //As an example, we will take 5 samples
-      result = rdc_field_update_all(rdc_handle, 0);
-      usleep(1000000);
-      }
-    } else { //standalone mode, do nothing
-      usleep(5000000); //sleep 5 seconds before fetch the stats
-    }
-    
-    //(4) stop the Slurm job 123, which will stop the watch
-    // Note: we do not have to stop the job to get stats. The rdc_job_get_stats can be called at any time before stop
-    result = rdc_job_stop_stats(rdc_handle, "123");
-    
-    //(5) Get the stats
-    rdc_job_info_t job_info;
-    result = rdc_job_get_stats(rdc_handle, "123", &job_info);
-    std::cout<<"Average Memory Utilization: " <<job_info.summary.memoryUtilization.average <<std::endl;
-    
-    //The cleanup and shutdown ....
-

docs/reference/api_ref.rst

@@ -1,11 +1,11 @@
 .. meta::
-  :description: documentation of the installation, configuration, and use of the ROCm Data Center tool 
-  :keywords: ROCm Data Center tool, RDC, ROCm, API, reference, data type, support
+  :description: The ROCm Data Center tool (RDC) addresses key infrastructure challenges regarding AMD GPUs in cluster and data center environments and simplifies their administration
+  :keywords: ROCm Data Center library, RDC library, RDC API, ROCm Data Center API
 
 .. _rdc-ref:
 
-******************************************
-API reference
-******************************************
+****************
+RDC API library
+****************
 
 .. doxygenindex::