[cherry-pick] use the required instruction to determine if the environment...

 [cherry-pick] use the required instruction to determine if the environment fits the sample code's required.  (#32766) (#33451)

1 put a instruction # required: gpu(for example) in the sample code
2 this piece of code will only run in the GPU-equipped CI pipelines, and be omitted in other pipelines.
3 the CI pipelines can specify its capacity by shell environment variable SAMPLE_CODE_TEST_CAPACITY

2.1 文档改版方案

see #32766 for more infomation

tools/check_file_diff_approvals.sh

@@ -52,7 +52,7 @@ API_FILES=("CMakeLists.txt"
            "python/paddle/fluid/tests/unittests/white_list/op_threshold_white_list.py"
            "python/paddle/fluid/tests/unittests/white_list/check_op_sequence_batch_1_input_white_list.py"
            "python/paddle/fluid/tests/unittests/white_list/no_grad_set_white_list.py"
-           "tools/wlist.json"
+           "tools/print_signatures.py"
            "tools/sampcd_processor.py"
            "paddle/scripts/paddle_build.bat"
            "tools/windows/run_unittests.sh"
@@ -80,11 +80,10 @@ function add_failed(){
     echo_list="${echo_list[@]}$1"
 }
 
-function run_test_sampcd_processor() {
+function run_tools_test() {
     CUR_PWD=$(pwd)
     cd ${PADDLE_ROOT}/tools
-    python test_sampcd_processor.py
-    python test_print_signatures.py
+    python $1
     cd ${CUR_PWD}
 }
 
@@ -141,12 +140,12 @@ for API_FILE in ${API_FILES[*]}; do
       elif [ "${API_FILE}" == "python/paddle/fluid/tests/unittests/white_list/no_grad_set_white_list.py" ];then
           echo_line="You must have one RD (Shixiaowei02 (Recommend), luotao1 or phlrain) approval for the python/paddle/fluid/tests/unittests/white_list/no_grad_set_white_list.py, which manages the white list of no_grad_set without value in operators. For more information, please refer to[https://github.com/PaddlePaddle/Paddle/wiki/It's-recommend-to-set-no_grad_set-to-be-None].\n"
           check_approval 1 39303645 6836917 43953930
-      elif [ "${API_FILE}" == "tools/wlist.json" ];then
-          echo_line="You must have one TPM (jzhang533) approval for the api whitelist for the tools/wlist.json.\n"
-          check_approval 1 29231
       elif [ "${API_FILE}" == "tools/sampcd_processor.py" ];then
           echo_line="test_sampcd_processor.py will be executed for changed sampcd_processor.py.\n"
-          run_test_sampcd_processor
+          run_tools_test test_sampcd_processor.py
+      elif [ "${API_FILE}" == "tools/print_signatures.py" ];then
+          echo_line="test_print_signatures.py will be executed for changed print_signatures.py.\n"
+          run_tools_test test_print_signatures.py
       elif [ "${API_FILE}" == "python/paddle/distributed/fleet/__init__.py" ]; then
 	      echo_line="You must have (fuyinno4 (Recommend), raindrops2sea) approval for ${API_FILE} changes"
 	      check_approval 1 35824027 38231817

tools/sampcd_processor.py

@@ -11,12 +11,20 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
+"""
+please make sure to run in the tools path
+usage: python sample_test.py {cpu or gpu} 
+    {cpu or gpu}: running in cpu version or gpu version
+
+for example, you can run cpu version python2 testing like this:
 
+    python sampcd_processor.py cpu 
+
+"""
 import os
 import sys
 import subprocess
 import multiprocessing
-import math
 import platform
 import inspect
 import json
@@ -24,16 +32,7 @@ import argparse
 import shutil
 import re
 import logging
-"""
-please make sure to run in the tools path
-usage: python sample_test.py {cpu or gpu} 
-    {cpu or gpu}: running in cpu version or gpu version
-
-for example, you can run cpu version python2 testing like this:
-
-    python sampcd_processor.py cpu 
-
-"""
+import time
 
 logger = logging.getLogger()
 if logger.handlers:
@@ -45,6 +44,7 @@ else:
 console.setFormatter(logging.Formatter("%(message)s"))
 
 RUN_ON_DEVICE = 'cpu'
+SAMPLE_CODE_TEST_CAPACITY = set()
 GPU_ID = 0
 methods = []
 whl_error = []
@@ -52,6 +52,15 @@ API_DEV_SPEC_FN = 'paddle/fluid/API_DEV.spec'
 API_PR_SPEC_FN = 'paddle/fluid/API_PR.spec'
 API_DIFF_SPEC_FN = 'dev_pr_diff_api.spec'
 SAMPLECODE_TEMPDIR = 'samplecode_temp'
+ENV_KEY_CODES_FRONTEND = 'CODES_INSERTED_INTO_FRONTEND'
+ENV_KEY_TEST_CAPACITY = 'SAMPLE_CODE_TEST_CAPACITY'
+SUMMARY_INFO = {
+    'success': [],
+    'failed': [],
+    'skiptest': [],
+    'nocodes': [],
+    # ... required not-match
+}
 
 
 def find_all(srcstr, substr):
@@ -75,32 +84,225 @@ def find_all(srcstr, substr):
     return indices
 
 
-def check_indent(cdline):
+def find_last_future_line_end(cbstr):
     """
-    to check the indent of a given code line
+    find the last `__future__` line.
 
-    to get the number of starting blank chars,
-    e.t. blankspaces and \t
+    Args:
+        docstr(str): docstring
+    Return:
+        index of the line end or None.
+    """
+    pat = re.compile('__future__.*\n')
+    lastmo = None
+    it = re.finditer(pat, cbstr)
+    while True:
+        try:
+            lastmo = next(it)
+        except StopIteration:
+            break
+    if lastmo:
+        return lastmo.end()
+    else:
+        return None
 
-    \t will be interpreted as 4 single blankspaces,
-    e.t. '\t'='    '
 
-    Args:
-        cdline(str) : a single line of code from the source file
+def extract_code_blocks_from_docstr(docstr):
+    """
+    extract code-blocks from the given docstring.
 
-    Returns:
-        int : the indent of the number of interpreted
-             blankspaces
+    DON'T include the multiline-string definition in code-blocks.
+    The *Examples* section must be the last.
+
+    Args:
+        docstr(str): docstring
+    Return:
+        code_blocks: A list of code-blocks, indent removed. 
+                     element {'name': the code-block's name, 'id': sequence id.
+                              'codes': codes, 'required': 'gpu'}
     """
-    indent = 0
-    for c in cdline:
-        if c == '\t':
-            indent += 4
-        elif c == ' ':
-            indent += 1
-        if c != ' ' and c != '\t':
+    code_blocks = []
+
+    mo = re.search(r"Examples:", docstr)
+    if mo is None:
+        return code_blocks
+    ds_list = docstr[mo.start():].replace("\t", '    ').split("\n")
+    lastlineindex = len(ds_list) - 1
+
+    cb_start_pat = re.compile(r"code-block::\s*python")
+    cb_param_pat = re.compile(r"^\s*:(\w+):\s*(\S*)\s*$")
+    cb_required_pat = re.compile(r"^\s*#\s*require[s|d]\s*:\s*(\S+)\s*$")
+
+    cb_info = {}
+    cb_info['cb_started'] = False
+    cb_info['cb_cur'] = []
+    cb_info['cb_cur_indent'] = -1
+    cb_info['cb_cur_name'] = None
+    cb_info['cb_cur_seq_id'] = 0
+    cb_info['cb_required'] = None
+
+    def _cb_started():
+        # nonlocal cb_started, cb_cur_name, cb_required, cb_cur_seq_id
+        cb_info['cb_started'] = True
+        cb_info['cb_cur_seq_id'] += 1
+        cb_info['cb_cur_name'] = None
+        cb_info['cb_required'] = None
+
+    def _append_code_block():
+        # nonlocal code_blocks, cb_cur, cb_cur_name, cb_cur_seq_id, cb_required
+        code_blocks.append({
+            'codes': inspect.cleandoc("\n".join(cb_info['cb_cur'])),
+            'name': cb_info['cb_cur_name'],
+            'id': cb_info['cb_cur_seq_id'],
+            'required': cb_info['cb_required'],
+        })
+
+    for lineno, linecont in enumerate(ds_list):
+        if re.search(cb_start_pat, linecont):
+            if not cb_info['cb_started']:
+                _cb_started()
+                continue
+            else:
+                # cur block end
+                if len(cb_info['cb_cur']):
+                    _append_code_block()
+                _cb_started()  # another block started
+                cb_info['cb_cur_indent'] = -1
+                cb_info['cb_cur'] = []
+        else:
+            if cb_info['cb_started']:
+                # handle the code-block directive's options
+                mo_p = cb_param_pat.match(linecont)
+                if mo_p:
+                    if mo_p.group(1) == 'name':
+                        cb_info['cb_cur_name'] = mo_p.group(2)
+                    continue
+                # read the required directive
+                mo_r = cb_required_pat.match(linecont)
+                if mo_r:
+                    cb_info['cb_required'] = mo_r.group(1)
+                # docstring end
+                if lineno == lastlineindex:
+                    mo = re.search(r"\S", linecont)
+                    if mo is not None and cb_info['cb_cur_indent'] <= mo.start(
+                    ):
+                        cb_info['cb_cur'].append(linecont)
+                    if len(cb_info['cb_cur']):
+                        _append_code_block()
                     break
-    return indent
+                # check indent for cur block start and end.
+                mo = re.search(r"\S", linecont)
+                if mo is None:
+                    continue
+                if cb_info['cb_cur_indent'] < 0:
+                    # find the first non empty line
+                    cb_info['cb_cur_indent'] = mo.start()
+                    cb_info['cb_cur'].append(linecont)
+                else:
+                    if cb_info['cb_cur_indent'] <= mo.start():
+                        cb_info['cb_cur'].append(linecont)
+                    else:
+                        if linecont[mo.start()] == '#':
+                            continue
+                        else:
+                            # block end
+                            if len(cb_info['cb_cur']):
+                                _append_code_block()
+                            cb_info['cb_started'] = False
+                            cb_info['cb_cur_indent'] = -1
+                            cb_info['cb_cur'] = []
+    return code_blocks
+
+
+def get_test_capacity():
+    """
+    collect capacities and set to SAMPLE_CODE_TEST_CAPACITY
+    """
+    global SAMPLE_CODE_TEST_CAPACITY  # write
+    global ENV_KEY_TEST_CAPACITY, RUN_ON_DEVICE  # readonly
+    if ENV_KEY_TEST_CAPACITY in os.environ:
+        for r in os.environ[ENV_KEY_TEST_CAPACITY].split(','):
+            rr = r.strip().lower()
+            if r:
+                SAMPLE_CODE_TEST_CAPACITY.add(rr)
+    if 'cpu' not in SAMPLE_CODE_TEST_CAPACITY:
+        SAMPLE_CODE_TEST_CAPACITY.add('cpu')
+
+    if RUN_ON_DEVICE:
+        SAMPLE_CODE_TEST_CAPACITY.add(RUN_ON_DEVICE)
+
+
+def is_required_match(requirestr, cbtitle='not-specified'):
+    """
+    search the required instruction in the code-block, and check it match the current running environment.
+    
+    environment values of equipped: cpu, gpu, xpu, distributed, skip
+    the 'skip' is the special flag to skip the test, so is_required_match will return False directly.
+
+    Args:
+        requirestr(str): the required string.
+        cbtitle(str): the title of the code-block.
+    returns:
+        True - yes, matched
+        False - not match
+        None - skipped  # trick
+    """
+    global SAMPLE_CODE_TEST_CAPACITY  # readonly
+    requires = set(['cpu'])
+    if requirestr:
+        for r in requirestr.split(','):
+            rr = r.strip().lower()
+            if rr:
+                requires.add(rr)
+    if 'skip' in requires or 'skiptest' in requires:
+        logger.info('%s: skipped', cbtitle)
+        return None
+
+    if all([
+            k in SAMPLE_CODE_TEST_CAPACITY for k in requires
+            if k not in ['skip', 'skiptest']
+    ]):
+        return True
+
+    logger.info('%s: the equipments [%s] not match the required [%s].', cbtitle,
+                ','.join(SAMPLE_CODE_TEST_CAPACITY), ','.join(requires))
+    return False
+
+
+def insert_codes_into_codeblock(codeblock, apiname='not-specified'):
+    """
+    insert some codes in the frontend and backend into the code-block.
+    """
+    global ENV_KEY_CODES_FRONTEND, GPU_ID, RUN_ON_DEVICE  # readonly
+    inserted_codes_f = ''
+    inserted_codes_b = ''
+    if ENV_KEY_CODES_FRONTEND in os.environ and os.environ[
+            ENV_KEY_CODES_FRONTEND]:
+        inserted_codes_f = os.environ[ENV_KEY_CODES_FRONTEND]
+    else:
+        cpu_str = '\nimport os\nos.environ["CUDA_VISIBLE_DEVICES"] = ""\n'
+        gpu_str = '\nimport os\nos.environ["CUDA_VISIBLE_DEVICES"] = "{}"\n'.format(
+            GPU_ID)
+        if 'required' in codeblock:
+            if codeblock['required'] is None or codeblock['required'] == 'cpu':
+                inserted_codes_f = cpu_str
+            elif codeblock['required'] == 'gpu':
+                inserted_codes_f = gpu_str
+        else:
+            if RUN_ON_DEVICE == "cpu":
+                inserted_codes_f = cpu_str
+            elif RUN_ON_DEVICE == "gpu":
+                inserted_codes_f = gpu_str
+    inserted_codes_b = '\nprint("{}\'s sample code (name:{}, id:{}) is executed successfully!")'.format(
+        apiname, codeblock['name'], codeblock['id'])
+
+    cb = codeblock['codes']
+    last_future_line_end = find_last_future_line_end(cb)
+    if last_future_line_end:
+        return cb[:last_future_line_end] + inserted_codes_f + cb[
+            last_future_line_end:] + inserted_codes_b
+    else:
+        return inserted_codes_f + cb + inserted_codes_b
 
 
 def sampcd_extract_to_file(srccom, name, htype="def", hname=""):
@@ -117,122 +319,111 @@ def sampcd_extract_to_file(srccom, name, htype="def", hname=""):
     Returns:
         sample_code_filenames(list of str)
     """
-    global GPU_ID, RUN_ON_DEVICE, SAMPLECODE_TEMPDIR
-    CODE_BLOCK_INTERDUCTORY = "code-block:: python"
+    global GPU_ID, RUN_ON_DEVICE, SAMPLECODE_TEMPDIR  # readonly
+    global SUMMARY_INFO  # update
 
-    sampcd_begins = find_all(srccom, CODE_BLOCK_INTERDUCTORY)
-    if len(sampcd_begins) == 0:
+    codeblocks = extract_code_blocks_from_docstr(srccom)
+    if len(codeblocks) == 0:
+        SUMMARY_INFO['nocodes'].append(name)
         # detect sample codes using >>> to format and consider this situation as wrong
-        print(htype, " name:", hname)
-        print("-----------------------")
+        logger.info(htype + " name:" + name)
+        logger.info("-----------------------")
         if srccom.find("Examples:") != -1:
-            print("----example code check----\n")
+            logger.info("----example code check----")
             if srccom.find(">>>") != -1:
-                print(
-                    "Deprecated sample code style:\n\n    Examples:\n\n        >>>codeline\n        >>>codeline\n\n\n ",
-                    "Please use '.. code-block:: python' to ",
-                    "format sample code.\n")
+                logger.warning(r"""Deprecated sample code style:
+    Examples:
+        >>>codeline
+        >>>codeline
+
+Please use '.. code-block:: python' to format the sample code.""")
                 return []
         else:
-            print("Error: No sample code!\n")
+            logger.warning("Error: No sample code!")
             return []
-    sample_code_filenames = []
-    for y in range(1, len(sampcd_begins) + 1):
-        sampcd_begin = sampcd_begins[y - 1]
-        sampcd = srccom[sampcd_begin + len(CODE_BLOCK_INTERDUCTORY) + 1:]
-        sampcd = sampcd.split("\n")
-        # remove starting empty lines
-        while sampcd[0].replace(' ', '').replace('\t', '') == '':
-            sampcd.pop(0)
-
-        # the minimum indent, which is the indent of the first
-        # non-empty line
-        min_indent = check_indent(sampcd[0])
-        sampcd_to_write = []
-        for i in range(0, len(sampcd)):
-            cdline = sampcd[i]
-            # handle empty lines or those only with spaces/tabs
-            if cdline.strip() == '':
-                continue
-            this_indent = check_indent(cdline)
-            if this_indent < min_indent:
-                break
-            else:
-                cdline = cdline.replace('\t', '    ')
-                sampcd_to_write.append(cdline[min_indent:])
-
-        sampcd = '\n'.join(sampcd_to_write)
-        if RUN_ON_DEVICE == "cpu":
-            sampcd = '\nimport os\nos.environ["CUDA_VISIBLE_DEVICES"] = ""\n' + sampcd
-        if RUN_ON_DEVICE == "gpu":
-            sampcd = '\nimport os\nos.environ["CUDA_VISIBLE_DEVICES"] = "{}"\n'.format(
-                GPU_ID) + sampcd
-        sampcd += '\nprint(' + '\"' + name + ' sample code is executed successfully!\")'
 
+    sample_code_filenames = []
+    for y, cb in enumerate(codeblocks):
+        matched = is_required_match(cb['required'], name)
+        # matched has three states:
+        # True - please execute it;
+        # None - no sample code found;
+        # False - it need other special equipment or environment.
+        # so, the following conditional statements are intentionally arranged.
+        if matched == True:
             tfname = os.path.join(SAMPLECODE_TEMPDIR, '{}_example{}'.format(
-            name, '.py' if len(sampcd_begins) == 1 else '_{}.py'.format(y)))
+                name, '.py'
+                if len(codeblocks) == 1 else '_{}.py'.format(y + 1)))
             with open(tfname, 'w') as tempf:
+                sampcd = insert_codes_into_codeblock(cb, name)
                 tempf.write(sampcd)
             sample_code_filenames.append(tfname)
+        elif matched is None:
+            logger.info('{}\' code block (name:{}, id:{}) is skipped.'.format(
+                name, cb['name'], cb['id']))
+            SUMMARY_INFO['skiptest'].append("{}-{}".format(name, cb['id']))
+        elif matched == False:
+            logger.info(
+                '{}\' code block (name:{}, id:{}) required({}) not match capacity({}).'.
+                format(name, cb['name'], cb['id'], cb['required'],
+                       SAMPLE_CODE_TEST_CAPACITY))
+            if cb['required'] not in SUMMARY_INFO:
+                SUMMARY_INFO[cb['required']] = []
+            SUMMARY_INFO[cb['required']].append("{}-{}".format(name, cb['id']))
+
     return sample_code_filenames
 
 
 def execute_samplecode(tfname):
     """
-    Execute a sample-code test.
+    Execute a sample-code test
 
     Args:
-        tfname: the filename of the samplecode.
+        tfname: the filename of the sample code
     
     Returns:
         result: success or not
         tfname: same as the input argument
-        msg: the stdout output of the samplecode executing.
+        msg: the stdout output of the sample code executing
+        time: time consumed by sample code
     """
     result = True
     msg = None
     if platform.python_version()[0] in ["2", "3"]:
         cmd = [sys.executable, tfname]
     else:
-        print("Error: fail to parse python version!")
+        logger.error("Error: fail to parse python version!")
         result = False
         exit(1)
 
-    # check required envisonment
-    with open(tfname, 'r') as f:
-        for line in f.readlines():
-            if re.match(r'#\s*required\s*:\s*(distributed|gpu|skip)', line):
-                result = True
-                return result, tfname, '{} is skipped. cause: {}'.format(tfname,
-                                                                         line)
-
-    logging.info('running %s', tfname)
-    print("\n----example code check----")
-    print("executing sample code .....", tfname)
+    logger.info("----example code check----")
+    logger.info("executing sample code: %s", tfname)
+    start_time = time.time()
     subprc = subprocess.Popen(
         cmd, stdout=subprocess.PIPE, stderr=subprocess.PIPE)
     output, error = subprc.communicate()
     msg = "".join(output.decode(encoding='utf-8'))
     err = "".join(error.decode(encoding='utf-8'))
+    end_time = time.time()
 
     if subprc.returncode != 0:
-        print("Sample code error found in ", tfname, ":")
-        print("-----------------------")
-        print(open(tfname).read())
-        print("-----------------------")
-        print("subprocess return code: ", str(subprc.returncode))
-        print("Error Raised from Sample Code ", tfname, " :")
-        print(err)
-        print(msg)
-        print("----example code check failed----\n")
-        logging.warning('%s error: %s', tfname, err)
-        logging.warning('%s msg: %s', tfname, msg)
+        with open(tfname, 'r') as f:
+            logger.warning("""Sample code error found in %s:
+-----------------------
+%s
+-----------------------
+subprocess return code: %d
+Error Raised from Sample Code:
+stderr: %s
+stdout: %s
+""", tfname, f.read(), subprc.returncode, err, msg)
+        logger.info("----example code check failed----")
         result = False
     else:
-        print("----example code check success----\n")
+        logger.info("----example code check success----")
 
     # msg is the returned code execution report
-    return result, tfname, msg
+    return result, tfname, msg, end_time - start_time
 
 
 def get_filenames():
@@ -317,35 +508,6 @@ def get_incrementapi():
                 f.write('\n')
 
 
-def get_wlist(fn="wlist.json"):
-    '''
-    this function will get the white list of API.
-
-    Returns:
-
-        wlist: a list of API that should not trigger the example check .
-
-    '''
-    wlist = []
-    wlist_file = []
-    # only white on CPU
-    gpu_not_white = []
-    with open(fn, 'r') as load_f:
-        load_dict = json.load(load_f)
-        for key in load_dict:
-            if key == 'wlist_dir':
-                for item in load_dict[key]:
-                    wlist_file.append(item["name"])
-            elif key == "gpu_not_white":
-                gpu_not_white = load_dict[key]
-            elif key == "wlist_api":
-                for item in load_dict[key]:
-                    wlist.append(item["name"])
-            else:
-                wlist = wlist + load_dict[key]
-    return wlist, wlist_file, gpu_not_white
-
-
 arguments = [
     # flags, dest, type, default, help
     ['--gpu_id', 'gpu_id', int, 0, 'GPU device id to use [0]'],
@@ -391,18 +553,15 @@ if __name__ == '__main__':
             ))
         logger.addHandler(logfHandler)
 
-    wlist, wlist_file, gpu_not_white = get_wlist()
-
     if args.mode == "gpu":
         GPU_ID = args.gpu_id
         logger.info("using GPU_ID %d", GPU_ID)
-        for _gnw in gpu_not_white:
-            wlist.remove(_gnw)
     elif args.mode != "cpu":
         logger.error("Unrecognized argument:%s, 'cpu' or 'gpu' is desired.",
                      args.mode)
         sys.exit("Invalid arguments")
     RUN_ON_DEVICE = args.mode
+    get_test_capacity()
     logger.info("API check -- Example Code")
     logger.info("sample_test running under python %s",
                 platform.python_version())
@@ -449,19 +608,50 @@ if __name__ == '__main__':
             if not temp[0]:
                 logger.info("In addition, mistakes found in sample codes: %s",
                             temp[1])
-                logger.info("error_methods: %s", str(temp[2]))
         logger.info("----------------------------------------------------")
         exit(1)
     else:
-        has_error = False
+        timeovered_test = {}
         for temp in result:
             if not temp[0]:
                 logger.info("In addition, mistakes found in sample codes: %s",
                             temp[1])
-                logger.info("error_methods: %s", str(temp[2]))
-                has_error = True
-        if has_error:
-            logger.info("Mistakes found in sample codes.")
-            logger.info("Please check sample codes.")
+                SUMMARY_INFO['failed'].append(temp[1])
+            else:
+                SUMMARY_INFO['success'].append(temp[1])
+            if temp[3] > 10:
+                timeovered_test[temp[1]] = temp[3]
+
+        if len(timeovered_test):
+            logger.info("%d sample codes ran time over 10s",
+                        len(timeovered_test))
+            if args.debug:
+                for k, v in timeovered_test.items():
+                    logger.info('{} - {}s'.format(k, v))
+        if len(SUMMARY_INFO['success']):
+            logger.info("%d sample codes ran success",
+                        len(SUMMARY_INFO['success']))
+        for k, v in SUMMARY_INFO.items():
+            if k not in ['success', 'failed', 'skiptest', 'nocodes']:
+                logger.info("%d sample codes required not match for %s",
+                            len(v), k)
+        if len(SUMMARY_INFO['skiptest']):
+            logger.info("%d sample codes skipped",
+                        len(SUMMARY_INFO['skiptest']))
+            if args.debug:
+                logger.info('\n'.join(SUMMARY_INFO['skiptest']))
+        if len(SUMMARY_INFO['nocodes']):
+            logger.info("%d apis don't have sample codes",
+                        len(SUMMARY_INFO['nocodes']))
+            if args.debug:
+                logger.info('\n'.join(SUMMARY_INFO['nocodes']))
+        if len(SUMMARY_INFO['failed']):
+            logger.info("%d sample codes ran failed",
+                        len(SUMMARY_INFO['failed']))
+            logger.info('\n'.join(SUMMARY_INFO['failed']))
+            logger.info(
+                "Mistakes found in sample codes. Please recheck the sample codes."
+            )
             exit(1)
+
     logger.info("Sample code check is successful!")

tools/test_sampcd_processor.py

@@ -20,15 +20,18 @@ import tempfile
 import shutil
 import sys
 import importlib
+import re
+import sampcd_processor
 from sampcd_processor import find_all
-from sampcd_processor import check_indent
 from sampcd_processor import get_api_md5
 from sampcd_processor import get_incrementapi
-from sampcd_processor import get_wlist
 from sampcd_processor import sampcd_extract_to_file
+from sampcd_processor import extract_code_blocks_from_docstr
 from sampcd_processor import execute_samplecode
-
-SAMPLECODE_TEMP_DIR = 'samplecode_temp'
+from sampcd_processor import find_last_future_line_end
+from sampcd_processor import insert_codes_into_codeblock
+from sampcd_processor import get_test_capacity
+from sampcd_processor import is_required_match
 
 
 class Test_find_all(unittest.TestCase):
@@ -43,27 +46,246 @@ class Test_find_all(unittest.TestCase):
                              find_all(' hello, world; hello paddle!', 'hello'))
 
 
-class Test_check_indent(unittest.TestCase):
-    def test_no_indent(self):
-        self.assertEqual(0, check_indent('hello paddle'))
+class Test_find_last_future_line_end(unittest.TestCase):
+    def test_no_instant(self):
+        samplecodes = """
+                print(10//3)
+        """
+        self.assertIsNone(find_last_future_line_end(samplecodes))
+
+    def test_1_instant(self):
+        samplecodes = """
+                from __future__ import print_function
+
+                print(10//3)
+        """
+        mo = re.search("print_function\n", samplecodes)
+        self.assertIsNotNone(mo)
+        self.assertGreaterEqual(
+            find_last_future_line_end(samplecodes), mo.end())
+
+    def test_2_instant(self):
+        samplecodes = """
+                from __future__ import print_function
+                from __future__ import division
+
+                print(10//3)
+        """
+        mo = re.search("division\n", samplecodes)
+        self.assertIsNotNone(mo)
+        self.assertGreaterEqual(
+            find_last_future_line_end(samplecodes), mo.end())
+
+
+class Test_extract_code_blocks_from_docstr(unittest.TestCase):
+    def test_no_samplecode(self):
+        docstr = """
+        placeholder
+        """
+        codeblocks = extract_code_blocks_from_docstr(docstr)
+        self.assertListEqual([], codeblocks)
+
+    def test_codeblock_before_examples_is_ignored(self):
+        docstr = """
+            .. code-block:: python
+
+                print(1+1)
+        Examples:
+        """
+        codeblocks = extract_code_blocks_from_docstr(docstr)
+        self.assertListEqual(codeblocks, [])
+
+    def test_1_samplecode(self):
+        docstr = """
+        Examples:
+            .. code-block:: python
+
+                print(1+1)
+        """
+        codeblocks = extract_code_blocks_from_docstr(docstr)
+        self.assertListEqual(codeblocks, [{
+            'codes': """print(1+1)""",
+            'name': None,
+            'id': 1,
+            'required': None,
+        }])
+
+    def test_2_samplecodes(self):
+        docstr = """
+        placeholder
+        Examples:
+            .. code-block:: python
+
+                print(1/0)
+
+            .. code-block:: python
+               :name: one_plus_one
+               :linenos:
+
+                # required: gpu
+                print(1+1)
+        """
+        codeblocks = extract_code_blocks_from_docstr(docstr)
+        self.assertListEqual(codeblocks, [{
+            'codes': """print(1/0)""",
+            'name': None,
+            'id': 1,
+            'required': None,
+        }, {
+            'codes': """# required: gpu
+print(1+1)""",
+            'name': 'one_plus_one',
+            'id': 2,
+            'required': 'gpu',
+        }])
+
+
+class Test_insert_codes_into_codeblock(unittest.TestCase):
+    def test_required_None(self):
+        codeblock = {
+            'codes': """print(1/0)""",
+            'name': None,
+            'id': 1,
+            'required': None,
+        }
+        self.assertEqual("""
+import os
+os.environ["CUDA_VISIBLE_DEVICES"] = ""
+print(1/0)
+print("not-specified's sample code (name:None, id:1) is executed successfully!")""",
+                         insert_codes_into_codeblock(codeblock))
+
+    def test_required_gpu(self):
+        codeblock = {
+            'codes': """# required: gpu
+print(1+1)""",
+            'name': None,
+            'id': 1,
+            'required': 'gpu',
+        }
+        self.assertEqual("""
+import os
+os.environ["CUDA_VISIBLE_DEVICES"] = "0"
+# required: gpu
+print(1+1)
+print("not-specified's sample code (name:None, id:1) is executed successfully!")""",
+                         insert_codes_into_codeblock(codeblock))
+
+    def test_from_future(self):
+        codeblock = {
+            'codes': """
+from __future__ import print_function
+from __future__ import division
+print(10//3)""",
+            'name': 'future',
+            'id': 1,
+            'required': None,
+        }
+        self.assertEqual("""
+from __future__ import print_function
+from __future__ import division
+
+import os
+os.environ["CUDA_VISIBLE_DEVICES"] = ""
+print(10//3)
+print("not-specified's sample code (name:future, id:1) is executed successfully!")""",
+                         insert_codes_into_codeblock(codeblock))
+
+
+def clear_capacity():
+    sampcd_processor.SAMPLE_CODE_TEST_CAPACITY = set()
+    sampcd_processor.RUN_ON_DEVICE = 'cpu'
+    if sampcd_processor.ENV_KEY_TEST_CAPACITY in os.environ:
+        del os.environ[sampcd_processor.ENV_KEY_TEST_CAPACITY]
 
-    def test_indent_4_spaces(self):
-        self.assertEqual(4, check_indent('    hello paddle'))
 
-    def test_indent_1_tab(self):
-        self.assertEqual(4, check_indent("\thello paddle"))
+class Test_get_test_capacity(unittest.TestCase):
+    def setUp(self):
+        clear_capacity()
+        get_test_capacity()
+
+    def tearDown(self):
+        clear_capacity()
+        get_test_capacity()
+
+    def test_NoEnvVar(self):
+        clear_capacity()
+        get_test_capacity()
+        self.assertCountEqual(['cpu', ],
+                              sampcd_processor.SAMPLE_CODE_TEST_CAPACITY)
+
+    def test_NoEnvVar_RUN_ON_DEVICE_gpu(self):
+        clear_capacity()
+        sampcd_processor.RUN_ON_DEVICE = 'gpu'
+        get_test_capacity()
+        self.assertCountEqual(['cpu', 'gpu'],
+                              sampcd_processor.SAMPLE_CODE_TEST_CAPACITY)
+
+    def test_EnvVar_gpu(self):
+        clear_capacity()
+        os.environ[sampcd_processor.ENV_KEY_TEST_CAPACITY] = 'gpu'
+        get_test_capacity()
+        self.assertCountEqual(['cpu', 'gpu'],
+                              sampcd_processor.SAMPLE_CODE_TEST_CAPACITY)
+
+    def test_EnvVar_gpu_and_distributed(self):
+        clear_capacity()
+        os.environ[sampcd_processor.ENV_KEY_TEST_CAPACITY] = 'gpu,distributed'
+        get_test_capacity()
+        self.assertCountEqual(['cpu', 'gpu', 'distributed'],
+                              sampcd_processor.SAMPLE_CODE_TEST_CAPACITY)
+
+
+class Test_is_required_match(unittest.TestCase):
+    def setUp(self):
+        clear_capacity()
+
+    def tearDown(self):
+        clear_capacity()
+        get_test_capacity()
+
+    def test_alldefault(self):
+        clear_capacity()
+        get_test_capacity()
+        self.assertTrue(is_required_match(''))
+        self.assertTrue(is_required_match(None))
+        self.assertTrue(is_required_match('cpu'))
+        self.assertFalse(is_required_match('gpu'))
+        self.assertIsNone(is_required_match('skiptest'))
+        self.assertIsNone(is_required_match('skip'))
+        self.assertIsNone(is_required_match('cpu,skiptest'))
+
+    def test_gpu_equipped(self):
+        clear_capacity()
+        os.environ[sampcd_processor.ENV_KEY_TEST_CAPACITY] = 'gpu'
+        get_test_capacity()
+        self.assertTrue(is_required_match('cpu'))
+        self.assertTrue(is_required_match('gpu'))
+        self.assertTrue(is_required_match('gpu,cpu'))
+        self.assertIsNone(is_required_match('skiptest'))
+        self.assertFalse(is_required_match('distributed'))
+
+    def test_gpu_distributed_equipped(self):
+        clear_capacity()
+        os.environ[sampcd_processor.ENV_KEY_TEST_CAPACITY] = 'gpu,distributed'
+        get_test_capacity()
+        self.assertTrue(is_required_match('cpu'))
+        self.assertTrue(is_required_match('gpu'))
+        self.assertTrue(is_required_match('distributed'))
+        self.assertFalse(is_required_match('xpu'))
+        self.assertIsNone(is_required_match('skiptest'))
 
 
 class Test_execute_samplecode(unittest.TestCase):
     def setUp(self):
-        if not os.path.exists(SAMPLECODE_TEMP_DIR):
-            os.mkdir(SAMPLECODE_TEMP_DIR)
-        self.successSampleCodeFile = os.path.join(SAMPLECODE_TEMP_DIR,
-                                                  'samplecode_success.py')
+        if not os.path.exists(sampcd_processor.SAMPLECODE_TEMPDIR):
+            os.mkdir(sampcd_processor.SAMPLECODE_TEMPDIR)
+        self.successSampleCodeFile = os.path.join(
+            sampcd_processor.SAMPLECODE_TEMPDIR, 'samplecode_success.py')
         with open(self.successSampleCodeFile, 'w') as f:
             f.write('print(1+1)')
-        self.failedSampleCodeFile = os.path.join(SAMPLECODE_TEMP_DIR,
-                                                 'samplecode_failed.py')
+        self.failedSampleCodeFile = os.path.join(
+            sampcd_processor.SAMPLECODE_TEMPDIR, 'samplecode_failed.py')
         with open(self.failedSampleCodeFile, 'w') as f:
             f.write('print(1/0)')
 
@@ -72,37 +294,41 @@ class Test_execute_samplecode(unittest.TestCase):
         os.remove(self.failedSampleCodeFile)
 
     def test_run_success(self):
-        result, tfname, msg = execute_samplecode(self.successSampleCodeFile)
+        result, tfname, msg, exec_time = execute_samplecode(
+            self.successSampleCodeFile)
         self.assertTrue(result)
         self.assertEqual(self.successSampleCodeFile, tfname)
         self.assertIsNotNone(msg)
         self.assertLess(msg.find('skipped'), 0)
+        self.assertLess(exec_time, 10)
 
     def test_run_failed(self):
-        result, tfname, msg = execute_samplecode(self.failedSampleCodeFile)
+        result, tfname, msg, exec_time = execute_samplecode(
+            self.failedSampleCodeFile)
         self.assertFalse(result)
         self.assertEqual(self.failedSampleCodeFile, tfname)
         self.assertIsNotNone(msg)
         self.assertLess(msg.find('skipped'), 0)
+        self.assertLess(exec_time, 10)
 
-    def test_testcases_skipped(self):
-        ...
-        tfname = os.path.join(SAMPLECODE_TEMP_DIR, 'samplecode_skipped.py')
-        with open(tfname, 'w') as f:
-            f.write("# required: distributed\nprint(1/0)")
-        result, _, msg = execute_samplecode(tfname)
-        self.assertTrue(result)
-        self.assertGreaterEqual(msg.find('skipped'), 0)
-        os.remove(tfname)
+
+def clear_summary_info():
+    for k in sampcd_processor.SUMMARY_INFO.keys():
+        sampcd_processor.SUMMARY_INFO[k].clear()
 
 
 class Test_sampcd_extract_to_file(unittest.TestCase):
     def setUp(self):
-        if not os.path.exists(SAMPLECODE_TEMP_DIR):
-            os.mkdir(SAMPLECODE_TEMP_DIR)
+        if not os.path.exists(sampcd_processor.SAMPLECODE_TEMPDIR):
+            os.mkdir(sampcd_processor.SAMPLECODE_TEMPDIR)
+        clear_capacity()
+        os.environ[sampcd_processor.ENV_KEY_TEST_CAPACITY] = 'gpu,distributed'
+        get_test_capacity()
 
     def tearDown(self):
-        shutil.rmtree(SAMPLECODE_TEMP_DIR)
+        shutil.rmtree(sampcd_processor.SAMPLECODE_TEMPDIR)
+        clear_capacity()
+        get_test_capacity()
 
     def test_1_samplecode(self):
         comments = """
@@ -113,9 +339,10 @@ class Test_sampcd_extract_to_file(unittest.TestCase):
         """
         funcname = 'one_plus_one'
         sample_code_filenames = sampcd_extract_to_file(comments, funcname)
-        self.assertCountEqual(
-            [os.path.join(SAMPLECODE_TEMP_DIR, funcname + '_example.py')],
-            sample_code_filenames)
+        self.assertCountEqual([
+            os.path.join(sampcd_processor.SAMPLECODE_TEMPDIR,
+                         funcname + '_example.py')
+        ], sample_code_filenames)
 
     def test_no_samplecode(self):
         comments = """
@@ -140,10 +367,64 @@ class Test_sampcd_extract_to_file(unittest.TestCase):
         funcname = 'one_plus_one'
         sample_code_filenames = sampcd_extract_to_file(comments, funcname)
         self.assertCountEqual([
-            os.path.join(SAMPLECODE_TEMP_DIR, funcname + '_example_1.py'),
-            os.path.join(SAMPLECODE_TEMP_DIR, funcname + '_example_2.py')
+            os.path.join(sampcd_processor.SAMPLECODE_TEMPDIR,
+                         funcname + '_example_1.py'),
+            os.path.join(sampcd_processor.SAMPLECODE_TEMPDIR,
+                         funcname + '_example_2.py')
         ], sample_code_filenames)
 
+    def test_2_samplecodes_has_skipped(self):
+        comments = """
+        placeholder
+        Examples:
+            .. code-block:: python
+
+                # required: skiptest
+                print(1/0)
+
+            .. code-block:: python
+
+                print(1+1)
+
+            .. code-block:: python
+
+                # required: gpu
+                print(1//1)
+
+            .. code-block:: python
+
+                # required: xpu
+                print(1//1)
+
+            .. code-block:: python
+
+                # required: distributed
+                print(1//1)
+
+            .. code-block:: python
+
+                # required: gpu
+                print(1//1)
+        """
+        funcname = 'one_plus_one'
+        clear_summary_info()
+        clear_capacity()
+        get_test_capacity()
+
+        sample_code_filenames = sampcd_extract_to_file(comments, funcname)
+        self.assertCountEqual([
+            os.path.join(sampcd_processor.SAMPLECODE_TEMPDIR,
+                         funcname + '_example_2.py')
+        ], sample_code_filenames)
+        self.assertCountEqual(sampcd_processor.SUMMARY_INFO['skiptest'],
+                              [funcname + '-1'])
+        self.assertCountEqual(sampcd_processor.SUMMARY_INFO['gpu'],
+                              [funcname + '-3', funcname + '-6'])
+        self.assertCountEqual(sampcd_processor.SUMMARY_INFO['xpu'],
+                              [funcname + '-4'])
+        self.assertCountEqual(sampcd_processor.SUMMARY_INFO['distributed'],
+                              [funcname + '-5'])
+
 
 class Test_get_api_md5(unittest.TestCase):
     def setUp(self):
@@ -208,55 +489,6 @@ class Test_get_incrementapi(unittest.TestCase):
             ], lines)
 
 
-class Test_get_wlist(unittest.TestCase):
-    def setUp(self):
-        self.tmpDir = tempfile.mkdtemp()
-        self.wlist_filename = os.path.join(self.tmpDir, 'wlist.json')
-        with open(self.wlist_filename, 'w') as f:
-            f.write(r'''
-{
-    "wlist_dir":[
-        {
-            "name":"../python/paddle/fluid/contrib",
-            "annotation":""
-        },
-        {
-            "name":"../python/paddle/verison.py",
-            "annotation":""
-        }
-    ],
-    "wlist_api":[
-        {
-            "name":"xxxxx",
-            "annotation":"not a real api, just for example"
-        }
-    ],
-    "wlist_temp_api":[
-        "to_tensor",
-        "save_persistables@dygraph/checkpoint.py"
-    ],
-    "gpu_not_white":[
-        "deformable_conv"
-    ]
-}
-''')
-
-    def tearDown(self):
-        os.remove(self.wlist_filename)
-        shutil.rmtree(self.tmpDir)
-
-    def test_get_wlist(self):
-        wlist, wlist_file, gpu_not_white = get_wlist(self.wlist_filename)
-        self.assertCountEqual(
-            ["xxxxx", "to_tensor",
-             "save_persistables@dygraph/checkpoint.py"], wlist)
-        self.assertCountEqual([
-            "../python/paddle/fluid/contrib",
-            "../python/paddle/verison.py",
-        ], wlist_file)
-        self.assertCountEqual(["deformable_conv"], gpu_not_white)
-
-
 # https://github.com/PaddlePaddle/Paddle/blob/develop/python/paddle/fluid/layers/ops.py
 # why? unabled to use the ast module. emmmmm
 

tools/wlist.json

-{
-    "wlist_dir":[
-        {
-            "name":"../python/paddle/fluid/contrib",
-            "annotation":""
-        },
-        {
-            "name":"../python/paddle/verison.py",
-            "annotation":""
-        },
-        {
-            "name":"../python/paddle/fluid/core_avx.py",
-            "annotation":""
-        },
-        {
-            "name":"../python/paddle/distributed",
-            "annotation":""
-        }
-    ],
-    "wlist_api":[
-        {
-            "name":"xxxxx",
-            "annotation":"not a real api, just for example"
-        },
-        {
-            "name":"squeeze_",
-            "annotation":"Inplace APIs don't need sample code. There is a special document introducing Inplace strategy"
-        },
-        {
-            "name":"unsqueeze_",
-            "annotation":"Inplace APIs don't need sample code. There is a special document introducing Inplace strategy"
-        },
-        {
-            "name":"reshape_",
-            "annotation":"Inplace APIs don't need sample code. There is a special document introducing Inplace strategy"
-        },
-        {
-            "name":"flatten_",
-            "annotation":"Inplace APIs don't need sample code. There is a special document introducing Inplace strategy"
-        },
-        {
-            "name":"scatter_",
-            "annotation":"Inplace APIs don't need sample code. There is a special document introducing Inplace strategy"
-        },
-        {
-            "name":"elu_",
-            "annotation":"Inplace APIs don't need sample code. There is a special document introducing Inplace strategy"
-        },
-        {
-            "name":"relu_",
-            "annotation":"Inplace APIs don't need sample code. There is a special document introducing Inplace strategy"
-        },
-        {
-            "name":"softmax_",
-            "annotation":"Inplace APIs don't need sample code. There is a special document introducing Inplace strategy"
-        },
-        {
-            "name":"tanh_",
-            "annotation":"Inplace APIs don't need sample code. There is a special document introducing Inplace strategy"
-        },
-        {
-            "name":"ceil_",
-            "annotation":"Inplace APIs don't need sample code. There is a special document introducing Inplace strategy"
-        },
-        {
-            "name":"floor_",
-            "annotation":"Inplace APIs don't need sample code. There is a special document introducing Inplace strategy"
-        },
-        {
-            "name":"exp_",
-            "annotation":"Inplace APIs don't need sample code. There is a special document introducing Inplace strategy"
-        },
-        {
-            "name":"reciprocal_",
-            "annotation":"Inplace APIs don't need sample code. There is a special document introducing Inplace strategy"
-        },
-        {
-            "name":"round_",
-            "annotation":"Inplace APIs don't need sample code. There is a special document introducing Inplace strategy"
-        },
-        {
-            "name":"sqrt_",
-            "annotation":"Inplace APIs don't need sample code. There is a special document introducing Inplace strategy"
-        },
-        {
-            "name":"rsqrt_",
-            "annotation":"Inplace APIs don't need sample code. There is a special document introducing Inplace strategy"
-        },
-        {
-            "name":"clip_",
-            "annotation":"Inplace APIs don't need sample code. There is a special document introducing Inplace strategy"
-        },
-        {
-            "name":"scale_",
-            "annotation":"Inplace APIs don't need sample code. There is a special document introducing Inplace strategy"
-        },
-        {
-            "name":"subtract_",
-            "annotation":"Inplace APIs don't need sample code. There is a special document introducing Inplace strategy"
-        },
-        {
-            "name":"add_",
-            "annotation":"Inplace APIs don't need sample code. There is a special document introducing Inplace strategy"
-        }
-    ],
-    "wlist_temp_api":[
-        "to_tensor",
-        "LRScheduler",
-        "ReduceOnPlateau",
-        "append_LARS",
-        "BuildStrategy.debug_graphviz_path",
-        "BuildStrategy.enable_sequential_execution",
-        "BuildStrategy.fuse_elewise_add_act_ops",
-        "BuildStrategy.fuse_relu_depthwise_conv",
-        "BuildStrategy.gradient_scale_strategy",
-        "BuildStrategy.reduce_strategy",
-        "BuildStrategy.remove_unnecessary_lock",
-        "BuildStrategy.sync_batch_norm",
-        "DynamicRNN.step_input",
-        "DynamicRNN.static_input",
-        "DynamicRNN.block",
-        "DynamicRNN.update_memory",
-        "DynamicRNN.output",
-        "transpiler.DistributeTranspilerConfig",
-        "transpiler.DistributeTranspilerConfig.slice_var_up",
-        "transpiler.DistributeTranspilerConfig.split_method",
-        "transpiler.DistributeTranspilerConfig.min_block_size",
-        "DistributeTranspilerConfig.slice_var_up",
-        "DistributeTranspilerConfig.split_method",
-        "ModelAverage.apply",
-        "ModelAverage.restore",
-        "DistributeTranspilerConfig",
-        "DistributeTranspilerConfig.min_block_size",
-        "ExecutionStrategy.allow_op_delay",
-        "load",
-        "Accuracy.update",
-        "ChunkEvaluator.update",
-        "ExecutionStrategy.num_iteration_per_drop_scope",
-        "ExecutionStrategy.num_threads",
-        "CompiledProgram._with_inference_optimize",
-        "CompositeMetric.add_metric",
-        "CompositeMetric.update",
-        "CompositeMetric.eval",
-        "DetectionMAP.get_map_var",
-        "MetricBase",
-        "MetricBase.reset",
-        "MetricBase.get_config",
-        "MetricBase.update",
-        "MetricBase.eval",
-        "Accuracy.eval",
-        "Auc.update",
-        "Auc.eval",
-        "EditDistance.update",
-        "EditDistance.eval",
-        "ExponentialMovingAverage.apply",
-        "ExponentialMovingAverage.restore",
-        "ExponentialMovingAverage.update",
-        "StaticRNN.step",
-        "StaticRNN.step_input",
-        "StaticRNN.step_output",
-        "StaticRNN.update_memory",
-        "DetectionMAP.reset",
-        "StaticRNN.output",
-        "cuda_places",
-        "CUDAPinnedPlace",
-        "CUDAPlace",
-        "Program.parse_from_string",
-        "Compressor",
-        "Compressor.config",
-        "Compressor.run",
-        "HDFSClient.upload",
-        "HDFSClient.download",
-        "HDFSClient.is_exist",
-        "HDFSClient.is_dir",
-        "HDFSClient.delete",
-        "HDFSClient.rename",
-        "HDFSClient.makedirs",
-        "HDFSClient.ls",
-        "HDFSClient.lsr",
-        "multi_download",
-        "multi_upload",
-        "TrainingDecoder.block",
-        "QuantizeTranspiler.training_transpile",
-        "QuantizeTranspiler.freeze_program",
-        "AutoMixedPrecisionLists",
-        "Uniform.sample",
-        "Uniform.log_prob",
-        "Uniform.entropy",
-        "Categorical.kl_divergence",
-        "Categorical.entropy",
-        "MultivariateNormalDiag.entropy",
-        "MultivariateNormalDiag.kl_divergence",
-        "RNNCell",
-        "RNNCell.call",
-        "RNNCell.get_initial_states",
-        "GRUCell.call",
-        "LSTMCell.call",
-        "Decoder",
-        "Decoder.initialize",
-        "Decoder.step",
-        "Decoder.finalize",
-        "fused_elemwise_activation",
-        "search_pyramid_hash",
-        "convert_dist_to_sparse_program",
-        "load_persistables_for_increment",
-        "load_persistables_for_inference",
-        "xmap_readers",
-        "Metric.reset",
-        "Metric.update",
-        "Metric.accumulate",
-        "Metric.name",
-        "Metric.compute",
-        "Accuracy.reset",
-        "Accuracy.update",
-        "Accuracy.accumulate",
-        "Accuracy.name",
-        "Accuracy.compute",
-        "Precision.reset",
-        "Precision.update",
-        "Precision.accumulate",
-        "Precision.name",
-        "Precision.compute",
-        "Recall.reset",
-        "Recall.update",
-        "Recall.accumulate",
-        "Recall.name",
-        "Recall.compute",
-        "Auc.reset",
-        "Auc.update",
-        "Auc.accumulate",
-        "Auc.name",
-        "Auc.compute",
-        "Callback.set_params",
-        "Callback.on_train_begin",
-        "Callback.on_train_end",
-        "Callback.on_eval_begin",
-        "Callback.on_eval_end",
-        "Callback.on_test_begin",
-        "Callback.on_test_end",
-        "Callback.on_epoch_begin",
-        "Callback.on_epoch_end",
-        "Callback.on_train_batch_begin",
-        "Callback.on_train_batch_end",
-        "Callback.on_eval_batch_begin",
-        "Callback.on_eval_batch_end",
-        "Callback.on_test_batch_begin",
-        "Callback.on_test_batch_end",
-        "Model.prepare",
-        "SimpleRNNCell",
-        "SimpleRNNCell.forward",
-        "LSTMCell",
-        "LSTMCell.forward",
-        "GRUCell",
-        "GRUCell.forward",
-        "SimpleRNN",
-        "GRU",
-        "LSTM",
-        "RNN",
-        "BiRNN",
-        "RNNCellBase",
-        "RNNCellBase.get_initial_states",
-        "gelu",
-        "erf",
-        "DecodeHelper",
-        "DecodeHelper.initialize",
-        "DecodeHelper.sample",
-        "DecodeHelper.next_inputs",
-        "TrainingHelper.initialize",
-        "TrainingHelper.sample",
-        "TrainingHelper.next_inputs",
-        "GreedyEmbeddingHelper.initialize",
-        "GreedyEmbeddingHelper.sample",
-        "GreedyEmbeddingHelper.next_inputs",
-        "LayerList.append",
-        "HDFSClient",
-        "InitState",
-        "TracedLayer",
-        "SampleEmbeddingHelper.sample",
-        "BasicDecoder.initialize",
-        "BasicDecoder.step",
-        "ParameterList.append",
-        "GreedyEmbeddingHelper",
-        "SampleEmbeddingHelper",
-        "BasicDecoder",
-        "lstm",
-        "partial_sum",
-        "StateCell",
-        "StateCell.compute_state",
-        "TrainingDecoder",
-        "TrainingDecoder.step_input",
-        "TrainingDecoder.static_input",
-        "TrainingDecoder.output",
-        "BeamSearchDecoder",
-        "GradClipByValue",
-        "GradClipByNorm",
-        "Variable.detach",
-        "Variable.numpy",
-        "Variable.set_value",
-        "Variable.gradient",
-        "BeamSearchDecoder.decode",
-        "BeamSearchDecoder.read_array",
-        "CompiledProgram",
-        "CompiledProgram.with_data_parallel",
-        "append_backward",
-        "guard",
-        "to_variable",
-        "op_freq_statistic",
-        "save_dygraph",
-        "load_dygraph",
-        "ParallelExecutor",
-        "ParallelExecutor.run",
-        "ParallelExecutor.drop_local_exe_scopes",
-        "GradClipByGlobalNorm",
-        "extend_with_decoupled_weight_decay",
-        "switch",
-        "Normal",
-        "memory_usage",
-        "decorate",
-        "PiecewiseDecay",
-        "InverseTimeDecay",
-        "PolynomialDecay",
-        "NoamDecay",
-        "start_profiler",
-        "profiler",
-        "tree_conv",
-        "multiclass_nms2",
-        "DataFeedDesc",
-        "Conv2D",
-        "Conv3D",
-        "Conv3DTranspose",
-        "Embedding",
-        "NCE",
-        "PRelu",
-        "BilinearTensorProduct",
-        "GroupNorm",
-        "SpectralNorm",
-        "TreeConv",
-        "prroi_pool",
-        "ChunkEvaluator",
-        "EditDistance",
-        "ErrorClipByValue",
-        "Program.clone",
-        "cuda_pinned_places",
-        "DataFeeder",
-        "elementwise_floordiv",
-        "Layer",
-        "Layer.create_parameter",
-        "Layer.create_variable",
-        "Layer.sublayers",
-        "Layer.add_parameter",
-        "Layer.add_sublayer",
-        "Layer.parameters",
-        "Tracer",
-        "Layer.full_name",
-        "InMemoryDataset",
-        "layer_norm",
-        "bipartite_match",
-        "double_buffer",
-        "cumsum",
-        "thresholded_relu",
-        "group_norm",
-        "random_crop",
-        "row_conv",
-        "hard_shrink",
-        "ssd_loss",
-        "retinanet_target_assign",
-        "InMemoryDataset.global_shuffle",
-        "InMemoryDataset.get_memory_data_size",
-        "DetectionMAP",
-        "hash",
-        "InMemoryDataset.set_queue_num",
-        "LayerNorm",
-        "Preprocessor",
-        "chunk_eval",
-        "GRUUnit",
-        "ExponentialMovingAverage",
-        "QueueDataset.global_shuffle",
-        "NumpyArrayInitializer",
-        "create_py_reader_by_data",
-        "InMemoryDataset.local_shuffle",
-        "InMemoryDataset.get_shuffle_data_size",
-        "size",
-        "edit_distance",
-        "nce",
-        "BilinearInitializer",
-        "NaturalExpDecay",
-        "noam_decay",
-        "retinanet_detection_output",
-        "Pool2D",
-        "PipelineOptimizer",
-        "generate_mask_labels",
-        "isfinite",
-        "InMemoryDataset.set_fleet_send_batch_size",
-        "cuda_profiler",
-        "unfold",
-        "Executor",
-        "InMemoryDataset.load_into_memory",
-        "ExponentialDecay",
-        "BatchNorm",
-        "deformable_conv",
-        "InMemoryDataset.preload_into_memory",
-        "py_reader",
-        "linear_lr_warmup",
-        "InMemoryDataset.wait_preload_done",
-        "CosineDecay",
-        "roi_perspective_transform",
-        "unique",
-        "ones_like",
-        "LambOptimizer",
-        "InMemoryDataset.release_memory",
-        "Conv2DTranspose",
-        "QueueDataset.local_shuffle",
-        "save_persistables@dygraph/checkpoint.py",
-        "load_persistables@dygraph/checkpoint.py",
-        "elementwise_pow",
-        "WeightedAverage.reset",
-        "ChunkEvaluator.eval",
-        "NCE.forward",
-        "elementwise_div",
-        "BilinearTensorProduct.forward",
-        "NoamDecay.step",
-        "elementwise_min",
-        "PiecewiseDecay.step",
-        "Conv3DTranspose.forward",
-        "elementwise_add",
-        "IfElse.output",
-        "IfElse.true_block",
-        "InverseTimeDecay.step",
-        "PolynomialDecay.step",
-        "Precision.eval",
-        "enabled",
-        "elementwise_max",
-        "stop_gperf_profiler",
-        "IfElse.false_block",
-        "WeightedAverage.add",
-        "Auc.trapezoid_area",
-        "elementwise_mul",
-        "GroupNorm.forward",
-        "SpectralNorm.forward",
-        "elementwise_sub",
-        "Switch.case",
-        "IfElse.input",
-        "prepare_context",
-        "PRelu.forward",
-        "Recall.update",
-        "start_gperf_profiler",
-        "TreeConv.forward",
-        "Conv2D.forward",
-        "Switch.default",
-        "elementwise_mod",
-        "Precision.update",
-        "WeightedAverage.eval",
-        "Conv3D.forward",
-        "Embedding.forward",
-        "Recall.eval",
-        "FC.forward",
-        "While.block",
-        "DGCMomentumOptimizer",
-        "ParallelEnv",
-        "spawn",
-        "init_parallel_env",
-        "DataParallel",
-        "DataParallel.scale_loss",
-        "DataParallel.apply_collective_grads",
-        "BasicLSTMCell.forward",
-        "BasicGRUCell.forward",
-        "RNN.forward",
-        "StackedRNNCell.forward",
-        "StackedLSTMCell.forward",
-        "LSTM.forward",
-        "BidirectionalRNN.forward",
-        "BidirectionalLSTM.forward",
-        "StackedGRUCell.forward",
-        "GRU.forward",
-        "BidirectionalGRU.forward",
-        "DynamicDecode.forward",
-        "Conv1dPoolLayer.forward",
-        "CNNEncoder.forward",
-        "TransformerCell.forward",
-        "TransformerBeamSearchDecoder.step",
-        "MultiHeadAttention.forward",
-        "MultiHeadAttention.cal_kv",
-        "FFN.forward",
-        "TransformerEncoderLayer.forward",
-        "TransformerEncoder.forward",
-        "TransformerDecoderLayer.forward",
-        "TransformerDecoder.forward",
-        "TransformerDecoder.prepare_static_cache",
-        "TransformerDecoder.prepare_incremental_cache",
-        "LinearChainCRF.forward",
-        "CRFDecoding.forward",
-        "SequenceTagging.forward",
-        "XPUPlace",
-        "is_compiled_with_xpu",
-        "xpu_places"
-    ],
-    "gpu_not_white":[
-        "deformable_conv",
-        "cuda_places",
-        "CUDAPinnedPlace",
-        "CUDAPlace",
-        "cuda_profiler",
-        "DGCMomentumOptimizer"
-    ]
-}