import into Eclipse

Create OpenCL module (initially copy of CUDA module).

Change occurrences of cuda to opencl

add OpenCL option to orio_main

add orcl script to launch orio with OpenCL support

Setup script copies orcl

set default build command

Create OpenCL module (initially copy of CUDA module).

Change occurrences of cuda to opencl

generate CUDA code without tinfo

import into Eclipse

import into Eclipse

Create OpenCL module (initially copy of CUDA module).

Change occurrences of cuda to opencl

add OpenCL option to orio_main

add orcl script to launch orio with OpenCL support

Setup script copies orcl

set default build command

Create OpenCL module (initially copy of CUDA module).

Change occurrences of cuda to opencl

generate CUDA code without tinfo

import into Eclipse

Create OpenCL module (initially copy of CUDA module).

import into Eclipse

Create OpenCL module (initially copy of CUDA module).

Change occurrences of cuda to opencl

add OpenCL option to orio_main

add orcl script to launch orio with OpenCL support

Setup script copies orcl

set default build command

Create OpenCL module (initially copy of CUDA module).

Change occurrences of cuda to opencl

generate CUDA code without tinfo

import into Eclipse

Create OpenCL module (initially copy of CUDA module).

Change occurrences of cuda to opencl

add OpenCL option to orio_main

add orcl script to launch orio with OpenCL support

Setup script copies orcl

import into Eclipse

Create OpenCL module (initially copy of CUDA module).

Change occurrences of cuda to opencl

add OpenCL option to orio_main

add orcl script to launch orio with OpenCL support

Setup script copies orcl

set default build command

Create OpenCL module (initially copy of CUDA module).

Change occurrences of cuda to opencl

generate CUDA code without tinfo

import into Eclipse

Create OpenCL module (initially copy of CUDA module).

Change occurrences of cuda to opencl

add OpenCL option to orio_main

add orcl script to launch orio with OpenCL support

Setup script copies orcl

set default build command

import into Eclipse

Create OpenCL module (initially copy of CUDA module).

Change occurrences of cuda to opencl

add OpenCL option to orio_main

add orcl script to launch orio with OpenCL support

Setup script copies orcl

set default build command

Create OpenCL module (initially copy of CUDA module).

Change occurrences of cuda to opencl

generate CUDA code without tinfo

import into Eclipse

Create OpenCL module (initially copy of CUDA module).

Change occurrences of cuda to opencl

add OpenCL option to orio_main

add orcl script to launch orio with OpenCL support

Setup script copies orcl

set default build command

generate CUDA code without tinfo

import into Eclipse

Change occurrences of cuda to opencl

add OpenCL option to orio_main

add orcl script to launch orio with OpenCL support

import into Eclipse

Create OpenCL module (initially copy of CUDA module).

Change occurrences of cuda to opencl

add OpenCL option to orio_main

add orcl script to launch orio with OpenCL support

Setup script copies orcl

set default build command

Create OpenCL module (initially copy of CUDA module).

Change occurrences of cuda to opencl

generate CUDA code without tinfo

import into Eclipse

Create OpenCL module (initially copy of CUDA module).

Change occurrences of cuda to opencl

add OpenCL option to orio_main

add orcl script to launch orio with OpenCL support

Setup script copies orcl

set default build command

generate CUDA code without tinfo

import into Eclipse

Change occurrences of cuda to opencl

add OpenCL option to orio_main

add orcl script to launch orio with OpenCL support

set default build command

Setup script copies orcl

OpenCL properties query

need to read properties before tranformation arguments

--amend

redirect props to file

process props from multiple platforms and devices

process props from multiple platforms and devices

import into Eclipse

Create OpenCL module (initially copy of CUDA module).

Change occurrences of cuda to opencl

add OpenCL option to orio_main

add orcl script to launch orio with OpenCL support

Setup script copies orcl

set default build command

Create OpenCL module (initially copy of CUDA module).

Change occurrences of cuda to opencl

generate CUDA code without tinfo

import into Eclipse

Create OpenCL module (initially copy of CUDA module).

Change occurrences of cuda to opencl

add OpenCL option to orio_main

add orcl script to launch orio with OpenCL support

Setup script copies orcl

set default build command

generate CUDA code without tinfo

import into Eclipse

Change occurrences of cuda to opencl

add OpenCL option to orio_main

add orcl script to launch orio with OpenCL support

set default build command

Setup script copies orcl

OpenCL properties query

need to read properties before tranformation arguments

--amend

redirect props to file

process props from multiple platforms and devices

process props from multiple platforms and devices

process props from multiple platforms and devices

handle platform properties

remove domain code from opencl transformation

import into Eclipse

Create OpenCL module (initially copy of CUDA module).

Change occurrences of cuda to opencl

add OpenCL option to orio_main

add orcl script to launch orio with OpenCL support

Setup script copies orcl

set default build command

Create OpenCL module (initially copy of CUDA module).

Change occurrences of cuda to opencl

generate CUDA code without tinfo

import into Eclipse

Create OpenCL module (initially copy of CUDA module).

Change occurrences of cuda to opencl

add OpenCL option to orio_main

add orcl script to launch orio with OpenCL support

Setup script copies orcl

set default build command

generate CUDA code without tinfo

import into Eclipse

Change occurrences of cuda to opencl

add OpenCL option to orio_main

add orcl script to launch orio with OpenCL support

set default build command

Setup script copies orcl

OpenCL properties query

need to read properties before tranformation arguments

--amend

redirect props to file

process props from multiple platforms and devices

process props from multiple platforms and devices

process props from multiple platforms and devices

handle platform properties

remove domain code from opencl transformation

remove domain code from opencl transformation

remove L1 cache setting, which is not available in OpenCL

remove L1 cache setting, which is not available in OpenCL

opencl boilerplate and kernel generation

opencl support in ptest_driver

add runtime error checking for OpenCL\

opencl works for matScale

some fixes to reductions

fix types in reduction kernels

add tests

skeleton file for performance measurement with TAU

use build command to build query code

pass tinfo to opencl submodule

only use local memory barrier, not global

add post-command option (for taudb_loadtrial)

metadata

vecHint, sizeHint options

Save OpenCL attributes as TAU metadata

generate unique filenames for profiles

allow filename twice in post-command

OpenCL compile flag arguments were in wrong position

put missing .iteritems() back (I don't know how it got deleted!)

fix transformation argument parsing for platform and device

Insert timer calls (need to fix this to only do this when using TAU)

Move end of execute timer to after blocking copy.

Add vecAXPY example.

Update tau_skeleton.c to use timers around OpenCL executions when OpenCL
is used.

vecMAXPY

WAXPY

ex14FF

Don't output TAU_METADATA except for OpenCL.

Spell 'language' correctly.

ex14FJ

viennacl

Author: nchaimov

.pydevproject

@@ -1,10 +1,8 @@
 <?xml version="1.0" encoding="UTF-8" standalone="no"?>
-<?eclipse-pydev version="1.0"?>
-
-<pydev_project>
-<pydev_property name="org.python.pydev.PYTHON_PROJECT_INTERPRETER">Default</pydev_property>
-<pydev_property name="org.python.pydev.PYTHON_PROJECT_VERSION">python 2.6</pydev_property>
+<?eclipse-pydev version="1.0"?><pydev_project>
+<pydev_property name="org.python.pydev.PYTHON_PROJECT_INTERPRETER">python27</pydev_property>
+<pydev_property name="org.python.pydev.PYTHON_PROJECT_VERSION">python 2.7</pydev_property>
 <pydev_pathproperty name="org.python.pydev.PROJECT_SOURCE_PATH">
-<path>/orio</path>
+<path>/Orio-svn</path>
 </pydev_pathproperty>
 </pydev_project>

orcl

@@ -0,0 +1,25 @@
+#!/usr/bin/env python 
+#
+# orf - Compile shell for Orio (generate OpenCL source code)
+#
+
+import os, sys, distutils.sysconfig
+
+# include the annotation tool in the Python's search path
+abs_path = os.path.realpath(sys.argv[0])
+cur_dir = os.path.dirname(abs_path)
+dirs=[]
+if os.path.basename(cur_dir) == 'bin':
+    # installed location
+    base_dir = os.path.join(os.path.dirname(cur_dir),'lib','python' + distutils.sysconfig.get_python_version(),'site-packages')
+    dirs = [base_dir, os.path.join(base_dir,'orio')]
+else:
+    # source tree (development) location
+    dirs = [os.path.join(cur_dir,'orio')]
+sys.path.extend(dirs)
+
+
+# dispatch to Orio's main
+import orio.main.orio_main
+orio.main.orio_main.start(sys.argv, orio.main.orio_main.OPENCL)
+

orio/main/cmd_line_opts.py

@@ -16,7 +16,9 @@
 
 Options:
   -c, --pre-command=<string>     Command string with which to prefix the execution of the 
-                                 Orio-built code, e.g., tauex
+                                 Orio-built code, e.g., tau_exec
+  --post-command=<string>        Command string to run after each execution of Orio-built code,
+                                 e.g., taudb_loadtrial
   -e, --erase-annot              remove annotations from the output
   -h, --help                     display this message
   -o <file>, --output=<file>     place the output in <file> (only valid when processing 
@@ -121,7 +123,7 @@ def parse(self, argv):
             opts, args = getopt.getopt(orioargv,
                                        'c:ehko:p:rs:vx',
                                        ['pre-command=', 'config=','configfile=', 'erase-annot', 'help', 'keep-temps',' output=', 
-                                       'output-prefix=', 'rename-objects', 'spec=', 'verbose', 'extern', 'validate'])
+                                       'output-prefix=', 'rename-objects', 'spec=', 'verbose', 'extern', 'validate', 'post-command='])
         except Exception, e:
             sys.stderr.write('Orio command-line error: %s' % e)
             sys.stderr.write(USAGE_MSG + '\n')
@@ -131,6 +133,8 @@ def parse(self, argv):
         for opt, arg in opts:
             if opt in ('-c', '--pre-command'):
                 cmdline['pre_cmd'] = arg
+            elif opt in ('--post-command'):
+                cmdline['post_cmd'] = arg
             elif opt in ('-e', '--erase-annot'):
                 cmdline['erase_annot'] = True
             elif opt in ('-h', '--help'):

orio/main/opt_driver.py

@@ -134,7 +134,7 @@ def __optimizeCodeFrag(self, cfrag, perf_params):
 
                 # apply code transformations
                 try:
-                  if self.lang == 'cuda':
+                  if self.lang == 'cuda' or self.lang == 'opencl':
                     transformation = mod_class(perf_params,
                                               cfrag.leader_ann.mod_code,
                                               optimized_body_code,

orio/main/orio_main.py

@@ -11,6 +11,7 @@
 C_CPP = 1
 FORTRAN = 2
 CUDA = 3
+OPENCL = 4
 
 #----------------------------------------------
 
@@ -26,6 +27,8 @@ def start(argv, lang):
         language = 'c'
     elif lang == CUDA:
         language = 'cuda'
+    elif lang == OPENCL:
+        language = 'opencl'
     else:
         sys.stderr.write('orio.main.main:  Language not supported at this time.')
         sys.exit(1)

orio/main/tspec/pparser.py

@@ -21,7 +21,7 @@
     'simplex_reflection_coef', 'simplex_expansion_coef',
     'simplex_contraction_coef', 'simplex_shrinkage_coef', 'simplex_local_distance', 'simplex_x0',
     'validation', 'validation_file', 'expected_output'
-    , 'macro'
+    , 'macro', 'performance_test_code', 'skeleton_test_code', 'skeleton_code_file'
 ]
 
 # map of reserved keywords
@@ -139,6 +139,7 @@ def p_def_type(p):
                 | INPUT_VARS
                 | SEARCH
                 | VALIDATION
+                | PERFORMANCE_TEST_CODE
     '''
     p[0] = (p[1], p.lineno(1))
 
@@ -207,6 +208,8 @@ def p_arg_type(p):
                 | SIMPLEX_X0
                 | VALIDATION_FILE
                 | EXPECTED_OUTPUT
+                | SKELETON_TEST_CODE
+                | SKELETON_CODE_FILE
     '''
     p[0] = (p[1], p.lineno(1))
 

orio/main/tuner/ptest_codegen.py

@@ -3,7 +3,7 @@
 #
 
 import random, re
-import skeleton_code
+import skeleton_code 
 from orio.main.util.globals import *
 from orio.main.tuner.skeleton_code import SEQ_TIMER
 
@@ -345,7 +345,11 @@ def generate(self, code_map):
         if not self.decl_file:
             prologue_code += ('%s();' % self.malloc_func_name) + '\n  '
         prologue_code += ('%s();' % self.init_func_name) + '\n'
-
+        
+        if Globals().language == 'opencl':
+            for (k, v) in Globals().metadata.iteritems():
+                prologue_code += 'TAU_METADATA("%s", "%s");\n' % (k, v)
+        
         # create code for the epilogue section
         epilogue_code = ''
         #if not self.decl_file:
@@ -372,6 +376,15 @@ def __init__(self, input_params, input_decls, decl_file, init_file, skeleton_cod
         super(PerfTestCodeGenCUDA, self).__init__(input_params, input_decls, decl_file, init_file, skeleton_code_file, language,
                  random_seed, use_parallel_search)
 
+# --------------------------------------------------------------------------------------
+class PerfTestCodeGenOpenCL(PerfTestCodeGen):
+
+    def __init__(self, input_params, input_decls, decl_file, init_file, skeleton_code_file, language='opencl',
+                 random_seed=None, use_parallel_search=False):
+        super(PerfTestCodeGenOpenCL, self).__init__(input_params, input_decls, decl_file, init_file, skeleton_code_file, language,
+                 random_seed, use_parallel_search)
+     
+
 
 # --------------------------------------------------------------------------------------
 

orio/main/tuner/ptest_driver.py

@@ -2,10 +2,11 @@
 # To compile and execute the performance-testing code to get the performance cost
 #
 
-import os, time, re
+import os, time, re, datetime, uuid
 
 from orio.main.util.globals import *
 
+
 #-----------------------------------------------------
 
 perftest_counter = 0
@@ -37,13 +38,13 @@ def __init__(self, tinfo, use_parallel_search, language="c", timing_code=''):
         global perftest_counter 
         perftest_counter += 1 
 
-        if not (language == 'c' or language == 'fortran' or language == 'cuda'):
+        if not (language == 'c' or language == 'fortran' or language == 'cuda' or language == 'opencl'):
             err('orio.main.tuner.ptest_driver: unknown output language specified: %s' % language)
         self.language = language
 
         self.__PTEST_FNAME=Globals().out_prefix+self.__PTEST_FNAME
 
-        if language == 'c': 
+        if language == 'c' or language == 'opencl': 
             self.ext = '.c'
         elif language == 'cuda':
             self.ext = '.cu'
@@ -55,7 +56,7 @@ def __init__(self, tinfo, use_parallel_search, language="c", timing_code=''):
         self.original_src_name = self.__PTEST_FNAME + '_original' + self.ext
         self.exe_name = self.__PTEST_FNAME + '.exe'
         self.original_exe_name = self.__PTEST_FNAME + '_original.exe'
-        if self.language == 'cuda':
+        if self.language == 'cuda' or self.language == 'opencl':
             self.obj_name = self.__PTEST_FNAME + '.o'
             self.original_obj_name = self.__PTEST_FNAME + '_original.o'
 
@@ -106,7 +107,7 @@ def __write(self, test_code, perf_param=None):
             err('orio.main.tuner.ptest_driver: cannot open file for writing: %s' % self.src_name)
 
         if self.first:
-            if self.language != 'cuda' and not self.tinfo.timer_file and not (os.path.exists(self.timer_file)):
+            if self.language != 'cuda' and self.language != 'opencl' and not self.tinfo.timer_file and not (os.path.exists(self.timer_file)):
                 # Generate the timing routine file
                 try: 
                     f = open(self.timer_file, 'w')
@@ -214,6 +215,10 @@ def __build(self, perf_param=None):
             if status:
                 err('orio.main.tuner.ptest_driver: failed to compile the test cuda code: "%s"' % cmd)
             cmd = ('%s %s -o %s %s' % (build_cmd, self.extra_compiler_opts, self.exe_name, self.obj_name))
+        elif self.language == 'opencl':
+            cmd = ('%s %s -o %s %s %s' % (build_cmd, self.extra_compiler_opts,
+                                             self.exe_name, self.src_name2, 
+                                             self.tinfo.libs))
         else:
             cmd = ('%s %s -o %s %s %s %s' % (build_cmd, self.extra_compiler_opts,
                                              self.exe_name, self.src_name2, 
@@ -245,6 +250,8 @@ def __build(self, perf_param=None):
     def __execute(self):
         '''Execute the test to get the performance costs'''
 
+        Globals().metadata['src_filenames'] = ",".join(Globals().src_filenames)
+
         # check if the executable exists
         if not os.path.exists(self.exe_name):
             err('orio.main.tuner.ptest_driver:  the executable of the test code does not exist')
@@ -295,6 +302,19 @@ def __execute(self):
             except Exception, e:
                 err('orio.main.tuner.ptest_driver: failed to execute the test code: "%s"\n --> %s: %s' % (cmd,e.__class__.__name__, e))
 
+            if Globals().post_cmd is not None:
+                try:
+                    cmd = Globals().post_cmd
+                    uniq = "profile-" + datetime.datetime.now().strftime("%y-%m-%d:%H:%S") + '-' + uuid.uuid4().hex
+                    if "%s" in cmd:
+                        cmd = cmd % (uniq, uniq)
+                    status = os.system(cmd) 
+                    if status:
+                        err('orio.main.tuner.ptest_driver: failed to execute the post-command: "%s"' % Globals().post_cmd)
+                except Exception, e:
+                    err('orio.main.tuner.ptest_driver: failed to execute the post-command: "%s"\n --> %s: %s' % (Globals().post_cmd,e.__class__.__name__, e))
+        
+                
             try:
                 if out:
                     #info('out: %s' % out)
@@ -374,7 +394,6 @@ def run(self, test_code, perf_param=None):
         @param test_code: the code for testing multiple coordinates in the search space
         @return: a dictionary of the times corresponding to each coordinate in the search space
         '''
-
         # write the testing code
         self.__write(test_code, perf_param=perf_param)
 

orio/main/tuner/skeleton_code.py

@@ -449,10 +449,10 @@ def __init__(self, code, use_parallel_search, language='c'):
             if use_parallel_search:
                 code = PAR_DEFAULT
             else:
-                if language == 'c':
-                    code = SEQ_DEFAULT
-                else:
+                if language == 'cuda':
                     code = SEQ_DEFAULT_CUDA
+                else:
+                    code = SEQ_DEFAULT
 
         self.code = code
         self.use_parallel_search = use_parallel_search

orio/main/tuner/tuner.py

@@ -66,7 +66,12 @@ def tune(self, module_body_code, line_no, cfrags):
                 c = orio.main.tuner.ptest_codegen.PerfTestCodeGenCUDA(prob_size, tinfo.ivar_decls, tinfo.ivar_decl_file,
                                                                   tinfo.ivar_init_file, tinfo.ptest_skeleton_code_file, self.odriver.lang,
                                                                   tinfo.random_seed, use_parallel_search)
-
+                
+            elif self.odriver.lang == 'opencl':
+                c = orio.main.tuner.ptest_codegen.PerfTestCodeGenCUDA(prob_size, tinfo.ivar_decls, tinfo.ivar_decl_file,
+                                                                  tinfo.ivar_init_file, tinfo.ptest_skeleton_code_file, self.odriver.lang,
+                                                                  tinfo.random_seed, use_parallel_search)
+                
             elif self.odriver.lang == 'fortran':
                 c = orio.main.tuner.ptest_codegen.PerfTestCodeGenFortran(prob_size, tinfo.ivar_decls, tinfo.ivar_decl_file,
                                                                          tinfo.ivar_init_file, tinfo.ptest_skeleton_code_file, self.odriver.lang,
@@ -124,6 +129,11 @@ def tune(self, module_body_code, line_no, cfrags):
                 iparams.sort(lambda x,y: cmp(x[0],y[0]))
                 for pname, pvalue in iparams:
                     info(' %s = %s' % (pname, pvalue))
+                    
+            iparams = ptcodegen.input_params[:]
+            iparams.sort(lambda x,y: cmp(x[0],y[0]))
+            for pname, pvalue in iparams:
+                Globals().metadata['size_' + pname] = pvalue
 
             # create the search engine
             search_eng = search_class({'cfrags':cfrags,