def declare_workitems(self,keys,kernel):
# The keys are strings
# print "\n\nDEC_WORKITEMS", keys, self.device_threads_dec, self.device_blocks_dec
if keys.find('tx') > -1:
kernel, self.device_threads_dec[0] = threads.tx(self.device_threads_dec[0], kernel)
if keys.find('ty') > -1:
kernel, self.device_threads_dec[1] = threads.ty(self.device_threads_dec[1], kernel)
if keys.find('tz') > -1:
kernel, self.device_threads_dec[2] = threads.tz(self.device_threads_dec[2], kernel)
if keys.find('bx') > -1:
kernel, self.device_blocks_dec[0] = blocks.bx(self.device_blocks_dec[0], kernel)
if keys.find('bx') > -1:
kernel, self.device_blocks_dec[1] = blocks.by(self.device_blocks_dec[1], kernel)
if keys.find('bz') > -1:
kernel, self.device_blocks_dec[2] = blocks.bz(self.device_blocks_dec[2], kernel)
return kernel
def declare_workitems(self,keys,kernel):
"""The work items are declared in the kernel depending on their
declaration in Urutu method"""
# The keys are strings
# print "\n\nDEC_WORKITEMS", keys, self.device_threads_dec, self.device_blocks_dec
if keys.find('tx') > -1:
kernel, self.device_threads_dec[0] = threads.tx(self.device_threads_dec[0], kernel)
if keys.find('ty') > -1:
kernel, self.device_threads_dec[1] = threads.ty(self.device_threads_dec[1], kernel)
if keys.find('tz') > -1:
kernel, self.device_threads_dec[2] = threads.tz(self.device_threads_dec[2], kernel)
if keys.find('bx') > -1:
kernel, self.device_blocks_dec[0] = blocks.bx(self.device_blocks_dec[0], kernel)
if keys.find('bx') > -1:
kernel, self.device_blocks_dec[1] = blocks.by(self.device_blocks_dec[1], kernel)
if keys.find('bz') > -1:
kernel, self.device_blocks_dec[2] = blocks.bz(self.device_blocks_dec[2], kernel)
return kernel
def declare_workitems(self,keys,kernel):
"""The work items are declared in the kernel depending on their
declaration in Urutu method"""
# The keys are strings
# print "\n\nDEC_WORKITEMS", keys, self.device_threads_dec, self.device_blocks_dec
if keys.find('tx') > -1:
kernel, self.device_threads_dec[0] = threads.tx(self.device_threads_dec[0], kernel)
if keys.find('ty') > -1:
kernel, self.device_threads_dec[1] = threads.ty(self.device_threads_dec[1], kernel)
if keys.find('tz') > -1:
kernel, self.device_threads_dec[2] = threads.tz(self.device_threads_dec[2], kernel)
if keys.find('bx') > -1:
kernel, self.device_blocks_dec[0] = blocks.bx(self.device_blocks_dec[0], kernel)
if keys.find('bx') > -1:
kernel, self.device_blocks_dec[1] = blocks.by(self.device_blocks_dec[1], kernel)
if keys.find('bz') > -1:
kernel, self.device_blocks_dec[2] = blocks.bz(self.device_blocks_dec[2], kernel)
return kernel
def declare_workitems(self, keys, kernel):
# The keys are strings
# print "\n\nDEC_WORKITEMS", keys, self.device_threads_dec, self.device_blocks_dec
if keys.find('tx') > -1:
kernel, self.device_threads_dec[0] = threads.tx(
self.device_threads_dec[0], kernel)
if keys.find('ty') > -1:
kernel, self.device_threads_dec[1] = threads.ty(
self.device_threads_dec[1], kernel)
if keys.find('tz') > -1:
kernel, self.device_threads_dec[2] = threads.tz(
self.device_threads_dec[2], kernel)
if keys.find('bx') > -1:
kernel, self.device_blocks_dec[0] = blocks.bx(
self.device_blocks_dec[0], kernel)
if keys.find('bx') > -1:
kernel, self.device_blocks_dec[1] = blocks.by(
self.device_blocks_dec[1], kernel)
if keys.find('bz') > -1:
kernel, self.device_blocks_dec[2] = blocks.bz(
self.device_blocks_dec[2], kernel)
return kernel
def inspect_it(self,sentence,kernel):
# print "Inside inspect_it()",sentence,kernel
"""The Urutu method translates to CUDA kernel here."""
phrase = sentence.split('\t')
if phrase.count('#') > 0:
return
tab = phrase.count('')
if tab < self.tabs and self.device_scope == False:
for j in range(self.tabs - tab):
kernel += "}\n"
self.tabs = phrase.count('')
sh = shlex.shlex(phrase[-1])
i = sh.get_token()
if i == '@' or i == 'def' or i == '' or i == '#' or i == '//' or i == '"""':
return kernel
stmt = []
while i is not sh.eof:
stmt.append(i)
i = sh.get_token()
if stmt.count('and') > 0:
stmt[stmt.index('and')] = " && "
if self.keys.count('tx') > 0 and self.var_nam.count('tx') == 0:
kernel, self.threads_dec[0] = threads.tx(self.threads_dec[0], kernel)
self.var_nam.append("tx")
self.type_vars.append("int")
if self.keys.count('ty') > 0 and self.var_nam.count('ty') == 0:
kernel, self.threads_dec[1] = threads.ty(self.threads_dec[1], kernel)
self.var_nam.append("ty")
self.type_vars.append("int")
if self.keys.count('tz') > 0 and self.var_nam.count('tz') == 0:
kernel, self.threads_dec[2] = threads.tz(self.threads_dec[2], kernel)
self.var_nam.append("tz")
self.type_vars.append("int")
if self.keys.count('bx') > 0 and self.var_nam.count('bx') == 0:
kernel, self.blocks_dec[0] = blocks.bx(self.blocks_dec[0], kernel)
self.var_nam.append("bx")
self.type_vars.append("int")
if self.keys.count('by') > 0 and self.var_nam.count('by') == 0:
kernel, self.blocks_dec[1] = blocks.by(self.blocks_dec[1], kernel)
self.var_nam.append("by")
self.type_vars.append("int")
if self.keys.count('bz') > 0 and self.var_nam.count('bz') == 0:
kernel, self.blocks_dec[2] = blocks.bz(self.blocks_dec[2], kernel)
self.var_nam.append("bz")
self.type_vars.append("int")
if stmt.count('Tx') == 1 or stmt.count('Ty') == 1 or stmt.count('Tz') == 1:
self.var_nam, self.var_val, self.threads, ker, self.type_vars= threads.threads_decl(stmt, self.var_nam, self.var_val, self.threads, self.type_vars)
kernel += ker
if stmt.count('=') > 0:
id_eq = stmt.index('=')
type_, ptr_, var_, val_ = self.checktype(stmt[:id_eq],stmt[id_eq+1:])
kernel += type_+ptr_+var_ + "= " + val_ +";\n"
return kernel
if stmt.count('Bx') == 1 or stmt.count('By') == 1 or stmt.count('Bz') == 1:
self.var_nam, self.var_val, self.blocks, ker, self.type_vars = blocks.blocks_decl(stmt, self.var_nam, self.var_val, self.blocks, self.type_vars)
kernel += ker
if stmt.count('=') > 0:
id_eq = stmt.index("=")
type_, ptr_, var_, val_ = self.checktype(stmt[:id_eq],stmt[id_eq+1:])
self.type_vars.append(type_)
self.var_nam.append(ptr_)
kernel += type_ + ptr_ + var_ + "= " + val_ +";\n"
return kernel
for j in self.device_func_name:
if stmt.count(j) > 0:
kernel += self.device_create_func(self.device_func_name.index(j),j, stmt)
kernel = self.device_body_buff + "}\n" + kernel
self.device_body_buff = ""
return kernel
if stmt[0] == '__global' or stmt[0] == '__shared' or stmt[0] == '__register' or stmt[0] == '__constant' :
self.decarrays(stmt)
return kernel
if stmt.count('for') > 0:
kernel += self._for(stmt, kernel)
return kernel
if stmt.count('if') > 0:
return kernel + grammar.keyword(stmt, kernel)
if stmt.count('else') > 0:
kernel = kernel + "else{\n"
return kernel
if stmt.count('cpu') < 1:
for j in stmt:
if self.moved.count(j) < 1 and self.arg_nam.count(j) > 0:
self.movedata(j)
if stmt.count('cpu') > 0:
self.kernel_final.append(kernel+"}")
cpu_empty = {'id':0, 'ln':0, 'return':[], 'src':"", 'args': [], 'htod': [], 'dtoh': []}
self.cpu_id = self.cpu_id + 1
cpu_empty['id'] = self.cpu_id
cpu_empty['ln'] = self.sentences.index(sentence)
eq_id = stmt.index('=')
if eq_id > 0:
cpu_empty['return'].append(stmt[0])
# if self.moved.count()
# If a data array is not sent to gpu, we use it in cpu function
func_name = stmt.index('cpu') + 2
for i in stmt[func_name+2:-1]:
if i is not ',':
if self.arguments.count(i) > 0:
cpu_empty['args'].append(i)
if self.moved.count(i) > 0:
cpu_empty['dtoh'].append(i)
else:
print "Pass the array accessed by the CPU function to Urutu Kernel"
stmt.remove(stmt[func_name-2])
stmt.remove(stmt[func_name-2])
cpu_empty['src'] = stmt
cpu_empty['isnext'] = False
if len(self.cpu_info) == 0:
self.run_gpu(self.kernel_final[-1])
self.gpu_id+=1
kernel = "__global__ void "+ self.global_func + "_" + str(self.gpu_id) + "(" + self.kernel_args + "){\n"
self.ismap.append(True)
self.threads_dec = [False, False, False]
self.blocks_dec = [False, False, False]
self.remove_workitems()
self.kernel = kernel
elif self.cpu_info[-1]['ln'] + 1 != cpu_empty['ln']:
self.cpu_info[-1]['isnext'] = False
print "Running on CPU"
self.run_cpu()
self.kernel_final.append(kernel+"}")
self.gpu_id+=1
kernel = "__global__ void "+ self.global_func + "_" + str(self.gpu_id) + "(" + self.kernel_args + "){\n"
self.ismap.append(True)
self.threads_dec = [False, False, False]
self.blocks_dec = [False, False, False]
self.remove_workitems()
self.kernel = kernel
elif self.cpu_info[-1]['ln'] + 1 == cpu_empty['ln']:
self.cpu_info[-1]['isnext'] = True
self.cpu_info.append(cpu_empty)
return kernel
if stmt.count('Urmod') > 0:
self.num_mod = self.num_mod + 1
self.modules.append(stmt[2:])
self.kernel_final.append(kernel+"}")
kernel = "__global__ void "+ self.global_func + "_" + str(len(self.modules)+1) + "(" + self.kernel_args + "){\n"
self.ismap.append(True)
self.threads_dec = [False, False, False]
self.blocks_dec = [False, False, False]
self.kernel = kernel
return kernel
else:
# print "Entering Checkvars",phrase
if(len(self.cpu_info)) > 0 and self.return_kernel == False:
self.run_cpu()
return self.checkvars(stmt,phrase[-1],kernel)
def inspect_it(self,sentence,kernel):
# print "Inside inspect_it()",sentence,kernel
"""The Urutu method translates to CUDA kernel here."""
phrase = sentence.split('\t')
if phrase.count('#') > 0:
return
tab = phrase.count('')
if tab < self.tabs and self.device_scope == False:
for j in range(self.tabs - tab):
kernel += "}\n"
self.tabs = phrase.count('')
sh = shlex.shlex(phrase[-1])
i = sh.get_token()
if i == '@' or i == 'def' or i == '' or i == '#' or i == '//' or i == '"""':
return kernel
stmt = []
while i is not sh.eof:
stmt.append(i)
i = sh.get_token()
if stmt.count('and') > 0:
stmt[stmt.index('and')] = " && "
if self.keys.count('tx') > 0 and self.var_nam.count('tx') == 0:
kernel, self.threads_dec[0] = threads.tx(self.threads_dec[0], kernel)
self.var_nam.append("tx")
self.type_vars.append("int")
if self.keys.count('ty') > 0 and self.var_nam.count('ty') == 0:
kernel, self.threads_dec[1] = threads.ty(self.threads_dec[1], kernel)
self.var_nam.append("ty")
self.type_vars.append("int")
if self.keys.count('tz') > 0 and self.var_nam.count('tz') == 0:
kernel, self.threads_dec[2] = threads.tz(self.threads_dec[2], kernel)
self.var_nam.append("tz")
self.type_vars.append("int")
if self.keys.count('bx') > 0 and self.var_nam.count('bx') == 0:
kernel, self.blocks_dec[0] = blocks.bx(self.blocks_dec[0], kernel)
self.var_nam.append("bx")
self.type_vars.append("int")
if self.keys.count('by') > 0 and self.var_nam.count('by') == 0:
kernel, self.blocks_dec[1] = blocks.by(self.blocks_dec[1], kernel)
self.var_nam.append("by")
self.type_vars.append("int")
if self.keys.count('bz') > 0 and self.var_nam.count('bz') == 0:
kernel, self.blocks_dec[2] = blocks.bz(self.blocks_dec[2], kernel)
self.var_nam.append("bz")
self.type_vars.append("int")
if stmt.count('Tx') == 1 or stmt.count('Ty') == 1 or stmt.count('Tz') == 1:
self.var_nam, self.var_val, self.threads, ker, self.type_vars= threads.threads_decl(stmt, self.var_nam, self.var_val, self.threads, self.type_vars)
kernel += ker
if stmt.count('=') > 0:
id_eq = stmt.index('=')
type_, ptr_, var_, val_ = self.checktype(stmt[:id_eq],stmt[id_eq+1:])
kernel += type_+ptr_+var_ + "= " + val_ +";\n"
return kernel
if stmt.count('Bx') == 1 or stmt.count('By') == 1 or stmt.count('Bz') == 1:
self.var_nam, self.var_val, self.blocks, ker, self.type_vars = blocks.blocks_decl(stmt, self.var_nam, self.var_val, self.blocks, self.type_vars)
kernel += ker
if stmt.count('=') > 0:
id_eq = stmt.index("=")
type_, ptr_, var_, val_ = self.checktype(stmt[:id_eq],stmt[id_eq+1:])
self.type_vars.append(type_)
self.var_nam.append(ptr_)
kernel += type_ + ptr_ + var_ + "= " + val_ +";\n"
return kernel
for j in self.device_func_name:
if stmt.count(j) > 0:
kernel += self.device_create_func(self.device_func_name.index(j),j, stmt)
kernel = self.device_body_buff + "}\n" + kernel
self.device_body_buff = ""
return kernel
if stmt[0] == '__global' or stmt[0] == '__shared' or stmt[0] == '__register' or stmt[0] == '__constant' :
self.decarrays(stmt)
return kernel
if stmt.count('for') > 0:
kernel += self._for(stmt, kernel)
return kernel
if stmt.count('if') > 0:
return kernel + grammar.keyword(stmt, kernel)
if stmt.count('else') > 0:
kernel = kernel + "else{\n"
return kernel
if stmt.count('cpu') < 1:
for j in stmt:
if self.moved.count(j) < 1 and self.arg_nam.count(j) > 0:
self.movedata(j)
if stmt.count('cpu') > 0:
self.kernel_final.append(kernel+"}")
cpu_empty = {'id':0, 'ln':0, 'return':[], 'src':"", 'args': [], 'htod': [], 'dtoh': []}
self.cpu_id = self.cpu_id + 1
cpu_empty['id'] = self.cpu_id
cpu_empty['ln'] = self.sentences.index(sentence)
eq_id = stmt.index('=')
if eq_id > 0:
cpu_empty['return'].append(stmt[0])
# if self.moved.count()
# If a data array is not sent to gpu, we use it in cpu function
func_name = stmt.index('cpu') + 2
for i in stmt[func_name+2:-1]:
if i is not ',':
if self.arguments.count(i) > 0:
cpu_empty['args'].append(i)
if self.moved.count(i) > 0:
cpu_empty['dtoh'].append(i)
else:
print "Pass the array accessed by the CPU function to Urutu Kernel"
stmt.remove(stmt[func_name-2])
stmt.remove(stmt[func_name-2])
cpu_empty['src'] = stmt
cpu_empty['isnext'] = False
if len(self.cpu_info) == 0:
self.run_gpu(self.kernel_final[-1])
self.gpu_id+=1
kernel = "__global__ void "+ self.global_func + "_" + str(self.gpu_id) + "(" + self.kernel_args + "){\n"
self.ismap.append(True)
self.threads_dec = [False, False, False]
self.blocks_dec = [False, False, False]
self.remove_workitems()
self.kernel = kernel
elif self.cpu_info[-1]['ln'] + 1 != cpu_empty['ln']:
self.cpu_info[-1]['isnext'] = False
print "Running on CPU"
self.run_cpu()
self.kernel_final.append(kernel+"}")
self.gpu_id+=1
kernel = "__global__ void "+ self.global_func + "_" + str(self.gpu_id) + "(" + self.kernel_args + "){\n"
self.ismap.append(True)
self.threads_dec = [False, False, False]
self.blocks_dec = [False, False, False]
self.remove_workitems()
self.kernel = kernel
elif self.cpu_info[-1]['ln'] + 1 == cpu_empty['ln']:
self.cpu_info[-1]['isnext'] = True
self.cpu_info.append(cpu_empty)
return kernel
if stmt.count('Urmod') > 0:
self.num_mod = self.num_mod + 1
self.modules.append(stmt[2:])
self.kernel_final.append(kernel+"}")
kernel = "__global__ void "+ self.global_func + "_" + str(len(self.modules)+1) + "(" + self.kernel_args + "){\n"
self.ismap.append(True)
self.threads_dec = [False, False, False]
self.blocks_dec = [False, False, False]
self.kernel = kernel
return kernel
else:
# print "Entering Checkvars",phrase
if(len(self.cpu_info)) > 0 and self.return_kernel == False:
self.run_cpu()
return self.checkvars(stmt,phrase[-1],kernel)
def inspect_it(self,sentence,kernel):
# print "Inside inspect_it()",sentence,kernel
phrase = sentence.split('\t')
if phrase.count('#') > 0:
return
tab = phrase.count('')
## if tab > self.tabs and tab != len(phrase):
# for j in range(tab - self.tabs):
# kernel = kernel + "{\n"
if tab < self.tabs:
for j in range(self.tabs - tab):
kernel = kernel + "}\n"
self.tabs = phrase.count('')
sh = shlex.shlex(phrase[-1])
i = sh.get_token()
if i == '@' or i == 'def' or i == '' or i == '#' or i == '//' or i == '"""':
return kernel
stmt = []
while i is not sh.eof:
stmt.append(i)
i = sh.get_token()
# print stmt
for j in self.device_func_name:
if stmt.count(j) > 0:
kernel += self.device_create_func(self.device_func_name.index(j),j, stmt)
kernel = self.device_body_buff + "}\n" + kernel
self.device_body_buff = ""
return kernel
if self.keys.count('tx') > 0 or self.keys.count("__shared"):
kernel, self.threads_dec[0] = threads.tx(self.threads_dec[0], kernel)
if self.keys.count('ty') > 0:
kernel, self.threads_dec[1] = threads.ty(self.threads_dec[1], kernel)
if self.keys.count('tz') > 0:
kernel, self.threads_dec[2] = threads.tz(self.threads_dec[2], kernel)
if self.keys.count('bx') > 0:
kernel, self.blocks_dec[0] = blocks.bx(self.blocks_dec[0], kernel)
if self.keys.count('bx') > 0:
kernel, self.blocks_dec[1] = blocks.by(self.blocks_dec[1], kernel)
if self.keys.count('bz') > 0:
kernel, self.blocks_dec[2] = blocks.bz(self.blocks_dec[2], kernel)
if stmt.count('Tx') == 1 or stmt.count('Ty') == 1 or stmt.count('Tz') == 1:
threads.threads_decl(stmt, self.var_nam, self.var_val, self.threads, self.type_vars)
return kernel
if stmt.count('Bx') == 1 or stmt.count('By') == 1 or stmt.count('Bz') == 1:
blocks.blocks_decl(stmt, self.var_nam, self.var_val, self.blocks)
return kernel
if stmt[0] == '__global' or stmt[0] == '__shared' or stmt[0] == '__register' or stmt[0] == '__constant' :
self.decarrays(stmt)
return kernel
if stmt.count('for') > 0:
kernel += self._for(stmt, kernel)
return kernel
if stmt.count('if') > 0:
return kernel + grammar.keyword(stmt, kernel)
if stmt.count('else') > 0:
kernel = kernel + "else{\n "
self.tabs+=1
return kernel
if stmt.count('Urmod') > 0:
self.modules.append(stmt[2:])
self.kernel_final.append(kernel+"}")
kernel = "__kernel void" + self.global_func + "_" + str(len(self.modules)+1) + "(" + self.kernel_args + "){\n"
kernel = "#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n" + kernel
self.threads_dec = [False, False, False]
self.blocks_dec = [False, False, False]
self.kernel = kernel
return kernel
else:
# print "Entering Checkvars"
return self.checkvars(stmt,phrase[-1],kernel)
def inspect_it(self, sentence, kernel):
# print "Inside inspect_it()",sentence,kernel
"""The Urutu method translates to CUDA kernel here."""
phrase = sentence.split("\t")
if phrase.count("#") > 0:
return
tab = phrase.count("")
if tab < self.tabs and self.device_scope == False:
for j in range(self.tabs - tab):
kernel += "}\n"
self.tabs = phrase.count("")
sh = shlex.shlex(phrase[-1])
i = sh.get_token()
if i == "@" or i == "def" or i == "" or i == "#" or i == "//" or i == '"""':
return kernel
stmt = []
while i is not sh.eof:
stmt.append(i)
i = sh.get_token()
if stmt.count("and") > 0:
stmt[stmt.index("and")] = " && "
if self.keys.count("tx") > 0 and self.var_nam.count("tx") == 0:
kernel, self.threads_dec[0] = threads.tx(self.threads_dec[0], kernel)
self.var_nam.append("tx")
self.type_vars.append("int")
if self.keys.count("ty") > 0 and self.var_nam.count("ty") == 0:
kernel, self.threads_dec[1] = threads.ty(self.threads_dec[1], kernel)
self.var_nam.append("ty")
self.type_vars.append("int")
if self.keys.count("tz") > 0 and self.var_nam.count("tz") == 0:
kernel, self.threads_dec[2] = threads.tz(self.threads_dec[2], kernel)
self.var_nam.append("tz")
self.type_vars.append("int")
if self.keys.count("bx") > 0 and self.var_nam.count("bx") == 0:
kernel, self.blocks_dec[0] = blocks.bx(self.blocks_dec[0], kernel)
self.var_nam.append("bx")
self.type_vars.append("int")
if self.keys.count("by") > 0 and self.var_nam.count("by") == 0:
kernel, self.blocks_dec[1] = blocks.by(self.blocks_dec[1], kernel)
self.var_nam.append("by")
self.type_vars.append("int")
if self.keys.count("bz") > 0 and self.var_nam.count("bz") == 0:
kernel, self.blocks_dec[2] = blocks.bz(self.blocks_dec[2], kernel)
self.var_nam.append("bz")
self.type_vars.append("int")
if stmt.count("Tx") == 1 or stmt.count("Ty") == 1 or stmt.count("Tz") == 1:
self.var_nam, self.var_val, self.threads, ker, self.type_vars = threads.threads_decl(
stmt, self.var_nam, self.var_val, self.threads, self.type_vars
)
kernel += ker
if stmt.count("=") > 0:
id_eq = stmt.index("=")
type_, ptr_, var_, val_ = self.checktype(stmt[:id_eq], stmt[id_eq + 1 :])
kernel += type_ + ptr_ + var_ + "= " + val_ + ";\n"
return kernel
if stmt.count("Bx") == 1 or stmt.count("By") == 1 or stmt.count("Bz") == 1:
self.var_nam, self.var_val, self.blocks, ker, self.type_vars = blocks.blocks_decl(
stmt, self.var_nam, self.var_val, self.blocks, self.type_vars
)
kernel += ker
if stmt.count("=") > 0:
id_eq = stmt.index("=")
type_, ptr_, var_, val_ = self.checktype(stmt[:id_eq], stmt[id_eq + 1 :])
self.type_vars.append(type_)
self.var_nam.append(ptr_)
kernel += type_ + ptr_ + var_ + "= " + val_ + ";\n"
return kernel
for j in self.device_func_name:
if stmt.count(j) > 0:
kernel += self.device_create_func(self.device_func_name.index(j), j, stmt)
kernel = self.device_body_buff + "}\n" + kernel
self.device_body_buff = ""
return kernel
if stmt[0] == "__global" or stmt[0] == "__shared" or stmt[0] == "__register" or stmt[0] == "__constant":
self.decarrays(stmt)
return kernel
if stmt.count("for") > 0:
kernel += self._for(stmt, kernel)
return kernel
if stmt.count("if") > 0:
return kernel + grammar.keyword(stmt, kernel)
if stmt.count("else") > 0:
kernel = kernel + "else{\n"
return kernel
if stmt.count("cpu") < 1:
for j in stmt:
if self.moved.count(j) < 1 and self.arg_nam.count(j) > 0:
self.movedata(j)
if stmt.count("cpu") > 0:
self.kernel_final.append(kernel + "}")
cpu_empty = {"id": 0, "ln": 0, "return": [], "src": "", "args": [], "htod": [], "dtoh": []}
self.cpu_id = self.cpu_id + 1
cpu_empty["id"] = self.cpu_id
cpu_empty["ln"] = self.sentences.index(sentence)
eq_id = stmt.index("=")
if eq_id > 0:
cpu_empty["return"].append(stmt[0])
# if self.moved.count()
# If a data array is not sent to gpu, we use it in cpu function
func_name = stmt.index("cpu") + 2
for i in stmt[func_name + 2 : -1]:
if i is not ",":
if self.arguments.count(i) > 0:
cpu_empty["args"].append(i)
if self.moved.count(i) > 0:
cpu_empty["dtoh"].append(i)
else:
print "Pass the array accessed by the CPU function to Urutu Kernel"
stmt.remove(stmt[func_name - 2])
stmt.remove(stmt[func_name - 2])
cpu_empty["src"] = stmt
cpu_empty["isnext"] = False
if len(self.cpu_info) == 0:
self.run_gpu(self.kernel_final[-1])
self.gpu_id += 1
kernel = (
"__global__ void " + self.global_func + "_" + str(self.gpu_id) + "(" + self.kernel_args + "){\n"
)
self.ismap.append(True)
self.threads_dec = [False, False, False]
self.blocks_dec = [False, False, False]
self.remove_workitems()
self.kernel = kernel
elif self.cpu_info[-1]["ln"] + 1 != cpu_empty["ln"]:
self.cpu_info[-1]["isnext"] = False
print "Running on CPU"
self.run_cpu()
self.kernel_final.append(kernel + "}")
self.gpu_id += 1
kernel = (
"__global__ void " + self.global_func + "_" + str(self.gpu_id) + "(" + self.kernel_args + "){\n"
)
self.ismap.append(True)
self.threads_dec = [False, False, False]
self.blocks_dec = [False, False, False]
self.remove_workitems()
self.kernel = kernel
elif self.cpu_info[-1]["ln"] + 1 == cpu_empty["ln"]:
self.cpu_info[-1]["isnext"] = True
self.cpu_info.append(cpu_empty)
return kernel
if stmt.count("Urmod") > 0:
self.num_mod = self.num_mod + 1
self.modules.append(stmt[2:])
self.kernel_final.append(kernel + "}")
kernel = (
"__global__ void "
+ self.global_func
+ "_"
+ str(len(self.modules) + 1)
+ "("
+ self.kernel_args
+ "){\n"
)
self.ismap.append(True)
self.threads_dec = [False, False, False]
self.blocks_dec = [False, False, False]
self.kernel = kernel
return kernel
else:
# print "Entering Checkvars",phrase
if (len(self.cpu_info)) > 0 and self.return_kernel == False:
self.run_cpu()
return self.checkvars(stmt, phrase[-1], kernel)
def inspect_it(self, sentence, kernel):
# print "Inside inspect_it()",sentence,kernel
phrase = sentence.split('\t')
if phrase.count('#') > 0:
return
tab = phrase.count('')
## if tab > self.tabs and tab != len(phrase):
# for j in range(tab - self.tabs):
# kernel = kernel + "{\n"
if tab < self.tabs:
for j in range(self.tabs - tab):
kernel = kernel + "}\n"
self.tabs = phrase.count('')
sh = shlex.shlex(phrase[-1])
i = sh.get_token()
if i == '@' or i == 'def' or i == '' or i == '#' or i == '//' or i == '"""':
return kernel
stmt = []
while i is not sh.eof:
stmt.append(i)
i = sh.get_token()
# print stmt
for j in self.device_func_name:
if stmt.count(j) > 0:
kernel += self.device_create_func(
self.device_func_name.index(j), j, stmt)
kernel = self.device_body_buff + "}\n" + kernel
self.device_body_buff = ""
return kernel
if self.keys.count('tx') > 0 or self.keys.count("__shared"):
kernel, self.threads_dec[0] = threads.tx(self.threads_dec[0],
kernel)
if self.keys.count('ty') > 0:
kernel, self.threads_dec[1] = threads.ty(self.threads_dec[1],
kernel)
if self.keys.count('tz') > 0:
kernel, self.threads_dec[2] = threads.tz(self.threads_dec[2],
kernel)
if self.keys.count('bx') > 0:
kernel, self.blocks_dec[0] = blocks.bx(self.blocks_dec[0], kernel)
if self.keys.count('bx') > 0:
kernel, self.blocks_dec[1] = blocks.by(self.blocks_dec[1], kernel)
if self.keys.count('bz') > 0:
kernel, self.blocks_dec[2] = blocks.bz(self.blocks_dec[2], kernel)
if stmt.count('Tx') == 1 or stmt.count('Ty') == 1 or stmt.count(
'Tz') == 1:
threads.threads_decl(stmt, self.var_nam, self.var_val,
self.threads, self.type_vars)
return kernel
if stmt.count('Bx') == 1 or stmt.count('By') == 1 or stmt.count(
'Bz') == 1:
blocks.blocks_decl(stmt, self.var_nam, self.var_val, self.blocks)
return kernel
if stmt[0] == '__global' or stmt[0] == '__shared' or stmt[
0] == '__register' or stmt[0] == '__constant':
self.decarrays(stmt)
return kernel
if stmt.count('for') > 0:
kernel += self._for(stmt, kernel)
return kernel
if stmt.count('if') > 0:
return kernel + grammar.keyword(stmt, kernel)
if stmt.count('else') > 0:
kernel = kernel + "else{\n "
self.tabs += 1
return kernel
if stmt.count('Urmod') > 0:
self.modules.append(stmt[2:])
self.kernel_final.append(kernel + "}")
kernel = "__kernel void" + self.global_func + "_" + str(
len(self.modules) + 1) + "(" + self.kernel_args + "){\n"
kernel = "#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n" + kernel
self.threads_dec = [False, False, False]
self.blocks_dec = [False, False, False]
self.kernel = kernel
return kernel
else:
# print "Entering Checkvars"
return self.checkvars(stmt, phrase[-1], kernel)