def cal(type,dep) :
dataset = 64
partition = 10
#T1 Fusion
filename ="FusedKernelT"+str(type)
if dep==1 :
filename= filename+"Dependency.txt"
else :
filename =filename+"NoDependency.txt"
change_stdout(filename)
if dep== 0:
info1 = json.loads(open("bilateralFilterKernel.json").read())
info2 = json.loads(open("depth2vertexKernel.json").read())
kernel1 = fw.Kernel(info1, dataset=dataset, partition=partition)
kernel2 = fw.Kernel(info2,dataset=dataset,partition=partition)
kernel_fused = kernel_fusion(kernel1,kernel2,type,0,0)
else :
info3 = json.loads(open("DependentKernel.json").read())
kernel3 = fw.Kernel(info3[0],dataset=dataset,partition=partition)
kernel4 = fw.Kernel(info3[1],dataset=dataset,partition=partition)
kernel_fused = kernel_fusion(kernel3,kernel4,type,0,0)
def initialize(self):
info = json.loads(open(self.info_file).read())
kernel = fw.Kernel(info)
if self.thread_coarsening_cpu > 1:
kernel.src_cpu = self.thread_coarsen(self.info_file,
self.thread_coarsening_cpu,
self.stride)
if self.thread_coarsening_gpu > 1:
kernel.src_gpu = self.thread_coarsen(self.info_file,
self.thread_coarsening_gpu,
self.stride)
kernel.src = kernel.src_cpu # Redundant assignment to avoid errors
kernel.macros = self.transform.macros
print kernel.local_args
for index in xrange(0, len(kernel.local_args)):
print index
kernel.local_args[index][
'size'] = self.transform.local_buffer_sizes[index]
for buffer_type in ['input', 'output']:
for index in xrange(len(kernel.buffer_info[buffer_type])):
kernel.buffer_info[buffer_type][index][
'size'] = self.transform.buffer_sizes[buffer_type][index]
kernel.buffer_info[buffer_type][index][
'chunk'] = self.transform.buffer_chunks[buffer_type][index]
kernel.local_work_size = self.transform.local_work_size
kernel.global_work_size = self.transform.global_work_size
kernel.local_chunk = self.transform.local_chunk
self.kernel = kernel
s_name = src_name[:-5] #name of Kernel
if args.log:
logging.basicConfig(level=logging.DEBUG)
cmd_qs, ctxs, gpus, cpus = fw.host_initialize(int(args.nGPU),
int(args.nCPU))
info_file = args.file
with open(info_file, "r") as f:
info = json.loads(f.read())
dataset = int(args.dataset_size)
st = time.time()
if args.dump_output_file:
fw.dump_output = True
if args.partition_class != None:
partition = int(args.partition_class)
kernel = fw.Kernel(info, dataset=dataset, partition=partition)
else:
kernel = fw.Kernel(info, dataset=dataset)
partition = info['partition']
name = s_name + '_' + str(partition) + '_' + str(dataset) + '_' + str(
time.time()).replace(".", "")
if args.log:
f_path = fw.SOURCE_DIR + 'logs/' + name + '_debug.log'
logging.basicConfig(filename=f_path, level=logging.DEBUG)
print "LOG file is saved at %s" % f_path
logging.debug('Number of GPUs : %d' % (len(gpus)))
logging.debug(gpus)
logging.debug('Number of CPUs : %d' % (len(cpus)))
logging.debug(cpus)
sched_start_time = datetime.datetime.now()
#####This is the main code
if __name__ == '__main__':
args = parse_arg(sys.argv[1:])
task_files = args.file
kernels = []
for task in open(task_files,"r").readlines():
task_src, partition, dataset = task.strip("\n").split(" ")
info = json.loads(open(fw.SOURCE_DIR + "info/" + task_src).read())
logging.debug( "Appending kernel" + task_src + " " + partition + " " + dataset)
kernels.append(fw.Kernel(info, partition=int(partition), dataset=int(dataset)))
for kernel1 in range(len(kernels)):
for kernel2 in range(len(kernels)) :
#fuse these two kernels and
name = "scheduling_" + args.select +"_" +str(time.time()).replace(".", "")
dump_dev =None
if args.dump_output_file:
fw.dump_output = True
if args.log:
f_path = fw.SOURCE_DIR + 'logs/' + name + '_debug.log'
logging.basicConfig(filename=f_path, level=logging.DEBUG)
if args.select == "baseline":