def build_unsharp(app_data, g_size=None, t_size=None):
pipe_data = app_data['pipe_data']
out_unsharp, inline_func = unsharp_mask(pipe_data)
R = pipe_data['R']
C = pipe_data['C']
threshold = pipe_data['threshold']
weight = pipe_data['weight']
live_outs = [out_unsharp]
pipe_name = app_data['app']
rows = app_data['rows'] - 2
cols = app_data['cols'] - 2
p_estimates = [(R, rows), (C, cols)]
p_constraints = [ Condition(R, "==", rows), \
Condition(C, "==", cols) ]
if (t_size == None):
t_size = [1, 8, 256]
if (g_size == None):
g_size = 4
opts = []
if app_data['early_free']:
opts += ['early_free']
if app_data['optimize_storage']:
opts += ['optimize_storage']
if app_data['pool_alloc']:
opts += ['pool_alloc']
if app_data['inline']:
opts += ['inline']
if app_data['multi-level-tiling']:
opts += ['multi-level-tiling']
if app_data['dpfuse']:
opts += ['dpfuse']
if app_data['logdpchoices']:
opts += ['logdpchoices']
if app_data['logmaxchildren']:
logmaxchildren = app_data['logmaxchildren']
pipe = buildPipeline(live_outs,
param_estimates=p_estimates,
param_constraints=p_constraints,
tile_sizes=t_size,
group_size=g_size,
options=opts,
pipe_name=pipe_name,
logMaxChildren=logmaxchildren)
return pipe
def build_unsharp(app_data):
pipe_data = app_data['pipe_data']
out_unsharp = unsharp_mask(pipe_data)
R = pipe_data['R']
C = pipe_data['C']
threshold = pipe_data['threshold']
weight = pipe_data['weight']
live_outs = [out_unsharp]
pipe_name = app_data['app']
rows = app_data['rows']-2
cols = app_data['cols']-2
p_estimates = [(R, rows), (C, cols)]
p_constraints = [ Condition(R, "==", rows), \
Condition(C, "==", cols) ]
t_size = [16, 16]
g_size = 11
opts = []
if app_data['early_free']:
opts += ['early_free']
if app_data['optimize_storage']:
opts += ['optimize_storage']
if app_data['pool_alloc']:
opts += ['pool_alloc']
pipe = buildPipeline(live_outs,
param_estimates=p_estimates,
param_constraints=p_constraints,
tile_sizes = t_size,
group_size = g_size,
options = opts,
pipe_name = pipe_name)
return pipe
def build_unsharp(app_data):
pipe_data = app_data['pipe_data']
out_unsharp = unsharp_mask(pipe_data)
R = pipe_data['R']
C = pipe_data['C']
threshold = pipe_data['threshold']
weight = pipe_data['weight']
live_outs = [out_unsharp]
pipe_name = app_data['app']
rows = app_data['rows'] - 2
cols = app_data['cols'] - 2
p_estimates = [(R, rows), (C, cols)]
p_constraints = [ Condition(R, "==", rows), \
Condition(C, "==", cols) ]
t_size = [16, 16]
g_size = 11
opts = []
if app_data['early_free']:
opts += ['early_free']
if app_data['optimize_storage']:
opts += ['optimize_storage']
if app_data['pool_alloc']:
opts += ['pool_alloc']
pipe = buildPipeline(live_outs,
param_estimates=p_estimates,
param_constraints=p_constraints,
tile_sizes=t_size,
group_size=g_size,
options=opts,
pipe_name=pipe_name)
return pipe
def auto_tune(app_data):
pipe_data = app_data['pipe_data']
app_name = app_data['app_name']
pipe_name = app_data['app']
pipe_func_name = "pipeline_" + pipe_name
out_unsharp, inline_func = unsharp_mask(pipe_data)
live_outs = [out_unsharp]
R = pipe_data['R']
C = pipe_data['C']
rows = app_data['rows'] - 2
cols = app_data['cols'] - 2
threshold = app_data['threshold']
weight = app_data['weight']
param_estimates = [(R, rows), (C, cols)]
param_constraints = [ Condition(R, "==", rows), \
Condition(C, "==", cols) ]
dst_path = "/tmp"
group_size_configs = [int(i * 4) for i in [0.2, 0.4, 0.5]]
#group_size_configs = [10]
tile_size_configs = []
tile_size_configs.append([64, 256])
tile_size_configs.append([64, 128])
tile_size_configs.append([32, 512])
tile_size_configs.append([32, 256])
tile_size_configs.append([32, 128])
tile_size_configs.append([32, 64])
tile_size_configs.append([16, 512])
tile_size_configs.append([16, 256])
tile_size_configs.append([16, 128])
tile_size_configs.append([16, 64])
tile_size_configs.append([8, 512])
tile_size_configs.append([8, 256])
tile_size_configs.append([8, 128])
tile_size_configs.append([8, 64])
tile_size_configs.append([8, 32])
# relative path to root directory from app dir
ROOT = app_data['ROOT']
opts = []
if app_data['early_free']:
opts += ['early_free']
if app_data['optimize_storage']:
opts += ['optimize_storage']
if app_data['pool_alloc']:
opts += ['pool_alloc']
if app_data['dpfuse']:
opts += ['dpfuse']
gen_compile_string(app_data)
cxx_string = app_data['cxx_string']
# Generate Variants for Tuning
# ============================
gen_config = {
"_tuner_app_name": app_name,
"_tuner_live_outs": live_outs,
"_tuner_param_constraints": param_constraints, #optional
"_tuner_param_estimates": param_estimates, #optional
"_tuner_tile_size_configs": tile_size_configs, #optional
"_tuner_group_size_configs": group_size_configs, #optional
"_tuner_opts": opts, #optional
"_tuner_dst_path": dst_path, # optional
"_tuner_cxx_string": cxx_string, # optional
"_tuner_root_path": ROOT, # needed if pool_alloc is set
"_tuner_debug_flag": True, # optional
"_tuner_opt_datadict": app_data
}
_tuner_src_path, _tuner_configs_count, _tuner_pipe = \
tuner.generate(gen_config)
'''
_tuner_src_path = '/tmp/PolycNUUEYoLt2Mage'
_tuner_configs_count = 75
_tuner_pipe = buildPipeline(live_outs)
'''
img_data = app_data['img_data']
IN = img_data['IN']
OUT = img_data['OUT']
pipe_args = {}
pipe_args['R'] = rows
pipe_args['C'] = cols
pipe_args['threshold'] = threshold
pipe_args['weight'] = weight
pipe_args['img'] = IN
pipe_args['mask'] = OUT
# Execute the generated variants
# ==============================
exec_config = {
"_tuner_app_name": app_name,
"_tuner_pipe": _tuner_pipe,
"_tuner_pipe_arg_data": pipe_args,
"_tuner_src_path": _tuner_src_path, # optional
"_tuner_configs_count": _tuner_configs_count, # optional
"_tuner_omp_threads": 4, # optional
"_tuner_nruns": int(app_data['app_args'].runs), # optional
"_tuner_debug_flag": True, # optional
#"_tuner_custom_executor": minimal_exec_mg,
"_tuner_app_data": app_data
}
tuner.execute(exec_config)
def auto_tune(app_data):
pipe_data = app_data['pipe_data']
app_name = app_data['app_name']
pipe_name = app_data['app']
pipe_func_name = "pipeline_"+pipe_name
out_unsharp = unsharp_mask(pipe_data)
live_outs = [out_unsharp]
R = pipe_data['R']
C = pipe_data['C']
rows = app_data['rows']-2
cols = app_data['cols']-2
threshold = app_data['threshold']
weight = app_data['weight']
param_estimates = [(R, rows), (C, cols)]
param_constraints = [ Condition(R, "==", rows), \
Condition(C, "==", cols) ]
dst_path = "/tmp"
group_size_configs = [3, 5, 7, 9, 10]
#group_size_configs = [10]
tile_size_configs = []
tile_size_configs.append([64, 256])
tile_size_configs.append([64, 128])
tile_size_configs.append([32, 512])
tile_size_configs.append([32, 256])
tile_size_configs.append([32, 128])
tile_size_configs.append([32, 64])
tile_size_configs.append([16, 512])
tile_size_configs.append([16, 256])
tile_size_configs.append([16, 128])
tile_size_configs.append([16, 64])
tile_size_configs.append([8, 512])
tile_size_configs.append([8, 256])
tile_size_configs.append([8, 128])
tile_size_configs.append([8, 64])
tile_size_configs.append([8, 32])
# relative path to root directory from app dir
ROOT = app_data['ROOT']
opts = []
if app_data['early_free']:
opts += ['early_free']
if app_data['optimize_storage']:
opts += ['optimize_storage']
if app_data['pool_alloc']:
opts += ['pool_alloc']
gen_compile_string(app_data)
cxx_string = app_data['cxx_string']
# Generate Variants for Tuning
# ============================
gen_config = {"_tuner_app_name": app_name,
"_tuner_live_outs": live_outs,
"_tuner_param_constraints": param_constraints, #optional
"_tuner_param_estimates": param_estimates, #optional
"_tuner_tile_size_configs": tile_size_configs, #optional
"_tuner_group_size_configs": group_size_configs, #optional
"_tuner_opts": opts, #optional
"_tuner_dst_path" : dst_path, # optional
"_tuner_cxx_string" : cxx_string, # optional
"_tuner_root_path" : ROOT, # needed if pool_alloc is set
"_tuner_debug_flag": True, # optional
"_tuner_opt_datadict": app_data
}
_tuner_src_path, _tuner_configs_count, _tuner_pipe = \
tuner.generate(gen_config)
'''
_tuner_src_path = '/tmp/PolycNUUEYoLt2Mage'
_tuner_configs_count = 75
_tuner_pipe = buildPipeline(live_outs)
'''
img_data = app_data['img_data']
IN = img_data['IN']
OUT = img_data['OUT']
pipe_args = {}
pipe_args['R'] = rows
pipe_args['C'] = cols
pipe_args['threshold'] = threshold
pipe_args['weight'] = weight
pipe_args['img'] = IN
pipe_args['mask'] = OUT
# Execute the generated variants
# ==============================
exec_config = {"_tuner_app_name": app_name,
"_tuner_pipe": _tuner_pipe,
"_tuner_pipe_arg_data": pipe_args,
"_tuner_src_path": _tuner_src_path, # optional
"_tuner_configs_count": _tuner_configs_count, # optional
"_tuner_omp_threads": 4, # optional
"_tuner_nruns": 1, # optional
"_tuner_debug_flag": True, # optional
#"_tuner_custom_executor": minimal_exec_mg,
"_tuner_app_data": app_data
}
tuner.execute(exec_config)