def argparser():
''' Argument parser. '''
ap = argparse.ArgumentParser()
ap.add_argument('net',
help='network name, should be a .py file under "nns". '
'Choices: {}.'.format(', '.join(all_networks())))
ap.add_argument('--batch', type=int, required=True, help='batch size')
ap.add_argument('--word', type=int, default=16, help='word size in bits')
ap.add_argument('--nodes',
type=int,
nargs=2,
required=True,
metavar=('H', 'W'),
help='Parallel node partitioning dimensions')
ap.add_argument('--array',
type=int,
nargs=2,
required=True,
metavar=('H', 'W'),
help='PE array dimensions')
ap.add_argument('--regf',
type=int,
required=True,
help='register file size in bytes per PE')
ap.add_argument('--gbuf',
type=int,
required=True,
help='global buffer size in bytes')
ap.add_argument('--bus-width',
type=int,
default=0,
help='array bus width in bits. set 0 to ignore')
ap.add_argument('--dram-bw',
type=float,
default='inf',
help='total DRAM bandwidth in bytes per cycle.')
ap.add_argument('--op-cost',
type=float,
default=1,
help='cost of arithmetic operation')
ap.add_argument('--hier-cost',
type=float,
nargs=4,
default=[200, 6, 2, 1],
metavar=('DRAM_COST', 'GBUF_COST', 'ITCN_COST',
'REGF_COST'),
help='cost of access to memory hierarchy')
ap.add_argument('--hop-cost',
type=float,
default=10,
help='cost of access through one NoC hop')
ap.add_argument('--unit-idle-cost',
type=float,
default=0,
help='static cost over all nodes for unit execution time')
ap.add_argument('--mem-type',
default='2D',
choices=['2D', '3D'],
help='memory type. "2D" has memory only on edge nodes; '
'"3D" has memory vertially on top of all nodes.')
ap.add_argument('--disable-bypass',
nargs='*',
default=[],
choices=['i', 'o', 'f'],
help='whether disallowing gbuf bypass for i (input), o '
'(output), or f (filter)')
ap.add_argument('--solve-loopblocking',
action='store_true',
help='Use analytical solver to choose loop blocking. '
'Otherwise use exhaustive search.')
ap.add_argument('--enable-access-forwarding',
action='store_true',
help='Each node fetches a subset of data and forwards to '
'other nodes.')
ap.add_argument('--enable-gbuf-sharing',
action='store_true',
help='Share gbuf capacity across nodes through NoC.')
ap.add_argument('--enable-save-writeback',
action='store_true',
help='Allow to save the writeback to memory for the '
'intermediate data between layers if able to '
'store the entire data set in on-chip buffers.')
ap.add_argument('--disable-interlayer-opt',
'--basic-interlayer-partition',
action='store_true',
help='Disable optimizations and only allow basic '
'inter-layer pipeline.')
ap.add_argument(
'--hybrid-partition',
'--hybrid-partition2d', # deprecated old name
action='store_true',
help='Use hybrid partition for layer for node mapping. '
'Otherwise use naive method based on layer type.')
ap.add_argument('--batch-partition',
action='store_true',
help='Allow partitioning batch, i.e., consider data '
'parallelism.')
ap.add_argument('--ifmaps-partition',
'--ifmap-partition',
action='store_true',
help='Allow partitioning ifmap channel dimension, which '
'requires extra data synchronization.')
ap.add_argument('--interlayer-partition',
'--inter-layer-partition',
action='store_true',
help='Allow partitioning resources across multiple layers '
'and process them simultaneously as an inter-layer '
'pipeline.')
ap.add_argument('--layer-pipeline-time-overhead',
type=float,
default=float('inf'),
help='maximum allowed execution time overhead due to '
'layer pipelining.')
ap.add_argument('--layer-pipeline-max-degree',
type=float,
default=float('inf'),
help='maximum allowed layer pipelining degree, i.e., '
'number of vertices in a pipeline segment.')
ap.add_argument('-g',
'--goal',
default='e',
choices=['e', 'd', 'ed', 'E', 'D', 'ED'],
help='Goal of optimization: E(nergy), D(elay), or ED.')
ap.add_argument('-t',
'--top',
type=int,
default=1,
help='Number of top schedules to keep during search.')
ap.add_argument('-p',
'--processes',
type=int,
default=multiprocessing.cpu_count() / 2,
help='Number of parallel processes to use for search.')
ap.add_argument('-v',
'--verbose',
action='store_true',
help='Show progress and details.')
return ap
def argparser():
''' Argument parser. '''
ap = argparse.ArgumentParser()
ap.add_argument('net',
help='network name, should be a .py file under "nns". '
'Choices: {}.'.format(', '.join(all_networks())))
ap.add_argument('--batch', type=int, required=True, help='batch size')
ap.add_argument('--word', type=int, default=16, help='word size in bits')
ap.add_argument('--nodes',
type=int,
nargs=2,
required=True,
metavar=('H', 'W'),
help='Parallel node partitioning dimensions')
ap.add_argument('--array',
type=int,
nargs=2,
required=True,
metavar=('H', 'W'),
help='PE array dimensions')
ap.add_argument('--regf',
type=int,
required=True,
help='register file size in bytes per PE')
ap.add_argument('--gbuf',
type=int,
required=True,
help='global buffer size in bytes')
ap.add_argument('--op-cost',
type=float,
default=1,
help='cost of arithmetic operation')
ap.add_argument('--hier-cost',
type=float,
nargs=4,
default=[200, 6, 2, 1],
metavar=('DRAM_COST', 'GBUF_COST', 'ITCN_COST',
'REGF_COST'),
help='cost of access to memory hierarchy')
ap.add_argument('--hop-cost',
type=float,
default=10,
help='cost of access through one NoC hop')
ap.add_argument('--unit-static-cost',
type=float,
default=0,
help='static cost for unit execution time')
ap.add_argument('--mem-type',
default='2D',
choices=['2D', '3D'],
help='memory type. "2D" has memory only on edge nodes; '
'"3D" has memory vertially on top of all nodes.')
ap.add_argument('--disable-bypass',
nargs='*',
default=[],
choices=['i', 'o', 'f'],
help='whether disallowing gbuf bypass for i (input), o '
'(output), or f (filter)')
ap.add_argument('--solve-loopblocking',
action='store_true',
help='Use analytical solver to choose loop blocking. '
'Otherwise use exhaustive search.')
ap.add_argument(
'--hybrid-partition',
'--hybrid-partition2d', # deprecated old name
action='store_true',
help='Use hybrid partition for layer for node mapping. '
'Otherwise use naive method based on layer type.')
ap.add_argument('--batch-partition',
action='store_true',
help='Allow partitioning batch, i.e., consider data '
'parallelism.')
ap.add_argument('--ifmaps-partition',
'--ifmap-partition',
action='store_true',
help='Allow partitioning ifmap channel dimension, which '
'requires extra data synchronization.')
ap.add_argument('-t',
'--top',
type=int,
default=1,
help='Number of top schedules to keep during search.')
ap.add_argument('-p',
'--processes',
type=int,
default=multiprocessing.cpu_count() / 2,
help='Number of parallel processes to use for search.')
ap.add_argument('-v',
'--verbose',
action='store_true',
help='Show progress and details.')
return ap
def test_import_network(self):
''' Get import_network. '''
for name in nns.all_networks():
network = nns.import_network(name)
self.assertIsInstance(network, Network)
def setUp(self):
self.net = {}
net = Network('net1')
# Linear.
net.set_input_layer(InputLayer(10, 1))
net.add('0', FCLayer(10, 20))
net.add('1', FCLayer(20, 30))
net.add('1p', PoolingLayer(30, 1, 1))
net.add('2', FCLayer(30, 40))
net.add('3', FCLayer(40, 50))
self.net[net.net_name] = net
net = Network('net2')
# Long linear.
net.set_input_layer(InputLayer(1, 1))
for idx in range(16):
net.add(str(idx), FCLayer(1, 1))
self.net[net.net_name] = net
net = Network('net3')
# Fork.
# /0-2\ /6- 7- 8\
# x 4-5 12
# \1-3/ \9-10-11/
net.set_input_layer(InputLayer(1, 1))
net.add('0', FCLayer(1, 1), prevs=net.INPUT_LAYER_KEY)
net.add('1', FCLayer(1, 1), prevs=net.INPUT_LAYER_KEY)
net.add('2', FCLayer(2, 1), prevs=('0', '1'))
net.add('2p', PoolingLayer(1, 1, 1))
net.add('3', FCLayer(2, 1), prevs=('0', '1'))
net.add('4', FCLayer(2, 1), prevs=('2p', '3'))
net.add('5', FCLayer(1, 1))
net.add('5p', PoolingLayer(1, 1, 1))
net.add('6', FCLayer(1, 1), prevs='5p')
net.add('7', FCLayer(1, 1))
net.add('8', FCLayer(1, 1))
net.add('9', FCLayer(1, 1), prevs='5p')
net.add('10', FCLayer(1, 1))
net.add('11', FCLayer(1, 1))
net.add('12', FCLayer(2, 1), prevs=('8', '11'))
self.net[net.net_name] = net
net = Network('net4')
# Complex fork.
# /5 \
# 0-1-2-3-4-6-7-8-10-14
# \9/
# \11-12 /
# \13 /
net.set_input_layer(InputLayer(1, 1))
net.add('0', FCLayer(1, 1))
net.add('1', FCLayer(1, 1))
net.add('2', FCLayer(1, 1))
net.add('3', FCLayer(1, 1))
net.add('4', FCLayer(1, 1))
net.add('5', FCLayer(1, 1), prevs='4')
net.add('6', FCLayer(1, 1), prevs='4')
net.add('7', FCLayer(1, 1))
net.add('8', FCLayer(1, 1), prevs='7')
net.add('9', FCLayer(1, 1), prevs='7')
net.add('10', FCLayer(1, 1))
net.add('10p', PoolingLayer(2, 1, 1), prevs=('8', '10'))
net.add('11', PoolingLayer(1, 1, 1), prevs='4')
net.add('12', FCLayer(1, 1))
net.add('13', PoolingLayer(1, 1, 1), prevs='4')
net.add('14', FCLayer(5, 1), prevs=('5', '10p', '12', '13'))
self.net[net.net_name] = net
net = Network('net5')
# Corner cases.
# ----\
# //1-2\ 7-8\
# 0-3-4-x 10-11-12
# \ \5/ 9 / \__/
# 6--/
net.set_input_layer(InputLayer(1, 1))
net.add('0', FCLayer(1, 1))
net.add('1', FCLayer(1, 1), prevs='0')
net.add('2', FCLayer(1, 1))
net.add('3', FCLayer(1, 1), prevs='0')
net.add('4', FCLayer(1, 1), prevs='3')
net.add('5', FCLayer(1, 1), prevs='3')
net.add('6', FCLayer(1, 1), prevs='0')
net.add('7', FCLayer(5, 1), prevs=('0', '2', '4', '5', '6'))
net.add('8', FCLayer(1, 1))
net.add('9', FCLayer(5, 1), prevs=('0', '2', '4', '5', '6'))
net.add('10', FCLayer(2, 1), prevs=('8', '9'))
net.add('11', FCLayer(1, 1))
net.add('12', FCLayer(2, 1), prevs=('10', '11'))
self.net[net.net_name] = net
net = Network('net6')
# Fmap sizes.
net.set_input_layer(InputLayer(1, 24))
net.add('0', ConvLayer(1, 1, 24, 3))
net.add('1', ConvLayer(1, 1, 12, 3, strd=2))
net.add('1p', PoolingLayer(1, 6, 2))
net.add('2', ConvLayer(1, 1, 6, 3))
net.add('3', ConvLayer(1, 1, 6, 3, strd=4), prevs=('0'))
self.net[net.net_name] = net
net = Network('net7')
# Topological order: see a visited vertex again.
# /---
# 0-1-\\
# \2--2p
net.set_input_layer(InputLayer(1, 1))
net.add('0', FCLayer(1, 1))
net.add('1', FCLayer(1, 1), prevs='0')
net.add('2', FCLayer(1, 1), prevs='0')
net.add('2p', PoolingLayer(3, 1, 1), prevs=('0', '1', '2'))
self.net[net.net_name] = net
net = Network('net8')
# Forward to the middle.
# /-\
# 0-1-2-2p-4-4p
# \-3------/
net.set_input_layer(InputLayer(1, 1))
net.add('0', FCLayer(1, 1))
net.add('1', FCLayer(1, 1), prevs='0')
net.add('2', FCLayer(1, 1), prevs='1')
net.add('2p', PoolingLayer(2, 1, 1), prevs=('1', '2'))
net.add('3', FCLayer(1, 1), prevs='0')
net.add('4', FCLayer(2, 1), prevs='2p')
net.add('4p', PoolingLayer(2, 1, 1), prevs=('3', '4'))
self.net[net.net_name] = net
net = Network('net9')
# Previous layers include input and others.
net.set_input_layer(InputLayer(1, 1))
net.add('0', FCLayer(1, 1))
net.add('1', FCLayer(2, 1), prevs=(net.INPUT_LAYER_KEY, '0'))
self.net[net.net_name] = net
# Real networks.
for net_name in all_networks():
self.net[net_name] = import_network(net_name)
self.batch_size = 16
self.resource = Resource(
proc_region=NodeRegion(origin=PhyDim2(0, 0),
dim=PhyDim2(8, 8),
type=NodeRegion.PROC),
dram_region=NodeRegion(origin=PhyDim2(0, 0),
dim=PhyDim2(8, 8),
type=NodeRegion.DRAM),
src_data_region=NodeRegion(origin=PhyDim2(0, 0),
dim=PhyDim2(8, 4),
type=NodeRegion.DRAM),
dst_data_region=NodeRegion(origin=PhyDim2(0, 4),
dim=PhyDim2(8, 4),
type=NodeRegion.DRAM),
dim_array=PhyDim2(16, 16),
size_gbuf=65536,
size_regf=64,
array_bus_width=float('inf'),
dram_bandwidth=float('inf'),
no_time_mux=False)
part = PartitionScheme(order=range(pe.NUM), pdims=[(1, 1)] * pe.NUM)
self.ofmap_layout = DataLayout(
frngs=(FmapRange((0, 0, 0, 0), (2, 4, 16, 16)), ),
regions=(NodeRegion(origin=PhyDim2(0, 0),
dim=PhyDim2(1, 1),
type=NodeRegion.DRAM), ),
parts=(part, ))
def test_all_networks(self):
''' Get all_networks. '''
self.assertIn('alex_net', nns.all_networks())
self.assertIn('vgg_net', nns.all_networks())
self.assertGreater(len(nns.all_networks()), 5)