def configDMAFPGA(self, rank_id):
binToSend = bytearray(np.array([self.parameters['neural_net'][rank_id]['rank_info']['ifm_baseaddr']], dtype=np.int64).byteswap().tobytes())
binToSend.extend(bytearray(np.array([self.parameters['neural_net'][rank_id]['command'][0]['ifm_width'] * self.parameters['neural_net'][rank_id]['command'][0]['ifm_height'] * self.parameters['neural_net'][rank_id]['command'][0]['ifm_depth'] * 2], dtype=np.int64).byteswap().tobytes()))
binToSend.extend(bytearray(np.array([self.parameters['neural_net'][rank_id]['rank_info']['ofm_baseaddr']], dtype=np.int64).byteswap().tobytes()))
binToSend.extend(bytearray(np.array([self.parameters['neural_net'][rank_id]['command'][len(self.parameters['neural_net'][rank_id]['command']) - 1]['ofm_height'] * self.parameters['neural_net'][rank_id]['command'][len(self.parameters['neural_net'][rank_id]['command'])-1] ['ofm_width'] * self.parameters['neural_net'][rank_id]['command'][len(self.parameters['neural_net'][rank_id]['command'])-1]['output_channels'] * 2], dtype=np.int64).byteswap().tobytes()))
GalapagosNet.binToStream(self, binToSend, self.parameters['neural_net'][rank_id]['rank_info']['dst_rank'], 'init' + str(rank_id))
def test(self, rank_id):
binToSend = bytearray(8) #address
print binToSend
print "len of long is " + str(len(binToSend))
b = bytearray.fromhex("0800 00 00 00 00 00 00")
binToSend.extend(bytearray(b)) #size
print "len of long is " + str(len(binToSend))
GalapagosNet.binToStream(
self, binToSend,
self.parameters['neural_net'][rank_id]['rank_info']['dst_rank'])
binToSend = bytearray(np.array([16]).tobytes()) #dma parameters
print len(bytearray(np.array([16]).tobytes()))
GalapagosNet.binToStream(
self, binToSend,
self.parameters['neural_net'][rank_id]['rank_info']['dst_rank'])
def configParametersFPGA(self, rank_id):
binToSend = bytearray(
np.array([
self.parameters['neural_net'][rank_id]['rank_info']
['biases_weights_baseaddr']
],
dtype=np.int64).byteswap().tobytes())
print len(binToSend)
length = 0
#calculate length of all parameters on fpga
for cmd_id, cmd in enumerate(
self.parameters['neural_net'][rank_id]['command']):
length = length + len(self.parameters['neural_net'][rank_id]
['command'][cmd_id]['configParameters'])
print "length of command " + str(cmd_id) + " is " + str(length)
print "length of command is " + str(length)
binToSend.extend(
np.array([length], dtype=np.int64).byteswap().tobytes())
print len(binToSend)
#send configuration parameters to all fpga_ranks
GalapagosNet.binToStream(
self, binToSend,
self.parameters['neural_net'][rank_id]['rank_info']['dst_rank'],
'init' + str(rank_id))
#send all weights for all commands on fpga
for cmd_id, cmd in enumerate(
self.parameters['neural_net'][rank_id]['command']):
#print binascii.hexlify(self.parameters['neural_net'][rank_id]['command'][cmd_id]['configParameters'])
GalapagosNet.binToStream(
self, self.parameters['neural_net'][rank_id]['command'][cmd_id]
['configParameters'], self.parameters['neural_net'][rank_id]
['rank_info']['dst_rank'], 'init' + str(rank_id))
def configCmdFPGA(self, rank_id):
# cmd_num_commands = np.array([1],dtype=np.int64).byteswap()
cmd_num_commands = np.array(
[len(self.parameters['neural_net'][rank_id]['command'])],
dtype=np.int64).byteswap()
for cmd_id, cmd in enumerate(
self.parameters['neural_net'][rank_id]['command']):
cmd_conv = np.array([
self.parameters['neural_net'][rank_id]['command'][cmd_id]
['ifm_height'], self.parameters['neural_net'][rank_id]
['command'][cmd_id]['ifm_width'], self.parameters['neural_net']
[rank_id]['command'][cmd_id]['kernel_height'],
self.parameters['neural_net'][rank_id]['command'][cmd_id]
['kernel_width'], self.parameters['neural_net'][rank_id]
['command'][cmd_id]['stride'], self.parameters['neural_net']
[rank_id]['command'][cmd_id]['pad'],
self.parameters['neural_net'][rank_id]['command'][cmd_id]
['ofm_height'], self.parameters['neural_net'][rank_id]
['command'][cmd_id]['ofm_width'], self.parameters['neural_net']
[rank_id]['command'][cmd_id]['ifm_slices'],
self.parameters['neural_net'][rank_id]['command'][cmd_id]
['ofm_slices'], self.parameters['neural_net'][rank_id]
['command'][cmd_id]['ofm_fragments'],
self.parameters['neural_net'][rank_id]['command'][cmd_id]
['ifm_mem_fragments']
],
dtype='uint16').byteswap()
cmd_addr = np.array([
self.parameters['neural_net'][rank_id]['command'][cmd_id]
['ifm_baseaddr'], self.parameters['neural_net'][rank_id]
['command'][cmd_id]['ifm_packet_length'],
self.parameters['neural_net'][rank_id]['command'][cmd_id]
['ifm_depth_offset'], self.parameters['neural_net'][rank_id]
['command'][cmd_id]['ifm_height_offset'],
self.parameters['neural_net'][rank_id]['command'][cmd_id]
['ofm_baseaddr'], self.parameters['neural_net'][rank_id]
['command'][cmd_id]['ofm_packet_length'],
self.parameters['neural_net'][rank_id]['command'][cmd_id]
['weights_baseaddr'], self.parameters['neural_net'][rank_id]
['command'][cmd_id]['weights_packet_length'],
self.parameters['neural_net'][rank_id]['command'][cmd_id]
['weight_depth_offset']
],
dtype='uint32').byteswap()
cmd_mode = np.array([0, 0], dtype='uint16')
cmd_pool = np.array([
0, 1, self.parameters['neural_net'][rank_id]['command'][cmd_id]
['pool_input_height'], self.parameters['neural_net'][rank_id]
['command'][cmd_id]['pool_input_width'],
self.parameters['neural_net'][rank_id]['command'][cmd_id]
['pool_kernel_height'], self.parameters['neural_net'][rank_id]
['command'][cmd_id]['pool_kernel_width'],
self.parameters['neural_net'][rank_id]['command'][cmd_id]
['pool_output_height'], self.parameters['neural_net'][rank_id]
['command'][cmd_id]['pool_output_width'],
self.parameters['neural_net'][rank_id]['command'][cmd_id]
['pool_stride'], 0
],
dtype='uint16').byteswap()
cmd_relu_bias = np.array([
self.parameters['neural_net'][rank_id]['command'][cmd_id]
['relu'], self.parameters['neural_net'][rank_id]['command']
[cmd_id]['bias']
],
dtype='uint8')
cmd_bias_addr = np.array([
self.parameters['neural_net'][rank_id]['command'][cmd_id]
['biases_weights_baseaddr']
],
dtype='uint32').byteswap()
cmd_rsvd = np.zeros((11, ), dtype='uint32')
if (cmd_id == 0):
#self.cmd = cmd_num_commands.tobytes() + \
# cmd_conv.tobytes() + \
# cmd_addr.tobytes() + \
# cmd_mode.tobytes() + \
# cmd_pool.tobytes() + \
# cmd_rsvd.tobytes()
self.cmd = cmd_mode.tobytes() + \
cmd_pool.tobytes() + \
cmd_rsvd.tobytes()
else:
self.cmd.extend(
bytearray(cmd_num_commands.tobytes() + cmd_conv.tobytes() +
cmd_addr.tobytes() + cmd_mode.tobytes() +
cmd_pool.tobytes() + cmd_rsvd.tobytes()))
self.cmd = bytearray(self.cmd)
self.cmd.extend(
bytearray(
np.array([
self.parameters['neural_net'][rank_id]['rank_info']
['batch_size']
],
dtype=np.int64).byteswap()))
self.cmd.extend(
bytearray(
np.array([
self.parameters['neural_net'][rank_id]['rank_info']
['num_ranks']
],
dtype=np.int64).byteswap()))
GalapagosNet.binToStream(
self, self.cmd,
self.parameters['neural_net'][rank_id]['rank_info']['dst_rank'],
'init' + str(rank_id))
# self.configDMAFPGA(rank_id)
# self.configCmdFPGA(rank_id)
def test(self, rank_id):
binToSend = bytearray(8) #address
print binToSend
print "len of long is " + str(len(binToSend))
b = bytearray.fromhex("0800 00 00 00 00 00 00")
binToSend.extend(bytearray(b)) #size
print "len of long is " + str(len(binToSend))
<<<<<<< HEAD
GalapagosNet.binToStream(self, binToSend, self.parameters['neural_net'][rank_id]['dst_rank'])
=======
GalapagosNet.binToStream(self, binToSend, self.parameters['neural_net'][rank_id]['rank_info']['dst_rank'])
>>>>>>> a83aeb56cba6bb8684d30caa1660636ec0f54f99
binToSend = bytearray(np.array([16]).tobytes()) #dma parameters
print len(bytearray(np.array([16]).tobytes()))
<<<<<<< HEAD
GalapagosNet.binToStream(self, binToSend, self.parameters['neural_net'][rank_id]['dst_rank'])
=======
GalapagosNet.binToStream(self, binToSend, self.parameters['neural_net'][rank_id]['rank_info']['dst_rank'])
>>>>>>> a83aeb56cba6bb8684d30caa1660636ec0f54f99