def cmp_stat (traffic, size, network, stat_dir, subnet, scale):
# extract mbnoc4 stat (subnet=4)
stat_out = []
if scale is True: scale_factor = MBNoC_collect_data.clk_scale_factor
else: scale_factor = 1
for sim_index in range (1, MBNoC_collect_data.SIM_COUNT + 1, 1):
stat = collect_stat.collect(sim_index, size, stat_dir, MBNoC_collect_data.insns_count)
if stat is None:
continue
inject_rate = "{:.2f}".format(float(stat['inject_rate'])/subnet) # must be scaled based on the flit size
latency = "{:.2f}".format(float(stat['pkt_net_latency'])/scale_factor)
deflect_flit_per_cycle = "{:.2f}".format(float(stat['deflect_flit_per_cycle'])/subnet)
throughput = "{:.2f}".format(float(stat['throughput'])/subnet*scale_factor)
stat_out.append((inject_rate, 0.00, latency, throughput, deflect_flit_per_cycle)) # energy is 0, unless I wanna have some fun to get it.
return stat_out
def evaluation():
global synth_entry_mbnoc
global synth_entry_bless
global final_stat
global break_down_power
global overall_power
global file_dir_bless
global file_dir_mbnoc
global max_defl
synth_entry_mbnoc = [] # important to reset the global list.
synth_entry_bless = [] # important to reset the global list.
if node == 16:
network = '4x4'
elif node == 64:
network = '8x8'
elif node == 256:
network = '16x16'
else:
raise Exception('Undefined Network Size')
if SIM_COUNT < 0:
raise Exception('Simulate 0 File ?')
if subnet < 1:
raise Exception('NO subnet!')
#file_dir_bless = file_dir_bless + network + '/'
#file_dir_mbnoc = file_dir_mbnoc + network + '/'
##insns_count = 10000001
component_count_mbnoc = [0, 16, 1, 5, 1, 1, 1, 4] # only consider the inter-router links, not include local and bypass link
component_count_mbnoc = [x*subnet for x in component_count_mbnoc]
component_count_bless = [0, 14, 1, 5, 1, 1, 1, 4]
port_on_compoment_mbnoc = [0, 1, 5, 1, 5, 5, 4, 1]
port_on_compoment_bless = [0, 1, 5, 1, 5, 0, 4, 1]
effective_count = 0
overall_cycle_bless = 0
overall_cycle_mbnoc = 0
stat_aggr = []
static_power_breakdown_mbnoc = [0, 0, 0, 0, 0, 0, 0, 0]
dynamic_power_breakdown_mbnoc = [0, 0, 0, 0, 0, 0, 0, 0]
overall_power_breakdown_mbnoc = [0, 0, 0, 0, 0, 0, 0, 0]
static_power_breakdown_bless = [0, 0, 0, 0, 0, 0, 0, 0]
dynamic_power_breakdown_bless = [0, 0, 0, 0, 0, 0, 0, 0]
overall_power_breakdown_bless = [0, 0, 0, 0, 0, 0, 0, 0]
for sim_index in range (1, SIM_COUNT+ 1, 1):
## Collect result
stat_bless = collect_stat.collect(sim_index, node, file_dir_bless, insns_count)
stat_mbnoc = collect_stat.collect(sim_index, node, file_dir_mbnoc, insns_count)
if (stat_bless == None or stat_mbnoc == None):
continue
## comupte weighted speedup
effective_count = effective_count + 1
ipc_alone = stat_bless['ipc']
ipc_share = stat_mbnoc['ipc']
active_cycle_mbnoc = stat_mbnoc['active_cycles']
active_cycle_bless = stat_bless['active_cycles']
accm_ws = 0
for i in range (0, node, 1):
accm_ws = float(ipc_share [i]) / float(ipc_alone[i]) + accm_ws
overall_cycle_bless = overall_cycle_bless + int(active_cycle_bless[i])
overall_cycle_mbnoc = overall_cycle_mbnoc + int(active_cycle_mbnoc[i])
w_speedup_mbnoc = float("{:.2f}".format(accm_ws))
throughput_bless = float(stat_bless['throughput'])
throughput_mbnoc = float(stat_mbnoc['throughput'])
normalized_throughput = float("{:.2f}".format(throughput_mbnoc / throughput_bless/subnet*clk_scale_factor)) # scaled by subnet size because of the flit size is smaller
deflect_flit_per_cycle_bless = float(stat_bless['deflect_flit_per_cycle'])
deflect_flit_per_cycle_mbnoc = float(stat_mbnoc['deflect_flit_per_cycle'])/subnet
if deflect_flit_per_cycle_bless > 0: normalized_deflection_rate = float("{:.2f}".format(deflect_flit_per_cycle_mbnoc / deflect_flit_per_cycle_bless)) # scaled by subnet size because of the flit size is smaller
else: normalized_deflection_rate = 1
deflect_per_flit_bless = float(stat_bless['deflect_per_flit'])
if max_defl < deflect_per_flit_bless: max_defl = deflect_per_flit_bless
mpki = float(stat_mbnoc['mpki'])
## energy computation
# event order: router, dff, portAllo, rc, xbar, local, permNet, link
traversal_mbnoc = float(stat_mbnoc['traversal'])
permute_mbnoc = float(stat_mbnoc['permute'])
link_mbnoc = traversal_mbnoc - float(stat_mbnoc['inj_flit'])*2 - float(stat_mbnoc['bypass'])
event_mbnoc = [0, traversal_mbnoc*3, traversal_mbnoc, traversal_mbnoc, traversal_mbnoc, traversal_mbnoc, permute_mbnoc, link_mbnoc]
actual_toggle_rate_mbnoc = power.comp_toggle_rate (event_mbnoc, component_count_mbnoc, port_on_compoment_mbnoc, overall_cycle_mbnoc)
link_toggle_rate_mbnoc = actual_toggle_rate_mbnoc.pop(7)
traversal_bless = float(stat_bless['traversal'])
permute_bless = float(stat_bless['permute'])
link_bless = traversal_bless - float(stat_bless['inj_flit'])*2
event_bless = [0, traversal_bless*3, traversal_bless, traversal_bless, traversal_bless, 0, permute_bless, link_bless]
actual_toggle_rate_bless = power.comp_toggle_rate (event_bless, component_count_bless, port_on_compoment_bless, overall_cycle_bless)
link_toggle_rate_bless = actual_toggle_rate_bless.pop(7)
simul_time_mbnoc = mbnoc_clk_period * overall_cycle_mbnoc # in second
simul_time_bless = bless_clk_period * overall_cycle_bless # in second
dynamic_breakdown_mbnoc = power.comp_dynamic_power (switch_mbnoc, internal_mbnoc, actual_toggle_rate_mbnoc, dynamic_mbnoc_link, link_toggle_rate_mbnoc, simul_time_mbnoc, default_toggle_rate)
dynamic_breakdown_bless = power.comp_dynamic_power (switch_bless, internal_bless, actual_toggle_rate_bless, dynamic_bless_link, link_toggle_rate_bless, simul_time_bless, default_toggle_rate)
static_breakdown_mbnoc = power.comp_static_power (static_mbnoc, component_count_mbnoc, simul_time_mbnoc)
static_breakdown_bless = power.comp_static_power (static_bless, component_count_bless, simul_time_bless)
overall_breakdown_mbnoc = [x+y for x,y in zip(static_breakdown_mbnoc, dynamic_breakdown_mbnoc)]
overall_breakdown_bless = [x+y for x,y in zip(static_breakdown_bless, dynamic_breakdown_bless)]
sum_dynamic_mbnoc = sum (dynamic_breakdown_mbnoc)
sum_dynamic_bless = sum (dynamic_breakdown_bless)
sum_static_mbnoc = sum (static_breakdown_mbnoc)
sum_static_bless = sum (static_breakdown_bless)
energy_mbnoc = sum_dynamic_mbnoc + sum_static_mbnoc # mbnoc power
energy_bless = sum_dynamic_bless + sum_static_bless # bless power
sum_energy_mbnoc = sum_dynamic_mbnoc + energy_mbnoc # mbnoc power
sum_energy_bless = sum_dynamic_bless + energy_bless # bless power
static_power_breakdown_mbnoc = [x+y for x,y in zip (static_power_breakdown_mbnoc, static_breakdown_mbnoc)]
static_power_breakdown_bless = [x+y for x,y in zip (static_power_breakdown_bless, static_breakdown_bless)]
dynamic_power_breakdown_mbnoc = [x+y for x,y in zip (dynamic_power_breakdown_mbnoc, dynamic_breakdown_mbnoc)]
dynamic_power_breakdown_bless = [x+y for x,y in zip (dynamic_power_breakdown_bless, dynamic_breakdown_bless)]
overall_power_breakdown_mbnoc = [x+y for x,y in zip (overall_power_breakdown_mbnoc, overall_breakdown_mbnoc)]
overall_power_breakdown_bless = [x+y for x,y in zip (overall_power_breakdown_bless, overall_breakdown_bless)]
normalized_energy = sum_energy_mbnoc / sum_energy_bless
stat_element = (mpki, w_speedup_mbnoc, float("{:.2f}".format(normalized_energy)), normalized_throughput, normalized_deflection_rate)
stat_aggr.append(stat_element)
###### For synthetic traffic
inject_rate_mbnoc = "{:.2f}".format(float(stat_mbnoc['inject_rate'])/subnet) # must be scaled based on the flit size
tot_latency_mbnoc = "{:.2f}".format(float(stat_mbnoc['pkt_tot_latency'])/clk_scale_factor)
synth_entry_mbnoc.append((inject_rate_mbnoc, "{:.2f}".format(energy_mbnoc), tot_latency_mbnoc, str("{:.2f}".format(throughput_mbnoc/subnet*clk_scale_factor)), str(deflect_flit_per_cycle_mbnoc)))
inject_rate_bless = "{:.2f}".format(float(stat_bless['inject_rate']))
tot_latency_bless = "{:.2f}".format(float(stat_bless['pkt_tot_latency']))
synth_entry_bless.append((inject_rate_bless, "{:.2f}".format(energy_bless), tot_latency_bless, str(throughput_bless), str(deflect_flit_per_cycle_bless)))
sorted_stat = sorted(stat_aggr, key=lambda tup: tup[0])
if effective_count is not 0:
static_power_breakdown_mbnoc = [float("{:.2f}".format(x / effective_count)) for x in static_power_breakdown_mbnoc]
dynamic_power_breakdown_mbnoc = [float("{:.2f}".format(x / effective_count)) for x in dynamic_power_breakdown_mbnoc]
overall_power_breakdown_mbnoc = [float("{:.2f}".format(x / effective_count)) for x in overall_power_breakdown_mbnoc]
static_power_breakdown_bless = [float("{:.2f}".format(x / effective_count)) for x in static_power_breakdown_bless]
dynamic_power_breakdown_bless = [float("{:.2f}".format(x / effective_count)) for x in dynamic_power_breakdown_bless]
overall_power_breakdown_bless = [float("{:.2f}".format(x / effective_count)) for x in overall_power_breakdown_bless]
overall_static_mbnoc = sum(static_power_breakdown_mbnoc)
overall_dynamic_mbnoc = sum(dynamic_power_breakdown_mbnoc)
overall_static_bless = sum(static_power_breakdown_bless)
overall_dynamic_bless = sum(dynamic_power_breakdown_bless)
else:
static_power_breakdown_mbnoc = 0
dynamic_power_breakdown_mbnoc = 0
overall_power_breakdown_mbnoc = 0
static_power_breakdown_bless = 0
dynamic_power_breakdown_bless = 0
overall_power_breakdown_bless = 0
overall_static_mbnoc = 0
overall_dynamic_mbnoc = 0
overall_static_bless = 0
overall_dynamic_bless = 0
overall_mbnoc = overall_static_mbnoc + overall_dynamic_mbnoc
overall_bless = overall_static_bless + overall_dynamic_bless
break_down_power = [static_power_breakdown_bless, static_power_breakdown_mbnoc,dynamic_power_breakdown_bless, dynamic_power_breakdown_mbnoc,overall_power_breakdown_bless,overall_power_breakdown_mbnoc]
overall_power = [overall_static_bless, overall_dynamic_bless, overall_bless, overall_static_mbnoc, overall_dynamic_mbnoc, overall_mbnoc]
#my_print.print_power_breakdown(break_down_power)
#my_print.print_power(overall_power)
## Categorize stat based on mpki
low_stat = []
medium_stat = []
high_stat = []
for element in sorted_stat:
if element[0] <= class_factor[0]:
low_stat.append(element)
elif element[0] <= class_factor[1]:
medium_stat.append(element)
else:
high_stat.append(element)
final_low_stat = final (low_stat)
final_med_stat = final (medium_stat)
final_high_stat = final (high_stat)
final_avg_stat = final (sorted_stat)
final_stat = [final_low_stat, final_med_stat, final_high_stat, final_avg_stat]
#final_output = my_print.print_final_stat(final_stat)
return final_stat