diff --git a/llvm/projects/soc_simulator/src/driver.py b/llvm/projects/soc_simulator/src/driver.py
index 6f224c6c62aaed338afe9732c39484f98957a6a2..85f17d687d645f99c6c95bf33a87b63004635b29 100644
--- a/llvm/projects/soc_simulator/src/driver.py
+++ b/llvm/projects/soc_simulator/src/driver.py
@@ -38,10 +38,10 @@ def parse_tensor_layer_file(layer_filename):
tensor_layer["H"] = float(layer_data[3])
tensor_layer["W"] = float(layer_data[4])
tensor_layer["Cout"] = float(layer_data[5])
- tensor_layer["Kh"] = float(layer_data[6])
- tensor_layer["Kw"] = float(layer_data[7])
- tensor_layer["Sh"] = float(layer_data[8])
- tensor_layer["Sw"] = float(layer_data[9])
+ tensor_layer["Kh"] = float(layer_data[7])
+ tensor_layer["Kw"] = float(layer_data[8])
+ tensor_layer["Sh"] = float(layer_data[9])
+ tensor_layer["Sw"] = float(layer_data[10])
elif is_fc(layer_name):
tensor_layer["RA"] = float(layer_data[1])
@@ -102,34 +102,31 @@ def parse_tensor_table(table_filename):
table_file.close()
fp16_swing = 8
-iterations = 10
class ApproxTypes:
FP16 = 0
FP32 = 1
PROMISE = 2
+def get_approx_type(approx_type):
+ if approx_type == 0:
+ return "fp16"
+ elif approx_type == 1:
+ return "fp32"
+ return "promise"
+
def is_promise(config_layer):
# TODO overhead in call to split?
return float(config_layer.split(' ')[0]) < fp16_swing
-# NOTE smart_dma is always true
+
def quantize(curr_layer, prev_layer, h2f_f2h_operation_ind, layer_data):
+ print(get_approx_type(curr_layer), get_approx_type(prev_layer))
if curr_layer == prev_layer or curr_layer == ApproxTypes.PROMISE \
or prev_layer == ApproxTypes.PROMISE: # No quantization needed
return 0.0, 0.0
layer_name = layer_data["Name"]
- '''
- if is_conv(layer_name):
- size = layer_data["N"] * layer_data["Cin"] * layer_data["H"] * layer_data["W"] \
- + layer_data["Cout"] * layer_data["Cin"] * layer_data["Kh"] * layer_data["Kw"]
- elif is_fc(layer_name):
- size = layer_data["RA"] * layer_data["CA"] + layer_data["RB"] * layer_data["CB"]
- elif not is_nml(layer_name):
- print("ERROR: Invalid layer name %s" % layer_name)
- exit(1)
- '''
# NOTE: Ignoring logic where curr == promise or prev == promise bc
# smartDMA is always true so we'd return near the beginning of the method
@@ -138,17 +135,23 @@ def quantize(curr_layer, prev_layer, h2f_f2h_operation_ind, layer_data):
# (which is why order matters in the tensor table)
tensor_op_row = tensor_table[layer_name][h2f_f2h_operation_ind]
if curr_layer == ApproxTypes.FP32:
- return tensor_op_row["h2f_time"], tensor_op_row["h2f_energy"]
+ time = tensor_op_row["h2f_time"]
+ energy = tensor_op_row["h2f_energy"]
elif curr_layer == ApproxTypes.FP16:
- return tensor_op_row["f2h_time"], tensor_op_row["f2h_energy"]
- assert(False) # Error: Should never reach this section
+ time = tensor_op_row["f2h_time"]
+ energy = tensor_op_row["f2h_energy"]
+ print("Quantization: (%f, %f)" % (time, energy))
+ return (time, energy)
def run_promise_simulation(swing, layer_data):
layer_name = layer_data["Name"]
patch_factor = 1
if is_conv(layer_name):
+ print("is conv")
+ print(layer_data["N"], layer_data["H"], layer_data["W"], layer_data["Sh"], layer_data["Sw"], layer_data["Cin"], layer_data["Kh"], layer_data["Kw"], layer_data["Cout"])
+
rows_a = layer_data["N"] * layer_data["H"] * layer_data["W"] \
/ (layer_data["Sh"] * layer_data["Sw"])
cols_a = layer_data["Cin"] * layer_data["Kh"] * layer_data["Kw"]
@@ -156,6 +159,7 @@ def run_promise_simulation(swing, layer_data):
cols_b = layer_data["Cout"]
patch_factor = layer_data["Kh"] * layer_data["Kw"]
elif is_fc(layer_name):
+ print("is fc")
rows_a = layer_data["RA"]
cols_a = layer_data["CA"]
rows_b = cols_a
@@ -163,7 +167,7 @@ def run_promise_simulation(swing, layer_data):
else:
print("PROMISE can't run whatever this layer is.")
exit(1)
-
+ #print("[%f x %f] x [%f x %f] : %f" % (rows_a, cols_a, rows_b, cols_b, swing))
# Run promise simulator
# TODO need to print time and energy in the ptm runner so we can pipe it
output = subprocess.Popen(["./ptm", str(rows_a), str(cols_a), str(rows_b), \
@@ -172,11 +176,11 @@ def run_promise_simulation(swing, layer_data):
total_time_energy = output.strip().split(',')
assert(len(total_time_energy) == 2)
+ print("PROMISE: (%s, %s)" % (total_time_energy[0], total_time_energy[1]))
return float(total_time_energy[0]), float(total_time_energy[1])
def run_gpu_simulation(curr_layer, layer_name, tensor_ind):
- print(layer_name, tensor_ind)
tensor_info = tensor_table[layer_name][tensor_ind]
if curr_layer == ApproxTypes.FP32:
conversion_time = tensor_info["fp32_time"]
@@ -219,11 +223,9 @@ def run_simulations(config_filename):
layer_name = layer_data["Name"]
if is_promise(config_layer):
- print("IS PROMISE")
print("Running layer %s on PROMISE" % layer_name)
curr_layer = ApproxTypes.PROMISE
quant_time, quant_energy = quantize(curr_layer, prev_layer, 0, layer_data)
-
# Compute
time, energy = run_promise_simulation(config_layer, layer_data)
aggregate_results[results_time_key][config_count] += time
@@ -235,23 +237,24 @@ def run_simulations(config_filename):
total_time = 0
total_energy = 0
- print(tensor_ops)
for tensor_ind, tensor_op in enumerate(tensor_ops): # sublevle
+ tensor_op = int(tensor_op)
if tensor_op == fp16_swing:
curr_layer = ApproxTypes.FP16
else:
curr_layer = ApproxTypes.FP32
quant_time, quant_energy = quantize(curr_layer, prev_layer, tensor_ind, \
layer_data)
+
conv_time, conv_energy = run_gpu_simulation(curr_layer, layer_name, tensor_ind)
total_time += conv_time
total_time += conv_energy
aggregate_results[results_time_key][config_count] += total_time
aggregate_results[results_energy_key][config_count] += total_energy
- print("DONE WITH LAYER")
prev_layer = curr_layer
config_count += 1
+ print("\n")
config_count += 1 # because we're storing the count and not the index
config_file.close()
@@ -295,4 +298,4 @@ if __name__ == "__main__":
parse_tensor_layer_file("/home/nvidia/Gitlab/hpvm/llvm/projects/hpvm-tensor-rt/build_mobilenet/mobilenet_layers.txt")
parse_tensor_table("/home/nvidia/Gitlab/hpvm/llvm/projects/hpvm-tensor-rt/build_pldi/mobilenet_results/mobilenet_tensors.txt")
run_simulations("/home/nvidia/Gitlab/hpvm/llvm/projects/hpvm-tensor-rt/build_mobilenet/mobilenet_conf2.txt")
- display_results("blah.txt")
+ #display_results("blah.txt")