-
Notifications
You must be signed in to change notification settings - Fork 1
/
predictors.py
274 lines (213 loc) · 11.2 KB
/
predictors.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
import sys
import math
import random
from prediction_elements import *
class OneLevel(BranchPredictor):
def __init__(self, num_state_bits, init_state_val, pht_size):
super().__init__(num_state_bits, init_state_val, pht_size)
def prediction_method(self, cutpc, actual_branch):
pht_address = cutpc
prediction = self.pattern_history_table[pht_address].get_state()
if actual_branch is 1:
self.pattern_history_table[pht_address].was_taken()
elif actual_branch is 0:
self.pattern_history_table[pht_address].was_not_taken()
return prediction
class TwoLevelGlobal(BranchPredictor):
def __init__(self, num_state_bits, init_state_val, pht_size):
super().__init__(num_state_bits, init_state_val, pht_size)
self.g_hist_reg_width = self.pht_numbits
self.global_branch_history = ShiftRegister(self.g_hist_reg_width)
def prediction_method(self, cutpc, actual_branch):
pht_address = self.addressing_method(cutpc, actual_branch)
prediction = self.pattern_history_table[pht_address].get_state()
self.global_branch_history.shift_in(actual_branch)
if actual_branch is 1:
self.pattern_history_table[pht_address].was_taken()
elif actual_branch is 0:
self.pattern_history_table[pht_address].was_not_taken()
return prediction
def addressing_method(self, cutpc, actual_branch):
return self.global_branch_history.get_current_val()
class GShare(TwoLevelGlobal):
def __init__(self, num_state_bits, init_state_val, pht_size):
super().__init__(num_state_bits, init_state_val, pht_size)
def addressing_method(self, cutpc, actual_branch):
return cutpc ^ self.global_branch_history.get_current_val()
def print_debug_stats(self):
print("\n---Debug---")
print("Bits in history register:\t\t", self.g_hist_reg_width)
print("Current values in global history reg:\t", self.global_branch_history.register)
print("Value of global history reg:\t\t", self.global_branch_history.get_current_val())
class TwoLevelLocal(BranchPredictor):
def __init__(self, num_state_bits, init_state_val, pht_size):
super().__init__(num_state_bits, init_state_val, pht_size)
self.g_hist_reg_width = self.pht_numbits
self.local_hist_reg_table_size = 128
self.reg_table_numbits = math.frexp(self.local_hist_reg_table_size)[1] - 1
self.cut_pc = [32, 32 - self.reg_table_numbits]
self.local_hist_reg_table = [ShiftRegister(self.g_hist_reg_width)
for i in range(self.local_hist_reg_table_size)]
def prediction_method(self, cutpc, actual_branch):
pht_address = self.local_hist_reg_table[cutpc].get_current_val()
prediction = self.pattern_history_table[pht_address].get_state()
self.local_hist_reg_table[cutpc].shift_in(actual_branch)
if actual_branch is 1:
self.pattern_history_table[pht_address].was_taken()
elif actual_branch is 0:
self.pattern_history_table[pht_address].was_not_taken()
return prediction
class TournamentPredictor:
def __init__(self, num_state_bits, init_state_val, pht_size):
offset = 0
self.pht_numbits = math.frexp(pht_size)[1] - 1
self.cut_pc = [self.pht_numbits + offset, offset]
gshare_predictor = GShare(num_state_bits, init_state_val, pht_size)
one_level_predictor = OneLevel(num_state_bits, init_state_val, pht_size)
self.predictors = [gshare_predictor, one_level_predictor]
self.meta_predictor = [StateCounter(num_state_bits, init_state_val)
for i in range(pht_size)]
init_basic_vars(self, num_state_bits, init_state_val, pht_size)
def predict(self, pc, actual_branch):
cutpc = get_from_bitrange(self.cut_pc, pc)
choosen_predictor = self.meta_predictor[cutpc].get_state()
predictions = [self.predictors[0].predict(pc, actual_branch), self.predictors[1].predict(pc, actual_branch)]
chosen_prediction = predictions[choosen_predictor]
if chosen_prediction == actual_branch:
self.good_predictions += 1
elif chosen_prediction is not None:
self.mispredictions += 1
elif chosen_prediction is None:
self.no_predictions += 1
if (predictions[0] == predictions[1]):
pass
elif (predictions[0] == actual_branch):
self.meta_predictor[cutpc].was_not_taken()
elif (predictions[1] == actual_branch):
self.meta_predictor[cutpc].was_taken()
def get_method_type(self):
return type(self).__name__.rstrip()
class TAGEPredictor:
def __init__(self, num_state_bits, init_state_val, num_base_entries):
base_predictor = TAGEBimodalBase(2, init_state_val, 4096)
# Init tagged predictors
tagged_predictors = []
for i in range (4):
tagged_predictors.append(TaggedTable(num_state_bits, init_state_val))
self.T = [base_predictor, tagged_predictors[0], tagged_predictors[1], #Predictor components, Ti
tagged_predictors[2], tagged_predictors[3]]
self.global_history_register = ShiftRegister(80)
init_basic_vars(self, num_state_bits, init_state_val, num_base_entries)
self.count = 0
self.msb_flip = True
def predict(self, pc, actual_branch):
predictions = []
tagged_predictors_index_tag = []
present_ghr_binstr = self.global_history_register.get_current_val_as_binstr()
# Base predictor 0
predictions.append(self.T[0].predict(pc, actual_branch))
# Tagged predictors 1-4
check_equal = []
tagged_predictors_index_tag = [self.index_tag_hash(pc, present_ghr_binstr, i) for i in range(1,5)]
for i in range(1,5):
predictions.append(self.T[i].predict(tagged_predictors_index_tag[i - 1][0], actual_branch))
equal = self.T[i].get_tag_at(tagged_predictors_index_tag[i - 1][0]) == tagged_predictors_index_tag[i - 1][1]
check_equal.append(equal)
provider_index = 0
for i in range(4,0,-1):
if check_equal[i - 1]:
overall_prediction = predictions[i]
provider_index = i
break
else:
overall_prediction = predictions[0]
altpred = 0
altpred_provider_index = 0
for i in range(provider_index-1,0,-1):
if check_equal[i - 1]:
altpred = predictions[i]
altpred_provider_index = i
break
else:
altpred = predictions[0]
if provider_index == 0:
self.T[0].update(pc, actual_branch)
else:
self.T[provider_index].update(tagged_predictors_index_tag[provider_index - 1][0], actual_branch)
# Update useful counter
if (altpred != overall_prediction) & (provider_index != 0):
if overall_prediction == actual_branch:
self.T[provider_index].useful_bits[tagged_predictors_index_tag[provider_index - 1][0]].was_taken()
elif overall_prediction is not None:
self.T[provider_index].useful_bits[tagged_predictors_index_tag[provider_index - 1][0]].was_not_taken()
if overall_prediction == actual_branch:
self.good_predictions += 1
elif overall_prediction is not None:
self.mispredictions += 1
# Replacement Policy
T_k_index = 0
T_j_index = 0
if provider_index != 4:
#for i in range(4,provider_index,-1):
for i in range(provider_index+1,5):
u_counter = self.T[i].useful_bits[tagged_predictors_index_tag[i-1][0]].state
if u_counter == 0:
T_k_index = i
break
else:
for tagged_component in self.T[1: (provider_index - 1)]:
for u_counter in tagged_component.useful_bits:
u_counter.was_not_taken()
if T_k_index >= 1:
for i in range(T_k_index - 1, 0,-1):
u_counter = self.T[i].useful_bits[tagged_predictors_index_tag[i-1][0]].state
if u_counter == 0:
T_j_index = i
break
else:
self.T[T_k_index].tags[tagged_predictors_index_tag[T_k_index-1][0]] = tagged_predictors_index_tag[T_k_index-1][1]
self.T[T_k_index].useful_bits[tagged_predictors_index_tag[T_k_index-1][0]].state = 0
self.T[T_k_index].counters[tagged_predictors_index_tag[T_k_index-1][0]].state = 4
if T_j_index != 0:
rand_num = random.randint(1,3)
if rand_num == 3:
self.T[T_j_index].tags[tagged_predictors_index_tag[T_j_index-1][0]] = tagged_predictors_index_tag[T_j_index-1][1]
self.T[T_j_index].useful_bits[tagged_predictors_index_tag[T_j_index-1][0]].state = 0
self.T[T_j_index].counters[tagged_predictors_index_tag[T_j_index-1][0]].state = 4
else:
self.T[T_k_index].tags[tagged_predictors_index_tag[T_k_index-1][0]] = tagged_predictors_index_tag[T_k_index-1][1]
self.T[T_k_index].useful_bits[tagged_predictors_index_tag[T_k_index-1][0]].state = 0
self.T[T_k_index].counters[tagged_predictors_index_tag[T_k_index-1][0]].state = 4
else:
self.no_predictions += 1
self.count += 1
if self.count == (256 * 1024):
if self.msb_flip:
for tagged_component in self.T[1:]:
for u_counter in tagged_component.useful_bits:
u_counter.state &= 1
else:
for tagged_component in self.T[1:]:
for u_counter in tagged_component.useful_bits:
u_counter.state &= 2
self.count = 0
self.msb_flip = not self.msb_flip
self.global_history_register.shift_in(actual_branch)
def index_tag_hash(self, pc, ghr_binstr, comp):
pc_off = 4
index_pc = get_from_bitrange([10+pc_off,0+pc_off], pc) ^ get_from_bitrange([20+pc_off,10+pc_off], pc)
index_ghr = binstr_get_from_bitrange([10,0],ghr_binstr)
tag_pc = get_from_bitrange([8+pc_off,0+pc_off], pc)
tag_R1 = binstr_get_from_bitrange([8,0], ghr_binstr)
tag_R2 = binstr_get_from_bitrange([7,0], ghr_binstr)
for i in range(1, 2**(comp - 1)):
index_ghr ^= binstr_get_from_bitrange([(i+1)*10,i*10],ghr_binstr)
for i in range(1, math.floor( ( (2**(comp - 1) * 10) / 8) ) ):
tag_R1 ^= binstr_get_from_bitrange([(i+1)*8,i*8],ghr_binstr)
for i in range(1, math.floor( ( (2**(comp - 1) * 10) / 7) ) ):
tag_R2 ^= binstr_get_from_bitrange([(i+1)*7,i*7],ghr_binstr)
index = index_pc ^ index_ghr
tag = tag_pc ^ tag_R1 ^ (tag_R2 << 1)
return [index, tag]
def get_method_type(self):
return type(self).__name__.rstrip()