Architecture of Computing Systems (Lecture Notes in Computer ...

Recommendations

Info

118 P. Petoumenos et al. IQ: n 4-instruction segments MLP-distance (instr) = 11 instructions miss other { { MLP-distance (segments) = 3 segments MLP info is detected in MDB (an n-entry circular queue) ... n ... n Fig. 3. MLP Distance Buffer (MDB) MLP-distance is measured in the number of segments between misses. misses i.e. possible dependencies between the instructions may cause their misses to be isolated. In any case, the “actual” MLP-distance will be less than or equal to the value produced by our approach. This in turn means we might miss some opportunities for downward IQ resizing but we will not incur performance degradation due to loss of MLP. Our experiments showed that for most benchmarks falsely assuming parallel misses causes few overestimations of the MLP-distance. Considering the simplicity of our mechanism, these results indicate a satisfactory level of performance. Measuring the MLP-distance allows us to control the IQ size for the relevant part of the code so as to not hurt MLP. We need, however, to relate this information back to the instruction stream so the next time we see the same part of the code we can react and correctly set the size of the IQ. 4.2 Associating MLP-Distance with Code For the purpose of managing the IQ in a MLP-aware fashion, we dynamically divide program execution into fragments and associate MLP-distance information with these fragments. Execution fragments should be comparable in size to the size of the IQ. Much shorter fragments would be sub-optimal since the information they carry will be used to manage the whole IQ, which contains multiple such fragments. This could lead to very frequent and —many times— conflicting resizing decisions. Longer fragments, such as program phases, also fail to provide us with the fine-grain information we need to quickly react to fast-changing MLP-distances in the instruction stream. To achieve the desired balance in the size of the fragments we use the notion of superpaths. A superpath is nothing more than an aggregation of sequentially executed basic blocks and loosely corresponds to a trace (in a trace cache) [13] or a hotspot [12]. The MLP-distance of a superpath is assigned by the MDB: when all of the MDB entries belonging to a superpath are “retired,” the longest MLP-distance stored in these entries is selected to update the MLP-distance of the superpath. Note that the instructions which establish this MLP-distance do not have to belong to the same superpath. An MLP-distance that straddles a number of superpaths affects all of them (if it is the maximum observed MLP-distance in each of them). The next time the same superpath is observed in dispatch, its stored information is retrieved to manage the IQ. A more detailed discussion about superpaths and how we actually implemented the aforementioned mechanisms can be found in Section 5.
MLP-Aware Instruction Queue Resizing: The Key to Power-Efficient Performance 119 4.3 Resizing Policy When we start tracking a superpath in dispatch, we check whether we have stored information about its behavior, including its MLP-distance information. If there is such information then we have an indication about the minimum IQ size which will not affect the MLP of this superpath. In many cases, however, there is no MLPdistance information available or the MLP-distance does not restrict IQ downsizing. The question is how much can we downsize the IQ is such cases? As explained in Section 3, an efficient IQ resizing scheme has to find the IQ size that minimizes energy consumption without hurting performance. This means that besides not hurting MLP we must also protect ILP. This gap can be filled by any of the existing IQ resizing techniques. For example, the ILP-feedback approach of [2] can provide a target IQ size that does not hurt ILP while the MLP-aware approach judges whether this size hurts MLP and if so it overrides the decision. For the rest of this paper the ILP-feedback information will be provided by the decision making mechanism in Folegnani and González [2]. This mechanism examines the participation of the youngest segment of the IQ to the IPC within a specific number of cycles. If the contribution of the youngest part is not important, namely the number of instructions that issue from this segment is below a threshold, the IQ size is decreased. In our case, we deactivate the youngest segment when it empties. The main idea in the Folegnani and González work is that if a segment contributes very little to ILP, deactivating it would not harm performance. However, this holds only for ILP —not for MLP. In other words, even if the contribution of a segment in issued instructions is very small, it can still have a significant impact on performance, if any of its issued instructions is involved in MLP. It is exactly for such cases where MLP-awareness makes all the difference. Further, in the Folegnani and González work, once the IQ is downsized, there is no way to detect whether the situation changes —all we can see is the contribution to ILP of the active portion of the IQ. Thus, periodically, the IQ is brought back to its full size, and the downsizing process starts again. In our case, the existence of MLP automatically upsizes the IQ, to a size that does not harm MLP. 5 Practical Implementation 5.1 IQ Segmentation To allow dynamic variation of the IQ size, we divide the IQ in a number of independent parts referred as segments. For example, we use an 128-entry IQ partitioned into eight, sixteen-entry segments. Bitline segmentation is used to implement the resizing of the structure [10]. The structure of the IQ follows the one in [2]. The IQ is a circular FIFO queue, with every new instruction inserted at the tail; retiring instructions are removed from the head. The difference in our case, is that individual segments can be deactivated. A segment is deactivated if instructions from the youngest segment contribute less than threshold instructions in a quantum of time (1000 cycles) and a segment is reactivated every 5 quanta. This inevitably leads to constraints that have to be met during the resizing process, similarly to those faced by Ponomarev et al. [11]:
Page 2 and 3:
Lecture Notes in Computer Science 5
Page 4 and 5:
Volume Editors Christian Müller-Sc
Page 6 and 7:
General Chair Organization Christia
Page 8 and 9:
Organization IX Hartmut Schmeck Kar
Page 10 and 11:
Keynote Table of Contents HyVM - Hy
Page 12 and 13:
Table of Contents XIII JetBench:AnO
Page 14 and 15:
How to Enhance a Superscalar Proces
Page 16 and 17:
4 J. Mische et al. The Real-time Vi
Page 18 and 19:
6 J. Mische et al. 4.1 Instruction
Page 20 and 21:
8 J. Mische et al. Additionally the
Page 22 and 23:
10 J. Mische et al. % of unused pip
Page 24 and 25:
12 J. Mische et al. % of cycles spe
Page 26 and 27:
14 J. Mische et al. 16. Lickly, B.,
Page 28 and 29:
16 G. Aşılıoğlu, E.M. Kaya, and
Page 30 and 31:
Page 32 and 33:
Page 34 and 35:
Page 36 and 37:
Page 38 and 39:
26 T.B. Preußer, P. Reichel, and R
Page 40 and 41:
Page 42 and 43:
Page 44 and 45:
Page 46 and 47:
Page 48 and 49:
Page 50 and 51:
38 P. Bellasi, W. Fornaciari, and D
Page 52 and 53:
Page 54 and 55:
Page 56 and 57:
Page 58 and 59:
Page 60 and 61:
Page 62 and 63:
50 J. Zeppenfeld and A. Herkersdorf
Page 64 and 65:
Page 66 and 67:
Page 68 and 69:
Page 70 and 71:
Page 72 and 73:
Page 74 and 75:
62 B. Jakimovski, B. Meyer, and E.
Page 76 and 77:
Page 78 and 79:
Page 80 and 81: 68 B. Jakimovski, B. Meyer, and E.
Page 86 and 87: 74 M. Bonn and H. Schmeck Fig. 1. J
Page 88 and 89: 76 M. Bonn and H. Schmeck 2.2 Node
Page 90 and 91: 78 M. Bonn and H. Schmeck Uptime-ba
Page 92 and 93: 80 M. Bonn and H. Schmeck 2.4 Simul
Page 94 and 95: 82 M. Bonn and H. Schmeck done rate
Page 96 and 97: 84 M. Bonn and H. Schmeck Fig. 8. J
Page 98 and 99: 86 M. Bonn and H. Schmeck tells the
Page 100 and 101: 88 J.-P. Steghöfer et al. � �
Page 102 and 103: 90 J.-P. Steghöfer et al. mechanis
Page 104 and 105: 92 J.-P. Steghöfer et al. resource
Page 106 and 107: 94 J.-P. Steghöfer et al. Choose a
Page 108 and 109: 96 J.-P. Steghöfer et al. 1. Defin
Page 110 and 111: 98 J.-P. Steghöfer et al. and all
Page 112 and 113: 100 J.-P. Steghöfer et al. 19. Kim
Page 114 and 115: 102 K. Kloch et al. large-scale sys
Page 116 and 117: 104 K. Kloch et al. a�t� 1.0 0.
Page 118 and 119: 106 K. Kloch et al. constant. This
Page 120 and 121: 108 K. Kloch et al. Relative number
Page 122 and 123: 110 K. Kloch et al. (a) infection r
Page 124 and 125: 112 K. Kloch et al. (ii) Phase of a
Page 126 and 127: 114 P. Petoumenos et al. Studying t
Page 128 and 129: 116 P. Petoumenos et al. % of Misse
Page 132 and 133: 120 P. Petoumenos et al. downsizing
Page 134 and 135: 122 P. Petoumenos et al. As long as
Page 136 and 137: 124 P. Petoumenos et al. comparable
Page 138 and 139: Exploiting Inactive Rename Slots fo
Page 140 and 141: 128 M. Kayaalp et al. In a supersca
Page 142 and 143: 130 M. Kayaalp et al. INSTRUCTION 1
Page 144 and 145: 132 M. Kayaalp et al. time. Alterna
Page 146 and 147: 134 M. Kayaalp et al. Fig. 3. Numbe
Page 148 and 149: 136 M. Kayaalp et al. The results o
Page 150 and 151: Efficient Transaction Nesting in Ha
Page 152 and 153: 140 Y. Liu et al. HTMs include TCC
Page 154 and 155: 142 Y. Liu et al. rollback T0 begin
Page 156 and 157: 144 Y. Liu et al. Processor core Pr
Page 158 and 159: 146 Y. Liu et al. 5.2 Results and A
Page 160 and 161: 148 Y. Liu et al. decreases, partia
Page 162 and 163: Decentralized Energy-Management to
Page 164 and 165: 152 B. Becker et al. Furthermore, t
Page 166 and 167: 154 B. Becker et al. An optimizing
Page 168 and 169: 156 B. Becker et al. smart-home man
Page 170 and 171: 158 B. Becker et al. freedom like w
Page 172 and 173: 160 B. Becker et al. Power [W] 4500
Page 174 and 175: EnergySaving Cluster Roll: Power Sa
Page 176 and 177: 164 M.F. Dolz et al. Themodulequeri
Page 178 and 179: 166 M.F. Dolz et al. This daemon al
Page 180 and 181:
168 M.F. Dolz et al. been submitted
Page 182 and 183:
170 M.F. Dolz et al. On the other h
Page 184 and 185:
172 M.F. Dolz et al. at the inactiv
Page 186 and 187:
Effect of the Degree of Neighborhoo
Page 188 and 189:
176 T. Abdullah et al. A zone based
Page 190 and 191:
178 T. Abdullah et al. A consumer/p
Page 192 and 193:
180 T. Abdullah et al. Messages 140
Page 194 and 195:
182 T. Abdullah et al. (except when
Page 196 and 197:
184 T. Abdullah et al. % Matchmakin
Page 198 and 199:
186 T. Abdullah et al. show that th
Page 200 and 201:
188 M. Schindewolf, D. Kramer, and
Page 202 and 203:
Page 204 and 205:
Page 206 and 207:
Page 208 and 209:
Page 210 and 211:
Page 212 and 213:
200 R. Plyaskin and A. Herkersdorf
Page 214 and 215:
Page 216 and 217:
Page 218 and 219:
Page 220 and 221:
Page 222 and 223:
Page 224 and 225:
212 M.Y. Qadri, D. Matichard, and K
Page 226 and 227:
Page 228 and 229:
Page 230 and 231:
Page 232 and 233:
Page 234 and 235:
A Tightly Coupled Accelerator Infra
Page 236 and 237:
224 F. Nowak and R. Buchty where A
Page 238 and 239:
226 F. Nowak and R. Buchty Fig. 3.
Page 240 and 241:
228 F. Nowak and R. Buchty Table 2.
Page 242 and 243:
230 F. Nowak and R. Buchty Table 4.
Page 244 and 245:
232 F. Nowak and R. Buchty 5.3 Comp
Page 246 and 247:
Optimizing Stencil Application on M
Page 248 and 249:
236 F. Xudong et al. compared to th
Page 250 and 251:
238 F. Xudong et al. stencil comput
Page 252 and 253:
240 F. Xudong et al. threads in the
Page 254 and 255:
242 F. Xudong et al. Speedup 14 12
Page 256 and 257:
244 F. Xudong et al. 5 Related Work
Page 258:
Abdullah, Tariq 174 Alima, Luc Onan
show all

Architecture of Computing Systems (Lecture Notes in Computer ...

You also want an ePaper? Increase the reach of your titles

Delete template?

Save as template?