176:
563:- which means that the system should prefetch 7 elements ahead. With the first iteration, i will be 0, so the system prefetches the 7th element. Now, with this arrangement, the first 7 accesses (i=0->6) will still be misses (under the simplifying assumption that each element of array1 is in a separate cache line of its own).
780:
171:. This buffer is called a stream buffer and is separate from the cache. The processor then consumes data/instructions from the stream buffer if the address associated with the prefetched blocks match the requested address generated by the program executing on the processor. The figure below illustrates this setup:
784:
While it appears that having perfect accuracy might imply that there are no misses, this is not the case. The prefetches themselves might result in new misses if the prefetched blocks are placed directly into the cache. Although these may be a small fraction of the total number of misses observed
294:
This class of prefetchers look for memory access streams that repeat over time. E.g. In this stream of memory accesses: N, A, B, C, E, G, H, A, B, C, I, J, K, A, B, C, L, M, N, O, A, B, C, ...; the stream A,B,C is repeating over time. Other design variation have tried to provide more efficient,
184:
Whenever the prefetch mechanism detects a miss on a memory block, say A, it allocates a stream to begin prefetching successive blocks from the missed block onward. If the stream buffer can hold 4 blocks, then the processor would prefetch A+1, A+2, A+3, A+4 and hold those in the allocated stream
90:
is typically accomplished by having a dedicated hardware mechanism in the processor that watches the stream of instructions or data being requested by the executing program, recognizes the next few elements that the program might need based on this stream and prefetches into the processor's
303:
Computer applications generate a variety of access patterns. The processor and memory subsystem architectures used to execute these applications further disambiguate the memory access patterns they generate. Hence, the effectiveness and efficiency of prefetching schemes often depend on the
712:
22:
is a technique used by computer processors to boost execution performance by fetching instructions or data from their original storage in slower memory to a faster local memory before it is actually needed (hence the term 'prefetch'). Most modern computer processors have fast and local
726:
659:
192:
This mechanism can be scaled up by adding multiple such 'stream buffers' - each of which would maintain a separate prefetch stream. For each new miss, there would be a new stream buffer allocated and it would operate in a similar way as described
185:
buffer. If the processor consumes A+1 next, then it shall be moved "up" from the stream buffer to the processor's cache. The first entry of the stream buffer would now be A+2 and so on. This pattern of prefetching successive blocks is called
579:
Software prefetching works well only with loops where there is regular array access as the programmer has to hand code the prefetch instructions, whereas hardware prefetchers work dynamically based on the program's behavior at
304:
application and the architectures used to execute them. Recent research has focused on building collaborative mechanisms to synergistically use multiple prefetching schemes for better prefetching coverage and accuracy.
667:
411:
At each iteration, the i element of the array "array1" is accessed. Therefore, the system can prefetch the elements that are going to be accessed in future iterations by inserting a "prefetch" instruction as shown
285:
In this case, the delta between the addresses of consecutive memory accesses is variable but still follows a pattern. Some prefetchers designs exploit this property to predict and prefetch for future accesses.
796:
Consider a for loop where each iteration takes 3 cycles to execute and the 'prefetch' operation takes 12 cycles. This implies that for the prefetched data to be useful, the system must start the prefetch
587:
Hardware prefetching also has less CPU overhead when compared to software prefetching. However, software prefetching can mitigate certain constraints of hardware prefetching, leading to improvements in
317:
Compiler directed prefetching is widely used within loops with a large number of iterations. In this technique, the compiler predicts future cache misses and inserts a prefetch instruction based on the
54:
fetches data before it is needed. Because data access patterns show less regularity than instruction patterns, accurate data prefetching is generally more challenging than instruction prefetching.
721:
Accuracy is the fraction of total prefetches that were useful - i.e. the ratio of the number of memory addresses prefetched were actually referenced by the program to the total prefetches done.
325:
These prefetches are non-blocking memory operations, i.e. these memory accesses do not interfere with actual memory accesses. They do not change the state of the processor or cause page faults.
1753:. Fourth International Conference on Architectural Support for Programming Languages and Operating Systems. Santa Clara, California, USA: Association for Computing Machinery. pp. 40–52.
35:, so prefetching data and then accessing it from caches is usually many orders of magnitude faster than accessing it directly from main memory. Prefetching can be done with non-blocking
775:{\displaystyle {\text{Prefetch Accuracy}}={\frac {\text{Cache Misses eliminated by prefetching}}{({\text{Useless Cache Prefetches}})+({\text{Cache Misses eliminated by prefetching}})}}}
610:
793:
The qualitative definition of timeliness is how early a block is prefetched versus when it is actually referenced. An example to further explain timeliness is as follows:
561:
829:
1084:. 17th annual international symposium on Computer Architecture – ISCA 1990. New York, New York, USA: Association for Computing Machinery Press. pp. 364–373.
515:
275:
255:
235:
169:
149:
97:
is typically accomplished by having the compiler analyze the code and insert additional "prefetch" instructions in the program during compilation itself.
1595:
1815:
1061:
707:{\displaystyle {\text{Total Cache Misses}}=({\text{Cache misses eliminated by prefetching}})+({\text{Cache misses not eliminated by prefetching}})}
1080:
Jouppi, Norman P. (1990). "Improving direct-mapped cache performance by the addition of a small fully-associative cache and prefetch buffers".
196:
The ideal depth of the stream buffer is something that is subject to experimentation against various benchmarks and depends on the rest of the
60:
fetches instructions before they need to be executed. The first mainstream microprocessors to use some form of instruction prefetch were the
1212:
Grannaes, Marius; Jahre, Magnus; Natvig, Lasse (2011). "Storage
Efficient Hardware Prefetching using Delta-Correlating Prediction Tables".
523:
loop. For instance, if one iteration of the loop takes 7 cycles to execute, and the cache miss penalty is 49 cycles then there should be
1170:. 21st Annual International Symposium on Computer Architecture. Chicago, Illinois, USA: IEEE Computer Society Press. pp. 24–33.
1376:
Shevgoor, Manjunath; Koladiya, Sahil; Balasubramonian, Rajeev; Wilkerson, Chris; Pugsley, Seth H.; Chishti, Zeshan (December 2015).
999:. 1991 ACM/IEEE Conference on Supercomputing. Albuquerque, New Mexico, USA: Association for Computing Machinery. pp. 176–186.
332:
1725:
1682:
1571:
1483:
1438:
1352:
1309:
1258:
209:
This type of prefetching monitors the delta between the addresses of the memory accesses and looks for patterns within it.
1123:
Chen, Tien-Fu; Baer, Jean-Loup (1995-05-01). "Effective hardware-based data prefetching for high-performance processors".
189:. It is mainly used when contiguous locations are to be prefetched. For example, it is used when prefetching instructions.
1419:
Kim, Jinchun; Pugsley, Seth H.; Gratz, Paul V.; Reddy, A.L. Narasimha; Wilkerson, Chris; Chishti, Zeshan (October 2016).
1766:
1193:
1022:
979:
581:
1526:
1395:
1107:
1513:. 35th Annual IEEE/ACM International Symposium on Microarchitecture, 2002. (MICRO-35). Proceedings. pp. 62–73.
123:
are one of the most common hardware based prefetching techniques in use. This technique was originally proposed by
1620:
Kim, Jinchun; Teran, Elvira; Gratz, Paul V.; Jiménez, Daniel A.; Pugsley, Seth H.; Wilkerson, Chris (2017-05-12).
922:"Adaptive priority-based cache replacement and prediction-based cache prefetching in edge computing environment"
1596:"Making Temporal Prefetchers Practical: The MISB Prefetcher – Research Articles – Arm Research – Arm Community"
28:
654:{\displaystyle Coverage={\frac {\text{Cache Misses eliminated by Prefetching}}{\text{Total Cache Misses}}}}
517:
depends on two factors, the cache miss penalty and the time it takes to execute a single iteration of the
1293:
Feedback
Directed Prefetching: Improving the Performance and Bandwidth-Efficiency of Hardware Prefetchers
1470:. ISCA 1997. ISCA 1997. New York, New York, USA: Association for Computing Machinery. pp. 252–263.
27:
in which prefetched data is held until it is required. The source for the prefetch operation is usually
1834:
1712:. 2020 ACM/IEEE 47th Annual International Symposium on Computer Architecture (ISCA). pp. 118–131.
855:
328:
One main advantage of software prefetching is that it reduces the number of compulsory cache misses.
112:
Stream buffers were developed based on the concept of "one block lookahead (OBL) scheme" proposed by
36:
1382:. 2015 48th Annual IEEE/ACM International Symposium on Microarchitecture (MICRO). pp. 141–152.
1296:. 2007 IEEE 13th International Symposium on High Performance Computer Architecture. pp. 63–74.
1222:
1176:
1090:
1005:
1709:
Bouquet of
Instruction Pointers: Instruction Pointer Classifier-based Spatial Hardware Prefetching
1425:. 2016 49th Annual IEEE/ACM International Symposium on Microarchitecture (MICRO). pp. 1–12.
921:
920:
Li, Chunlin; Song, Mingyang; Du, Shaofeng; Wang, Xiaohai; Zhang, Min; Luo, Youlong (2020-09-01).
127:
in 1990 and many variations of this method have been developed since. The basic idea is that the
1217:
1171:
1085:
1000:
526:
1622:"Kill the Program Counter: Reconstructing Program Behavior in the Processor Cache Hierarchy"
1558:. MICRO-46. New York, New York, USA: Association for Computing Machinery. pp. 247–259.
1245:. ICS 2009. New York, New York, USA: Association for Computing Machinery. pp. 499–500.
800:
31:. Because of their design, accessing cache memories is typically much faster than accessing
1082:
Proceedings of the 17th annual international symposium on
Computer Architecture – ISCA 1990
1055:
840:
331:
The following example shows how a prefetch instruction would be added into code to improve
605:
Coverage is the fraction of total misses that are eliminated because of prefetching, i.e.
74:(four bytes). In recent years, all high-performance processors use prefetching techniques.
8:
1783:
1731:
1688:
1577:
1532:
1489:
1444:
1401:
1358:
1315:
1264:
1148:
941:
902:
500:
260:
240:
220:
154:
134:
1809:
1762:
1735:
1721:
1678:
1643:
1567:
1522:
1479:
1434:
1391:
1348:
1305:
1254:
1189:
1140:
1103:
1018:
975:
945:
894:
197:
1692:
1556:
Proceedings of the 46th Annual IEEE/ACM International
Symposium on Microarchitecture
1405:
1362:
1268:
1798:
1754:
1713:
1670:
1633:
1559:
1536:
1514:
1471:
1448:
1426:
1383:
1340:
1319:
1297:
1246:
1181:
1152:
1132:
1095:
1043:
1040:
Software methods for improvement of cache performance on supercomputer applications
1010:
933:
906:
886:
845:
1581:
1493:
1339:. 2017 IEEE International Conference on Computer Design (ICCD). pp. 373–376.
1717:
319:
175:
1781:
1707:
1667:
Proceedings of the 45th Annual
International Symposium on Computer Architecture
1518:
1508:
1468:
Proceedings of the 24th Annual
International Symposium on Computer Architecture
1430:
1420:
1377:
1334:
1291:
937:
113:
1828:
1647:
1301:
1185:
1144:
898:
179:
A typical stream buffer setup as originally proposed by Norman Jouppi in 1990
124:
1802:
1674:
1638:
1621:
1563:
1387:
1250:
1552:"Linearizing irregular memory accesses for improved correlated prefetching"
974:. Boca Raton, Florida: CRC Press, Taylor & Francis Group. p. 163.
1758:
1662:
1551:
1475:
1238:
1099:
1014:
890:
1344:
1283:
217:
In this pattern, consecutive memory accesses are made to blocks that are
120:
83:
Cache prefetching can be accomplished either by hardware or by software.
32:
1463:
1375:
576:
intervention, hardware prefetching requires special hardware mechanisms.
257:
and uses it to compute the memory address for prefetching. E.g.: If the
850:
128:
1136:
1047:
1287:
24:
997:
An
Effective On-chip Preloading Scheme to Reduce Data Access Penalty
47:
Cache prefetching can either fetch data or instructions into cache.
573:
68:
1507:
Collins, J.; Sair, S.; Calder, B.; Tullsen, D.M. (November 2002).
1243:
Proceedings of the 23rd
International Conference on Supercomputing
566:
151:
subsequent addresses) are fetched into a separate buffer of depth
1669:. ISCA '18. Los Angeles, California: IEEE Press. pp. 83–95.
1749:
Callahan, David; Kennedy, Ken; Porterfield, Allan (1991-01-01).
785:
without any prefetching, this is a non-zero number of misses.
1663:"Division of labor: a more effective approach to prefetching"
237:
addresses apart. In this case, the prefetcher calculates the
71:
61:
1336:
T2: A Highly
Accurate and Energy Efficient Stride Prefetcher
64:
1748:
1168:
Evaluating Stream Buffers As a Secondary Cache Replacement
78:
42:
16:
Computer processing technique to boost memory performance
1782:
Lee, Jaekyu and Kim, Hyesoon and Vuduc, Richard (2012),
1506:
597:
There are three main metrics to judge cache prefetching
831:
iterations prior to its usage to maintain timeliness.
1239:"Access map pattern matching for data cache prefetch"
1237:
Ishii, Yasuo; Inaba, Mary; Hiraki, Kei (2009-06-08).
803:
729:
670:
613:
529:
503:
263:
243:
223:
157:
137:
1619:
1418:
1706:Pakalapati, Samuel; Panda, Biswabandan (May 2020).
1333:Kondguli, Sushant; Huang, Michael (November 2017).
1281:
1211:
1784:"When Prefetching Works, When It Doesn't, and Why"
823:
774:
706:
653:
572:While software prefetching requires programmer or
555:
509:
307:
269:
249:
229:
163:
143:
101:
1165:
289:
1826:
1661:Kondguli, Sushant; Huang, Michael (2018-06-02).
1379:Efficiently prefetching complex address patterns
1236:
877:Smith, Alan Jay (1982-09-01). "Cache Memories".
312:
1705:
972:Fundamentals of parallel multicore architecture
592:
567:Comparison of hardware and software prefetching
1461:
277:is 4, the address to be prefetched would A+4.
1660:
1332:
1166:Palacharla, S.; Kessler, R. E. (1994-01-01).
995:Baer, Jean-Loup; Chen, Tien-Fu (1991-01-01).
1814:: CS1 maint: multiple names: authors list (
1060:: CS1 maint: multiple names: authors list (
926:Journal of Network and Computer Applications
919:
298:
280:
1462:Joseph, Doug; Grunwald, Dirk (1997-05-01).
1422:Path confidence based lookahead prefetching
1550:Jain, Akanksha; Lin, Calvin (2013-12-07).
1207:
1205:
1038:Kennedy, Porterfield, Allan (1989-01-01).
698:Cache misses not eliminated by prefetching
1637:
1221:
1175:
1089:
1004:
1214:Journal of Instruction-Level Parallelism
1122:
994:
322:and execution time of the instructions.
174:
1549:
1202:
1037:
969:
79:Hardware vs. software cache prefetching
1827:
1079:
763:Cache Misses eliminated by prefetching
741:Cache Misses eliminated by prefetching
684:Cache misses eliminated by prefetching
644:Cache Misses eliminated by Prefetching
204:
43:Data vs. instruction cache prefetching
1464:"Prefetching using Markov predictors"
876:
1075:
1073:
1071:
965:
963:
961:
959:
957:
955:
872:
870:
338:Consider a for loop as shown below:
13:
1510:Pointer cache assisted prefetching
212:
14:
1846:
1068:
952:
867:
106:
1282:Srinath, Santhosh; Mutlu, Onur;
1775:
1742:
1699:
1654:
1613:
1588:
1543:
1500:
1455:
1412:
1369:
1326:
1275:
1230:
308:Methods of software prefetching
102:Methods of hardware prefetching
1791:ACM Trans. Archit. Code Optim.
1159:
1125:IEEE Transactions on Computers
1116:
1031:
988:
913:
766:
758:
752:
744:
701:
693:
687:
679:
290:Irregular temporal prefetching
1:
861:
788:
313:Compiler directed prefetching
1718:10.1109/ISCA45697.2020.00021
593:Metrics of cache prefetching
295:performant implementations.
7:
1042:(Thesis). Rice University.
834:
716:
600:
497:Here, the prefetch stride,
10:
1851:
1519:10.1109/MICRO.2002.1176239
1431:10.1109/MICRO.2016.7783763
938:10.1016/j.jnca.2020.102715
95:Software based prefetching
88:Hardware based prefetching
37:cache control instructions
856:Cache control instruction
299:Collaborative prefetching
281:Irregular spatial strides
1302:10.1109/HPCA.2007.346185
1186:10.1109/ISCA.1994.288164
749:Useless Cache Prefetches
556:{\displaystyle k=49/7=7}
414:
340:
1803:10.1145/2133382.2133384
1675:10.1109/ISCA.2018.00018
1639:10.1145/3093336.3037701
1564:10.1145/2540708.2540730
1388:10.1145/2830772.2830793
1251:10.1145/1542275.1542349
58:Instruction prefetching
825:
824:{\displaystyle 12/3=4}
776:
708:
655:
557:
511:
271:
251:
231:
187:Sequential Prefetching
180:
165:
145:
1759:10.1145/106972.106979
1476:10.1145/264107.264207
1100:10.1145/325164.325162
1015:10.1145/125826.125932
970:Solihin, Yan (2016).
891:10.1145/356887.356892
826:
777:
709:
656:
558:
512:
272:
252:
232:
178:
166:
146:
1751:Software Prefetching
1345:10.1109/ICCD.2017.64
841:Prefetch input queue
801:
727:
668:
611:
527:
501:
261:
241:
221:
155:
135:
67:(six bytes) and the
1626:ACM SIGPLAN Notices
205:Strided prefetching
821:
772:
704:
673:Total Cache Misses
651:
647:Total Cache Misses
553:
507:
267:
247:
227:
181:
161:
141:
1835:Cache (computing)
1727:978-1-7281-4661-4
1684:978-1-5386-5984-7
1600:community.arm.com
1573:978-1-4503-2638-4
1485:978-0-89791-901-2
1440:978-1-5090-3508-3
1354:978-1-5386-2254-4
1311:978-1-4244-0804-7
1290:(February 2007).
1260:978-1-60558-498-0
1137:10.1109/12.381947
770:
764:
750:
742:
733:
732:Prefetch Accuracy
699:
685:
674:
649:
648:
645:
510:{\displaystyle k}
333:cache performance
270:{\displaystyle s}
250:{\displaystyle s}
230:{\displaystyle s}
198:microarchitecture
164:{\displaystyle k}
144:{\displaystyle k}
20:Cache prefetching
1842:
1820:
1819:
1813:
1805:
1788:
1779:
1773:
1772:
1746:
1740:
1739:
1703:
1697:
1696:
1658:
1652:
1651:
1641:
1617:
1611:
1610:
1608:
1607:
1592:
1586:
1585:
1547:
1541:
1540:
1504:
1498:
1497:
1459:
1453:
1452:
1416:
1410:
1409:
1373:
1367:
1366:
1330:
1324:
1323:
1279:
1273:
1272:
1234:
1228:
1227:
1225:
1209:
1200:
1199:
1179:
1163:
1157:
1156:
1120:
1114:
1113:
1093:
1077:
1066:
1065:
1059:
1051:
1035:
1029:
1028:
1008:
992:
986:
985:
967:
950:
949:
917:
911:
910:
879:ACM Comput. Surv
874:
846:Link prefetching
830:
828:
827:
822:
811:
781:
779:
778:
773:
771:
769:
765:
762:
751:
748:
740:
739:
734:
731:
713:
711:
710:
705:
700:
697:
686:
683:
675:
672:
660:
658:
657:
652:
650:
646:
643:
642:
562:
560:
559:
554:
543:
516:
514:
513:
508:
493:
490:
487:
484:
481:
478:
475:
472:
469:
466:
463:
460:
457:
454:
451:
448:
445:
442:
439:
436:
433:
430:
427:
424:
421:
418:
407:
404:
401:
398:
395:
392:
389:
386:
383:
380:
377:
374:
371:
368:
365:
362:
359:
356:
353:
350:
347:
344:
276:
274:
273:
268:
256:
254:
253:
248:
236:
234:
233:
228:
170:
168:
167:
162:
150:
148:
147:
142:
52:Data prefetching
1850:
1849:
1845:
1844:
1843:
1841:
1840:
1839:
1825:
1824:
1823:
1807:
1806:
1786:
1780:
1776:
1769:
1747:
1743:
1728:
1704:
1700:
1685:
1659:
1655:
1618:
1614:
1605:
1603:
1594:
1593:
1589:
1574:
1548:
1544:
1529:
1505:
1501:
1486:
1460:
1456:
1441:
1417:
1413:
1398:
1374:
1370:
1355:
1331:
1327:
1312:
1280:
1276:
1261:
1235:
1231:
1223:10.1.1.229.3483
1210:
1203:
1196:
1164:
1160:
1121:
1117:
1110:
1078:
1069:
1053:
1052:
1036:
1032:
1025:
993:
989:
982:
968:
953:
918:
914:
875:
868:
864:
837:
807:
802:
799:
798:
791:
761:
747:
743:
738:
730:
728:
725:
724:
719:
696:
682:
671:
669:
666:
665:
641:
612:
609:
608:
603:
595:
569:
539:
528:
525:
524:
502:
499:
498:
495:
494:
491:
488:
485:
482:
479:
476:
473:
470:
467:
464:
461:
458:
455:
452:
449:
446:
443:
440:
437:
434:
431:
428:
425:
422:
419:
416:
409:
408:
405:
402:
399:
396:
393:
390:
387:
384:
381:
378:
375:
372:
369:
366:
363:
360:
357:
354:
351:
348:
345:
342:
315:
310:
301:
292:
283:
262:
259:
258:
242:
239:
238:
222:
219:
218:
215:
213:Regular strides
207:
156:
153:
152:
136:
133:
132:
109:
104:
81:
45:
17:
12:
11:
5:
1848:
1838:
1837:
1822:
1821:
1774:
1768:978-0897913805
1767:
1741:
1726:
1698:
1683:
1653:
1632:(4): 737–749.
1612:
1602:. 24 June 2019
1587:
1572:
1542:
1527:
1499:
1484:
1454:
1439:
1411:
1396:
1368:
1353:
1325:
1310:
1274:
1259:
1229:
1201:
1195:978-0818655104
1194:
1177:10.1.1.92.3031
1158:
1131:(5): 609–623.
1115:
1108:
1091:10.1.1.37.6114
1067:
1030:
1024:978-0897914598
1023:
1006:10.1.1.642.703
987:
981:978-1482211184
980:
951:
912:
885:(3): 473–530.
865:
863:
860:
859:
858:
853:
848:
843:
836:
833:
820:
817:
814:
810:
806:
790:
787:
768:
760:
757:
754:
746:
737:
718:
715:
703:
695:
692:
689:
681:
678:
640:
637:
634:
631:
628:
625:
622:
619:
616:
602:
599:
594:
591:
590:
589:
585:
577:
568:
565:
552:
549:
546:
542:
538:
535:
532:
506:
415:
341:
314:
311:
309:
306:
300:
297:
291:
288:
282:
279:
266:
246:
226:
214:
211:
206:
203:
202:
201:
194:
190:
173:
172:
160:
140:
117:
114:Alan Jay Smith
108:
107:Stream buffers
105:
103:
100:
99:
98:
92:
80:
77:
76:
75:
55:
44:
41:
15:
9:
6:
4:
3:
2:
1847:
1836:
1833:
1832:
1830:
1817:
1811:
1804:
1800:
1796:
1792:
1785:
1778:
1770:
1764:
1760:
1756:
1752:
1745:
1737:
1733:
1729:
1723:
1719:
1715:
1711:
1710:
1702:
1694:
1690:
1686:
1680:
1676:
1672:
1668:
1664:
1657:
1649:
1645:
1640:
1635:
1631:
1627:
1623:
1616:
1601:
1597:
1591:
1583:
1579:
1575:
1569:
1565:
1561:
1557:
1553:
1546:
1538:
1534:
1530:
1528:0-7695-1859-1
1524:
1520:
1516:
1512:
1511:
1503:
1495:
1491:
1487:
1481:
1477:
1473:
1469:
1465:
1458:
1450:
1446:
1442:
1436:
1432:
1428:
1424:
1423:
1415:
1407:
1403:
1399:
1397:9781450340342
1393:
1389:
1385:
1381:
1380:
1372:
1364:
1360:
1356:
1350:
1346:
1342:
1338:
1337:
1329:
1321:
1317:
1313:
1307:
1303:
1299:
1295:
1294:
1289:
1288:Patt, Yale N.
1285:
1278:
1270:
1266:
1262:
1256:
1252:
1248:
1244:
1240:
1233:
1224:
1219:
1215:
1208:
1206:
1197:
1191:
1187:
1183:
1178:
1173:
1169:
1162:
1154:
1150:
1146:
1142:
1138:
1134:
1130:
1126:
1119:
1111:
1109:0-89791-366-3
1105:
1101:
1097:
1092:
1087:
1083:
1076:
1074:
1072:
1063:
1057:
1049:
1045:
1041:
1034:
1026:
1020:
1016:
1012:
1007:
1002:
998:
991:
983:
977:
973:
966:
964:
962:
960:
958:
956:
947:
943:
939:
935:
931:
927:
923:
916:
908:
904:
900:
896:
892:
888:
884:
880:
873:
871:
866:
857:
854:
852:
849:
847:
844:
842:
839:
838:
832:
818:
815:
812:
808:
804:
794:
786:
782:
755:
735:
722:
714:
690:
676:
662:
638:
635:
632:
629:
626:
623:
620:
617:
614:
606:
598:
586:
583:
578:
575:
571:
570:
564:
550:
547:
544:
540:
536:
533:
530:
522:
521:
504:
413:
339:
336:
334:
329:
326:
323:
321:
305:
296:
287:
278:
264:
244:
224:
210:
199:
195:
191:
188:
183:
182:
177:
158:
138:
131:address (and
130:
126:
125:Norman Jouppi
122:
118:
115:
111:
110:
96:
93:
89:
86:
85:
84:
73:
70:
66:
63:
59:
56:
53:
50:
49:
48:
40:
38:
34:
30:
26:
21:
1794:
1790:
1777:
1750:
1744:
1708:
1701:
1666:
1656:
1629:
1625:
1615:
1604:. Retrieved
1599:
1590:
1555:
1545:
1509:
1502:
1467:
1457:
1421:
1414:
1378:
1371:
1335:
1328:
1292:
1284:Kim, Hyesoon
1277:
1242:
1232:
1216:(13): 1–16.
1213:
1167:
1161:
1128:
1124:
1118:
1081:
1039:
1033:
996:
990:
971:
929:
925:
915:
882:
878:
795:
792:
783:
723:
720:
663:
607:
604:
596:
588:performance.
519:
518:
496:
410:
337:
330:
327:
324:
320:miss penalty
316:
302:
293:
284:
216:
208:
186:
94:
87:
82:
57:
51:
46:
25:cache memory
19:
18:
1056:cite thesis
33:main memory
29:main memory
1606:2022-03-16
1048:1911/19069
932:: 102715.
862:References
851:Prefetcher
789:Timeliness
129:cache miss
1736:218683672
1648:0362-1340
1218:CiteSeerX
1172:CiteSeerX
1145:0018-9340
1086:CiteSeerX
1001:CiteSeerX
946:219506427
899:0360-0300
200:involved.
1829:Category
1810:citation
1797:: 1–29,
1693:50777324
1406:15294463
1363:11055312
1269:37841036
835:See also
717:Accuracy
601:Coverage
574:compiler
462:prefetch
69:Motorola
1537:5608519
1449:1097472
1320:6909725
1153:1450745
907:6023466
664:where,
582:runtime
121:buffers
119:Stream
1765:
1734:
1724:
1691:
1681:
1646:
1582:196831
1580:
1570:
1535:
1525:
1494:434419
1492:
1482:
1447:
1437:
1404:
1394:
1361:
1351:
1318:
1308:
1267:
1257:
1220:
1192:
1174:
1151:
1143:
1106:
1088:
1021:
1003:
978:
944:
905:
897:
486:array1
474:array1
468:array1
412:below:
400:array1
388:array1
193:above.
91:cache.
1787:(PDF)
1732:S2CID
1689:S2CID
1578:S2CID
1533:S2CID
1490:S2CID
1445:S2CID
1402:S2CID
1359:S2CID
1316:S2CID
1265:S2CID
1149:S2CID
942:S2CID
903:S2CID
72:68000
62:Intel
1816:link
1763:ISBN
1722:ISBN
1679:ISBN
1644:ISSN
1568:ISBN
1523:ISBN
1480:ISBN
1435:ISBN
1392:ISBN
1349:ISBN
1306:ISBN
1255:ISBN
1190:ISBN
1141:ISSN
1104:ISBN
1062:link
1019:ISBN
976:ISBN
895:ISSN
444:1024
441:<
370:1024
367:<
65:8086
1799:doi
1755:doi
1714:doi
1671:doi
1634:doi
1560:doi
1515:doi
1472:doi
1427:doi
1384:doi
1341:doi
1298:doi
1247:doi
1182:doi
1133:doi
1096:doi
1044:hdl
1011:doi
934:doi
930:165
887:doi
520:for
423:int
417:for
349:int
343:for
1831::
1812:}}
1808:{{
1793:,
1789:,
1761:.
1730:.
1720:.
1687:.
1677:.
1665:.
1642:.
1630:52
1628:.
1624:.
1598:.
1576:.
1566:.
1554:.
1531:.
1521:.
1488:.
1478:.
1466:.
1443:.
1433:.
1400:.
1390:.
1357:.
1347:.
1314:.
1304:.
1286:;
1263:.
1253:.
1241:.
1204:^
1188:.
1180:.
1147:.
1139:.
1129:44
1127:.
1102:.
1094:.
1070:^
1058:}}
1054:{{
1017:.
1009:.
954:^
940:.
928:.
924:.
901:.
893:.
883:14
881:.
869:^
805:12
661:,
537:49
471:);
453:++
379:++
335:.
39:.
1818:)
1801::
1795:9
1771:.
1757::
1738:.
1716::
1695:.
1673::
1650:.
1636::
1609:.
1584:.
1562::
1539:.
1517::
1496:.
1474::
1451:.
1429::
1408:.
1386::
1365:.
1343::
1322:.
1300::
1271:.
1249::
1226:.
1198:.
1184::
1155:.
1135::
1112:.
1098::
1064:)
1050:.
1046::
1027:.
1013::
984:.
948:.
936::
909:.
889::
819:4
816:=
813:3
809:/
767:)
759:(
756:+
753:)
745:(
736:=
702:)
694:(
691:+
688:)
680:(
677:=
639:=
636:e
633:g
630:a
627:r
624:e
621:v
618:o
615:C
584:.
551:7
548:=
545:7
541:/
534:=
531:k
505:k
492:}
489:;
483:*
480:2
477:=
465:(
459:{
456:)
450:i
447:;
438:i
435:;
432:0
429:=
426:i
420:(
406:}
403:;
397:*
394:2
391:=
385:{
382:)
376:i
373:;
364:i
361:;
358:0
355:=
352:i
346:(
265:s
245:s
225:s
159:k
139:k
116:.
Text is available under the Creative Commons Attribution-ShareAlike License. Additional terms may apply.