wormhole branch prediction using multi dimensional
play

Wormhole branch prediction using multi-dimensional histories Jorge - PowerPoint PPT Presentation

Dec 07, 2023 •519 likes •1.12k views

Wormhole branch prediction using multi-dimensional histories Jorge Albericio, Joshua San Miguel , Natalie Enright Jerger, and Andreas Moshovos 2 Motivation foreach frame: foreach object: if distance( object , p ) < threshold: { /* do

  1. Wormhole branch prediction using multi-dimensional histories Jorge Albericio, Joshua San Miguel , 
 Natalie Enright Jerger, and Andreas Moshovos

  2. 2

  3. Motivation foreach frame: foreach object: if distance( object , p ) < threshold: { /* do something */ } 3

  4. Motivation foreach frame: foreach object: if distance( object , p ) < threshold: { /* do something */ } 3

  5. Motivation foreach frame: foreach object: if distance( object , p ) < threshold: { /* do something */ } 3

  6. Motivation foreach frame: foreach object: if distance( object , p ) < threshold: { /* do something */ } 3

  7. Multidimensional histories 1 0 1 0 0 1 0 1 0 0 1 0 1 0 0 foreach frame: foreach object: if distance( object , p ) < threshold: { /* do something */ } 4

  8. Multidimensional histories 0 1 2 3 4 0 1 2 3 4 0 1 2 3 t t t t t t t t t t t t t t c c c c c c c c c c c c c c e e e e e e e e e e e e e e j j j j j j j j j j j j j j b b b b b b b b b b b b b b O O O O O O O O O O O O O O 1 0 1 0 0 1 0 1 0 0 1 0 1 0 0 foreach frame: foreach object: if distance( object , p ) < threshold: { /* do something */ } 4

  9. Multidimensional histories 0 1 2 3 4 0 1 2 3 4 0 1 2 3 t t t t t t t t t t t t t t c c c c c c c c c c c c c c e e e e e e e e e e e e e e j j j j j j j j j j j j j j b b b b b b b b b b b b b b O O O O O O O O O O O O O O 1 0 1 0 0 1 0 1 0 0 1 0 1 0 0 Frame 0 Frame 1 Frame 2 foreach frame: foreach object: if distance( object , p ) < threshold: { /* do something */ } 4

  10. Multidimensional histories 0 1 2 3 4 0 1 2 3 4 0 1 2 3 t t t t t t t t t t t t t t c c c c c c c c c c c c c c e e e e e e e e e e e e e e j j j j j j j j j j j j j j b b b b b b b b b b b b b b O O O O O O O O O O O O O O 1 0 1 0 0 1 0 1 0 0 1 0 1 0 ? 0 Frame 0 Frame 1 Frame 2 foreach frame: foreach object: if distance( object , p ) < threshold: { /* do something */ } 4

  11. Multidimensional histories 0 1 2 3 4 0 1 2 3 4 0 1 2 3 t t t t t t t t t t t t t t c c c c c c c c c c c c c c e e e e e e e e e e e e e e j j j j j j j j j j j j j j b b b b b b b b b b b b b b O O O O O O O O O O O O O O 1 0 1 0 0 1 0 1 0 0 1 0 1 0 ? 0 Frame 0 Frame 1 Frame 2 foreach frame: foreach object: if distance( object , p ) < threshold: { /* do something */ } 4

  12. Multidimensional histories 0 1 2 3 4 0 1 2 3 4 0 1 2 3 t t t t t t t t t t t t t t c c c c c c c c c c c c c c e e e e e e e e e e e e e e j j j j j j j j j j j j j j b b b b b b b b b b b b b b O O O O O O O O O O O O O O 1 0 1 0 0 1 0 1 0 0 1 0 1 0 0 Frame 0 Frame 1 Frame 2 foreach frame: foreach object: if distance( object , p ) < threshold: { /* do something */ } 4

  13. Multidimensional histories 0 1 2 3 4 t t t t t c c c c c e e e e e j j j j j b b b b b O O O O O 1 0 1 0 0 Frames 1 0 1 0 0 1 0 1 0 0 foreach frame: foreach object: if distance( object , p ) < threshold: { /* do something */ } 4

  14. Multidimensional histories 0 1 2 3 4 t t t t t c c c c c e e e e e j j j j j b b b b b O O O O O 1 0 1 0 0 Frames 1 0 1 0 0 1 0 1 0 0 foreach frame: foreach object: if distance( object , p ) < threshold: { /* do something */ } 4

  15. Summary • Wormhole predictor : 
 ISL-TAGE Side-predictor on top of Wormhole ISL-TAGE pred. • Able to capture columns, diagonals, and other patterns predictions • 3.995 , 2.498 , and 2.014 MPKI in the 4KB, 32KB, and unlimited competition tracks 5

  16. Detecting diagonals for j in N: for k in N: if j != k: /*Calculation*/ 6

  17. Detecting diagonals for j in N: for k in N: if j != k: /*Calculation*/ 6

  18. Detecting diagonals for j in N: for k in N: if j != k: /*Calculation*/ 1 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 1 0 0 0 0 6

  19. Detecting diagonals for j in N: for k in N: if j != k: /*Calculation*/ 1 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 1 0 0 0 0 j=0 j=1 j=2 6

  20. Detecting diagonals 1 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 1 0 0 0 0 j=0 j=1 j=2 for j in N: //Loop 1 for k in N: //Loop 2 if j != k: /*Calculation*/ 7

  21. Detecting diagonals Inner loop (k) Iterations 1 0 0 0 0 0 0 Outer loop (j) 0 1 0 0 0 0 0 Iterations 0 0 1 0 0 0 0 for j in N: //Loop 1 for k in N: //Loop 2 if j != k: /*Calculation*/ 7

  22. Multidimensional histories 1 0 0 0 0 0 0 Outer loop (j) 0 1 0 0 0 0 0 Iterations Inner loop (k) for j in N: Iterations for k in N: if j != k: /*Calculation*/ 8

  23. Multidimensional histories 1 0 0 0 0 0 0 Outer loop (j) ? 0 1 0 0 0 0 0 Iterations Inner loop (k) for j in N: Iterations for k in N: if j != k: /*Calculation*/ 8

  24. Multidimensional histories Saturating counters 1 0 0 0 0 0 0 Outer loop (j) ? 0 1 0 0 0 0 0 Iterations Inner loop (k) for j in N: Iterations for k in N: if j != k: /*Calculation*/ 8

  25. Multidimensional histories Saturating counters 1 0 1 0 0 0 0 0 0 0 Outer loop (j) ? 0 0 1 0 0 0 0 0 Iterations Inner loop (k) for j in N: Iterations for k in N: if j != k: /*Calculation*/ 8

  26. Multidimensional histories Saturating counters 0100 x 1 0 1 0 0 0 0 0 0 0 Outer loop (j) ? 0 0 1 0 0 0 0 0 Iterations Inner loop (k) for j in N: Iterations for k in N: if j != k: /*Calculation*/ 8

  27. Multidimensional histories Saturating counters 0100 x 1 0 1 0 0 0 0 0 0 0 Outer loop (j) 1 ? 0 0 1 0 0 0 0 0 Iterations Inner loop (k) for j in N: Iterations for k in N: if j != k: /*Calculation*/ 8

  28. Multidimensional histories Saturating counters 0100 x 1 1 0 1 0 0 0 0 0 0 0 Outer loop (j) 1 ? 0 0 1 0 0 0 0 0 Iterations Inner loop (k) for j in N: Iterations for k in N: if j != k: /*Calculation*/ 8

  29. Multidimensional histories Saturating counters x 1 0 1 0 0 0 0 0 0 Outer loop (j) 0 0 0 0 1 0 0 0 Iterations 0 0 0 1 0 0 0 Inner loop (k) for j in N: Iterations for k in N: if j != k: /*Calculation*/ 9

  30. Multidimensional histories Saturating counters x 1 0 1 0 0 0 0 0 0 Outer loop (j) 0 0 0 0 1 0 0 0 Iterations 0 ? 0 0 1 0 0 0 Inner loop (k) for j in N: Iterations for k in N: if j != k: /*Calculation*/ 9

  31. Multidimensional histories Saturating counters x 1 0 1 0 0 0 0 0 0 Outer loop (j) 1 0 0 0 0 0 0 1 0 0 0 Iterations 0 0 ? 0 0 1 0 0 0 Inner loop (k) for j in N: Iterations for k in N: if j != k: /*Calculation*/ 9

  32. Multidimensional histories Saturating counters 0100 x 1 0 1 0 0 0 0 0 0 Outer loop (j) 1 0 0 0 0 0 0 1 0 0 0 Iterations 0 0 ? 0 0 1 0 0 0 Inner loop (k) for j in N: Iterations for k in N: if j != k: /*Calculation*/ 9

  33. Multidimensional histories Saturating counters 0100 x 1 Prediction: 1 0 1 0 0 0 0 0 0 Outer loop (j) 1 0 0 0 0 0 0 1 0 0 0 Iterations 0 0 ? 0 0 1 0 0 0 Inner loop (k) for j in N: Iterations for k in N: if j != k: /*Calculation*/ 9

  34. Multidimensional histories Saturating counters 0100 x 1 Prediction: 1 0 1 0 0 0 0 0 0 Outer loop (j) 1 0 0 0 0 0 0 1 0 0 0 Iterations 1 0 0 ? 0 0 1 0 0 0 Inner loop (k) for j in N: Iterations for k in N: if j != k: /*Calculation*/ 9

  35. Multidimensional histories Saturating counters 0100 x 1 Prediction: 1 0 1 0 0 0 0 0 0 Outer loop (j) 1 0 0 0 0 0 0 1 0 0 0 Iterations 1 0 0 ? 0 0 1 0 0 0 Inner loop (k) for j in N: Iterations for k in N: if j != k: /*Calculation*/ 9

  36. Wormhole predictor and ISL-TAGE • On top of ISL-TAGE 
 ISL-TAGE (Best performer from last CBP) Loop Wormhole tagged pred. pred. comp. bimodal • TAGE: bimodal + tagged Stat. pred. corr. • Statistical corrector predictions • Loop predictor 10

  37. Wormhole predictor and ISL-TAGE • On top of ISL-TAGE 
 ISL-TAGE (Best performer from last CBP) Loop Wormhole tagged pred. pred. comp. bimodal • TAGE: bimodal + tagged Stat. pred. corr. • Statistical corrector predictions • Loop predictor Detect difficult branches 10

  38. Wormhole predictor and ISL-TAGE • On top of ISL-TAGE 
 ISL-TAGE (Best performer from last CBP) Loop Wormhole tagged pred. pred. comp. bimodal • TAGE: bimodal + tagged Stat. pred. corr. • Statistical corrector predictions • Loop predictor Detect difficult branches Loop dimensionality 10

Recommend

CS 104 Computer Organization and Design Branch Prediction CS104:Branch Prediction 1 Branch

CS 104 Computer Organization and Design Branch Prediction CS104:Branch Prediction 1 Branch

CS 104 Computer Organization and Design Branch Prediction CS104:Branch Prediction 1 Branch Prediction Quick Overview App App App Now that we know about SRAMs System software Mem CPU I/O CS104: Branch Prediction 2 Branch

677 views • 21 slides
1 Predictor for a Single Branch Branch History Table of 1-bit Predictor BHT also Called Branch

1 Predictor for a Single Branch Branch History Table of 1-bit Predictor BHT also Called Branch

Reducing Branch Penalty Branch penalty in dynamically scheduled processors: wasted cycles due to pipeline flushing on mis- predicted branches Lecture 9: Branch Prediction I Reduce branch penalty: 1. Prediction analysis, 1-bit predictor,

378 views • 3 slides
Branch Prediction Branch Prediction vs vs Execution Time Execution Time Prediction

Branch Prediction Branch Prediction vs vs Execution Time Execution Time Prediction

Branch Prediction Branch Prediction vs vs Execution Time Execution Time Prediction Prediction Jakob Engblom, PhD Jakob Engblom, PhD Uppsala Unive University rsity &amp; Virtutech Inc. &amp; Virtutech Inc. Uppsala virtutech virtu

377 views • 26 slides
Visualizing Multi-dimensional Data S E T H H O R R I G A N C O M P U T E R V I S U A L I Z A T

Visualizing Multi-dimensional Data S E T H H O R R I G A N C O M P U T E R V I S U A L I Z A T

Visualizing Multi-dimensional Data S E T H H O R R I G A N C O M P U T E R V I S U A L I Z A T I O N F A L L 2 0 0 8 Motivation Multi-dimensional datasets are common Digital cameras Wall-street stocks Motor vehicles

571 views • 19 slides
1 Branch History Table of 1-bit Predictor 1-bit BHT Weakness BHT also Called Branch Example: in

1 Branch History Table of 1-bit Predictor 1-bit BHT Weakness BHT also Called Branch Example: in

Reducing Branch Penalty Branch penalty in dynamically scheduled processors: wasted cycles due to pipeline flushing on mis- predicted branches Lecture 9: Branch Prediction Reduce branch penalty: 1. Basic idea, saturating counter, BHT, Predict

248 views • 3 slides
Is branch prediction important for performance? Daniel J. Bernstein Spectre paper: Modern

Is branch prediction important for performance? Daniel J. Bernstein Spectre paper: Modern

1 Is branch prediction important for performance? Daniel J. Bernstein Spectre paper: Modern processors use branch prediction and speculative execution to maximize performance. Wikipedia: Branch predictors play a critical role in

1.61k views • 43 slides
Multi-Dimensional LSTM Networks for Video Prediction Wonmin Byeon NVIDIA Research March 29, 2018

Multi-Dimensional LSTM Networks for Video Prediction Wonmin Byeon NVIDIA Research March 29, 2018

Multi-Dimensional LSTM Networks for Video Prediction Multi-Dimensional LSTM Networks for Video Prediction Wonmin Byeon NVIDIA Research March 29, 2018 Wonmin Byeon | NVIDIA Research | March 29, 2018 1 / 44 Multi-Dimensional LSTM Networks for

739 views • 71 slides
Securing Neighbor Discovery the wormhole attack centralized and decentralized wormhole

Securing Neighbor Discovery the wormhole attack centralized and decentralized wormhole

Securing Neighbor Discovery the wormhole attack centralized and decentralized wormhole detection mechanisms Security and Cooperation in Wireless Networks Georg-August University Gttingen Introduction many wireless networking

639 views • 32 slides
Is branch prediction The article cited by Wikipedia important for performance? says: Branch

Is branch prediction The article cited by Wikipedia important for performance? says: Branch

1 2 Is branch prediction The article cited by Wikipedia important for performance? says: Branch predictor (BP) is an essential component in modern Daniel J. Bernstein processors since high BP accuracy can improve performance and Spectre

810 views • 57 slides
Today Arrays One-dimensional Machine-Level Programming IV: Data Multi-dimensional

Today Arrays One-dimensional Machine-Level Programming IV: Data Multi-dimensional

Today Arrays One-dimensional Machine-Level Programming IV: Data Multi-dimensional (nested) Multi-level Structures CSci 2021: Machine Architecture and Organization Allocation March 2nd-4th, 2020 Access Your instructor:

224 views • 6 slides
Data Structures in Memory! Arrays One dimensional Multi dimensional (nested) M

Data Structures in Memory! Arrays One dimensional Multi dimensional (nested) M

University of Washington Data Structures in Memory! Arrays One dimensional Multi dimensional (nested) M l i di i l ( d) Multi level Structs Alignment Unions 23 April 2012 Data Structures 1 University of

631 views • 33 slides
Bottom-up Estimation and Top-down Prediction for Multi-level Models: Solar Energy Prediction

Bottom-up Estimation and Top-down Prediction for Multi-level Models: Solar Energy Prediction

Bottom-up Estimation and Top-down Prediction for Multi-level Models: Solar Energy Prediction Combining Information from Multiple Sources Jae-Kwang Kim Department of Statistics, Iowa State University Ross-Royall Symposium: Johns Hopkins

857 views • 37 slides
Multi-Dimensional Gas Flows Tai-Ping Liu Academia Sinica, Taiwan Stanford University Final

Multi-Dimensional Gas Flows Tai-Ping Liu Academia Sinica, Taiwan Stanford University Final

Multi-Dimensional Gas Flows Tai-Ping Liu Academia Sinica, Taiwan Stanford University Final Conference HCDTE, July 18-22, SISSA Tai-Ping Liu Multi-Dimensional Gas Flows Multi-Dimensional Gas Flows Euler equations in gas dynamics t + x

502 views • 25 slides
Branch Prediction Philipp Koehn 11 October 2019 Philipp Koehn Computer Systems Fundamentals:

Branch Prediction Philipp Koehn 11 October 2019 Philipp Koehn Computer Systems Fundamentals:

Branch Prediction Philipp Koehn 11 October 2019 Philipp Koehn Computer Systems Fundamentals: Branch Prediction 11 October 2019 Control Hazard 1 Also called branch hazard Selection of next instruction depends on outcome of previous

459 views • 26 slides
ECE 2162 Branch Prediction Control Dependencies Branches are very frequent Approx. 20%

ECE 2162 Branch Prediction Control Dependencies Branches are very frequent Approx. 20%

ECE 2162 Branch Prediction Control Dependencies Branches are very frequent Approx. 20% of all instructions Can not wait until we know where it goes Long pipelines Branch outcome known after B cycles No scheduling past the

826 views • 34 slides
Multi-dimensional Dependency Grammar as Graph Description Ralph Debusmann and Gert Smolka

Multi-dimensional Dependency Grammar as Graph Description Ralph Debusmann and Gert Smolka

Multi-dimensional Dependency Grammar as Graph Description Multi-dimensional Dependency Grammar as Graph Description Ralph Debusmann and Gert Smolka Programming Systems Lab, Saarbrcken, Germany FLAIRS-19, May 11th, 2006 Multi-dimensional

698 views • 48 slides
Hawking Radiation around the Wormhole Sung-Won Kim (Ewha Womans University) In collaboration

Hawking Radiation around the Wormhole Sung-Won Kim (Ewha Womans University) In collaboration

Hawking Radiation around the Wormhole Sung-Won Kim (Ewha Womans University) In collaboration with S. Hayward APS2012 - Yukawa 1 Contents Motivation Surface Gravity Wormhole Temperature Hamilton-Jacobi Equation Summary

510 views • 12 slides
F From STEM EELS to multi dimensional and STEM EELS t lti di i l d multi signal

F From STEM EELS to multi dimensional and STEM EELS t lti di i l d multi signal

Taipei, November 22, 2011 F From STEM EELS to multi dimensional and STEM EELS t lti di i l d multi signal electron microscopy Christian Colliex Laboratoire de Physique des Solides, Bldg 510 L b i d h i d S lid ld 10

842 views • 59 slides
branch prediction 1 last time what happens with TLB in access patterns overlapping TLB and

branch prediction 1 last time what happens with TLB in access patterns overlapping TLB and

branch prediction 1 last time what happens with TLB in access patterns overlapping TLB and cache index lookup overview of caches and page table lookups and TLB generally 3 an OOO pipeline combined with register-ready info to issue

1.36k views • 100 slides
Branch Prediction Tackles problem of stalls from control

Branch Prediction Tackles problem of stalls from control

Branch Prediction Tackles problem of stalls from control dependencies Vital for mul5ple issue architectures Branches arrive up to N 5mes faster when

553 views • 21 slides
Multi- -dimensional Data and dimensional Data and Spatial Range Spatial Range Multi Query in

Multi- -dimensional Data and dimensional Data and Spatial Range Spatial Range Multi Query in

Multi- -dimensional Data and dimensional Data and Spatial Range Spatial Range Multi Query in Sensor Networks Query in Sensor Networks Jie Gao Computer Science Department Stony Brook University 1 Papers Papers [Li03a] X. Li, Y. J.

615 views • 40 slides
Genomic Prediction and Selection for Multi-Environments with Big Data using the BGLR statistical

Genomic Prediction and Selection for Multi-Environments with Big Data using the BGLR statistical

Genomic Prediction and Selection for Multi-Environments with Big Data using the BGLR statistical package Biometrics and Statistics Unit, Global Maize and Wheat programs June, 2015. CIMMyT Genomic Prediction and Selection for Multi-Environments

471 views • 26 slides
Signal Rate Inference for Dimensional Faust Multi-Dimensional Faust Y. Orlarey P. Jouvelot

Signal Rate Inference for Dimensional Faust Multi-Dimensional Faust Y. Orlarey P. Jouvelot

Signal Rate Inference for Multi- Signal Rate Inference for Dimensional Faust Multi-Dimensional Faust Y. Orlarey P. Jouvelot Signals Yann Orlarey 1 Pierre Jouvelot 2 Types/rates Inference 1 Grame, France Correctness Algorithm 2 MINES

525 views • 15 slides
Wormhole: A Fast Ordered Index for In-memory Data Management(II) Main Paper : Wormhole: A Fast

Wormhole: A Fast Ordered Index for In-memory Data Management(II) Main Paper : Wormhole: A Fast

Wormhole: A Fast Ordered Index for In-memory Data Management(II) Main Paper : Wormhole: A Fast Ordered Index for In-memory Data Management Authors: Wu, Xingbo, Fan Ni, and Song Jiang. Published in: In Proceedings of the Fourteenth EuroSys

774 views • 13 slides

More recommend