Veerle Desmet Lieven Eeckhout Koen De Bosschere

1
Veerle Desmet Lieven Eeckhout Koen De Bosschere

• Using Decision Trees to Improve Program-Based and
  Profile-Based Static Branch Prediction

2
Control flow
High-level program code

• if (input gt 0)
• for (i0 ilt10 i)
• / a loop ) /
• else
• printf(input lt 0)

compiler architecture
3
Branch Prediction

• Static
• fixed prediction per static branch
• predicts most frequently direction
• backward branches typically taken
• e.g 1 misprediction for 10 predictions
• Applications
• compiler optimizations
• embedded systems
• for guiding dynamic branch prediction

taken
not-taken
4
Outline

• Introduction
• Decision Trees
• Extended feature set
• Heuristics
• Ordering
• Two new heuristics
• Results
• Conclusion

5
Decision trees

• C4.5
• Set examples
• Choose best split (information gainratio)
• Divide examples in subtrees
• Repeat for subtrees
• Classification model

features
6
Features

• Literature Calder et al.
• branch direction, branch type, successor basic
  block is a loopheader, postdominance
  relations,...
• New candidate features
• number of instructions in successor basic block
• dependency distance
• number of incoming edges

2.8
7
Fair evaluation
branch direction
forward
backward
C4.5
branch type
taken
beq bgt
bne
taken
not-taken
Programs
Prediction model
SPECint2000 SPECint95
Evaluation
8
Outline

• Introduction
• Decision Trees
• Extended feature set
• Heuristics
• Automatic Ordering
• Two new heuristics
• Results
• Conclusion

9
Loops
Heuristic

• Loops
• for, while,...
• 11 static
• 35 dynamic
• Loops are repeated
• 81 correct
• 88 static upper limit

10
Non-loops
Heuristics

• Opcode
• Negative numbers denote error values
• Pointer
• Pointer ? NULL
• Pointers differ
• Loopheader
• Loops are entered
• ...

11
Heuristic ordering
Priority

• Loop ? pointer ? call ? opcode ? return ? store ?
  loopheader ? guard random
• Ball Larus
• Optimal ordering out of 8! possibilities
• Dependent on program set
• Automated way

12
Optimal tree
loop
opcode
heuristic
call
heuristic
return
heuristic
loopheader
heuristic
store
not-taken
pointer
taken
not-taken
taken
SPECint2000 SPECint95
13
New heuristic 1
Postdominating heuristic

• If one successor is postdominator
• Predict the non-postdominating successor
• if-block without else-block if-block will be
  executed

if (input 0)
if-block
postdominating block
NEW
14
New heuristic 2
NEW
Dependency distance heuristic

• number of instructions between branch and its
  register defining instruction

... cmplt r1, 10, r2 add r3, r4, r5 bne r2
ld r3, (r1) beq r2
distance 2
undefined distance
15
Final ordering
loop
opcode
heuristic
call
heuristic
return
heuristic
loopheader
heuristic
store
not-taken
postdom
taken
no
yes
not-taken
distance
lt3
undefined
3
taken
not-taken
SPECint2000 SPECint95
16
IPM

• Instructions Per Mispredicted branch
• how many instructions, including correctly
  predicted branches, one passes on average before
  encountering a mispredicted branch Fisher
  Freudenberger

HIGHER-IS-BETTER
17
Results
Ball Larus order 31.3 IPM
Higher-is-better
Decision Tree order 32.1 IPM
postdominating heuristic 34.7 IPM
dependency distance heuristic 37.1 IPM
100
90
80
70
60
IPM
50
40
30
20
10
0
18.5
li
go
vpr
gcc
mcf
perl
gap
gzip
twolf
bzip2
ijpeg
crafty
vortex
parser
average
m88ksim
compress
18
Conclusion

• This presentation
• Set extra features
• provides 2.8 gain in IPM
• Decision trees for
• heuristic ordering
• automatic extracting new heuristics
• 18.5 gain in IPM
• In paper also
• Coverage for heuristics
• Profile-based branch prediction
• up to 11 gain in IPM

19

• Paper
• proceedings pp. 336352
• Presentation
• www.elis.ugent.be/vdesmet

VEERLE DESMET Ghent University
20
Veerle Desmet Lieven Eeckhout Koen De Bosschere

• Using Decision Trees to Improve Program-Based and
  Profile-Based Static Branch Prediction