Compiling Imperative and Object-Oriented Languages - Garbage Collection

1. Garbage Collection Guido Wachsmuth Delft Course IN4303 University of Technology Compiler Construction Challenge the future

2. Recap register allocation How can we assign registers to local variables and temporaries? • perform liveness analysis • build interference graph • colour interference graph What to do if the graph is not colourable? • keep local variables in memory How to handle move instructions efficiently? • coalesce nodes safely Garbage Collection 2

3. Overview today’s lecture garbage collection Garbage Collection 3

4. Overview today’s lecture garbage collection mark & sweep • mark reachable records • sweep unmarked records Garbage Collection 3

5. Overview today’s lecture garbage collection mark & sweep • mark reachable records • sweep unmarked records reference counting • deallocate records with count 0 Garbage Collection 3

6. Overview today’s lecture garbage collection mark & sweep • mark reachable records • sweep unmarked records reference counting • deallocate records with count 0 copy collection • copy reachable records • consider generations of records Garbage Collection 3

7. I garbage collection Garbage Collection 4

8. the heap Garbage Collection 5

9. Recap: Java Virtual Machine the heap method area stack pc: 00 constant pool optop: 00 local variables 00 12 ldc 00 4303 4303 00 00 002A 002A 01 00 00 01 0000 0004 01 01 02 19 aload 02 02 02 03 00 00 03 03 03 04 12 ldc 04 04 04 05 01 01 05 05 05 06 2E iaload 06 06 06 heap 4303 4303 "Compilers" 002A 002A [20,01,40,02,42] Garbage Collection 6

10. Recap: Java Virtual Machine the heap method area stack pc: 00 02 constant pool optop: 00 01 local variables 00 12 ldc 00 4303 4303 00 4303 4303 00 002A 002A 01 00 00 01 0000 0004 01 01 02 19 aload 02 02 02 03 00 00 03 03 03 04 12 ldc 04 04 04 05 01 01 05 05 05 06 2E iaload 06 06 06 heap 4303 4303 "Compilers" 002A 002A [20,01,40,02,42] Garbage Collection 6

11. Recap: Java Virtual Machine the heap method area stack pc: 00 04 02 constant pool optop: 00 02 01 local variables 00 12 ldc 00 4303 4303 00 4303 4303 00 002A 002A 01 00 00 01 0000 0004 01 002A 002A 01 02 19 aload 02 02 02 03 00 00 03 03 03 04 12 ldc 04 04 04 05 01 01 05 05 05 06 2E iaload 06 06 06 heap 4303 4303 "Compilers" 002A 002A [20,01,40,02,42] Garbage Collection 6

12. Recap: Java Virtual Machine the heap method area stack pc: 00 06 04 02 constant pool optop: 00 03 02 01 local variables 00 12 ldc 00 4303 4303 00 4303 4303 00 002A 002A 01 00 00 01 0000 0004 01 002A 002A 01 02 19 aload 02 02 0000 0004 02 03 00 00 03 03 03 04 12 ldc 04 04 04 05 01 01 05 05 05 06 2E iaload 06 06 06 heap 4303 4303 "Compilers" 002A 002A [20,01,40,02,42] Garbage Collection 6

13. Recap: Java Virtual Machine the heap method area stack pc: 00 07 06 04 02 constant pool optop: 00 02 01 local variables 00 12 ldc 00 4303 4303 00 4303 4303 00 002A 002A 01 00 00 01 0000 0004 01 002A 002A 0000 0042 01 02 19 aload 02 02 0000 0004 02 03 00 00 03 03 03 04 12 ldc 04 04 04 05 01 01 05 05 05 06 2E iaload 06 06 06 heap 4303 4303 "Compilers" 002A 002A [20,01,40,02,42] Garbage Collection 6

14. Heap example 12 ● ● 15 ● ● ● 7 37 ● ● 59 ● ● ● 9 20 ● ● ● 37 ● p q r Garbage Collection 7

15. II mark & sweep Garbage Collection 8

16. Mark & sweep idea mark • mark reachable records • start at variables (roots) • follow references sweep • marked records: unmark • unmarked records: deallocate • linked list of free records Garbage Collection 9

17. Marking example 12 ● ● 15 ● ● ● 7 37 ● ● 59 ● ● ● 9 20 ● ● ● 37 ● p q r Garbage Collection 10

18. Marking example 12 ● ● 15 ● ● ● 7 37 ● ● 59 ● ● ● 9 20 ● ● ● 37 ● roots p q r Garbage Collection 10

48. Sweeping example 12 ● ● 15 ● ● ● 7 37 ● ● 59 ● ● ● 9 20 ● ● ● 37 ● p q r Garbage Collection 11

49. Sweeping example 12 ● ● 15 ● ● ● 7 37 ● ● 59 ● ● ● 9 20 ● ● ● ● 37 ● freelist f p q r Garbage Collection 11

60. Marking algorithms function DFS(x) function DFS(x) if pointer(x) & !marked[x] if pointer(x) & !marked[x] marked[x] = true marked[x] = true t = 1 ; stack[t] = x foreach f in fields(x) while t > 0 DFS(f) x = stack[t] ; t = t - 1 foreach f in fields(x) if pointer(f) & !marked[f] marked[f] = true t = t + 1 ; stack[t] = f Garbage Collection 12

61. Recap: Marking pointer reversals 12 ● ● 15 ● ● ● 7 37 ● ● 59 ● ● ● 9 20 ● ● ● 37 ● p q r Garbage Collection 13

91. Mark & sweep notes sweeping • independent of marking algorithm • several freelists (per record size) • split free records for allocation fragmentation • external: many free records of small size • internal: too-large record with unused memory inside Garbage Collection 14

92. III reference counts Garbage Collection 15

93. Reference counts idea counts • how many pointers point to each record? • store with each record counting • extra instructions deallocate • put on freelist • recursive deallocation on allocation Garbage Collection 16

94. Reference counts example 12 ● ● 15 ● ● ● 7 37 ● ● 59 ● ● ● 9 20 ● ● ● 37 ● p q r Garbage Collection 17

95. Reference counts example 12 1 ● ● 15 1 ● ● ●17 37 2 ● ● 59 1 ● ● ●19 20 1 ● ● ● 37 ● p q r Garbage Collection 17

96. Reference counts example 12 1 ● ● 15 1 ● ● ●17 37 2 ● ● 59 1 ● ● ●19 20 1 ● ● ● 37 ● p q r Garbage Collection 17

97. Reference counts notes cycles • memory leaks • break cycles explicitly • occasional mark & sweep collection expensive • fetch, decrease, store old reference counter • possible deallocation • fetch, increase, store new reference counter Garbage Collection 18

98. coffee break Garbage Collection 19

99. IV copy collections Garbage Collection 20

100. Copy collections idea spaces • fromspace & tospace • switch roles after copy copy • traverse reachability graph • copy from fromspace to tospace • fromspace unreachable, free memory • tospace compact, no fragmentation Garbage Collection 21

101. Copy collection example 12 ● ● 15 ● ● ● 7 37 ● ● 59 ● ● ● 9 20 ● ● ● 37 ● p q r Garbage Collection 22

102. Copy collection example 12 ● ● 15 ● ● ● 7 37 ● ● 59 ● ● ● 9 20 ● ● ● 37 ● p q r Garbage Collection 22

103. Copy collection example 12 ● ● 15 ● ● ● 7 37 ● ● 59 ● ● ● 9 20 ● ● ● 37 ● p q r 15 ● ● Garbage Collection 22

104. Copy collection example 12 ● ● 15 ● ● ● ● 7 37 ● ● 59 ● ● ● 9 20 ● ● ● 37 ● p q r 15 ● ● Garbage Collection 22

107. Copy collection example 12 ● ● 15 ● ● ● ● 7 37 ● ● 59 ● ● ● 9 20 ● ● ● 37 ● p q r 15 ● ● 37 ● ● Garbage Collection 22

108. Copy collection example 12 ● ● 15 ● ● ● ● 7 37 ● ● ● 59 ● ● ● 9 20 ● ● ● 37 ● p q r 15 ● ● 37 ● ● Garbage Collection 22

111. Copy collection 12 ● ● 15 ● ● ● ● 7 37 ● ● ● 59 ● ● ● 9 20 ● ● ● 37 ● p q r 15 ● ● 37 ● ● Garbage Collection 23

114. Copy collection 12 ● ● 15 ● ● ● ● 7 37 ● ● ● 59 ● ● ● 9 20 ● ● ● 37 ● p q r 15 ● ● 37 ● ● 12 ● ● Garbage Collection 23

115. Copy collection 12 ● ● ● 15 ● ● ● ● 7 37 ● ● ● 59 ● ● ● 9 20 ● ● ● 37 ● p q r 15 ● ● 37 ● ● 12 ● ● Garbage Collection 23

121. Copy collection 12 ● ● ● 15 ● ● ● ● 7 37 ● ● ● 59 ● ● ● 9 20 ● ● ● 37 ● p q r 15 ● ● 37 ● ● 12 ● ● 20 ● ● Garbage Collection 23

122. Copy collection 12 ● ● ● 15 ● ● ● ● 7 37 ● ● ● 59 ● ● ● 9 20 ● ● ● ● 37 ● p q r 15 ● ● 37 ● ● 12 ● ● 20 ● ● Garbage Collection 23

125. Copy collection 12 ● ● ● 15 ● ● ● ● 7 37 ● ● ● 59 ● ● ● 9 20 ● ● ● ● 37 ● p q r 15 ● ● 37 ● ● 12 ● ● 20 ● ● 59 ● ● Garbage Collection 23

126. Copy collection 12 ● ● ● 15 ● ● ● ● 7 37 ● ● ● 59 ● ● ● ● 9 20 ● ● ● ● 37 ● p q r 15 ● ● 37 ● ● 12 ● ● 20 ● ● 59 ● ● Garbage Collection 23

138. Copy collection ● 37 ● p q r 15 ● ● 37 ● ● 12 ● ● 20 ● ● 59 ● ● Garbage Collection 23

139. Copy collections algorithms function Forward(p) function BFS() if fromspace(p) next = scan = start(tospace) if tospace(fields(p)[1]) return fields(p)[1] foreach r in root() r = Forward(r) else while scan < next foreach f in fields(p) next.f = f foreach f in fields(scan) f = Forward(f) fields(p)[1] = next next = next + size(p) scan = scan + size(scan) return fields(p)[1] else return p Garbage Collection 24

140. Copy collections locality adjacent records • likely to be unrelated pointers to records in records • likely to be accessed • likely to be far apart solution • depth-first copy: slow pointer reversals • hybrid copy algorithm Garbage Collection 25

141. Copy collections generational collection generations • young data: likely to die soon • old data: likely to survive for more collections • divide heap, collect younger generations more frequently collection • roots: variables & pointers from older to younger generations • preserve pointers to old generations • promote objects to older generations Garbage Collection 26

142. V summary Garbage Collection 27

143. Summary lessons learned How can we collect unreachable records on the heap? • reference counts • mark reachable records, sweep unreachable records • copy reachable records How can we reduce heap space needed for garbage collection? • pointer-reversal • breadth-first search • hybrid algorithms Garbage Collection 28

144. Literature learn more Andrew W. Appel, Jens Palsberg: Modern Compiler Implementation in Java, 2nd edition. 2002 Garbage Collection 29

145. Outlook coming next compiler components and their generators • Lecture 11: Lexical Analysis • Lecture 12: Syntactical Analysis • Lecture 13: SDF inside • Lecture 14: Static Analysis Lab Nov 19 & Nov 26 • test cases • finish Lab 2, do reference resolution next • finish Lab 3, do hover help next • content completion Garbage Collection 30

146. questions Garbage Collection 31

147. credits Garbage Collection 32

148. Pictures copyrights Slide 1: Trash by Vladimer Shioshvili, some rights reserved Slide 5: Gravel Pile by kenjonbro, some rights reserved Slide 19: Typhoo by Dominica Williamson, some rights reserved Slide 30: Romantic Pigeon Date by Harald Hoyer, some rights reserved Slide 31: Questions by Oberazzi, some rights reserved Slide 32: Too Much Credit by Andres Rueda, some rights reserved Garbage Collection 33

Compiling Imperative and Object-Oriented Languages - Garbage Collection

Recommended

More Related Content

More from Guido Wachsmuth (16)

Recently uploaded (20)

Compiling Imperative and Object-Oriented Languages - Garbage Collection

Editor's Notes