The bytecode mumbo-jumbo

1. @rrafols the bytecode mumbo-jumbo #perfmatters

2. @rrafols Disclaimer: This presentation contains bytecode Content is my own experimentation and might differ on other environments

3. @rrafols Competitions Work 2015 Local winner 2016+ Ambassador Runner up Demoscene Speaker Author Entrepreneurship

4. @rrafols Our friend the java compiler.

5. @rrafols *.java → [javac] → *.class

6. @rrafols Or, for example, on Android

7. @rrafols *.java → [javac] → *.class *.class → [dx] → dex file

8. @rrafols *.java → [javac] → *.class *.class → [dx] → dex dex → [dexopt] → opt. dex dex → [dex2oat] → native

9. @rrafols but change is coming! Jack & Jill

10. @rrafols but change is coming! Jack & Jill

11. @rrafols *.java → [jack] → dex file

12. @rrafols Let’s focus on javac Javac vs other compilers

13. @rrafols Compilers Produces optimized code for the target platform

14. @rrafols javac Does not produce optimized code*

15. @rrafols javac Does not know on which architecture the code will be executed

16. @rrafols Source: Oracle

17. @rrafols For this reason Java bytecode & operations are stack based

18. @rrafols Easy to interpret But not the most performant solution

19. @rrafols Quick example Stack based integer addition

20. @rrafols j = j + i

21. @rrafols Java bytecode

22. @rrafols iload_3 iload_2 iadd istore_2

23. @rrafols Register based approach

24. @rrafols add r01, r02, r01 or add eax, ebx

25. @rrafols Let’s make things interesting… j = j + i + k + w + h * 2 + p * p;

26. @rrafols Java bytecode

27. @rrafols 0: iload_2 1: iload_1 2: iadd 3: iload_3 4: iadd 5: iload 4 7: iadd 8: iload 5 10: iconst_2 11: imul 12: iadd 13: iload 6 15: iload 6 17: imul 18: iadd 19: istore_2 j = j + i + k + w + h * 2 + p * p; Local vars 1: i 2: j 3: k 4: w 5: h 6: p

34. @rrafols Register based approach

35. @rrafols add r01, r02, r01 add r01, r03, r01 add r01, r04, r01 mul r07, r05, #2 add r01, r07, r01 mul r08, r06, r06 add r02, r08, r01 j = j + i + k + w + h * 2 + p * p; r01: i r02: j r03: k r04: w r05: h r06: p

36. @rrafols Java VM (JVM) Only the JVM knows the architecture where is running. In this case, for example, we used up to 8 registers

37. @rrafols Java VM (JVM) All optimizations are left to be done by the JVM

38. @rrafols Maybe takes this concept a bit too far...

39. @rrafols Imagine this simple C code #include <stdio.h> int main() { int a = 10; int b = 1 + 2 + 3 + 4 + 5 + 6 + a; printf("%dn", b); }

40. @rrafols GCC compiler #include <stdio.h> int main() { int a = 10; int b = 1 + 2 + 3 + 4 + 5 + 6 + a; printf("%dn", b); } … movl $31, %esi call _printf … * Using gcc & -O2 compiler option

41. @rrafols javac public static void main(String args[]) { int a = 10; int b = 1 + 2 + 3 + 4 + 5 + 6 + a; System.out.println(b); } 0: bipush 10 2: istore_1 3: bipush 21 5: iload_1 6: iadd 7: istore_2 ...

42. @rrafols Let's do a small change #include <stdio.h> int main() { int a = 10; int b = 1 + 2 + 3 + 4 + 5 + a + 6; printf("%dn", b); }

43. @rrafols GCC compiler #include <stdio.h> int main() { int a = 10; int b = 1 + 2 + 3 + 4 + 5 + a + 6; printf("%dn", b); } … movl $31, %esi call _printf … * Using gcc & -O2 compiler option

44. @rrafols javac public static void main(String args[]) { int a = 10; int b = 1 + 2 + 3 + 4 + 5 + a + 6; System.out.println(b); } 0: bipush 10 2: istore_1 3: bipush 15 5: iload_1 6: iadd 7: bipush 6 9: iadd 10: istore_2

45. @rrafols Let's do another quick change.. public static void main(String args[]) { int a = 10; int b = a + 1 + 2 + 3 + 4 + 5 + 6; System.out.println(b); }

46. @rrafols javac 0: bipush 10 2: istore_1 3: iload_1 4: iconst_1 5: iadd 6: iconst_2 7: iadd 8: iconst_3 9: iadd 10: iconst_4 11: iadd 12: iconst_5 13: iadd 14: bipush 6 16: iadd 17: istore_2 public static void main(String args[]) { int a = 10; int b = a + 1 + 2 + 3 + 4 + 5 + 6; System.out.println(b); }

47. @rrafols On Android there is jack to the rescue...

48. @rrafols jack public static void main(String args[]) { int a = 10; int b = a + 1 + 2 + 3 + 4 + 5 + 6; System.out.println(b); } ... 0: const/16 v0, #int 31 2: sget-object v1, Ljava/lang/System; 4: invoke-virtual {v1, v0} 7: return-void ...

49. @rrafols And on Java there is the JIT compiler to the rescue

50. @rrafols JIT assembly output public static void main(String args[]) { int a = 10; int b = a + 1 + 2 + 3 + 4 + 5 + 6; System.out.println(b); } ... 0x00000001104b2bff: mov eax, 0x0001f ... 0: bipush 10 2: istore_1 3: iload_1 4: iconst_1 5: iadd 6: iconst_2 7: iadd 8: iconst_3 9: iadd 10: iconst_4 11: iadd 12: iconst_5 13: iadd 14: bipush 6 16: iadd 17: istore_2

51. @rrafols Language additions Thinks to consider

52. @rrafols Autoboxing Transparent to the developer but compiler adds some 'extra' code

53. @rrafols Autoboxing long total = 0; for(int i = 0; i < N; i++) { total += i; } 0: lconst_0 1: lstore_1 2: iconst_0 3: istore_3 4: iload_3 5: ldc #8 // N 7: if_icmpge 21 10: lload_1 11: iload_3 12: i2l 13: ladd 14: lstore_1 15: iinc 3, 1 18: goto 4

56. @rrafols Autoboxing long total = 0; for(int i = 0; i < N; i++) { total += i; } 0: lconst_0 1: lstore_1 2: iconst_0 3: istore_3 4: iload_3 5: ldc #8 // N 7: if_icmpge 21 10: lload_1 11: iload_3 12: i2l 13: ladd 14: lstore_1 15: iinc 3, 1 18: goto 4 21:

60. @rrafols Autoboxing Long total = 0; for(Integer i = 0; i < N; i++) { total += i; }

61. @rrafols Autoboxing Long total = 0; for(Integer i = 0; i < N; i++) { total += i; } 00: lconst_0 01: invokestatic #7 // java/lang/Long.valueOf:(J)Ljava/la 04: astore_1 05: iconst_0 06: invokestatic #8 // java/lang/Integer.valueOf:(I)Ljava/ 09: astore_2 10: aload_2 11: invokevirtual #9 // java/lang/Integer.intValue:()I 14: sipush N 17: if_icmpge 54 20: aload_1 21: invokevirtual #10 // java/lang/Long.longValue:()J 24: aload_2 25: invokevirtual #9 // java/lang/Integer.intValue:()I 28: i2l 29: ladd 30: invokestatic #7 // java/lang/Long.valueOf:(J)Ljava/la 33: astore_1 34: aload_2 35: astore_3 36: aload_2 37: invokevirtual #9 // java/lang/Integer.intValue:()I 40: iconst_1 41: iadd 42: invokestatic #8 // java/lang/Integer.valueOf:(I)Ljava/ 45: dup 46: astore_2 47: astore 4 49: aload_3 50: pop 51: goto 10

68. @rrafols Autoboxing Long total = 0; for(Integer i = 0; i < N; i++) { total += i; } // ? 00: lconst_0 01: invokestatic #7 // Method java/lang/Long.valueOf:(J)L 04: astore_1 05: iconst_0 06: invokestatic #8 // Method java/lang/Integer.valueOf:(I 09: astore_2 10: aload_2 11: invokevirtual #9 // Method java/lang/Integer.intValue: 14: sipush N 17: if_icmpge 54 20: aload_1 21: invokevirtual #10 // Method java/lang/Long.longValue:( 24: aload_2 25: invokevirtual #9 // Method java/lang/Integer.intValue: 28: i2l 29: ladd 30: invokestatic #7 // Method java/lang/Long.valueOf:(J)L 33: astore_1 34: aload_2 35: astore_3 36: aload_2 37: invokevirtual #9 // Method java/lang/Integer.intValue: 40: iconst_1 41: iadd 42: invokestatic #8 // Method java/lang/Integer.valueOf:(I 45: dup 46: astore_2 47: astore 4 49: aload_3 50: pop 51: goto 10

69. @rrafols Autoboxing This is what that code is actually doing: Long total = Long.valueOf(0); for(Integer i = Integer.valueOf(0); i.intValue() < N; i = Integer.valueOf(i.intValue() + 1)) { total = Long.valueOf(total.longValue() + (long)i.intValue()) }

70. @rrafols Autoboxing Object creation Long total = Long.valueOf(0); for(Integer i = Integer.valueOf(0); i.intValue() < N; i = Integer.valueOf(i.intValue() + 1)) { total = Long.valueOf(total.longValue() + (long)i.intValue()) }

71. @rrafols Autoboxing What about Jack?

72. @rrafols Autoboxing Jack does not help in this situation

73. @rrafols Autoboxing What about the JIT compiler?

74. @rrafols Autoboxing Let's run that loop N times (on my desktop computer) N = 10.000.000.000

75. @rrafols Autoboxing

76. @rrafols Autoboxing Let’s try it on Android Dalvik VM & ART

77. @rrafols Autoboxing

78. @rrafols Language Additions Use them wisely!

79. @rrafols Sorting No bytecode mumbo-jumbo here

80. @rrafols Let's sort some numbers… Arrays.sort(...)

81. @rrafols Difference between sorting primitive types & objects

82. @rrafols Using int & Integer

83. @rrafols Sorting objects is a stable sort Default java algorithm: TimSort adaptation

84. @rrafols Sorting primitives does not require to be stable sort Default java algorithm: Dual-Pivot quicksort

85. @rrafols Sorting Use primitive types as much as possible

86. @rrafols Loops What is going on behind the scenes

87. @rrafols Loops - List ArrayList<Integer> list = new … static long loopStandardList() { long result = 0; for(int i = 0; i < list.size(); i++) { result += list.get(i); } return result; }

88. @rrafols ArrayList<Integer> list = new … static long loopStandardList() { long result = 0; for(int i = 0; i < list.size(); i++) { result += list.get(i); } return result; } 07: lload_0 08: getstatic list 11: iload_2 12: invokevirtual java/util/ArrayList.get 15: checkcast java/lang/Integer 18: invokevirtual java/lang/Integer.intValue 21: i2l 22: ladd 23: lstore_0 24: iinc 2, 1 27: iload_2 28: getstatic list 31: invokevirtual java/util/ArrayList.size 34: if_icmplt 7 Loops - List

89. @rrafols ArrayList<Integer> list = new … static long loopStandardList() { long result = 0; for(int i = 0; i < list.size(); i++) { result += list.get(i); } return result; } Loops - List 07: lload_0 08: getstatic list 11: iload_2 12: invokevirtual java/util/ArrayList.get 15: checkcast java/lang/Integer 18: invokevirtual java/lang/Integer.intValue 21: i2l 22: ladd 23: lstore_0 24: iinc 2, 1 27: iload_2 28: getstatic list 31: invokevirtual java/util/ArrayList.size 34: if_icmplt 7

92. @rrafols Loops - foreach ArrayList<Integer> list = new … static long loopForeachList() { long result = 0; for(int v : list) { result += v; } return result; }

93. @rrafols ArrayList<Integer> list = new … static long loopForeachList() { long result = 0; for(int v : list) { result += v; } return result; } 12: aload_3 13: invokeinterface java/util/Iterator.next 18: checkcast java/lang/Integer 21: invokevirtual java/lang/Integer.intValue 24: istore_2 25: lload_0 26: iload_2 27: i2l 28: ladd 29: lstore_0 30: aload_3 31: invokeinterface java/util/Iterator.hasNext 36: ifne 12 Loops - foreach

94. @rrafols ArrayList<Integer> list = new … static long loopForeachList() { long result = 0; for(int v : list) { result += v; } return result; } Loops - foreach 12: aload_3 13: invokeinterface java/util/Iterator.next 18: checkcast java/lang/Integer 21: invokevirtual java/lang/Integer.intValue 24: istore_2 25: lload_0 26: iload_2 27: i2l 28: ladd 29: lstore_0 30: aload_3 31: invokeinterface java/util/Iterator.hasNext 36: ifne 12

95. @rrafols ArrayList<Integer> list = new … static long loopForeachList() { long result = 0; for(int v : list) { result += v; } return result; } Loops - foreach 12: aload_3 13: invokeinterface java/util/Iterator.next 18: checkcast java/lang/Integer 21: invokevirtual java/lang/Integer.intValue 24: istore_2 25: lload_0 26: iload_2 27: i2l 28: ladd 29: lstore_0 30: aload_3 31: invokeinterface java/util/Iterator.hasNext 36: ifne 12

96. @rrafols Loops - Array static int[] array = new ... static long loopStandardArray() { long result = 0; for(int i = 0; i < array.length; i++) { result += array[i]; } return result; }

97. @rrafols static int[] array = new ... static long loopStandardArray() { long result = 0; for(int i = 0; i < array.length; i++) { result += array[i]; } return result; } 07: lload_0 08: getstatic array 11: iload_2 12: iaload 13: i2l 14: ladd 15: lstore_0 16: iinc 2, 1 19: iload_2 20: getstatic array 23: arraylength 24: if_icmplt 7 Loops - Array

98. @rrafols static int[] array = new ... static long loopStandardArray() { long result = 0; for(int i = 0; i < array.length; i++) { result += array[i]; } return result; } Loops - Array 07: lload_0 08: getstatic array 11: iload_2 12: iaload 13: i2l 14: ladd 15: lstore_0 16: iinc 2, 1 19: iload_2 20: getstatic array 23: arraylength 24: if_icmplt 7

99. @rrafols static int[] array = new ... static long loopStandardArray() { long result = 0; for(int i = 0; i < array.length; i++) { result += array[i]; } return result; } Loops - Array 07: lload_0 08: getstatic array 11: iload_2 12: iaload 13: i2l 14: ladd 15: lstore_0 16: iinc 2, 1 19: iload_2 20: getstatic array 23: arraylength 24: if_icmplt 7

100. @rrafols Loops - size cached static int[] array = new ... static long loopStandardArray () { long result = 0; int length = array.length; for(int i = 0; i < length; i++) { result += array[i]; } return result; }

101. @rrafols static int[] array = new ... static long loopStandardArray () { long result = 0; int length = array.length; for(int i = 0; i < length; i++) { result += array[i]; } return result; } 12: lload_0 13: getstatic array 16: iload_3 17: iaload 18: i2l 19: ladd 20: lstore_0 21: iinc 3, 1 24: iload_3 25: iload_2 26: if_icmplt 12 Loops - size cached

102. @rrafols static int[] array = new ... static long loopStandardArray () { long result = 0; int length = array.length; for(int i = 0; i < length; i++) { result += array[i]; } return result; } Loops - size cached 12: lload_0 13: getstatic array 16: iload_3 17: iaload 18: i2l 19: ladd 20: lstore_0 21: iinc 3, 1 24: iload_3 25: iload_2 26: if_icmplt 12

103. @rrafols Loops - backwards static int[] array = new ... static long loopStandardArray () { long result = 0; for(int i = array.length - 1; i >= 0; i--) { result += array[i]; } return result; }

104. @rrafols static int[] array = new ... static long loopStandardArray () { long result = 0; for(int i = array.length - 1; i >= 0; i--) { result += array[i]; } return result; } 12: lload_0 13: getstatic array 16: iload_2 17: iaload 18: i2l 19: ladd 20: lstore_0 21: iinc 2, -1 24: iload_2 25: ifge 12 Loops - backwards

105. @rrafols static int[] array = new ... static long loopStandardArray () { long result = 0; for(int i = array.length - 1; i >= 0; i--) { result += array[i]; } return result; } Loops - backwards 12: lload_0 13: getstatic array 16: iload_2 17: iaload 18: i2l 19: ladd 20: lstore_0 21: iinc 2, -1 24: iload_2 25: ifge 12

106. @rrafols

107. @rrafols

108. @rrafols Loops – foreach II static long loopForeachArray(int[] array) { long result = 0; for(int v : array) { result += v; } return result; }

109. @rrafols Loops – foreach II static long loopForeachArray(int[] array) { long result = 0; for(int v : array) { result += v; } return result; } 00: lconst_0 01: lstore_1 02: aload_0 03: dup 04: astore 6 06: arraylength 07: istore 5 09: iconst_0 10: istore 4 12: goto 29 15: aload 6 17: iload 4 19: iaload 20: istore_3 21: lload_1 22: iload_3 23: i2l 24: ladd 25: lstore_1 26: iinc 4, 1 29: iload 4 31: iload 5 33: if_icmplt 15 0 array 4 - 1 - 5 - 2 - 6 - 3 - 7 -

110. @rrafols Loops – foreach II static long loopForeachArray(int[] array) { long result = 0; for(int v : array) { result += v; } return result; } 00: lconst_0 01: lstore_1 02: aload_0 03: dup 04: astore 6 06: arraylength 07: istore 5 09: iconst_0 10: istore 4 12: goto 29 15: aload 6 17: iload 4 19: iaload 20: istore_3 21: lload_1 22: iload_3 23: i2l 24: ladd 25: lstore_1 26: iinc 4, 1 29: iload 4 31: iload 5 33: if_icmplt 15 0 array 4 - 1 0 (result) 5 - 2 - 6 - 3 - 7 -

111. @rrafols Loops – foreach II static long loopForeachArray(int[] array) { long result = 0; for(int v : array) { result += v; } return result; } 00: lconst_0 01: lstore_1 02: aload_0 03: dup 04: astore 6 06: arraylength 07: istore 5 09: iconst_0 10: istore 4 12: goto 29 15: aload 6 17: iload 4 19: iaload 20: istore_3 21: lload_1 22: iload_3 23: i2l 24: ladd 25: lstore_1 26: iinc 4, 1 29: iload 4 31: iload 5 33: if_icmplt 15 0 array 4 - 1 0 (result) 5 - 2 - 6 array 3 - 7 -

112. @rrafols Loops – foreach II static long loopForeachArray(int[] array) { long result = 0; for(int v : array) { result += v; } return result; } 00: lconst_0 01: lstore_1 02: aload_0 03: dup 04: astore 6 06: arraylength 07: istore 5 09: iconst_0 10: istore 4 12: goto 29 15: aload 6 17: iload 4 19: iaload 20: istore_3 21: lload_1 22: iload_3 23: i2l 24: ladd 25: lstore_1 26: iinc 4, 1 29: iload 4 31: iload 5 33: if_icmplt 15 0 array 4 - 1 0 (result) 5 array.length 2 - 6 array 3 - 7 -

113. @rrafols Loops – foreach II static long loopForeachArray(int[] array) { long result = 0; for(int v : array) { result += v; } return result; } 00: lconst_0 01: lstore_1 02: aload_0 03: dup 04: astore 6 06: arraylength 07: istore 5 09: iconst_0 10: istore 4 12: goto 29 15: aload 6 17: iload 4 19: iaload 20: istore_3 21: lload_1 22: iload_3 23: i2l 24: ladd 25: lstore_1 26: iinc 4, 1 29: iload 4 31: iload 5 33: if_icmplt 15 0 array 4 0 (loop index) 1 0 (result) 5 array.length 2 - 6 array 3 - 7 -

114. @rrafols Loops – foreach II static long loopForeachArray(int[] array) { long result = 0; for(int v : array) { result += v; } return result; } 00: lconst_0 01: lstore_1 02: aload_0 03: dup 04: astore 6 06: arraylength 07: istore 5 09: iconst_0 10: istore 4 12: goto 29 15: aload 6 17: iload 4 19: iaload 20: istore_3 21: lload_1 22: iload_3 23: i2l 24: ladd 25: lstore_1 26: iinc 4, 1 29: iload 4 31: iload 5 33: if_icmplt 15 0 array 4 0 (loop index) 1 0 (result) 5 array.length 2 - 6 array 3 - 7 -

115. @rrafols Loops – foreach II static long loopForeachArray(int[] array) { long result = 0; for(int v : array) { result += v; } return result; } 00: lconst_0 01: lstore_1 02: aload_0 03: dup 04: astore 6 06: arraylength 07: istore 5 09: iconst_0 10: istore 4 12: goto 29 15: aload 6 17: iload 4 19: iaload 20: istore_3 21: lload_1 22: iload_3 23: i2l 24: ladd 25: lstore_1 26: iinc 4, 1 29: iload 4 31: iload 5 33: if_icmplt 15 0 array 4 0 (loop index) 1 0 (result) 5 array.length 2 - 6 array 3 array[index] 7 -

116. @rrafols Loops – foreach II static long loopForeachArray(int[] array) { long result = 0; for(int v : array) { result += v; } return result; } 00: lconst_0 01: lstore_1 02: aload_0 03: dup 04: astore 6 06: arraylength 07: istore 5 09: iconst_0 10: istore 4 12: goto 29 15: aload 6 17: iload 4 19: iaload 20: istore_3 21: lload_1 22: iload_3 23: i2l 24: ladd 25: lstore_1 26: iinc 4, 1 29: iload 4 31: iload 5 33: if_icmplt 15 0 array 4 0 (loop index) 1 0 + array[index] 5 array.length 2 - 6 array 3 array[index] 7 -

117. @rrafols Loops – foreach II static long loopForeachArray(int[] array) { long result = 0; for(int v : array) { result += v; } return result; } 00: lconst_0 01: lstore_1 02: aload_0 03: dup 04: astore 6 06: arraylength 07: istore 5 09: iconst_0 10: istore 4 12: goto 29 15: aload 6 17: iload 4 19: iaload 20: istore_3 21: lload_1 22: iload_3 23: i2l 24: ladd 25: lstore_1 26: iinc 4, 1 29: iload 4 31: iload 5 33: if_icmplt 15 0 array 4 0 (loop index) + 1 1 0 + array[index] 5 array.length 2 - 6 array 3 array[index] 7 -

118. @rrafols

119. @rrafols

120. @rrafols Loops Use arrays instead of lists

121. @rrafols Loops When using lists, avoid foreach or iterator constructions if performance is a requirement

122. @rrafols Manual bytecode optimization Worth it?

123. @rrafols foreach loop 0: lconst_0 1: lstore_1 2: aload_0 3: dup 4: astore 6 6: arraylength 7: istore 5 9: iconst_0 10: istore 4 12: goto 29 15: aload 6 17: iload 4 19: iaload 20: istore_3 21: lload_1 22: iload_3 23: i2l 24: ladd 25: lstore_1 26: iinc 4, 1 29: iload 4 31: iload 5 33: if_icmplt 15 0: aload_0 1: arraylength 2: istore_3 3: iconst_0 4: istore_1 5: lconst_0 6: goto 17 9: aload_0 10: iload_1 11: iaload 12: i2l 13: ladd 14: iinc 1, 1 17: iload_1 18: iload_3 19: if_icmplt 9 Manual bytecode optimization

124. @rrafols Manual bytecode optimization 0: aload_0 1: arraylength 2: istore_3 3: iconst_0 4: istore_1 5: lconst_0 6: goto 17 9: aload_0 10: iload_1 11: iaload 12: i2l 13: ladd 14: iinc 1, 1 17: iload_1 18: iload_3 19: if_icmplt 9 Local variables 0 array 3 - 1 - 4 - 2 - 5 - Stack - - -

125. @rrafols 0: aload_0 1: arraylength 2: istore_3 3: iconst_0 4: istore_1 5: lconst_0 6: goto 17 9: aload_0 10: iload_1 11: iaload 12: i2l 13: ladd 14: iinc 1, 1 17: iload_1 18: iload_3 19: if_icmplt 9 Local variables 0 array 3 array.length 1 - 4 - 2 - 5 - Stack - - - Manual bytecode optimization

126. @rrafols 0: aload_0 1: arraylength 2: istore_3 3: iconst_0 4: istore_1 5: lconst_0 6: goto 17 9: aload_0 10: iload_1 11: iaload 12: i2l 13: ladd 14: iinc 1, 1 17: iload_1 18: iload_3 19: if_icmplt 9 Local variables 0 array 3 array.length 1 0 (index) 4 - 2 - 5 - Stack - - - Manual bytecode optimization

127. @rrafols 0: aload_0 1: arraylength 2: istore_3 3: iconst_0 4: istore_1 5: lconst_0 6: goto 17 9: aload_0 10: iload_1 11: iaload 12: i2l 13: ladd 14: iinc 1, 1 17: iload_1 18: iload_3 19: if_icmplt 9 Local variables 0 array 3 array.length 1 0 (index) 4 - 2 - 5 - Stack 0 (result) - - Manual bytecode optimization

128. @rrafols 0: aload_0 1: arraylength 2: istore_3 3: iconst_0 4: istore_1 5: lconst_0 6: goto 17 9: aload_0 10: iload_1 11: iaload 12: i2l 13: ladd 14: iinc 1, 1 17: iload_1 18: iload_3 19: if_icmplt 9 Local variables 0 array 3 array.length 1 0 (index) 4 - 2 - 5 - Stack 0 (result) array[index] - Manual bytecode optimization

129. @rrafols 0: aload_0 1: arraylength 2: istore_3 3: iconst_0 4: istore_1 5: lconst_0 6: goto 17 9: aload_0 10: iload_1 11: iaload 12: i2l 13: ladd 14: iinc 1, 1 17: iload_1 18: iload_3 19: if_icmplt 9 Local variables 0 array 3 array.length 1 0 (index) 4 - 2 - 5 - Stack 0 + array[index] - - Manual bytecode optimization

130. @rrafols 0: aload_0 1: arraylength 2: istore_3 3: iconst_0 4: istore_1 5: lconst_0 6: goto 17 9: aload_0 10: iload_1 11: iaload 12: i2l 13: ladd 14: iinc 1, 1 17: iload_1 18: iload_3 19: if_icmplt 9 Local variables 0 array 3 array.length 1 0 (index) + 1 4 - 2 - 5 - Stack 0 + array[index] - - Manual bytecode optimization

131. @rrafols

132. @rrafols Worth it? Only in very specific, rare, unique, special, peculiar cases. Too much effort involved.

133. @rrafols Calling a method Is there an overhead?

134. @rrafols Overhead of calling a method for(int i = 0; i < N; i++) { setVal(getVal() + 1); } for(int i = 0; i < N; i++) { val = val + 1; } vs

135. @rrafols

136. @rrafols String concatenation The evil + sign

137. @rrafols String concatenation String str = ""; for(int i = 0; i < N; i++) { str += OTHER_STR; }

138. @rrafols String concatenation String str = ""; for(int i = 0; i < N; i++) { str += OTHER_STR; } 0: ldc String 2: astore_1 3: iconst_0 4: istore_2 5: iload_2 6: sipush N 9: if_icmpge 40 12: new class java/lang/StringBuilder 15: dup 16: invokespecial java/lang/StringBuilder."<init>" 19: aload_1 20: invokevirtual java/lang/StringBuilder.append 23: aload_0 24: getfield OTHER_STR 27: invokevirtual java/lang/StringBuilder.append 30: invokevirtual java/lang/StringBuilder.toString 33: astore_1 34: iinc 2, 1 37: goto 5

144. @rrafols String str = ""; for(int i = 0; i < N; i++) { StringBuilder sb = new StringBuilder(); sb.append(str); sb.append(OTHER_STR); str = sb.toString(); } String concatenation

145. @rrafols Object creation: String str = ""; for(int i = 0; i < N; i++) { StringBuilder sb = new StringBuilder(); sb.append(str); sb.append(OTHER_STR); str = sb.toString(); } String concatenation

146. @rrafols String concatenation alternatives

147. @rrafols String.concat() • Concat cost is O(N) + O(M) • Concat returns a new String Object. String str = ""; for(int i = 0; i < N; i++) { str = str.concat(OTHER_STR); }

148. @rrafols String.concat() Object creation: String str = ""; for(int i = 0; i < N; i++) { str = str.concat(OTHER_STR); }

149. @rrafols StringBuilder • StringBuilder.append cost is O(M) [M being the length of appended String] StringBuilder sb = new StringBuilder() for(int i = 0; i < N; i++) { sb.append(OTHER_STR); } str = sb.toString();

150. @rrafols StringBuilder sb = new StringBuilder() for(int i = 0; i < N; i++) { sb.append(OTHER_STR); } str = sb.toString(); 0: ldc String 2: astore_1 3: new java/lang/StringBuilder 6: dup 7: invokespecial java/lang/StringBuilder."<init>" 10: astore_2 11: iconst_0 12: istore_3 13: iload_3 14: sipush N 17: if_icmpge 35 20: aload_2 21: aload_0 22: getfield OTHER_STR 25: invokevirtual java/lang/StringBuilder.append 28: pop 29: iinc 3, 1 32: goto 13

151. @rrafols 0: ldc String 2: astore_1 3: new java/lang/StringBuilder 6: dup 7: invokespecial java/lang/StringBuilder."<init>" 10: astore_2 11: iconst_0 12: istore_3 13: iload_3 14: sipush N 17: if_icmpge 35 20: aload_2 21: aload_0 22: getfield OTHER_STR 25: invokevirtual java/lang/StringBuilder.append 28: pop 29: iinc 3, 1 32: goto 13 sb = new StringBuilder() for(int i = 0; i < N; i++) { sb.append(OTHER_STR); } str = sb.toString(); StringBuilder

152. @rrafols Object creation: StringBuilder sb = new StringBuilder(); for(int i = 0; i < N; i++) { sb.append(OTHER_STR); } str = sb.toString(); StringBuilder

153. @rrafols String concatenation Use StringBuilder (properly) as much as possible. StringBuffer is the thread safe implementation.

154. @rrafols Strings in case statements

155. @rrafols public void taskStateMachine(String status) { switch(status) { case "PENDING": System.out.println("Status pending"); break; case "EXECUTING": System.out.println("Status executing"); break; } }

156. @rrafols

157. @rrafols

158. @rrafols Tooling

159. @rrafols Tooling - Disassembler Java • javap -c <classfile> Android: •Dexdump -d <dexfile>

160. @rrafols Tooling - Assembler Krakatau https://github.com/Storyyeller/Krakatau

161. @rrafols Tooling – Disassembler - ART adb pull /data/dalvik- cache/arm/data@app@<package>-1@base apk@classes.dex gobjdump -D <file>

162. @rrafols Tooling – Disassembler - ART adb shell oatdump --oat-file=/data/dalvik- cache/arm/data@app@<package>- 1@base.apk@classes.dex

163. @rrafols Tooling – PrintAssembly - JIT -XX:+UnlockDiagnosticVMOptions -XX:+PrintAssembly -XX:CompileCommand=print,com.raimon.test.Test::method Under the Hood of the JVM: From Bytecode Through the JIT to Assembly http://alblue.bandlem.com/2016/09/javaone-hotspot.html

164. @rrafols Always measure example - yuv2rgb optimization

165. @rrafols Source: Wikipedia

166. @rrafols

167. @rrafols Slightly optimized version precalc tables, fixed point operations, 2 pixels per loop…

168. @rrafols

169. @rrafols

170. @rrafols Lets compare: Normal, minified, minified with optimizations & jack Minified = obfuscated using Proguard

171. @rrafols Normal Minified Minified & optimized Jack 0 2000 4000 6000 8000 10000 12000 14000 16000 18000 20000 non-optimized optimized

172. @rrafols Performance measurements Avoid doing multiple tests in one run JIT might be evil!

173. @rrafols Thank you! http://blog.rafols.org @rrafols https://es.linkedin.com/in/raimonrafols

The bytecode mumbo-jumbo

Recommandé

Recommandé

Contenu connexe

Tendances

Tendances (20)

Similaire à The bytecode mumbo-jumbo

Similaire à The bytecode mumbo-jumbo (20)

Plus de Raimon Ràfols

Plus de Raimon Ràfols (7)

The bytecode mumbo-jumbo