Tweet
Tweet

Slide 1

Slide 1 text

Writing a basic x86-64 JIT compiler from scratch in stock Python Christian Stigen Larsen, 2018-01-23 https://csl.name

Slide 2

Slide 2 text

No content

Slide 3

Slide 3 text

from jitcompiler import jit @jit def foo(a, b): return a*a - b*b

Slide 4

Slide 4 text

from jitcompiler import * @jit def foo(a, b): return a*a - b*b

Slide 5

Slide 5 text

from jitcompiler import * @jit def foo(a, b): return a*a - b*b

Slide 6

Slide 6 text

from jitcompiler import * @jit def foo(a, b): return a*a - b*b

Slide 7

Slide 7 text

from jitcompiler import * @jit def foo(a, b): return a*a - b*b >>> foo(2, 3)

Slide 8

Slide 8 text

from jitcompiler import * @jit def foo(a, b): return a*a - b*b >>> foo(2, 3) --- JIT-compiling

Slide 9

Slide 9 text

from jitcompiler import * @jit def foo(a, b): return a*a - b*b >>> foo(2, 3) --- JIT-compiling -5

Slide 10

Slide 10 text

from jitcompiler import * @jit def foo(a, b): return a*a - b*b >>> foo(2, 3) --- JIT-compiling -5 >>> foo(3, 4)

Slide 11

Slide 11 text

from jitcompiler import * @jit def foo(a, b): return a*a - b*b >>> foo(2, 3) --- JIT-compiling -5 >>> foo(3, 4) -7

Slide 12

Slide 12 text

No content

Slide 13

Slide 13 text

>>> print(disassemble(foo)) 0x100b1d000 48 89 fb mov rbx, rdi 0x100b1d003 48 89 f8 mov rax, rdi 0x100b1d006 48 0f af c3 imul rax, rbx 0x100b1d00a 50 push rax 0x100b1d00b 48 89 f3 mov rbx, rsi 0x100b1d00e 48 89 f0 mov rax, rsi 0x100b1d011 48 0f af c3 imul rax, rbx 0x100b1d015 48 89 c3 mov rbx, rax 0x100b1d018 58 pop rax 0x100b1d019 48 29 d8 sub rax, rbx 0x100b1d01c c3 ret

Slide 14

Slide 14 text

>>> print(disassemble(foo)) 0x100b1d000 48 89 fb mov rbx, rdi 0x100b1d003 48 89 f8 mov rax, rdi 0x100b1d006 48 0f af c3 imul rax, rbx 0x100b1d00a 50 push rax 0x100b1d00b 48 89 f3 mov rbx, rsi 0x100b1d00e 48 89 f0 mov rax, rsi 0x100b1d011 48 0f af c3 imul rax, rbx 0x100b1d015 48 89 c3 mov rbx, rax 0x100b1d018 58 pop rax 0x100b1d019 48 29 d8 sub rax, rbx 0x100b1d01c c3 ret

Slide 15

Slide 15 text

No content

Slide 16

Slide 16 text

Strategy Python Bound Name def foo

Slide 17

Slide 17 text

Python Function Strategy Bytecode Python Bound Name def foo

Slide 18

Slide 18 text

Python Function Strategy Python Bound Name Bytecode Bytecode def foo

Slide 19

Slide 19 text

Python Function Strategy Python Bound Name Bytecode Bytecode Bytecode def foo

Slide 20

Slide 20 text

Python Function Strategy Python Bound Name Bytecode Bytecode Bytecode IR def foo

Slide 21

Slide 21 text

Python Function Strategy Python Bound Name Bytecode Bytecode Bytecode IR Optimized IR def foo

Slide 22

Slide 22 text

Python Function Strategy Python Bound Name Bytecode Bytecode Bytecode IR Optimized IR x86-64 Machine Code def foo

Slide 23

Slide 23 text

Python Function Strategy Python Bound Name Memory Page Bytes … Read + Write Bytecode Bytecode Bytecode IR Optimized IR x86-64 Machine Code def foo

Slide 24

Slide 24 text

Python Function Strategy Python Bound Name Memory Page Bytes … Read + Write Bytecode x86-64 Machine Code def foo

Slide 25

Slide 25 text

Python Function Strategy Python Bound Name Memory Page Bytes … Read + Write Bytecode x86-64 Machine Code Read + Execute def foo

Slide 26

Slide 26 text

Strategy Python Bound Name Memory Page Bytes … Read + Write x86-64 Machine Code Read + Execute FFI def foo

Slide 27

Slide 27 text

Intermission: Why JIT?

Slide 28

Slide 28 text

Intermission: Why JIT? Because we can, and it’s awesome.

Slide 29

Slide 29 text

No content

Slide 30

Slide 30 text

Python Bytecode def foo(a, b): return a*a - b*b

Slide 31

Slide 31 text

Python Bytecode def foo(a, b): return a*a - b*b >>> import dis >>> dis.dis(foo)

Slide 32

Slide 32 text

Python Bytecode def foo(a, b): return a*a - b*b >>> import dis >>> dis.dis(foo) 2 0 LOAD_FAST 0 (a) 3 LOAD_FAST 0 (a) 6 BINARY_MULTIPLY 7 LOAD_FAST 1 (b) 10 LOAD_FAST 1 (b) 13 BINARY_MULTIPLY 14 BINARY_SUBTRACT 15 RETURN_VALUE

Slide 33

Slide 33 text

Python Bytecode def foo(a, b): return a*a - b*b >>> import dis >>> dis.dis(foo) 2 0 LOAD_FAST 0 (a) 3 LOAD_FAST 0 (a) 6 BINARY_MULTIPLY 7 LOAD_FAST 1 (b) 10 LOAD_FAST 1 (b) 13 BINARY_MULTIPLY 14 BINARY_SUBTRACT 15 RETURN_VALUE

Slide 34

Slide 34 text

Python Bytecode def foo(a, b): return a*a - b*b >>> import dis >>> dis.dis(foo) 2 0 LOAD_FAST 0 (a) 3 LOAD_FAST 0 (a) 6 BINARY_MULTIPLY 7 LOAD_FAST 1 (b) 10 LOAD_FAST 1 (b) 13 BINARY_MULTIPLY 14 BINARY_SUBTRACT 15 RETURN_VALUE

Slide 35

Slide 35 text

Python Bytecode def foo(a, b): return a*a - b*b >>> import dis >>> dis.dis(foo) 2 0 LOAD_FAST 0 (a) 3 LOAD_FAST 0 (a) 6 BINARY_MULTIPLY 7 LOAD_FAST 1 (b) 10 LOAD_FAST 1 (b) 13 BINARY_MULTIPLY 14 BINARY_SUBTRACT 15 RETURN_VALUE

Slide 36

Slide 36 text

Python Bytecode def foo(a, b): return a*a - b*b >>> import dis >>> dis.dis(foo) 2 0 LOAD_FAST 0 (a) 3 LOAD_FAST 0 (a) 6 BINARY_MULTIPLY 7 LOAD_FAST 1 (b) 10 LOAD_FAST 1 (b) 13 BINARY_MULTIPLY 14 BINARY_SUBTRACT 15 RETURN_VALUE

Slide 37

Slide 37 text

Python Bytecode def foo(a, b): return a*a - b*b >>> import dis >>> dis.dis(foo) 2 0 LOAD_FAST 0 (a) 3 LOAD_FAST 0 (a) 6 BINARY_MULTIPLY 7 LOAD_FAST 1 (b) 10 LOAD_FAST 1 (b) 13 BINARY_MULTIPLY 14 BINARY_SUBTRACT 15 RETURN_VALUE

Slide 38

Slide 38 text

Python Bytecode def foo(a, b): return a*a - b*b >>> import dis >>> dis.dis(foo) 2 0 LOAD_FAST 0 (a) 3 LOAD_FAST 0 (a) 6 BINARY_MULTIPLY 7 LOAD_FAST 1 (b) 10 LOAD_FAST 1 (b) 13 BINARY_MULTIPLY 14 BINARY_SUBTRACT 15 RETURN_VALUE

Slide 39

Slide 39 text

Python Bytecode def foo(a, b): return a*a - b*b >>> import dis >>> dis.dis(foo) 2 0 LOAD_FAST 0 (a) 3 LOAD_FAST 0 (a) 6 BINARY_MULTIPLY 7 LOAD_FAST 1 (b) 10 LOAD_FAST 1 (b) 13 BINARY_MULTIPLY 14 BINARY_SUBTRACT 15 RETURN_VALUE

Slide 40

Slide 40 text

0 LOAD_FAST 0 (a) 3 LOAD_FAST 0 (a) 6 BINARY_MULTIPLY 7 LOAD_FAST 1 (b) 10 LOAD_FAST 1 (b) 13 BINARY_MULTIPLY 14 BINARY_SUBTRACT 15 RETURN_VALUE

Slide 41

Slide 41 text

0 LOAD_FAST 0 (a) 3 LOAD_FAST 0 (a) 6 BINARY_MULTIPLY 7 LOAD_FAST 1 (b) 10 LOAD_FAST 1 (b) 13 BINARY_MULTIPLY 14 BINARY_SUBTRACT 15 RETURN_VALUE

Slide 42

Slide 42 text

0 LOAD_FAST 0 (a) 3 LOAD_FAST 0 (a) 6 BINARY_MULTIPLY 7 LOAD_FAST 1 (b) 10 LOAD_FAST 1 (b) 13 BINARY_MULTIPLY 14 BINARY_SUBTRACT 15 RETURN_VALUE a

Slide 43

Slide 43 text

0 LOAD_FAST 0 (a) 3 LOAD_FAST 0 (a) 6 BINARY_MULTIPLY 7 LOAD_FAST 1 (b) 10 LOAD_FAST 1 (b) 13 BINARY_MULTIPLY 14 BINARY_SUBTRACT 15 RETURN_VALUE a a

Slide 44

Slide 44 text

0 LOAD_FAST 0 (a) 3 LOAD_FAST 0 (a) 6 BINARY_MULTIPLY 7 LOAD_FAST 1 (b) 10 LOAD_FAST 1 (b) 13 BINARY_MULTIPLY 14 BINARY_SUBTRACT 15 RETURN_VALUE a*a

Slide 45

Slide 45 text

0 LOAD_FAST 0 (a) 3 LOAD_FAST 0 (a) 6 BINARY_MULTIPLY 7 LOAD_FAST 1 (b) 10 LOAD_FAST 1 (b) 13 BINARY_MULTIPLY 14 BINARY_SUBTRACT 15 RETURN_VALUE a*a b

Slide 46

Slide 46 text

0 LOAD_FAST 0 (a) 3 LOAD_FAST 0 (a) 6 BINARY_MULTIPLY 7 LOAD_FAST 1 (b) 10 LOAD_FAST 1 (b) 13 BINARY_MULTIPLY 14 BINARY_SUBTRACT 15 RETURN_VALUE a*a b b

Slide 47

Slide 47 text

0 LOAD_FAST 0 (a) 3 LOAD_FAST 0 (a) 6 BINARY_MULTIPLY 7 LOAD_FAST 1 (b) 10 LOAD_FAST 1 (b) 13 BINARY_MULTIPLY 14 BINARY_SUBTRACT 15 RETURN_VALUE a*a b*b

Slide 48

Slide 48 text

0 LOAD_FAST 0 (a) 3 LOAD_FAST 0 (a) 6 BINARY_MULTIPLY 7 LOAD_FAST 1 (b) 10 LOAD_FAST 1 (b) 13 BINARY_MULTIPLY 14 BINARY_SUBTRACT 15 RETURN_VALUE a*a-b*b

Slide 49

Slide 49 text

0 LOAD_FAST 0 (a) 3 LOAD_FAST 0 (a) 6 BINARY_MULTIPLY 7 LOAD_FAST 1 (b) 10 LOAD_FAST 1 (b) 13 BINARY_MULTIPLY 14 BINARY_SUBTRACT 15 RETURN_VALUE a*a-b*b

Slide 50

Slide 50 text

0 LOAD_FAST 0 (a)

Slide 51

Slide 51 text

0 LOAD_FAST 0 (a) >>> foo.__code__.co_code ‘|\x00\x00|\x00\x00\x14|\x01\x00|\x01\x00\x14\x185’

Slide 52

Slide 52 text

0 LOAD_FAST 0 (a) >>> foo.__code__.co_code ‘|\x00\x00|\x00\x00\x14|\x01\x00|\x01\x00\x14\x185’

Slide 53

Slide 53 text

0 LOAD_FAST 0 (a) >>> foo.__code__.co_code ‘|\x00\x00|\x00\x00\x14|\x01\x00|\x01\x00\x14\x185’

Slide 54

Slide 54 text

0 LOAD_FAST 0 (a) >>> foo.__code__.co_code ‘|\x00\x00|\x00\x00\x14|\x01\x00|\x01\x00\x14\x185’

Slide 55

Slide 55 text

0 LOAD_FAST 0 (a) >>> foo.__code__.co_code ‘|\x00\x00|\x00\x00\x14|\x01\x00|\x01\x00\x14\x185’ >>> bytecode = map(ord, foo.__code__.co_code)

Slide 56

Slide 56 text

0 LOAD_FAST 0 (a) >>> foo.__code__.co_code ‘|\x00\x00|\x00\x00\x14|\x01\x00|\x01\x00\x14\x185’ >>> bytecode = map(ord, foo.__code__.co_code) >>> opcode = bytecode[0]

Slide 57

Slide 57 text

0 LOAD_FAST 0 (a) >>> foo.__code__.co_code ‘|\x00\x00|\x00\x00\x14|\x01\x00|\x01\x00\x14\x185’ >>> bytecode = map(ord, foo.__code__.co_code) >>> opcode = bytecode[0] >>> opcode 124

Slide 58

Slide 58 text

0 LOAD_FAST 0 (a) >>> foo.__code__.co_code ‘|\x00\x00|\x00\x00\x14|\x01\x00|\x01\x00\x14\x185’ >>> bytecode = map(ord, foo.__code__.co_code) >>> opcode = bytecode[0] >>> opcode 124 >>> dis.opname[opcode]

Slide 59

Slide 59 text

0 LOAD_FAST 0 (a) >>> foo.__code__.co_code ‘|\x00\x00|\x00\x00\x14|\x01\x00|\x01\x00\x14\x185’ >>> bytecode = map(ord, foo.__code__.co_code) >>> opcode = bytecode[0] >>> opcode 124 >>> dis.opname[opcode] ‘LOAD_FAST’

Slide 60

Slide 60 text

0 LOAD_FAST 0 (a) >>> foo.__code__.co_code ‘|\x00\x00|\x00\x00\x14|\x01\x00|\x01\x00\x14\x185’ >>> bytecode = map(ord, foo.__code__.co_code) >>> opcode = bytecode[0] >>> opcode 124 >>> dis.opname[opcode] ‘LOAD_FAST’ >>> arg = bytecode[1] | bytecode[2] << 8

Slide 61

Slide 61 text

0 LOAD_FAST 0 (a) >>> foo.__code__.co_code ‘|\x00\x00|\x00\x00\x14|\x01\x00|\x01\x00\x14\x185’ >>> bytecode = map(ord, foo.__code__.co_code) >>> opcode = bytecode[0] >>> opcode 124 >>> dis.opname[opcode] ‘LOAD_FAST’ >>> arg = bytecode[1] | bytecode[2] << 8 >>> arg 0

Slide 62

Slide 62 text

0 LOAD_FAST 0 (a) >>> foo.__code__.co_code ‘|\x00\x00|\x00\x00\x14|\x01\x00|\x01\x00\x14\x185’ >>> bytecode = map(ord, foo.__code__.co_code) >>> opcode = bytecode[0] >>> opcode 124 >>> dis.opname[opcode] ‘LOAD_FAST’ >>> arg = bytecode[1] | bytecode[2] << 8 >>> arg 0 >>> foo.__code__.co_varnames[arg] ‘a’

Slide 63

Slide 63 text

No content

Slide 64

Slide 64 text

>>> def bar(): return 123

Slide 65

Slide 65 text

>>> dis.dis(bar) 1 0 LOAD_CONST 1 (123) 3 RETURN_VALUE >>> def bar(): return 123

Slide 66

Slide 66 text

>>> dis.dis(bar) 1 0 LOAD_CONST 1 (123) 3 RETURN_VALUE >>> def bar(): return 123

Slide 67

Slide 67 text

>>> dis.dis(bar) 1 0 LOAD_CONST 1 (123) 3 RETURN_VALUE >>> def bar(): return 123 >>> bar.__code__.co_consts[1] 123

Slide 68

Slide 68 text

>>> dis.dis(bar) 1 0 LOAD_CONST 1 (123) 3 RETURN_VALUE >>> def bar(): return 123 >>> bar.__code__.co_consts[1] 123

Slide 69

Slide 69 text

No content

Slide 70

Slide 70 text

class Compiler(object): # …

Slide 71

Slide 71 text

class Compiler(object): # … def fetch(self): byte = self.bytecode[self.index]

Slide 72

Slide 72 text

class Compiler(object): # … def fetch(self): byte = self.bytecode[self.index] self.index += 1 return byte

Slide 73

Slide 73 text

class Compiler(object): # … def fetch(self): byte = self.bytecode[self.index] self.index += 1 return byte def decode(self): opcode = self.fetch() opname = dis.opname[opcode]

Slide 74

Slide 74 text

class Compiler(object): # … def fetch(self): byte = self.bytecode[self.index] self.index += 1 return byte def decode(self): opcode = self.fetch() opname = dis.opname[opcode] if takes_arg(opname): arg = self.fetch() | self.fetch() << 8 else: arg = None

Slide 75

Slide 75 text

class Compiler(object): # … def fetch(self): byte = self.bytecode[self.index] self.index += 1 return byte def decode(self): opcode = self.fetch() opname = dis.opname[opcode] if takes_arg(opname): arg = self.fetch() | self.fetch() << 8 else: arg = None return opname, arg

Slide 76

Slide 76 text

No content

Slide 77

Slide 77 text

def compile(self): while self.index < len(self.bytecode): op, arg = self.decode() if op == “LOAD_FAST”: yield “push”, self.variable(arg), None elif …

Slide 78

Slide 78 text

yield “push”, self.variable(arg), None

Slide 79

Slide 79 text

yield “push”, self.variable(arg), None

Slide 80

Slide 80 text

yield “push”, self.variable(arg), None IR & Assembly

Slide 81

Slide 81 text

yield “push”, self.variable(arg), None IR & Assembly rax rbx rcx … rsi rdi … Registers return address Stack

Slide 82

Slide 82 text

push yield “push”, self.variable(arg), None IR & Assembly rax rbx rcx … rsi rdi … Registers return address Stack

Slide 83

Slide 83 text

push yield “push”, self.variable(arg), None IR & Assembly rax rbx rcx … rsi rdi … Registers return address Stack

Slide 84

Slide 84 text

push yield “push”, self.variable(arg), None IR & Assembly rax rbx rcx … rsi rdi … Registers return address Stack

Slide 85

Slide 85 text

push yield “push”, self.variable(arg), None IR & Assembly rax rbx rcx … rsi rdi … Registers return address Stack

Slide 86

Slide 86 text

push yield “push”, self.variable(arg), None IR & Assembly rax rbx rcx … rsi rdi … Registers return address Stack def variable(self, index): passing_order = (“rdi”, “rsi”, “rdx”, “rcx”) return passing_order[index]

Slide 87

Slide 87 text

push yield “push”, self.variable(arg), None IR & Assembly rax rbx rcx … rsi rdi … Registers return address Stack

Slide 88

Slide 88 text

push yield “push”, self.variable(arg), None IR & Assembly rax rbx rcx … rsi rdi … Registers return address Stack rdi

Slide 89

Slide 89 text

No content

Slide 90

Slide 90 text

def baz(a, b): a = b

Slide 91

Slide 91 text

def baz(a, b): a = b LOAD_FAST 1 (b) STORE_FAST 0 (a) LOAD_CONST 0 (None) RETURN_VALUE

Slide 92

Slide 92 text

LOAD_FAST 1 def baz(a, b): a = b LOAD_FAST 1 (b) STORE_FAST 0 (a) LOAD_CONST 0 (None) RETURN_VALUE

Slide 93

Slide 93 text

LOAD_FAST 1 push rsi def baz(a, b): a = b LOAD_FAST 1 (b) STORE_FAST 0 (a) LOAD_CONST 0 (None) RETURN_VALUE

Slide 94

Slide 94 text

LOAD_FAST 1 push rsi STORE_FAST 0 def baz(a, b): a = b LOAD_FAST 1 (b) STORE_FAST 0 (a) LOAD_CONST 0 (None) RETURN_VALUE

Slide 95

Slide 95 text

LOAD_FAST 1 push rsi STORE_FAST 0 pop rax mov rdi, rax def baz(a, b): a = b LOAD_FAST 1 (b) STORE_FAST 0 (a) LOAD_CONST 0 (None) RETURN_VALUE

Slide 96

Slide 96 text

LOAD_FAST 1 push rsi STORE_FAST 0 pop rax mov rdi, rax def baz(a, b): a = b LOAD_FAST 1 (b) STORE_FAST 0 (a) LOAD_CONST 0 (None) RETURN_VALUE LOAD_CONST 0

Slide 97

Slide 97 text

LOAD_FAST 1 push rsi STORE_FAST 0 pop rax mov rdi, rax def baz(a, b): a = b LOAD_FAST 1 (b) STORE_FAST 0 (a) LOAD_CONST 0 (None) RETURN_VALUE LOAD_CONST 0 imm rax, 0 push rax

Slide 98

Slide 98 text

LOAD_FAST 1 push rsi STORE_FAST 0 pop rax mov rdi, rax def baz(a, b): a = b LOAD_FAST 1 (b) STORE_FAST 0 (a) LOAD_CONST 0 (None) RETURN_VALUE LOAD_CONST 0 imm rax, 0 push rax RETURN_VALUE

Slide 99

Slide 99 text

LOAD_FAST 1 push rsi STORE_FAST 0 pop rax mov rdi, rax def baz(a, b): a = b LOAD_FAST 1 (b) STORE_FAST 0 (a) LOAD_CONST 0 (None) RETURN_VALUE LOAD_CONST 0 imm rax, 0 push rax RETURN_VALUE pop rax ret

Slide 100

Slide 100 text

push rsi pop rax mov rdi, rax def baz(a, b): a = b imm rax, 0 push rax pop rax ret

Slide 101

Slide 101 text

push rsi pop rax mov rdi, rax def baz(a, b): a = b imm rax, 0 push rax pop rax ret >>> baz(11, 22)

Slide 102

Slide 102 text

push rsi pop rax mov rdi, rax def baz(a, b): a = b imm rax, 0 push rax pop rax ret >>> baz(11, 22) rax rsi rdi return address

Slide 103

Slide 103 text

push rsi pop rax mov rdi, rax def baz(a, b): a = b imm rax, 0 push rax pop rax ret >>> baz(11, 22) rax rsi 22 rdi 11 return address

Slide 104

Slide 104 text

push rsi pop rax mov rdi, rax def baz(a, b): a = b imm rax, 0 push rax pop rax ret >>> baz(11, 22) rax rsi 22 rdi 11 return address

Slide 105

Slide 105 text

push rsi pop rax mov rdi, rax def baz(a, b): a = b imm rax, 0 push rax pop rax ret >>> baz(11, 22) rax rsi 22 rdi 11 return address

Slide 106

Slide 106 text

push rsi pop rax mov rdi, rax def baz(a, b): a = b imm rax, 0 push rax pop rax ret >>> baz(11, 22) rax rsi 22 rdi 11 return address 22

Slide 107

Slide 107 text

push rsi pop rax mov rdi, rax def baz(a, b): a = b imm rax, 0 push rax pop rax ret >>> baz(11, 22) rax rsi 22 rdi 11 return address 22

Slide 108

Slide 108 text

push rsi pop rax mov rdi, rax def baz(a, b): a = b imm rax, 0 push rax pop rax ret >>> baz(11, 22) rax 22 rsi 22 rdi 11 return address

Slide 109

Slide 109 text

push rsi pop rax mov rdi, rax def baz(a, b): a = b imm rax, 0 push rax pop rax ret >>> baz(11, 22) rax 22 rsi 22 rdi 11 return address

Slide 110

Slide 110 text

push rsi pop rax mov rdi, rax def baz(a, b): a = b imm rax, 0 push rax pop rax ret >>> baz(11, 22) rax 22 rsi 22 rdi 22 return address

Slide 111

Slide 111 text

push rsi pop rax mov rdi, rax def baz(a, b): a = b imm rax, 0 push rax pop rax ret >>> baz(11, 22) rax 22 rsi 22 rdi 22 return address

Slide 112

Slide 112 text

push rsi pop rax mov rdi, rax def baz(a, b): a = b imm rax, 0 push rax pop rax ret >>> baz(11, 22) rax 0 rsi 22 rdi 22 return address

Slide 113

Slide 113 text

push rsi pop rax mov rdi, rax def baz(a, b): a = b imm rax, 0 push rax pop rax ret >>> baz(11, 22) rax 0 rsi 22 rdi 22 return address

Slide 114

Slide 114 text

push rsi pop rax mov rdi, rax def baz(a, b): a = b imm rax, 0 push rax pop rax ret >>> baz(11, 22) rax 0 rsi 22 rdi 22 return address 0

Slide 115

Slide 115 text

push rsi pop rax mov rdi, rax def baz(a, b): a = b imm rax, 0 push rax pop rax ret >>> baz(11, 22) rax 0 rsi 22 rdi 22 return address 0

Slide 116

Slide 116 text

push rsi pop rax mov rdi, rax def baz(a, b): a = b imm rax, 0 push rax pop rax ret >>> baz(11, 22) rax 0 rsi 22 rdi 22 return address

Slide 117

Slide 117 text

push rsi pop rax mov rdi, rax def baz(a, b): a = b imm rax, 0 push rax pop rax ret >>> baz(11, 22) rax 0 rsi 22 rdi 22 return address

Slide 118

Slide 118 text

push rsi pop rax mov rdi, rax def baz(a, b): a = b imm rax, 0 push rax pop rax ret >>> baz(11, 22) rax 0 rsi 22 rdi 22

Slide 119

Slide 119 text

push rsi pop rax mov rdi, rax def baz(a, b): a = b imm rax, 0 push rax pop rax ret >>> baz(11, 22) rax 0 rsi 22 rdi 22

Slide 120

Slide 120 text

push rsi pop rax mov rdi, rax def baz(a, b): a = b imm rax, 0 push rax pop rax ret >>> baz(11, 22) rax 0 rsi 22 rdi 22 LOAD_FAST push STORE_FAST pop rax mov , rax LOAD_CONST imm rax push rax BINARY_MULTIPLY pop rax pop rbx imul rax, rbx push rax BINARY_ADD / INPLACE_ADD pop rax pop rbx add rax, rbx push rax BINARY_SUBTRACT / INPLACE_SUBTRACT pop rbx pop rax sub rax, rbx push rax UNARY_NEGATIVE pop rax neg rax push rax RETURN_VALUE pop rax ret

Slide 121

Slide 121 text

push rsi pop rax mov rdi, rax def baz(a, b): a = b imm rax, 0 push rax pop rax ret >>> baz(11, 22) rax 0 rsi 22 rdi 22 LOAD_FAST push STORE_FAST pop rax mov , rax LOAD_CONST imm rax push rax BINARY_MULTIPLY pop rax pop rbx imul rax, rbx push rax BINARY_ADD / INPLACE_ADD pop rax pop rbx add rax, rbx push rax BINARY_SUBTRACT / INPLACE_SUBTRACT pop rbx pop rax sub rax, rbx push rax UNARY_NEGATIVE pop rax neg rax push rax RETURN_VALUE pop rax ret

Slide 122

Slide 122 text

push rsi pop rax mov rdi, rax def baz(a, b): a = b imm rax, 0 push rax pop rax ret >>> baz(11, 22) rax 0 rsi 22 rdi 22 LOAD_FAST push STORE_FAST pop rax mov , rax LOAD_CONST imm rax push rax BINARY_MULTIPLY pop rax pop rbx imul rax, rbx push rax BINARY_ADD / INPLACE_ADD pop rax pop rbx add rax, rbx push rax BINARY_SUBTRACT / INPLACE_SUBTRACT pop rbx pop rax sub rax, rbx push rax UNARY_NEGATIVE pop rax neg rax push rax RETURN_VALUE pop rax ret

Slide 123

Slide 123 text

push rsi pop rax mov rdi, rax def baz(a, b): a = b imm rax, 0 push rax pop rax ret >>> baz(11, 22) rax 0 rsi 22 rdi 22 LOAD_FAST push STORE_FAST pop rax mov , rax LOAD_CONST imm rax push rax BINARY_MULTIPLY pop rax pop rbx imul rax, rbx push rax BINARY_ADD / INPLACE_ADD pop rax pop rbx add rax, rbx push rax BINARY_SUBTRACT / INPLACE_SUBTRACT pop rbx pop rax sub rax, rbx push rax UNARY_NEGATIVE pop rax neg rax push rax RETURN_VALUE pop rax ret

Slide 124

Slide 124 text

push rsi pop rax mov rdi, rax def baz(a, b): a = b imm rax, 0 push rax pop rax ret >>> baz(11, 22) rax 0 rsi 22 rdi 22 LOAD_FAST push STORE_FAST pop rax mov , rax LOAD_CONST imm rax push rax BINARY_MULTIPLY pop rax pop rbx imul rax, rbx push rax BINARY_ADD / INPLACE_ADD pop rax pop rbx add rax, rbx push rax BINARY_SUBTRACT / INPLACE_SUBTRACT pop rbx pop rax sub rax, rbx push rax UNARY_NEGATIVE pop rax neg rax push rax RETURN_VALUE pop rax ret

Slide 125

Slide 125 text

push rsi pop rax mov rdi, rax def baz(a, b): a = b imm rax, 0 push rax pop rax ret >>> baz(11, 22) rax 0 rsi 22 rdi 22 LOAD_FAST push STORE_FAST pop rax mov , rax LOAD_CONST imm rax push rax BINARY_MULTIPLY pop rax pop rbx imul rax, rbx push rax BINARY_ADD / INPLACE_ADD pop rax pop rbx add rax, rbx push rax BINARY_SUBTRACT / INPLACE_SUBTRACT pop rbx pop rax sub rax, rbx push rax UNARY_NEGATIVE pop rax neg rax push rax RETURN_VALUE pop rax ret

Slide 126

Slide 126 text

push rsi pop rax mov rdi, rax def baz(a, b): a = b imm rax, 0 push rax pop rax ret >>> baz(11, 22) rax 0 rsi 22 rdi 22 LOAD_FAST push STORE_FAST pop rax mov , rax LOAD_CONST imm rax push rax BINARY_MULTIPLY pop rax pop rbx imul rax, rbx push rax BINARY_ADD / INPLACE_ADD pop rax pop rbx add rax, rbx push rax BINARY_SUBTRACT / INPLACE_SUBTRACT pop rbx pop rax sub rax, rbx push rax UNARY_NEGATIVE pop rax neg rax push rax RETURN_VALUE pop rax ret

Slide 127

Slide 127 text

push rsi pop rax mov rdi, rax def baz(a, b): a = b imm rax, 0 push rax pop rax ret >>> baz(11, 22) rax 0 rsi 22 rdi 22 LOAD_FAST push STORE_FAST pop rax mov , rax LOAD_CONST imm rax push rax BINARY_MULTIPLY pop rax pop rbx imul rax, rbx push rax BINARY_ADD / INPLACE_ADD pop rax pop rbx add rax, rbx push rax BINARY_SUBTRACT / INPLACE_SUBTRACT pop rbx pop rax sub rax, rbx push rax UNARY_NEGATIVE pop rax neg rax push rax RETURN_VALUE pop rax ret

Slide 128

Slide 128 text

push rsi pop rax mov rdi, rax def baz(a, b): a = b imm rax, 0 push rax pop rax ret >>> baz(11, 22) rax 0 rsi 22 rdi 22 LOAD_FAST push STORE_FAST pop rax mov , rax LOAD_CONST imm rax push rax BINARY_MULTIPLY pop rax pop rbx imul rax, rbx push rax BINARY_ADD / INPLACE_ADD pop rax pop rbx add rax, rbx push rax BINARY_SUBTRACT / INPLACE_SUBTRACT pop rbx pop rax sub rax, rbx push rax UNARY_NEGATIVE pop rax neg rax push rax RETURN_VALUE pop rax ret

Slide 129

Slide 129 text

push rsi pop rax mov rdi, rax def baz(a, b): a = b imm rax, 0 push rax pop rax ret >>> baz(11, 22) rax 0 rsi 22 rdi 22

Slide 130

Slide 130 text

push rsi pop rax mov rdi, rax imm rax, 0 push rax pop rax ret

Slide 131

Slide 131 text

push rsi pop rax mov rdi, rax imm rax, 0 push rax pop rax ret

Slide 132

Slide 132 text

push rsi pop rax mov rdi, rax imm rax, 0 push rax pop rax ret

Slide 133

Slide 133 text

push rsi pop rax mov rdi, rax imm rax, 0 push rax pop rax ret

Slide 134

Slide 134 text

push rsi pop rax mov rdi, rax imm rax, 0 push rax pop rax ret mov rax, rsi

Slide 135

Slide 135 text

push rsi pop rax mov rdi, rax imm rax, 0 push rax pop rax ret mov rax, rsi

Slide 136

Slide 136 text

push rsi pop rax mov rdi, rax imm rax, 0 push rax pop rax ret mov rax, rsi Peephole Optimization

Slide 137

Slide 137 text

push rsi pop rax mov rdi, rax imm rax, 0 push rax pop rax ret mov rax, rsi Peephole Optimization

Slide 138

Slide 138 text

push rsi pop rax mov rdi, rax imm rax, 0 push rax pop rax ret mov rax, rsi mov rdi, rax Peephole Optimization

Slide 139

Slide 139 text

push rsi pop rax mov rdi, rax imm rax, 0 push rax pop rax ret mov rax, rsi mov rdi, rax Peephole Optimization

Slide 140

Slide 140 text

push rsi pop rax mov rdi, rax imm rax, 0 push rax pop rax ret mov rax, rsi mov rdi, rax imm rax, 0 Peephole Optimization

Slide 141

Slide 141 text

push rsi pop rax mov rdi, rax imm rax, 0 push rax pop rax ret mov rax, rsi mov rdi, rax imm rax, 0 Peephole Optimization

Slide 142

Slide 142 text

push rsi pop rax mov rdi, rax imm rax, 0 push rax pop rax ret mov rax, rsi mov rdi, rax imm rax, 0 Peephole Optimization

Slide 143

Slide 143 text

push rsi pop rax mov rdi, rax imm rax, 0 push rax pop rax ret mov rax, rsi mov rdi, rax imm rax, 0 ret Peephole Optimization

Slide 144

Slide 144 text

mov rax, rsi mov rdi, rax imm rax, 0 ret Peephole Optimization

Slide 145

Slide 145 text

mov rdi, rsi imm rax, 0 ret Peephole Optimization

Slide 146

Slide 146 text

mov rdi, rsi imm rax, 0 ret Peephole Optimization

Slide 147

Slide 147 text

mov rdi, rsi imm rax, 0 ret Peephole Optimization

Slide 148

Slide 148 text

No content

Slide 149

Slide 149 text

class Assembler(object): # … def emit(self, *args): for code in args: self.block[self.index] = code self.index += 1 Machine Code Generation

Slide 150

Slide 150 text

class Assembler(object): # … def emit(self, *args): for code in args: self.block[self.index] = code self.index += 1 Machine Code Generation def push(self, a, dummy): self.emit(0x50 | self.registers(a))

Slide 151

Slide 151 text

class Assembler(object): # … def emit(self, *args): for code in args: self.block[self.index] = code self.index += 1 Machine Code Generation def push(self, a, dummy): self.emit(0x50 | self.registers(a))

Slide 152

Slide 152 text

class Assembler(object): # … def emit(self, *args): for code in args: self.block[self.index] = code self.index += 1 Machine Code Generation def push(self, a, dummy): self.emit(0x50 | self.registers(a)) def registers(self, a): order = (“rax”, “rcx”, …) return order.index(a)

Slide 153

Slide 153 text

class Assembler(object): # … def emit(self, *args): for code in args: self.block[self.index] = code self.index += 1 Machine Code Generation def push(self, a, dummy): self.emit(0x50 | self.registers(a)) def registers(self, a): order = (“rax”, “rcx”, …) return order.index(a)

Slide 154

Slide 154 text

Machine Code Generation

Slide 155

Slide 155 text

def add(self, a, b): self.emit(0x48, 0x01, 0xc0 | self.registers(a, b)) def immediate(self, a, constant): self.emit(0x48, 0xb8 | self.registers(a), *self.little_endian(constant)) Machine Code Generation

Slide 156

Slide 156 text

def add(self, a, b): self.emit(0x48, 0x01, 0xc0 | self.registers(a, b)) def immediate(self, a, constant): self.emit(0x48, 0xb8 | self.registers(a), *self.little_endian(constant)) Machine Code Generation

Slide 157

Slide 157 text

def add(self, a, b): self.emit(0x48, 0x01, 0xc0 | self.registers(a, b)) def immediate(self, a, constant): self.emit(0x48, 0xb8 | self.registers(a), *self.little_endian(constant)) Machine Code Generation

Slide 158

Slide 158 text

def add(self, a, b): self.emit(0x48, 0x01, 0xc0 | self.registers(a, b)) def immediate(self, a, constant): self.emit(0x48, 0xb8 | self.registers(a), *self.little_endian(constant)) Machine Code Generation

Slide 159

Slide 159 text

def add(self, a, b): self.emit(0x48, 0x01, 0xc0 | self.registers(a, b)) def immediate(self, a, constant): self.emit(0x48, 0xb8 | self.registers(a), *self.little_endian(constant)) Machine Code Generation

Slide 160

Slide 160 text

Machine Code Generation

Slide 161

Slide 161 text

def ret(self, a, b): self.emit(0xc3) def push(self, a, _): self.emit(0x50 | self.registers(a)) def pop(self, a, _): self.emit(0x58 | self.registers(a)) def imul(self, a, b): self.emit(0x48, 0x0f, 0xaf, 0xc0 | self.registers(a, b)) def add(self, a, b): self.emit(0x48, 0x01, 0xc0 | self.registers(b, a)) def sub(self, a, b): self.emit(0x48, 0x29, 0xc0 | self.registers(b, a)) def neg(self, a, _): self.emit(0x48, 0xf7, 0xd8 | self.register(a)) def mov(self, a, b): self.emit(0x48, 0x89, 0xc0 | self.registers(b, a)) def immediate(self, a, number): self.emit(0x48, 0xb8 | self.registers(a), *self.little_endian(number)) Machine Code Generation

Slide 162

Slide 162 text

No content

Slide 163

Slide 163 text

Memory Management

Slide 164

Slide 164 text

Memory Management • Allocate one page of memory using mmap

Slide 165

Slide 165 text

Memory Management • Allocate one page of memory using mmap • Use mprotect to change from R+W to R+X

Slide 166

Slide 166 text

Memory Management

Slide 167

Slide 167 text

Memory Management def create_block(size): ptr = mmap(0, size, MMAP.PROT_WRIT | MMAP.PROT_READ | MMAP.MAP_ANONYMOUS, 0, 0) if ptr == MAP_FAILED: raise RuntimeError(…) return ptr

Slide 168

Slide 168 text

Memory Management def create_block(size): ptr = mmap(0, size, MMAP.PROT_WRIT | MMAP.PROT_READ | MMAP.MAP_ANONYMOUS, 0, 0) if ptr == MAP_FAILED: raise RuntimeError(…) return ptr

Slide 169

Slide 169 text

Memory Management def create_block(size): ptr = mmap(0, size, MMAP.PROT_WRIT | MMAP.PROT_READ | MMAP.MAP_ANONYMOUS, 0, 0) if ptr == MAP_FAILED: raise RuntimeError(…) return ptr

Slide 170

Slide 170 text

Memory Management def create_block(size): ptr = mmap(0, size, MMAP.PROT_WRIT | MMAP.PROT_READ | MMAP.MAP_ANONYMOUS, 0, 0) if ptr == MAP_FAILED: raise RuntimeError(…) return ptr

Slide 171

Slide 171 text

Memory Management

Slide 172

Slide 172 text

Memory Management def make_executable(block, size): if mprotect(block, size, MMAP.PROT_READ | MMAP.PROT_EXEC) != 0: raise RuntimeError(…)

Slide 173

Slide 173 text

Memory Management def make_executable(block, size): if mprotect(block, size, MMAP.PROT_READ | MMAP.PROT_EXEC) != 0: raise RuntimeError(…)

Slide 174

Slide 174 text

No content

Slide 175

Slide 175 text

FFI

Slide 176

Slide 176 text

FFI • Use ctypes as FFI

Slide 177

Slide 177 text

FFI • Use ctypes as FFI • Grab machine code address, create signature and bind to function.

Slide 178

Slide 178 text

FFI • Use ctypes as FFI • Grab machine code address, create signature and bind to function. • Ctypes also seem to save and restore registers

Slide 179

Slide 179 text

FFI • Use ctypes as FFI • Grab machine code address, create signature and bind to function. • Ctypes also seem to save and restore registers • More overhead: Marshalling arguments

Slide 180

Slide 180 text

FFI

Slide 181

Slide 181 text

FFI signature = ctypes.CFUNCTYPE([ctypes.c_int64, ctypes.c_int64]) signature.restype = ctypes.c_int64 func = signature(assembler.address) print(func(1, 2))

Slide 182

Slide 182 text

FFI signature = ctypes.CFUNCTYPE([ctypes.c_int64, ctypes.c_int64]) signature.restype = ctypes.c_int64 func = signature(assembler.address) print(func(1, 2))

Slide 183

Slide 183 text

FFI signature = ctypes.CFUNCTYPE([ctypes.c_int64, ctypes.c_int64]) signature.restype = ctypes.c_int64 func = signature(assembler.address) print(func(1, 2))

Slide 184

Slide 184 text

FFI signature = ctypes.CFUNCTYPE([ctypes.c_int64, ctypes.c_int64]) signature.restype = ctypes.c_int64 func = signature(assembler.address) print(func(1, 2))

Slide 185

Slide 185 text

FFI signature = ctypes.CFUNCTYPE([ctypes.c_int64, ctypes.c_int64]) signature.restype = ctypes.c_int64 func = signature(assembler.address) print(func(1, 2))

Slide 186

Slide 186 text

No content

Slide 187

Slide 187 text

The “@jit” Decorator def jit(func): def front(*args, **kw): if not hasattr(front, “func”): front.func = compile_native(func) return front.func(*args, **kw) return front @jit def foo(a, b): return a*a - b*b

Slide 188

Slide 188 text

The “@jit” Decorator def jit(func): def front(*args, **kw): if not hasattr(front, “func”): front.func = compile_native(func) return front.func(*args, **kw) return front @jit def foo(a, b): return a*a - b*b

Slide 189

Slide 189 text

Demo

Slide 190

Slide 190 text

Things we haven’t touched on

Slide 191

Slide 191 text

Things we haven’t touched on • Code generation: SSA, branching

Slide 192

Slide 192 text

Things we haven’t touched on • Code generation: SSA, branching • Optimization: TAC, register allocation

Slide 193

Slide 193 text

Things we haven’t touched on • Code generation: SSA, branching • Optimization: TAC, register allocation • Compatibility: Python object system

Slide 194

Slide 194 text

Things we haven’t touched on • Code generation: SSA, branching • Optimization: TAC, register allocation • Compatibility: Python object system • Performance gotchas

Slide 195

Slide 195 text

Pointers • Full details on blog post at
 https://csl.name/post/python-compiler/ • PeachPy • Truffle + Graal, LLVM, libjit, etc. • NUMBA

Slide 196

Slide 196 text

Thanks y’all!