SLIDE 1
FROM SYSTEM F TO TYPED ASSEMBLY LANGUAGE
Greg Morrisett, David Walker, Karl Crary & Neal Glew TOPLAS 1999
Presentation by: Drew Zagieboylo/Matthew Milano
TYPED ASSEMBLY LANGUAGE
FROM SYSTEM F TO TYPED ASSEMBLY LANGUAGE Greg Morrisett, David - - PowerPoint PPT Presentation
FROM SYSTEM F TO TYPED ASSEMBLY LANGUAGE Greg Morrisett, David - - PowerPoint PPT Presentation
FROM SYSTEM F TO TYPED ASSEMBLY LANGUAGE Greg Morrisett, David Walker, Karl Crary & Neal Glew TOPLAS 1999 Presentation by: Drew Zagieboylo/Matthew Milano TYPED ASSEMBLY LANGUAGE TYPED ASSEMBLY LANGUAGE TYPED ASSEMBLY LANGUAGE TYPED
SLIDE 2
SLIDE 3
TYPED ASSEMBLY LANGUAGE
SLIDE 4
TYPED ASSEMBLY LANGUAGE
SLIDE 5
TYPED ASSEMBLY LANGUAGE
SLIDE 6
WHY DO WE WANT TAL?
SLIDE 7
TYPE SYSTEMS ALL THE WAY!!
SLIDE 8
x86
NO TYPES :(
TYPED INTERMEDIATE LANGUAGES
➤ TIL ➤ Throughout the 90’s (and today!) ➤ Benefits of Types (efficiency + soundness) ➤ Target Language is Untyped ML TIL …
TYPES!
SLIDE 9
HOW TO GUARANTEE SAFETY W/ UNTYPED AND UNTRUSTED CODE?
SLIDE 10
PROOF-CARRYING CODE
➤ George Necula (POPL ’97) ➤ Compiler Produces:
- 1. Program
- 2. Proof
➤ First-Order Predicate
Logic Based
➤ Difficult to Build
Compilers
SLIDE 11
TYPED ASSEMBLY LANGUAGE
➤ Extend benefits of types all the way to the target ➤ Types as implementation of Proof-Carrying Code
SLIDE 12
TYPED ASSEMBLY LANGUAGE - FEATURES
➤ RISC-style language ➤ Types: ➤ Code types ➤ Pointer Types ➤ Existential Type Constructor ➤ Security: ➤ No pointer forging! ➤ Control Flow Integrity ➤ Other: ➤ Memory Allocation
SLIDE 13
SYSTEM F TO TAL
➤ Show that TAL is expressive
SLIDE 14
SYSTEM F TO TAL
➤ CPS Conversion
SLIDE 15
CPS TRANSLATION
➤ Continuation Passing Style ➤ Translate to near-linear series of let bindings & calls ➤ Removes function call stack
Abstraction Translation Application Translation
SLIDE 16
SYSTEM F TO
➤ Continuation Passing Style
λK (fix f(n : int) : int . if0 (n,1,n × f(n − 1))) 6 λF (fix f(n : int, k : (int) → void) . λK if0(n, k(1), (6,λ(n : int) . halt[int]n) f(x, λ(y : int) . let z = n × y in k(z)))) let x = n − 1in
SLIDE 17
SYSTEM F TO TAL
➤ Closure Conversion
SLIDE 18
POLYMORPHIC CLOSURE CONVERSION
➤ Generate Explicit Closures ➤ Implements Encapsulation ➤ New Syntax ➤ Existential Types ➤ Packing/Unpacking
➤ Uses Type Erasure* ➤ Function bodies type-check w/o environment type info ➤ Pack is a no-op at runtime
τ, σ ::= . . . |∃α . τ u ::= . . . |v[τ]| pack[τ1, v] as τ2 d ::= . . . |[α, x] = unpack v
SLIDE 19
TO
➤ Polymorphic Closure Conversion
λK λC
Function Type Translation Application Translation
SLIDE 20
SYSTEM F TO TAL
➤ Hoisting
SLIDE 21
HOISTING
➤ Separating Code Definition & Program ➤ Much like real memory layout ➤ Closures make this easy! ➤ Bind fix statements to variables, pointing to code
SLIDE 22
TO
➤ Polymorphic Closure Conversion ➤ Factorial(6)
λK λC
SLIDE 23
SYSTEM F TO TAL
➤ Memory Allocation
SLIDE 24
ALLOCATION
➤ Assembly language doesn’t have Tuples! ➤ Need to allocate memory for tuples (and initialize!)
➤ x = (v1, v2)
A[[⟨τ1, . . . , τn⟩]] ≜ ⟨A[[τ1]]1, . . . , A[[τn]]1⟩
SLIDE 25
ALLOCATION
λH λA
SLIDE 26
SYSTEM F TO TAL
➤ Code Generation
SLIDE 27
SYSTEM F TO TAL
➤ Code Generation ➤ Mostly direct translation to assembly ➤ Function types annotate registers
➤ unpack is just a mov instruction w/ type erasure ➤ malloc is abstract
SLIDE 28
TAL IMPLEMENTATION
➤ TALx86 : IA32 ISA ➤ Variation from Paper: ➤ Other data types (arrays, floats, etc.) ➤ Not CPS -> Uses Explicit Stack ➤ Implements malloc and unpack instructions ➤ Modules with Type Interfaces ➤ Some optimizations ➤ Register-sized objects vs. “large objects” ➤ Cross-module optimization
SLIDE 29
CONCLUSIONS
➤ System F -> TAL ➤ We can have security and expressivity ➤ Utilizes many PL techniques ➤ Type-directed Compilation ➤ Formalism omits many optimizations (other work) ➤ Future Work & Impact ➤ Cyclone (low level, typed language) ➤ (and then Rust)
SLIDE 30
THANK YOU!
SLIDE 31
POLYMORPHIC CC - TWICE EXAMPLE
SLIDE 32
POLYMORPHIC CC - TWICE EXAMPLE