Linear Regions Are All You Need - PowerPoint PPT Presentation

About This Presentation
Title:

Linear Regions Are All You Need

Description:

Objects are individually allocated in a region. constant-time operation ... low-level interface with manifest cost model. range of memory management options ... – PowerPoint PPT presentation

Number of Views:50
Avg rating:3.0/5.0
Slides: 76
Provided by: csCor
Category:

less

Transcript and Presenter's Notes

Title: Linear Regions Are All You Need


1
Linear Regions Are All You Need
  • Matthew Fluet
  • Cornell University
  • Greg Morrisett Amal Ahmed
  • Harvard University

2
Memory Management
  • Dynamic allocation pervasive in computation

3
Memory Management
  • Dynamic allocation pervasive in computation
  • Region-based Memory Management
  • Memory is divided into regions
  • Objects are individually allocated in a region
  • constant-time operation
  • All objects in a region are deallocated together
  • constant-time operation

4
Application Cyclone
  • Cyclone Safe-C Project
  • type-safety
  • with the virtues of C
  • low-level interface with manifest cost model
  • range of memory management options
  • regions are an organizing principle

5
Cyclone Regions
6
Application Cyclone
  • MediaNET
  • TCP benchmark (packet forwarding)
  • Cyclone v.0.1 (lexical regions garbage
    collector)
  • High water mark 840 KB
  • 130 collections
  • Basic throughput 50 MB/s
  • Cyclone v.0.5 (unique pointers dynamic regions)
  • High water mark 8 KB
  • 0 collections
  • Basic throughput 74MB/s

7
Application Cyclone
  • MediaNET
  • TCP benchmark (packet forwarding)
  • Cyclone v.0.1 (lexical regions garbage
    collector)
  • High water mark 840 KB
  • 130 collections
  • Basic throughput 50 MB/s
  • Cyclone v.0.5 (unique pointers dynamic regions)
  • High water mark 8 KB
  • 0 collections
  • Basic throughput 74MB/s

8
Cyclone Regions
Proving type safety of Cyclone is a nightmare!!
9
Cyclone Regions
Goal simple model where we can easily encode the
key features of Cyclone in a target language
witha simpler type system.
10
Cyclone Regions
Linear RegionsAre All You Need
11
Outline
  • Introduction
  • Monadic Type System (FRGN) ICFP04
  • Substructural Type System (lrgnUL)
  • Translation Sketch
  • Conclusion

12
Monadic Type System for Regions ICFP04
  • Extend the runST trick to nested regions L-PJ
    94
  • Polymorphic type system ensures safety
  • Key insights (FRGN)
  • Effects map to an indexed monadic type
  • Region subtyping witnessed by types
  • Sufficient for encoding Tofte-Talpin region
    calculusand core Cyclone region features

13
RGN monad Types
  • Monadic type
  • RGN s t
  • computations in stack of regions s returning
    values of type t a stack transformer

14
RGN monad Operations
  • Monadic unit and bind
  • returnRGN
  • 8s,a. a ! RGN s a
  • thenRGN
  • 8s,a,b. RGN s a ! (a ! RGN s b) ! RGN s b

15
RGN monad Operations
  • Monadic unit and bind
  • returnRGN
  • 8s,a. a ! RGN s a
  • thenRGN
  • 8s,a,b. RGN s a ! (a ! RGN s b) ! RGN s b

16
RGN monad Operations
  • Monadic unit and bind
  • returnRGN
  • 8s,a. a ! RGN s a
  • thenRGN
  • 8s,a,b. RGN s a ! (a ! RGN s b) ! RGN s b

17
RGN monad Types
  • Reference type
  • Ref s t
  • values of type t allocated in region
  • at the top of the stack of regions s

18
RGN monad Operations
  • Create and read region allocated values
  • new
  • 8s,a. a ! RGN s (Ref s a)
  • read
  • 8s,a. Ref s a ! RGN s a

19
RGN monad Operations
  • Create and read region allocated values
  • new
  • 8s,a. a ! RGN s (Ref s a)
  • read
  • 8s,a. Ref s a ! RGN s a

20
RGN monad Encapsulation
  • Encapsulate and run a monadic computation
  • runRGN
  • 8a. (8s. RGN s a) ! a

21
RGN monad Encapsulation
  • Encapsulate and run a monadic computation
  • runRGN
  • 8a. (8s. RGN s a) ! a

22
RGN monad Encapsulation
  • Encapsulate and run a monadic computation
  • runRGN
  • 8a. (8s. RGN s a) ! a

for all stacks ) no assumptions about stack of
regions
23
RGN monad Encapsulation
  • Encapsulate and run a monadic computation
  • runRGN
  • 8a. (8s. RGN s a) ! a

for all stacks ) no assumptions about stack of
regions
24
RGN monad Encapsulation
  • Encapsulate and run a monadic computation
  • runRGN
  • 8a. (8s. RGN s a) ! a

result is independent of stack ) s 62 frv(a)
)region values dont escape
for all stacks ) no assumptions about stack of
regions
25
RGN monad Example
runRGN ( Ls1. do a à new s1 1 c à runRGN (
Ls2. do b à new s2 7 z new
s1 z ) c )
r1
26
RGN monad Example
runRGN ( Ls1. do a à new s1 1 c à runRGN (
Ls2. do b à new s2 7 z new
s1 z ) c )
r1
a 1
inputallocated in first region
27
RGN monad Example
runRGN ( Ls1. do a à new s1 1 c à runRGN (
Ls2. do b à new s2 7 z new
s1 z ) c )
r2
r1
a 1
input allocated in first region
28
RGN monad Example
temporary allocated in second region
runRGN ( Ls1. do a à new s1 1 c à runRGN (
Ls2. do b à new s2 7 z new
s1 z ) c )
r2
b 7
r1
a 1
inputallocated in first region
29
RGN monad Example
temporary allocated in second region
runRGN ( Ls1. do a à new s1 1 c à runRGN (
Ls2. do b à new s2 7 z new
s1 z ) c )
r2
b 7
r1
a 1
input and outputallocated in first region
c 8
30
RGN monad Example
temporary allocated in second region
runRGN ( Ls1. do a à new s1 1 c à runRGN (
Ls2. do b à new s2 7 z new
s1 z ) c )
r1
a 1
input and outputallocated in first region
c 8
31
RGN monad Example
runRGN ( Ls1. do a à new s1 1 c à runRGN (
Ls2. do b à new s2 7 z new
s1 z ) c )
allocating in younger region requires RGN s2 t
type
allocating in older region requires RGN s1 t type
32
RGN monad Witnesses
  • Witness type
  • Pf(s1 s2)
  • type-level proof that the stack of regions s1
  • is a substack of the stack of regions s2

33
RGN monad Witnesses
  • Witness operations
  • coerceRGN
  • 8s1,s2,a. Pf(s1 s2) ! RGN s1 a ! RGN s2 a
  • transSub
  • 8s1,s2,s3. Pf(s1 s2) ! Pf(s2 s3) ! Pf(s1
    s3)

34
RGN monad Regions
  • Regions are created and destroyedwith a
    lexically scoped construct
  • letRGN
  • 8s1,a. (8s2. Pf(s1 s2) ! RGN s2 a) ! RGN s1 a

35
RGN monad Regions
  • Regions are created and destroyedwith a
    lexically scoped construct
  • letRGN
  • 8s1,a. (8s2. Pf(s1 s2) ! RGN s2 a) ! RGN s1 a

36
RGN monad Example
letRGN ( Ls1. lpf1. do a à new s1 1 c Ã
letRGN ( Ls2. lpf2. do b à new s2 7
z coerceRgn pf (new s1 z )) c )
r2
b 7
r1
a 1
c 8
37
Limitations of LIFO Regions
  • Lexical scope is ill-suited for
  • iterative computations
  • Conways Game of Life copying GC
  • CPS-based computations
  • event-based computations

38
Limitations of LIFO Regions
  • Lexical scope is ill-suited for
  • iterative computations
  • Conways Game of Life copying GC
  • CPS-based computations
  • event-based computations
  • But, lexical scope was ensuring that the stack
    of regions was used in a single-threaded manner

39
Substructural Type Systems
  • Provide core mechanisms to restrict the number
    and order of uses of data and operations
  • generalization of linear type systems

40
Substructural Type System lUL
  • Qualifiers
  • q U j L
  • PreTypes
  • t 1 j t1 t2 j t1 ! t2 j 8a.t j 9a.t
  • Types
  • t qt

41
Substructural Type System lUL
  • Qualifiers
  • q U j L
  • PreTypes
  • t 1 j t1 t2 j t1 ! t2 j 8a.t j 9a.t
  • Types
  • t qt

How maythe value be used?
42
Substructural Type System lUL
  • Qualifiers
  • q U j L
  • PreTypes
  • t 1 j t1 t2 j t1 ! t2 j 8a.t j 9a.t
  • Types
  • t qt

How often maythe value be used?
How maythe value be used?
43
Substructural Qualifiers
Linear
must be used exactly once
Unrestricted Drop Copy
may be used an arbitrary of times
44
Substructural Type System for Regions
  • Provide core mechanisms to restrict the number
    and order of uses of data and operations
  • generalization of linear type systems
  • Key insights (lrgnUL)
  • Separate region names from region liveness
  • Region liveness witnessed by types
  • Sufficient for encoding FRGN calculusand
    advanced Cyclone region features

45
lrgnUL lUL Regions
  • PreTypes
  • t j cap r j ref r t j 8r.t j 9r.t

capability for region rmediates all access to
a regionfor allocating, reading, and writing
46
lrgnUL Region Primitives
  • Regions are created and destroyedwith separate
    operations
  • newrgn
  • U1 ! (9r. Lcap r)
  • freergn
  • 8r. (Lcap r ! U1)

47
lrgnUL Region Primitives
  • Regions are created and destroyedwith separate
    operations
  • newrgn
  • U1 ! (9r. Lcap r)
  • freergn
  • 8r. (Lcap r ! U1)

Produces a capability.
Consumes a capability.
48
lrgnUL Region Primitives
  • Regions are created and destroyedwith separate
    operations
  • newrgn
  • U1 ! (9r. Lcap r)
  • freergn
  • 8r. (Lcap r ! U1)

49
lrgnUL Region Primitives
  • new
  • 8r,a. ((Lcap r Ua) ! (Lcap r Uref r Ua))
  • read
  • 8r,a. ((Lcap r Uref r Ua) !
  • (Lcap r Ua))

50
lrgnUL Region Primitives
  • new
  • 8r,a. ((Lcap r Ua) ! (Lcap r Uref r Ua))
  • read
  • 8r,a. ((Lcap r Uref r Ua) !
  • (Lcap r Ua))

Requires a capability.
Returns a capability.
51
lrgnUL Region Primitives
  • new
  • 8r,a. ((Lcap r Ua) ! (Lcap r Uref r Ua)
  • read
  • 8r,a. ((Lcap r Uref r Ua) !
  • (Lcap r Ua)

52
Translation FRGN to lrgnUL, Types
  • RGN s t U(s ! L(s t))

53
Translation FRGN to lrgnUL, Types
  • RGN s t s ! (s t)
  • operational behavior of monad is
    store/stack-passing

54
Translation FRGN to lrgnUL, Types
  • RGN s t s ! (s t)
  • operational behavior of monad is
    store/stack-passing

55
Translation FRGN to lrgnUL, Types
  • RGN s t s ! (s t)
  • operational behavior of monad is
    store/stack-passing
  • represent stack of regionsas a sequence of
    linear capabilities,formed out of nested linear
    tuples

56
Translation FRGN to lrgnUL, Types
  • RGN s t s ! (s t)
  • operational behavior of monad is
    store/stack-passing
  • represent stack of regionsas a sequence of
    linear capabilities,formed out of nested linear
    tuples

57
Translation FRGN to lrgnUL, Ops
  • returnRGN s t e let res t e
    in Ulstks. Lhstk,resi
  • thenRGN s ta tb e1 e2 let f RGN
    s ta e1 in let g ta ! RGN s tb e2
    in Ulstks. let hstk,resi f stk in g res stk

58
Translation FRGN to lrgnUL, Ops
  • returnRGN s t e let res t e
    in Ulstks. Lhstk,resi
  • thenRGN s ta tb e1 e2 let f RGN
    s ta e1 in let g ta ! RGN s tb e2
    in Ulstks. let hstk,resi f stk in g res stk

Store-passing encoding
59
Translation FRGN to lrgnUL, Types
  • Pf(s1 s2) U(9s. Iso(s2, L(s1 s)))

60
Translation FRGN to lrgnUL, Types
  • Pf(s1 s2) U(9s. Iso(s2, L(s1 s)))
  • Isomorphism between s2 and L(s1 s), for some
    slack s

61
Translation FRGN to lrgnUL, Types
  • Pf(s1 s2) U(9s. Iso(s2, L(s1 s)))
  • Isomorphism between s2 and L(s1 s), for some
    slack s
  • Proof that s1 is a substack of s2 is persistent
  • Liveness of s1 and s2 is ephemeral

62
Translation FRGN to lrgnUL, Types
  • Ref s t U(9r. U(U(9s. Iso(s, L(s Lcap
    r))) Ref s t U(9r. U( Uref r t))

63
Translation FRGN to lrgnUL, Types
  • Ref s t 9r. (9s. Iso(s, L(s Lcap r))
    Ref s t 9r. ( Uref r t)
  • Existential fixes region r

64
Translation FRGN to lrgnUL, Types
  • Ref s t 9r. (9s. Iso(s, L(s Lcap r))
    Ref s t 9r. ( Uref r t)
  • Existential fixes region r
  • Isomorphism witnesses membership of r in s

65
Translation FRGN to lrgnUL, Ops
  • letRGN s1 t e let f 8s2.
    Pf(s1s2) ! Hnd s2 ! RGN s2 t e
    in Ulstk1s1.let pack(r,hcap,hndi) newrgn Lhi
    in Ulstk1s1.let stk2 Lhstk1,capi
    in Ulstk1s1.let id Ulstk L(s1 Lcap r).stk
    in Ulstk1s1.let pwit Upack(Lcap r,Uhid,idi)
    in Ulstk1s1.let phnd Upack(r,UhUpack(s1,Uhid,i
    di),hndi) in Ulstk1s1.let hstk2,resi f L(s1
    Lcap r) pwit phnd stk2 in Ulstk1s1.let
    hstk1,capi stk2 in Ulstk1s1.let hi freergn
    r Lhcap,hndi in Ulstk1s1.Lhstk1,resi

66
Translation FRGN to lrgnUL, Ops
Stack-passing encoding
  • letRGN s1 t e let f 8s2.
    Pf(s1s2) ! RGN s2 t e in Ulstk1s1.let
    pack(r,cap) newrgn Uhi in Ulstk1s1.let stk2
    Lhstk1,capi in Ulstk1s1.let id Ulstk L(s1
    Lcap r).stk in Ulstk1s1.let pwit Upack(Lcap
    r,Uhid,idi) in Ulstk1s1.let hstk2,resi f
    L(s1 Lcap r) pwit stk2 in Ulstk1s1.let
    hstk1,capi stk2 in Ulstk1s1.let hi freergn
    r cap in Ulstk1s1.Lhstk1,resi

67
Translation FRGN to lrgnUL, Ops
Create destroy a new region
  • letRGN s1 t e let f 8s2.
    Pf(s1s2) ! RGN s2 t e in Ulstk1s1.let
    pack(r,cap) newrgn Uhi in Ulstk1s1.let stk2
    Lhstk1,capi in Ulstk1s1.let id Ulstk L(s1
    Lcap r).stk in Ulstk1s1.let pwit Upack(Lcap
    r,Uhid,idi) in Ulstk1s1.let hstk2,resi f
    L(s1 Lcap r) pwit stk2 in Ulstk1s1.let
    hstk1,capi stk2 in Ulstk1s1.let hi freergn
    r cap in Ulstk1s1.Lhstk1,resi

68
Translation FRGN to lrgnUL, Ops
Construct rep.of new stack
  • letRGN s1 t e let f 8s2.
    Pf(s1s2) ! RGN s2 t e in Ulstk1s1.let
    pack(r,cap) newrgn Uhi in Ulstk1s1.let stk2
    Lhstk1,capi in Ulstk1s1.let id Ulstk L(s1
    Lcap r).stk in Ulstk1s1.let pwit Upack(Lcap
    r,Uhid,idi) in Ulstk1s1.let hstk2,resi f
    L(s1 Lcap r) pwit stk2 in Ulstk1s1.let
    hstk1,capi stk2 in Ulstk1s1.let hi freergn
    r cap in Ulstk1s1.Lhstk1,resi

69
Translation FRGN to lrgnUL, Ops
Run comp. and recover old stack and new cap
  • letRGN s1 t e let f 8s2.
    Pf(s1s2) ! RGN s2 t e in Ulstk1s1.let
    pack(r,cap) newrgn Uhi in Ulstk1s1.let stk2
    Lhstk1,capi in Ulstk1s1.let id Ulstk L(s1
    Lcap r).stk in Ulstk1s1.let pwit Upack(Lcap
    r,Uhid,idi) in Ulstk1s1.let hstk2,resi f
    L(s1 Lcap r) pwit stk2 in Ulstk1s1.let
    hstk1,capi stk2 in Ulstk1s1.let hi freergn
    r cap in Ulstk1s1.Lhstk1,resi

70
Translation FRGN to lrgnUL, Ops
Construct isomorphism
  • letRGN s1 t e let f 8s2.
    Pf(s1s2) ! RGN s2 t e in Ulstk1s1.let
    pack(r,cap) newrgn Uhi in Ulstk1s1.let stk2
    Lhstk1,capi in Ulstk1s1.let id Ulstk L(s1
    Lcap r).stk in Ulstk1s1.let pwit Upack(Lcap
    r,Uhid,idi) in Ulstk1s1.let hstk2,resi f
    L(s1 Lcap r) pwit stk2 in Ulstk1s1.let
    hstk1,capi stk2 in Ulstk1s1.let hi freergn
    r cap in Ulstk1s1.Lhstk1,resi

71
Translation FRGN to lrgnUL, Ops
  • letRGN s1 t e let f 8s2.
    Pf(s1s2) ! RGN s2 t e in Ulstk1s1.let
    pack(r,cap) newrgn Uhi in Ulstk1s1.let stk2
    Lhstk1,capi in Ulstk1s1.let id Ulstk L(s1
    Lcap r).stk in Ulstk1s1.let pwit Upack(Lcap
    r,Uhid,idi) in Ulstk1s1.let hstk2,resi f
    L(s1 Lcap r) pwit stk2 in Ulstk1s1.let
    hstk1,capi stk2 in Ulstk1s1.let hi freergn
    r cap in Ulstk1s1.Lhstk1,resi

72
Encoding Cyclone Features
  • Many of Cyclones features fit into this
    framework
  • Lexical Regions
  • Dynamic Regions
  • Heap
  • Reaps
  • Unique Pointers
  • See paper for more details.

73
Future Work
  • In practice, capabilities shouldnt be values
  • Encode results of region analyses
  • Aiken et.al. PLDI95, Henglein et.al. PPDP01
  • Combine convenience of monadic encapsulation with
    power of substructural threading

74
Conclusion
  • Substructural type systems applicable to encoding
    region-based memory management
  • target-level language exposes commonalitiesin
    source-level language features
  • Scope vs. Lifetime
  • Lexical scope of region name
  • Un-scoped lifetime of region capability
  • Unrestricted witnesses vs. Linear capabilities

75
(No Transcript)
Write a Comment
User Comments (0)
About PowerShow.com