1 / 19

Advanced Compilers CMPSCI 710 Spring 2003 Data flow analysis

Advanced Compilers CMPSCI 710 Spring 2003 Data flow analysis. Emery Berger University of Massachusetts, Amherst. Data flow analysis. Framework for proving facts about program at each point Point: entry or exit from block (or CFG edge) Lots of “small facts”

vachel
Télécharger la présentation

Advanced Compilers CMPSCI 710 Spring 2003 Data flow analysis

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Advanced CompilersCMPSCI 710Spring 2003Data flow analysis Emery Berger University of Massachusetts, Amherst

  2. Data flow analysis • Framework for proving facts about program at each point • Point: entry or exit from block (or CFG edge) • Lots of “small facts” • Little or no interaction between facts • Based on all paths through program • Includes infeasible paths

  3. Infeasible Paths Example a = 1; if (a == 0) { a = 1; } if (a == 0) { a = 2; } • Infeasible paths never actually taken by program, regardless of input • Undecidable to distinguish from feasible

  4. Data Flow-Based Optimizations • Dead variable elimination • a = 3; print a; x = 12; halt) a = 3; print a; halt • Copy propagation • x = y; … use of x ) …use of y • Partial redundancy • a = 3*c + d; b = 3*c ) b = 3*c; a=b+d • Constant propagation • a = 3; b = 2; c = a+b ) a = 3; b = 2; c = 5

  5. Data Flow Analysis • Define lattice to represent facts • Attach meaning to lattice values • Associate transfer function to each node • (f:L!L) • Initialize values at each program point • Iterate through program until fixed point

  6. Lattice-Related Definitions • Meet function: u • commutative and associative • x u x = x • Unique bottom? and top > element • x u? = ? • x u> = x • Ordering: x v y iff x u y = x • Function f is monotone if 8 x, y: • x v y implies f(x) v f(y)

  7. Bit-Vector Lattice • Meet = bit-vector logical and > 111 111 u 111 = 011 u 111 = 000 u 101 = 100 u 011 = 011 u 110 = 001 u 110 = 011 101 110 001 010 100 ? 000

  8. Constant Propagation Lattice > … -2 -1 0 1 2 … ? • Meet rules: • a u> = a • a u? = ? • constant u constant = constant (if equal) • constant u constant = ?(if not equal) • Define obvious transfer functions for arithmetic

  9. Iterative Data Flow Analysis • Initialize non-entry nodes to > • Identity element for meet function • If node function is monotone: • Each re-evaluation of node moves down the lattice, if it moves at all • If height of lattice is finite, must terminate

  10. Constant Propagation Example I • Two choices of “point”: • Compute at edges • maximal information • Compute on entry • must “meet” data from all incident edges • loses information x = true; if (x) then else a = 3; b = 2; a = 2; b = 3; c = a+b;

  11. Constant Propagation Example II • Vector for (x,a,b,c) • Init values to > • Iterate forwards x = true; if (x) (true,>,>,>) (>,>,>,>) (>,>,>,>) (true,>,>,>) then else a = 3; b = 2; a = 2; b = 3; (>,>,>,>) (>,>,>,>) (true,3,2,>) (true,2,3,>) c = a+b; (>,>,>,>) (true,?,?,5)

  12. Accuracy: MOP vs. MFP • We want “meet-over-all-paths” solution, but paths can be infinite if there are loops • Best we can do in general: • Maximum Fixed Point solution =largest solution, ordered by v, that is fixed point of iterative computation • Provides “most information” • More conservative than MOP

  13. Distributive Problems • f is distributiveiff • f(x u y) = f(x) u f(y) • Doing meet early doesn’t reduce precision • Non-distributive problems: • constant propagation • Distributive problems: • MFP = MOP • reaching definitions, live variables

  14. Reaching Definitions • Definition: each assignment to variable • defs(v) represents set of all definitions of v • Assume all variables scalars • No pointers • No arrays • A definition reaches given point if 9path to that point such that variable may have value from definition

  15. Data Flow Functions • Kill(S): facts not true after S just because they were true before • example: redefinition of variable (assignment) • Gen(S): facts true after S regardless of facts true before S • example: assigned values not killed in S • In(S): dataflow info on entry to S • In(S) = [p 2PRED(S)Out(p) • example: definitions that reach S • Out(S): dataflow info on exit from S • Out(S) = Gen(S) [ (In(S) – Kill(S)) • example: reaching defs after S

  16. For Reaching Definitions • For reaching defs, u= [ • Gen(d: v = exp) = {d} • “on exit from block d, generate new definition” • Kill(d: v = exp) = defs(v) • “on exit from block d, definitions of v are killed” • Computing In(S) • If S has one predecessor P, In(S) = Out(P) • Otherwise: In(S) = uP inPRED(S)Out(P) • Out(Entry) = {}

  17. Reaching Definitions Example Entry parameter a; parameter b; x = a*b; y = a*b; while (y > a+b) { a = a+1; x = a+b; } 1: parameter a; 2: parameter b; 3: x=a*b; 4: y=a*b; if y > a+b defs(x) = defs(y) = defs(a) = defs(b) = 5: a = a+1; Exit 6: x = a+b;

  18. Analysis Direction • Forwards analysis: • start with Entry, compute towards Exit • Backwards analysis: • start with Exit, compute towards Entry • In(S) = Gen(S) [ (Out(S) – Kill(S)) • Out(S) = uF in SUCC(S) In(F) • Backwards problems: • Live variables: which variables might be read before overwritten or discarded

  19. Next Time • More data flow analysis

More Related