Intermediate Representations Part II

Transcription

1 Intermediate Representations Part II

2 Types of Intermediate Representations Three major categories Structural Linear Hybrid

3 Directed Acyclic Graph A directed acyclic graph (DAG) is an AST with a unique node for each value - z w / x * x 2 2 y z x 2 * y w x / 2

4 Directed Acyclic Graph A directed acyclic graph (DAG) is an AST with a unique node for each value - z w / x * 2 y z x 2 * y w x / 2

5 Stack Machine Code Used for many dynamic languages, e.g., Java Example: x - 2 * y becomes push x! push 2! push y! multiply! subtract!

6 Stack Machine Code Operations take operands from a stack Compact form Introduced names are implicit, not explicit Simple to generate and execute code

7 Stack Machine Code Advantages Result is stored in a temporary! Explicit name for result. x - 2 * y Multiply pops two items off of stack and pushes result! Implicit name for result push 2! push y! multiply! push x! subtract!

8 Stack Machine Code Advantages push 2! push y! multiply! push x! subtract! 2 stack!

9 Stack Machine Code Advantages push 2! push y! multiply! push x! subtract! y 2 stack!

10 Stack Machine Code Advantages push 2! push y! multiply! push x! subtract! 2y stack!

11 Stack Machine Code Advantages push 2! push y! multiply! push x! subtract! x 2y stack!

12 Stack Machine Code Advantages push 2! push y! multiply! push x! subtract! x-2y stack!

13 Three Address Code Different representations of three address code In general, three address code has statements of the form: x y op z with 1 operator (op ) and (at most) 3 names (x, y, & z)

14 Three Address Code Example: z x - 2 * y becomes t 2 * y! z x - t! Explicit name for result.

15 Three Address Code Advantages Resembles many real (RISC) machines Introduces a new set of names Compact form

16 Three Address Code: Quadruples Naïve representation of three address code Table of k * 4 small integers Destination Two operands load r1, y! loadi r2, 2! mult r3, r2, r1! load r4, x! sub r5, r4, r3! RISC assembly code load! 1! y! loadi! 2! 2! mult! 3! 2! 1! load! 4! x! sub! 5! 4! 3! Array of instructions

17 Three Address Code: Array of Pointers Index causes level of indirection Easy (and cheap) to reorder Easy to add (delete) instructions

18 Three Address Code: Array of Pointers Index causes level of indirection Easy (and cheap) to reorder Easy to add (delete) instructions

19 Control-flow Graph Models the transfer of control in the procedure Nodes in the graph are basic blocks Can be represented with quads or any other linear representation Edges in the graph represent control flow Example a 2 b 5 if (x = y) a 3 b 4 Basic blocks Maximal length sequences of straight-line code c a * b

20 Control-Flow Graphs Node: an instruction or sequence of instructions (a basic block) Two instructions i, j in same basic block iff execution of i guarantees execution of j Directed edge: potential flow of control Distinguished start node Entry First instruction in program

21 Identifying Basic Blocks Input: sequence of instructions instr(i) Identify leaders: first instruction of basic block Iterate: add subsequent instructions to basic block until we reach another leader

22 Basic Block Partition Algorithm leaders = instr(1) // first instruction for i = 1 to n // iterate thru all instrs if instr(i) is a branch leaders = leaders targets of instr(i) leaders = leaders instr(i+1) // instr after branch worklist = leaders While worklist not empty x = first instruction in worklist worklist = worklist {x} block(x) = {x} for (i = x + 1; i <= n && i not in leaders; i++) block(x) = block(x) {i}

23 CFG Example

24 CFG Example

25 CFG Example This is a join point.

26 Single Static Assignment: High-Level Definition Each assignment to variable given a unique name All uses reached by that assignment are renamed

27 Single Static Assignment: High-Level Definition Each assignment to variable given a unique name All uses reached by that assignment are renamed Easy for straight-line code V 4 V 0 4 V + 5 V V 6 V 1 6 V + 7 V Before SSA After SSA

28 Single Static Assignment: High-Level Definition Each assignment to variable given a unique name All uses reached by that assignment are renamed Easy for straight-line code V 4 V 0 4 V + 5 V V 6 V 1 6 V + 7 V Before SSA After SSA

29 Single Static Assignment: High-Level Definition Each assignment to a variable is given a unique name All uses reached by that assignment are renamed Easy for straight-line code V 4 V 0 4 V + 5 V V 6 V 1 6 V + 7 V Before SSA After SSA

30 What About Control Flow? B1 if ( ) B1 if ( ) B2 B3 B2 B3 X 5 X 3 X 0 5 X 1 3 B4 Y X B4 X 2 Φ(X 0, X 1 ) Y 0 X 2 Before SSA After SSA Φ-nodes: combine values from distinct edges (join points)

31 Φ-Functions At each join, add special assignment: φ function : operands indicate which assignments reach join j th operand = j th predecessor If control reaches join from j th predecessor, then value of φ( ) is value of j th operand B1 if ( ) B2 B3 X 0 5 X 1 3 B4 X 2 Φ(X 0, X 1 ) Y 0 X 2

32 What About Control Flow? B1 X 1 B1 X 0 1 B2 X X + 1 B2 X 1 Φ(X 0, X 2 ) X 2 X Before SSA After SSA

33 Class Example if P x=v then v = 4 else v = 6

34 Class Example if P x=v then v = 4 else v = 6 if P then v 0 = 4 else v 1 = 6 V 2 = φ(v 0, v 1 ) X 0 = V 2

35 Another SSA Example Original! x! y! while (x < k)! x x + 1! y y + x!

36 Another SSA Example Original! x! y! while (x < k)! x x + 1! y y + x! Same code using gotos!!x!!!y!!!if (x >= k)! goto next! loop: x x + 1! y y + x!!! if(x < k)! goto loop! next:!

37 Another SSA Example Original! x! y! while (x < k)! x x + 1! y y + x! Same code using gotos!!x!!!y!!!if (x >= k)! goto next! loop: x x + 1! y y + x!!! if(x < k)! goto loop! next:!

38 Another SSA Example Same code using gotos SSA-form!!x!!!y!!!if (x >= k)! goto next! loop: x x + 1! y y + x!!! if(x < k)! goto loop! next:!!!x 0!!!y 0!!!if (x 0 >= k) goto next! loop: x 1 φ(x 0,x 2 )!!!! y 1 φ(y 0,y 2 )!!x 2 x 1 + 1!!y 2 y 1 + x 2!!! if(x 2 < k) goto loop! next:!

39 Another SSA Example Same code using gotos SSA-form!!x!!!y!!!if (x >= k)! goto next! loop: x x + 1! y y + x!!! if(x < k)! goto loop! next:!!!x 0!!!y 0!!!if (x 0 >= k) goto next! loop: x 1 φ(x 0,x 2 )!!!! y 1 φ(y 0,y 2 )!!x 2 x 1 + 1!!y 2 y 1 + x 2!!! if(x 2 < k) goto loop! next:!

40 Another SSA Example Same code using gotos SSA-form!!x!!!y!!!if (x >= k)! goto next! loop: x x + 1! y y + x!!! if(x < k)! goto loop! next:!!!x 0!!!y 0!!!if (x 0 >= k) goto next! loop: x 1 φ(x 0,x 2 )!!!! y 1 φ(y 0,y 2 )!!x 2 x 1 + 1!!y 2 y 1 + x 2!!! if(x 2 < k) goto loop! next:!

41 Another SSA Example Same code using gotos!!x!!!y!!!if (x >= k)! goto next! loop: x x + 1! y y + x!!! if(x < k)! goto loop! next:! SSA-form!!x 0!!!y 0!!!if (x 0 >= k) goto next! loop: x 1 φ(x 0,x 2 )!!!! y 1 φ(y 0,y 2 )!!x 2 x 1 + 1!!y 2 y 1 + x 2!!! if(x 2 < k) goto loop! next:!

42 Static Single Assignment Form Same code using gotos SSA-form!!x!!!y!!!if (x >= k)! goto next! loop: x x + 1! y y + x!!! if(x < k)! goto loop! next:! This is new! Inserted! at points where diff.! control flow merge.!!x 0!!!y 0!!!if (x 0 >= k) goto next! loop: x 1 φ(x 0,x 2 )!!!! y 1 φ(y 0,y 2 )!!x 2 x 1 + 1!!y 2 y 1 + x 2!!! if(x 2 < k) goto loop! next:!

43 Static Single Assignment Form Same code using gotos SSA-form!!x!!!y!!!if (x >= k)! goto next! loop: x x + 1! y y + x!!! if(x < k)! goto loop! next:! Keeps single assign.! property.!!x 0!!!y 0!!!if (x 0 >= k) goto next! loop: x 1 φ(x 0,x 2 )!!!! y 1 φ(y 0,y 2 )!!x 2 x 1 + 1!!y 2 y 1 + x 2!!! if(x 2 < k) goto loop! next:!

44 Dependence Graph

45 Dependence Graph

46 Dependence Graph

47 Dependence Graph

48 Using Multiple Representations Source Code Front End IR 1 Middle IR 2 Middle IR 3 End End Back End Target Code Repeatedly lower the level of the intermediate representation Each intermediate representation is suited towards certain optimizations Example: the Open64 compiler WHIRL intermediate format Consists of 5 different IRs that are progressively more detailed and less abstract

49 Memory Models Two major models Register-to-register model Keep all values that can legally be stored in a register in registers Ignore machine limitations on number of registers Compiler back-end must insert loads and stores Memory-to-memory model Keep all values in memory Only promote values to registers directly before they are used Compiler back-end can remove loads and stores Compilers for RISC machines usually use register-to-register Reflects programming model Easier to determine when registers are used

50 The Rest of the Story Representing the code is only part of an IR There are other necessary components Symbol table Constant table Representation, type Storage class, offset Storage map Overall storage layout Overlap information Virtual register assignments

51 Symbol Tables Classic approach to building a symbol table uses hashing Personal preference: a two-table scheme Sparse index to reduce chance of collisions Dense table to hold actual data See B.3 in EaC for a longer explanation Easy to expand, to traverse, to read & write from/to files Use chains in index to handle collisions Collision occurs when h() returns a slot in the sparse index that is already full. h( foe ) h( fee ) NextSlot fum float Stack-like growth scalar fee integer scalar fie char * array Comp 412, Fall Sparse index Dense table

52 Hash-less Symbol Tables Classic approach to building a symbol table uses hashing Some concern about worst-case behavior Collisions in the hash function can lead to linear search Some authors advocate perfect hash for keyword lookup Automata theory lets us avoid worst-case behavior Collision occurs when h() returns a slot in the sparse index that is already full. h( foe ) My favorite hash table organization Stack-like growth h( fee ) fum float scalar NextSlot fee integer scalar Comp 412, Fall Sparse index fie char * array Dense table

53 Hash-less Symbol Tables One alternative is Paige & Cai s multiset discrimination Order the name space offline Assign indices to each name Replace the names in the input with their encoded indices Digression on page 241 of EaC Using DFA techniques, we can build a guaranteed linear-time replacement for the hash function h DFA that results from a list of words is acyclic RE looks like r 1 r 2 r 3 r k Could process input twice, once to build DFA, once to use it We can do even better Comp 412, Fall

54 Hash-less Symbol Tables Classic approach to building a symbol table uses hashing Some concern about worst-case behavior Collisions in the hash function can lead to linear search Some authors advocate perfect hash for keyword lookup Automata theory lets us avoid worst-case behavior Replace the hash function, h, and the sparse index with an efficient direct map, d, d( fie ) d( fee ) d( fum ) d( foe ) Stack-like growth fum float scalar NextSlot fee integer scalar Comp 412, Fall Sparse index fie char * array Dense table

55 Hash-less Symbol Tables Incremental construction of an acyclic DFA To add a word, run it through the DFA At some point, it will face a transition to the error state At that point, start building states & transitions to recognize it Requires a memory access per character in the key If DFA grows too large, memory access costs become excessive For small key sets (e.g., names in a procedure), not a problem Optimizations Last state on each path can be explicit Substantial reduction in memory costs Instantiate when path is lengthened Trade off granularity against size of state representation Encode capitalization separately Bit strings tied to final state? Comp 412, Fall