7/23/2016

Program Analysis Recap https://www.cse.iitb.ac.in/~karkare/cs618/ • Optimizations Code Optimizations – To improve efficiency of generated (time, space, resources …) Amey Karkare – Maintain semantic equivalence Dept of Computer Science and Engg • Two levels IIT Kanpur – Machine Independent Visiting IIT Bombay [email protected] – Machine Dependent [email protected]

2

Peephole Optimization examples… • target code often contains redundant Redundant loads and stores instructions and suboptimal constructs • Consider the code sequence • examine a short sequence of target instruction (peephole) and replace by a Move R , a shorter or faster sequence 0 Move a, R0 • peephole is a small moving window on the target systems • Instruction 2 can always be removed if it does not have a label.

3 4

1 7/23/2016

Peephole optimization examples… example …

Unreachable code constant propagation • Consider following code sequence if 0 <> 1 goto L2 int debug = 0 print information if (debug) { L2: print debugging info } Evaluate boolean expression. Since if condition is always true the code becomes this may be translated as goto L2 if debug == 1 goto L1 goto L2 print debugging information L1: print debugging info L2: L2: The print statement is now unreachable. Therefore, the code Eliminate jumps becomes if debug != 1 goto L2 print debugging information L2: L2: 5 6

Peephole optimization examples… Peephole optimization examples…

• flow of control: replace jump over jumps • – Replace X^2 by X*X goto L2 goto L1 – Replace multiplication by left shift by … … … – Replace division by right shift … L1: goto L2 L1 : goto L2 • Use faster machine instructions replace Add #1,R • Simplify algebraic expressions remove x := x+0 or x:=x*1 by Inc R

7 8

2 7/23/2016

Proposed Evaluation

Assignments 5%‐10% Course Project 30%‐40% Course Logistics ( Proposal 5% ) ( Report 15% ) ( Implementation & Presentation 15% ) Mid semester exam 10%‐20% End semester exam 25%‐35% Quizzes/Class Participation 5% Audit: Stuff in BLACK

9 10

Machine Independent Optimizations

• of optimizations – Local Machine Independent Intraprocedural Code Optimizations – Global – Interprocedural

12

3 7/23/2016

Local Optimizations Global Optimizations

• Restricted to a basic block • Typically restricted within a procedure/function • Simplifies the analysis – Could be restricted to a smaller scope, e.g. a loop • Not all optimizations can be applied locally • Most implement up to global – E.g. Loop optimizations optimizations • Gains are also limited – Well founded theory • Simplify global/interprocedural – Practical gains optimizations

13 14

Interprocedural Optimizations

• Spans multiple procedures, files – In some cases multiple languages! A Catalogue of Code • Not as popular as global optimizations Optimizations – No single theory applicable to multiple scenarios – Time consuming

15 16

4 7/23/2016

Compile‐time Evaluation Compile‐time Evaluation

• Move run‐time actions to compile‐time • Constant Propagation • : – Replace a variable by its “constant” value Volume = 4/3*PI*r*r*r; – Compute 4/3*PI at i = 5; i = 5; … Replaced by … – Applied very frequently for linearizing indices of j = i*4; j = 5*4; multidimensional arrays … … – How/When can we apply it? – May result in application of constant folding – How/When can we apply it?

17 18

Common Subexpression Elimination Copy Propagation

• Reuse a computation if already “available” • Replace a variable by another – If they are guaranteed to have same value x = u+v; t0 = u+v; i = k; i= k; … Replaced by x = t0; y = u+v+w; … … Replaced by … … y = t0+w; j = i*4; j = k*4; … … … • How to do it? – May result in dead code, common subexpr, … • When can we do it? – How to apply it? – When can we apply it?

19 20

5 7/23/2016

Code Movement Code Movement

• Move the code in a program • Code size reduction • Benefits: Suppose op generates a large number of machine – Code size reduction instructions – Reduction in the frequency of execution t1 = x op y; if (a < b) if (a < b) • Allowed only if the meaning of the program u = x op y; Replaced by u = t1; else does not change. else v = x op y; – May result in dead code, common subexpr, … v = t1; – How/When can we apply it?

21 22

Code Movement Loop Invariant Code Movement • Execution frequency reduction • Execution frequency reduction if (a < b) if (a < b) { u = …; u= …; Replaced by for (…) { t3 = a+b; else t2 = x*y; for (…) { … Replaced by v = x*y; } else { u = a+b; … w = x*y; t2 = x*y; … u = t3; v = t2; } … } } w = t2; • How/When can we do it? • How/When can we do it?

23 24

6 7/23/2016

Code Movement Other optimizations • Safety of code motion • – Remove unreachable, unused code. • Profitability of code motion – Can we always do it? • Strength reduction – Use of low strength operators in place of high strength operators. • i*i instead of i^2, pow(i,2) • i<<1 instead of i*2 – Typically performed for integers only (Why?)

25 26

Data Flow Analysis Data Flow Abstraction • Class of techniques to derive information • Flow graph about flow of data – Graph representation of paths that – along program execution paths program may exercise during execution • Used to answer questions such as: – Typically one graph per procedure – whether two identical expressions evaluate to – Graphs for separate procedure have to same value be combined/connected for • used in common subexpression elimination interprocedural analysis – whether the result of an assignment is used later • Later! • used by dead code elimination • Single procedure, single flow graph for now.

27 28

7 7/23/2016

Data Flow Abstraction • Basic Blocks (bb) • Input state/Output state for Stmt –Program point before/after a stmt –Denoted IN[s] and OUT[s] –Within a basic block: • Program point after a stmt is same as the program point before the next stmt

29

8