1. Introduction to Program Optimizations Chapter 11 Mooly Sagiv

2. Outline Undecidability of optimizations Optimization Criteria Placement of optimizations Early Optimizations –Constant-Folding –Scalar replacement of aggregates –Algebraic simplification –Value numbering

3. Undecidability of Optimizations “Optimization” = More efficient generated code It is generally undecidable to determine: –If a given optimization always/sometimes leads to improvement on a given program –If a given optimization is applicable Solutions: –Allow optimizations to slow some programs –Focus on improvements for “reasonable” programs –Apply optimization conservatively

4. Example: Loop Invariant Code Motion L1: if … goto L2 … t  y/z r  r + t … goto L1 L2: t  y/z L1: if … goto L2 … r  r + t … goto L1 L2:

5. Optimization Criteria Execution-time Code space Data space

6. Profiling Instrument the code to measure certain “resource” consumption (profile information) Optimize the next compile session with profile information

7. More Dimensions for Data-Flow Analysis Flow-sensitive Context-sensitive May vs. Must

8. “Important” Optimizations Loop optimizations Memory hierarchy optimizations –Global register optimizations –Instruction and data cache –Virtual memory Interprocedural optimizations (OO programs) Tail call optimizations (recursive programs)

9. Importance of Individual Optimizations Group I: Operate on loops, important for all programs –constant folding –algebraic simplifications and reassociations –global value numbering –sparse conditional constant propagation –common-subexpression elimination, loop-invariant code motions, partial redundency elimination –strength reduction –induction variable elimination –dead code elimination –unreachable code elimination –graph coloring register allocation –software pipelining, loop unrolling, register renaming –branch and basic block scheduling

10. Importance of Individual Optimizations Group II: Operate on loops, important for many programs –Local and global copy propagation –leaf-routine optimization –machine idioms and instruction combining –branch optimization and loop inversion (while  repeat) –unnecessary bound-checking elimination –branch predication

11. Importance of Individual Optimizations Group III: apply to whole procedures, help other optimizations –procedure integration –tail-call elimination –in-line expansion –shrink wrapping (move/decrease prologue and epilogue) –scalar replacement of aggregates –additional control flow optimizations (straithening, if simplification, unswitching, conditional moves) Group IV: Decrease code size –code hoisting –tail merging

12. Order of Optimizations and Repetitions Transformations are order-dependent No “Optimal order” May need to repeat certain transformations

13. Textbook Order Scalar replacement of array references Data-cache optimizations A HIR Global value numbering … C MIR|LIR Procedure integration … B HIR|MIR In-line expansion … D LIR Interprocedural register allocation … E link-time constant-folding simplifications

14. Procedure integration Tail-call elimination Scalar replacement of aggregates Sparse conditional constant propagation Interprocedural constant propagation Procedure specialization and cloning Sparse conditional constant propagation Early High Level Optimizations (B) HIR|MIR constant-folding simplifications

15. MIR|LIR(C) constant-folding simplifications Global value numbering Local and global copy propagation Sparse conditional constant propagation Dead code elimination Common Subexpression Elimination Loop invariant code motion Partial redundency Elimination Dead code elimination code hoisting induction variable strength reduction linear function test replacement induction variable removal bound checking elimination, control flow optimizations

16. LIR(D) constant-folding simplifications Inline expansion Leaf-routine optimizations Shrink wrapping Machine idioms Tail merging Branch optimization and conditional moves Dead code elimination Software pipelining, … Instruction Scheduling 1 Register allocation Instruction Scheduling 2 Intraprocedural I-cache optimizations Instruction prefetching Data prefertching Branch predication

17. Link-time optimizations(E) Interprocedural register allocation Aggregation global references Interprcudural I-cache optimizations

18. Scalar replacement of array references Data-cache optimizations Procedure integration Tail-call elimination Scalar replacement of aggregates Sparse constant propagation Interprocedural constant propagation Procedure specialization and cloning Sparse conditional constant propagation Global value numbering Local and global copy propagation Sparse conditional constant propagation Dead code elimination Common Subexpression Elimination Loop invariant code motion Partial redundency Elimination Inline expansion Leaf-routine optimizations Instruction Scheduling 1 Register allocation Instruction Scheduling 2 Intraprocedural I-cache optimizations Instruction prefetching Data prefertching Branch predication Interprocedural register allocation Interprocedural I-cache optimization