1. Rahul Sharma, Eric Schkufza, Berkeley Churchill, Alex Aiken

2.  Prove two programs are equivalent  Compiler optimizations  Validate refactorings  Cross checking different implementations  Old and well studied problem  Undecidable in general  Major challenge: prove equivalence of loops  Straight line programs relatively easy

3.  Prove equivalence of two binaries … while … … while … … Trustworthy Compiler CompCert, gcc –O0 Optimizing Compiler gcc –O3, icc –O3

4. Straight Line Code Straight Line Code Trustworthy Compiler CompCert, gcc –O0 STOKE (ASPLOS 13) Random mutations … while … … while … …

5.  Do not support “while” loops: [CHR00], [FH02], [FH05], [AEF + 05], [SBC + 05], [MSF06]  Do not reason about termination: [SDE + 08], [GS09], [RE11], [LHM + 12], [PY13], [LMS + 13]  Translation validation: [Nec00],[GZB05], …  Need information from the compiler

6.  Decompose proof movq 8(rsp), rdi #rdi != 0 movq 8(rsp), rdi decq rdi movq rdi, 8(rsp) retq movq 8(rsp), r9 #r9 != 0 decq r9retq a a’ bb’ cc’

7.  Given a simulation relation, proofs for loops reduce to proofs for loop free fragments  Use decision procedures  Main challenge: infer a simulation relation  Infer synchronization points  Infer invariants  We use compilers as black boxes  Mine relations from concrete executions

8.  Run some tests to get data  From executions, unit tests, random tests, etc.

9. B retq B’ retq 2nn B;B n

10.  Attempt to detect synchronization points  Number of times program points are executed  Values align movq 8(rsp), rdi #rdi != 0 movq 8(rsp), rdi decq rdi movq rdi, 8(rsp) retq movq 8(rsp), r9 #r9 != 0 decq r9retq n 1n n+1 n

11.  Invariants are restricted to equalities  Infer invariants from observed data values 8(rsp)rdi 22 11 00 movq 8(rsp), rdi #rdi != 0 movq 8(rsp), rdi decq rdi movq rdi, 8(rsp) retq

12.  Invariants are restricted to equalities  Infer invariants from observed data values 8(rsp)rdir9’ 222 111 000 movq 8(rsp), r9 #r9 != 0 decq r9retq

13. 8(rsp)rdir9’ 222 111 000

14.  The executions are synchronized  The invariants are maintained movq 8(rsp), rdi #rdi != 0 movq 8(rsp), rdi decq rdi movq rdi, 8(rsp) retq movq 8(rsp), r9 #r9 != 0 decq r9retq a a’ bb’ cc’ States equal Live outs equal

15.  The executions are synchronized  The invariants are maintained  Queries in quantifier free bitvector arithmetic  Complete SMT solvers!  Incorporate counter-examples in relations  Sound but not complete  If checking succeeds then equivalent  Can fail to infer a sound simulation relation

16.  Insufficient data to infer a sound relation  Expressiveness of invariants  Inequalities, quantifiers, etc.  Expressiveness of SMT solver  Floating point, multiply, divide, etc.

17.  Run tests and generate data  https://github.com/eschkufz/x64asm https://github.com/eschkufz/x64asm  Nullspace computation  libIML: integer matrix library  SMT solver: Z3

18.  Compute kernel inside OpenSSL  Validating CompCert against gcc  Stochastic optimization for loops

19.  Multiplication kernel  Extensive performance tests  Run the kernel ~15 million times  Choose 16 random tests for inference  Compile with gcc –O0 and gcc –O3  Successfully prove equivalence

22. ProgramStoke vs gcc -O0Stoke vs gcc –O3 Bansal1.58X1.04X SAXPY9.22X1.48X

23.  Prove equivalence of loops in two stages  Infer simulation relation  Check the inferred relation using SMT solvers  Use runtime data for inference  No change required to the compilers  Better verifiers lead to better optimizers

24.  M. D. Ernst, J. H. Perkins, P. J. Guo, S. McCamant, C. Pacheco, M. S. Tschantz, and C. Xiao. The Daikon system for dynamic detection of likely invariants. Sci. Comput. Program., 69(1-3):35–45, 2007  T. Nguyen, D. Kapur, W. Weimer, and S. Forrest. Using dynamic analysis to discover polynomial and array invariants. ICSE 2012  P. Garg, C. Löding, P. Madhusudan, D. Neider: Learning Universally Quantified Invariants of Linear Data Structures. CAV 2013  R. Sharma, S. Gupta, B. Hariharan, A. Aiken, P. Liang, A. V. Nori: A Data Driven Approach for Algebraic Loop Invariants. ESOP 2013  R. Sharma, S. Gupta, B. Hariharan, A. Aiken, A. V. Nori: Verification as Learning Geometric Concepts. SAS 2013  A.V. Nori, R. Sharma: Termination proofs from tests. ESEC/SIGSOFT FSE 2013