Scaling Program Synthesis by Exploiting Existing Code James - PowerPoint PPT Presentation

Scaling Program Synthesis by Exploiting Existing Code James Bornholt Emina Torlak University of Washington

Synthesis: write programs automatically Syntax Semantics

Synthesis: write programs automatically Syntax Semantics

Synthesis: write programs automatically Target Behavior f ( x ) = 4 x + 1 Syntax Semantics

Synthesis: write programs automatically 4*x + 1 Target Behavior f ( x ) = 4 x + 1 Syntax Semantics

Synthesis: write programs automatically 4*x + 1 (x<<2) + 1 Target Behavior f ( x ) = 4 x + 1 x + x + x + x + 1 Syntax Semantics

Synthesis: write programs automatically Target Behavior Permutation ( L, f ( L )) ∧ Sorted ( f ( L )) Syntax Semantics

Synthesis: write programs automatically Quicksort Target Behavior Permutation ( L, f ( L )) ∧ Sorted ( f ( L )) Bubble Sort Syntax Semantics

The success of synthesis End-user programming by example [FlashFill, POPL’11] Cache coherence protocols [Transit, PLDI’13] Parallel browser layout engines [PPoPP’13] Compilers for new spatial architectures [Chlorophyll, PLDI’14]

The success of synthesis End-user programming by example [FlashFill, POPL’11] s i s Cache coherence protocols e h t n y S [Transit, PLDI’13] turns t n e t n i l e v e l - h g i h Parallel browser layout into l i a t e d l e engines [PPoPP’13] v e l - w o l Compilers for new spatial architectures [Chlorophyll, PLDI’14]

The success of synthesis Approximate Computing quality bounds into End-user programming approximate programs by example [FlashFill, POPL’11] s i s Cache coherence protocols e h t n y S [Transit, PLDI’13] turns t n e t n i l e v e l - h g i h Parallel browser layout into l i a t e d l e engines [PPoPP’13] v e l - w o l Compilers for new spatial architectures [Chlorophyll, PLDI’14]

The success of synthesis Approximate Computing quality bounds into End-user programming approximate programs by example [FlashFill, POPL’11] s Hardware i s Cache coherence protocols e h t n y S Synthesis [Transit, PLDI’13] turns t n e t n i l e v programs e l - h g i h into Parallel browser layout into l hardware designs i a t e d l e engines [PPoPP’13] v e l - w o l Compilers for new spatial architectures [Chlorophyll, PLDI’14]

The success of synthesis Approximate Computing quality bounds into End-user programming approximate programs by example [FlashFill, POPL’11] s Hardware i s Cache coherence protocols e h t n y S Synthesis [Transit, PLDI’13] turns t n e t n i l e v programs e l - h g i h into Parallel browser layout into l hardware designs i a t e d l e engines [PPoPP’13] v e l - w o l Compilers for new spatial Black Box architectures [Chlorophyll, PLDI’14] Systems observed behaviors into speci fj cations

Approximate Computing Not all applications require quality bounds perfect accuracy into approximate programs

Approximate Computing Not all applications require quality bounds perfect accuracy into approximate programs

Approximate Computing Not all applications require quality bounds perfect accuracy into approximate programs Video and image quality

Approximate Computing Not all applications require quality bounds perfect accuracy into approximate programs Video and image quality Sensors and simulation Machine learning

Approximate Computing Not all applications require quality bounds perfect accuracy into approximate programs Precise Implementation

Approximate Computing Not all applications require quality bounds perfect accuracy into approximate programs Precise Implementation Desired Quality

Approximate Computing Not all applications require quality bounds perfect accuracy into approximate programs Precise Implementation Approximate Compiler Desired Quality

Approximate Computing Not all applications require quality bounds perfect accuracy into approximate programs Precise Implementation Approximate Approximate Program Compiler Desired Quality

Approximate Computing Not all applications require quality bounds perfect accuracy into approximate programs Precise Implementation Approximate Approximate Program Program Synthesis † Desired Quality † Bornholt, Torlak, Ceze, Grossman. Approximate Program Synthesis . At WAX’15, collocated with PLDI’15.

Hardware Synthesis Synthesizing circuits from programs high-level languages into hardware designs

Hardware Synthesis Synthesizing circuits from programs high-level languages into hardware designs module example(input a, b, c, output y); assign y = ~a & ~b & ~c | a & ~b & ~c | a & ~b & c; endmodule

Hardware Synthesis Synthesizing circuits from programs high-level languages into hardware designs module example(input a, b, c, output y); assign y = ~a & ~b & ~c | a & ~b & ~c | a & ~b & c; endmodule

Hardware Synthesis Synthesizing circuits from programs high-level languages into hardware designs module example(input a, b, c, output y); assign y = ~a & ~b & ~c | a & ~b & ~c | a & ~b & c; endmodule

Hardware Synthesis Synthesizing circuits from programs high-level languages into hardware designs High-Level Synthesis (HLS)

Hardware Synthesis Synthesizing circuits from programs high-level languages into hardware designs Place and route Timing closure High-Level Synthesis (HLS) Crosstalk and feedback Quantum!

Hardware Synthesis Synthesizing circuits from programs high-level languages into hardware designs Mark Wyse UW grad student and HLS extraordinaire

Hardware Synthesis Synthesizing circuits from programs high-level languages into hardware designs Mark Wyse UW grad student and HLS extraordinaire float dist(float a[3], float b[3]) { float r = 0; r += (a[0] - b[0]) * (a[0] - b[0]); r += (a[1] - b[1]) * (a[1] - b[1]); r += (a[2] - b[2]) * (a[2] - b[2]); return sqrt(r); }

Black Box Systems De fj ning system behavior observed behaviors by observation into speci fj cations

Black Box Systems De fj ning system behavior observed behaviors by observation into speci fj cations

Black Box Systems De fj ning system behavior observed behaviors by observation into speci fj cations

Black Box Systems De fj ning system behavior observed behaviors by observation into speci fj cations

Black Box Systems De fj ning system behavior observed behaviors by observation into speci fj cations Programming by example

Black Box Systems De fj ning system behavior observed behaviors by observation into speci fj cations Programming Synthesizing x86 by example instruction specs Godefroid and Taly. Automated Synthesis of Symbolic Instruction Encodings from I/O Samples. PLDI’12.

Approximate Hardware Black Box Computing Synthesis Systems quality bounds programs observed behaviors into into into approximate programs hardware designs speci fj cations

Machine Learning and Synthesis

Learning programs from examples Black Box Systems observed behaviors into speci fj cations ^ ϕ ( x i , P ( x i )) ∃ P. x i ∈ X

Learning programs from examples Black Box Systems observed behaviors into speci fj cations ^ ϕ ( x i , P ( x i )) ∃ P. x i ∈ X FlashFill

Learning programs from examples FlashFill Version Space Algebra

Machine learning and synthesis Statistical Program Machine Synthesis Learning

Machine learning and synthesis Millions of examples/parameters Statistical Program Machine Synthesis Learning

Machine learning and synthesis 100 instructions Millions of examples/parameters Statistical Program Machine Synthesis Learning

Machine learning and synthesis Approximate Computing Hardware Synthesis Black Box Systems 100 instructions Millions of examples/parameters Statistical Program Machine Synthesis Learning

Machine learning and synthesis Approximate Computing Hardware Synthesis Black Box Systems 100 instructions Millions of examples/parameters Statistical Program Machine Synthesis Learning

Programs are not uniformly distributed.

Programs are not uniformly distributed. 80000 Stack Over fm ow mentions 60000 40000 20000 0 malloc longjmp

Programs are not uniformly distributed. 80000 Stack Over fm ow mentions 60000 40000 20000 0 malloc longjmp

Programs are not uniformly distributed. 80000 Stack Over fm ow mentions 60000 40000 20000 0 malloc longjmp

Programs are not uniformly distributed. 80000 mov Stack Over fm ow mentions mov call 60000 test jmp 40000 call lea 20000 push sub push 0 malloc longjmp

Component-Based Synthesis

Synthesis of Loop-Free Programs Gulwani, Jha, Tiwari, and Venkatesan. Synthesis of loop-free programs. PLDI’11.

Synthesis of Loop-Free Programs � x + y ⌫ f ( x, y ) = 2 Gulwani, Jha, Tiwari, and Venkatesan. Synthesis of loop-free programs. PLDI’11.