Polly: Polyhedral optimizations for LLVM

Polly is a polyhedral optimizer for LLVM. Using an abstract mathematical representation it analyzes and optimizes the memory access pattern of a program. This includes data-locality optimizations for cache locality as well as automatic parallelization for thread-level and SIMD parallelism. Our overall goal is an integrated optimizer for data-locality and parallelism that takes advantage of multi-cores, cache hierarchies, short vector instructions as well as dedicated accelerators.

WARNING: Polly started as a research project and larger parts of it are still open research projects. Even though we aim for a robust, production quality implementation, not all parts of Polly are there yet. We invite you to contribute - both as industry collaborator, who may want to use parts of Polly in production, as well as researcher, who is more interested in using his expertise to work on some of the open research projects.

News

2012

February

Polly is an official LLVM project, reachable at http://polly.llvm.org

January

Improved support for the isl scheduling optimizer

Polly can now automatically optimize all polybench 2.0 kernels without the help of an external optimizer. The compile time is reasonable and we can show notable speedups for various kernels.

2011

November

Talk at the LLVM Developer Meeting 2011

New SCEV parser
(Allows parameters in array subscript and max/signextend)

October

Polly can use the isl schedule optimizer
(The optimizer is similar to the one in Pluto, but it is part of isl)

August

Use Polly as clang plugin

July

Polly builder as part of the LLVM Buildbots

June

Tobias is founded for three years by a Google Europe Fellowship in Efficient Computing.

May

Tobias' diploma thesis and Raghesh's master thesis. See our list of publications.

April

Polly moves to the LLVM infrastructure (svn, bugtracker)

March

Presentation at CGO/IMPACT

Polly can compile polybench 2.0 with vectorization and OpenMP code generation

Februar

pollycc - a script to automatically compile with polyhedral optimizations

Januar

Basic OpenMP support, Alias analysis integration, Pluto/POCC support


2010

Dezember

Basic vectorization support

November

Talk at the LLVM Developer Meeting

October

Dependency analysis

Finished Phase 1 - Get something working

Support scalar dependences and sequential SCoPs

August

RegionInfo pass committed to LLVM

llvm-test suite compiles

July

Code generation works for normal SCoPs.

June

OpenSCoP import/export works (as far as openscop is finished).

May

The CLooG AST can be parsed.

April

SCoPs can automatically be detected.

March

The RegionInfo framework is almost completed.

February

Translate a simple loop to Polly-IR and regenerate a loop structure with CLooG works.

ISL and CLooG are integrated.

January

The RegionInfo pass is finished.


2009

End of the year

Work on the infrastructure started.