MattPD

LoopInterchange: version runtime-bounded outer epilogues

#224804 | 3524fcccffa51d87c6223e5cf8db1bf788ef626d runtime versioning

Context: Improving SPECfp2000/SWIM2000 performance for Flang.

Background: #224196 added outer-epilogue fission, a transformation distributing the trailing code of each outer iteration into its own loop so that the remaining nest can be interchanged. Its walkthrough documents its legality: recovery of a row stride W from exact byte offsets, and a dependence proof across the two loops created by distribution, holding while the outer loop's trip count T satisfies T <= W. A plan is the complete, non-mutating record of a single candidate decision. LoopInterchange builds the plan first and applies it later. The

MattPD

LoopInterchange: extract statically bounded outer epilogues

#224196 | 35703509 NFC helpers | 11bcd2fa preparation | c0ec2d79 apply

Context: Improving the SPECfp2000/SWIM2000 benchmark performance for Flang.

Background: SPEC CPU2000 171.swim does not interchange under Flang/LLVM (related earlier work & comment: sebpop reported that it did not interchange under Flang because LLVM could not infer the number of iterations). #199511 ("[LoopInterchange] Supported partially-perfect Loop Nests") then landed the candidate-chain r

MattPD

Implicit intrinsic names and module accessibility

The evidence favors accepting both compilation layouts in llvm/llvm-project#219570. The issue's expected-behavior section instead calls for rejection.

Standards conformance should determine the intended behavior. The implementation discussion below concerns the effort needed to achieve that behavior.

The case

The producer module is implicit_intrinsic_source. The facade module is implicit_intrinsic_facade.

MattPD

LoopInterchange: route inner subnests inside non-linear loop trees

#214920 | 4a6f6018

Context: Improving the SPECfp2000/SWIM2000 benchmark performance for Flang.

Background: SPEC CPU2000 171.swim does not interchange under Flang/LLVM (related earlier work & comment: sebpop reported that it did not interchange under Flang because LLVM could not infer the number of iterations).

This patch addresses a different blocker in the same benchmark. Section 5 states what remains.

MattPD

C++ links: compilers - optimization - Induction Variables & Scalar Evolution (SCEV)

See also:

• Compilers
• Compilers Correctness

---

Readings

MattPD

C++ links: compilers - optimization - inlining

See also:

• Compilers
• Compilers Correctness

---

• An Optimization-Driven Incremental Inline Substitution Algorithm for Just-in-Time Compilers

MattPD

rust-analyzer

• https://rust-analyzer.github.io/
• Architecture: https://rust-analyzer.github.io/book/contributing/architecture.html
• Explaining rust-analyzer: https://www.youtube.com/playlist?list=PLhb66M_x9UmrqXhQuIpWC5VgTdrGxMx3y

Incremental Packrat Parsing

• SLE 2017
• Patrick Dubroy and Alessandro Warth
• https://ohmlang.github.io/pubs/sle2017/incremental-packrat-parsing.pdf

MattPD

An intermediate representation for speculative optimizations in a dynamic compiler

• VMIL 2013
• Gilles Duboscq, Thomas Würthinger, Lukas Stadler, Christian Wimmer, Doug Simon, Hanspeter Mössenböck
• https://dl.acm.org/doi/abs/10.1145/2542142.2542143

Correctness of speculative optimizations with dynamic deoptimization

• POPL 2018
• Olivier Flückiger, Gabriel Scherer, Ming-Ho Yee, Aviral Goel, Amal Ahmed, Jan Vitek
• https://dl.acm.org/doi/abs/10.1145/3158137

MattPD

• Teaching Compilers: https://danghica.blogspot.com/2020/04/teaching-compilers.html
• Thought Experiment: An Introductory Compilers Class: https://semantic-domain.blogspot.com/2020/02/thought-experiment-introductory.html
• Undergrad Compilers from the Hive Mind: https://eschew.wordpress.com/2020/01/26/undergrad-compilers-from-the-hive-mind/
• ChocoPy: A Programming Language for Compilers Courses: https://chocopy.org/
• Bril: An Intermediate Language for Teaching Compilers: https://www.cs.cornell.edu/~asampson/blog/bril.html
• My First Fifteen Compilers: https://blog.sigplan.org/2019/07/09/my-first-fifteen-compilers/
• Teaching Compilers Backward: https://blog.sigplan.org/2021/02/23/teaching-compilers-backward/
• Teaching and Learning Compilers Incrementally
  • ICFP 2023 Tutorial; Jeremy G. Siek
• https://www.youtube.com/watch?v=UJURb9H_q3A

MattPD

C++ links: Type Erasure

https://github.com/MattPD/cpplinks / C++ / Type Erasure

(draft; work in progress)

Type Erasure

• Back to Basics: Type Erasure
  • CppCon 2019; Arthur O'Dwyer