streamly/docs/Build.md

Build and Optimization Guide

Building

Compiler (GHC) Versions

GHC 8.6 and above are recommended. For best performance use GHC 8.8 or 8.10 along with fusion-plugin (see below). Benchmarks show that GHC 8.8 has significantly better performance than GHC 8.6 in many cases.

GHC 9.0 has some performance issues, please see this issue for details. However, upcoming minor version updates may fix some of these issues.

Memory requirements

Building streamly itself may require upto 4GB memory. Depending on the size of the application you may require 1-16GB memory to build. For most applications up to 8GB of memory should be sufficient.

To reduce the memory footprint you may want to break big modules into smaller ones and reduce unnecessary inlining on large functions. You can also use the -Rghc-timing GHC option to report the memory usage during compilation.

See the "Build times and space considerations" section below for more details.

Compilation Options

Recommended Options

Add fusion-plugin to the build-depends section of your program in the cabal file and use the following GHC options:

  -O2
  -fdicts-strict
  -fmax-worker-args=16
  -fspec-constr-recursive=16
  -fplugin Fusion.Plugin

Important Notes:

1. fusion-plugin can improve performance significantly by better stream fusion, many cases. If the perform regresses due to fusion-plugin please open an issue. You may remove the -fplugin option for regular builds but it is recommended for deployment builds and performance benchmarking. Note, for GHC 8.4 or lower fusion-plugin cannot be used.
2. In certain cases it is possible that GHC takes too long to compile with -fspec-constr-recursive=16, if that happens please reduce the value or remove that option.
3. At the very least -O -fdicts-strict compilation options are absolutely required to avoid issues in some cases. For example, the program main = S.drain $ S.concatMap S.fromList $ S.repeat [] may hog memory without these options.

See Explanation for details about these flags.

Explanation

• -fdicts-strict is needed to avoid a GHC issue leading to memory leak in some cases.

• -fspec-constr-recursive is needed for better stream fusion by enabling the SpecConstr optimization in more cases. Large values used with this flag may lead to huge compilation times and code bloat, if that happens please avoid it or use a lower value (e.g. 3 or 4).

• -fmax-worker-args is needed for better stream fusion by enabling the SpecConstr optimization in some important cases.

• -fplugin=Fusion.Plugin enables predictable stream fusion optimization in certain cases by helping the compiler inline internal bindings and therefore enabling case-of-case optimization. In some cases, especially in some file IO benchmarks, it can make a difference of 5-10x better performance.

Multi-core Parallelism

Concurrency without a threaded runtime may be a bit more efficient. Do not use threaded runtime unless you really need multi-core parallelism. To get multi-core parallelism use the following GHC options:

-threaded -with-rtsopts "-N"

Platform Specific Features

Streamly supports Linux, macOS and Windows operating systems. Some modules and functionality may depend on specific OS kernel features. Features/modules may get disabled if the kernel/OS does not support it.

Linux

• File system events notification module is supported only for kernel versions 2.6.36 onwards.

Mac OSX

• File system events notification module requires macOS 10.7+ with Xcode/macOS SDK installed (depends on Cocoa framework). However, we only test on latest three versions of the OS.

Performance Optimizations

A "closed loop" is any streamly code that generates a stream using unfold (or conceptually any stream generation combinator) and ends up eliminating it with a fold (conceptually any stream elimination combinator). It is essentially a loop processing multiple elements in a stream sequence, just like a for or while loop in imperative programming.

Closed loops are generated in a modular fashion by stream generation, transformation and elimination combinators in streamly. Combinators transfer data to the next stream pipeline stage using data constructors. These data constructors are eliminated by the compiler using stream fusion optimizations, generating a very efficient loop.

However, stream fusion optimization depends on proper inlining of the combinators involved. The fusion-plugin package mentioned earlier fills gaps for several optimizations that GHC does not perform automatically. It automatically inlines the internal definitions that involve the constructors we want to eliminate. In some cases fusion-plugin may not help and programmer may have to annotate the code manually for complete fusion. In this section we mention some of the cases where programmer annotation may help in stream fusion.

Remember, you need to worry about performance only where it matters, try to optimize the fast path and not everything blindly.

INLINE annotations

It may help to add INLINE annotations on any intermediate functions involved in a closed loop. In some cases you may have to add an inline phase as well as described below.

Usually GHC has three inline phases - the first phase is pahse-2, the second phase is phase-1 and the last one is phase-0.

Early INLINE

Generally, you only have to inline the combinators or functions participating in a loop and not the whole loop itself. But sometimes you may want to inline the whole loop itself inside a larger function. In most cases you can just add an INLINE pragma on the function containing the loop. But you may need some special considerations in some (not common) cases.

In some cases you may have to use INLINE[2] instead of INLINE which means inline the function early in phase-2. This may sometimes be needed on the because the performance of several combinators in streamly depends on getting inlined in phase-2 and if you use a plain INLINE annotation GHC may decide to delay the inlining in some cases. This is not very common but may be needed sometimes. Perhaps GHC can be fixed or we can resolve this using fusion-plugin in future.

Delayed INLINE

When a function is passed to a higher order function e.g. a function passed to concatMap or unfoldMany then we want the function to be inlined after the higher order is inlined so that proper fusion of the higher order function can occur. For such cases we usually add INLINE[1] on the function being passed to instruct GHC not to inline it too early.

Strictness annotations

• Strictness annotations on data, specially the data used as accumulator in folds and scans, can help in improving performance.
• Strictness annotations on function arguments can help the compiler unbox constructors in certain cases, improving performance.
• Sometimes using -XStrict extension can help improve performance, if so you may be missing some strictness annotations. -XStrict can be used as an aid to detect missing annotations, using it blindly may regress performance.

Use tail recursion

Do not use a strict foldr or lazy foldl unless you know what you are doing. Use lazy foldr for lazily transforming the stream and strict foldl for reducing the stream. If you are manually writing recursive code, try to use tail recursion where possible.

Build times and space considerations

Haskell, being a pure functional language, confers special powers on GHC. It allows GHC to do whole program optimization. In a closed loop all the components of the loop are inlined and GHC fuses them together, performs many optimizing transformations and churns out an optimized fused loop code. Let's call it whole-loop-optimization.

To be able to fuse the loop by whole-loop-optimization all the parts of the loop must be operated on by GHC at the same time to fuse them together. The amount of time and memory required to do so depends on the size of the loop. Huge loops can take a lot of time and memory. We have seen GHC take 4-5 GB of memory when a lot of combinators are used in a single module.

If a module takes too much time and space we can break it into multiple modules moving some non-inlined parts in another module. There is another advantage of breaking large modules, it can take advantage of parallel compilation if they do not depend on each other.