Crypto.Lol.Benchmarks

Benchmarks for individual layers of the Lol stack (Tensor, CycRep, Cyc), plus default parameters and helper functions for diplaying results.

Default parameters for benchmarks

defaultLolBenches :: (forall m r. (Fact m, NFData r) => NFData (t m r), _) => Proxy t -> Proxy h -> [Benchmark]

Benchmark parameters reported in the paper. We suggest running these benchmarks to quickly compare performance on your system or with your Tensor backend.

oneIdxBenches :: forall t m r gen. _ => Proxy '(m, r) -> Proxy t -> Proxy gen -> Benchmark

Collection of all single-index operations at all levels of the library.

twoIdxBenches :: forall t m m' r. _ => Proxy '(m, m', r) -> Proxy t -> Benchmark

Collection of all inter-ring operations at all levels of the library.

Benchmarks for different layers of Lol

tensorBenches1 :: forall (t :: Factored -> * -> *) (m :: Factored) (r :: *) gen. (Fact m, _) => Proxy '(t, m, r) -> Proxy gen -> Benchmark

Benchmarks for single-index Tensor operations. There must be a CRT basis for \(O_m\) over r. These cover the same functions as cycBenches1, but may have different performance due to how GHC interacts with Lol.

tensorBenches2 :: forall (t :: Factored -> * -> *) (m :: Factored) (m' :: Factored) (r :: *). _ => Proxy '(t, m, m', r) -> Benchmark

Benchmarks for inter-ring Tensor operations. There must be a CRT basis for \(O_{m'}\) over r. These cover the same functions as cycBenches1, but may have different performance due to how GHC interacts with Lol.

cycRepBenches1 :: forall t m r gen. _ => Proxy '(t, m, r) -> Proxy gen -> Benchmark

Benchmarks for single-index CycRep operations. There must be a CRT basis for \(O_m\) over r. These cover the same functions as cycBenches1, but may have different performance due to how GHC interacts with Lol.

cycRepBenches2 :: forall t m m' r. _ => Proxy '(t, m, m', r) -> Benchmark

Benchmarks for inter-ring CycRep operations. There must be a CRT basis for \(O_{m'}\) over r. These cover the same functions as cycBenches2, but may have different performance due to how GHC interacts with Lol.

cycBenches1 :: forall t m r gen. _ => Proxy '(t, m, r) -> Proxy gen -> Benchmark

Benchmarks for single-index Cyc operations. There must be a CRT basis for \(O_m\) over r.

cycBenches2 :: forall t m m' r. _ => Proxy '(t, m, m', r) -> Benchmark

Benchmarks for inter-ring Cyc operations. There must be a CRT basis for \(O_{m'}\) over r.

Utilities for creating benchmarks

mkBench :: NFData b => String -> (a -> b) -> a -> Benchmark

Make a Benchmark from a function and its input

mkBenchIO :: NFData b => String -> IO b -> Benchmark

Make a Benchmark from an IO function and its input

bgroup

Arguments

:: String

A name to identify the group of benchmarks.

-> [Benchmark]

Benchmarks to group under this name.

-> Benchmark

Group several benchmarks together under a common name.

data Benchmark

Specification of a collection of benchmarks and environments. A benchmark may consist of:

• An environment that creates input data for benchmarks, created with env.
• A single Benchmarkable item with a name, created with bench.
• A (possibly nested) group of Benchmarks, created with bgroup.

Instances

Show Benchmark

Defined in Criterion.Measurement.Types

Methods

showsPrec :: Int -> Benchmark -> ShowS

show :: Benchmark -> String

showList :: [Benchmark] -> ShowS

Utilities for showing benchmark results

module Crypto.Lol.Utils.ShowType

data Verb

Verbosity of benchmark output.

Constructors

Progress

prints a '.' when each benchmark completes

Abridged

prints a one-line summary for each benchmark

Full

prints full criterion output for each benchmark

Instances

Eq Verb

Defined in Crypto.Lol.Utils.PrettyPrint

Methods

(==) :: Verb -> Verb -> Bool

(/=) :: Verb -> Verb -> Bool

prettyBenchesTable :: TableOpts -> Benchmark -> IO ()

Takes benchmark options an a benchmark group nested as params/level/op, and prints a table comparing operations across all selected levels of Lol.

defaultTableOpts :: Maybe String -> TableOpts

Runs all benchmarks with verbosity Progress.

data TableOpts

Options for the simple benchmark format.

Constructors

TOpts

Character width of row labels

Fields

• verb :: Verb

  Verbosity

• level :: String

  Which level of Lol to benchmark

• benches :: [String]

  Which operations to benchmark. The empty list means run all benchmarks.

• params :: [String]

  Which parameters to benchmark. The empty list means run all parameters.

• colWidth :: Int

  Character width of data columns

• testNameWidth :: Int

prettyBenchesDiagnostic :: DiagnosticOpts -> Benchmark -> IO ()

Takes benchmark options an a benchmark group nested as params/level/op, and prints a table comparing operations across all selected levels of Lol.

defaultDiagnosticOpts :: DiagnosticOpts

Runs all benchmarks with verbosity Progress.

data DiagnosticOpts

Options for the diagnostic benchmark printout.

Constructors

DOpts

Character width of row labels

Fields

• verb :: Verb

  Verbosity

• levels :: [String]

  Which levels of Lol to benchmark. The empty list means run all levels.

• benches :: [String]

  Which operations to benchmark. The empty list means run all benchmarks.

• redThreshold :: Double

  How many times larger a benchmark must be (compared to the minimum benchmark for that parameter, across all levels), to be printed in red

• colWidth :: Int

  Character width of data columns

• testNameWidth :: Int