rrnewton

{-# LANGUAGE ScopedTypeVariables #-}
{-# LANGUAGE TypeOperators #-}
{-# LANGUAGE FlexibleContexts #-}
{-# LANGUAGE FlexibleInstances #-}
{-# LANGUAGE CPP #-}

-- Naive translation of shen_anova1_batch.m into Accelerate.

module Main (main) where 

rrnewton

import Data.Complex

-- This should be the biggest representable double:
big :: Double
big = 1.7976931348623157e308

-- This uses OCaml's approach at overflow-avoidance:
mag :: Complex Double -> Double
mag (x:+y) = 

rrnewton

> module Main where

How to write hybrid CPU/GPU programs with Haskell
-------------------------------------------------

What's better than programming a GPU with a high-level,
Haskell-embedded DSL (domain-specific-language)?  Well, perhaps
writing portable CPU/GPU programs that utilize both pieces of

rrnewton

{-# LANGUAGE RankNTypes, GeneralizedNewtypeDeriving, CPP #-}

import Data.Vector.Mutable as MV
import qualified  Data.Vector as V -- ((!), freeze)
import Control.Monad.ST
import Control.Monad.Primitive
import Prelude hiding (read)


type Splitter a = a -> (a,a)

rrnewton

This is the average latency as measured by a single trial run of 100K ping/pongs with several different transports.

    PingTCP:           7.34s   >80%CPU 
    PingTCPTransport:  19s     ~60%CPU
    PingPipes:         6.36s   >=100%CPU

The machine is a 3.1 ghz Intel Westmere (Dell workstation) running RHEL 6.2.

The PingTCP is a baseline that uses the raw Network.Socket interface rather than the Network.Transport abstraction.