tmcdonell

import LLVM.General.Context
import LLVM.General.Module                                      as LLVM
import LLVM.General.Target
import LLVM.General.ExecutionEngine
import LLVM.General.PassManager

import LLVM.General.AST.Name                                    ( Name(..) )
import LLVM.General.AST.DataLayout                              ( DataLayout, Endianness(..), defaultDataLayout )

import Control.Monad

tmcdonell

module Main (main) where

import Control.Monad

import Data.Array.Accelerate as A hiding ((++),(>->))
import Data.Array.Accelerate.CUDA
import Data.Array.Accelerate.IO as A

import Data.Vector.Storable as S

tmcdonell

> stack setup --upgrade-cabal --verbose
Version 0.1.4.0, Git revision dac8cfe51802e08edb523b80c51cf35e21933513 (1722 commits) X86_64
2015-09-21 14:35:46.098374: [debug] Checking for project config at: /Users/trevor/Dropbox/Projects/stack-bug/stack.yaml @(stack_256QTnIkPXj66qDsz01mSu:Stack.Config src/Stack/Config.hs:466:9)
2015-09-21 14:35:46.101455: [debug] Loading project config file stack.yaml @(stack_256QTnIkPXj66qDsz01mSu:Stack.Config src/Stack/Config.hs:489:13)
2015-09-21 14:35:46.102182: [debug] Trying to decode /Users/trevor/.stack/build-plan-cache/x86_64-osx/lts-2.22.cache @(stack_256QTnIkPXj66qDsz01mSu:Data.Binary.VersionTagged src/Data/Binary/VersionTagged.hs:77:5)
2015-09-21 14:35:46.107005: [debug] Success decoding /Users/trevor/.stack/build-plan-cache/x86_64-osx/lts-2.22.cache @(stack_256QTnIkPXj66qDsz01mSu:Data.Binary.VersionTagged src/Data/Binary/VersionTagged.hs:86:13)
2015-09-21 14:35:46.107324: [debug] Run process: ghc --info @(stack_256QTnIkPXj66qDsz01mSu:System.Process.Read src/System/Proc

tmcdonell

fib 10: 969 terms
benchmarking fib/10/down/manual
time                 10.90 μs   (10.77 μs .. 11.02 μs)
                     0.999 R²   (0.998 R² .. 0.999 R²)
mean                 10.84 μs   (10.72 μs .. 10.97 μs)
std dev              445.2 ns   (367.4 ns .. 545.6 ns)
variance introduced by outliers: 50% (severely inflated)

benchmarking fib/10/down/ghostbuster
time                 17.23 μs   (17.09 μs .. 17.41 μs)

tmcdonell

-- pretty :: PP.Pretty a => a -> String
-- pretty = PP.prettyPrintStyleMode style mode
--   where
--     mode   = PP.defaultMode { PP.classIndent = 2, PP.whereIndent = 2, PP.caseIndent = 2, PP.layout = PP.PPOffsideRule }
--     style  = PP.style { PP.mode = PP.PageMode, PP.lineLength = 1000 }

{-# LANGUAGE GADTs #-}
{-# LANGUAGE ScopedTypeVariables #-}
{-# LANGUAGE BangPatterns #-}
{-# OPTIONS_GHC -Wall #-}

tmcdonell

{-# LANGUAGE BangPatterns #-}
{-# LANGUAGE GADTs #-}
{-# LANGUAGE ScopedTypeVariables #-}
-- {-# OPTIONS_GHC -fdefer-type-errors #-}
{-# OPTIONS_GHC -fno-warn-unused-binds #-}
{-# OPTIONS_GHC -fno-warn-name-shadowing #-}
{-# OPTIONS_GHC -fno-warn-incomplete-patterns #-}
{-# OPTIONS_GHC -fno-warn-unused-matches #-}

module Ghostbuster where

tmcdonell

//
// Generated by LLVM NVPTX Back-End
//

.version 3.1
.target sm_30
.address_size 64

	// .globl	__nv_fminf
.visible .func  (.param .b32 func_retval0) fminf

tmcdonell

{-# LANGUAGE TypeOperators       #-}
{-# LANGUAGE ScopedTypeVariables #-}

import Prelude                          as P
import Data.Array.Accelerate            as A


ok :: Acc (Vector (Point Float))
   -> Acc (Vector (Cluster Float))
   -> Acc (Vector (Cluster Float))

tmcdonell

//
// Generated by LLVM NVPTX Back-End
//

.version 3.1
.target sm_20
.address_size 64

	// .globl	map
                                        // @map

tmcdonell

; ModuleID = 'map'
target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-v16:16:16-v32:32:32-v64:64:64-v128:128:128-f32:32:32-f64:64:64-n16:32:64"
target triple = "nvptx64-nvidia-cuda"

; Function Attrs: nounwind
define void @map(i64* noalias nocapture %out.ad0, i64 %out.sh0, i64* noalias nocapture readonly %fv0.ad0, i64 %fv0.sh0) #0 {
entry:
  %0 = trunc i64 %out.sh0 to i32
  %1 = tail call i32 @llvm.nvvm.read.ptx.sreg.nctaid.x() #2
  %2 = tail call i32 @llvm.nvvm.read.ptx.sreg.ntid.x() #2