diff --git a/Foreign/CUDA/Internal/Offsets.hsc b/Foreign/CUDA/Internal/Offsets.hsc --- a/Foreign/CUDA/Internal/Offsets.hsc +++ b/Foreign/CUDA/Internal/Offsets.hsc @@ -11,7 +11,7 @@ -- Runtime API -------------------------------------------------------------------------------- -#include +#include "cbits/stubs.h" devNameOffset, devMaxThreadDimOffset, devMaxGridSizeOffset :: Int diff --git a/Foreign/CUDA/Runtime/Device.chs b/Foreign/CUDA/Runtime/Device.chs --- a/Foreign/CUDA/Runtime/Device.chs +++ b/Foreign/CUDA/Runtime/Device.chs @@ -28,7 +28,7 @@ ) where -#include +#include "cbits/stubs.h" {# context lib="cudart" #} -- Friends diff --git a/Foreign/CUDA/Runtime/Error.chs b/Foreign/CUDA/Runtime/Error.chs --- a/Foreign/CUDA/Runtime/Error.chs +++ b/Foreign/CUDA/Runtime/Error.chs @@ -31,7 +31,6 @@ import System.IO.Unsafe #include "cbits/stubs.h" -#include {# context lib="cudart" #} diff --git a/Foreign/CUDA/Runtime/Event.chs b/Foreign/CUDA/Runtime/Event.chs --- a/Foreign/CUDA/Runtime/Event.chs +++ b/Foreign/CUDA/Runtime/Event.chs @@ -20,7 +20,7 @@ ) where -#include +#include "cbits/stubs.h" {# context lib="cudart" #} -- Friends diff --git a/Foreign/CUDA/Runtime/Exec.chs b/Foreign/CUDA/Runtime/Exec.chs --- a/Foreign/CUDA/Runtime/Exec.chs +++ b/Foreign/CUDA/Runtime/Exec.chs @@ -21,7 +21,6 @@ ) where #include "cbits/stubs.h" -#include {# context lib="cudart" #} -- Friends diff --git a/Foreign/CUDA/Runtime/Marshal.chs b/Foreign/CUDA/Runtime/Marshal.chs --- a/Foreign/CUDA/Runtime/Marshal.chs +++ b/Foreign/CUDA/Runtime/Marshal.chs @@ -38,7 +38,7 @@ ) where -#include +#include "cbits/stubs.h" {# context lib="cudart" #} -- Friends diff --git a/Foreign/CUDA/Runtime/Stream.chs b/Foreign/CUDA/Runtime/Stream.chs --- a/Foreign/CUDA/Runtime/Stream.chs +++ b/Foreign/CUDA/Runtime/Stream.chs @@ -18,7 +18,7 @@ ) where -#include +#include "cbits/stubs.h" {# context lib="cudart" #} -- Friends diff --git a/Foreign/CUDA/Runtime/Texture.chs b/Foreign/CUDA/Runtime/Texture.chs --- a/Foreign/CUDA/Runtime/Texture.chs +++ b/Foreign/CUDA/Runtime/Texture.chs @@ -29,7 +29,7 @@ import Foreign import Foreign.C -#include +#include "cbits/stubs.h" {# context lib="cudart" #} #c diff --git a/Foreign/CUDA/Runtime/Utils.chs b/Foreign/CUDA/Runtime/Utils.chs --- a/Foreign/CUDA/Runtime/Utils.chs +++ b/Foreign/CUDA/Runtime/Utils.chs @@ -12,7 +12,7 @@ module Foreign.CUDA.Runtime.Utils (runtimeVersion, driverVersion) where -#include +#include "cbits/stubs.h" {# context lib="cudart" #} -- Friends diff --git a/cbits/stubs.h b/cbits/stubs.h --- a/cbits/stubs.h +++ b/cbits/stubs.h @@ -5,14 +5,21 @@ #ifndef C_STUBS_H #define C_STUBS_H +#if defined(mingw32_TARGET_OS) +#include +#undef CUDARTAPI +#define CUDARTAPI __stdcall +#endif + /* * We need to work around some shortcomings in the C parser of c2hs by disabling advanced attributes etc on Apple platforms. */ -#ifdef __APPLE__ +#ifdef __APPLE__ #define _ANSI_SOURCE #define __AVAILABILITY__ #define __OSX_AVAILABLE_STARTING(_mac, _iphone) -#define __OSX_AVAILABLE_BUT_DEPRECATED(_macIntro, _macDep, _iphoneIntro, _iphoneDep) +#define __OSX_AVAILABLE_BUT_DEPRECATED(_macIntro, _macDep, _iphoneIntro, _iphoneDep) +#define __OSX_AVAILABLE_BUT_DEPRECATED_MSG(_osxIntro, _osxDep, _iosIntro, _iosDep, _msg) #endif #include diff --git a/configure b/configure --- a/configure +++ b/configure @@ -624,6 +624,7 @@ ac_subst_vars='LTLIBOBJS LIBOBJS +cuda_ghci_libs cuda_frameworks cuda_libraries cuda_lib_path @@ -1655,6 +1656,52 @@ } # ac_fn_c_check_header_compile +# ac_fn_c_check_decl LINENO SYMBOL VAR INCLUDES +# --------------------------------------------- +# Tests whether SYMBOL is declared in INCLUDES, setting cache variable VAR +# accordingly. +ac_fn_c_check_decl () +{ + as_lineno=${as_lineno-"$1"} as_lineno_stack=as_lineno_stack=$as_lineno_stack + as_decl_name=`echo $2|sed 's/ *(.*//'` + as_decl_use=`echo $2|sed -e 's/(/((/' -e 's/)/) 0&/' -e 's/,/) 0& (/g'` + { $as_echo "$as_me:${as_lineno-$LINENO}: checking whether $as_decl_name is declared" >&5 +$as_echo_n "checking whether $as_decl_name is declared... " >&6; } +if eval \${$3+:} false; then : + $as_echo_n "(cached) " >&6 +else + cat confdefs.h - <<_ACEOF >conftest.$ac_ext +/* end confdefs.h. */ +$4 +int +main () +{ +#ifndef $as_decl_name +#ifdef __cplusplus + (void) $as_decl_use; +#else + (void) $as_decl_name; +#endif +#endif + + ; + return 0; +} +_ACEOF +if ac_fn_c_try_compile "$LINENO"; then : + eval "$3=yes" +else + eval "$3=no" +fi +rm -f core conftest.err conftest.$ac_objext conftest.$ac_ext +fi +eval ac_res=\$$3 + { $as_echo "$as_me:${as_lineno-$LINENO}: result: $ac_res" >&5 +$as_echo "$ac_res" >&6; } + eval $as_lineno_stack; ${as_lineno_stack:+:} unset as_lineno + +} # ac_fn_c_check_decl + # ac_fn_c_try_link LINENO # ----------------------- # Try to link conftest.$ac_ext, and return whether this succeeded. @@ -3097,8 +3144,26 @@ # that, otherwise choose the default # if test "$NVCC" != ""; then - cuda_prefix="$(dirname "$(dirname "$NVCC")")" - cuda_c2hsflags="--cpp="$NVCC" --cppopts=-E --cppopts=-arch=sm_20 " + case $target in + *mingw* ) + # We need a Windows-style path, but mingw doesn't include cygpath, + # thus this hack. + NVCC=`cd $(dirname $NVCC) && pwd -W`/$(basename nvcc) + ;; + esac + + cuda_prefix="$(dirname "$(dirname "$NVCC")")" + + case $target in + *mingw* ) + cuda_c2hsflags="--cppopts=-E --cppopts=-arch=sm_20 --cppopts="-D${target_os}_TARGET_OS=1" " + CPPFLAGS+=" "-D${target_os}_TARGET_OS=1" " + ;; + * ) + cuda_c2hsflags="--cpp="$NVCC" --cppopts=-E --cppopts=-arch=sm_20 " + ;; + esac + else echo "WARNING: Cannot find the CUDA compiler 'nvcc'. This is likely to cause problems." cuda_prefix="/usr/local/cuda" @@ -3118,6 +3183,16 @@ #LDFLAGS+=" -Xlinker -rpath ${cuda_lib_path} " #CPPFLAGS+=" -arch ${target_cpu} " ;; + mingw*) + case $target_cpu in + x86_64*) + cuda_lib_path="${cuda_prefix}/lib/x64" + ;; + *) + cuda_lib_path="${cuda_prefix}/lib/Win32" + ;; + esac + ;; *) case $target_cpu in x86_64*) @@ -3607,7 +3682,52 @@ done -{ $as_echo "$as_me:${as_lineno-$LINENO}: checking for library containing cuDriverGetVersion" >&5 + +case $target in + *mingw32* ) + # AC_SEARCH_LIBS doesn't work on Win32 with functions that use the + # stdcall calling convention, so we use AC_CHECK_DELCS instead. + LIBS+="-lcuda -lcudart" + ac_fn_c_check_decl "$LINENO" "cuDriverGetVersion" "ac_cv_have_decl_cuDriverGetVersion" "#include +" +if test "x$ac_cv_have_decl_cuDriverGetVersion" = xyes; then : + ac_have_decl=1 +else + ac_have_decl=0 +fi + +cat >>confdefs.h <<_ACEOF +#define HAVE_DECL_CUDRIVERGETVERSION $ac_have_decl +_ACEOF +if test $ac_have_decl = 1; then : + +else + as_fn_error $? "could not find CUDA driver library${longerror}" "$LINENO" 5 +fi + + ac_fn_c_check_decl "$LINENO" "cudaRuntimeGetVersion" "ac_cv_have_decl_cudaRuntimeGetVersion" "#include +" +if test "x$ac_cv_have_decl_cudaRuntimeGetVersion" = xyes; then : + ac_have_decl=1 +else + ac_have_decl=0 +fi + +cat >>confdefs.h <<_ACEOF +#define HAVE_DECL_CUDARUNTIMEGETVERSION $ac_have_decl +_ACEOF +if test $ac_have_decl = 1; then : + +else + as_fn_error $? "could not find CUDA runtime library${longerror}" "$LINENO" 5 +fi + + + cuda_ghci_libs+="`nm ${cuda_lib_path}/cuda.lib | head -2 | tail -1 | sed -e 's/://' -e 's/.dll//'` " + cuda_ghci_libs+="`nm ${cuda_lib_path}/cudart.lib | head -2 | tail -1 | sed -e 's/://' -e 's/.dll//'` " + ;; + *) + { $as_echo "$as_me:${as_lineno-$LINENO}: checking for library containing cuDriverGetVersion" >&5 $as_echo_n "checking for library containing cuDriverGetVersion... " >&6; } if ${ac_cv_search_cuDriverGetVersion+:} false; then : $as_echo_n "(cached) " >&6 @@ -3665,7 +3785,7 @@ as_fn_error $? "could not find CUDA driver library${longerror}" "$LINENO" 5 fi -{ $as_echo "$as_me:${as_lineno-$LINENO}: checking for library containing cudaRuntimeGetVersion" >&5 + { $as_echo "$as_me:${as_lineno-$LINENO}: checking for library containing cudaRuntimeGetVersion" >&5 $as_echo_n "checking for library containing cudaRuntimeGetVersion... " >&6; } if ${ac_cv_search_cudaRuntimeGetVersion+:} false; then : $as_echo_n "(cached) " >&6 @@ -3723,6 +3843,8 @@ as_fn_error $? "could not find CUDA runtime library${longerror}" "$LINENO" 5 fi + ;; +esac # Populate the buildinfo, with the search paths and any target specific options # @@ -3738,6 +3860,7 @@ for x in $LIBS; do cuda_libraries+="${x#-l} " done + diff --git a/configure.ac b/configure.ac --- a/configure.ac +++ b/configure.ac @@ -22,8 +22,26 @@ # that, otherwise choose the default # if test "$NVCC" != ""; then - cuda_prefix="$(dirname "$(dirname "$NVCC")")" - cuda_c2hsflags="--cpp="$NVCC" --cppopts=-E --cppopts=-arch=sm_20 " + case $target in + *mingw* ) + # We need a Windows-style path, but mingw doesn't include cygpath, + # thus this hack. + NVCC=`cd $(dirname $NVCC) && pwd -W`/$(basename nvcc) + ;; + esac + + cuda_prefix="$(dirname "$(dirname "$NVCC")")" + + case $target in + *mingw* ) + cuda_c2hsflags="--cppopts=-E --cppopts=-arch=sm_20 --cppopts="-D${target_os}_TARGET_OS=1" " + CPPFLAGS+=" "-D${target_os}_TARGET_OS=1" " + ;; + * ) + cuda_c2hsflags="--cpp="$NVCC" --cppopts=-E --cppopts=-arch=sm_20 " + ;; + esac + else echo "WARNING: Cannot find the CUDA compiler 'nvcc'. This is likely to cause problems." cuda_prefix="/usr/local/cuda" @@ -43,6 +61,16 @@ #LDFLAGS+=" -Xlinker -rpath ${cuda_lib_path} " #CPPFLAGS+=" -arch ${target_cpu} " ;; + mingw*) + case $target_cpu in + x86_64*) + cuda_lib_path="${cuda_prefix}/lib/x64" + ;; + *) + cuda_lib_path="${cuda_prefix}/lib/Win32" + ;; + esac + ;; *) case $target_cpu in x86_64*) @@ -115,9 +143,24 @@ # CUDA 5.5 on OS X (Mavericks) — I guess, because the library is now only a stub. # AC_CHECK_HEADERS([cuda.h cuda_runtime_api.h], [], [AC_MSG_ERROR(could not find CUDA headers${longerror})]) -AC_SEARCH_LIBS(cuDriverGetVersion, cuda, [], [AC_MSG_ERROR(could not find CUDA driver library${longerror})]) -AC_SEARCH_LIBS(cudaRuntimeGetVersion, cudart, [], [AC_MSG_ERROR(could not find CUDA runtime library${longerror})]) +case $target in + *mingw32* ) + # AC_SEARCH_LIBS doesn't work on Win32 with functions that use the + # stdcall calling convention, so we use AC_CHECK_DELCS instead. + LIBS+="-lcuda -lcudart" + AC_CHECK_DECLS([cuDriverGetVersion], [], [AC_MSG_ERROR(could not find CUDA driver library${longerror})], [[#include ]]) + AC_CHECK_DECLS([cudaRuntimeGetVersion], [], [AC_MSG_ERROR(could not find CUDA runtime library${longerror})], [[#include ]]) + + cuda_ghci_libs+="`nm ${cuda_lib_path}/cuda.lib | head -2 | tail -1 | sed -e 's/://' -e 's/.dll//'` " + cuda_ghci_libs+="`nm ${cuda_lib_path}/cudart.lib | head -2 | tail -1 | sed -e 's/://' -e 's/.dll//'` " + ;; + *) + AC_SEARCH_LIBS(cuDriverGetVersion, cuda, [], [AC_MSG_ERROR(could not find CUDA driver library${longerror})]) + AC_SEARCH_LIBS(cudaRuntimeGetVersion, cudart, [], [AC_MSG_ERROR(could not find CUDA runtime library${longerror})]) + ;; +esac + # Populate the buildinfo, with the search paths and any target specific options # cuda_ccflags="$CPPFLAGS " @@ -140,5 +183,6 @@ AC_SUBST([cuda_lib_path]) AC_SUBST([cuda_libraries]) AC_SUBST([cuda_frameworks]) +AC_SUBST([cuda_ghci_libs]) AC_OUTPUT diff --git a/cuda.cabal b/cuda.cabal --- a/cuda.cabal +++ b/cuda.cabal @@ -1,5 +1,5 @@ Name: cuda -Version: 0.6.5.0 +Version: 0.6.5.1 Synopsis: FFI binding to the CUDA interface for programming NVIDIA GPUs Description: The CUDA library provides a direct, general purpose C-like SPMD programming @@ -13,7 +13,11 @@ The configure script will look for your CUDA installation in the standard places, and if the nvcc compiler is found in your PATH, relative to that. . - This release is for version 6.0 of the CUDA toolkit. + This release is for version 6.5 of the CUDA toolkit. + . + [/NOTE:/] + . + Due to a bug in nvcc, you must use c2hs-0.17.2. See tmcdonell/cuda#18. License: BSD3 License-file: LICENSE @@ -81,8 +85,7 @@ ghc-prof-options: -fprof-auto -fprof-cafs -Test-suite deviceQuery - Type: exitcode-stdio-1.0 +Executable nvidia-device-query Main-is: DeviceQuery.hs hs-source-dirs: examples/src/deviceQueryDrv diff --git a/examples/src/deviceQueryDrv/DeviceQuery.hs b/examples/src/deviceQueryDrv/DeviceQuery.hs --- a/examples/src/deviceQueryDrv/DeviceQuery.hs +++ b/examples/src/deviceQueryDrv/DeviceQuery.hs @@ -1,5 +1,7 @@ {-# LANGUAGE RecordWildCards #-} +module Main where + import Numeric import Control.Monad import Text.Printf @@ -31,11 +33,11 @@ statDevice :: DeviceProperties -> IO () statDevice dev@DeviceProperties{..} = let - DeviceResources{..} = deviceResources dev + DeviceResources{..} = deviceResources dev - pad v = take width $ v ++ repeat ' ' - width = maximum $ map (length . fst) props - table = nest 2 $ vcat $ map (\(k,v) -> text (pad k) <+> v) props + pad v = take width $ v ++ repeat ' ' + width = maximum $ map (length . fst) props + table = nest 2 $ vcat $ map (\(k,v) -> text (pad k) <+> v) props grid (x,y) = int x <+> char 'x' <+> int y cube (x,y,z) = int x <+> char 'x' <+> int y <+> char 'x' <+> int z @@ -55,9 +57,9 @@ ,("Maximum threads per block:", int maxThreadsPerBlock) ,("Maximum grid dimensions:", cube maxGridSize) ,("Maximum block dimensions:", cube maxBlockSize) - ,("GPU clock rate:", text $ showFreq clockRate) - ,("Memory clock rate:", text $ showFreq memClockRate) - ,("Memory bus width:", text $ showFreq memBusWidth) + ,("GPU clock rate:", text . showFreq $ clockRate * 1000) + ,("Memory clock rate:", text . showFreq $ memClockRate * 1000) + ,("Memory bus width:", int memBusWidth <> text "-bit") ,("L2 cache size:", text $ showBytes cacheMemL2) ,("Maximum texture dimensions", empty) ,(" 1D:", int maxTextureDim1D)