It seemed to me that 'nvcc' was not able to refer the proper x10
header files and libries' path (e.g., Inull/include.)
Would you be able to give me some comments how the x10c++'s post nvcc
compile step makes the right path?
Best regards,
Myoungkyu
=============================================================
svn: r19811
$ export CPLUS_INCLUDE_PATH=/usr/local/cuda-3.2/cuda/include
$ export LD_LIBRARY_PATH=/usr/local/cuda-3.2/cuda/lib:
$ ant dist -DX10RT_CUDA=true -DNO_CHECKS=true -Doptimize=true
=============================================================
$ '/home/mksong/x10-trunk-cuda/x10.dist/bin/x10c++' -O -NO_CHECKS
-STATIC_CALLS CUDAMatMul.x10 -o CUDAMatMul -report postcompile=5
Output files: [CUDAMatMul.h, CUDAMatMul.cu, CUDAMatMul__MyRail.h,
CUDAMatMul.cc]
Executing post-compiler nvcc --cubin -Xptxas -v -Inull/include CUDAMatMul.cu
Executing post-compiler g++ -g
-I/home/mksong/x10-trunk-cuda/x10.dist/include
-I/home/mksong/x10-trunk-cuda/x10.dist/stdlib/include
-I/home/mksong/x10-trunk-cuda/x10.dist/samples/CUDA -I. -O2
-finline-functions -DNO_TRACING -Wno-long-long -Wno-unused-parameter
-DNO_CHECKS -DX10_USE_BDWGC -pthread -o
/home/mksong/x10-trunk-cuda/x10.dist/samples/CUDA/CUDAMatMul
CUDAMatMul.cc xxx_main_xxx.cc
-L/home/mksong/x10-trunk-cuda/x10.dist/stdlib/lib -lgc -lx10
-L/home/mksong/x10-trunk-cuda/x10.dist/lib -lx10rt_sockets -ldl -lm
-lpthread -Wl,--rpath
-Wl,/home/mksong/x10-trunk-cuda/x10.dist/stdlib/lib -Wl,--rpath
-Wl,/home/mksong/x10-trunk-cuda/x10.dist/lib -Wl,-export-dynamic -lrt
x10c++: CUDAMatMul.cu:1:27: error: x10aux/config.h: No such file or
directory
CUDAMatMul.cu:2:34: error: x10aux/cuda_kernel.cuh: No such file
or directory
x10c++: Non-zero return code: 1
x10c++: Found @CUDA annotation, but not compiling for GPU because nvcc
could not be run (check your $PATH).
$
$ nvcc --cubin -Xptxas -v -Inull/include CUDAMatMul.cu
CUDAMatMul.cu:1:27: error: x10aux/config.h: No such file or directory
CUDAMatMul.cu:2:34: error: x10aux/cuda_kernel.cuh: No such file or directory
=============================================================
<21:03:28> mksong@cuda:x10-trunk-cuda$ find . | xargs grep -i "nvcc"
...
./x10.compiler/src/x10cuda/visit/CUDACodeGenerator.java:
String[]
nvccCmd = { "nvcc", "--cubin", "-Xptxas", "-v",
...
-------------------------------------------------------------------------------------------------------------------------
>I don't know if -DX10RT_CUDA is sufficient as I always do -DX10RT_CUDA=true
>
>I don't know why it can't find x10aux/config.h, the nvcc options must be
>wrong. Can you run x10c++ with -report postcompile=5 and show the nvcc
>line?
>
>How come you were able to run it if it failed compile?
>
>By the way that behaviour of CUDAMatMul is consistent with it running on the
>CPU instead of on the GPU (and hitting thread limits that are much lower on
>the CPU).
-------------------------------------------------------------------------------------------------------------------------
> Myoungkyu,
>
> The 'x10c' and 'x10c++' scripts use their own path to figure out the
> location of the X10 distribution. However, when you invoke them via
> the PATH, they are unable to figure that out (because their own path
> is empty). Always specify a path (either relative or absolute) to
> the 'x10c' or 'x10c++' invocations, e.g., 'x10.dist/bin/x10c++' or
> '/home/mksong/x10-trunk-cuda/x10.dist/bin/x10c++'.
>
> And yes, you also need -DX10RT_CUDA=true.
> Igor
-------------------------------------------------------------------------------------------------------------------------
On Fri, Jan 28, 2011 at 11:57 PM, Myoungkyu Song <mks...@vt.edu> wrote:
> Hi,
>
> I tried to test a CUDA application on the x10.dist, however I have
> gotten the following error messages, while X10C++ generated *.cc/h/cu
> and a.out.
>
> ====================================================
> $ svn co https://x10.svn.sourceforge.net/svnroot/x10/trunk x10-trunk-cuda
> $ ant -Doptimize=true -DX10RT_CUDA squeakyclean dist
> ====================================================
> $ x10c++ -O -NO_CHECKS -STATIC_CALLS CUDAMatMul.x10 -o CUDAMatMul
> x10c++: CUDAMatMul.cu:1:27: error: x10aux/config.h: No such file or
> directory
> CUDAMatMul.cu:2:34: error: x10aux/cuda_kernel.cuh: No such file or
> directory
> x10c++: Non-zero return code: 1
> x10c++: Found @CUDA annotation, but not compiling for GPU because nvcc
> could not be run (check your $PATH).
> $
> $ find . -name "config.h" -print
> ./x10.dist/stdlib/include/x10aux/config.h
> ./x10.runtime/src-cpp/bdwgc/src/libatomic_ops-1.2/src/config.h
> ./x10.runtime/src-cpp/x10aux/config.h
> $
> $ find . -name "cuda_kernel.cuh" -print
> ./x10.dist/stdlib/include/x10aux/cuda_kernel.cuh
> ./x10.runtime/src-cpp/x10aux/cuda_kernel.cuh
> $
> $ echo $PATH;
>
> /usr/local/cuda-3.2/cuda/bin:/home/mksong/x10-trunk-cuda/x10.dist/bin:/home/mksong/install/apache-ant-1.8.2/bin:/home/mksong/install/jdk1.6.0_23/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin:/usr/games:/usr/local/cuda/bin
> $
> $ echo $LD_LIBRARY_PATH
> /usr/local/cuda-3.2/cuda/lib:/usr/local/cuda/lib
> $
> $ which nvcc; nvcc --version
> /usr/local/cuda-3.2/cuda/bin/nvcc
> nvcc: NVIDIA (R) Cuda compiler driver
> Copyright (c) 2005-2010 NVIDIA Corporation
> Built on Wed_Nov__3_16:16:57_PDT_2010
> Cuda compilation tools, release 3.2, V0.2.1221
> $ ls -la
> total 2795
> drwxr-xr-x 3 mksong mksong 480 2011-01-28 23:36 .
> drwxr-xr-x 7 mksong mksong 712 2011-01-28 21:15 ..
> -rw-r--r-- 1 mksong mksong 15149 2011-01-28 21:15 CUDA3DFD.x10
> -rw-r--r-- 1 mksong mksong 8147 2011-01-28 21:15 CUDABlackScholes.x10
> -rw-r--r-- 1 mksong mksong 5514 2011-01-28 21:15 CUDAKernelTest.x10
> -rwxr-xr-x 1 mksong mksong 1416681 2011-01-28 23:36 CUDAMatMul
> -rw-r--r-- 1 mksong mksong 708913 2011-01-28 23:35 CUDAMatMul.cc
> -rw-r--r-- 1 mksong mksong 617308 2011-01-28 23:35 CUDAMatMul.cu
> -rw-r--r-- 1 mksong mksong 5109 2011-01-28 23:35 CUDAMatMul.h
> -rw-r--r-- 1 mksong mksong 6065 2011-01-28 23:35 CUDAMatMul__MyRail.h
> -rw-r--r-- 1 mksong mksong 37179 2011-01-28 21:15 CUDAMatMul.x10
> -rw-r--r-- 1 mksong mksong 1808 2011-01-28 21:15 CUDATopology.x10
> -rw-r--r-- 1 mksong mksong 12110 2011-01-28 21:15 KMeansCUDA.x10
> drwxr-xr-x 6 mksong mksong 240 2011-01-28 21:15 .svn
> -rw-r--r-- 1 mksong mksong 199 2011-01-28 23:36 xxx_main_xxx.cc
> $
> $ X10RT_ACCELS=ALL runx10 CUDAMatMul
> GC Warning: Repeated allocation of very large block (appr. size 67112960):
> May lead to memory leak and poor performance.
>
> testing sgemm( 'N', 'N', n, n, n, ... )
>
> TOO MANY THREADS... ABORTING
> terminate called after throwing an instance of 'int'
> Aborted
> $
> ====================================================
> * OS - Linux version 2.6.24-24-generic (Ubuntu 8.04.2)
> * Java - 1.6.0_23
> * g++ - 4.2.4
> * x10c++ - 2.1.1 (r19798 | yzibin | 2011-01-28 20:20:10 -0500)
> * cuda - 3.2, V0.2.1221
> $ lspci | grep VGA
> 01:00.0 VGA compatible controller: nVidia Corporation Unknown device
> 06d1 (rev a3)
> 02:00.0 VGA compatible controller: nVidia Corporation Unknown device
> 06d1 (rev a3)
> ====================================================
>
> So, could you give me some comments about the above problems?
>
> Best regards,
> Myoungkyu Song
------------------------------------------------------------------------------
Special Offer-- Download ArcSight Logger for FREE (a $49 USD value)!
Finally, a world-class log management solution at an even better price-free!
Download using promo code Free_Logger_4_Dev2Dev. Offer expires
February 28th, so secure your free ArcSight Logger TODAY!
http://p.sf.net/sfu/arcsight-sfd2d
_______________________________________________
X10-users mailing list
X10-users@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/x10-users
