Luca Salucci
2014-02-04 17:10:35 UTC
Thanks for the clarification! It is really useful. While compiling I am still having troubles. Let me explain.
my project folder hierarchy is the following:
.:
cuGML GML.x10
./cuGML:
cuSparseCSC.x10 cuVectorMult.x10 cuVector.x10
I use the following command to compile:
x10c++ -O -sourcepath cuGML -classpath <path_to_gml>/x10.gml/lib/native_gml.jar
-x10lib <path_to_gml>/x10.gml/native_gml.properties -post '# # # -L /usr/lib/libblas
-L /usr/lib/lapack -lblas -llapack' GML.x10 -o GML
My project extends the GML library (that I provide as .jar through the -classpath)
and thus makes use of BLAS and LAPACK that i link in the post-compilation.
I hope this is right, if not let me know please!
Well if I compile it to target the host everything is ok, but if I add the @CUDA directive
to the methods I want to run on the CUDA device I get the following:
x10c++: gcc: error: cuGML.cuVector.cu: No such file or directory
gcc: warning: -x c++ after last input file has no effect
gcc: fatal error: no input files
compilation terminated.
x10c++: Non-zero return code: 4
x10c++: Found @CUDA annotation, but not compiling for GPU because nvcc could not be run (check your $PATH).
x10c++: gcc: error: cuGML.cuVector.cu: No such file or directory
gcc: warning: -x c++ after last input file has no effect
gcc: fatal error: no input files
compilation terminated.
x10c++: Non-zero return code: 4
x10c++: Found @CUDA annotation, but not compiling for GPU because nvcc could not be run (check your $PATH).
x10c++: gcc: error: cuGML.cuVector.cu: No such file or directory
gcc: warning: -x c++ after last input file has no effect
gcc: fatal error: no input files
compilation terminated.
x10c++: Non-zero return code: 4
x10c++: Found @CUDA annotation, but not compiling for GPU because nvcc could not be run (check your $PATH).
x10c++: gcc: error: cuGML.cuVector.cu: No such file or directory
gcc: warning: -x c++ after last input file has no effect
gcc: fatal error: no input files
compilation terminated.
x10c++: Non-zero return code: 4
x10c++: Found @CUDA annotation, but not compiling for GPU because nvcc could not be run (check your $PATH).
x10c++: gcc: error: cuGML.cuVector.cu: No such file or directory
gcc: warning: -x c++ after last input file has no effect
gcc: fatal error: no input files
compilation terminated.
x10c++: Non-zero return code: 4
x10c++: Found @CUDA annotation, but not compiling for GPU because nvcc could not be run (check your $PATH).
x10c++: gcc: error: cuGML.cuVector.cu: No such file or directory
gcc: warning: -x c++ after last input file has no effect
gcc: fatal error: no input files
compilation terminated.
x10c++: Non-zero return code: 4
x10c++: Found @CUDA annotation, but not compiling for GPU because nvcc could not be run (check your $PATH).
x10c++: gcc: error: cuGML.cuVector.cu: No such file or directory
gcc: warning: -x c++ after last input file has no effect
gcc: fatal error: no input files
compilation terminated.
x10c++: Non-zero return code: 4
x10c++: Found @CUDA annotation, but not compiling for GPU because nvcc could not be run (check your $PATH).
x10c++: gcc: error: cuGML.cuVector.cu: No such file or directory
gcc: warning: -x c++ after last input file has no effect
gcc: fatal error: no input files
compilation terminated.
x10c++: Non-zero return code: 4
x10c++: Found @CUDA annotation, but not compiling for GPU because nvcc could not be run (check your $PATH).
x10c++: 15 dynamically checked calls or field accesses, run with -VERBOSE_CHECKS for more details.
I do not understand why it does not find the cuGML.<class>.cu files since they are present!
Indeed the project directory after compilation is in this state:
.:
cuGML GML GML.cc GML.h GML.x10 xxx_main_xxx.cc
./cuGML:
cuSparseCSC.cc cuSparseCSC.h cuSparseCSC.x10 cuVector.cc cuVector.cu cuVector__CUDADeviceNotFoundException.h
cuVector.h cuVectorMult.cc cuVectorMult.h cuVectorMult.x10 cuVector.x10
The GML program is produced but no .cubin file is generated for the methods to be run on the CUDA device, thus if i run it it crashes.
N.B. I do not know why but the last version of X10 (2.4.1) was not working on the machines on which I work!
After a bunch of work and configuration I found that the 2.3.1 version was working thus this is the version I am actually using.
I really do hope you do not tell me this is unsolvable otherwise my Master Thesis would be a flop.
Thank for your help!
Luca Salucci
my project folder hierarchy is the following:
.:
cuGML GML.x10
./cuGML:
cuSparseCSC.x10 cuVectorMult.x10 cuVector.x10
I use the following command to compile:
x10c++ -O -sourcepath cuGML -classpath <path_to_gml>/x10.gml/lib/native_gml.jar
-x10lib <path_to_gml>/x10.gml/native_gml.properties -post '# # # -L /usr/lib/libblas
-L /usr/lib/lapack -lblas -llapack' GML.x10 -o GML
My project extends the GML library (that I provide as .jar through the -classpath)
and thus makes use of BLAS and LAPACK that i link in the post-compilation.
I hope this is right, if not let me know please!
Well if I compile it to target the host everything is ok, but if I add the @CUDA directive
to the methods I want to run on the CUDA device I get the following:
x10c++: gcc: error: cuGML.cuVector.cu: No such file or directory
gcc: warning: -x c++ after last input file has no effect
gcc: fatal error: no input files
compilation terminated.
x10c++: Non-zero return code: 4
x10c++: Found @CUDA annotation, but not compiling for GPU because nvcc could not be run (check your $PATH).
x10c++: gcc: error: cuGML.cuVector.cu: No such file or directory
gcc: warning: -x c++ after last input file has no effect
gcc: fatal error: no input files
compilation terminated.
x10c++: Non-zero return code: 4
x10c++: Found @CUDA annotation, but not compiling for GPU because nvcc could not be run (check your $PATH).
x10c++: gcc: error: cuGML.cuVector.cu: No such file or directory
gcc: warning: -x c++ after last input file has no effect
gcc: fatal error: no input files
compilation terminated.
x10c++: Non-zero return code: 4
x10c++: Found @CUDA annotation, but not compiling for GPU because nvcc could not be run (check your $PATH).
x10c++: gcc: error: cuGML.cuVector.cu: No such file or directory
gcc: warning: -x c++ after last input file has no effect
gcc: fatal error: no input files
compilation terminated.
x10c++: Non-zero return code: 4
x10c++: Found @CUDA annotation, but not compiling for GPU because nvcc could not be run (check your $PATH).
x10c++: gcc: error: cuGML.cuVector.cu: No such file or directory
gcc: warning: -x c++ after last input file has no effect
gcc: fatal error: no input files
compilation terminated.
x10c++: Non-zero return code: 4
x10c++: Found @CUDA annotation, but not compiling for GPU because nvcc could not be run (check your $PATH).
x10c++: gcc: error: cuGML.cuVector.cu: No such file or directory
gcc: warning: -x c++ after last input file has no effect
gcc: fatal error: no input files
compilation terminated.
x10c++: Non-zero return code: 4
x10c++: Found @CUDA annotation, but not compiling for GPU because nvcc could not be run (check your $PATH).
x10c++: gcc: error: cuGML.cuVector.cu: No such file or directory
gcc: warning: -x c++ after last input file has no effect
gcc: fatal error: no input files
compilation terminated.
x10c++: Non-zero return code: 4
x10c++: Found @CUDA annotation, but not compiling for GPU because nvcc could not be run (check your $PATH).
x10c++: gcc: error: cuGML.cuVector.cu: No such file or directory
gcc: warning: -x c++ after last input file has no effect
gcc: fatal error: no input files
compilation terminated.
x10c++: Non-zero return code: 4
x10c++: Found @CUDA annotation, but not compiling for GPU because nvcc could not be run (check your $PATH).
x10c++: 15 dynamically checked calls or field accesses, run with -VERBOSE_CHECKS for more details.
I do not understand why it does not find the cuGML.<class>.cu files since they are present!
Indeed the project directory after compilation is in this state:
.:
cuGML GML GML.cc GML.h GML.x10 xxx_main_xxx.cc
./cuGML:
cuSparseCSC.cc cuSparseCSC.h cuSparseCSC.x10 cuVector.cc cuVector.cu cuVector__CUDADeviceNotFoundException.h
cuVector.h cuVectorMult.cc cuVectorMult.h cuVectorMult.x10 cuVector.x10
The GML program is produced but no .cubin file is generated for the methods to be run on the CUDA device, thus if i run it it crashes.
N.B. I do not know why but the last version of X10 (2.4.1) was not working on the machines on which I work!
After a bunch of work and configuration I found that the 2.3.1 version was working thus this is the version I am actually using.
I really do hope you do not tell me this is unsolvable otherwise my Master Thesis would be a flop.
Thank for your help!
Luca Salucci