diff options
Diffstat (limited to 'bench/btl/libs')
44 files changed, 3972 insertions, 0 deletions
diff --git a/bench/btl/libs/BLAS/CMakeLists.txt b/bench/btl/libs/BLAS/CMakeLists.txt new file mode 100644 index 000000000..de42fe047 --- /dev/null +++ b/bench/btl/libs/BLAS/CMakeLists.txt @@ -0,0 +1,60 @@ + +find_package(ATLAS) +if (ATLAS_FOUND) + btl_add_bench(btl_atlas main.cpp) + if(BUILD_btl_atlas) + target_link_libraries(btl_atlas ${ATLAS_LIBRARIES}) + set_target_properties(btl_atlas PROPERTIES COMPILE_FLAGS "-DCBLASNAME=ATLAS -DHAS_LAPACK=1") + endif(BUILD_btl_atlas) +endif (ATLAS_FOUND) + +find_package(MKL) +if (MKL_FOUND) + btl_add_bench(btl_mkl main.cpp) + if(BUILD_btl_mkl) + target_link_libraries(btl_mkl ${MKL_LIBRARIES}) + set_target_properties(btl_mkl PROPERTIES COMPILE_FLAGS "-DCBLASNAME=INTEL_MKL -DHAS_LAPACK=1") + endif(BUILD_btl_mkl) +endif (MKL_FOUND) + + +find_package(GOTO2) +if (GOTO2_FOUND) + btl_add_bench(btl_goto2 main.cpp) + if(BUILD_btl_goto2) + target_link_libraries(btl_goto2 ${GOTO_LIBRARIES} ) + set_target_properties(btl_goto2 PROPERTIES COMPILE_FLAGS "-DCBLASNAME=GOTO2") + endif(BUILD_btl_goto2) +endif (GOTO2_FOUND) + +find_package(GOTO) +if (GOTO_FOUND) + if(GOTO2_FOUND) + btl_add_bench(btl_goto main.cpp OFF) + else() + btl_add_bench(btl_goto main.cpp) + endif() + if(BUILD_btl_goto) + target_link_libraries(btl_goto ${GOTO_LIBRARIES} ) + set_target_properties(btl_goto PROPERTIES COMPILE_FLAGS "-DCBLASNAME=GOTO") + endif(BUILD_btl_goto) +endif (GOTO_FOUND) + +find_package(ACML) +if (ACML_FOUND) + btl_add_bench(btl_acml main.cpp) + if(BUILD_btl_acml) + target_link_libraries(btl_acml ${ACML_LIBRARIES} ) + set_target_properties(btl_acml PROPERTIES COMPILE_FLAGS "-DCBLASNAME=ACML -DHAS_LAPACK=1") + endif(BUILD_btl_acml) +endif (ACML_FOUND) + +if(Eigen_SOURCE_DIR AND CMAKE_Fortran_COMPILER_WORKS) + # we are inside Eigen and blas/lapack interface is compilable + include_directories(${Eigen_SOURCE_DIR}) + btl_add_bench(btl_eigenblas main.cpp) + if(BUILD_btl_eigenblas) + target_link_libraries(btl_eigenblas eigen_blas eigen_lapack ) + set_target_properties(btl_eigenblas PROPERTIES COMPILE_FLAGS "-DCBLASNAME=EigenBLAS") + endif() +endif() diff --git a/bench/btl/libs/BLAS/blas.h b/bench/btl/libs/BLAS/blas.h new file mode 100644 index 000000000..28f3a4e57 --- /dev/null +++ b/bench/btl/libs/BLAS/blas.h @@ -0,0 +1,675 @@ +#ifndef BLAS_H +#define BLAS_H + +#define BLASFUNC(FUNC) FUNC##_ + +#ifdef __WIN64__ +typedef long long BLASLONG; +typedef unsigned long long BLASULONG; +#else +typedef long BLASLONG; +typedef unsigned long BLASULONG; +#endif + +int BLASFUNC(xerbla)(const char *, int *info, int); + +float BLASFUNC(sdot) (int *, float *, int *, float *, int *); +float BLASFUNC(sdsdot)(int *, float *, float *, int *, float *, int *); + +double BLASFUNC(dsdot) (int *, float *, int *, float *, int *); +double BLASFUNC(ddot) (int *, double *, int *, double *, int *); +double BLASFUNC(qdot) (int *, double *, int *, double *, int *); + +#if defined(F_INTERFACE_GFORT) && !defined(__64BIT__) +int BLASFUNC(cdotu) (int *, float * , int *, float *, int *); +int BLASFUNC(cdotc) (int *, float *, int *, float *, int *); +void BLASFUNC(zdotu) (double *, int *, double *, int *, double *, int *); +void BLASFUNC(zdotc) (double *, int *, double *, int *, double *, int *); +void BLASFUNC(xdotu) (double *, int *, double *, int *, double *, int *); +void BLASFUNC(xdotc) (double *, int *, double *, int *, double *, int *); +#elif defined(F_INTERFACE_F2C) || \ + defined(F_INTERFACE_PGI) || \ + defined(F_INTERFACE_GFORT) || \ + (defined(F_INTERFACE_PATHSCALE) && defined(__64BIT__)) +void BLASFUNC(cdotu) (float *, int *, float * , int *, float *, int *); +void BLASFUNC(cdotc) (float *, int *, float *, int *, float *, int *); +void BLASFUNC(zdotu) (double *, int *, double *, int *, double *, int *); +void BLASFUNC(zdotc) (double *, int *, double *, int *, double *, int *); +void BLASFUNC(xdotu) (double *, int *, double *, int *, double *, int *); +void BLASFUNC(xdotc) (double *, int *, double *, int *, double *, int *); +#else +std::complex<float> BLASFUNC(cdotu) (int *, float *, int *, float *, int *); +std::complex<float> BLASFUNC(cdotc) (int *, float *, int *, float *, int *); +std::complex<double> BLASFUNC(zdotu) (int *, double *, int *, double *, int *); +std::complex<double> BLASFUNC(zdotc) (int *, double *, int *, double *, int *); +double BLASFUNC(xdotu) (int *, double *, int *, double *, int *); +double BLASFUNC(xdotc) (int *, double *, int *, double *, int *); +#endif + +int BLASFUNC(cdotuw) (int *, float *, int *, float *, int *, float*); +int BLASFUNC(cdotcw) (int *, float *, int *, float *, int *, float*); +int BLASFUNC(zdotuw) (int *, double *, int *, double *, int *, double*); +int BLASFUNC(zdotcw) (int *, double *, int *, double *, int *, double*); + +int BLASFUNC(saxpy) (int *, float *, float *, int *, float *, int *); +int BLASFUNC(daxpy) (int *, double *, double *, int *, double *, int *); +int BLASFUNC(qaxpy) (int *, double *, double *, int *, double *, int *); +int BLASFUNC(caxpy) (int *, float *, float *, int *, float *, int *); +int BLASFUNC(zaxpy) (int *, double *, double *, int *, double *, int *); +int BLASFUNC(xaxpy) (int *, double *, double *, int *, double *, int *); +int BLASFUNC(caxpyc)(int *, float *, float *, int *, float *, int *); +int BLASFUNC(zaxpyc)(int *, double *, double *, int *, double *, int *); +int BLASFUNC(xaxpyc)(int *, double *, double *, int *, double *, int *); + +int BLASFUNC(scopy) (int *, float *, int *, float *, int *); +int BLASFUNC(dcopy) (int *, double *, int *, double *, int *); +int BLASFUNC(qcopy) (int *, double *, int *, double *, int *); +int BLASFUNC(ccopy) (int *, float *, int *, float *, int *); +int BLASFUNC(zcopy) (int *, double *, int *, double *, int *); +int BLASFUNC(xcopy) (int *, double *, int *, double *, int *); + +int BLASFUNC(sswap) (int *, float *, int *, float *, int *); +int BLASFUNC(dswap) (int *, double *, int *, double *, int *); +int BLASFUNC(qswap) (int *, double *, int *, double *, int *); +int BLASFUNC(cswap) (int *, float *, int *, float *, int *); +int BLASFUNC(zswap) (int *, double *, int *, double *, int *); +int BLASFUNC(xswap) (int *, double *, int *, double *, int *); + +float BLASFUNC(sasum) (int *, float *, int *); +float BLASFUNC(scasum)(int *, float *, int *); +double BLASFUNC(dasum) (int *, double *, int *); +double BLASFUNC(qasum) (int *, double *, int *); +double BLASFUNC(dzasum)(int *, double *, int *); +double BLASFUNC(qxasum)(int *, double *, int *); + +int BLASFUNC(isamax)(int *, float *, int *); +int BLASFUNC(idamax)(int *, double *, int *); +int BLASFUNC(iqamax)(int *, double *, int *); +int BLASFUNC(icamax)(int *, float *, int *); +int BLASFUNC(izamax)(int *, double *, int *); +int BLASFUNC(ixamax)(int *, double *, int *); + +int BLASFUNC(ismax) (int *, float *, int *); +int BLASFUNC(idmax) (int *, double *, int *); +int BLASFUNC(iqmax) (int *, double *, int *); +int BLASFUNC(icmax) (int *, float *, int *); +int BLASFUNC(izmax) (int *, double *, int *); +int BLASFUNC(ixmax) (int *, double *, int *); + +int BLASFUNC(isamin)(int *, float *, int *); +int BLASFUNC(idamin)(int *, double *, int *); +int BLASFUNC(iqamin)(int *, double *, int *); +int BLASFUNC(icamin)(int *, float *, int *); +int BLASFUNC(izamin)(int *, double *, int *); +int BLASFUNC(ixamin)(int *, double *, int *); + +int BLASFUNC(ismin)(int *, float *, int *); +int BLASFUNC(idmin)(int *, double *, int *); +int BLASFUNC(iqmin)(int *, double *, int *); +int BLASFUNC(icmin)(int *, float *, int *); +int BLASFUNC(izmin)(int *, double *, int *); +int BLASFUNC(ixmin)(int *, double *, int *); + +float BLASFUNC(samax) (int *, float *, int *); +double BLASFUNC(damax) (int *, double *, int *); +double BLASFUNC(qamax) (int *, double *, int *); +float BLASFUNC(scamax)(int *, float *, int *); +double BLASFUNC(dzamax)(int *, double *, int *); +double BLASFUNC(qxamax)(int *, double *, int *); + +float BLASFUNC(samin) (int *, float *, int *); +double BLASFUNC(damin) (int *, double *, int *); +double BLASFUNC(qamin) (int *, double *, int *); +float BLASFUNC(scamin)(int *, float *, int *); +double BLASFUNC(dzamin)(int *, double *, int *); +double BLASFUNC(qxamin)(int *, double *, int *); + +float BLASFUNC(smax) (int *, float *, int *); +double BLASFUNC(dmax) (int *, double *, int *); +double BLASFUNC(qmax) (int *, double *, int *); +float BLASFUNC(scmax) (int *, float *, int *); +double BLASFUNC(dzmax) (int *, double *, int *); +double BLASFUNC(qxmax) (int *, double *, int *); + +float BLASFUNC(smin) (int *, float *, int *); +double BLASFUNC(dmin) (int *, double *, int *); +double BLASFUNC(qmin) (int *, double *, int *); +float BLASFUNC(scmin) (int *, float *, int *); +double BLASFUNC(dzmin) (int *, double *, int *); +double BLASFUNC(qxmin) (int *, double *, int *); + +int BLASFUNC(sscal) (int *, float *, float *, int *); +int BLASFUNC(dscal) (int *, double *, double *, int *); +int BLASFUNC(qscal) (int *, double *, double *, int *); +int BLASFUNC(cscal) (int *, float *, float *, int *); +int BLASFUNC(zscal) (int *, double *, double *, int *); +int BLASFUNC(xscal) (int *, double *, double *, int *); +int BLASFUNC(csscal)(int *, float *, float *, int *); +int BLASFUNC(zdscal)(int *, double *, double *, int *); +int BLASFUNC(xqscal)(int *, double *, double *, int *); + +float BLASFUNC(snrm2) (int *, float *, int *); +float BLASFUNC(scnrm2)(int *, float *, int *); + +double BLASFUNC(dnrm2) (int *, double *, int *); +double BLASFUNC(qnrm2) (int *, double *, int *); +double BLASFUNC(dznrm2)(int *, double *, int *); +double BLASFUNC(qxnrm2)(int *, double *, int *); + +int BLASFUNC(srot) (int *, float *, int *, float *, int *, float *, float *); +int BLASFUNC(drot) (int *, double *, int *, double *, int *, double *, double *); +int BLASFUNC(qrot) (int *, double *, int *, double *, int *, double *, double *); +int BLASFUNC(csrot) (int *, float *, int *, float *, int *, float *, float *); +int BLASFUNC(zdrot) (int *, double *, int *, double *, int *, double *, double *); +int BLASFUNC(xqrot) (int *, double *, int *, double *, int *, double *, double *); + +int BLASFUNC(srotg) (float *, float *, float *, float *); +int BLASFUNC(drotg) (double *, double *, double *, double *); +int BLASFUNC(qrotg) (double *, double *, double *, double *); +int BLASFUNC(crotg) (float *, float *, float *, float *); +int BLASFUNC(zrotg) (double *, double *, double *, double *); +int BLASFUNC(xrotg) (double *, double *, double *, double *); + +int BLASFUNC(srotmg)(float *, float *, float *, float *, float *); +int BLASFUNC(drotmg)(double *, double *, double *, double *, double *); + +int BLASFUNC(srotm) (int *, float *, int *, float *, int *, float *); +int BLASFUNC(drotm) (int *, double *, int *, double *, int *, double *); +int BLASFUNC(qrotm) (int *, double *, int *, double *, int *, double *); + +/* Level 2 routines */ + +int BLASFUNC(sger)(int *, int *, float *, float *, int *, + float *, int *, float *, int *); +int BLASFUNC(dger)(int *, int *, double *, double *, int *, + double *, int *, double *, int *); +int BLASFUNC(qger)(int *, int *, double *, double *, int *, + double *, int *, double *, int *); +int BLASFUNC(cgeru)(int *, int *, float *, float *, int *, + float *, int *, float *, int *); +int BLASFUNC(cgerc)(int *, int *, float *, float *, int *, + float *, int *, float *, int *); +int BLASFUNC(zgeru)(int *, int *, double *, double *, int *, + double *, int *, double *, int *); +int BLASFUNC(zgerc)(int *, int *, double *, double *, int *, + double *, int *, double *, int *); +int BLASFUNC(xgeru)(int *, int *, double *, double *, int *, + double *, int *, double *, int *); +int BLASFUNC(xgerc)(int *, int *, double *, double *, int *, + double *, int *, double *, int *); + +int BLASFUNC(sgemv)(char *, int *, int *, float *, float *, int *, + float *, int *, float *, float *, int *); +int BLASFUNC(dgemv)(char *, int *, int *, double *, double *, int *, + double *, int *, double *, double *, int *); +int BLASFUNC(qgemv)(char *, int *, int *, double *, double *, int *, + double *, int *, double *, double *, int *); +int BLASFUNC(cgemv)(char *, int *, int *, float *, float *, int *, + float *, int *, float *, float *, int *); +int BLASFUNC(zgemv)(char *, int *, int *, double *, double *, int *, + double *, int *, double *, double *, int *); +int BLASFUNC(xgemv)(char *, int *, int *, double *, double *, int *, + double *, int *, double *, double *, int *); + +int BLASFUNC(strsv) (char *, char *, char *, int *, float *, int *, + float *, int *); +int BLASFUNC(dtrsv) (char *, char *, char *, int *, double *, int *, + double *, int *); +int BLASFUNC(qtrsv) (char *, char *, char *, int *, double *, int *, + double *, int *); +int BLASFUNC(ctrsv) (char *, char *, char *, int *, float *, int *, + float *, int *); +int BLASFUNC(ztrsv) (char *, char *, char *, int *, double *, int *, + double *, int *); +int BLASFUNC(xtrsv) (char *, char *, char *, int *, double *, int *, + double *, int *); + +int BLASFUNC(stpsv) (char *, char *, char *, int *, float *, float *, int *); +int BLASFUNC(dtpsv) (char *, char *, char *, int *, double *, double *, int *); +int BLASFUNC(qtpsv) (char *, char *, char *, int *, double *, double *, int *); +int BLASFUNC(ctpsv) (char *, char *, char *, int *, float *, float *, int *); +int BLASFUNC(ztpsv) (char *, char *, char *, int *, double *, double *, int *); +int BLASFUNC(xtpsv) (char *, char *, char *, int *, double *, double *, int *); + +int BLASFUNC(strmv) (char *, char *, char *, int *, float *, int *, + float *, int *); +int BLASFUNC(dtrmv) (char *, char *, char *, int *, double *, int *, + double *, int *); +int BLASFUNC(qtrmv) (char *, char *, char *, int *, double *, int *, + double *, int *); +int BLASFUNC(ctrmv) (char *, char *, char *, int *, float *, int *, + float *, int *); +int BLASFUNC(ztrmv) (char *, char *, char *, int *, double *, int *, + double *, int *); +int BLASFUNC(xtrmv) (char *, char *, char *, int *, double *, int *, + double *, int *); + +int BLASFUNC(stpmv) (char *, char *, char *, int *, float *, float *, int *); +int BLASFUNC(dtpmv) (char *, char *, char *, int *, double *, double *, int *); +int BLASFUNC(qtpmv) (char *, char *, char *, int *, double *, double *, int *); +int BLASFUNC(ctpmv) (char *, char *, char *, int *, float *, float *, int *); +int BLASFUNC(ztpmv) (char *, char *, char *, int *, double *, double *, int *); +int BLASFUNC(xtpmv) (char *, char *, char *, int *, double *, double *, int *); + +int BLASFUNC(stbmv) (char *, char *, char *, int *, int *, float *, int *, float *, int *); +int BLASFUNC(dtbmv) (char *, char *, char *, int *, int *, double *, int *, double *, int *); +int BLASFUNC(qtbmv) (char *, char *, char *, int *, int *, double *, int *, double *, int *); +int BLASFUNC(ctbmv) (char *, char *, char *, int *, int *, float *, int *, float *, int *); +int BLASFUNC(ztbmv) (char *, char *, char *, int *, int *, double *, int *, double *, int *); +int BLASFUNC(xtbmv) (char *, char *, char *, int *, int *, double *, int *, double *, int *); + +int BLASFUNC(stbsv) (char *, char *, char *, int *, int *, float *, int *, float *, int *); +int BLASFUNC(dtbsv) (char *, char *, char *, int *, int *, double *, int *, double *, int *); +int BLASFUNC(qtbsv) (char *, char *, char *, int *, int *, double *, int *, double *, int *); +int BLASFUNC(ctbsv) (char *, char *, char *, int *, int *, float *, int *, float *, int *); +int BLASFUNC(ztbsv) (char *, char *, char *, int *, int *, double *, int *, double *, int *); +int BLASFUNC(xtbsv) (char *, char *, char *, int *, int *, double *, int *, double *, int *); + +int BLASFUNC(ssymv) (char *, int *, float *, float *, int *, + float *, int *, float *, float *, int *); +int BLASFUNC(dsymv) (char *, int *, double *, double *, int *, + double *, int *, double *, double *, int *); +int BLASFUNC(qsymv) (char *, int *, double *, double *, int *, + double *, int *, double *, double *, int *); +int BLASFUNC(csymv) (char *, int *, float *, float *, int *, + float *, int *, float *, float *, int *); +int BLASFUNC(zsymv) (char *, int *, double *, double *, int *, + double *, int *, double *, double *, int *); +int BLASFUNC(xsymv) (char *, int *, double *, double *, int *, + double *, int *, double *, double *, int *); + +int BLASFUNC(sspmv) (char *, int *, float *, float *, + float *, int *, float *, float *, int *); +int BLASFUNC(dspmv) (char *, int *, double *, double *, + double *, int *, double *, double *, int *); +int BLASFUNC(qspmv) (char *, int *, double *, double *, + double *, int *, double *, double *, int *); +int BLASFUNC(cspmv) (char *, int *, float *, float *, + float *, int *, float *, float *, int *); +int BLASFUNC(zspmv) (char *, int *, double *, double *, + double *, int *, double *, double *, int *); +int BLASFUNC(xspmv) (char *, int *, double *, double *, + double *, int *, double *, double *, int *); + +int BLASFUNC(ssyr) (char *, int *, float *, float *, int *, + float *, int *); +int BLASFUNC(dsyr) (char *, int *, double *, double *, int *, + double *, int *); +int BLASFUNC(qsyr) (char *, int *, double *, double *, int *, + double *, int *); +int BLASFUNC(csyr) (char *, int *, float *, float *, int *, + float *, int *); +int BLASFUNC(zsyr) (char *, int *, double *, double *, int *, + double *, int *); +int BLASFUNC(xsyr) (char *, int *, double *, double *, int *, + double *, int *); + +int BLASFUNC(ssyr2) (char *, int *, float *, + float *, int *, float *, int *, float *, int *); +int BLASFUNC(dsyr2) (char *, int *, double *, + double *, int *, double *, int *, double *, int *); +int BLASFUNC(qsyr2) (char *, int *, double *, + double *, int *, double *, int *, double *, int *); +int BLASFUNC(csyr2) (char *, int *, float *, + float *, int *, float *, int *, float *, int *); +int BLASFUNC(zsyr2) (char *, int *, double *, + double *, int *, double *, int *, double *, int *); +int BLASFUNC(xsyr2) (char *, int *, double *, + double *, int *, double *, int *, double *, int *); + +int BLASFUNC(sspr) (char *, int *, float *, float *, int *, + float *); +int BLASFUNC(dspr) (char *, int *, double *, double *, int *, + double *); +int BLASFUNC(qspr) (char *, int *, double *, double *, int *, + double *); +int BLASFUNC(cspr) (char *, int *, float *, float *, int *, + float *); +int BLASFUNC(zspr) (char *, int *, double *, double *, int *, + double *); +int BLASFUNC(xspr) (char *, int *, double *, double *, int *, + double *); + +int BLASFUNC(sspr2) (char *, int *, float *, + float *, int *, float *, int *, float *); +int BLASFUNC(dspr2) (char *, int *, double *, + double *, int *, double *, int *, double *); +int BLASFUNC(qspr2) (char *, int *, double *, + double *, int *, double *, int *, double *); +int BLASFUNC(cspr2) (char *, int *, float *, + float *, int *, float *, int *, float *); +int BLASFUNC(zspr2) (char *, int *, double *, + double *, int *, double *, int *, double *); +int BLASFUNC(xspr2) (char *, int *, double *, + double *, int *, double *, int *, double *); + +int BLASFUNC(cher) (char *, int *, float *, float *, int *, + float *, int *); +int BLASFUNC(zher) (char *, int *, double *, double *, int *, + double *, int *); +int BLASFUNC(xher) (char *, int *, double *, double *, int *, + double *, int *); + +int BLASFUNC(chpr) (char *, int *, float *, float *, int *, float *); +int BLASFUNC(zhpr) (char *, int *, double *, double *, int *, double *); +int BLASFUNC(xhpr) (char *, int *, double *, double *, int *, double *); + +int BLASFUNC(cher2) (char *, int *, float *, + float *, int *, float *, int *, float *, int *); +int BLASFUNC(zher2) (char *, int *, double *, + double *, int *, double *, int *, double *, int *); +int BLASFUNC(xher2) (char *, int *, double *, + double *, int *, double *, int *, double *, int *); + +int BLASFUNC(chpr2) (char *, int *, float *, + float *, int *, float *, int *, float *); +int BLASFUNC(zhpr2) (char *, int *, double *, + double *, int *, double *, int *, double *); +int BLASFUNC(xhpr2) (char *, int *, double *, + double *, int *, double *, int *, double *); + +int BLASFUNC(chemv) (char *, int *, float *, float *, int *, + float *, int *, float *, float *, int *); +int BLASFUNC(zhemv) (char *, int *, double *, double *, int *, + double *, int *, double *, double *, int *); +int BLASFUNC(xhemv) (char *, int *, double *, double *, int *, + double *, int *, double *, double *, int *); + +int BLASFUNC(chpmv) (char *, int *, float *, float *, + float *, int *, float *, float *, int *); +int BLASFUNC(zhpmv) (char *, int *, double *, double *, + double *, int *, double *, double *, int *); +int BLASFUNC(xhpmv) (char *, int *, double *, double *, + double *, int *, double *, double *, int *); + +int BLASFUNC(snorm)(char *, int *, int *, float *, int *); +int BLASFUNC(dnorm)(char *, int *, int *, double *, int *); +int BLASFUNC(cnorm)(char *, int *, int *, float *, int *); +int BLASFUNC(znorm)(char *, int *, int *, double *, int *); + +int BLASFUNC(sgbmv)(char *, int *, int *, int *, int *, float *, float *, int *, + float *, int *, float *, float *, int *); +int BLASFUNC(dgbmv)(char *, int *, int *, int *, int *, double *, double *, int *, + double *, int *, double *, double *, int *); +int BLASFUNC(qgbmv)(char *, int *, int *, int *, int *, double *, double *, int *, + double *, int *, double *, double *, int *); +int BLASFUNC(cgbmv)(char *, int *, int *, int *, int *, float *, float *, int *, + float *, int *, float *, float *, int *); +int BLASFUNC(zgbmv)(char *, int *, int *, int *, int *, double *, double *, int *, + double *, int *, double *, double *, int *); +int BLASFUNC(xgbmv)(char *, int *, int *, int *, int *, double *, double *, int *, + double *, int *, double *, double *, int *); + +int BLASFUNC(ssbmv)(char *, int *, int *, float *, float *, int *, + float *, int *, float *, float *, int *); +int BLASFUNC(dsbmv)(char *, int *, int *, double *, double *, int *, + double *, int *, double *, double *, int *); +int BLASFUNC(qsbmv)(char *, int *, int *, double *, double *, int *, + double *, int *, double *, double *, int *); +int BLASFUNC(csbmv)(char *, int *, int *, float *, float *, int *, + float *, int *, float *, float *, int *); +int BLASFUNC(zsbmv)(char *, int *, int *, double *, double *, int *, + double *, int *, double *, double *, int *); +int BLASFUNC(xsbmv)(char *, int *, int *, double *, double *, int *, + double *, int *, double *, double *, int *); + +int BLASFUNC(chbmv)(char *, int *, int *, float *, float *, int *, + float *, int *, float *, float *, int *); +int BLASFUNC(zhbmv)(char *, int *, int *, double *, double *, int *, + double *, int *, double *, double *, int *); +int BLASFUNC(xhbmv)(char *, int *, int *, double *, double *, int *, + double *, int *, double *, double *, int *); + +/* Level 3 routines */ + +int BLASFUNC(sgemm)(char *, char *, int *, int *, int *, float *, + float *, int *, float *, int *, float *, float *, int *); +int BLASFUNC(dgemm)(char *, char *, int *, int *, int *, double *, + double *, int *, double *, int *, double *, double *, int *); +int BLASFUNC(qgemm)(char *, char *, int *, int *, int *, double *, + double *, int *, double *, int *, double *, double *, int *); +int BLASFUNC(cgemm)(char *, char *, int *, int *, int *, float *, + float *, int *, float *, int *, float *, float *, int *); +int BLASFUNC(zgemm)(char *, char *, int *, int *, int *, double *, + double *, int *, double *, int *, double *, double *, int *); +int BLASFUNC(xgemm)(char *, char *, int *, int *, int *, double *, + double *, int *, double *, int *, double *, double *, int *); + +int BLASFUNC(cgemm3m)(char *, char *, int *, int *, int *, float *, + float *, int *, float *, int *, float *, float *, int *); +int BLASFUNC(zgemm3m)(char *, char *, int *, int *, int *, double *, + double *, int *, double *, int *, double *, double *, int *); +int BLASFUNC(xgemm3m)(char *, char *, int *, int *, int *, double *, + double *, int *, double *, int *, double *, double *, int *); + +int BLASFUNC(sge2mm)(char *, char *, char *, int *, int *, + float *, float *, int *, float *, int *, + float *, float *, int *); +int BLASFUNC(dge2mm)(char *, char *, char *, int *, int *, + double *, double *, int *, double *, int *, + double *, double *, int *); +int BLASFUNC(cge2mm)(char *, char *, char *, int *, int *, + float *, float *, int *, float *, int *, + float *, float *, int *); +int BLASFUNC(zge2mm)(char *, char *, char *, int *, int *, + double *, double *, int *, double *, int *, + double *, double *, int *); + +int BLASFUNC(strsm)(char *, char *, char *, char *, int *, int *, + float *, float *, int *, float *, int *); +int BLASFUNC(dtrsm)(char *, char *, char *, char *, int *, int *, + double *, double *, int *, double *, int *); +int BLASFUNC(qtrsm)(char *, char *, char *, char *, int *, int *, + double *, double *, int *, double *, int *); +int BLASFUNC(ctrsm)(char *, char *, char *, char *, int *, int *, + float *, float *, int *, float *, int *); +int BLASFUNC(ztrsm)(char *, char *, char *, char *, int *, int *, + double *, double *, int *, double *, int *); +int BLASFUNC(xtrsm)(char *, char *, char *, char *, int *, int *, + double *, double *, int *, double *, int *); + +int BLASFUNC(strmm)(char *, char *, char *, char *, int *, int *, + float *, float *, int *, float *, int *); +int BLASFUNC(dtrmm)(char *, char *, char *, char *, int *, int *, + double *, double *, int *, double *, int *); +int BLASFUNC(qtrmm)(char *, char *, char *, char *, int *, int *, + double *, double *, int *, double *, int *); +int BLASFUNC(ctrmm)(char *, char *, char *, char *, int *, int *, + float *, float *, int *, float *, int *); +int BLASFUNC(ztrmm)(char *, char *, char *, char *, int *, int *, + double *, double *, int *, double *, int *); +int BLASFUNC(xtrmm)(char *, char *, char *, char *, int *, int *, + double *, double *, int *, double *, int *); + +int BLASFUNC(ssymm)(char *, char *, int *, int *, float *, float *, int *, + float *, int *, float *, float *, int *); +int BLASFUNC(dsymm)(char *, char *, int *, int *, double *, double *, int *, + double *, int *, double *, double *, int *); +int BLASFUNC(qsymm)(char *, char *, int *, int *, double *, double *, int *, + double *, int *, double *, double *, int *); +int BLASFUNC(csymm)(char *, char *, int *, int *, float *, float *, int *, + float *, int *, float *, float *, int *); +int BLASFUNC(zsymm)(char *, char *, int *, int *, double *, double *, int *, + double *, int *, double *, double *, int *); +int BLASFUNC(xsymm)(char *, char *, int *, int *, double *, double *, int *, + double *, int *, double *, double *, int *); + +int BLASFUNC(csymm3m)(char *, char *, int *, int *, float *, float *, int *, + float *, int *, float *, float *, int *); +int BLASFUNC(zsymm3m)(char *, char *, int *, int *, double *, double *, int *, + double *, int *, double *, double *, int *); +int BLASFUNC(xsymm3m)(char *, char *, int *, int *, double *, double *, int *, + double *, int *, double *, double *, int *); + +int BLASFUNC(ssyrk)(char *, char *, int *, int *, float *, float *, int *, + float *, float *, int *); +int BLASFUNC(dsyrk)(char *, char *, int *, int *, double *, double *, int *, + double *, double *, int *); +int BLASFUNC(qsyrk)(char *, char *, int *, int *, double *, double *, int *, + double *, double *, int *); +int BLASFUNC(csyrk)(char *, char *, int *, int *, float *, float *, int *, + float *, float *, int *); +int BLASFUNC(zsyrk)(char *, char *, int *, int *, double *, double *, int *, + double *, double *, int *); +int BLASFUNC(xsyrk)(char *, char *, int *, int *, double *, double *, int *, + double *, double *, int *); + +int BLASFUNC(ssyr2k)(char *, char *, int *, int *, float *, float *, int *, + float *, int *, float *, float *, int *); +int BLASFUNC(dsyr2k)(char *, char *, int *, int *, double *, double *, int *, + double*, int *, double *, double *, int *); +int BLASFUNC(qsyr2k)(char *, char *, int *, int *, double *, double *, int *, + double*, int *, double *, double *, int *); +int BLASFUNC(csyr2k)(char *, char *, int *, int *, float *, float *, int *, + float *, int *, float *, float *, int *); +int BLASFUNC(zsyr2k)(char *, char *, int *, int *, double *, double *, int *, + double*, int *, double *, double *, int *); +int BLASFUNC(xsyr2k)(char *, char *, int *, int *, double *, double *, int *, + double*, int *, double *, double *, int *); + +int BLASFUNC(chemm)(char *, char *, int *, int *, float *, float *, int *, + float *, int *, float *, float *, int *); +int BLASFUNC(zhemm)(char *, char *, int *, int *, double *, double *, int *, + double *, int *, double *, double *, int *); +int BLASFUNC(xhemm)(char *, char *, int *, int *, double *, double *, int *, + double *, int *, double *, double *, int *); + +int BLASFUNC(chemm3m)(char *, char *, int *, int *, float *, float *, int *, + float *, int *, float *, float *, int *); +int BLASFUNC(zhemm3m)(char *, char *, int *, int *, double *, double *, int *, + double *, int *, double *, double *, int *); +int BLASFUNC(xhemm3m)(char *, char *, int *, int *, double *, double *, int *, + double *, int *, double *, double *, int *); + +int BLASFUNC(cherk)(char *, char *, int *, int *, float *, float *, int *, + float *, float *, int *); +int BLASFUNC(zherk)(char *, char *, int *, int *, double *, double *, int *, + double *, double *, int *); +int BLASFUNC(xherk)(char *, char *, int *, int *, double *, double *, int *, + double *, double *, int *); + +int BLASFUNC(cher2k)(char *, char *, int *, int *, float *, float *, int *, + float *, int *, float *, float *, int *); +int BLASFUNC(zher2k)(char *, char *, int *, int *, double *, double *, int *, + double*, int *, double *, double *, int *); +int BLASFUNC(xher2k)(char *, char *, int *, int *, double *, double *, int *, + double*, int *, double *, double *, int *); +int BLASFUNC(cher2m)(char *, char *, char *, int *, int *, float *, float *, int *, + float *, int *, float *, float *, int *); +int BLASFUNC(zher2m)(char *, char *, char *, int *, int *, double *, double *, int *, + double*, int *, double *, double *, int *); +int BLASFUNC(xher2m)(char *, char *, char *, int *, int *, double *, double *, int *, + double*, int *, double *, double *, int *); + +int BLASFUNC(sgemt)(char *, int *, int *, float *, float *, int *, + float *, int *); +int BLASFUNC(dgemt)(char *, int *, int *, double *, double *, int *, + double *, int *); +int BLASFUNC(cgemt)(char *, int *, int *, float *, float *, int *, + float *, int *); +int BLASFUNC(zgemt)(char *, int *, int *, double *, double *, int *, + double *, int *); + +int BLASFUNC(sgema)(char *, char *, int *, int *, float *, + float *, int *, float *, float *, int *, float *, int *); +int BLASFUNC(dgema)(char *, char *, int *, int *, double *, + double *, int *, double*, double *, int *, double*, int *); +int BLASFUNC(cgema)(char *, char *, int *, int *, float *, + float *, int *, float *, float *, int *, float *, int *); +int BLASFUNC(zgema)(char *, char *, int *, int *, double *, + double *, int *, double*, double *, int *, double*, int *); + +int BLASFUNC(sgems)(char *, char *, int *, int *, float *, + float *, int *, float *, float *, int *, float *, int *); +int BLASFUNC(dgems)(char *, char *, int *, int *, double *, + double *, int *, double*, double *, int *, double*, int *); +int BLASFUNC(cgems)(char *, char *, int *, int *, float *, + float *, int *, float *, float *, int *, float *, int *); +int BLASFUNC(zgems)(char *, char *, int *, int *, double *, + double *, int *, double*, double *, int *, double*, int *); + +int BLASFUNC(sgetf2)(int *, int *, float *, int *, int *, int *); +int BLASFUNC(dgetf2)(int *, int *, double *, int *, int *, int *); +int BLASFUNC(qgetf2)(int *, int *, double *, int *, int *, int *); +int BLASFUNC(cgetf2)(int *, int *, float *, int *, int *, int *); +int BLASFUNC(zgetf2)(int *, int *, double *, int *, int *, int *); +int BLASFUNC(xgetf2)(int *, int *, double *, int *, int *, int *); + +int BLASFUNC(sgetrf)(int *, int *, float *, int *, int *, int *); +int BLASFUNC(dgetrf)(int *, int *, double *, int *, int *, int *); +int BLASFUNC(qgetrf)(int *, int *, double *, int *, int *, int *); +int BLASFUNC(cgetrf)(int *, int *, float *, int *, int *, int *); +int BLASFUNC(zgetrf)(int *, int *, double *, int *, int *, int *); +int BLASFUNC(xgetrf)(int *, int *, double *, int *, int *, int *); + +int BLASFUNC(slaswp)(int *, float *, int *, int *, int *, int *, int *); +int BLASFUNC(dlaswp)(int *, double *, int *, int *, int *, int *, int *); +int BLASFUNC(qlaswp)(int *, double *, int *, int *, int *, int *, int *); +int BLASFUNC(claswp)(int *, float *, int *, int *, int *, int *, int *); +int BLASFUNC(zlaswp)(int *, double *, int *, int *, int *, int *, int *); +int BLASFUNC(xlaswp)(int *, double *, int *, int *, int *, int *, int *); + +int BLASFUNC(sgetrs)(char *, int *, int *, float *, int *, int *, float *, int *, int *); +int BLASFUNC(dgetrs)(char *, int *, int *, double *, int *, int *, double *, int *, int *); +int BLASFUNC(qgetrs)(char *, int *, int *, double *, int *, int *, double *, int *, int *); +int BLASFUNC(cgetrs)(char *, int *, int *, float *, int *, int *, float *, int *, int *); +int BLASFUNC(zgetrs)(char *, int *, int *, double *, int *, int *, double *, int *, int *); +int BLASFUNC(xgetrs)(char *, int *, int *, double *, int *, int *, double *, int *, int *); + +int BLASFUNC(sgesv)(int *, int *, float *, int *, int *, float *, int *, int *); +int BLASFUNC(dgesv)(int *, int *, double *, int *, int *, double*, int *, int *); +int BLASFUNC(qgesv)(int *, int *, double *, int *, int *, double*, int *, int *); +int BLASFUNC(cgesv)(int *, int *, float *, int *, int *, float *, int *, int *); +int BLASFUNC(zgesv)(int *, int *, double *, int *, int *, double*, int *, int *); +int BLASFUNC(xgesv)(int *, int *, double *, int *, int *, double*, int *, int *); + +int BLASFUNC(spotf2)(char *, int *, float *, int *, int *); +int BLASFUNC(dpotf2)(char *, int *, double *, int *, int *); +int BLASFUNC(qpotf2)(char *, int *, double *, int *, int *); +int BLASFUNC(cpotf2)(char *, int *, float *, int *, int *); +int BLASFUNC(zpotf2)(char *, int *, double *, int *, int *); +int BLASFUNC(xpotf2)(char *, int *, double *, int *, int *); + +int BLASFUNC(spotrf)(char *, int *, float *, int *, int *); +int BLASFUNC(dpotrf)(char *, int *, double *, int *, int *); +int BLASFUNC(qpotrf)(char *, int *, double *, int *, int *); +int BLASFUNC(cpotrf)(char *, int *, float *, int *, int *); +int BLASFUNC(zpotrf)(char *, int *, double *, int *, int *); +int BLASFUNC(xpotrf)(char *, int *, double *, int *, int *); + +int BLASFUNC(slauu2)(char *, int *, float *, int *, int *); +int BLASFUNC(dlauu2)(char *, int *, double *, int *, int *); +int BLASFUNC(qlauu2)(char *, int *, double *, int *, int *); +int BLASFUNC(clauu2)(char *, int *, float *, int *, int *); +int BLASFUNC(zlauu2)(char *, int *, double *, int *, int *); +int BLASFUNC(xlauu2)(char *, int *, double *, int *, int *); + +int BLASFUNC(slauum)(char *, int *, float *, int *, int *); +int BLASFUNC(dlauum)(char *, int *, double *, int *, int *); +int BLASFUNC(qlauum)(char *, int *, double *, int *, int *); +int BLASFUNC(clauum)(char *, int *, float *, int *, int *); +int BLASFUNC(zlauum)(char *, int *, double *, int *, int *); +int BLASFUNC(xlauum)(char *, int *, double *, int *, int *); + +int BLASFUNC(strti2)(char *, char *, int *, float *, int *, int *); +int BLASFUNC(dtrti2)(char *, char *, int *, double *, int *, int *); +int BLASFUNC(qtrti2)(char *, char *, int *, double *, int *, int *); +int BLASFUNC(ctrti2)(char *, char *, int *, float *, int *, int *); +int BLASFUNC(ztrti2)(char *, char *, int *, double *, int *, int *); +int BLASFUNC(xtrti2)(char *, char *, int *, double *, int *, int *); + +int BLASFUNC(strtri)(char *, char *, int *, float *, int *, int *); +int BLASFUNC(dtrtri)(char *, char *, int *, double *, int *, int *); +int BLASFUNC(qtrtri)(char *, char *, int *, double *, int *, int *); +int BLASFUNC(ctrtri)(char *, char *, int *, float *, int *, int *); +int BLASFUNC(ztrtri)(char *, char *, int *, double *, int *, int *); +int BLASFUNC(xtrtri)(char *, char *, int *, double *, int *, int *); + +int BLASFUNC(spotri)(char *, int *, float *, int *, int *); +int BLASFUNC(dpotri)(char *, int *, double *, int *, int *); +int BLASFUNC(qpotri)(char *, int *, double *, int *, int *); +int BLASFUNC(cpotri)(char *, int *, float *, int *, int *); +int BLASFUNC(zpotri)(char *, int *, double *, int *, int *); +int BLASFUNC(xpotri)(char *, int *, double *, int *, int *); + +#endif diff --git a/bench/btl/libs/BLAS/blas_interface.hh b/bench/btl/libs/BLAS/blas_interface.hh new file mode 100644 index 000000000..651054632 --- /dev/null +++ b/bench/btl/libs/BLAS/blas_interface.hh @@ -0,0 +1,83 @@ +//===================================================== +// File : blas_interface.hh +// Author : L. Plagne <laurent.plagne@edf.fr)> +// Copyright (C) EDF R&D, lun sep 30 14:23:28 CEST 2002 +//===================================================== +// +// This program is free software; you can redistribute it and/or +// modify it under the terms of the GNU General Public License +// as published by the Free Software Foundation; either version 2 +// of the License, or (at your option) any later version. +// +// This program is distributed in the hope that it will be useful, +// but WITHOUT ANY WARRANTY; without even the implied warranty of +// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +// GNU General Public License for more details. +// You should have received a copy of the GNU General Public License +// along with this program; if not, write to the Free Software +// Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. +// +#ifndef blas_PRODUIT_MATRICE_VECTEUR_HH +#define blas_PRODUIT_MATRICE_VECTEUR_HH + +#include <c_interface_base.h> +#include <complex> +extern "C" +{ +#include "blas.h" + + // Cholesky Factorization +// void spotrf_(const char* uplo, const int* n, float *a, const int* ld, int* info); +// void dpotrf_(const char* uplo, const int* n, double *a, const int* ld, int* info); + void ssytrd_(char *uplo, const int *n, float *a, const int *lda, float *d, float *e, float *tau, float *work, int *lwork, int *info ); + void dsytrd_(char *uplo, const int *n, double *a, const int *lda, double *d, double *e, double *tau, double *work, int *lwork, int *info ); + void sgehrd_( const int *n, int *ilo, int *ihi, float *a, const int *lda, float *tau, float *work, int *lwork, int *info ); + void dgehrd_( const int *n, int *ilo, int *ihi, double *a, const int *lda, double *tau, double *work, int *lwork, int *info ); + + // LU row pivoting +// void dgetrf_( int *m, int *n, double *a, int *lda, int *ipiv, int *info ); +// void sgetrf_(const int* m, const int* n, float *a, const int* ld, int* ipivot, int* info); + // LU full pivoting + void sgetc2_(const int* n, float *a, const int *lda, int *ipiv, int *jpiv, int*info ); + void dgetc2_(const int* n, double *a, const int *lda, int *ipiv, int *jpiv, int*info ); +#ifdef HAS_LAPACK +#endif +} + +#define MAKE_STRING2(S) #S +#define MAKE_STRING(S) MAKE_STRING2(S) + +#define CAT2(A,B) A##B +#define CAT(A,B) CAT2(A,B) + + +template<class real> class blas_interface; + + +static char notrans = 'N'; +static char trans = 'T'; +static char nonunit = 'N'; +static char lower = 'L'; +static char right = 'R'; +static char left = 'L'; +static int intone = 1; + + + +#define SCALAR float +#define SCALAR_PREFIX s +#include "blas_interface_impl.hh" +#undef SCALAR +#undef SCALAR_PREFIX + + +#define SCALAR double +#define SCALAR_PREFIX d +#include "blas_interface_impl.hh" +#undef SCALAR +#undef SCALAR_PREFIX + +#endif + + + diff --git a/bench/btl/libs/BLAS/blas_interface_impl.hh b/bench/btl/libs/BLAS/blas_interface_impl.hh new file mode 100644 index 000000000..0e84df038 --- /dev/null +++ b/bench/btl/libs/BLAS/blas_interface_impl.hh @@ -0,0 +1,151 @@ + +#define BLAS_FUNC(NAME) CAT(CAT(SCALAR_PREFIX,NAME),_) + +template<> class blas_interface<SCALAR> : public c_interface_base<SCALAR> +{ + +public : + + static SCALAR fone; + static SCALAR fzero; + + static inline std::string name() + { + return MAKE_STRING(CBLASNAME); + } + + static inline void matrix_vector_product(gene_matrix & A, gene_vector & B, gene_vector & X, int N){ + BLAS_FUNC(gemv)(¬rans,&N,&N,&fone,A,&N,B,&intone,&fzero,X,&intone); + } + + static inline void symv(gene_matrix & A, gene_vector & B, gene_vector & X, int N){ + BLAS_FUNC(symv)(&lower, &N,&fone,A,&N,B,&intone,&fzero,X,&intone); + } + + static inline void syr2(gene_matrix & A, gene_vector & B, gene_vector & X, int N){ + BLAS_FUNC(syr2)(&lower,&N,&fone,B,&intone,X,&intone,A,&N); + } + + static inline void ger(gene_matrix & A, gene_vector & X, gene_vector & Y, int N){ + BLAS_FUNC(ger)(&N,&N,&fone,X,&intone,Y,&intone,A,&N); + } + + static inline void rot(gene_vector & A, gene_vector & B, SCALAR c, SCALAR s, int N){ + BLAS_FUNC(rot)(&N,A,&intone,B,&intone,&c,&s); + } + + static inline void atv_product(gene_matrix & A, gene_vector & B, gene_vector & X, int N){ + BLAS_FUNC(gemv)(&trans,&N,&N,&fone,A,&N,B,&intone,&fzero,X,&intone); + } + + static inline void matrix_matrix_product(gene_matrix & A, gene_matrix & B, gene_matrix & X, int N){ + BLAS_FUNC(gemm)(¬rans,¬rans,&N,&N,&N,&fone,A,&N,B,&N,&fzero,X,&N); + } + + static inline void transposed_matrix_matrix_product(gene_matrix & A, gene_matrix & B, gene_matrix & X, int N){ + BLAS_FUNC(gemm)(¬rans,¬rans,&N,&N,&N,&fone,A,&N,B,&N,&fzero,X,&N); + } + +// static inline void ata_product(gene_matrix & A, gene_matrix & X, int N){ +// ssyrk_(&lower,&trans,&N,&N,&fone,A,&N,&fzero,X,&N); +// } + + static inline void aat_product(gene_matrix & A, gene_matrix & X, int N){ + BLAS_FUNC(syrk)(&lower,¬rans,&N,&N,&fone,A,&N,&fzero,X,&N); + } + + static inline void axpy(SCALAR coef, const gene_vector & X, gene_vector & Y, int N){ + BLAS_FUNC(axpy)(&N,&coef,X,&intone,Y,&intone); + } + + static inline void axpby(SCALAR a, const gene_vector & X, SCALAR b, gene_vector & Y, int N){ + BLAS_FUNC(scal)(&N,&b,Y,&intone); + BLAS_FUNC(axpy)(&N,&a,X,&intone,Y,&intone); + } + + static inline void cholesky(const gene_matrix & X, gene_matrix & C, int N){ + int N2 = N*N; + BLAS_FUNC(copy)(&N2, X, &intone, C, &intone); + char uplo = 'L'; + int info = 0; + BLAS_FUNC(potrf)(&uplo, &N, C, &N, &info); + if(info!=0) std::cerr << "potrf_ error " << info << "\n"; + } + + static inline void partial_lu_decomp(const gene_matrix & X, gene_matrix & C, int N){ + int N2 = N*N; + BLAS_FUNC(copy)(&N2, X, &intone, C, &intone); + char uplo = 'L'; + int info = 0; + int * ipiv = (int*)alloca(sizeof(int)*N); + BLAS_FUNC(getrf)(&N, &N, C, &N, ipiv, &info); + if(info!=0) std::cerr << "getrf_ error " << info << "\n"; + } + + static inline void trisolve_lower(const gene_matrix & L, const gene_vector& B, gene_vector & X, int N){ + BLAS_FUNC(copy)(&N, B, &intone, X, &intone); + BLAS_FUNC(trsv)(&lower, ¬rans, &nonunit, &N, L, &N, X, &intone); + } + + static inline void trisolve_lower_matrix(const gene_matrix & L, const gene_matrix& B, gene_matrix & X, int N){ + BLAS_FUNC(copy)(&N, B, &intone, X, &intone); + BLAS_FUNC(trsm)(&right, &lower, ¬rans, &nonunit, &N, &N, &fone, L, &N, X, &N); + } + + static inline void trmm(gene_matrix & A, gene_matrix & B, gene_matrix & X, int N){ + BLAS_FUNC(trmm)(&left, &lower, ¬rans,&nonunit, &N,&N,&fone,A,&N,B,&N); + } + + #ifdef HAS_LAPACK + + static inline void lu_decomp(const gene_matrix & X, gene_matrix & C, int N){ + int N2 = N*N; + BLAS_FUNC(copy)(&N2, X, &intone, C, &intone); + char uplo = 'L'; + int info = 0; + int * ipiv = (int*)alloca(sizeof(int)*N); + int * jpiv = (int*)alloca(sizeof(int)*N); + BLAS_FUNC(getc2)(&N, C, &N, ipiv, jpiv, &info); + } + + + + static inline void hessenberg(const gene_matrix & X, gene_matrix & C, int N){ + { + int N2 = N*N; + int inc = 1; + BLAS_FUNC(copy)(&N2, X, &inc, C, &inc); + } + int info = 0; + int ilo = 1; + int ihi = N; + int bsize = 64; + int worksize = N*bsize; + SCALAR* d = new SCALAR[N+worksize]; + BLAS_FUNC(gehrd)(&N, &ilo, &ihi, C, &N, d, d+N, &worksize, &info); + delete[] d; + } + + static inline void tridiagonalization(const gene_matrix & X, gene_matrix & C, int N){ + { + int N2 = N*N; + int inc = 1; + BLAS_FUNC(copy)(&N2, X, &inc, C, &inc); + } + char uplo = 'U'; + int info = 0; + int ilo = 1; + int ihi = N; + int bsize = 64; + int worksize = N*bsize; + SCALAR* d = new SCALAR[3*N+worksize]; + BLAS_FUNC(sytrd)(&uplo, &N, C, &N, d, d+N, d+2*N, d+3*N, &worksize, &info); + delete[] d; + } + + #endif // HAS_LAPACK + +}; + +SCALAR blas_interface<SCALAR>::fone = SCALAR(1); +SCALAR blas_interface<SCALAR>::fzero = SCALAR(0); diff --git a/bench/btl/libs/BLAS/c_interface_base.h b/bench/btl/libs/BLAS/c_interface_base.h new file mode 100644 index 000000000..515d8dcfc --- /dev/null +++ b/bench/btl/libs/BLAS/c_interface_base.h @@ -0,0 +1,73 @@ + +#ifndef BTL_C_INTERFACE_BASE_H +#define BTL_C_INTERFACE_BASE_H + +#include "utilities.h" +#include <vector> + +template<class real> class c_interface_base +{ + +public: + + typedef real real_type; + typedef std::vector<real> stl_vector; + typedef std::vector<stl_vector > stl_matrix; + + typedef real* gene_matrix; + typedef real* gene_vector; + + static void free_matrix(gene_matrix & A, int N){ + delete A; + } + + static void free_vector(gene_vector & B){ + delete B; + } + + static inline void matrix_from_stl(gene_matrix & A, stl_matrix & A_stl){ + int N = A_stl.size(); + A = new real[N*N]; + for (int j=0;j<N;j++) + for (int i=0;i<N;i++) + A[i+N*j] = A_stl[j][i]; + } + + static inline void vector_from_stl(gene_vector & B, stl_vector & B_stl){ + int N = B_stl.size(); + B = new real[N]; + for (int i=0;i<N;i++) + B[i] = B_stl[i]; + } + + static inline void vector_to_stl(gene_vector & B, stl_vector & B_stl){ + int N = B_stl.size(); + for (int i=0;i<N;i++) + B_stl[i] = B[i]; + } + + static inline void matrix_to_stl(gene_matrix & A, stl_matrix & A_stl){ + int N = A_stl.size(); + for (int j=0;j<N;j++){ + A_stl[j].resize(N); + for (int i=0;i<N;i++) + A_stl[j][i] = A[i+N*j]; + } + } + + static inline void copy_vector(const gene_vector & source, gene_vector & cible, int N){ + for (int i=0;i<N;i++) + cible[i]=source[i]; + } + + static inline void copy_matrix(const gene_matrix & source, gene_matrix & cible, int N){ + for (int j=0;j<N;j++){ + for (int i=0;i<N;i++){ + cible[i+N*j] = source[i+N*j]; + } + } + } + +}; + +#endif diff --git a/bench/btl/libs/BLAS/main.cpp b/bench/btl/libs/BLAS/main.cpp new file mode 100644 index 000000000..8347c9f0b --- /dev/null +++ b/bench/btl/libs/BLAS/main.cpp @@ -0,0 +1,73 @@ +//===================================================== +// File : main.cpp +// Author : L. Plagne <laurent.plagne@edf.fr)> +// Copyright (C) EDF R&D, lun sep 30 14:23:28 CEST 2002 +//===================================================== +// +// This program is free software; you can redistribute it and/or +// modify it under the terms of the GNU General Public License +// as published by the Free Software Foundation; either version 2 +// of the License, or (at your option) any later version. +// +// This program is distributed in the hope that it will be useful, +// but WITHOUT ANY WARRANTY; without even the implied warranty of +// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +// GNU General Public License for more details. +// You should have received a copy of the GNU General Public License +// along with this program; if not, write to the Free Software +// Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. +// +#include "utilities.h" +#include "blas_interface.hh" +#include "bench.hh" +#include "basic_actions.hh" + +#include "action_cholesky.hh" +#include "action_lu_decomp.hh" +#include "action_partial_lu.hh" +#include "action_trisolve_matrix.hh" + +#ifdef HAS_LAPACK +#include "action_hessenberg.hh" +#endif + +BTL_MAIN; + +int main() +{ + + bench<Action_axpy<blas_interface<REAL_TYPE> > >(MIN_AXPY,MAX_AXPY,NB_POINT); + bench<Action_axpby<blas_interface<REAL_TYPE> > >(MIN_AXPY,MAX_AXPY,NB_POINT); + + bench<Action_matrix_vector_product<blas_interface<REAL_TYPE> > >(MIN_MV,MAX_MV,NB_POINT); + bench<Action_atv_product<blas_interface<REAL_TYPE> > >(MIN_MV,MAX_MV,NB_POINT); + bench<Action_symv<blas_interface<REAL_TYPE> > >(MIN_MV,MAX_MV,NB_POINT); + bench<Action_syr2<blas_interface<REAL_TYPE> > >(MIN_MV,MAX_MV,NB_POINT); + + bench<Action_ger<blas_interface<REAL_TYPE> > >(MIN_MV,MAX_MV,NB_POINT); + bench<Action_rot<blas_interface<REAL_TYPE> > >(MIN_AXPY,MAX_AXPY,NB_POINT); + + bench<Action_matrix_matrix_product<blas_interface<REAL_TYPE> > >(MIN_MM,MAX_MM,NB_POINT); +// bench<Action_ata_product<blas_interface<REAL_TYPE> > >(MIN_MM,MAX_MM,NB_POINT); + bench<Action_aat_product<blas_interface<REAL_TYPE> > >(MIN_MM,MAX_MM,NB_POINT); + + bench<Action_trisolve<blas_interface<REAL_TYPE> > >(MIN_MM,MAX_MM,NB_POINT); + bench<Action_trisolve_matrix<blas_interface<REAL_TYPE> > >(MIN_MM,MAX_MM,NB_POINT); + + bench<Action_trmm<blas_interface<REAL_TYPE> > >(MIN_MM,MAX_MM,NB_POINT); + + bench<Action_cholesky<blas_interface<REAL_TYPE> > >(MIN_MM,MAX_MM,NB_POINT); + bench<Action_partial_lu<blas_interface<REAL_TYPE> > >(MIN_MM,MAX_MM,NB_POINT); + + #ifdef HAS_LAPACK + bench<Action_lu_decomp<blas_interface<REAL_TYPE> > >(MIN_MM,MAX_MM,NB_POINT); + bench<Action_hessenberg<blas_interface<REAL_TYPE> > >(MIN_MM,MAX_MM,NB_POINT); + bench<Action_tridiagonalization<blas_interface<REAL_TYPE> > >(MIN_MM,MAX_MM,NB_POINT); + #endif + + //bench<Action_lu_solve<blas_LU_solve_interface<REAL_TYPE> > >(MIN_LU,MAX_LU,NB_POINT); + + return 0; +} + + diff --git a/bench/btl/libs/STL/CMakeLists.txt b/bench/btl/libs/STL/CMakeLists.txt new file mode 100644 index 000000000..4cfc2dcf2 --- /dev/null +++ b/bench/btl/libs/STL/CMakeLists.txt @@ -0,0 +1,2 @@ + +btl_add_bench(btl_STL main.cpp OFF) diff --git a/bench/btl/libs/STL/STL_interface.hh b/bench/btl/libs/STL/STL_interface.hh new file mode 100644 index 000000000..93e76bd55 --- /dev/null +++ b/bench/btl/libs/STL/STL_interface.hh @@ -0,0 +1,244 @@ +//===================================================== +// File : STL_interface.hh +// Author : L. Plagne <laurent.plagne@edf.fr)> +// Copyright (C) EDF R&D, lun sep 30 14:23:24 CEST 2002 +//===================================================== +// +// This program is free software; you can redistribute it and/or +// modify it under the terms of the GNU General Public License +// as published by the Free Software Foundation; either version 2 +// of the License, or (at your option) any later version. +// +// This program is distributed in the hope that it will be useful, +// but WITHOUT ANY WARRANTY; without even the implied warranty of +// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +// GNU General Public License for more details. +// You should have received a copy of the GNU General Public License +// along with this program; if not, write to the Free Software +// Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. +// +#ifndef STL_INTERFACE_HH +#define STL_INTERFACE_HH +#include <string> +#include <vector> +#include "utilities.h" + +using namespace std; + +template<class real> +class STL_interface{ + +public : + + typedef real real_type ; + + typedef std::vector<real> stl_vector; + typedef std::vector<stl_vector > stl_matrix; + + typedef stl_matrix gene_matrix; + + typedef stl_vector gene_vector; + + static inline std::string name( void ) + { + return "STL"; + } + + static void free_matrix(gene_matrix & A, int N){} + + static void free_vector(gene_vector & B){} + + static inline void matrix_from_stl(gene_matrix & A, stl_matrix & A_stl){ + A = A_stl; + } + + static inline void vector_from_stl(gene_vector & B, stl_vector & B_stl){ + B = B_stl; + } + + static inline void vector_to_stl(gene_vector & B, stl_vector & B_stl){ + B_stl = B ; + } + + + static inline void matrix_to_stl(gene_matrix & A, stl_matrix & A_stl){ + A_stl = A ; + } + + static inline void copy_vector(const gene_vector & source, gene_vector & cible, int N){ + for (int i=0;i<N;i++){ + cible[i]=source[i]; + } + } + + + static inline void copy_matrix(const gene_matrix & source, gene_matrix & cible, int N){ + for (int i=0;i<N;i++) + for (int j=0;j<N;j++) + cible[i][j]=source[i][j]; + } + +// static inline void ata_product(const gene_matrix & A, gene_matrix & X, int N) +// { +// real somme; +// for (int j=0;j<N;j++){ +// for (int i=0;i<N;i++){ +// somme=0.0; +// for (int k=0;k<N;k++) +// somme += A[i][k]*A[j][k]; +// X[j][i]=somme; +// } +// } +// } + + static inline void aat_product(const gene_matrix & A, gene_matrix & X, int N) + { + real somme; + for (int j=0;j<N;j++){ + for (int i=0;i<N;i++){ + somme=0.0; + if(i>=j) + { + for (int k=0;k<N;k++){ + somme+=A[k][i]*A[k][j]; + } + X[j][i]=somme; + } + } + } + } + + + static inline void matrix_matrix_product(const gene_matrix & A, const gene_matrix & B, gene_matrix & X, int N) + { + real somme; + for (int j=0;j<N;j++){ + for (int i=0;i<N;i++){ + somme=0.0; + for (int k=0;k<N;k++) + somme+=A[k][i]*B[j][k]; + X[j][i]=somme; + } + } + } + + static inline void matrix_vector_product(gene_matrix & A, gene_vector & B, gene_vector & X, int N) + { + real somme; + for (int i=0;i<N;i++){ + somme=0.0; + for (int j=0;j<N;j++) + somme+=A[j][i]*B[j]; + X[i]=somme; + } + } + + static inline void symv(gene_matrix & A, gene_vector & B, gene_vector & X, int N) + { + for (int j=0; j<N; ++j) + X[j] = 0; + for (int j=0; j<N; ++j) + { + real t1 = B[j]; + real t2 = 0; + X[j] += t1 * A[j][j]; + for (int i=j+1; i<N; ++i) { + X[i] += t1 * A[j][i]; + t2 += A[j][i] * B[i]; + } + X[j] += t2; + } + } + + static inline void syr2(gene_matrix & A, gene_vector & B, gene_vector & X, int N) + { + for (int j=0; j<N; ++j) + { + for (int i=j; i<N; ++i) + A[j][i] += B[i]*X[j] + B[j]*X[i]; + } + } + + static inline void ger(gene_matrix & A, gene_vector & X, gene_vector & Y, int N) + { + for (int j=0; j<N; ++j) + { + for (int i=j; i<N; ++i) + A[j][i] += X[i]*Y[j]; + } + } + + static inline void atv_product(gene_matrix & A, gene_vector & B, gene_vector & X, int N) + { + real somme; + for (int i=0;i<N;i++){ + somme = 0.0; + for (int j=0;j<N;j++) + somme += A[i][j]*B[j]; + X[i] = somme; + } + } + + static inline void axpy(real coef, const gene_vector & X, gene_vector & Y, int N){ + for (int i=0;i<N;i++) + Y[i]+=coef*X[i]; + } + + static inline void axpby(real a, const gene_vector & X, real b, gene_vector & Y, int N){ + for (int i=0;i<N;i++) + Y[i] = a*X[i] + b*Y[i]; + } + + static inline void trisolve_lower(const gene_matrix & L, const gene_vector & B, gene_vector & X, int N){ + copy_vector(B,X,N); + for(int i=0; i<N; ++i) + { + X[i] /= L[i][i]; + real tmp = X[i]; + for (int j=i+1; j<N; ++j) + X[j] -= tmp * L[i][j]; + } + } + + static inline real norm_diff(const stl_vector & A, const stl_vector & B) + { + int N=A.size(); + real somme=0.0; + real somme2=0.0; + + for (int i=0;i<N;i++){ + real diff=A[i]-B[i]; + somme+=diff*diff; + somme2+=A[i]*A[i]; + } + return somme/somme2; + } + + static inline real norm_diff(const stl_matrix & A, const stl_matrix & B) + { + int N=A[0].size(); + real somme=0.0; + real somme2=0.0; + + for (int i=0;i<N;i++){ + for (int j=0;j<N;j++){ + real diff=A[i][j] - B[i][j]; + somme += diff*diff; + somme2 += A[i][j]*A[i][j]; + } + } + + return somme/somme2; + } + + static inline void display_vector(const stl_vector & A) + { + int N=A.size(); + for (int i=0;i<N;i++){ + INFOS("A["<<i<<"]="<<A[i]<<endl); + } + } + +}; + +#endif diff --git a/bench/btl/libs/STL/main.cpp b/bench/btl/libs/STL/main.cpp new file mode 100644 index 000000000..4e73328ef --- /dev/null +++ b/bench/btl/libs/STL/main.cpp @@ -0,0 +1,42 @@ +//===================================================== +// File : main.cpp +// Author : L. Plagne <laurent.plagne@edf.fr)> +// Copyright (C) EDF R&D, lun sep 30 14:23:23 CEST 2002 +//===================================================== +// +// This program is free software; you can redistribute it and/or +// modify it under the terms of the GNU General Public License +// as published by the Free Software Foundation; either version 2 +// of the License, or (at your option) any later version. +// +// This program is distributed in the hope that it will be useful, +// but WITHOUT ANY WARRANTY; without even the implied warranty of +// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +// GNU General Public License for more details. +// You should have received a copy of the GNU General Public License +// along with this program; if not, write to the Free Software +// Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. +// +#include "utilities.h" +#include "STL_interface.hh" +#include "bench.hh" +#include "basic_actions.hh" + +BTL_MAIN; + +int main() +{ + bench<Action_axpy<STL_interface<REAL_TYPE> > >(MIN_AXPY,MAX_AXPY,NB_POINT); + bench<Action_axpby<STL_interface<REAL_TYPE> > >(MIN_AXPY,MAX_AXPY,NB_POINT); + bench<Action_matrix_vector_product<STL_interface<REAL_TYPE> > >(MIN_MV,MAX_MV,NB_POINT); + bench<Action_atv_product<STL_interface<REAL_TYPE> > >(MIN_MV,MAX_MV,NB_POINT); + bench<Action_symv<STL_interface<REAL_TYPE> > >(MIN_MV,MAX_MV,NB_POINT); + bench<Action_syr2<STL_interface<REAL_TYPE> > >(MIN_MV,MAX_MV,NB_POINT); + bench<Action_matrix_matrix_product<STL_interface<REAL_TYPE> > >(MIN_MM,MAX_MM,NB_POINT); + bench<Action_ata_product<STL_interface<REAL_TYPE> > >(MIN_MM,MAX_MM,NB_POINT); + bench<Action_aat_product<STL_interface<REAL_TYPE> > >(MIN_MM,MAX_MM,NB_POINT); + + return 0; +} + + diff --git a/bench/btl/libs/blitz/CMakeLists.txt b/bench/btl/libs/blitz/CMakeLists.txt new file mode 100644 index 000000000..880ab7338 --- /dev/null +++ b/bench/btl/libs/blitz/CMakeLists.txt @@ -0,0 +1,17 @@ + +find_package(Blitz) + +if (BLITZ_FOUND) + include_directories(${BLITZ_INCLUDES}) + + btl_add_bench(btl_blitz btl_blitz.cpp) + if (BUILD_btl_blitz) + target_link_libraries(btl_blitz ${BLITZ_LIBRARIES}) + endif (BUILD_btl_blitz) + + btl_add_bench(btl_tiny_blitz btl_tiny_blitz.cpp OFF) + if (BUILD_btl_tiny_blitz) + target_link_libraries(btl_tiny_blitz ${BLITZ_LIBRARIES}) + endif (BUILD_btl_tiny_blitz) + +endif (BLITZ_FOUND) diff --git a/bench/btl/libs/blitz/blitz_LU_solve_interface.hh b/bench/btl/libs/blitz/blitz_LU_solve_interface.hh new file mode 100644 index 000000000..dcb9f567f --- /dev/null +++ b/bench/btl/libs/blitz/blitz_LU_solve_interface.hh @@ -0,0 +1,192 @@ +//===================================================== +// File : blitz_LU_solve_interface.hh +// Author : L. Plagne <laurent.plagne@edf.fr)> +// Copyright (C) EDF R&D, lun sep 30 14:23:31 CEST 2002 +//===================================================== +// +// This program is free software; you can redistribute it and/or +// modify it under the terms of the GNU General Public License +// as published by the Free Software Foundation; either version 2 +// of the License, or (at your option) any later version. +// +// This program is distributed in the hope that it will be useful, +// but WITHOUT ANY WARRANTY; without even the implied warranty of +// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +// GNU General Public License for more details. +// You should have received a copy of the GNU General Public License +// along with this program; if not, write to the Free Software +// Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. +// +#ifndef BLITZ_LU_SOLVE_INTERFACE_HH +#define BLITZ_LU_SOLVE_INTERFACE_HH + +#include "blitz/array.h" +#include <vector> + +BZ_USING_NAMESPACE(blitz) + +template<class real> +class blitz_LU_solve_interface : public blitz_interface<real> +{ + +public : + + typedef typename blitz_interface<real>::gene_matrix gene_matrix; + typedef typename blitz_interface<real>::gene_vector gene_vector; + + typedef blitz::Array<int,1> Pivot_Vector; + + inline static void new_Pivot_Vector(Pivot_Vector & pivot,int N) + { + + pivot.resize(N); + + } + + inline static void free_Pivot_Vector(Pivot_Vector & pivot) + { + + return; + + } + + + static inline real matrix_vector_product_sliced(const gene_matrix & A, gene_vector B, int row, int col_start, int col_end) + { + + real somme=0.; + + for (int j=col_start ; j<col_end+1 ; j++){ + + somme+=A(row,j)*B(j); + + } + + return somme; + + } + + + + + static inline real matrix_matrix_product_sliced(gene_matrix & A, int row, int col_start, int col_end, gene_matrix & B, int row_shift, int col ) + { + + real somme=0.; + + for (int j=col_start ; j<col_end+1 ; j++){ + + somme+=A(row,j)*B(j+row_shift,col); + + } + + return somme; + + } + + inline static void LU_factor(gene_matrix & LU, Pivot_Vector & pivot, int N) + { + + ASSERT( LU.rows()==LU.cols() ) ; + int index_max = 0 ; + real big = 0. ; + real theSum = 0. ; + real dum = 0. ; + // Get the implicit scaling information : + gene_vector ImplicitScaling( N ) ; + for( int i=0; i<N; i++ ) { + big = 0. ; + for( int j=0; j<N; j++ ) { + if( abs( LU( i, j ) )>=big ) big = abs( LU( i, j ) ) ; + } + if( big==0. ) { + INFOS( "blitz_LU_factor::Singular matrix" ) ; + exit( 0 ) ; + } + ImplicitScaling( i ) = 1./big ; + } + // Loop over columns of Crout's method : + for( int j=0; j<N; j++ ) { + for( int i=0; i<j; i++ ) { + theSum = LU( i, j ) ; + theSum -= matrix_matrix_product_sliced(LU, i, 0, i-1, LU, 0, j) ; + // theSum -= sum( LU( i, Range( fromStart, i-1 ) )*LU( Range( fromStart, i-1 ), j ) ) ; + LU( i, j ) = theSum ; + } + + // Search for the largest pivot element : + big = 0. ; + for( int i=j; i<N; i++ ) { + theSum = LU( i, j ) ; + theSum -= matrix_matrix_product_sliced(LU, i, 0, j-1, LU, 0, j) ; + // theSum -= sum( LU( i, Range( fromStart, j-1 ) )*LU( Range( fromStart, j-1 ), j ) ) ; + LU( i, j ) = theSum ; + if( (ImplicitScaling( i )*abs( theSum ))>=big ) { + dum = ImplicitScaling( i )*abs( theSum ) ; + big = dum ; + index_max = i ; + } + } + // Interchanging rows and the scale factor : + if( j!=index_max ) { + for( int k=0; k<N; k++ ) { + dum = LU( index_max, k ) ; + LU( index_max, k ) = LU( j, k ) ; + LU( j, k ) = dum ; + } + ImplicitScaling( index_max ) = ImplicitScaling( j ) ; + } + pivot( j ) = index_max ; + if ( LU( j, j )==0. ) LU( j, j ) = 1.e-20 ; + // Divide by the pivot element : + if( j<N ) { + dum = 1./LU( j, j ) ; + for( int i=j+1; i<N; i++ ) LU( i, j ) *= dum ; + } + } + + } + + inline static void LU_solve(const gene_matrix & LU, const Pivot_Vector pivot, gene_vector &B, gene_vector X, int N) + { + + // Pour conserver le meme header, on travaille sur X, copie du second-membre B + X = B.copy() ; + ASSERT( LU.rows()==LU.cols() ) ; + firstIndex indI ; + // Forward substitution : + int ii = 0 ; + real theSum = 0. ; + for( int i=0; i<N; i++ ) { + int ip = pivot( i ) ; + theSum = X( ip ) ; + // theSum = B( ip ) ; + X( ip ) = X( i ) ; + // B( ip ) = B( i ) ; + if( ii ) { + theSum -= matrix_vector_product_sliced(LU, X, i, ii-1, i-1) ; + // theSum -= sum( LU( i, Range( ii-1, i-1 ) )*X( Range( ii-1, i-1 ) ) ) ; + // theSum -= sum( LU( i, Range( ii-1, i-1 ) )*B( Range( ii-1, i-1 ) ) ) ; + } else if( theSum ) { + ii = i+1 ; + } + X( i ) = theSum ; + // B( i ) = theSum ; + } + // Backsubstitution : + for( int i=N-1; i>=0; i-- ) { + theSum = X( i ) ; + // theSum = B( i ) ; + theSum -= matrix_vector_product_sliced(LU, X, i, i+1, N) ; + // theSum -= sum( LU( i, Range( i+1, toEnd ) )*X( Range( i+1, toEnd ) ) ) ; + // theSum -= sum( LU( i, Range( i+1, toEnd ) )*B( Range( i+1, toEnd ) ) ) ; + // Store a component of the solution vector : + X( i ) = theSum/LU( i, i ) ; + // B( i ) = theSum/LU( i, i ) ; + } + + } + +}; + +#endif diff --git a/bench/btl/libs/blitz/blitz_interface.hh b/bench/btl/libs/blitz/blitz_interface.hh new file mode 100644 index 000000000..a67c47c75 --- /dev/null +++ b/bench/btl/libs/blitz/blitz_interface.hh @@ -0,0 +1,147 @@ +//===================================================== +// File : blitz_interface.hh +// Author : L. Plagne <laurent.plagne@edf.fr)> +// Copyright (C) EDF R&D, lun sep 30 14:23:30 CEST 2002 +// Copyright (C) 2008 Gael Guennebaud <gael.guennebaud@inria.fr> +//===================================================== +// +// This program is free software; you can redistribute it and/or +// modify it under the terms of the GNU General Public License +// as published by the Free Software Foundation; either version 2 +// of the License, or (at your option) any later version. +// +// This program is distributed in the hope that it will be useful, +// but WITHOUT ANY WARRANTY; without even the implied warranty of +// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +// GNU General Public License for more details. +// You should have received a copy of the GNU General Public License +// along with this program; if not, write to the Free Software +// Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. +// +#ifndef BLITZ_INTERFACE_HH +#define BLITZ_INTERFACE_HH + +#include <blitz/blitz.h> +#include <blitz/array.h> +#include <blitz/vector-et.h> +#include <blitz/vecwhere.h> +#include <blitz/matrix.h> +#include <vector> + +BZ_USING_NAMESPACE(blitz) + +template<class real> +class blitz_interface{ + +public : + + typedef real real_type ; + + typedef std::vector<real> stl_vector; + typedef std::vector<stl_vector > stl_matrix; + + typedef blitz::Array<real, 2> gene_matrix; + typedef blitz::Array<real, 1> gene_vector; +// typedef blitz::Matrix<real, blitz::ColumnMajor> gene_matrix; +// typedef blitz::Vector<real> gene_vector; + + static inline std::string name() { return "blitz"; } + + static void free_matrix(gene_matrix & A, int N){} + + static void free_vector(gene_vector & B){} + + static inline void matrix_from_stl(gene_matrix & A, stl_matrix & A_stl){ + A.resize(A_stl[0].size(),A_stl.size()); + for (int j=0; j<A_stl.size() ; j++){ + for (int i=0; i<A_stl[j].size() ; i++){ + A(i,j)=A_stl[j][i]; + } + } + } + + static inline void vector_from_stl(gene_vector & B, stl_vector & B_stl){ + B.resize(B_stl.size()); + for (int i=0; i<B_stl.size() ; i++){ + B(i)=B_stl[i]; + } + } + + static inline void vector_to_stl(gene_vector & B, stl_vector & B_stl){ + for (int i=0; i<B_stl.size() ; i++){ + B_stl[i]=B(i); + } + } + + static inline void matrix_to_stl(gene_matrix & A, stl_matrix & A_stl){ + int N=A_stl.size(); + for (int j=0;j<N;j++){ + A_stl[j].resize(N); + for (int i=0;i<N;i++) + A_stl[j][i] = A(i,j); + } + } + + static inline void matrix_matrix_product(const gene_matrix & A, const gene_matrix & B, gene_matrix & X, int N) + { + firstIndex i; + secondIndex j; + thirdIndex k; + X = sum(A(i,k) * B(k,j), k); + } + + static inline void ata_product(const gene_matrix & A, gene_matrix & X, int N) + { + firstIndex i; + secondIndex j; + thirdIndex k; + X = sum(A(k,i) * A(k,j), k); + } + + static inline void aat_product(const gene_matrix & A, gene_matrix & X, int N) + { + firstIndex i; + secondIndex j; + thirdIndex k; + X = sum(A(i,k) * A(j,k), k); + } + + static inline void matrix_vector_product(gene_matrix & A, gene_vector & B, gene_vector & X, int N) + { + firstIndex i; + secondIndex j; + X = sum(A(i,j)*B(j),j); + } + + static inline void atv_product(gene_matrix & A, gene_vector & B, gene_vector & X, int N) + { + firstIndex i; + secondIndex j; + X = sum(A(j,i) * B(j),j); + } + + static inline void axpy(const real coef, const gene_vector & X, gene_vector & Y, int N) + { + firstIndex i; + Y = Y(i) + coef * X(i); + //Y += coef * X; + } + + static inline void copy_matrix(const gene_matrix & source, gene_matrix & cible, int N){ + cible = source; + //cible.template operator=<gene_matrix>(source); +// for (int i=0;i<N;i++){ +// for (int j=0;j<N;j++){ +// cible(i,j)=source(i,j); +// } +// } + } + + static inline void copy_vector(const gene_vector & source, gene_vector & cible, int N){ + //cible.template operator=<gene_vector>(source); + cible = source; + } + +}; + +#endif diff --git a/bench/btl/libs/blitz/btl_blitz.cpp b/bench/btl/libs/blitz/btl_blitz.cpp new file mode 100644 index 000000000..16d2b5951 --- /dev/null +++ b/bench/btl/libs/blitz/btl_blitz.cpp @@ -0,0 +1,51 @@ +//===================================================== +// File : main.cpp +// Author : L. Plagne <laurent.plagne@edf.fr)> +// Copyright (C) EDF R&D, lun sep 30 14:23:30 CEST 2002 +//===================================================== +// +// This program is free software; you can redistribute it and/or +// modify it under the terms of the GNU General Public License +// as published by the Free Software Foundation; either version 2 +// of the License, or (at your option) any later version. +// +// This program is distributed in the hope that it will be useful, +// but WITHOUT ANY WARRANTY; without even the implied warranty of +// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +// GNU General Public License for more details. +// You should have received a copy of the GNU General Public License +// along with this program; if not, write to the Free Software +// Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. +// +#include "utilities.h" +#include "blitz_interface.hh" +#include "blitz_LU_solve_interface.hh" +#include "bench.hh" +#include "action_matrix_vector_product.hh" +#include "action_matrix_matrix_product.hh" +#include "action_axpy.hh" +#include "action_lu_solve.hh" +#include "action_ata_product.hh" +#include "action_aat_product.hh" +#include "action_atv_product.hh" + +BTL_MAIN; + +int main() +{ + + bench<Action_matrix_vector_product<blitz_interface<REAL_TYPE> > >(MIN_MV,MAX_MV,NB_POINT); + bench<Action_atv_product<blitz_interface<REAL_TYPE> > >(MIN_MV,MAX_MV,NB_POINT); + + bench<Action_matrix_matrix_product<blitz_interface<REAL_TYPE> > >(MIN_MM,MAX_MM,NB_POINT); + bench<Action_ata_product<blitz_interface<REAL_TYPE> > >(MIN_MM,MAX_MM,NB_POINT); + bench<Action_aat_product<blitz_interface<REAL_TYPE> > >(MIN_MM,MAX_MM,NB_POINT); + + bench<Action_axpy<blitz_interface<REAL_TYPE> > >(MIN_AXPY,MAX_AXPY,NB_POINT); + + //bench<Action_lu_solve<blitz_LU_solve_interface<REAL_TYPE> > >(MIN_LU,MAX_LU,NB_POINT); + + return 0; +} + + diff --git a/bench/btl/libs/blitz/btl_tiny_blitz.cpp b/bench/btl/libs/blitz/btl_tiny_blitz.cpp new file mode 100644 index 000000000..9fddde752 --- /dev/null +++ b/bench/btl/libs/blitz/btl_tiny_blitz.cpp @@ -0,0 +1,38 @@ +//===================================================== +// File : main.cpp +// Author : L. Plagne <laurent.plagne@edf.fr)> +// Copyright (C) EDF R&D, lun sep 30 14:23:30 CEST 2002 +//===================================================== +// +// This program is free software; you can redistribute it and/or +// modify it under the terms of the GNU General Public License +// as published by the Free Software Foundation; either version 2 +// of the License, or (at your option) any later version. +// +// This program is distributed in the hope that it will be useful, +// but WITHOUT ANY WARRANTY; without even the implied warranty of +// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +// GNU General Public License for more details. +// You should have received a copy of the GNU General Public License +// along with this program; if not, write to the Free Software +// Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. +// +#include "utilities.h" +#include "tiny_blitz_interface.hh" +#include "static/bench_static.hh" +#include "action_matrix_vector_product.hh" +#include "action_matrix_matrix_product.hh" +#include "action_axpy.hh" + +BTL_MAIN; + +int main() +{ + bench_static<Action_axpy,tiny_blitz_interface>(); + bench_static<Action_matrix_matrix_product,tiny_blitz_interface>(); + bench_static<Action_matrix_vector_product,tiny_blitz_interface>(); + + return 0; +} + + diff --git a/bench/btl/libs/blitz/tiny_blitz_interface.hh b/bench/btl/libs/blitz/tiny_blitz_interface.hh new file mode 100644 index 000000000..6b26db72d --- /dev/null +++ b/bench/btl/libs/blitz/tiny_blitz_interface.hh @@ -0,0 +1,106 @@ +//===================================================== +// File : tiny_blitz_interface.hh +// Author : L. Plagne <laurent.plagne@edf.fr)> +// Copyright (C) EDF R&D, lun sep 30 14:23:30 CEST 2002 +//===================================================== +// +// This program is free software; you can redistribute it and/or +// modify it under the terms of the GNU General Public License +// as published by the Free Software Foundation; either version 2 +// of the License, or (at your option) any later version. +// +// This program is distributed in the hope that it will be useful, +// but WITHOUT ANY WARRANTY; without even the implied warranty of +// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +// GNU General Public License for more details. +// You should have received a copy of the GNU General Public License +// along with this program; if not, write to the Free Software +// Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. +// +#ifndef TINY_BLITZ_INTERFACE_HH +#define TINY_BLITZ_INTERFACE_HH + +#include "blitz/array.h" +#include "blitz/tiny.h" +#include "blitz/tinymat.h" +#include "blitz/tinyvec.h" +#include <blitz/tinyvec-et.h> + +#include <vector> + +BZ_USING_NAMESPACE(blitz) + +template<class real, int SIZE> +class tiny_blitz_interface +{ + +public : + + typedef real real_type ; + + typedef std::vector<real> stl_vector; + typedef std::vector<stl_vector > stl_matrix; + + typedef TinyVector<real,SIZE> gene_vector; + typedef TinyMatrix<real,SIZE,SIZE> gene_matrix; + + static inline std::string name() { return "tiny_blitz"; } + + static void free_matrix(gene_matrix & A, int N){} + + static void free_vector(gene_vector & B){} + + static inline void matrix_from_stl(gene_matrix & A, stl_matrix & A_stl){ + for (int j=0; j<A_stl.size() ; j++) + for (int i=0; i<A_stl[j].size() ; i++) + A(i,j)=A_stl[j][i]; + } + + static inline void vector_from_stl(gene_vector & B, stl_vector & B_stl){ + for (int i=0; i<B_stl.size() ; i++) + B(i) = B_stl[i]; + } + + static inline void vector_to_stl(gene_vector & B, stl_vector & B_stl){ + for (int i=0; i<B_stl.size() ; i++) + B_stl[i] = B(i); + } + + static inline void matrix_to_stl(gene_matrix & A, stl_matrix & A_stl){ + int N = A_stl.size(); + for (int j=0;j<N;j++) + { + A_stl[j].resize(N); + for (int i=0;i<N;i++) + A_stl[j][i] = A(i,j); + } + } + + static inline void copy_matrix(const gene_matrix & source, gene_matrix & cible, int N){ + for (int j=0;j<N;j++) + for (int i=0;i<N;i++) + cible(i,j) = source(i,j); + } + + static inline void copy_vector(const gene_vector & source, gene_vector & cible, int N){ + for (int i=0;i<N;i++){ + cible(i) = source(i); + } + } + + static inline void matrix_matrix_product(const gene_matrix & A, const gene_matrix & B, gene_matrix & X, int N){ + X = product(A,B); + } + + static inline void matrix_vector_product(gene_matrix & A, gene_vector & B, gene_vector & X, int N){ + X = product(A,B); + } + + static inline void axpy(const real coef, const gene_vector & X, gene_vector & Y, int N){ + Y += coef * X; + } + +}; + + +#endif diff --git a/bench/btl/libs/eigen2/CMakeLists.txt b/bench/btl/libs/eigen2/CMakeLists.txt new file mode 100644 index 000000000..a2e8fc6da --- /dev/null +++ b/bench/btl/libs/eigen2/CMakeLists.txt @@ -0,0 +1,19 @@ + +find_package(Eigen2) + +if(EIGEN2_FOUND) + + include_directories(BEFORE ${EIGEN2_INCLUDE_DIR}) + btl_add_bench(btl_eigen2_linear main_linear.cpp) + btl_add_bench(btl_eigen2_vecmat main_vecmat.cpp) + btl_add_bench(btl_eigen2_matmat main_matmat.cpp) + btl_add_bench(btl_eigen2_adv main_adv.cpp ) + + btl_add_target_property(btl_eigen2_linear COMPILE_FLAGS "-fno-exceptions -DBTL_PREFIX=eigen2") + btl_add_target_property(btl_eigen2_vecmat COMPILE_FLAGS "-fno-exceptions -DBTL_PREFIX=eigen2") + btl_add_target_property(btl_eigen2_matmat COMPILE_FLAGS "-fno-exceptions -DBTL_PREFIX=eigen2") + btl_add_target_property(btl_eigen2_adv COMPILE_FLAGS "-fno-exceptions -DBTL_PREFIX=eigen2") + + btl_add_bench(btl_tiny_eigen2 btl_tiny_eigen2.cpp OFF) + +endif() # EIGEN2_FOUND diff --git a/bench/btl/libs/eigen2/btl_tiny_eigen2.cpp b/bench/btl/libs/eigen2/btl_tiny_eigen2.cpp new file mode 100644 index 000000000..d1515be84 --- /dev/null +++ b/bench/btl/libs/eigen2/btl_tiny_eigen2.cpp @@ -0,0 +1,46 @@ +//===================================================== +// Copyright (C) 2008 Gael Guennebaud <gael.guennebaud@inria.fr> +//===================================================== +// +// This program is free software; you can redistribute it and/or +// modify it under the terms of the GNU General Public License +// as published by the Free Software Foundation; either version 2 +// of the License, or (at your option) any later version. +// +// This program is distributed in the hope that it will be useful, +// but WITHOUT ANY WARRANTY; without even the implied warranty of +// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +// GNU General Public License for more details. +// You should have received a copy of the GNU General Public License +// along with this program; if not, write to the Free Software +// Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. +// +#include "utilities.h" +#include "eigen3_interface.hh" +#include "static/bench_static.hh" +#include "action_matrix_vector_product.hh" +#include "action_matrix_matrix_product.hh" +#include "action_axpy.hh" +#include "action_lu_solve.hh" +#include "action_ata_product.hh" +#include "action_aat_product.hh" +#include "action_atv_product.hh" +#include "action_cholesky.hh" +#include "action_trisolve.hh" + +BTL_MAIN; + +int main() +{ + + bench_static<Action_axpy,eigen2_interface>(); + bench_static<Action_matrix_matrix_product,eigen2_interface>(); + bench_static<Action_matrix_vector_product,eigen2_interface>(); + bench_static<Action_atv_product,eigen2_interface>(); + bench_static<Action_cholesky,eigen2_interface>(); + bench_static<Action_trisolve,eigen2_interface>(); + + return 0; +} + + diff --git a/bench/btl/libs/eigen2/eigen2_interface.hh b/bench/btl/libs/eigen2/eigen2_interface.hh new file mode 100644 index 000000000..47fe58135 --- /dev/null +++ b/bench/btl/libs/eigen2/eigen2_interface.hh @@ -0,0 +1,168 @@ +//===================================================== +// Copyright (C) 2008 Gael Guennebaud <g.gael@free.fr> +//===================================================== +// +// This program is free software; you can redistribute it and/or +// modify it under the terms of the GNU General Public License +// as published by the Free Software Foundation; either version 2 +// of the License, or (at your option) any later version. +// +// This program is distributed in the hope that it will be useful, +// but WITHOUT ANY WARRANTY; without even the implied warranty of +// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +// GNU General Public License for more details. +// You should have received a copy of the GNU General Public License +// along with this program; if not, write to the Free Software +// Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. +// +#ifndef EIGEN2_INTERFACE_HH +#define EIGEN2_INTERFACE_HH +// #include <cblas.h> +#include <Eigen/Core> +#include <Eigen/Cholesky> +#include <Eigen/LU> +#include <Eigen/QR> +#include <vector> +#include "btl.hh" + +using namespace Eigen; + +template<class real, int SIZE=Dynamic> +class eigen2_interface +{ + +public : + + enum {IsFixedSize = (SIZE!=Dynamic)}; + + typedef real real_type; + + typedef std::vector<real> stl_vector; + typedef std::vector<stl_vector> stl_matrix; + + typedef Eigen::Matrix<real,SIZE,SIZE> gene_matrix; + typedef Eigen::Matrix<real,SIZE,1> gene_vector; + + static inline std::string name( void ) + { + #if defined(EIGEN_VECTORIZE_SSE) + if (SIZE==Dynamic) return "eigen2"; else return "tiny_eigen2"; + #elif defined(EIGEN_VECTORIZE_ALTIVEC) + if (SIZE==Dynamic) return "eigen2"; else return "tiny_eigen2"; + #else + if (SIZE==Dynamic) return "eigen2_novec"; else return "tiny_eigen2_novec"; + #endif + } + + static void free_matrix(gene_matrix & A, int N) {} + + static void free_vector(gene_vector & B) {} + + static BTL_DONT_INLINE void matrix_from_stl(gene_matrix & A, stl_matrix & A_stl){ + A.resize(A_stl[0].size(), A_stl.size()); + + for (int j=0; j<A_stl.size() ; j++){ + for (int i=0; i<A_stl[j].size() ; i++){ + A.coeffRef(i,j) = A_stl[j][i]; + } + } + } + + static BTL_DONT_INLINE void vector_from_stl(gene_vector & B, stl_vector & B_stl){ + B.resize(B_stl.size(),1); + + for (int i=0; i<B_stl.size() ; i++){ + B.coeffRef(i) = B_stl[i]; + } + } + + static BTL_DONT_INLINE void vector_to_stl(gene_vector & B, stl_vector & B_stl){ + for (int i=0; i<B_stl.size() ; i++){ + B_stl[i] = B.coeff(i); + } + } + + static BTL_DONT_INLINE void matrix_to_stl(gene_matrix & A, stl_matrix & A_stl){ + int N=A_stl.size(); + + for (int j=0;j<N;j++){ + A_stl[j].resize(N); + for (int i=0;i<N;i++){ + A_stl[j][i] = A.coeff(i,j); + } + } + } + + static inline void matrix_matrix_product(const gene_matrix & A, const gene_matrix & B, gene_matrix & X, int N){ + X = (A*B).lazy(); + } + + static inline void transposed_matrix_matrix_product(const gene_matrix & A, const gene_matrix & B, gene_matrix & X, int N){ + X = (A.transpose()*B.transpose()).lazy(); + } + + static inline void ata_product(const gene_matrix & A, gene_matrix & X, int N){ + X = (A.transpose()*A).lazy(); + } + + static inline void aat_product(const gene_matrix & A, gene_matrix & X, int N){ + X = (A*A.transpose()).lazy(); + } + + static inline void matrix_vector_product(const gene_matrix & A, const gene_vector & B, gene_vector & X, int N){ + X = (A*B)/*.lazy()*/; + } + + static inline void atv_product(gene_matrix & A, gene_vector & B, gene_vector & X, int N){ + X = (A.transpose()*B)/*.lazy()*/; + } + + static inline void axpy(real coef, const gene_vector & X, gene_vector & Y, int N){ + Y += coef * X; + } + + static inline void axpby(real a, const gene_vector & X, real b, gene_vector & Y, int N){ + Y = a*X + b*Y; + } + + static inline void copy_matrix(const gene_matrix & source, gene_matrix & cible, int N){ + cible = source; + } + + static inline void copy_vector(const gene_vector & source, gene_vector & cible, int N){ + cible = source; + } + + static inline void trisolve_lower(const gene_matrix & L, const gene_vector& B, gene_vector& X, int N){ + X = L.template marked<LowerTriangular>().solveTriangular(B); + } + + static inline void trisolve_lower_matrix(const gene_matrix & L, const gene_matrix& B, gene_matrix& X, int N){ + X = L.template marked<LowerTriangular>().solveTriangular(B); + } + + static inline void cholesky(const gene_matrix & X, gene_matrix & C, int N){ + C = X.llt().matrixL(); +// C = X; +// Cholesky<gene_matrix>::computeInPlace(C); +// Cholesky<gene_matrix>::computeInPlaceBlock(C); + } + + static inline void lu_decomp(const gene_matrix & X, gene_matrix & C, int N){ + C = X.lu().matrixLU(); +// C = X.inverse(); + } + + static inline void tridiagonalization(const gene_matrix & X, gene_matrix & C, int N){ + C = Tridiagonalization<gene_matrix>(X).packedMatrix(); + } + + static inline void hessenberg(const gene_matrix & X, gene_matrix & C, int N){ + C = HessenbergDecomposition<gene_matrix>(X).packedMatrix(); + } + + + +}; + +#endif diff --git a/bench/btl/libs/eigen2/main_adv.cpp b/bench/btl/libs/eigen2/main_adv.cpp new file mode 100644 index 000000000..fe3368925 --- /dev/null +++ b/bench/btl/libs/eigen2/main_adv.cpp @@ -0,0 +1,44 @@ +//===================================================== +// Copyright (C) 2008 Gael Guennebaud <gael.guennebaud@inria.fr> +//===================================================== +// +// This program is free software; you can redistribute it and/or +// modify it under the terms of the GNU General Public License +// as published by the Free Software Foundation; either version 2 +// of the License, or (at your option) any later version. +// +// This program is distributed in the hope that it will be useful, +// but WITHOUT ANY WARRANTY; without even the implied warranty of +// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +// GNU General Public License for more details. +// You should have received a copy of the GNU General Public License +// along with this program; if not, write to the Free Software +// Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. +// +#include "utilities.h" +#include "eigen2_interface.hh" +#include "bench.hh" +#include "action_trisolve.hh" +#include "action_trisolve_matrix.hh" +#include "action_cholesky.hh" +#include "action_hessenberg.hh" +#include "action_lu_decomp.hh" +// #include "action_partial_lu.hh" + +BTL_MAIN; + +int main() +{ + bench<Action_trisolve<eigen2_interface<REAL_TYPE> > >(MIN_MM,MAX_MM,NB_POINT); + bench<Action_trisolve_matrix<eigen2_interface<REAL_TYPE> > >(MIN_MM,MAX_MM,NB_POINT); + bench<Action_cholesky<eigen2_interface<REAL_TYPE> > >(MIN_MM,MAX_MM,NB_POINT); + bench<Action_lu_decomp<eigen2_interface<REAL_TYPE> > >(MIN_MM,MAX_MM,NB_POINT); +// bench<Action_partial_lu<eigen2_interface<REAL_TYPE> > >(MIN_MM,MAX_MM,NB_POINT); + + bench<Action_hessenberg<eigen2_interface<REAL_TYPE> > >(MIN_MM,MAX_MM,NB_POINT); + bench<Action_tridiagonalization<eigen2_interface<REAL_TYPE> > >(MIN_MM,MAX_MM,NB_POINT); + + return 0; +} + + diff --git a/bench/btl/libs/eigen2/main_linear.cpp b/bench/btl/libs/eigen2/main_linear.cpp new file mode 100644 index 000000000..c17d16c08 --- /dev/null +++ b/bench/btl/libs/eigen2/main_linear.cpp @@ -0,0 +1,34 @@ +//===================================================== +// Copyright (C) 2008 Gael Guennebaud <gael.guennebaud@inria.fr> +//===================================================== +// +// This program is free software; you can redistribute it and/or +// modify it under the terms of the GNU General Public License +// as published by the Free Software Foundation; either version 2 +// of the License, or (at your option) any later version. +// +// This program is distributed in the hope that it will be useful, +// but WITHOUT ANY WARRANTY; without even the implied warranty of +// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +// GNU General Public License for more details. +// You should have received a copy of the GNU General Public License +// along with this program; if not, write to the Free Software +// Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. +// +#include "utilities.h" +#include "eigen2_interface.hh" +#include "bench.hh" +#include "basic_actions.hh" + +BTL_MAIN; + +int main() +{ + + bench<Action_axpy<eigen2_interface<REAL_TYPE> > >(MIN_AXPY,MAX_AXPY,NB_POINT); + bench<Action_axpby<eigen2_interface<REAL_TYPE> > >(MIN_AXPY,MAX_AXPY,NB_POINT); + + return 0; +} + + diff --git a/bench/btl/libs/eigen2/main_matmat.cpp b/bench/btl/libs/eigen2/main_matmat.cpp new file mode 100644 index 000000000..cd9dc9cb0 --- /dev/null +++ b/bench/btl/libs/eigen2/main_matmat.cpp @@ -0,0 +1,35 @@ +//===================================================== +// Copyright (C) 2008 Gael Guennebaud <gael.guennebaud@inria.fr> +//===================================================== +// +// This program is free software; you can redistribute it and/or +// modify it under the terms of the GNU General Public License +// as published by the Free Software Foundation; either version 2 +// of the License, or (at your option) any later version. +// +// This program is distributed in the hope that it will be useful, +// but WITHOUT ANY WARRANTY; without even the implied warranty of +// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +// GNU General Public License for more details. +// You should have received a copy of the GNU General Public License +// along with this program; if not, write to the Free Software +// Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. +// +#include "utilities.h" +#include "eigen2_interface.hh" +#include "bench.hh" +#include "basic_actions.hh" + +BTL_MAIN; + +int main() +{ + bench<Action_matrix_matrix_product<eigen2_interface<REAL_TYPE> > >(MIN_MM,MAX_MM,NB_POINT); +// bench<Action_ata_product<eigen2_interface<REAL_TYPE> > >(MIN_MM,MAX_MM,NB_POINT); + bench<Action_aat_product<eigen2_interface<REAL_TYPE> > >(MIN_MM,MAX_MM,NB_POINT); +// bench<Action_trmm<eigen2_interface<REAL_TYPE> > >(MIN_MM,MAX_MM,NB_POINT); + + return 0; +} + + diff --git a/bench/btl/libs/eigen2/main_vecmat.cpp b/bench/btl/libs/eigen2/main_vecmat.cpp new file mode 100644 index 000000000..8b66cd2d9 --- /dev/null +++ b/bench/btl/libs/eigen2/main_vecmat.cpp @@ -0,0 +1,36 @@ +//===================================================== +// Copyright (C) 2008 Gael Guennebaud <gael.guennebaud@inria.fr> +//===================================================== +// +// This program is free software; you can redistribute it and/or +// modify it under the terms of the GNU General Public License +// as published by the Free Software Foundation; either version 2 +// of the License, or (at your option) any later version. +// +// This program is distributed in the hope that it will be useful, +// but WITHOUT ANY WARRANTY; without even the implied warranty of +// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +// GNU General Public License for more details. +// You should have received a copy of the GNU General Public License +// along with this program; if not, write to the Free Software +// Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. +// +#include "utilities.h" +#include "eigen2_interface.hh" +#include "bench.hh" +#include "basic_actions.hh" + +BTL_MAIN; + +int main() +{ + bench<Action_matrix_vector_product<eigen2_interface<REAL_TYPE> > >(MIN_MV,MAX_MV,NB_POINT); + bench<Action_atv_product<eigen2_interface<REAL_TYPE> > >(MIN_MV,MAX_MV,NB_POINT); +// bench<Action_symv<eigen2_interface<REAL_TYPE> > >(MIN_MV,MAX_MV,NB_POINT); +// bench<Action_syr2<eigen2_interface<REAL_TYPE> > >(MIN_MV,MAX_MV,NB_POINT); +// bench<Action_ger<eigen2_interface<REAL_TYPE> > >(MIN_MV,MAX_MV,NB_POINT); + + return 0; +} + + diff --git a/bench/btl/libs/eigen3/CMakeLists.txt b/bench/btl/libs/eigen3/CMakeLists.txt new file mode 100644 index 000000000..00cae23d3 --- /dev/null +++ b/bench/btl/libs/eigen3/CMakeLists.txt @@ -0,0 +1,65 @@ + + +if((NOT EIGEN3_INCLUDE_DIR) AND Eigen_SOURCE_DIR) + # unless EIGEN3_INCLUDE_DIR is defined, let's use current Eigen version + set(EIGEN3_INCLUDE_DIR ${Eigen_SOURCE_DIR}) + set(EIGEN3_FOUND TRUE) +else() + find_package(Eigen3) +endif() + +if (EIGEN3_FOUND) + + include_directories(${EIGEN3_INCLUDE_DIR}) + btl_add_bench(btl_eigen3_linear main_linear.cpp) + btl_add_bench(btl_eigen3_vecmat main_vecmat.cpp) + btl_add_bench(btl_eigen3_matmat main_matmat.cpp) + btl_add_bench(btl_eigen3_adv main_adv.cpp ) + + btl_add_target_property(btl_eigen3_linear COMPILE_FLAGS "-fno-exceptions -DBTL_PREFIX=eigen3") + btl_add_target_property(btl_eigen3_vecmat COMPILE_FLAGS "-fno-exceptions -DBTL_PREFIX=eigen3") + btl_add_target_property(btl_eigen3_matmat COMPILE_FLAGS "-fno-exceptions -DBTL_PREFIX=eigen3") + btl_add_target_property(btl_eigen3_adv COMPILE_FLAGS "-fno-exceptions -DBTL_PREFIX=eigen3") + + option(BTL_BENCH_NOGCCVEC "also bench Eigen explicit vec without GCC's auto vec" OFF) + if(CMAKE_COMPILER_IS_GNUCXX AND BTL_BENCH_NOGCCVEC) + btl_add_bench(btl_eigen3_nogccvec_linear main_linear.cpp) + btl_add_bench(btl_eigen3_nogccvec_vecmat main_vecmat.cpp) + btl_add_bench(btl_eigen3_nogccvec_matmat main_matmat.cpp) + btl_add_bench(btl_eigen3_nogccvec_adv main_adv.cpp ) + + btl_add_target_property(btl_eigen3_nogccvec_linear COMPILE_FLAGS "-fno-exceptions -fno-tree-vectorize -DBTL_PREFIX=eigen3_nogccvec") + btl_add_target_property(btl_eigen3_nogccvec_vecmat COMPILE_FLAGS "-fno-exceptions -fno-tree-vectorize -DBTL_PREFIX=eigen3_nogccvec") + btl_add_target_property(btl_eigen3_nogccvec_matmat COMPILE_FLAGS "-fno-exceptions -fno-tree-vectorize -DBTL_PREFIX=eigen3_nogccvec") + btl_add_target_property(btl_eigen3_nogccvec_adv COMPILE_FLAGS "-fno-exceptions -fno-tree-vectorize -DBTL_PREFIX=eigen3_nogccvec") + endif() + + + if(NOT BTL_NOVEC) + btl_add_bench(btl_eigen3_novec_linear main_linear.cpp OFF) + btl_add_bench(btl_eigen3_novec_vecmat main_vecmat.cpp OFF) + btl_add_bench(btl_eigen3_novec_matmat main_matmat.cpp OFF) + btl_add_bench(btl_eigen3_novec_adv main_adv.cpp OFF) + btl_add_target_property(btl_eigen3_novec_linear COMPILE_FLAGS "-fno-exceptions -DEIGEN_DONT_VECTORIZE -DBTL_PREFIX=eigen3_novec") + btl_add_target_property(btl_eigen3_novec_vecmat COMPILE_FLAGS "-fno-exceptions -DEIGEN_DONT_VECTORIZE -DBTL_PREFIX=eigen3_novec") + btl_add_target_property(btl_eigen3_novec_matmat COMPILE_FLAGS "-fno-exceptions -DEIGEN_DONT_VECTORIZE -DBTL_PREFIX=eigen3_novec") + btl_add_target_property(btl_eigen3_novec_adv COMPILE_FLAGS "-fno-exceptions -DEIGEN_DONT_VECTORIZE -DBTL_PREFIX=eigen3_novec") + +# if(BUILD_btl_eigen3_adv) +# target_link_libraries(btl_eigen3_adv ${MKL_LIBRARIES}) +# endif(BUILD_btl_eigen3_adv) + + endif(NOT BTL_NOVEC) + + btl_add_bench(btl_tiny_eigen3 btl_tiny_eigen3.cpp OFF) + + if(NOT BTL_NOVEC) + btl_add_bench(btl_tiny_eigen3_novec btl_tiny_eigen3.cpp OFF) + btl_add_target_property(btl_tiny_eigen3_novec COMPILE_FLAGS "-DBTL_PREFIX=eigen3_tiny") + + if(BUILD_btl_tiny_eigen3_novec) + btl_add_target_property(btl_tiny_eigen3_novec COMPILE_FLAGS "-DEIGEN_DONT_VECTORIZE -DBTL_PREFIX=eigen3_tiny_novec") + endif(BUILD_btl_tiny_eigen3_novec) + endif(NOT BTL_NOVEC) + +endif (EIGEN3_FOUND) diff --git a/bench/btl/libs/eigen3/btl_tiny_eigen3.cpp b/bench/btl/libs/eigen3/btl_tiny_eigen3.cpp new file mode 100644 index 000000000..d1515be84 --- /dev/null +++ b/bench/btl/libs/eigen3/btl_tiny_eigen3.cpp @@ -0,0 +1,46 @@ +//===================================================== +// Copyright (C) 2008 Gael Guennebaud <gael.guennebaud@inria.fr> +//===================================================== +// +// This program is free software; you can redistribute it and/or +// modify it under the terms of the GNU General Public License +// as published by the Free Software Foundation; either version 2 +// of the License, or (at your option) any later version. +// +// This program is distributed in the hope that it will be useful, +// but WITHOUT ANY WARRANTY; without even the implied warranty of +// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +// GNU General Public License for more details. +// You should have received a copy of the GNU General Public License +// along with this program; if not, write to the Free Software +// Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. +// +#include "utilities.h" +#include "eigen3_interface.hh" +#include "static/bench_static.hh" +#include "action_matrix_vector_product.hh" +#include "action_matrix_matrix_product.hh" +#include "action_axpy.hh" +#include "action_lu_solve.hh" +#include "action_ata_product.hh" +#include "action_aat_product.hh" +#include "action_atv_product.hh" +#include "action_cholesky.hh" +#include "action_trisolve.hh" + +BTL_MAIN; + +int main() +{ + + bench_static<Action_axpy,eigen2_interface>(); + bench_static<Action_matrix_matrix_product,eigen2_interface>(); + bench_static<Action_matrix_vector_product,eigen2_interface>(); + bench_static<Action_atv_product,eigen2_interface>(); + bench_static<Action_cholesky,eigen2_interface>(); + bench_static<Action_trisolve,eigen2_interface>(); + + return 0; +} + + diff --git a/bench/btl/libs/eigen3/eigen3_interface.hh b/bench/btl/libs/eigen3/eigen3_interface.hh new file mode 100644 index 000000000..31bcc1f93 --- /dev/null +++ b/bench/btl/libs/eigen3/eigen3_interface.hh @@ -0,0 +1,240 @@ +//===================================================== +// Copyright (C) 2008 Gael Guennebaud <gael.guennebaud@inria.fr> +//===================================================== +// +// This program is free software; you can redistribute it and/or +// modify it under the terms of the GNU General Public License +// as published by the Free Software Foundation; either version 2 +// of the License, or (at your option) any later version. +// +// This program is distributed in the hope that it will be useful, +// but WITHOUT ANY WARRANTY; without even the implied warranty of +// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +// GNU General Public License for more details. +// You should have received a copy of the GNU General Public License +// along with this program; if not, write to the Free Software +// Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. +// +#ifndef EIGEN3_INTERFACE_HH +#define EIGEN3_INTERFACE_HH + +#include <Eigen/Eigen> +#include <vector> +#include "btl.hh" + +using namespace Eigen; + +template<class real, int SIZE=Dynamic> +class eigen3_interface +{ + +public : + + enum {IsFixedSize = (SIZE!=Dynamic)}; + + typedef real real_type; + + typedef std::vector<real> stl_vector; + typedef std::vector<stl_vector> stl_matrix; + + typedef Eigen::Matrix<real,SIZE,SIZE> gene_matrix; + typedef Eigen::Matrix<real,SIZE,1> gene_vector; + + static inline std::string name( void ) + { + return EIGEN_MAKESTRING(BTL_PREFIX); + } + + static void free_matrix(gene_matrix & A, int N) {} + + static void free_vector(gene_vector & B) {} + + static BTL_DONT_INLINE void matrix_from_stl(gene_matrix & A, stl_matrix & A_stl){ + A.resize(A_stl[0].size(), A_stl.size()); + + for (int j=0; j<A_stl.size() ; j++){ + for (int i=0; i<A_stl[j].size() ; i++){ + A.coeffRef(i,j) = A_stl[j][i]; + } + } + } + + static BTL_DONT_INLINE void vector_from_stl(gene_vector & B, stl_vector & B_stl){ + B.resize(B_stl.size(),1); + + for (int i=0; i<B_stl.size() ; i++){ + B.coeffRef(i) = B_stl[i]; + } + } + + static BTL_DONT_INLINE void vector_to_stl(gene_vector & B, stl_vector & B_stl){ + for (int i=0; i<B_stl.size() ; i++){ + B_stl[i] = B.coeff(i); + } + } + + static BTL_DONT_INLINE void matrix_to_stl(gene_matrix & A, stl_matrix & A_stl){ + int N=A_stl.size(); + + for (int j=0;j<N;j++){ + A_stl[j].resize(N); + for (int i=0;i<N;i++){ + A_stl[j][i] = A.coeff(i,j); + } + } + } + + static inline void matrix_matrix_product(const gene_matrix & A, const gene_matrix & B, gene_matrix & X, int N){ + X.noalias() = A*B; + } + + static inline void transposed_matrix_matrix_product(const gene_matrix & A, const gene_matrix & B, gene_matrix & X, int N){ + X.noalias() = A.transpose()*B.transpose(); + } + +// static inline void ata_product(const gene_matrix & A, gene_matrix & X, int N){ +// X.noalias() = A.transpose()*A; +// } + + static inline void aat_product(const gene_matrix & A, gene_matrix & X, int N){ + X.template triangularView<Lower>().setZero(); + X.template selfadjointView<Lower>().rankUpdate(A); + } + + static inline void matrix_vector_product(const gene_matrix & A, const gene_vector & B, gene_vector & X, int N){ + X.noalias() = A*B; + } + + static inline void symv(const gene_matrix & A, const gene_vector & B, gene_vector & X, int N){ + X.noalias() = (A.template selfadjointView<Lower>() * B); +// internal::product_selfadjoint_vector<real,0,LowerTriangularBit,false,false>(N,A.data(),N, B.data(), 1, X.data(), 1); + } + + template<typename Dest, typename Src> static void triassign(Dest& dst, const Src& src) + { + typedef typename Dest::Scalar Scalar; + typedef typename internal::packet_traits<Scalar>::type Packet; + const int PacketSize = sizeof(Packet)/sizeof(Scalar); + int size = dst.cols(); + for(int j=0; j<size; j+=1) + { +// const int alignedEnd = alignedStart + ((innerSize-alignedStart) & ~packetAlignedMask); + Scalar* A0 = dst.data() + j*dst.stride(); + int starti = j; + int alignedEnd = starti; + int alignedStart = (starti) + internal::first_aligned(&A0[starti], size-starti); + alignedEnd = alignedStart + ((size-alignedStart)/(2*PacketSize))*(PacketSize*2); + + // do the non-vectorizable part of the assignment + for (int index = starti; index<alignedStart ; ++index) + { + if(Dest::Flags&RowMajorBit) + dst.copyCoeff(j, index, src); + else + dst.copyCoeff(index, j, src); + } + + // do the vectorizable part of the assignment + for (int index = alignedStart; index<alignedEnd; index+=PacketSize) + { + if(Dest::Flags&RowMajorBit) + dst.template copyPacket<Src, Aligned, Unaligned>(j, index, src); + else + dst.template copyPacket<Src, Aligned, Unaligned>(index, j, src); + } + + // do the non-vectorizable part of the assignment + for (int index = alignedEnd; index<size; ++index) + { + if(Dest::Flags&RowMajorBit) + dst.copyCoeff(j, index, src); + else + dst.copyCoeff(index, j, src); + } + //dst.col(j).tail(N-j) = src.col(j).tail(N-j); + } + } + + static EIGEN_DONT_INLINE void syr2(gene_matrix & A, gene_vector & X, gene_vector & Y, int N){ + // internal::product_selfadjoint_rank2_update<real,0,LowerTriangularBit>(N,A.data(),N, X.data(), 1, Y.data(), 1, -1); + for(int j=0; j<N; ++j) + A.col(j).tail(N-j) += X[j] * Y.tail(N-j) + Y[j] * X.tail(N-j); + } + + static EIGEN_DONT_INLINE void ger(gene_matrix & A, gene_vector & X, gene_vector & Y, int N){ + for(int j=0; j<N; ++j) + A.col(j) += X * Y[j]; + } + + static EIGEN_DONT_INLINE void rot(gene_vector & A, gene_vector & B, real c, real s, int N){ + internal::apply_rotation_in_the_plane(A, B, JacobiRotation<real>(c,s)); + } + + static inline void atv_product(gene_matrix & A, gene_vector & B, gene_vector & X, int N){ + X.noalias() = (A.transpose()*B); + } + + static inline void axpy(real coef, const gene_vector & X, gene_vector & Y, int N){ + Y += coef * X; + } + + static inline void axpby(real a, const gene_vector & X, real b, gene_vector & Y, int N){ + Y = a*X + b*Y; + } + + static EIGEN_DONT_INLINE void copy_matrix(const gene_matrix & source, gene_matrix & cible, int N){ + cible = source; + } + + static EIGEN_DONT_INLINE void copy_vector(const gene_vector & source, gene_vector & cible, int N){ + cible = source; + } + + static inline void trisolve_lower(const gene_matrix & L, const gene_vector& B, gene_vector& X, int N){ + X = L.template triangularView<Lower>().solve(B); + } + + static inline void trisolve_lower_matrix(const gene_matrix & L, const gene_matrix& B, gene_matrix& X, int N){ + X = L.template triangularView<Upper>().solve(B); + } + + static inline void trmm(const gene_matrix & L, const gene_matrix& B, gene_matrix& X, int N){ + X.noalias() = L.template triangularView<Lower>() * B; + } + + static inline void cholesky(const gene_matrix & X, gene_matrix & C, int N){ + C = X; + internal::llt_inplace<real,Lower>::blocked(C); + //C = X.llt().matrixL(); +// C = X; +// Cholesky<gene_matrix>::computeInPlace(C); +// Cholesky<gene_matrix>::computeInPlaceBlock(C); + } + + static inline void lu_decomp(const gene_matrix & X, gene_matrix & C, int N){ + C = X.fullPivLu().matrixLU(); + } + + static inline void partial_lu_decomp(const gene_matrix & X, gene_matrix & C, int N){ + Matrix<DenseIndex,1,Dynamic> piv(N); + DenseIndex nb; + C = X; + internal::partial_lu_inplace(C,piv,nb); +// C = X.partialPivLu().matrixLU(); + } + + static inline void tridiagonalization(const gene_matrix & X, gene_matrix & C, int N){ + typename Tridiagonalization<gene_matrix>::CoeffVectorType aux(N-1); + C = X; + internal::tridiagonalization_inplace(C, aux); + } + + static inline void hessenberg(const gene_matrix & X, gene_matrix & C, int N){ + C = HessenbergDecomposition<gene_matrix>(X).packedMatrix(); + } + + + +}; + +#endif diff --git a/bench/btl/libs/eigen3/main_adv.cpp b/bench/btl/libs/eigen3/main_adv.cpp new file mode 100644 index 000000000..efe5857e4 --- /dev/null +++ b/bench/btl/libs/eigen3/main_adv.cpp @@ -0,0 +1,44 @@ +//===================================================== +// Copyright (C) 2008 Gael Guennebaud <gael.guennebaud@inria.fr> +//===================================================== +// +// This program is free software; you can redistribute it and/or +// modify it under the terms of the GNU General Public License +// as published by the Free Software Foundation; either version 2 +// of the License, or (at your option) any later version. +// +// This program is distributed in the hope that it will be useful, +// but WITHOUT ANY WARRANTY; without even the implied warranty of +// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +// GNU General Public License for more details. +// You should have received a copy of the GNU General Public License +// along with this program; if not, write to the Free Software +// Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. +// +#include "utilities.h" +#include "eigen3_interface.hh" +#include "bench.hh" +#include "action_trisolve.hh" +#include "action_trisolve_matrix.hh" +#include "action_cholesky.hh" +#include "action_hessenberg.hh" +#include "action_lu_decomp.hh" +#include "action_partial_lu.hh" + +BTL_MAIN; + +int main() +{ + bench<Action_trisolve<eigen3_interface<REAL_TYPE> > >(MIN_MM,MAX_MM,NB_POINT); + bench<Action_trisolve_matrix<eigen3_interface<REAL_TYPE> > >(MIN_MM,MAX_MM,NB_POINT); + bench<Action_cholesky<eigen3_interface<REAL_TYPE> > >(MIN_MM,MAX_MM,NB_POINT); + bench<Action_lu_decomp<eigen3_interface<REAL_TYPE> > >(MIN_MM,MAX_MM,NB_POINT); + bench<Action_partial_lu<eigen3_interface<REAL_TYPE> > >(MIN_MM,MAX_MM,NB_POINT); + + bench<Action_hessenberg<eigen3_interface<REAL_TYPE> > >(MIN_MM,MAX_MM,NB_POINT); + bench<Action_tridiagonalization<eigen3_interface<REAL_TYPE> > >(MIN_MM,MAX_MM,NB_POINT); + + return 0; +} + + diff --git a/bench/btl/libs/eigen3/main_linear.cpp b/bench/btl/libs/eigen3/main_linear.cpp new file mode 100644 index 000000000..e8538b7d0 --- /dev/null +++ b/bench/btl/libs/eigen3/main_linear.cpp @@ -0,0 +1,35 @@ +//===================================================== +// Copyright (C) 2008 Gael Guennebaud <gael.guennebaud@inria.fr> +//===================================================== +// +// This program is free software; you can redistribute it and/or +// modify it under the terms of the GNU General Public License +// as published by the Free Software Foundation; either version 2 +// of the License, or (at your option) any later version. +// +// This program is distributed in the hope that it will be useful, +// but WITHOUT ANY WARRANTY; without even the implied warranty of +// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +// GNU General Public License for more details. +// You should have received a copy of the GNU General Public License +// along with this program; if not, write to the Free Software +// Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. +// +#include "utilities.h" +#include "eigen3_interface.hh" +#include "bench.hh" +#include "basic_actions.hh" + +BTL_MAIN; + +int main() +{ + + bench<Action_axpy<eigen3_interface<REAL_TYPE> > >(MIN_AXPY,MAX_AXPY,NB_POINT); + bench<Action_axpby<eigen3_interface<REAL_TYPE> > >(MIN_AXPY,MAX_AXPY,NB_POINT); + bench<Action_rot<eigen3_interface<REAL_TYPE> > >(MIN_AXPY,MAX_AXPY,NB_POINT); + + return 0; +} + + diff --git a/bench/btl/libs/eigen3/main_matmat.cpp b/bench/btl/libs/eigen3/main_matmat.cpp new file mode 100644 index 000000000..926fa2b01 --- /dev/null +++ b/bench/btl/libs/eigen3/main_matmat.cpp @@ -0,0 +1,35 @@ +//===================================================== +// Copyright (C) 2008 Gael Guennebaud <gael.guennebaud@inria.fr> +//===================================================== +// +// This program is free software; you can redistribute it and/or +// modify it under the terms of the GNU General Public License +// as published by the Free Software Foundation; either version 2 +// of the License, or (at your option) any later version. +// +// This program is distributed in the hope that it will be useful, +// but WITHOUT ANY WARRANTY; without even the implied warranty of +// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +// GNU General Public License for more details. +// You should have received a copy of the GNU General Public License +// along with this program; if not, write to the Free Software +// Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. +// +#include "utilities.h" +#include "eigen3_interface.hh" +#include "bench.hh" +#include "basic_actions.hh" + +BTL_MAIN; + +int main() +{ + bench<Action_matrix_matrix_product<eigen3_interface<REAL_TYPE> > >(MIN_MM,MAX_MM,NB_POINT); +// bench<Action_ata_product<eigen3_interface<REAL_TYPE> > >(MIN_MM,MAX_MM,NB_POINT); + bench<Action_aat_product<eigen3_interface<REAL_TYPE> > >(MIN_MM,MAX_MM,NB_POINT); + bench<Action_trmm<eigen3_interface<REAL_TYPE> > >(MIN_MM,MAX_MM,NB_POINT); + + return 0; +} + + diff --git a/bench/btl/libs/eigen3/main_vecmat.cpp b/bench/btl/libs/eigen3/main_vecmat.cpp new file mode 100644 index 000000000..0dda444cf --- /dev/null +++ b/bench/btl/libs/eigen3/main_vecmat.cpp @@ -0,0 +1,36 @@ +//===================================================== +// Copyright (C) 2008 Gael Guennebaud <gael.guennebaud@inria.fr> +//===================================================== +// +// This program is free software; you can redistribute it and/or +// modify it under the terms of the GNU General Public License +// as published by the Free Software Foundation; either version 2 +// of the License, or (at your option) any later version. +// +// This program is distributed in the hope that it will be useful, +// but WITHOUT ANY WARRANTY; without even the implied warranty of +// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +// GNU General Public License for more details. +// You should have received a copy of the GNU General Public License +// along with this program; if not, write to the Free Software +// Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. +// +#include "utilities.h" +#include "eigen3_interface.hh" +#include "bench.hh" +#include "basic_actions.hh" + +BTL_MAIN; + +int main() +{ + bench<Action_matrix_vector_product<eigen3_interface<REAL_TYPE> > >(MIN_MV,MAX_MV,NB_POINT); + bench<Action_atv_product<eigen3_interface<REAL_TYPE> > >(MIN_MV,MAX_MV,NB_POINT); + bench<Action_symv<eigen3_interface<REAL_TYPE> > >(MIN_MV,MAX_MV,NB_POINT); + bench<Action_syr2<eigen3_interface<REAL_TYPE> > >(MIN_MV,MAX_MV,NB_POINT); + bench<Action_ger<eigen3_interface<REAL_TYPE> > >(MIN_MV,MAX_MV,NB_POINT); + + return 0; +} + + diff --git a/bench/btl/libs/gmm/CMakeLists.txt b/bench/btl/libs/gmm/CMakeLists.txt new file mode 100644 index 000000000..bc2586243 --- /dev/null +++ b/bench/btl/libs/gmm/CMakeLists.txt @@ -0,0 +1,6 @@ + +find_package(GMM) +if (GMM_FOUND) + include_directories(${GMM_INCLUDES}) + btl_add_bench(btl_gmm main.cpp) +endif (GMM_FOUND) diff --git a/bench/btl/libs/gmm/gmm_LU_solve_interface.hh b/bench/btl/libs/gmm/gmm_LU_solve_interface.hh new file mode 100644 index 000000000..dcb9f567f --- /dev/null +++ b/bench/btl/libs/gmm/gmm_LU_solve_interface.hh @@ -0,0 +1,192 @@ +//===================================================== +// File : blitz_LU_solve_interface.hh +// Author : L. Plagne <laurent.plagne@edf.fr)> +// Copyright (C) EDF R&D, lun sep 30 14:23:31 CEST 2002 +//===================================================== +// +// This program is free software; you can redistribute it and/or +// modify it under the terms of the GNU General Public License +// as published by the Free Software Foundation; either version 2 +// of the License, or (at your option) any later version. +// +// This program is distributed in the hope that it will be useful, +// but WITHOUT ANY WARRANTY; without even the implied warranty of +// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +// GNU General Public License for more details. +// You should have received a copy of the GNU General Public License +// along with this program; if not, write to the Free Software +// Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. +// +#ifndef BLITZ_LU_SOLVE_INTERFACE_HH +#define BLITZ_LU_SOLVE_INTERFACE_HH + +#include "blitz/array.h" +#include <vector> + +BZ_USING_NAMESPACE(blitz) + +template<class real> +class blitz_LU_solve_interface : public blitz_interface<real> +{ + +public : + + typedef typename blitz_interface<real>::gene_matrix gene_matrix; + typedef typename blitz_interface<real>::gene_vector gene_vector; + + typedef blitz::Array<int,1> Pivot_Vector; + + inline static void new_Pivot_Vector(Pivot_Vector & pivot,int N) + { + + pivot.resize(N); + + } + + inline static void free_Pivot_Vector(Pivot_Vector & pivot) + { + + return; + + } + + + static inline real matrix_vector_product_sliced(const gene_matrix & A, gene_vector B, int row, int col_start, int col_end) + { + + real somme=0.; + + for (int j=col_start ; j<col_end+1 ; j++){ + + somme+=A(row,j)*B(j); + + } + + return somme; + + } + + + + + static inline real matrix_matrix_product_sliced(gene_matrix & A, int row, int col_start, int col_end, gene_matrix & B, int row_shift, int col ) + { + + real somme=0.; + + for (int j=col_start ; j<col_end+1 ; j++){ + + somme+=A(row,j)*B(j+row_shift,col); + + } + + return somme; + + } + + inline static void LU_factor(gene_matrix & LU, Pivot_Vector & pivot, int N) + { + + ASSERT( LU.rows()==LU.cols() ) ; + int index_max = 0 ; + real big = 0. ; + real theSum = 0. ; + real dum = 0. ; + // Get the implicit scaling information : + gene_vector ImplicitScaling( N ) ; + for( int i=0; i<N; i++ ) { + big = 0. ; + for( int j=0; j<N; j++ ) { + if( abs( LU( i, j ) )>=big ) big = abs( LU( i, j ) ) ; + } + if( big==0. ) { + INFOS( "blitz_LU_factor::Singular matrix" ) ; + exit( 0 ) ; + } + ImplicitScaling( i ) = 1./big ; + } + // Loop over columns of Crout's method : + for( int j=0; j<N; j++ ) { + for( int i=0; i<j; i++ ) { + theSum = LU( i, j ) ; + theSum -= matrix_matrix_product_sliced(LU, i, 0, i-1, LU, 0, j) ; + // theSum -= sum( LU( i, Range( fromStart, i-1 ) )*LU( Range( fromStart, i-1 ), j ) ) ; + LU( i, j ) = theSum ; + } + + // Search for the largest pivot element : + big = 0. ; + for( int i=j; i<N; i++ ) { + theSum = LU( i, j ) ; + theSum -= matrix_matrix_product_sliced(LU, i, 0, j-1, LU, 0, j) ; + // theSum -= sum( LU( i, Range( fromStart, j-1 ) )*LU( Range( fromStart, j-1 ), j ) ) ; + LU( i, j ) = theSum ; + if( (ImplicitScaling( i )*abs( theSum ))>=big ) { + dum = ImplicitScaling( i )*abs( theSum ) ; + big = dum ; + index_max = i ; + } + } + // Interchanging rows and the scale factor : + if( j!=index_max ) { + for( int k=0; k<N; k++ ) { + dum = LU( index_max, k ) ; + LU( index_max, k ) = LU( j, k ) ; + LU( j, k ) = dum ; + } + ImplicitScaling( index_max ) = ImplicitScaling( j ) ; + } + pivot( j ) = index_max ; + if ( LU( j, j )==0. ) LU( j, j ) = 1.e-20 ; + // Divide by the pivot element : + if( j<N ) { + dum = 1./LU( j, j ) ; + for( int i=j+1; i<N; i++ ) LU( i, j ) *= dum ; + } + } + + } + + inline static void LU_solve(const gene_matrix & LU, const Pivot_Vector pivot, gene_vector &B, gene_vector X, int N) + { + + // Pour conserver le meme header, on travaille sur X, copie du second-membre B + X = B.copy() ; + ASSERT( LU.rows()==LU.cols() ) ; + firstIndex indI ; + // Forward substitution : + int ii = 0 ; + real theSum = 0. ; + for( int i=0; i<N; i++ ) { + int ip = pivot( i ) ; + theSum = X( ip ) ; + // theSum = B( ip ) ; + X( ip ) = X( i ) ; + // B( ip ) = B( i ) ; + if( ii ) { + theSum -= matrix_vector_product_sliced(LU, X, i, ii-1, i-1) ; + // theSum -= sum( LU( i, Range( ii-1, i-1 ) )*X( Range( ii-1, i-1 ) ) ) ; + // theSum -= sum( LU( i, Range( ii-1, i-1 ) )*B( Range( ii-1, i-1 ) ) ) ; + } else if( theSum ) { + ii = i+1 ; + } + X( i ) = theSum ; + // B( i ) = theSum ; + } + // Backsubstitution : + for( int i=N-1; i>=0; i-- ) { + theSum = X( i ) ; + // theSum = B( i ) ; + theSum -= matrix_vector_product_sliced(LU, X, i, i+1, N) ; + // theSum -= sum( LU( i, Range( i+1, toEnd ) )*X( Range( i+1, toEnd ) ) ) ; + // theSum -= sum( LU( i, Range( i+1, toEnd ) )*B( Range( i+1, toEnd ) ) ) ; + // Store a component of the solution vector : + X( i ) = theSum/LU( i, i ) ; + // B( i ) = theSum/LU( i, i ) ; + } + + } + +}; + +#endif diff --git a/bench/btl/libs/gmm/gmm_interface.hh b/bench/btl/libs/gmm/gmm_interface.hh new file mode 100644 index 000000000..3ea303c1b --- /dev/null +++ b/bench/btl/libs/gmm/gmm_interface.hh @@ -0,0 +1,144 @@ +//===================================================== +// Copyright (C) 2008 Gael Guennebaud <gael.guennebaud@inria.fr> +//===================================================== +// +// This program is free software; you can redistribute it and/or +// modify it under the terms of the GNU General Public License +// as published by the Free Software Foundation; either version 2 +// of the License, or (at your option) any later version. +// +// This program is distributed in the hope that it will be useful, +// but WITHOUT ANY WARRANTY; without even the implied warranty of +// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +// GNU General Public License for more details. +// You should have received a copy of the GNU General Public License +// along with this program; if not, write to the Free Software +// Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. +// +#ifndef GMM_INTERFACE_HH +#define GMM_INTERFACE_HH + +#include <gmm/gmm.h> +#include <vector> + +using namespace gmm; + +template<class real> +class gmm_interface { + +public : + + typedef real real_type ; + + typedef std::vector<real> stl_vector; + typedef std::vector<stl_vector > stl_matrix; + + typedef gmm::dense_matrix<real> gene_matrix; + typedef stl_vector gene_vector; + + static inline std::string name( void ) + { + return "gmm"; + } + + static void free_matrix(gene_matrix & A, int N){ + return ; + } + + static void free_vector(gene_vector & B){ + return ; + } + + static inline void matrix_from_stl(gene_matrix & A, stl_matrix & A_stl){ + A.resize(A_stl[0].size(),A_stl.size()); + + for (int j=0; j<A_stl.size() ; j++){ + for (int i=0; i<A_stl[j].size() ; i++){ + A(i,j) = A_stl[j][i]; + } + } + } + + static inline void vector_from_stl(gene_vector & B, stl_vector & B_stl){ + B = B_stl; + } + + static inline void vector_to_stl(gene_vector & B, stl_vector & B_stl){ + B_stl = B; + } + + static inline void matrix_to_stl(gene_matrix & A, stl_matrix & A_stl){ + int N=A_stl.size(); + + for (int j=0;j<N;j++){ + A_stl[j].resize(N); + for (int i=0;i<N;i++){ + A_stl[j][i] = A(i,j); + } + } + } + + static inline void matrix_matrix_product(const gene_matrix & A, const gene_matrix & B, gene_matrix & X, int N){ + gmm::mult(A,B, X); + } + + static inline void transposed_matrix_matrix_product(const gene_matrix & A, const gene_matrix & B, gene_matrix & X, int N){ + gmm::mult(gmm::transposed(A),gmm::transposed(B), X); + } + + static inline void ata_product(const gene_matrix & A, gene_matrix & X, int N){ + gmm::mult(gmm::transposed(A),A, X); + } + + static inline void aat_product(const gene_matrix & A, gene_matrix & X, int N){ + gmm::mult(A,gmm::transposed(A), X); + } + + static inline void matrix_vector_product(gene_matrix & A, gene_vector & B, gene_vector & X, int N){ + gmm::mult(A,B,X); + } + + static inline void atv_product(gene_matrix & A, gene_vector & B, gene_vector & X, int N){ + gmm::mult(gmm::transposed(A),B,X); + } + + static inline void axpy(const real coef, const gene_vector & X, gene_vector & Y, int N){ + gmm::add(gmm::scaled(X,coef), Y); + } + + static inline void axpby(real a, const gene_vector & X, real b, gene_vector & Y, int N){ + gmm::add(gmm::scaled(X,a), gmm::scaled(Y,b), Y); + } + + static inline void copy_matrix(const gene_matrix & source, gene_matrix & cible, int N){ + gmm::copy(source,cible); + } + + static inline void copy_vector(const gene_vector & source, gene_vector & cible, int N){ + gmm::copy(source,cible); + } + + static inline void trisolve_lower(const gene_matrix & L, const gene_vector& B, gene_vector & X, int N){ + gmm::copy(B,X); + gmm::lower_tri_solve(L, X, false); + } + + static inline void partial_lu_decomp(const gene_matrix & X, gene_matrix & R, int N){ + gmm::copy(X,R); + std::vector<int> ipvt(N); + gmm::lu_factor(R, ipvt); + } + + static inline void hessenberg(const gene_matrix & X, gene_matrix & R, int N){ + gmm::copy(X,R); + gmm::Hessenberg_reduction(R,X,false); + } + + static inline void tridiagonalization(const gene_matrix & X, gene_matrix & R, int N){ + gmm::copy(X,R); + gmm::Householder_tridiagonalization(R,X,false); + } + +}; + +#endif diff --git a/bench/btl/libs/gmm/main.cpp b/bench/btl/libs/gmm/main.cpp new file mode 100644 index 000000000..1f0c051eb --- /dev/null +++ b/bench/btl/libs/gmm/main.cpp @@ -0,0 +1,51 @@ +//===================================================== +// Copyright (C) 2008 Gael Guennebaud <gael.guennebaud@inria.fr> +//===================================================== +// +// This program is free software; you can redistribute it and/or +// modify it under the terms of the GNU General Public License +// as published by the Free Software Foundation; either version 2 +// of the License, or (at your option) any later version. +// +// This program is distributed in the hope that it will be useful, +// but WITHOUT ANY WARRANTY; without even the implied warranty of +// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +// GNU General Public License for more details. +// You should have received a copy of the GNU General Public License +// along with this program; if not, write to the Free Software +// Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. +// +#include "utilities.h" +#include "gmm_interface.hh" +#include "bench.hh" +#include "basic_actions.hh" +#include "action_hessenberg.hh" +#include "action_partial_lu.hh" + +BTL_MAIN; + +int main() +{ + + bench<Action_axpy<gmm_interface<REAL_TYPE> > >(MIN_AXPY,MAX_AXPY,NB_POINT); + bench<Action_axpby<gmm_interface<REAL_TYPE> > >(MIN_AXPY,MAX_AXPY,NB_POINT); + + bench<Action_matrix_vector_product<gmm_interface<REAL_TYPE> > >(MIN_MV,MAX_MV,NB_POINT); + bench<Action_atv_product<gmm_interface<REAL_TYPE> > >(MIN_MV,MAX_MV,NB_POINT); + + bench<Action_matrix_matrix_product<gmm_interface<REAL_TYPE> > >(MIN_MM,MAX_MM,NB_POINT); +// bench<Action_ata_product<gmm_interface<REAL_TYPE> > >(MIN_MM,MAX_MM,NB_POINT); +// bench<Action_aat_product<gmm_interface<REAL_TYPE> > >(MIN_MM,MAX_MM,NB_POINT); + + bench<Action_trisolve<gmm_interface<REAL_TYPE> > >(MIN_MM,MAX_MM,NB_POINT); + //bench<Action_lu_solve<blitz_LU_solve_interface<REAL_TYPE> > >(MIN_LU,MAX_LU,NB_POINT); + + bench<Action_partial_lu<gmm_interface<REAL_TYPE> > >(MIN_MM,MAX_MM,NB_POINT); + + bench<Action_hessenberg<gmm_interface<REAL_TYPE> > >(MIN_MM,MAX_MM,NB_POINT); + bench<Action_tridiagonalization<gmm_interface<REAL_TYPE> > >(MIN_MM,MAX_MM,NB_POINT); + + return 0; +} + + diff --git a/bench/btl/libs/mtl4/.kdbgrc.main b/bench/btl/libs/mtl4/.kdbgrc.main new file mode 100644 index 000000000..fed082f7f --- /dev/null +++ b/bench/btl/libs/mtl4/.kdbgrc.main @@ -0,0 +1,12 @@ +[General] +DebuggerCmdStr= +DriverName=GDB +FileVersion=1 +OptionsSelected= +ProgramArgs= +TTYLevel=7 +WorkingDirectory= + +[Memory] +ColumnWidths=80,0 +NumExprs=0 diff --git a/bench/btl/libs/mtl4/CMakeLists.txt b/bench/btl/libs/mtl4/CMakeLists.txt new file mode 100644 index 000000000..14b47a808 --- /dev/null +++ b/bench/btl/libs/mtl4/CMakeLists.txt @@ -0,0 +1,6 @@ + +find_package(MTL4) +if (MTL4_FOUND) + include_directories(${MTL4_INCLUDE_DIR}) + btl_add_bench(btl_mtl4 main.cpp) +endif (MTL4_FOUND) diff --git a/bench/btl/libs/mtl4/main.cpp b/bench/btl/libs/mtl4/main.cpp new file mode 100644 index 000000000..96fcfb9c9 --- /dev/null +++ b/bench/btl/libs/mtl4/main.cpp @@ -0,0 +1,46 @@ +//===================================================== +// Copyright (C) 2008 Gael Guennebaud <gael.guennebaud@inria.fr> +//===================================================== +// +// This program is free software; you can redistribute it and/or +// modify it under the terms of the GNU General Public License +// as published by the Free Software Foundation; either version 2 +// of the License, or (at your option) any later version. +// +// This program is distributed in the hope that it will be useful, +// but WITHOUT ANY WARRANTY; without even the implied warranty of +// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +// GNU General Public License for more details. +// You should have received a copy of the GNU General Public License +// along with this program; if not, write to the Free Software +// Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. +// +#include "utilities.h" +#include "mtl4_interface.hh" +#include "bench.hh" +#include "basic_actions.hh" +#include "action_cholesky.hh" +// #include "action_lu_decomp.hh" + +BTL_MAIN; + +int main() +{ + + bench<Action_axpy<mtl4_interface<REAL_TYPE> > >(MIN_AXPY,MAX_AXPY,NB_POINT); + bench<Action_axpby<mtl4_interface<REAL_TYPE> > >(MIN_AXPY,MAX_AXPY,NB_POINT); + + bench<Action_matrix_vector_product<mtl4_interface<REAL_TYPE> > >(MIN_MV,MAX_MV,NB_POINT); + bench<Action_atv_product<mtl4_interface<REAL_TYPE> > >(MIN_MV,MAX_MV,NB_POINT); + bench<Action_matrix_matrix_product<mtl4_interface<REAL_TYPE> > >(MIN_MM,MAX_MM,NB_POINT); +// bench<Action_ata_product<mtl4_interface<REAL_TYPE> > >(MIN_MM,MAX_MM,NB_POINT); +// bench<Action_aat_product<mtl4_interface<REAL_TYPE> > >(MIN_MM,MAX_MM,NB_POINT); + + bench<Action_trisolve<mtl4_interface<REAL_TYPE> > >(MIN_MM,MAX_MM,NB_POINT); +// bench<Action_cholesky<mtl4_interface<REAL_TYPE> > >(MIN_MM,MAX_MM,NB_POINT); +// bench<Action_lu_decomp<mtl4_interface<REAL_TYPE> > >(MIN_MM,MAX_MM,NB_POINT); + + return 0; +} + + diff --git a/bench/btl/libs/mtl4/mtl4_LU_solve_interface.hh b/bench/btl/libs/mtl4/mtl4_LU_solve_interface.hh new file mode 100644 index 000000000..dcb9f567f --- /dev/null +++ b/bench/btl/libs/mtl4/mtl4_LU_solve_interface.hh @@ -0,0 +1,192 @@ +//===================================================== +// File : blitz_LU_solve_interface.hh +// Author : L. Plagne <laurent.plagne@edf.fr)> +// Copyright (C) EDF R&D, lun sep 30 14:23:31 CEST 2002 +//===================================================== +// +// This program is free software; you can redistribute it and/or +// modify it under the terms of the GNU General Public License +// as published by the Free Software Foundation; either version 2 +// of the License, or (at your option) any later version. +// +// This program is distributed in the hope that it will be useful, +// but WITHOUT ANY WARRANTY; without even the implied warranty of +// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +// GNU General Public License for more details. +// You should have received a copy of the GNU General Public License +// along with this program; if not, write to the Free Software +// Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. +// +#ifndef BLITZ_LU_SOLVE_INTERFACE_HH +#define BLITZ_LU_SOLVE_INTERFACE_HH + +#include "blitz/array.h" +#include <vector> + +BZ_USING_NAMESPACE(blitz) + +template<class real> +class blitz_LU_solve_interface : public blitz_interface<real> +{ + +public : + + typedef typename blitz_interface<real>::gene_matrix gene_matrix; + typedef typename blitz_interface<real>::gene_vector gene_vector; + + typedef blitz::Array<int,1> Pivot_Vector; + + inline static void new_Pivot_Vector(Pivot_Vector & pivot,int N) + { + + pivot.resize(N); + + } + + inline static void free_Pivot_Vector(Pivot_Vector & pivot) + { + + return; + + } + + + static inline real matrix_vector_product_sliced(const gene_matrix & A, gene_vector B, int row, int col_start, int col_end) + { + + real somme=0.; + + for (int j=col_start ; j<col_end+1 ; j++){ + + somme+=A(row,j)*B(j); + + } + + return somme; + + } + + + + + static inline real matrix_matrix_product_sliced(gene_matrix & A, int row, int col_start, int col_end, gene_matrix & B, int row_shift, int col ) + { + + real somme=0.; + + for (int j=col_start ; j<col_end+1 ; j++){ + + somme+=A(row,j)*B(j+row_shift,col); + + } + + return somme; + + } + + inline static void LU_factor(gene_matrix & LU, Pivot_Vector & pivot, int N) + { + + ASSERT( LU.rows()==LU.cols() ) ; + int index_max = 0 ; + real big = 0. ; + real theSum = 0. ; + real dum = 0. ; + // Get the implicit scaling information : + gene_vector ImplicitScaling( N ) ; + for( int i=0; i<N; i++ ) { + big = 0. ; + for( int j=0; j<N; j++ ) { + if( abs( LU( i, j ) )>=big ) big = abs( LU( i, j ) ) ; + } + if( big==0. ) { + INFOS( "blitz_LU_factor::Singular matrix" ) ; + exit( 0 ) ; + } + ImplicitScaling( i ) = 1./big ; + } + // Loop over columns of Crout's method : + for( int j=0; j<N; j++ ) { + for( int i=0; i<j; i++ ) { + theSum = LU( i, j ) ; + theSum -= matrix_matrix_product_sliced(LU, i, 0, i-1, LU, 0, j) ; + // theSum -= sum( LU( i, Range( fromStart, i-1 ) )*LU( Range( fromStart, i-1 ), j ) ) ; + LU( i, j ) = theSum ; + } + + // Search for the largest pivot element : + big = 0. ; + for( int i=j; i<N; i++ ) { + theSum = LU( i, j ) ; + theSum -= matrix_matrix_product_sliced(LU, i, 0, j-1, LU, 0, j) ; + // theSum -= sum( LU( i, Range( fromStart, j-1 ) )*LU( Range( fromStart, j-1 ), j ) ) ; + LU( i, j ) = theSum ; + if( (ImplicitScaling( i )*abs( theSum ))>=big ) { + dum = ImplicitScaling( i )*abs( theSum ) ; + big = dum ; + index_max = i ; + } + } + // Interchanging rows and the scale factor : + if( j!=index_max ) { + for( int k=0; k<N; k++ ) { + dum = LU( index_max, k ) ; + LU( index_max, k ) = LU( j, k ) ; + LU( j, k ) = dum ; + } + ImplicitScaling( index_max ) = ImplicitScaling( j ) ; + } + pivot( j ) = index_max ; + if ( LU( j, j )==0. ) LU( j, j ) = 1.e-20 ; + // Divide by the pivot element : + if( j<N ) { + dum = 1./LU( j, j ) ; + for( int i=j+1; i<N; i++ ) LU( i, j ) *= dum ; + } + } + + } + + inline static void LU_solve(const gene_matrix & LU, const Pivot_Vector pivot, gene_vector &B, gene_vector X, int N) + { + + // Pour conserver le meme header, on travaille sur X, copie du second-membre B + X = B.copy() ; + ASSERT( LU.rows()==LU.cols() ) ; + firstIndex indI ; + // Forward substitution : + int ii = 0 ; + real theSum = 0. ; + for( int i=0; i<N; i++ ) { + int ip = pivot( i ) ; + theSum = X( ip ) ; + // theSum = B( ip ) ; + X( ip ) = X( i ) ; + // B( ip ) = B( i ) ; + if( ii ) { + theSum -= matrix_vector_product_sliced(LU, X, i, ii-1, i-1) ; + // theSum -= sum( LU( i, Range( ii-1, i-1 ) )*X( Range( ii-1, i-1 ) ) ) ; + // theSum -= sum( LU( i, Range( ii-1, i-1 ) )*B( Range( ii-1, i-1 ) ) ) ; + } else if( theSum ) { + ii = i+1 ; + } + X( i ) = theSum ; + // B( i ) = theSum ; + } + // Backsubstitution : + for( int i=N-1; i>=0; i-- ) { + theSum = X( i ) ; + // theSum = B( i ) ; + theSum -= matrix_vector_product_sliced(LU, X, i, i+1, N) ; + // theSum -= sum( LU( i, Range( i+1, toEnd ) )*X( Range( i+1, toEnd ) ) ) ; + // theSum -= sum( LU( i, Range( i+1, toEnd ) )*B( Range( i+1, toEnd ) ) ) ; + // Store a component of the solution vector : + X( i ) = theSum/LU( i, i ) ; + // B( i ) = theSum/LU( i, i ) ; + } + + } + +}; + +#endif diff --git a/bench/btl/libs/mtl4/mtl4_interface.hh b/bench/btl/libs/mtl4/mtl4_interface.hh new file mode 100644 index 000000000..3795ac61e --- /dev/null +++ b/bench/btl/libs/mtl4/mtl4_interface.hh @@ -0,0 +1,144 @@ +//===================================================== +// Copyright (C) 2008 Gael Guennebaud <gael.guennebaud@inria.fr> +//===================================================== +// +// This program is free software; you can redistribute it and/or +// modify it under the terms of the GNU General Public License +// as published by the Free Software Foundation; either version 2 +// of the License, or (at your option) any later version. +// +// This program is distributed in the hope that it will be useful, +// but WITHOUT ANY WARRANTY; without even the implied warranty of +// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +// GNU General Public License for more details. +// You should have received a copy of the GNU General Public License +// along with this program; if not, write to the Free Software +// Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. +// +#ifndef MTL4_INTERFACE_HH +#define MTL4_INTERFACE_HH + +#include <boost/numeric/mtl/mtl.hpp> +#include <boost/numeric/mtl/utility/range_generator.hpp> +// #include <boost/numeric/mtl/operation/cholesky.hpp> +#include <vector> + +using namespace mtl; + +template<class real> +class mtl4_interface { + +public : + + typedef real real_type ; + + typedef std::vector<real> stl_vector; + typedef std::vector<stl_vector > stl_matrix; + + typedef mtl::dense2D<real, mtl::matrix::parameters<mtl::tag::col_major> > gene_matrix; + typedef mtl::dense_vector<real> gene_vector; + + static inline std::string name() { return "mtl4"; } + + static void free_matrix(gene_matrix & A, int N){ + return ; + } + + static void free_vector(gene_vector & B){ + return ; + } + + static inline void matrix_from_stl(gene_matrix & A, stl_matrix & A_stl){ + A.change_dim(A_stl[0].size(), A_stl.size()); + + for (int j=0; j<A_stl.size() ; j++){ + for (int i=0; i<A_stl[j].size() ; i++){ + A(i,j) = A_stl[j][i]; + } + } + } + + static inline void vector_from_stl(gene_vector & B, stl_vector & B_stl){ + B.change_dim(B_stl.size()); + for (int i=0; i<B_stl.size() ; i++){ + B[i] = B_stl[i]; + } + } + + static inline void vector_to_stl(gene_vector & B, stl_vector & B_stl){ + for (int i=0; i<B_stl.size() ; i++){ + B_stl[i] = B[i]; + } + } + + static inline void matrix_to_stl(gene_matrix & A, stl_matrix & A_stl){ + int N=A_stl.size(); + for (int j=0;j<N;j++){ + A_stl[j].resize(N); + for (int i=0;i<N;i++){ + A_stl[j][i] = A(i,j); + } + } + } + + static inline void matrix_matrix_product(const gene_matrix & A, const gene_matrix & B, gene_matrix & X, int N){ + X = (A*B); +// morton_dense<double, doppled_64_row_mask> C(N,N); +// C = B; +// X = (A*C); + } + + static inline void transposed_matrix_matrix_product(const gene_matrix & A, const gene_matrix & B, gene_matrix & X, int N){ + X = (trans(A)*trans(B)); + } + +// static inline void ata_product(const gene_matrix & A, gene_matrix & X, int N){ +// X = (trans(A)*A); +// } + + static inline void aat_product(const gene_matrix & A, gene_matrix & X, int N){ + X = (A*trans(A)); + } + + static inline void matrix_vector_product(gene_matrix & A, gene_vector & B, gene_vector & X, int N){ + X = (A*B); + } + + static inline void atv_product(gene_matrix & A, gene_vector & B, gene_vector & X, int N){ + X = (trans(A)*B); + } + + static inline void axpy(const real coef, const gene_vector & X, gene_vector & Y, int N){ + Y += coef * X; + } + + static inline void axpby(real a, const gene_vector & X, real b, gene_vector & Y, int N){ + Y = a*X + b*Y; + } + +// static inline void cholesky(const gene_matrix & X, gene_matrix & C, int N){ +// C = X; +// recursive_cholesky(C); +// } + +// static inline void lu_decomp(const gene_matrix & X, gene_matrix & R, int N){ +// R = X; +// std::vector<int> ipvt(N); +// lu_factor(R, ipvt); +// } + + static inline void trisolve_lower(const gene_matrix & L, const gene_vector& B, gene_vector & X, int N){ + X = lower_trisolve(L, B); + } + + static inline void copy_matrix(const gene_matrix & source, gene_matrix & cible, int N){ + cible = source; + } + + static inline void copy_vector(const gene_vector & source, gene_vector & cible, int N){ + cible = source; + } + +}; + +#endif diff --git a/bench/btl/libs/tvmet/CMakeLists.txt b/bench/btl/libs/tvmet/CMakeLists.txt new file mode 100644 index 000000000..25b565b97 --- /dev/null +++ b/bench/btl/libs/tvmet/CMakeLists.txt @@ -0,0 +1,6 @@ + +find_package(Tvmet) +if (TVMET_FOUND) + include_directories(${TVMET_INCLUDE_DIR}) + btl_add_bench(btl_tvmet main.cpp OFF) +endif (TVMET_FOUND) diff --git a/bench/btl/libs/tvmet/main.cpp b/bench/btl/libs/tvmet/main.cpp new file mode 100644 index 000000000..633215c43 --- /dev/null +++ b/bench/btl/libs/tvmet/main.cpp @@ -0,0 +1,40 @@ +//===================================================== +// File : main.cpp +// Author : L. Plagne <laurent.plagne@edf.fr)> +// Copyright (C) EDF R&D, lun sep 30 14:23:30 CEST 2002 +//===================================================== +// +// This program is free software; you can redistribute it and/or +// modify it under the terms of the GNU General Public License +// as published by the Free Software Foundation; either version 2 +// of the License, or (at your option) any later version. +// +// This program is distributed in the hope that it will be useful, +// but WITHOUT ANY WARRANTY; without even the implied warranty of +// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +// GNU General Public License for more details. +// You should have received a copy of the GNU General Public License +// along with this program; if not, write to the Free Software +// Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. +// +#include "utilities.h" +#include "tvmet_interface.hh" +#include "static/bench_static.hh" +#include "action_matrix_vector_product.hh" +#include "action_matrix_matrix_product.hh" +#include "action_atv_product.hh" +#include "action_axpy.hh" + +BTL_MAIN; + +int main() +{ + bench_static<Action_axpy,tvmet_interface>(); + bench_static<Action_matrix_matrix_product,tvmet_interface>(); + bench_static<Action_matrix_vector_product,tvmet_interface>(); + bench_static<Action_atv_product,tvmet_interface>(); + + return 0; +} + + diff --git a/bench/btl/libs/tvmet/tvmet_interface.hh b/bench/btl/libs/tvmet/tvmet_interface.hh new file mode 100644 index 000000000..b441ada21 --- /dev/null +++ b/bench/btl/libs/tvmet/tvmet_interface.hh @@ -0,0 +1,104 @@ +//===================================================== +// File : tvmet_interface.hh +// Author : L. Plagne <laurent.plagne@edf.fr)> +// Copyright (C) EDF R&D, lun sep 30 14:23:30 CEST 2002 +//===================================================== +// +// This program is free software; you can redistribute it and/or +// modify it under the terms of the GNU General Public License +// as published by the Free Software Foundation; either version 2 +// of the License, or (at your option) any later version. +// +// This program is distributed in the hope that it will be useful, +// but WITHOUT ANY WARRANTY; without even the implied warranty of +// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +// GNU General Public License for more details. +// You should have received a copy of the GNU General Public License +// along with this program; if not, write to the Free Software +// Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. +// +#ifndef TVMET_INTERFACE_HH +#define TVMET_INTERFACE_HH + +#include <tvmet/tvmet.h> +#include <tvmet/Vector.h> +#include <tvmet/Matrix.h> + +#include <vector> + +using namespace tvmet; + +template<class real, int SIZE> +class tvmet_interface{ + +public : + + typedef real real_type ; + + typedef std::vector<real> stl_vector; + typedef std::vector<stl_vector > stl_matrix; + + typedef Vector<real,SIZE> gene_vector; + typedef Matrix<real,SIZE,SIZE> gene_matrix; + + static inline std::string name() { return "tiny_tvmet"; } + + static void free_matrix(gene_matrix & A, int N){} + + static void free_vector(gene_vector & B){} + + static inline void matrix_from_stl(gene_matrix & A, stl_matrix & A_stl){ + for (int j=0; j<A_stl.size() ; j++) + for (int i=0; i<A_stl[j].size() ; i++) + A(i,j) = A_stl[j][i]; + } + + static inline void vector_from_stl(gene_vector & B, stl_vector & B_stl){ + for (int i=0; i<B_stl.size() ; i++) + B[i]=B_stl[i]; + } + + static inline void vector_to_stl(gene_vector & B, stl_vector & B_stl){ + for (int i=0; i<B_stl.size() ; i++){ + B_stl[i]=B[i]; + } + } + + static inline void matrix_to_stl(gene_matrix & A, stl_matrix & A_stl){ + int N = A_stl.size(); + for (int j=0;j<N;j++){ + A_stl[j].resize(N); + for (int i=0;i<N;i++) + A_stl[j][i] = A(i,j); + } + } + + + static inline void copy_matrix(const gene_matrix & source, gene_matrix & cible, int N){ + cible = source; + } + + static inline void copy_vector(const gene_vector & source, gene_vector & cible, int N){ + cible = source; + } + + static inline void matrix_matrix_product(const gene_matrix & A, const gene_matrix & B, gene_matrix & X, int N){ + X = prod(A,B); + } + + static inline void matrix_vector_product(gene_matrix & A, gene_vector & B, gene_vector & X, int N){ + X = prod(A,B); + } + + static inline void atv_product(gene_matrix & A, gene_vector & B, gene_vector & X, int N){ + X = prod(trans(A),B); + } + + static inline void axpy(const real coef, const gene_vector & X, gene_vector & Y, int N){ + Y+=coef*X; + } + +}; + + +#endif diff --git a/bench/btl/libs/ublas/CMakeLists.txt b/bench/btl/libs/ublas/CMakeLists.txt new file mode 100644 index 000000000..bdb58bea1 --- /dev/null +++ b/bench/btl/libs/ublas/CMakeLists.txt @@ -0,0 +1,7 @@ + +find_package(Boost) +if (Boost_FOUND) + include_directories(${Boost_INCLUDE_DIRS}) + include_directories(${Boost_INCLUDES}) + btl_add_bench(btl_ublas main.cpp) +endif (Boost_FOUND) diff --git a/bench/btl/libs/ublas/main.cpp b/bench/btl/libs/ublas/main.cpp new file mode 100644 index 000000000..e2e77ee1f --- /dev/null +++ b/bench/btl/libs/ublas/main.cpp @@ -0,0 +1,44 @@ +//===================================================== +// File : main.cpp +// Author : L. Plagne <laurent.plagne@edf.fr)> +// Copyright (C) EDF R&D, lun sep 30 14:23:27 CEST 2002 +//===================================================== +// +// This program is free software; you can redistribute it and/or +// modify it under the terms of the GNU General Public License +// as published by the Free Software Foundation; either version 2 +// of the License, or (at your option) any later version. +// +// This program is distributed in the hope that it will be useful, +// but WITHOUT ANY WARRANTY; without even the implied warranty of +// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +// GNU General Public License for more details. +// You should have received a copy of the GNU General Public License +// along with this program; if not, write to the Free Software +// Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. +// +#include "utilities.h" +#include "ublas_interface.hh" +#include "bench.hh" +#include "basic_actions.hh" + +BTL_MAIN; + +int main() +{ + bench<Action_axpy<ublas_interface<REAL_TYPE> > >(MIN_AXPY,MAX_AXPY,NB_POINT); + bench<Action_axpby<ublas_interface<REAL_TYPE> > >(MIN_AXPY,MAX_AXPY,NB_POINT); + + bench<Action_matrix_vector_product<ublas_interface<REAL_TYPE> > >(MIN_MV,MAX_MV,NB_POINT); + bench<Action_atv_product<ublas_interface<REAL_TYPE> > >(MIN_MV,MAX_MV,NB_POINT); + + bench<Action_matrix_matrix_product<ublas_interface<REAL_TYPE> > >(MIN_MM,MAX_MM,NB_POINT); +// bench<Action_ata_product<ublas_interface<REAL_TYPE> > >(MIN_MM,MAX_MM,NB_POINT); +// bench<Action_aat_product<ublas_interface<REAL_TYPE> > >(MIN_MM,MAX_MM,NB_POINT); + + bench<Action_trisolve<ublas_interface<REAL_TYPE> > >(MIN_MM,MAX_MM,NB_POINT); + + return 0; +} + + diff --git a/bench/btl/libs/ublas/ublas_interface.hh b/bench/btl/libs/ublas/ublas_interface.hh new file mode 100644 index 000000000..95cad5195 --- /dev/null +++ b/bench/btl/libs/ublas/ublas_interface.hh @@ -0,0 +1,141 @@ +//===================================================== +// File : ublas_interface.hh +// Author : L. Plagne <laurent.plagne@edf.fr)> +// Copyright (C) EDF R&D, lun sep 30 14:23:27 CEST 2002 +//===================================================== +// +// This program is free software; you can redistribute it and/or +// modify it under the terms of the GNU General Public License +// as published by the Free Software Foundation; either version 2 +// of the License, or (at your option) any later version. +// +// This program is distributed in the hope that it will be useful, +// but WITHOUT ANY WARRANTY; without even the implied warranty of +// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +// GNU General Public License for more details. +// You should have received a copy of the GNU General Public License +// along with this program; if not, write to the Free Software +// Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. +// +#ifndef UBLAS_INTERFACE_HH +#define UBLAS_INTERFACE_HH + +#include <boost/numeric/ublas/vector.hpp> +#include <boost/numeric/ublas/matrix.hpp> +#include <boost/numeric/ublas/io.hpp> +#include <boost/numeric/ublas/triangular.hpp> + +using namespace boost::numeric; + +template <class real> +class ublas_interface{ + +public : + + typedef real real_type ; + + typedef std::vector<real> stl_vector; + typedef std::vector<stl_vector> stl_matrix; + + typedef typename boost::numeric::ublas::matrix<real,boost::numeric::ublas::column_major> gene_matrix; + typedef typename boost::numeric::ublas::vector<real> gene_vector; + + static inline std::string name( void ) { return "ublas"; } + + static void free_matrix(gene_matrix & A, int N) {} + + static void free_vector(gene_vector & B) {} + + static inline void matrix_from_stl(gene_matrix & A, stl_matrix & A_stl){ + A.resize(A_stl.size(),A_stl[0].size()); + for (int j=0; j<A_stl.size() ; j++) + for (int i=0; i<A_stl[j].size() ; i++) + A(i,j)=A_stl[j][i]; + } + + static inline void vector_from_stl(gene_vector & B, stl_vector & B_stl){ + B.resize(B_stl.size()); + for (int i=0; i<B_stl.size() ; i++) + B(i)=B_stl[i]; + } + + static inline void vector_to_stl(gene_vector & B, stl_vector & B_stl){ + for (int i=0; i<B_stl.size() ; i++) + B_stl[i]=B(i); + } + + static inline void matrix_to_stl(gene_matrix & A, stl_matrix & A_stl){ + int N=A_stl.size(); + for (int j=0;j<N;j++) + { + A_stl[j].resize(N); + for (int i=0;i<N;i++) + A_stl[j][i]=A(i,j); + } + } + + static inline void copy_vector(const gene_vector & source, gene_vector & cible, int N){ + for (int i=0;i<N;i++){ + cible(i) = source(i); + } + } + + static inline void copy_matrix(const gene_matrix & source, gene_matrix & cible, int N){ + for (int i=0;i<N;i++){ + for (int j=0;j<N;j++){ + cible(i,j) = source(i,j); + } + } + } + + static inline void matrix_vector_product_slow(gene_matrix & A, gene_vector & B, gene_vector & X, int N){ + X = prod(A,B); + } + + static inline void matrix_matrix_product_slow(gene_matrix & A, gene_matrix & B, gene_matrix & X, int N){ + X = prod(A,B); + } + + static inline void axpy_slow(const real coef, const gene_vector & X, gene_vector & Y, int N){ + Y+=coef*X; + } + + // alias free assignements + + static inline void matrix_vector_product(gene_matrix & A, gene_vector & B, gene_vector & X, int N){ + X.assign(prod(A,B)); + } + + static inline void atv_product(gene_matrix & A, gene_vector & B, gene_vector & X, int N){ + X.assign(prod(trans(A),B)); + } + + static inline void matrix_matrix_product(gene_matrix & A, gene_matrix & B, gene_matrix & X, int N){ + X.assign(prod(A,B)); + } + + static inline void axpy(const real coef, const gene_vector & X, gene_vector & Y, int N){ + Y.plus_assign(coef*X); + } + + static inline void axpby(real a, const gene_vector & X, real b, gene_vector & Y, int N){ + Y = a*X + b*Y; + } + + static inline void ata_product(gene_matrix & A, gene_matrix & X, int N){ + // X = prod(trans(A),A); + X.assign(prod(trans(A),A)); + } + + static inline void aat_product(gene_matrix & A, gene_matrix & X, int N){ + // X = prod(A,trans(A)); + X.assign(prod(A,trans(A))); + } + + static inline void trisolve_lower(const gene_matrix & L, const gene_vector& B, gene_vector & X, int N){ + X = solve(L, B, ublas::lower_tag ()); + } + +}; + +#endif |