diff --git a/.gitlab/build.sh b/.gitlab/build.sh
index bd7b9259e6b1c7ff8388906d6c550b130da28699..de06d28ccbb1424496d8169a107b7c404ae247bb 100755
--- a/.gitlab/build.sh
+++ b/.gitlab/build.sh
@@ -1,11 +1,5 @@
#!/usr/bin/env bash
-
-fatal() {
- echo "$0: error occurred, exit"
- exit 1
-}
-
-set -x
+set -ex
SCAN=""
@@ -29,14 +23,14 @@ if [[ "$SYSTEM" != "windows" ]]; then
# to avoid the Accelerate framework and get Openblas we use BLA_PREFER_PKGCONFIG
cmake -B build-${VERSION} -S . \
-DBUILD_SHARED_LIBS=ON -DCMAKE_INSTALL_PREFIX=$PWD/install-${VERSION} \
- -DMORSE_ENABLE_COVERAGE=OFF -DBLA_PREFER_PKGCONFIG=ON || fatal
+ -DMORSE_ENABLE_COVERAGE=OFF -DBLA_PREFER_PKGCONFIG=ON
else
- source .gitlab-ci-env.sh $CHAM_CI_ENV_ARG || fatal
+ source .gitlab-ci-env.sh $CHAM_CI_ENV_ARG
if [[ $CI_COMMIT_REF_NAME == $CI_DEFAULT_BRANCH ]]
then
SCAN="scan-build -plist --intercept-first --exclude CMakeFiles --analyze-headers -o analyzer_reports "
fi
- eval '${SCAN}cmake -B build-${VERSION} -S . -C cmake_modules/gitlab-ci-initial-cache.cmake $BUILD_OPTIONS' || fatal
+ eval '${SCAN}cmake -B build-${VERSION} -S . -C cmake_modules/gitlab-ci-initial-cache.cmake $BUILD_OPTIONS'
fi
else
# on windows the mpi_f08 interface is missing, see https://www.scivision.dev/windows-mpi-msys2/
@@ -44,9 +38,41 @@ else
# directory can reach more than 10Go
cmake -GNinja -B build-${VERSION} -S . \
-DBUILD_SHARED_LIBS=ON -DCMAKE_INSTALL_PREFIX=$PWD/install-${VERSION} \
- -DCHAMELEON_USE_MPI=OFF \
- || fatal
+ -DCHAMELEON_USE_MPI=OFF
+fi
+eval '${SCAN}cmake --build build-${VERSION} -j 4'
+cmake --install build-${VERSION}
+
+#
+# Check link to chameleon
+#
+cd .gitlab/check_link/
+
+# Set the compiler
+if [[ "$SYSTEM" == "macosx" ]]; then
+ export CC=clang
+ if brew ls --versions chameleon > /dev/null; then brew remove --force --ignore-dependencies chameleon; fi
+else
+ export CC=gcc
+fi
+export FC=gfortran
+
+# Set the path variables
+if [[ "$SYSTEM" == "linux" ]]; then
+ export LIBRARY_PATH=$PWD/../../install-${VERSION}/lib:$LIBRARY_PATH
+ export LD_LIBRARY_PATH=$PWD/../../install-${VERSION}/lib:$LD_LIBRARY_PATH
+elif [[ "$SYSTEM" == "macosx" ]]; then
+ export LIBRARY_PATH=$PWD/../../install-${VERSION}/lib:$LIBRARY_PATH
+ export DYLD_LIBRARY_PATH=$PWD/../../install-${VERSION}/lib:$DYLD_LIBRARY_PATH
+elif [[ "$SYSTEM" == "windows" ]]; then
+ export PATH="/c/Windows/WinSxS/x86_microsoft-windows-m..namespace-downlevel_31bf3856ad364e35_10.0.19041.1_none_21374cb0681a6320":$PATH
+ export PATH=$PWD/../../install-${VERSION}/bin:$PATH
fi
-eval '${SCAN}cmake --build build-${VERSION} -j 4' || fatal
-cmake --install build-${VERSION} || fatal
-rm -r install-${VERSION} || fatal
+
+# 1) using cmake:
+./link_cmake.sh $PWD/../../install-${VERSION}
+# 2) using pkg-config:
+./link_pkgconfig.sh $PWD/../../install-${VERSION}
+
+cd ../..
+rm -r install-${VERSION}
diff --git a/.gitlab/check_link/CMakeLists.txt b/.gitlab/check_link/CMakeLists.txt
new file mode 100644
index 0000000000000000000000000000000000000000..33bf1f657a49c231b2600dc3aaaba3efb22be665
--- /dev/null
+++ b/.gitlab/check_link/CMakeLists.txt
@@ -0,0 +1,37 @@
+###
+#
+# @copyright 2013-2024 Bordeaux INP, CNRS (LaBRI UMR 5800), Inria,
+# Univ. Bordeaux. All rights reserved.
+#
+# @version 1.3.0
+# @author Mathieu Faverge
+# @author Florent Pruvost
+# @date 2024-06-13
+#
+###
+cmake_minimum_required(VERSION 3.5)
+
+project(LINK_CHAMELEON C CXX Fortran)
+
+# look for CHAMELEON on the system
+# Hint: use CMAKE_PREFIX_PATH (env. var. or cmake var.) to the installation
+# directory of CHAMELEON if not installed in a standard path
+find_package(CHAMELEON REQUIRED)
+
+if (TARGET CHAMELEON::chameleon)
+ message(STATUS "CHAMELEON_BIN_DIR ${CHAMELEON_BIN_DIR}")
+ message(STATUS "CHAMELEON_INC_DIR ${CHAMELEON_INC_DIR}")
+ message(STATUS "CHAMELEON_LIB_DIR ${CHAMELEON_LIB_DIR}")
+else()
+ message(FATAL_ERROR "Target CHAMELEON::chameleon is not found, check your CHAMELEONConfig.cmake.")
+endif()
+
+set(CMAKE_BUILD_RPATH ${CHAMELEON_LIB_DIR})
+
+add_executable(link_chameleon_c ${CMAKE_CURRENT_SOURCE_DIR}/../../example/link_chameleon/link_chameleon.c)
+target_link_libraries(link_chameleon_c PRIVATE CHAMELEON::chameleon)
+
+# Launch executables with ctest
+enable_testing()
+include(CTest)
+add_test(link_chameleon_c link_chameleon_c)
diff --git a/.gitlab/check_link/link_cmake.sh b/.gitlab/check_link/link_cmake.sh
new file mode 100755
index 0000000000000000000000000000000000000000..613218956c689e3187ae72e99207f2d51a2b2305
--- /dev/null
+++ b/.gitlab/check_link/link_cmake.sh
@@ -0,0 +1,27 @@
+#!/usr/bin/env bash
+###
+#
+# @file link_cmake.sh
+# @copyright 2023-2024 Bordeaux INP, CNRS (LaBRI UMR 5800), Inria,
+# Univ. Bordeaux. All rights reserved.
+#
+# @version 1.3.0
+# @author Florent Pruvost
+# @author Mathieu Faverge
+# @date 2023-12-07
+#
+# Check that linking with the project is ok when using cmake.
+#
+###
+set -ex
+
+if [ $# -lt 1 ]
+then
+ echo "usage: ./link_cmake.sh path_to_install"
+ exit 1
+fi
+cmake -B build -DCMAKE_PREFIX_PATH=$1
+cmake --build build --verbose
+if [[ "$VERSION" != "starpu_simgrid" ]]; then
+ ctest --test-dir build --verbose
+fi
diff --git a/.gitlab/check_link/link_pkgconfig.sh b/.gitlab/check_link/link_pkgconfig.sh
new file mode 100755
index 0000000000000000000000000000000000000000..f7fc478c6e3145d2cb61586ba86eb2fbd6bdfd95
--- /dev/null
+++ b/.gitlab/check_link/link_pkgconfig.sh
@@ -0,0 +1,58 @@
+#!/usr/bin/env bash
+###
+#
+# @file link_pkgconfig.sh
+# @copyright 2023-2024 Bordeaux INP, CNRS (LaBRI UMR 5800), Inria,
+# Univ. Bordeaux. All rights reserved.
+#
+# @version 1.3.0
+# @author Florent Pruvost
+# @author Mathieu Faverge
+# @date 2023-12-07
+#
+# Check that linking with the project is ok when using pkg-config.
+#
+###
+set -ex
+
+static=""
+chameleon_path=""
+
+while [ $# -gt 0 ]
+do
+ case $1 in
+ --static | -static | -s )
+ static="--static"
+ ;;
+ * )
+ chameleon_path=$1
+ ;;
+ esac
+ shift
+done
+
+if [ -z $chameleon_path ]
+then
+ echo """
+usage: ./link_pkgconfig.sh path_to_chameleon_install [--static|-static|-s]
+ use the --static parameter if chameleon is static (.a)
+ env. var. CC and FC must be defined to C and Fortran90 compilers
+"""
+ exit 1
+fi
+
+export PKG_CONFIG_PATH=$chameleon_path/lib/pkgconfig:$PKG_CONFIG_PATH
+
+mkdir -p build
+cd build
+
+FLAGS=`pkg-config $static --cflags chameleon`
+if [[ "$SYSTEM" == "macosx" ]]; then
+ FLAGS="-Wl,-rpath,$chameleon_path/lib $FLAGS"
+fi
+LIBS=`pkg-config $static --libs chameleon`
+$CC $FLAGS ../../../example/link_chameleon/link_chameleon.c $LIBS -o link_chameleon_c
+
+if [[ "$VERSION" != "starpu_simgrid" ]]; then
+ ./link_chameleon_c
+fi
diff --git a/CMakeLists.txt b/CMakeLists.txt
index 44b3fff3f6699f82de1df192d6980aa0f137c7b0..3892d9b94325f20b38d2f8a1b13099786e466d8c 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -262,8 +262,10 @@ endif()
# Option to activate or not simulation mode (use Simgrid through StarPU)
# ----------------------------------------------------------------------
+cmake_dependent_option(CHAMELEON_SIMULATION
+"Enable simulation mode using Simgrid through StarPU" OFF
+"CHAMELEON_SCHED_STARPU" OFF)
if(CHAMELEON_SCHED_STARPU)
- option(CHAMELEON_SIMULATION "Enable simulation mode using Simgrid through StarPU" OFF)
if (NOT CHAMELEON_SIMULATION)
message("-- ${BoldGreen}CHAMELEON_SIMULATION is set to OFF, turn it ON to use"
" SIMULATION mode (only with StarPU compiled with SimGrid)${ColourReset}")
@@ -529,6 +531,12 @@ if(NOT CHAMELEON_SIMULATION)
morse_export_imported_target(HIP HIP hip chameleon)
morse_export_imported_target(HIP HIPBLAS hipblas chameleon)
+ if (CHAMELEON_USE_HIP_ROC)
+ morse_export_imported_target(roc hipblas rochipblas chameleon)
+ morse_export_imported_target(hip host hiphost chameleon)
+ morse_export_imported_target(hip device hipdevice chameleon)
+ morse_export_imported_target(hip amdhip64 hipamdhip64 chameleon)
+ endif()
else(HIP_FOUND)
message(FATAL_ERROR "CHAMELEON_USE_HIP requires HIP to be found")
endif (HIP_FOUND)
diff --git a/cmake_modules/CHAMELEONConfig.cmake.in b/cmake_modules/CHAMELEONConfig.cmake.in
index b1f50a43812a77f5e92319868cb5bb5ed9bf1db6..2346b326cbf320c2eb6ca3d556ff330b32448e8d 100644
--- a/cmake_modules/CHAMELEONConfig.cmake.in
+++ b/cmake_modules/CHAMELEONConfig.cmake.in
@@ -37,6 +37,16 @@ if (@CHAMELEON_USE_CUDA@ AND NOT @CHAMELEON_SIMULATION@)
include("${CMAKE_CURRENT_LIST_DIR}/cudaTargets.cmake")
include("${CMAKE_CURRENT_LIST_DIR}/cublasTargets.cmake")
endif()
+if (@CHAMELEON_USE_HIP@ AND NOT @CHAMELEON_SIMULATION@)
+ include("${CMAKE_CURRENT_LIST_DIR}/hipTargets.cmake")
+ include("${CMAKE_CURRENT_LIST_DIR}/hipblasTargets.cmake")
+ if (@CHAMELEON_USE_HIP_ROC@)
+ include("${CMAKE_CURRENT_LIST_DIR}/rochipblasTargets.cmake")
+ include("${CMAKE_CURRENT_LIST_DIR}/hiphostTargets.cmake")
+ include("${CMAKE_CURRENT_LIST_DIR}/hipdeviceTargets.cmake")
+ include("${CMAKE_CURRENT_LIST_DIR}/hipamdhip64Targets.cmake")
+ endif()
+endif()
# add the targets file
if (@CHAMELEON_USE_HMATOSS@)
diff --git a/cmake_modules/GenPkgConfig.cmake b/cmake_modules/GenPkgConfig.cmake
index 4cdb04163484a6610d57bf769196b2e7bd4b2a8d..37bbfb8c1e4fccb932d5e4303bb268f3529c1359 100644
--- a/cmake_modules/GenPkgConfig.cmake
+++ b/cmake_modules/GenPkgConfig.cmake
@@ -128,10 +128,20 @@ MACRO(GENERATE_PKGCONFIG_FILE)
list(APPEND CHAMELEON_PKGCONFIG_LIBS_PRIVATE "${QUARK_LIBRARIES_DEP}")
elseif(CHAMELEON_SCHED_STARPU)
list(APPEND CHAMELEON_PKGCONFIG_LIBS -lchameleon_starpu)
- if ( CHAMELEON_USE_MPI )
- list(APPEND CHAMELEON_PKGCONFIG_REQUIRED_PRIVATE starpumpi-${CHAMELEON_STARPU_VERSION})
+ if ( CHAMELEON_SIMULATION )
+ # need PUBLIC here because simulation mode with starpu requires to replace the
+ # main function by a starpu main one
+ if ( CHAMELEON_USE_MPI )
+ list(APPEND CHAMELEON_PKGCONFIG_REQUIRED starpumpi-${CHAMELEON_STARPU_VERSION})
+ else()
+ list(APPEND CHAMELEON_PKGCONFIG_REQUIRED starpu-${CHAMELEON_STARPU_VERSION})
+ endif()
else()
- list(APPEND CHAMELEON_PKGCONFIG_REQUIRED_PRIVATE starpu-${CHAMELEON_STARPU_VERSION})
+ if ( CHAMELEON_USE_MPI )
+ list(APPEND CHAMELEON_PKGCONFIG_REQUIRED_PRIVATE starpumpi-${CHAMELEON_STARPU_VERSION})
+ else()
+ list(APPEND CHAMELEON_PKGCONFIG_REQUIRED_PRIVATE starpu-${CHAMELEON_STARPU_VERSION})
+ endif()
endif()
endif()
diff --git a/cmake_modules/morse_cmake b/cmake_modules/morse_cmake
index 92987016f5342650fa8b88553708c6dad44052bd..34e01580a425eb5fc13f5495a9a602fb26a3a589 160000
--- a/cmake_modules/morse_cmake
+++ b/cmake_modules/morse_cmake
@@ -1 +1 @@
-Subproject commit 92987016f5342650fa8b88553708c6dad44052bd
+Subproject commit 34e01580a425eb5fc13f5495a9a602fb26a3a589
diff --git a/example/link_chameleon/link_chameleon.c b/example/link_chameleon/link_chameleon.c
index 97027293790aeb99fd208a52b1a368b57e557807..bde80da784c96f084a7d493b543d0e4a88b114ae 100644
--- a/example/link_chameleon/link_chameleon.c
+++ b/example/link_chameleon/link_chameleon.c
@@ -2,295 +2,59 @@
*
* @file link_chameleon.c
*
- * @copyright 2009-2014 The University of Tennessee and The University of
- * Tennessee Research Foundation. All rights reserved.
- * @copyright 2012-2024 Bordeaux INP, CNRS (LaBRI UMR 5800), Inria,
+ * @copyright 2024-2024 Bordeaux INP, CNRS (LaBRI UMR 5800), Inria,
* Univ. Bordeaux. All rights reserved.
*
***
*
* @brief Chameleon link_chameleon example
*
- * @version 1.2.0
+ * @version 1.3.0
* @author Florent Pruvost
* @author Mathieu Faverge
- * @date 2022-02-22
+ * @date 2024-06-14
*
*/
-#include <math.h>
-#include <stdio.h>
-#include <stdlib.h>
-#include <string.h>
-#include <assert.h>
-
-#if defined( _WIN32 ) || defined( _WIN64 )
-#include <windows.h>
-#else /* Non-Windows */
-#include <unistd.h>
-#include <sys/resource.h>
-#endif
-
-
-/* Common functions for all steps of the tutorial */
-
-static void get_thread_count(int *thrdnbr) {
-#if defined WIN32 || defined WIN64
- sscanf( getenv( "NUMBER_OF_PROCESSORS" ), "%d", thrdnbr );
-#else
- *thrdnbr = sysconf(_SC_NPROCESSORS_ONLN);
-#endif
-}
-
-static int startswith(const char *s, const char *prefix) {
- size_t n = strlen( prefix );
- if (strncmp( s, prefix, n ))
- return 0;
- return 1;
-}
-
-
-/* define complexity of algorithms - see Lawn 41 page 120 */
-#define FMULS_POTRF(__n) ((double)(__n) * (((1. / 6.) * (double)(__n) + 0.5) * (double)(__n) + (1. / 3.)))
-#define FADDS_POTRF(__n) ((double)(__n) * (((1. / 6.) * (double)(__n) ) * (double)(__n) - (1. / 6.)))
-#define FMULS_TRSM(__m, __n) (0.5 * (double)(__n) * (double)(__m) * ((double)(__m)+1.))
-#define FADDS_TRSM(__m, __n) (0.5 * (double)(__n) * (double)(__m) * ((double)(__m)-1.))
-
-#include <coreblas/lapacke.h>
#include <chameleon.h>
-
-/* Integer parameters for step1 */
-enum iparam_step1 {
- IPARAM_THRDNBR, /* Number of cores */
- IPARAM_N, /* Number of columns of the matrix */
- IPARAM_NRHS, /* Number of RHS */
- /* End */
- IPARAM_SIZEOF
-};
-
-/* Specific routines used in step1.c main program */
-
-/**
- * Initialize integer parameters
- */
-static void init_iparam(int iparam[IPARAM_SIZEOF]){
- iparam[IPARAM_THRDNBR ] = -1;
- iparam[IPARAM_N ] = 500;
- iparam[IPARAM_NRHS ] = 1;
- }
-
-/**
- * Print how to use the program
- */
-static void show_help(char *prog_name) {
- printf( "Usage:\n%s [options]\n\n", prog_name );
- printf( "Options are:\n"
- " --help Show this help\n"
- "\n"
- " --n=X dimension (N). (default: 500)\n"
- " --nrhs=X number of RHS. (default: 1)\n"
- "\n"
- " --threads=X Number of CPU workers (default: _SC_NPROCESSORS_ONLN)\n"
- "\n");
-}
-
-/**
- * Read arguments following step1 program call
- */
-static void read_args(int argc, char *argv[], int *iparam){
- int i;
- for (i = 1; i < argc && argv[i]; ++i) {
- if ( startswith( argv[i], "--help") || startswith( argv[i], "-help") ||
- startswith( argv[i], "--h") || startswith( argv[i], "-h") ) {
- show_help( argv[0] );
- exit(0);
- } else if (startswith( argv[i], "--n=" )) {
- sscanf( strchr( argv[i], '=' ) + 1, "%d", &(iparam[IPARAM_N]) );
- } else if (startswith( argv[i], "--nrhs=" )) {
- sscanf( strchr( argv[i], '=' ) + 1, "%d", &(iparam[IPARAM_NRHS]) );
- } else if (startswith( argv[i], "--threads=" )) {
- sscanf( strchr( argv[i], '=' ) + 1, "%d", &(iparam[IPARAM_THRDNBR]) );
- } else {
- fprintf( stderr, "Unknown option: %s\n", argv[i] );
- }
- }
-}
-
+#include <stdlib.h>
/**
- * Print a header message to summarize main parameters
- */
-static void print_header(char *prog_name, int * iparam) {
-#if defined(CHAMELEON_SIMULATION)
- double eps = 0.;
-#else
- double eps = LAPACKE_dlamch_work( 'e' );
-#endif
-
- printf( "#\n"
- "# CHAMELEON %d.%d.%d, %s\n"
- "# Nb threads: %d\n"
- "# Nb gpus: %d\n"
- "# N: %d\n"
- "# NB: %d\n"
- "# IB: %d\n"
- "# eps: %e\n"
- "#\n",
- CHAMELEON_VERSION_MAJOR,
- CHAMELEON_VERSION_MINOR,
- CHAMELEON_VERSION_MICRO,
- prog_name,
- iparam[IPARAM_THRDNBR],
- 0,
- iparam[IPARAM_N],
- 128,
- 32,
- eps );
-
- printf( "# M N K/NRHS seconds Gflop/s\n");
- printf( "#%7d %7d %7d ", iparam[IPARAM_N], iparam[IPARAM_N], iparam[IPARAM_NRHS]);
- fflush( stdout );
- return;
-}
-
-/*
- * test external application link with chameleon
+ * @brief Matrix multiplication example to check library linking with Chameleon
*/
-int main(int argc, char *argv[]) {
-
- size_t i, j;
- size_t N; // matrix order
- size_t NRHS; // number of RHS vectors
- int NCPU; // number of cores to use
- int NGPU; // number of gpus (cuda devices) to use
- int UPLO = ChamUpper; // where is stored L
-
- /* declarations to time the program and evaluate performances */
- double fmuls, fadds, flops, gflops, cpu_time;
-
- /* variable to check the numerical results */
- double anorm, bnorm, xnorm, eps, res;
- int hres;
-
- int major, minor, patch;
- CHAMELEON_Version(&major, &minor, &patch);
-
- /* initialize some parameters with default values */
- int iparam[IPARAM_SIZEOF];
- memset(iparam, 0, IPARAM_SIZEOF*sizeof(int));
- init_iparam(iparam);
-
- /* read arguments */
- read_args(argc, argv, iparam);
- N = iparam[IPARAM_N];
- NRHS = iparam[IPARAM_NRHS];
-
- /* compute the algorithm complexity to evaluate performances */
- fadds = (double)( FADDS_POTRF(N) + 2 * FADDS_TRSM(N,NRHS) );
- fmuls = (double)( FMULS_POTRF(N) + 2 * FMULS_TRSM(N,NRHS) );
- flops = 1e-9 * (fmuls + fadds);
- gflops = 0.0;
- cpu_time = 0.0;
-
- /* initialize the number of thread if not given by the user in argv
- * It makes sense only if this program is linked with pthread and
- * multithreaded BLAS and LAPACK */
- if ( iparam[IPARAM_THRDNBR] == -1 ) {
- get_thread_count( &(iparam[IPARAM_THRDNBR]) );
- }
- NCPU = iparam[IPARAM_THRDNBR];
- NGPU = 0;
-
- /* print informations to user */
- print_header( argv[0], iparam);
-
- /* Initialize CHAMELEON with main parameters */
- CHAMELEON_Init( NCPU, NGPU );
-
- /*
- * allocate memory for our data using a C macro (see step1.h)
- * - matrix A : size N x N
- * - set of RHS vectors B : size N x NRHS
- * - set of solutions vectors X : size N x NRHS
- */
- double *A = malloc( N * N * sizeof(double) );
- double *Acpy = malloc( N * N * sizeof(double) );
- double *B = malloc( N * NRHS * sizeof(double) );
- double *X = malloc( N * NRHS * sizeof(double) );
-
- /* generate A matrix with random values such that it is spd */
- CHAMELEON_dplgsy( (double)N, ChamUpperLower, N, A, N, 51 );
-
- /* generate RHS */
- CHAMELEON_dplrnt( N, NRHS, B, N, 5673 );
-
- /* copy A before facto. in order to check the result */
- memcpy(Acpy, A, N*N*sizeof(double));
-
- /* copy B in X before solving */
- memcpy(X, B, N*NRHS*sizeof(double));
-
- /************************************************************/
- /* solve the system AX = B using the Cholesky factorization */
- /************************************************************/
-
- cpu_time = -CHAMELEON_timer();
-
- /* Cholesky facorization:
- * A is replaced by its factorization L or L^T depending on uplo */
- CHAMELEON_dpotrf( UPLO, N, A, N );
-
- /* Solve:
- * B is stored in X on entry, X contains the result on exit.
- * Forward and back substitutions
- */
- CHAMELEON_dpotrs(UPLO, N, NRHS, A, N, X, N);
-
- cpu_time += CHAMELEON_timer();
-
- /* print informations to user */
- gflops = flops / cpu_time;
- printf( "%9.3f %9.2f\n", cpu_time, gflops);
- fflush( stdout );
-
- /************************************************************/
- /* check if solve is correct i.e. AX-B = 0 */
- /************************************************************/
-
- /* compute norms to check the result */
- anorm = CHAMELEON_dlange( ChamInfNorm, N, N, Acpy, N);
- bnorm = CHAMELEON_dlange( ChamInfNorm, N, NRHS, B, N);
- xnorm = CHAMELEON_dlange( ChamInfNorm, N, NRHS, X, N);
-
- /* compute A*X-B, store the result in B */
- CHAMELEON_dgemm(ChamNoTrans, ChamNoTrans,
- N, NRHS, N, 1.0, Acpy, N, X, N, -1.0, B, N);
- res = CHAMELEON_dlange( ChamInfNorm, N, NRHS, B, N);
-
- /* check residual and print a message */
- eps = LAPACKE_dlamch_work( 'e' );
-
- /*
- * if hres = 0 then the test succeed
- * else the test failed
- */
- hres = 0;
- hres = ( res / N / eps / (anorm * xnorm + bnorm ) > 100.0 );
- printf( " ||Ax-b|| ||A|| ||x|| ||b|| ||Ax-b||/N/eps/(||A||||x||+||b||) RETURN\n");
- if (hres)
- printf( "%8.5e %8.5e %8.5e %8.5e %8.5e FAILURE \n",
- res, anorm, xnorm, bnorm,
- res / N / eps / (anorm * xnorm + bnorm ));
- else
- printf( "%8.5e %8.5e %8.5e %8.5e %8.5e SUCCESS \n",
- res, anorm, xnorm, bnorm,
- res / N / eps / (anorm * xnorm + bnorm ));
-
- /* deallocate data */
- free(A);
- free(Acpy);
- free(B);
- free(X);
+int main() {
+ /* number of threads */
+ int NT = 2;
+ /* number of gpus */
+ int NG = 0;
+ /* size of the matrices */
+ int N = 16;
+ /* size of the tiles */
+ int NB = 2;
+ /* seeds for A and B initialization */
+ int seedA = 33;
+ int seedB = 47;
+
+ CHAMELEON_Init( NT, NG );
+ CHAMELEON_Set( CHAMELEON_TILE_SIZE, NB );
+ int NMPI = CHAMELEON_Comm_size();
+ int GRID_P = NMPI;
+ int GRID_Q = 1;
+
+ CHAM_desc_t *descA = NULL, *descB = NULL, *descC = NULL;
+ CHAMELEON_Desc_Create(&descA, NULL, ChamRealFloat,
+ NB, NB, NB*NB, N, N, 0, 0, N, N, GRID_P, GRID_Q);
+ CHAMELEON_Desc_Create(&descB, NULL, ChamRealFloat,
+ NB, NB, NB*NB, N, N, 0, 0, N, N, GRID_P, GRID_Q);
+ CHAMELEON_Desc_Create(&descC, NULL, ChamRealFloat,
+ NB, NB, NB*NB, N, N, 0, 0, N, N, GRID_P, GRID_Q);
+
+ CHAMELEON_splrnt_Tile( descA, 33 );
+ CHAMELEON_splrnt_Tile( descB, 47 );
+ CHAMELEON_sgemm_Tile(ChamNoTrans, ChamNoTrans, 1.0, descA, descB, 0.0, descC);
+
+ CHAMELEON_Desc_Destroy( &descA );
+ CHAMELEON_Desc_Destroy( &descB );
+ CHAMELEON_Desc_Destroy( &descC );
- /* Finalize CHAMELEON */
CHAMELEON_Finalize();
return EXIT_SUCCESS;
diff --git a/gpuhipblas/compute/CMakeLists.txt b/gpuhipblas/compute/CMakeLists.txt
index abaa7b40a6df131c6349f0d75c96a5c9304009ce..3c6349e809d97fbdc117798ac58e91826047fa72 100644
--- a/gpuhipblas/compute/CMakeLists.txt
+++ b/gpuhipblas/compute/CMakeLists.txt
@@ -70,7 +70,8 @@ target_include_directories(gpuhipblas PUBLIC
$<INSTALL_INTERFACE:include>)
set_property(TARGET gpuhipblas PROPERTY INSTALL_NAME_DIR "${CMAKE_INSTALL_PREFIX}/lib")
-target_link_libraries(gpuhipblas PRIVATE coreblas HIP::HIPBLAS)
+target_link_libraries(gpuhipblas PRIVATE coreblas)
+target_link_libraries(gpuhipblas PUBLIC HIP::HIPBLAS)
target_link_libraries(gpuhipblas PUBLIC MORSE::M)
# export target coreblas