Commit b706f13d authored by Giovanni La Mura's avatar Giovanni La Mura
Browse files

Merge branch 'master' into script_devel to update to offload trapping

parents b8036aa5 3c2b3d5e

.gitlab-ci.yml

+3 −3
Original line number Diff line number Diff line
@@ -78,13 +78,13 @@ compatibility_stage: 
      - CXX=g++-14 FC=gfortran-14 ./configure

      - make wipe

      - make -j

      - echo "Running make with refinement with gnu compilers version 14..."

      - echo "Running make with gnu compilers version 14..."

      - cd ..

      - rm -rf build_gnu14

      - mkdir build_gnu14_refine

      - cd build_gnu14_refine

      - cp -r ../build/* .

      - CXX=g++-14 FC=gfortran-14 ./configure --enable-refinement

      - CXX=g++-14 FC=gfortran-14 ./configure

      - make wipe

      - make -j

      #- echo "Running make with flang version 16 and clang version 16..."
@@ -173,7 +173,7 @@ building_stage: 
      - cat /etc/os-release

      - cd build

      - echo "Configuring with default compilers (MAGMA disabled)..."

      - ./configure --without-magma --without-cublas --disable-offload --enable-refinement --enable-shared

      - ./configure --without-magma --without-cublas --disable-offload --enable-shared

      - make wipe

      - echo "Building the default configuration..."

      - make -j

COPYING

+2 −0
Original line number Diff line number Diff line
@@ -617,3 +617,5 @@ reviewing courts shall apply local law that most closely approximates 
an absolute waiver of all civil liability in connection with the

Program, unless a warranty or assumption of liability accompanies a

copy of the Program in return for a fee.

build/configure.sh

+23 −27
Original line number Diff line number Diff line
@@ -21,21 +21,20 @@ NVTXFLAGS="" 
OMPMODE="auto"

OFFLOAD="auto"

OFFLOADFLAGS=""

REFINEFLAGS=""

# End of default configuration settings



# Function declarations

function guess_cxx {

    # Guess the name of the C++ compiler

    result=$(which mpicxx)

    result=$(which mpicxx 2>/dev/null)

    if [ "x$result" = "x" ]; then

	result=$(which g++)

	result=$(which g++ 2>/dev/null)

    fi

    if [ "x$result" = "x" ]; then

	result=$(which clang)

	result=$(which clang 2>/dev/null)

    fi

    if [ "x$result" = "x" ]; then

	result=$(which icpx)

	result=$(which icpx 2>/dev/null)

    fi

    if [ "x$result" = "x" ]; then

	result="none"
@@ -45,21 +44,21 @@ function guess_cxx { 


function guess_fc {

    # Guess the name of the FORTRAN compiler

    result=$(which mpif90)

    result=$(which mpif90 2>/dev/null)

    if [ "x$result" = "x" ]; then

	result=$(which gfortran)

	result=$(which gfortran 2>/dev/null)

    fi

    if [ "x$result" = "x" ]; then

	result=$(which f77)

	result=$(which f77 2>/dev/null)

    fi

    if [ "x$result" = "x" ]; then

	result=$(which flang)

	result=$(which flang 2>/dev/null)

    fi

    if [ "x$result" = "x" ]; then

	result=$(which flang-new)

	result=$(which flang-new 2>/dev/null)

    fi

    if [ "x$result" = "x" ]; then

	result=$(which ifx)

	result=$(which ifx 2>/dev/null)

    fi

    if [ "x$result" = "x" ]; then

	result="none"
@@ -87,9 +86,6 @@ function print_help { 
    echo "--enable-nvtx                     Enable NVTX profiling tools.                   "

    echo "--disable-nvtx                    Disable NVTX profiling tools (DEFAULT).        "

    echo "--enable-optimize=OPT             Use OPT-level compiler optimization.           "

    echo "--enable-refinement               Use iterative refinement of matrix inversion.  "

    echo "--disable-refinement              Do not iterate refinement of matrix inversion  "

    echo "                                  (DEFAULT).                                     "

    echo "--enable-shared                   Use shared libraries (default is static).      "

    echo "--disable-shared                  Use static libraries (DEFAULT).                "

    echo "--help                            Print this help and exit.                      "
@@ -198,10 +194,6 @@ do 
	fi

	FC_OPT=$opt_level

	CXX_OPT=$opt_level

    elif [ "x$cut_arg" = "x--enable-refinement" ]; then

	REFINEFLAGS=" -DUSE_REFINEMENT"

    elif [ "x$cut_arg" = "x--disable-refinement" ]; then

	REFINEFLAGS=""

    elif [ "x$cut_arg" = "x--enable-shared" ]; then

	LIBMODE="shared"

    elif [ "x$cut_arg" = "x--disable-shared" ]; then
@@ -634,6 +626,11 @@ if [ "x$CUBLAS" != "xno" ]; then 
		cuda_pkg=$(for i in "${pkg_array[@]}"; do echo "$i" | cut --delimiter=" " -f1; done | grep cublas)

    		CUDAFLAGS=$(pkg-config --cflags ${cuda_pkg})

    		CUDALDFLAGS=$(pkg-config --libs ${cuda_pkg})

	    else

		# CUBLAS not detected

		CUBLAS="no"

		CUDAFLAGS=""

		CUDALDFLAGS=""

            fi # end of CUBLAS runtime decision tree

            declare -a pkg_array=$(pkg-config --list-all | grep cudart)

            for i in "${pkg_array[@]}"; do echo "$i" | cut --delimiter=" " -f1; done | grep cudart > /dev/null
@@ -647,13 +644,17 @@ if [ "x$CUBLAS" != "xno" ]; then 
            echo $CUDALDFLAGS | grep cublas > /dev/null

            cudart_check=$?

            if [ "x${cudart_check}" != "x0" ]; then

		if [ "x${CUBLAS}" != "xno" ]; then

		    CUDALDFLAGS="$CUDALDFLAGS -lcublas"

		fi

            fi

            echo $CUDALDFLAGS | grep cudart > /dev/null

            cudart_check=$?

            if [ "x${cudart_check}" != "x0" ]; then

		if [ "x${CUBLAS}" != "xno" ]; then

		    CUDALDFLAGS="$CUDALDFLAGS -lcudart"

		fi

            fi

	else

            # pkg-config is not available

            if [ -f /usr/local/cuda/include/cuda.h ]; then
@@ -813,7 +814,7 @@ EOF 
    fi

    if [ "x$result" = "x0" ]; then

	echo "yes"

	OFFLOADFLAGS=" -fcf-protection=check -foffload=nvptx-none=\"-O${CXX_OPT}${CXX_DBG} -fcf-protection=check -fopt-info -lm -latomic -gomp\""

	OFFLOADFLAGS=" -DUSE_TARGET_OFFLOAD -fcf-protection=check -foffload=nvptx-none=\"-O${CXX_OPT}${CXX_DBG} -fcf-protection=check -fopt-info -lm -latomic -lgomp\""

	if [ "x${OMPFLAGS}" = "x" ]; then

	    OFFLOADFLAGS="${OFFLOADFLAGS} -fopenmp"

	fi
@@ -826,7 +827,7 @@ else 
fi

# End of offload checks

if [ "x$CXXFLAGS" = "x" ]; then

    CXXFLAGS="-O${CXX_OPT}${CXX_DBG}${CLANGFLAGS}${INCLUDEFLAGS}${HDF5FLAGS}${OMPFLAGS}${MPIFLAGS}${LAPACKFLAGS}${CUBLASFLAGS}${MAGMAFLAGS}${REFINEFLAGS}${DEBUGFLAGS}${OFFLOADFLAGS}${NVTXFLAGS}"

    CXXFLAGS="-O${CXX_OPT}${CXX_DBG}${CLANGFLAGS}${INCLUDEFLAGS}${HDF5FLAGS}${OMPFLAGS}${MPIFLAGS}${LAPACKFLAGS}${CUBLASFLAGS}${MAGMAFLAGS}${DEBUGFLAGS}${OFFLOADFLAGS}${NVTXFLAGS}"

fi

if [ "x$CXXLDFLAGS" = "x" ]; then

    if [ "x$LIBMODE" = "xstatic" ]; then
@@ -885,11 +886,6 @@ if [ "x${NVTXFLAGS}" = "x" ]; then 
else

    echo "INFO: NVTX profiling is enabled."

fi

if [ "x${REFINEFLAGS}" = "x" ]; then

    echo "INFO: iterative matrix inversion refinement is disabled."

else

    echo "INFO: iterative matrix inversion refinement is enabled."

fi

if [ "x${OFFLOADFLAGS}" = "x" ]; then

    echo "INFO: GPU offload through OpenMP is disabled."

else

src/cluster/cluster.cpp

+37 −21
Original line number Diff line number Diff line 
/* Copyright (C) 2024   INAF - Osservatorio Astronomico di Cagliari

/* Copyright (C) 2025   INAF - Osservatorio Astronomico di Cagliari



   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
@@ -757,9 +757,15 @@ int cluster_jxi488_cycle( 
	+ ").\n";

      logger->log(message, LOG_WARN);

    } else if (recommended_li < cid->c1->li) {

      if (gconf->dyn_order_flag > 0) {

	message = "INFO: lowering internal order from " + to_string(cid->c1->li) + " to "

	  + to_string(recommended_li) + " for scale iteration " + to_string(jxi488) + ".\n";

	logger->log(message, LOG_INFO);

      } else {

	message = "WARNING: internal order " + to_string(cid->c1->li) + " too high for scale iteration "

	  + to_string(jxi488) + ".\n";

	logger->log(message, LOG_WARN);

      }

    }

    if (recommended_le > cid->c1->le) {

      message = "WARNING: external order " + to_string(cid->c1->le) + " for scale iteration "
@@ -767,11 +773,18 @@ int cluster_jxi488_cycle( 
	+ ").\n";

      logger->log(message, LOG_WARN);

    } else if (recommended_le < cid->c1->le) {

      if (gconf->dyn_order_flag > 0) {

	message = "INFO: lowering external order from " + to_string(cid->c1->le) + " to "

	  + to_string(recommended_le) + " for scale iteration " + to_string(jxi488) + ".\n";

	logger->log(message, LOG_INFO);

      } else {

	message = "INFO: external order " + to_string(cid->c1->le) + " too high for scale iteration "

	  + to_string(jxi488) + ".\n";

	logger->log(message, LOG_WARN);

      }

    }

    if (recommended_li < max_li || recommended_le < max_le) {

      if (gconf->dyn_order_flag > 0) {

	int new_li = (recommended_li < max_li) ? recommended_li : max_li;

	int new_le = (recommended_le < max_le) ? recommended_le : max_le;

	cid->update_orders(sconf->_rcf, new_li, new_le);
@@ -780,6 +793,7 @@ int cluster_jxi488_cycle( 
	cid->refinemode = 2;

      }

    }

  }

  int li = cid->c1->li;

  int le = cid->c1->le;

  int lm = cid->c1->lm;
@@ -883,8 +897,10 @@ int cluster_jxi488_cycle( 
  double actualaccuracy = cid->accuracygoal;

  invert_matrix(cid->am, ndit, jer, cid->maxrefiters, actualaccuracy, cid->refinemode, output_path, jxi488, mxndm, cid->proc_device);

  // in principle, we should check whether the returned actualaccuracy is indeed lower than the accuracygoal, and do something about it if not

#ifdef USE_REFINEMENT

  if (gconf->refine_flag > 0) {

    if (cid->refinemode==2) {

      message = "DEBUG: iterative refinement enabled at run-time.\n";

      logger->log(message, LOG_DEBG);

      message = "INFO: calibration obtained accuracy " + to_string(actualaccuracy) + " (" + to_string(cid->accuracygoal) + " requested) in " + to_string(cid->maxrefiters) + " refinement iterations\n";

      logger->log(message);

      if (actualaccuracy > 1e-1) {
@@ -892,7 +908,7 @@ int cluster_jxi488_cycle( 
	exit(1);

      }

    }

#endif

  }

  cid->refinemode = 0;

#ifdef DEBUG_AM

  VirtualAsciiFile *outam2 = new VirtualAsciiFile();

src/cluster/np_cluster.cpp

+1 −1
Original line number Diff line number Diff line 
/* Copyright (C) 2024   INAF - Osservatorio Astronomico di Cagliari

/* Copyright (C) 2025   INAF - Osservatorio Astronomico di Cagliari



   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