Merge branch 'master' into fix1150

.github/workflows/fpga-ci.yml

@@ -20,9 +20,11 @@ jobs:
         rm -rf .dacecache tests/.dacecache
         . /opt/setupenv
         python -m pip install --upgrade pip
-        pip install pytest-xdist flake8 coverage codecov
+        pip install pytest-xdist flake8 coverage
         pip uninstall -y dace
         pip install -e ".[testing]"
+        curl -Os https://uploader.codecov.io/latest/linux/codecov
+        chmod +x codecov
 
     - name: Run FPGA Tests
       run: |
@@ -35,7 +37,7 @@ jobs:
         reachable=0
         ping -W 2 -c 1 codecov.io || reachable=$?
         if [ $reachable -eq 0 ]; then
-          codecov
+          ./codecov
         else
           echo "Codecov.io is unreachable"
         fi

.github/workflows/general-ci.yml

@@ -12,7 +12,7 @@ jobs:
     runs-on: ubuntu-latest
     strategy:
       matrix:
-        python-version: [3.7,'3.10']
+        python-version: [3.7,'3.11']
         simplify: [0,1,autoopt]
 
     steps:
@@ -31,8 +31,10 @@ jobs:
         sudo apt-get install -y libpapi-dev papi-tools  # Instrumentation dependencies
         sudo apt-get install -y verilator # RTL simulation dependencies
         python -m pip install --upgrade pip
-        pip install flake8 pytest-xdist coverage codecov
+        pip install flake8 pytest-xdist coverage
         pip install -e ".[testing]"
+        curl -Os https://uploader.codecov.io/latest/linux/codecov
+        chmod +x codecov
 
     - name: Test dependencies
       run: |
@@ -52,7 +54,7 @@ jobs:
             export DACE_optimizer_automatic_simplification=${{ matrix.simplify }}
         fi
         pytest -n auto --cov-report=xml --cov=dace --tb=short -m "not gpu and not verilator and not tensorflow and not mkl and not sve and not papi and not mlir and not lapack and not fpga and not mpi and not rtl_hardware and not scalapack and not datainstrument"
-        codecov
+        ./codecov
 
     - name: Test OpenBLAS LAPACK
       run: |
@@ -68,17 +70,17 @@ jobs:
             export DACE_optimizer_automatic_simplification=${{ matrix.simplify }}
         fi
         pytest -n 1 --cov-report=xml --cov=dace --tb=short -m "lapack"
-        codecov
+        ./codecov
 
     - name: Run other tests
       run: |
         export NOSTATUSBAR=1
-        export DACE_testing_serialization=1
+        export DACE_testing_serialization=0
         export DACE_testing_deserialize_exception=1
         export DACE_cache=single
         export DACE_optimizer_automatic_simplification=${{ matrix.simplify }}
         export PYTHON_BINARY="coverage run --source=dace --parallel-mode"
         ./tests/polybench_test.sh
         ./tests/xform_test.sh
         coverage combine .; coverage report; coverage xml
-        codecov
+        ./codecov

.github/workflows/gpu-ci.yml

@@ -20,10 +20,12 @@ jobs:
         rm -rf .dacecache tests/.dacecache
         . /opt/setupenv
         python -m pip install --upgrade pip
-        pip install flake8 pytest-xdist coverage codecov
+        pip install flake8 pytest-xdist coverage
         pip install mpi4py
         pip uninstall -y dace
         pip install -e ".[testing]"
+        curl -Os https://uploader.codecov.io/latest/linux/codecov
+        chmod +x codecov
 
     - name: Test dependencies
       run: |
@@ -52,7 +54,7 @@ jobs:
         reachable=0
         ping -W 2 -c 1 codecov.io || reachable=$?
         if [ $reachable -eq 0 ]; then
-          codecov
+          ./codecov
         else
           echo "Codecov.io is unreachable"
         fi

.github/workflows/heterogeneous-ci.yml

@@ -20,10 +20,12 @@ jobs:
         rm -rf .dacecache tests/.dacecache
         . /opt/setupenv
         python -m pip install --upgrade pip
-        pip install flake8 pytest-xdist coverage codecov
+        pip install flake8 pytest-xdist coverage
         pip install mpi4py pytest-mpi
         pip uninstall -y dace
         pip install -e ".[testing]"
+        curl -Os https://uploader.codecov.io/latest/linux/codecov
+        chmod +x codecov
 
     - name: Test dependencies
       run: |
@@ -73,7 +75,7 @@ jobs:
         reachable=0
         ping -W 2 -c 1 codecov.io || reachable=$?
         if [ $reachable -eq 0 ]; then
-          codecov
+          ./codecov
         else
           echo "Codecov.io is unreachable"
         fi

AUTHORS

@@ -32,6 +32,10 @@ Neville Walo
 Lukas Trümper
 Cliff Hodel
 Tiancheng Chen
+Reid Wahl
 Yihang Luo
+Alexandru Calotoiu
+Phillip Lane
+Samuel Martin
 
 and other contributors listed in https://github.com/spcl/dace/graphs/contributors

dace/cli/daceprof.py

@@ -227,7 +227,8 @@ def make_sequential(sdfg: dace.SDFG):
             for n, _ in sdfg.all_nodes_recursive():
                 if isinstance(n, dace.nodes.EntryNode):
                     sched = getattr(n, 'schedule', False)
-                    if sched == dace.ScheduleType.CPU_Multicore or sched == dace.ScheduleType.Default:
+                    if sched in (dace.ScheduleType.CPU_Multicore, dace.ScheduleType.CPU_Persistent,
+                                 dace.ScheduleType.Default):
                         n.schedule = dace.ScheduleType.Sequential
 
         registered.append(dace.hooks.register_sdfg_call_hook(before_hook=make_sequential))

dace/cli/fcdc.py

@@ -0,0 +1,62 @@
+# Copyright 2019-2023 ETH Zurich and the DaCe authors. All rights reserved.
+""" Simple Fortran SDFG command-line compiler. """
+
+import dace
+import os
+import sys
+import argparse
+import shutil
+from dace.frontend.fortran import fortran_parser
+
+def main():
+    # Command line options parser
+    parser = argparse.ArgumentParser(description='Fortran to SDFG command-line transpiler.')
+
+    # Required argument for Fortran file path
+    parser.add_argument('filepath', help='<PATH TO FORTRAN FILE>', type=str)
+
+    # Optional argument for output location
+    parser.add_argument('-o',
+                        '--out',
+                        type=str,
+                        help='If provided, saves library as the given file or in the specified path, '
+                        'together with a header file.')
+
+    parser.add_argument('-O',
+                        '--optimize',
+                        dest='optimize',
+                        action='store_true',
+                        help="If set, invokes the command-line optimization"
+                        " interface",
+                        default=False)
+
+    args = parser.parse_args()
+
+    filepath = args.filepath
+    if not os.path.isfile(filepath):
+        print('Fortran file', filepath, 'not found')
+        exit(1)
+
+    outpath = args.out
+
+    # Load SDFG
+    sdfg = fortran_parser.create_sdfg_from_fortran_file(filepath)
+
+    if args.optimize:
+        sdfg.optimize()
+
+    # Compile SDFG
+    sdfg.compile(outpath)
+
+    # Copying header file to optional path
+    if outpath is not None:
+        source = os.path.join(sdfg.build_folder, 'include', sdfg.name + '.h')
+        if os.path.isdir(outpath):
+            outpath = os.path.join(outpath, sdfg.name + '.h')
+        else:
+            outpath = os.path.join(os.path.dirname(outpath), sdfg.name + '.h')
+        shutil.copyfile(source, outpath)
+
+
+if __name__ == '__main__':
+    main()

dace/codegen/codegen.py

@@ -31,7 +31,7 @@ def generate_headers(sdfg: SDFG, frame: framecode.DaCeCodeGenerator) -> str:
     exit_params = (sdfg.name, sdfg.name)
     proto += 'typedef void * %sHandle_t;\n' % sdfg.name
     proto += 'extern "C" %sHandle_t __dace_init_%s(%s);\n' % init_params
-    proto += 'extern "C" void __dace_exit_%s(%sHandle_t handle);\n' % exit_params
+    proto += 'extern "C" int __dace_exit_%s(%sHandle_t handle);\n' % exit_params
     proto += 'extern "C" void __program_%s(%sHandle_t handle%s);\n' % params
     return proto
 
@@ -69,15 +69,16 @@ def generate_dummy(sdfg: SDFG, frame: framecode.DaCeCodeGenerator) -> str:
 
 int main(int argc, char **argv) {{
     {sdfg.name}Handle_t handle;
+    int err;
 {allocations}
 
     handle = __dace_init_{sdfg.name}({init_params});
     __program_{sdfg.name}(handle{params});
-    __dace_exit_{sdfg.name}(handle);
+    err = __dace_exit_{sdfg.name}(handle);
 
 {deallocations}
 
-    return 0;
+    return err;
 }}
 '''
 
@@ -177,11 +178,6 @@ def generate_code(sdfg, validate=True) -> List[CodeObject]:
                 shutil.move(f"{tmp_dir}/test2.sdfg", "test2.sdfg")
                 raise RuntimeError('SDFG serialization failed - files do not match')
 
-        # Run with the deserialized version
-        # NOTE: This means that all subsequent modifications to `sdfg`
-        # are not reflected outside of this function (e.g., library
-        # node expansion).
-        sdfg = sdfg2
 
     # Before generating the code, run type inference on the SDFG connectors
     infer_types.infer_connector_types(sdfg)

dace/codegen/common.py

@@ -1,11 +1,12 @@
-# Copyright 2019-2022 ETH Zurich and the DaCe authors. All rights reserved.
+# Copyright 2019-2023 ETH Zurich and the DaCe authors. All rights reserved.
 import ast
 from copy import deepcopy
 import ctypes.util
 from dace import config, data, dtypes, sdfg as sd, symbolic
 from dace.sdfg import SDFG
 from dace.properties import CodeBlock
 from dace.codegen import cppunparse
+from dace.codegen.tools import gpu_runtime
 from functools import lru_cache
 from io import StringIO
 import os
@@ -73,7 +74,7 @@ def update_persistent_desc(desc: data.Data, sdfg: SDFG):
     Replaces the symbols used in a persistent data descriptor according to NestedSDFG's symbol mapping.
     The replacement happens recursively up to the top-level SDFG.
     """
-    if (desc.lifetime == dtypes.AllocationLifetime.Persistent and sdfg.parent
+    if (desc.lifetime in (dtypes.AllocationLifetime.Persistent, dtypes.AllocationLifetime.External) and sdfg.parent
             and any(str(s) in sdfg.parent_nsdfg_node.symbol_mapping for s in desc.free_symbols)):
         newdesc = deepcopy(desc)
         csdfg = sdfg
@@ -144,3 +145,29 @@ def _try_execute(cmd: str) -> bool:
                        'set the DaCe configuration entry ``compiler.cuda.backend`` '
                        'or the ``DACE_compiler_cuda_backend`` environment variable '
                        'to either "cuda" or "hip".')
+
+
+@lru_cache()
+def get_gpu_runtime() -> gpu_runtime.GPURuntime:
+    """
+    Returns the GPU runtime library (CUDA / HIP) if exists. The result is cached for performance.
+    """
+    backend = get_gpu_backend()
+    if backend == 'cuda':
+        libpath = ctypes.util.find_library('cudart')
+        if os.name == 'nt' and not libpath:  # Windows-based search
+            for version in (12, 11, 10, 9):
+                libpath = ctypes.util.find_library(f'cudart64_{version}0')
+                if libpath:
+                    break
+    elif backend == 'hip':
+        libpath = ctypes.util.find_library('amdhip64')
+    else:
+        raise RuntimeError(f'Cannot obtain GPU runtime library for backend {backend}')
+
+    if not libpath:
+        envname = 'PATH' if os.name == 'nt' else 'LD_LIBRARY_PATH'
+        raise RuntimeError(f'GPU runtime library for {backend} not found. Please set the {envname} '
+                           'environment variable to point to the libraries.')
+
+    return gpu_runtime.GPURuntime(backend, libpath)