mfc.sh: refactor running cases from within mfc.sh

Author: henryleberre

misc/run-phoenix-release-gpu.sh

@@ -17,7 +17,7 @@ set -x
 . ./mfc.sh load -c p -m GPU
 
 gpu_count=$(nvidia-smi -L | wc -l)        # number of GPUs on node
-gpu_ids=$(seq -s ',' 0 $(($gpu_count-1))) # 0,1,2,...,gpu_count-1
+gpu_ids=$(seq -s ' ' 0 $(($gpu_count-1))) # 0,1,2,...,gpu_count-1
 
 ./mfc.sh test -a -b mpirun -j $(nproc) \
               --gpu -g $gpu_ids

toolchain/mfc/args.py

@@ -65,11 +65,14 @@ def add_common_arguments(p, mask = None):
             for target in DEPENDENCY_TARGETS:
                 p.add_argument(f"--no-{target.name}", action="store_true", help=f"Do not build the {target.name} dependency. Use the system's instead.")
 
+        if "g" not in mask:
+            p.add_argument("-g", "--gpus", nargs="+", type=int, default=[0], help="(GPU) List of GPU #s to use.")
+
     # === BUILD ===
-    add_common_arguments(build)
+    add_common_arguments(build, "g")
 
     # === CLEAN ===
-    add_common_arguments(clean, "j")
+    add_common_arguments(clean, "jg")
 
     binaries = [ b.bin for b in BINARIES ]
 
@@ -82,7 +85,6 @@ def add_common_arguments(p, mask = None):
     test.add_argument("-b", "--binary",       choices=binaries, type=str, default=None, help="(Serial) Override MPI execution binary")
     test.add_argument("-r", "--relentless",   action="store_true", default=False, help="Run all tests, even if multiple fail.")
     test.add_argument("-a", "--test-all",     action="store_true", default=False, help="Run the Post Process Tests too.")
-    test.add_argument("-g", "--gpus",         type=str, default="0", help="(GPU) Comma separated list of GPU #s to use.")
     test.add_argument("-%", "--percent",      type=int, default=100, help="Percentage of tests to run.")
     test.add_argument("-m", "--max-attempts", type=int, default=3, help="Maximum number of attempts to run a test.")
 
@@ -120,7 +122,7 @@ def add_common_arguments(p, mask = None):
     add_common_arguments(bench, "t")
 
     # === COUNT ===
-    add_common_arguments(count)
+    add_common_arguments(count, "g")
 
     args: dict = vars(parser.parse_args())
 
@@ -150,8 +152,4 @@ def append_defaults_to_data(name: str, parser):
         if args[e] is not None:
             args[e] = os.path.abspath(args[e])
 
-    # Turn GPU ID list into a comma separated string
-    if "gpus" in args:
-        args["gpus"] = [int(g) for g in args["gpus"].split(",")]
-
     return args

toolchain/mfc/common.py

@@ -29,9 +29,12 @@ class MFCException(Exception):
     pass
 
 
-def system(command: typing.List[str], no_exception: bool = False, exception_text=None, on_error=lambda: None, cwd=None, stdout=None, stderr=None) -> int:
+def system(command: typing.List[str], no_exception: bool = False, exception_text=None, on_error=lambda: None, cwd=None, stdout=None, stderr=None, env: dict = None) -> int:
     cmd = [ str(x) for x in command if not isspace(str(x)) ]
 
+    if env is None:
+        env = os.environ.copy()
+
     if stdout != subprocess.DEVNULL:
         cons.print(no_indent=True)
 
@@ -40,7 +43,7 @@ def system(command: typing.List[str], no_exception: bool = False, exception_text
     if stdout != subprocess.DEVNULL:
         cons.print(no_indent=True)
 
-    r = subprocess.run(cmd, cwd=cwd, stdout=stdout, stderr=stderr)
+    r = subprocess.run(cmd, cwd=cwd, stdout=stdout, stderr=stderr, env=env)
 
     if r.returncode != 0:
         on_error()

toolchain/mfc/count.py

@@ -1,6 +1,7 @@
 import os, glob, typing, typing
 
-from .common  import MFC_ROOTDIR
+from .state   import ARG
+from .common  import MFC_ROOTDIR, format_list_to_string
 from .printer import cons
 
 import rich.table
@@ -20,12 +21,13 @@ def handle_dir(dirpath: str) -> typing.Tuple[typing.List[typing.Tuple[str, int]]
     return (files, total)
 
 def count():
-    cons.print("[bold]Counting lines of code in [magenta]MFC[/magenta][/bold] (excluding whitespace lines)")
-    cons.print()
+    target_str_list = format_list_to_string(ARG('targets'), 'magenta')
+
+    cons.print(f"[bold]Counting lines of code in {target_str_list}[/bold] (excluding whitespace lines)")
     cons.indent()
 
     total = 0
-    for codedir in ['common', 'pre_process', 'simulation', 'post_process']:
+    for codedir in ['common'] + ARG("targets"):
         dirfiles, dircount = handle_dir(os.path.join(MFC_ROOTDIR, 'src', codedir))        
         table = rich.table.Table(show_header=True, box=rich.table.box.SIMPLE)
         table.add_column(f"File (in [magenta]{codedir}[/magenta])", justify="left")
@@ -38,7 +40,7 @@ def count():
 
         cons.raw.print(table)
 
-    cons.print(f"[bold]Total MFC lines: [bold cyan]{total}[/bold cyan].[/bold]")
+    cons.print(f"[bold]Total {target_str_list} lines: [bold cyan]{total}[/bold cyan].[/bold]")
     cons.print()
     cons.unindent()
 

toolchain/mfc/run/engines.py

@@ -74,7 +74,6 @@ def get_args(self) -> str:
 
     def get_exec_cmd(self, target: MFCTarget) -> typing.List[str]:
         cmd = []
-
         if ARG("mpi"):
             cmd += [self.mpibin.bin] + self.mpibin.gen_params() + ARG("flags")[:]
 
@@ -148,7 +147,13 @@ def run(self, targets: typing.List[MFCTarget]) -> None:
 
             if not ARG("dry_run"):
                 start_time = time.monotonic()
-                system(self.get_exec_cmd(target), cwd=self.input.case_dirpath)
+                system(
+                    self.get_exec_cmd(target), cwd=self.input.case_dirpath, 
+                    env={
+                        **os.environ.copy(),
+                        'CUDA_VISIBLE_DEVICES': ','.join([str(_) for _ in ARG('gpus')])
+                    }
+                )
                 end_time   = time.monotonic()
                 cons.print(no_indent=True)
 

toolchain/mfc/run/run.py

@@ -67,3 +67,12 @@ def run_target(target: MFCTarget):
 
 def run() -> None:
     run_targets([ get_target(_) for _ in ARG("targets")])
+
+
+def run_targets_with(targets: typing.List[MFCTarget]):
+    pass
+
+
+def run_target_with(target: MFCTarget):
+    run_targets_with([target])
+

toolchain/mfc/test/case.py

@@ -78,7 +78,6 @@
     'sigV'                          : 0.1,
     'rhoRV'                         : 0.0,
 
-
     'Monopole'                      : 'F',
     'num_mono'                      : 1,
     'Mono(1)%loc(1)'                : 0.5,