Merge pull request #31227 from loganharbour/31135_remove_devices

Remove `--libtorch-device` and `--device`

Author: loganharbour

framework/include/base/MooseApp.h

@@ -1519,6 +1519,12 @@ class MooseApp : public ConsoleStreamInterface,
    */
   bool runInputs();
 
+  /**
+   * Helper that reports an error if the given capability is reserved and
+   * should not be added via addCapability().
+   */
+  static void checkReservedCapability(const std::string & capability);
+
   /// General storage for custom RestartableData that can be added to from outside applications
   std::unordered_map<RestartableDataMapName, std::pair<RestartableDataMap, std::string>>
       _restartable_meta_data;

framework/src/base/MooseApp.C

@@ -368,11 +368,7 @@ MooseApp::validParams()
   params.addParam<bool>(
       "automatic_automatic_scaling", false, "Whether to turn on automatic scaling by default");
 
-  MooseEnum compute_device_type("cpu cuda mps hip ceed-cpu ceed-cuda ceed-hip", "cpu");
-  params.addCommandLineParam<MooseEnum>("libtorch_device",
-                                        "--libtorch-device",
-                                        compute_device_type,
-                                        "Deprecated. Use --compute-device.");
+  const MooseEnum compute_device_type("cpu cuda mps hip ceed-cpu ceed-cuda ceed-hip", "cpu");
   params.addCommandLineParam<MooseEnum>(
       "compute_device",
       "--compute-device",
@@ -517,9 +513,7 @@ MooseApp::MooseApp(const InputParameters & parameters)
     _initial_backup(getParam<std::unique_ptr<Backup> *>("_initial_backup"))
 #ifdef MOOSE_LIBTORCH_ENABLED
     ,
-    _libtorch_device(determineLibtorchDeviceType(isParamSetByUser("libtorch_device")
-                                                     ? getParam<MooseEnum>("libtorch_device")
-                                                     : getParam<MooseEnum>("compute_device")))
+    _libtorch_device(determineLibtorchDeviceType(getParam<MooseEnum>("compute_device")))
 #endif
 #ifdef MOOSE_MFEM_ENABLED
     ,
@@ -533,13 +527,6 @@ MooseApp::MooseApp(const InputParameters & parameters)
             : std::set<std::string>{})
 #endif
 {
-  if (isParamSetByUser("libtorch_device"))
-  {
-    mooseDeprecated("--libtorch-device has been renamed --compute-device.");
-    if (isParamSetByUser("compute_device"))
-      mooseError("Remove the redundant --libtorch-device option.");
-  }
-
   if (&parameters != &_pars)
   {
     const auto show_trace = Moose::show_trace;
@@ -801,8 +788,6 @@ MooseApp::MooseApp(const InputParameters & parameters)
 std::optional<MooseEnum>
 MooseApp::getComputeDevice() const
 {
-  if (isParamSetByUser("libtorch_device"))
-    return getParam<MooseEnum>("libtorch_device");
   if (isParamSetByUser("compute_device"))
     return getParam<MooseEnum>("compute_device");
   return {};
@@ -2371,6 +2356,19 @@ MooseApp::runInputs()
   return false;
 }
 
+void
+MooseApp::checkReservedCapability(const std::string & capability)
+{
+  // The list of these capabilities should match those within
+  // Tester.checkRunnableBase() in the TestHarness
+  static const std::set<std::string> reserved{
+      "scale_refine", "valgrind", "recover", "heavy", "mpi_procs", "num_threads", "compute_device"};
+  if (reserved.count(capability))
+    mooseError("MooseApp::addCapability(): The capability \"",
+               capability,
+               "\" is reserved and may not be registered by an application.");
+}
+
 void
 MooseApp::setOutputPosition(const Point & p)
 {
@@ -3530,7 +3528,7 @@ MooseApp::constructingMeshGenerators() const
 torch::DeviceType
 MooseApp::determineLibtorchDeviceType(const MooseEnum & device_enum) const
 {
-  const auto pname = isParamSetByUser("libtorch_device") ? "--libtorch-device" : "--compute-device";
+  const auto pname = "--compute-device";
   if (device_enum == "cuda")
   {
 #ifdef __linux__
@@ -3592,12 +3590,14 @@ MooseApp::addCapability(const std::string & capability,
                         CapabilityUtils::Type value,
                         const std::string & doc)
 {
+  checkReservedCapability(capability);
   Moose::Capabilities::getCapabilityRegistry().add(capability, value, doc);
 }
 
 void
 MooseApp::addCapability(const std::string & capability, const char * value, const std::string & doc)
 {
+  checkReservedCapability(capability);
   Moose::Capabilities::getCapabilityRegistry().add(capability, std::string(value), doc);
 }
 

python/TestHarness/TestHarness.py

@@ -1146,8 +1146,6 @@ def parseCLArgs(self, argv):
         parser.add_argument('--distributed-mesh', action='store_true', dest='distributed_mesh', help='Pass "--distributed-mesh" to executable')
 
         parser.add_argument('--compute-device', action='store', dest='compute_device', type=str, choices=TestHarness.validComputeDevices(), default='cpu', help='Run libtorch or MFEM tests with this compute device; device availability depends on library support and compilation settings')
-        parser.add_argument('--libtorch-device', action='store', dest='compute_device', type=str, choices=TestHarness.validComputeDevices(), help='Deprecated, use --compute-device instead')
-        parser.add_argument('--device', action='store', dest='compute_device', type=str, choices=TestHarness.validComputeDevices(), help='Deprecated, use --compute-device instead')
 
         parser.add_argument('--error', action='store_true', help='Run the tests with warnings as errors (Pass "--error" to executable)')
         parser.add_argument('--error-unused', action='store_true', help='Run the tests with errors on unused parameters (Pass "--error-unused" to executable)')

python/TestHarness/testers/RunApp.py

@@ -53,10 +53,9 @@ def validParams():
         # Valgrind
         params.addParam('valgrind', 'NORMAL', "Set to (NONE, NORMAL, HEAVY) to determine which configurations where valgrind will run.")
 
-        params.addParam('libtorch_devices', "The devices to use for this libtorch test ('CPU', 'CUDA', 'MPS')")
         device_list_str = "', '".join(d.upper() for d in TestHarness.validComputeDevices())
         device_param_doc = f"The devices to use for this libtorch or MFEM test ('{device_list_str}'); device availability depends on library support and compilation settings; default ('CPU')"
-        params.addParam('devices', ['CPU'], device_param_doc)
+        params.addParam('compute_devices', ['CPU'], device_param_doc)
 
         return params
 
@@ -79,12 +78,7 @@ def __init__(self, name, params):
             if params['no_additional_cli_args']:
                 raise Exception('The parameters "command_proxy" and "no_additional_cli_args" cannot be supplied together')
 
-        if params.isValid('libtorch_devices'):
-            for value in params['libtorch_devices']:
-                if value.lower() not in ['cpu', 'cuda', 'mps']:
-                    raise Exception(f'Unknown libtorch_device "{value}"')
-
-        for value in params['devices']:
+        for value in params['compute_devices']:
             if value.lower() not in TestHarness.validComputeDevices():
                 raise Exception(f'Unknown device "{value}"')
 
@@ -130,12 +124,7 @@ def checkRunnable(self, options):
                 self.setStatus(self.skip)
                 return False
 
-        if self.specs.isValid('libtorch_devices'):
-            devices_lower = [x.lower() for x in self.specs['libtorch_devices']]
-        elif self.specs.isValid('devices'):
-            devices_lower = [x.lower() for x in self.specs['devices']]
-        else:
-            devices_lower = [x.lower() for x in self.specs['compute_devices']]
+        devices_lower = [x.lower() for x in self.specs['compute_devices']]
         if options.compute_device not in devices_lower:
             self.addCaveats(f'{options.compute_device} not in compute devices')
             self.setStatus(self.skip)

python/TestHarness/testers/Tester.py

@@ -581,6 +581,8 @@ def augment(key, val_doc):
                     raise ValueError(f"Capability {key} is defined by the app, but it is a reserved dynamic test harness capability. This is an application bug.")
                 capabilities[key] = val_doc
 
+            # NOTE: If you add to this list, add the capability name as a reserved
+            # capability within MooseApp::checkReservedCapability()
             augment('scale_refine', [options.scaling, 'The number of refinements to do when scaling'])
             if options.valgrind_mode == '':
                 augment('valgrind', [False, 'Not running with valgrind'])

test/tests/mfem/ics/tests

@@ -9,7 +9,7 @@
     ignored_items = "root['VTKFile']['@version']"
     requirement = 'The system shall have the ability to set initial conditions on a scalar variable.'
     capabilities = 'mfem'
-    devices = 'cpu cuda'
+    compute_devices = 'cpu cuda'
     max_parallel = 1 # schemadiff with multiple ranks
     recover = false
   []
@@ -21,7 +21,7 @@
     ignored_items = "root['VTKFile']['@version']"
     requirement = 'The system shall have the ability to set initial conditions on a vector MFEM variable.'
     capabilities = 'mfem'
-    devices = 'cpu cuda'
+    compute_devices = 'cpu cuda'
     max_parallel = 1 # schemadiff with multiple ranks
     recover = false
   []
@@ -34,7 +34,7 @@
     ignored_items = "root['VTKFile']['@version']"
     requirement = 'The system shall have the ability to set initial conditions on a scalar variable in transient simulations.'
     capabilities = 'mfem'
-    devices = 'cpu cuda'
+    compute_devices = 'cpu cuda'
     max_parallel = 1 # schemadiff with multiple ranks
     recover = false
   []

test/tests/mfem/kernels/tests

@@ -9,7 +9,7 @@
     ignored_items = "root['VTKFile']['@version']"
     requirement = 'The system shall have the ability to solve a definite Maxwell problem with Nedelec elements of the first kind using MFEM.'
     capabilities = 'mfem'
-    devices = 'cpu' # schemadiff with cuda
+    compute_devices = 'cpu' # schemadiff with cuda
     max_parallel = 1 # schemadiff with multiple ranks
     recover = false
     max_threads = 1
@@ -29,7 +29,7 @@
                'Outputs/ParaViewDataCollection/file_base=OutputData/CurlCurlLOR'
     requirement = 'The system shall have the ability to solve a LOR definite Maxwell problem with Nedelec elements of the first kind using MFEM.'
     capabilities = 'mfem'
-    devices = 'cpu cuda'
+    compute_devices = 'cpu cuda'
     valgrind = heavy
     max_parallel = 1 # schemadiff with multiple ranks
     recover = false
@@ -45,7 +45,7 @@
     ignored_items = "root['VTKFile']['@version']"
     requirement = 'The system shall have the ability to solve a diffusion problem using MFEM.'
     capabilities = 'mfem'
-    devices = 'cpu cuda'
+    compute_devices = 'cpu cuda'
     max_parallel = 1 # schemadiff with multiple ranks
     recover = false
     expect_out = 'Parallel Type:.*distributed.*Mesh Dimension.*3.*Spatial Dimension.*3.*Elems.*2476.*Num Subdomains.*1.*Solver.*MFEMHypreGMRES'
@@ -60,7 +60,7 @@
     ignored_items = "root['VTKFile']['@version']"
     requirement = 'The system shall have the ability to solve a diffusion problem with partial assembly using MFEM.'
     capabilities = 'mfem'
-    devices = 'cpu cuda'
+    compute_devices = 'cpu cuda'
     max_parallel = 1 # schemadiff with multiple ranks
     recover = false
   []
@@ -72,7 +72,7 @@
     cli_args = 'Solver/type=MFEMCGSolver Solver/preconditioner=jacobi '
                'Executioner/device=ceed-cpu '
                'Executioner/assembly_level=none'
-    devices = 'ceed-cpu'
+    compute_devices = 'ceed-cpu'
     ignored_items = "root['VTKFile']['@version']"
     requirement = 'The system shall have the ability to solve a diffusion problem with matrix-free assembly using MFEM.'
     capabilities = 'mfem'
@@ -91,7 +91,7 @@
                'Outputs/ParaViewDataCollection/file_base=OutputData/DiffusionLOR'
     requirement = 'The system shall have the ability to solve a diffusion problem with Low-Order-Refined preconditioning set up from MOOSE and produce the same result as a native MFEM run.'
     capabilities = 'mfem'
-    devices = 'cpu cuda'
+    compute_devices = 'cpu cuda'
     max_parallel = 1 # schemadiff with multiple ranks
     recover = false
     abs_zero = 1e-6
@@ -104,7 +104,7 @@
     ignored_items = "root['VTKFile']['@version']"
     requirement = 'The system shall have the ability to solve a grad-div problem with Raviart-Thomas elements using MFEM.'
     capabilities = 'mfem'
-    devices = 'cpu cuda'
+    compute_devices = 'cpu cuda'
     max_parallel = 1 # schemadiff with multiple ranks
     recover = false
     errors = 'foobarbaz' # allow cuda errors I guess?
@@ -122,7 +122,7 @@
                'Outputs/ParaViewDataCollection/file_base=OutputData/GradDivLOR'
     requirement = 'The system shall have the ability to solve a LOR grad-div problem with Raviart-Thomas elements using MFEM.'
     capabilities = 'mfem'
-    devices = 'cpu cuda'
+    compute_devices = 'cpu cuda'
     valgrind = heavy
     max_parallel = 1 # schemadiff with multiple ranks
     recover = false
@@ -140,7 +140,7 @@
     ignored_items = "root['VTKFile']['@version']"
     requirement = 'The system shall have the ability to solve a transient heat conduction problem using MFEM.'
     capabilities = 'mfem'
-    devices = 'cpu cuda'
+    compute_devices = 'cpu cuda'
     max_parallel = 1 # schemadiff with multiple ranks
     recover = false
   []
@@ -158,7 +158,7 @@
     ignored_items = "root['VTKFile']['@version']"
     requirement = 'The system shall have the ability to solve a transient heat conduction problem with element assembly using MFEM.'
     capabilities = 'mfem'
-    devices = 'cpu cuda'
+    compute_devices = 'cpu cuda'
     valgrind = none
     max_parallel = 1 # schemadiff with multiple ranks
     recover = false
@@ -175,7 +175,7 @@
                'Solver/type=MFEMCGSolver Solver/preconditioner=jacobi '
                'Executioner/assembly_level=none '
                'Executioner/device=ceed-cpu'
-    devices = 'ceed-cpu'
+    compute_devices = 'ceed-cpu'
     ignored_items = "root['VTKFile']['@version']"
     requirement = 'The system shall have the ability to solve a transient heat conduction problem with matrix-free assembly using MFEM.'
     capabilities = 'mfem'
@@ -198,7 +198,7 @@
                'Outputs/ParaViewDataCollection/file_base=OutputData/HeatConductionLOR'
     requirement = 'The system shall have the ability to solve a transient heat conduction problem with Low-Order-Refined preconditioning set up from MOOSE and produce the same result as a native MFEM run.'
     capabilities = 'mfem'
-    devices = 'cpu cuda'
+    compute_devices = 'cpu cuda'
     max_parallel = 1 # schemadiff with multiple ranks
     recover = false
   []
@@ -210,7 +210,7 @@
     ignored_items = "root['VTKFile']['@version']"
     requirement = 'The system shall have the ability to solve a transient heat conduction problem with a heat transfer coefficient on one boundary using MFEM.'
     capabilities = 'mfem'
-    devices = 'cpu cuda'
+    compute_devices = 'cpu cuda'
     max_parallel = 1 # schemadiff with multiple ranks
     recover = false
   []
@@ -222,7 +222,7 @@
     ignored_items = "root['VTKFile']['@version']"
     requirement = 'The system shall have the ability to solve a linear elasticity problem using MFEM.'
     capabilities = 'mfem'
-    devices = 'cpu cuda'
+    compute_devices = 'cpu cuda'
     max_parallel = 1 # schemadiff with multiple ranks
     recover = false
   []
@@ -234,7 +234,7 @@
     ignored_items = "root['VTKFile']['@version']"
     requirement = 'The system shall have the ability to solve a linear elasticity problem for a beam deformed under gravitational loads using MFEM.'
     capabilities = 'mfem'
-    devices = 'cpu' # schemadiff with cuda
+    compute_devices = 'cpu' # schemadiff with cuda
     max_parallel = 1 # schemadiff with multiple ranks
     recover = false
     max_threads = 1
@@ -247,7 +247,7 @@
     requirement = 'The system shall have the ability to solve for an irrotational vortex in two dimensions using MFEM.'
     abs_zero = 1e-6
     capabilities = 'mfem'
-    devices = 'cpu cuda'
+    compute_devices = 'cpu cuda'
     recover = false
     max_parallel = 1 # MOOSE-MFEM-HypreGMRES doesn't appear to work with multiple ranks with cuda
   []
@@ -258,7 +258,7 @@
     requirement = 'The system shall have the ability to solve a mixed Darcy problem in two dimensions using MFEM.'
     abs_zero = 1e-6
     capabilities = 'mfem'
-    devices = 'cpu cuda'
+    compute_devices = 'cpu cuda'
     recover = false
   []
 []

test/tests/mfem/outputs/tests

@@ -17,7 +17,7 @@
     requirement = 'The system shall have the ability to write out MFEM solutions to '
                   'ParaView, VisIt and Conduit data collections.'
     capabilities = 'mfem'
-    devices = 'cpu cuda'
+    compute_devices = 'cpu cuda'
     recover = false
     max_parallel = 1 # MOOSE-MFEM-HypreGMRES doesn't appear to work with multiple ranks with cuda
   []

test/tests/mfem/submeshes/tests

@@ -9,7 +9,7 @@
     ignored_items = "root['VTKFile']['@version']"
     requirement = 'MOOSE shall have the ability to solve a diffusion problem defined on an MFEM domain restricted submesh.'
     capabilities = 'mfem'
-    devices = 'cpu cuda'
+    compute_devices = 'cpu cuda'
     max_parallel = 1 # schemadiff with multiple ranks
     recover = false
   []
@@ -21,7 +21,7 @@
     ignored_items = "root['VTKFile']['@version']"
     requirement = 'MOOSE shall have the ability to solve an FE problem defined on an MFEM boundary restricted submesh.'
     capabilities = 'mfem'
-    devices = 'cpu cuda'
+    compute_devices = 'cpu cuda'
     max_parallel = 1 # schemadiff with multiple ranks
     recover = false
   []
@@ -33,7 +33,7 @@
     ignored_items = "root['VTKFile']['@version']"
     requirement = 'MOOSE shall have the ability to transfer a domain restricted submesh variable data to a corresponding variable defined on the parent mesh.'
     capabilities = 'mfem'
-    devices = 'cpu cuda'
+    compute_devices = 'cpu cuda'
     max_parallel = 1 # schemadiff with multiple ranks
     recover = false
   []

test/tests/mfem/transfers/h1_mfem_parent_mfem_sub/tests

@@ -7,7 +7,7 @@
       type = RunApp
       input = parent.i
       capabilities = 'mfem'
-      devices = 'cpu cuda'
+      compute_devices = 'cpu cuda'
       recover = false
       detail = 'running the inputs, and '
     []
@@ -30,7 +30,7 @@
       type = RunApp
       input = sub.i
       capabilities = 'mfem'
-      devices = 'cpu cuda'
+      compute_devices = 'cpu cuda'
       recover = false
       cli_args = 'MultiApps/active=subapp Transfers/active=to_sub '
                  'subapp:MultiApps/active="" subapp:Transfers/active=""'