Merge pull request #2597 from devitocodes/deviceid-envvar

api: Add DEVITO_DEVICE_DEFAULT env var

Author: FabioLuporini

FAQ.md

@@ -292,6 +292,9 @@ Used to select a specific "backend compiler". The backend compiler takes as inpu
 #### DEVITO_PLATFORM
 This environment variable is mostly needed when running on GPUs, to ask Devito to generate code for a particular device (see for example this [tutorial](https://github.com/devitocodes/devito/blob/main/examples/gpu/01_diffusion_with_openmp_offloading.ipynb)). Can be also used to specify CPU architectures such as Intel's -- Haswell, Broadwell, SKL and KNL -- ARM, AMD, and Power. Often one can ignore this variable because Devito typically does a decent job at auto-detecting the underlying platform.
 
+#### DEVITO_DEVICEID
+Specify the default device to use when running on a GPU. This is useful when multiple GPUs are available on the system. The default is `0`, which means the first device. Use `1` for the second device, and so on.
+
 #### DEVITO_LANGUAGE
 Specify the generated code language. The default is `C`, which means sequential C. Use `openmp` to emit C+OpenMP or `openacc` for C+OpenACC.
 

devito/arch/archinfo.py

@@ -1023,7 +1023,12 @@ def supports(self, query, language=None):
             return False
 
         cc = get_nvidia_cc()
-        if query == 'async-loads' and cc >= 80:
+        if cc is None:
+            # Something off with `get_nvidia_cc` on this system
+            warning(f"Couldn't establish if `query={query}` is supported on this "
+                    "system. Assuming it is not.")
+            return False
+        elif query == 'async-loads' and cc >= 80:
             # Asynchronous pipeline loads -- introduced in Ampere
             return True
         elif query in ('tma', 'thread-block-cluster') and cc >= 90:

devito/parameters.py

@@ -147,6 +147,7 @@ def _signature_items(self):
     'DEVITO_OPT': 'opt',
     'DEVITO_MPI': 'mpi',
     'DEVITO_TOPOLOGY': 'topology',
+    'DEVITO_DEVICEID': 'deviceid',
     'DEVITO_LANGUAGE': 'language',
     'DEVITO_AUTOTUNING': 'autotuning',
     'DEVITO_LOGGING': 'log-level',

devito/types/dimension.py

@@ -916,7 +916,7 @@ def __init_finalize__(self, name, parent=None, factor=None, condition=None,
         super().__init_finalize__(name, parent)
 
         # Process subsampling factor
-        if factor is None or factor == 1:
+        if factor is None:
             self._factor = None
         elif is_number(factor):
             self._factor = int(factor)
@@ -937,16 +937,18 @@ def uses_symbolic_factor(self):
     def factor_data(self):
         if isinstance(self.factor, Constant):
             return self.factor.data
-        else:
+        elif self.factor is not None:
             return self.factor
+        else:
+            return 1
 
     @property
     def spacing(self):
         return self.factor_data * self.parent.spacing
 
     @property
     def factor(self):
-        return self._factor if self.uses_symbolic_factor else 1
+        return self._factor
 
     @cached_property
     def symbolic_factor(self):