misc: Support more types of errors

devitocodes · Dec 17, 2024 · df6e14c · df6e14c
1 parent 0b7e2ef
commit df6e14c
Show file tree

Hide file tree

Showing 2 changed files with 14 additions and 0 deletions.
diff --git a/devito/operator/operator.py b/devito/operator/operator.py
@@ -696,6 +696,18 @@ def _postprocess_errors(self, retval):
             raise ExecutionError("Detected nan/inf in some output Functions")
         elif retval == error_mapper['KernelLaunch']:
             raise ExecutionError("Kernel launch failed")
+        elif retval == error_mapper['KernelLaunchOutOfResources']:
+            raise ExecutionError(
+                "Kernel launch failed due to insufficient resources. This may be "
+                "due to excessive register pressure in one of the Operator "
+                "kernels. Try supplying a smaller `par-tile` value."
+            )
+        elif retval == error_mapper['KernelLaunchUnknown']:
+            raise ExecutionError(
+                "Kernel launch failed due to an unknown error. This might "
+                "simply indicate memory corruption, but also, in a more unlikely "
+                "case, a hardware issue. Please report this issue to the "
+                "Devito team.")
         else:
             raise ExecutionError("An error occurred during execution")
 

diff --git a/devito/passes/iet/errors.py b/devito/passes/iet/errors.py
@@ -108,4 +108,6 @@ class Retval(LocalObject, Expr):
 error_mapper = {
     'Stability': 100,
     'KernelLaunch': 200,
+    'KernelLaunchOutOfResources': 201,
+    'KernelLaunchUnknown': 202,
 }