From 7f9739c266a3c8467c154787aa1e6089440bd1be Mon Sep 17 00:00:00 2001
From: "copilot-swe-agent[bot]" <198982749+Copilot@users.noreply.github.com>
Date: Tue, 10 Mar 2026 15:00:32 +0000
Subject: [PATCH 1/4] Initial plan


From 185fe1c9260f7481cc9c42e74480c89c20a37358 Mon Sep 17 00:00:00 2001
From: "copilot-swe-agent[bot]" <198982749+Copilot@users.noreply.github.com>
Date: Tue, 10 Mar 2026 15:14:27 +0000
Subject: [PATCH 2/4] Move known_callables from ASTBuilderBase to TargetBase

Co-authored-by: inducer <352067+inducer@users.noreply.github.com>
---
 loopy/library/reduction.py |  4 ++--
 loopy/target/__init__.py   | 12 ++++++++++--
 loopy/target/c/__init__.py | 21 ++++++++++++---------
 loopy/target/cuda.py       | 16 ++++++----------
 loopy/target/opencl.py     | 13 +++++++------
 loopy/target/pyopencl.py   | 23 ++++++++++++-----------
 loopy/target/python.py     |  7 -------
 loopy/tools.py             |  2 +-
 loopy/translation_unit.py  |  2 +-
 9 files changed, 51 insertions(+), 49 deletions(-)

diff --git a/loopy/library/reduction.py b/loopy/library/reduction.py
index 38ebc1c70..adb5a0bfa 100644
--- a/loopy/library/reduction.py
+++ b/loopy/library/reduction.py
@@ -368,7 +368,7 @@ def __call__(self,
         from loopy.translation_unit import add_callable_to_table
 
         # getting the callable 'max' from target
-        max_scalar_callable = target.get_device_ast_builder().known_callables["max"]
+        max_scalar_callable = target.known_callables["max"]
 
         # type specialize the callable
         max_scalar_callable, callables_table = max_scalar_callable.with_types(
@@ -404,7 +404,7 @@ def __call__(self,
         from loopy.translation_unit import add_callable_to_table
 
         # getting the callable 'min' from target
-        min_scalar_callable = target.get_device_ast_builder().known_callables["min"]
+        min_scalar_callable = target.known_callables["min"]
 
         # type specialize the callable
         min_scalar_callable, callables_table = min_scalar_callable.with_types(
diff --git a/loopy/target/__init__.py b/loopy/target/__init__.py
index 9476c0e2a..3785067ce 100644
--- a/loopy/target/__init__.py
+++ b/loopy/target/__init__.py
@@ -191,6 +191,15 @@ def get_kernel_executor(
         """
         raise NotImplementedError()
 
+    @property
+    def known_callables(self):
+        """
+        Returns a mapping from function ids to corresponding
+        :class:`loopy.kernel.function_interface.InKernelCallable` for the
+        function ids known to *self*.
+        """
+        return {}
+
 
 @dataclass(frozen=True)
 class ASTBuilderBase(ABC, Generic[ASTType]):
@@ -208,8 +217,7 @@ def known_callables(self):
         :class:`loopy.kernel.function_interface.InKernelCallable` for the
         function ids known to *self.target*.
         """
-        # FIXME: @inducer: Do we need to move this to TargetBase?
-        return {}
+        return dict(self.target.known_callables)
 
     def symbol_manglers(self):
         return []
diff --git a/loopy/target/c/__init__.py b/loopy/target/c/__init__.py
index a2885b7bf..41b64d710 100644
--- a/loopy/target/c/__init__.py
+++ b/loopy/target/c/__init__.py
@@ -509,6 +509,13 @@ def get_host_ast_builder(self):
     def get_device_ast_builder(self):
         return CFamilyASTBuilder(self)
 
+    @property
+    @override
+    def known_callables(self):
+        callables = super().known_callables
+        callables.update(get_c_callables())
+        return callables
+
     # {{{ types
 
     @memoize_method
@@ -890,13 +897,6 @@ def preamble_generators(self):
                     lambda preamble_info: _preamble_generator(
                           preamble_info, self.preamble_function_qualifier)])
 
-    @property
-    @override
-    def known_callables(self):
-        callables = super().known_callables
-        callables.update(get_c_callables())
-        return callables
-
     # }}}
 
     # {{{ code generation
@@ -1606,15 +1606,18 @@ class CWithGNULibcTarget(CTarget):
     def get_device_ast_builder(self):
         return CWithGNULibcASTBuilder(self)
 
-
-class CWithGNULibcASTBuilder(CASTBuilder):
     @property
+    @override
     def known_callables(self):
         callables = super().known_callables
         callables.update(get_gnu_libc_callables())
         return callables
 
 
+class CWithGNULibcASTBuilder(CASTBuilder):
+    pass
+
+
 class ExecutableCWithGNULibcTarget(ExecutableCTarget):
     def get_device_ast_builder(self):
         return CWithGNULibcASTBuilder(self)
diff --git a/loopy/target/cuda.py b/loopy/target/cuda.py
index 23cd2dab0..614786a0f 100644
--- a/loopy/target/cuda.py
+++ b/loopy/target/cuda.py
@@ -249,6 +249,12 @@ def split_kernel_at_global_barriers(self):
     def get_device_ast_builder(self):
         return CUDACASTBuilder(self)
 
+    @property
+    def known_callables(self):
+        callables = super().known_callables
+        callables.update(get_cuda_callables())
+        return callables
+
     # {{{ types
 
     @memoize_method
@@ -330,16 +336,6 @@ class CUDACASTBuilder(CFamilyASTBuilder):
 
     preamble_function_qualifier = "inline __device__"
 
-    # {{{ library
-
-    @property
-    def known_callables(self):
-        callables = super().known_callables
-        callables.update(get_cuda_callables())
-        return callables
-
-    # }}}
-
     # {{{ top-level codegen
 
     def get_function_declaration(
diff --git a/loopy/target/opencl.py b/loopy/target/opencl.py
index ed05c7628..0d17a2ca1 100644
--- a/loopy/target/opencl.py
+++ b/loopy/target/opencl.py
@@ -632,6 +632,13 @@ def split_kernel_at_global_barriers(self):
     def get_device_ast_builder(self):
         return OpenCLCASTBuilder(self)
 
+    @property
+    @override
+    def known_callables(self):
+        callables = super().known_callables
+        callables.update(get_opencl_callables())
+        return callables
+
     @memoize_method
     def get_dtype_registry(self) -> DTypeRegistry:
         from loopy.target.c.compyte.dtypes import (
@@ -673,12 +680,6 @@ def vector_dtype(self, base, count):
 class OpenCLCASTBuilder(CFamilyASTBuilder):
     # {{{ library
 
-    @property
-    def known_callables(self):
-        callables = super().known_callables
-        callables.update(get_opencl_callables())
-        return callables
-
     def symbol_manglers(self):
         return (
                 [*super().symbol_manglers(), opencl_symbol_mangler])
diff --git a/loopy/target/pyopencl.py b/loopy/target/pyopencl.py
index 9d88394c9..55f046ed4 100644
--- a/loopy/target/pyopencl.py
+++ b/loopy/target/pyopencl.py
@@ -584,6 +584,18 @@ def get_host_ast_builder(self):
     def get_device_ast_builder(self):
         return PyOpenCLCASTBuilder(self)
 
+    @property
+    @override
+    def known_callables(self):
+        from loopy.library.random123 import get_random123_callables
+
+        # order matters: e.g. prefer our abs() over that of the
+        # superclass
+        callables = super().known_callables
+        callables.update(get_pyopencl_callables())
+        callables.update(get_random123_callables(self))
+        return callables
+
     # {{{ types
 
     @override
@@ -1224,17 +1236,6 @@ def get_function_declaration(
 
     # {{{ library
 
-    @property
-    def known_callables(self):
-        from loopy.library.random123 import get_random123_callables
-
-        # order matters: e.g. prefer our abs() over that of the
-        # superclass
-        callables = super().known_callables
-        callables.update(get_pyopencl_callables())
-        callables.update(get_random123_callables(self.target))
-        return callables
-
     def preamble_generators(self):
         return ([pyopencl_preamble_generator, *super().preamble_generators()])
 
diff --git a/loopy/target/python.py b/loopy/target/python.py
index 3b4b9795f..1fd336da0 100644
--- a/loopy/target/python.py
+++ b/loopy/target/python.py
@@ -190,13 +190,6 @@ class PythonASTBuilderBase(ASTBuilderBase[Generable]):
     """A Python host AST builder for integration with PyOpenCL.
     """
 
-    @property
-    def known_callables(self):
-        from loopy.target.c import get_c_callables
-        callables = super().known_callables
-        callables.update(get_c_callables())
-        return callables
-
     def preamble_generators(self):
         return (
                 [*super().preamble_generators(), _base_python_preamble_generator])
diff --git a/loopy/tools.py b/loopy/tools.py
index 8ab419585..777cc38a2 100644
--- a/loopy/tools.py
+++ b/loopy/tools.py
@@ -674,7 +674,7 @@ def __init__(self, rule_mapping_context, callables_table, target):
     @cached_property
     def known_callables(self):
         from loopy.kernel.function_interface import CallableKernel
-        return (frozenset(self.target.get_device_ast_builder().known_callables)
+        return (frozenset(self.target.known_callables)
                 | {name
                    for name, clbl in self.callables_table.items()
                    if isinstance(clbl, CallableKernel)})
diff --git a/loopy/translation_unit.py b/loopy/translation_unit.py
index 9d27e24a4..fbb20b91b 100644
--- a/loopy/translation_unit.py
+++ b/loopy/translation_unit.py
@@ -873,7 +873,7 @@ def resolve_callables(t_unit: TranslationUnit) -> TranslationUnit:
     # get registered callables
     known_callables = dict(t_unit.callables_table)
     # get target specific callables
-    known_callables.update(t_unit.target.get_device_ast_builder().known_callables)
+    known_callables.update(t_unit.target.known_callables)
     # get loopy specific callables
     known_callables.update(get_loopy_callables())
 

From 91bbd26372f9a30c738502b76ed48f19ce380a69 Mon Sep 17 00:00:00 2001
From: "copilot-swe-agent[bot]" <198982749+Copilot@users.noreply.github.com>
Date: Tue, 10 Mar 2026 15:28:24 +0000
Subject: [PATCH 3/4] Split known_callables into known_host_callables and
 known_device_callables on TargetBase

Co-authored-by: inducer <352067+inducer@users.noreply.github.com>
---
 loopy/library/reduction.py |  4 ++--
 loopy/target/__init__.py   | 21 +++++++++++++++++----
 loopy/target/c/__init__.py |  8 ++++----
 loopy/target/cuda.py       |  4 ++--
 loopy/target/opencl.py     |  4 ++--
 loopy/target/pyopencl.py   | 12 ++++++++++--
 loopy/target/python.py     |  4 ++++
 loopy/tools.py             |  2 +-
 loopy/translation_unit.py  |  2 +-
 9 files changed, 43 insertions(+), 18 deletions(-)

diff --git a/loopy/library/reduction.py b/loopy/library/reduction.py
index adb5a0bfa..660cbf482 100644
--- a/loopy/library/reduction.py
+++ b/loopy/library/reduction.py
@@ -368,7 +368,7 @@ def __call__(self,
         from loopy.translation_unit import add_callable_to_table
 
         # getting the callable 'max' from target
-        max_scalar_callable = target.known_callables["max"]
+        max_scalar_callable = target.known_device_callables["max"]
 
         # type specialize the callable
         max_scalar_callable, callables_table = max_scalar_callable.with_types(
@@ -404,7 +404,7 @@ def __call__(self,
         from loopy.translation_unit import add_callable_to_table
 
         # getting the callable 'min' from target
-        min_scalar_callable = target.known_callables["min"]
+        min_scalar_callable = target.known_device_callables["min"]
 
         # type specialize the callable
         min_scalar_callable, callables_table = min_scalar_callable.with_types(
diff --git a/loopy/target/__init__.py b/loopy/target/__init__.py
index 3785067ce..ec6eadca0 100644
--- a/loopy/target/__init__.py
+++ b/loopy/target/__init__.py
@@ -192,11 +192,20 @@ def get_kernel_executor(
         raise NotImplementedError()
 
     @property
-    def known_callables(self):
+    def known_host_callables(self):
         """
         Returns a mapping from function ids to corresponding
         :class:`loopy.kernel.function_interface.InKernelCallable` for the
-        function ids known to *self*.
+        function ids known to *self* for host code generation.
+        """
+        return {}
+
+    @property
+    def known_device_callables(self):
+        """
+        Returns a mapping from function ids to corresponding
+        :class:`loopy.kernel.function_interface.InKernelCallable` for the
+        function ids known to *self* for device code generation.
         """
         return {}
 
@@ -215,9 +224,9 @@ def known_callables(self):
         """
         Returns a mapping from function ids to corresponding
         :class:`loopy.kernel.function_interface.InKernelCallable` for the
-        function ids known to *self.target*.
+        function ids known to *self.target* for device code generation.
         """
-        return dict(self.target.known_callables)
+        return dict(self.target.known_device_callables)
 
     def symbol_manglers(self):
         return []
@@ -359,6 +368,10 @@ def __str__(self):
 
 
 class DummyHostASTBuilder(ASTBuilderBase[None]):
+    @property
+    def known_callables(self):
+        return dict(self.target.known_host_callables)
+
     def get_function_definition(self, codegen_state, codegen_result,
             schedule_index, function_decl, function_body):
         return function_body
diff --git a/loopy/target/c/__init__.py b/loopy/target/c/__init__.py
index 41b64d710..c6803f3fd 100644
--- a/loopy/target/c/__init__.py
+++ b/loopy/target/c/__init__.py
@@ -511,8 +511,8 @@ def get_device_ast_builder(self):
 
     @property
     @override
-    def known_callables(self):
-        callables = super().known_callables
+    def known_device_callables(self):
+        callables = super().known_device_callables
         callables.update(get_c_callables())
         return callables
 
@@ -1608,8 +1608,8 @@ def get_device_ast_builder(self):
 
     @property
     @override
-    def known_callables(self):
-        callables = super().known_callables
+    def known_device_callables(self):
+        callables = super().known_device_callables
         callables.update(get_gnu_libc_callables())
         return callables
 
diff --git a/loopy/target/cuda.py b/loopy/target/cuda.py
index 614786a0f..5e07da951 100644
--- a/loopy/target/cuda.py
+++ b/loopy/target/cuda.py
@@ -250,8 +250,8 @@ def get_device_ast_builder(self):
         return CUDACASTBuilder(self)
 
     @property
-    def known_callables(self):
-        callables = super().known_callables
+    def known_device_callables(self):
+        callables = super().known_device_callables
         callables.update(get_cuda_callables())
         return callables
 
diff --git a/loopy/target/opencl.py b/loopy/target/opencl.py
index 0d17a2ca1..e434d608a 100644
--- a/loopy/target/opencl.py
+++ b/loopy/target/opencl.py
@@ -634,8 +634,8 @@ def get_device_ast_builder(self):
 
     @property
     @override
-    def known_callables(self):
-        callables = super().known_callables
+    def known_device_callables(self):
+        callables = super().known_device_callables
         callables.update(get_opencl_callables())
         return callables
 
diff --git a/loopy/target/pyopencl.py b/loopy/target/pyopencl.py
index 55f046ed4..dbea17765 100644
--- a/loopy/target/pyopencl.py
+++ b/loopy/target/pyopencl.py
@@ -586,16 +586,24 @@ def get_device_ast_builder(self):
 
     @property
     @override
-    def known_callables(self):
+    def known_device_callables(self):
         from loopy.library.random123 import get_random123_callables
 
         # order matters: e.g. prefer our abs() over that of the
         # superclass
-        callables = super().known_callables
+        callables = super().known_device_callables
         callables.update(get_pyopencl_callables())
         callables.update(get_random123_callables(self))
         return callables
 
+    @property
+    @override
+    def known_host_callables(self):
+        from loopy.target.c import get_c_callables
+        callables = super().known_host_callables
+        callables.update(get_c_callables())
+        return callables
+
     # {{{ types
 
     @override
diff --git a/loopy/target/python.py b/loopy/target/python.py
index 1fd336da0..5de59b38f 100644
--- a/loopy/target/python.py
+++ b/loopy/target/python.py
@@ -190,6 +190,10 @@ class PythonASTBuilderBase(ASTBuilderBase[Generable]):
     """A Python host AST builder for integration with PyOpenCL.
     """
 
+    @property
+    def known_callables(self):
+        return dict(self.target.known_host_callables)
+
     def preamble_generators(self):
         return (
                 [*super().preamble_generators(), _base_python_preamble_generator])
diff --git a/loopy/tools.py b/loopy/tools.py
index 777cc38a2..685293558 100644
--- a/loopy/tools.py
+++ b/loopy/tools.py
@@ -674,7 +674,7 @@ def __init__(self, rule_mapping_context, callables_table, target):
     @cached_property
     def known_callables(self):
         from loopy.kernel.function_interface import CallableKernel
-        return (frozenset(self.target.known_callables)
+        return (frozenset(self.target.known_device_callables)
                 | {name
                    for name, clbl in self.callables_table.items()
                    if isinstance(clbl, CallableKernel)})
diff --git a/loopy/translation_unit.py b/loopy/translation_unit.py
index fbb20b91b..cfb2e9b12 100644
--- a/loopy/translation_unit.py
+++ b/loopy/translation_unit.py
@@ -873,7 +873,7 @@ def resolve_callables(t_unit: TranslationUnit) -> TranslationUnit:
     # get registered callables
     known_callables = dict(t_unit.callables_table)
     # get target specific callables
-    known_callables.update(t_unit.target.known_callables)
+    known_callables.update(t_unit.target.known_device_callables)
     # get loopy specific callables
     known_callables.update(get_loopy_callables())
 

From 844de53791e7e82ce885a91eb8cbcfb910d38785 Mon Sep 17 00:00:00 2001
From: "copilot-swe-agent[bot]" <198982749+Copilot@users.noreply.github.com>
Date: Tue, 10 Mar 2026 16:30:05 +0000
Subject: [PATCH 4/4] Fix ExecutableCWithGNULibcTarget missing
 known_device_callables override

Co-authored-by: inducer <352067+inducer@users.noreply.github.com>
---
 loopy/target/c/__init__.py | 7 +++++++
 1 file changed, 7 insertions(+)

diff --git a/loopy/target/c/__init__.py b/loopy/target/c/__init__.py
index c6803f3fd..6c5520bfc 100644
--- a/loopy/target/c/__init__.py
+++ b/loopy/target/c/__init__.py
@@ -1622,6 +1622,13 @@ class ExecutableCWithGNULibcTarget(ExecutableCTarget):
     def get_device_ast_builder(self):
         return CWithGNULibcASTBuilder(self)
 
+    @property
+    @override
+    def known_device_callables(self):
+        callables = super().known_device_callables
+        callables.update(get_gnu_libc_callables())
+        return callables
+
 # }}}
 
 # vim: foldmethod=marker