get_cached_kernel_executor -> get_cached_kernel

Author: inducer

sumpy/e2e.py

@@ -282,7 +282,7 @@ def __call__(self, actx: PyOpenCLArrayContext, **kwargs):
         src_rscale = centers.dtype.type(kwargs.pop("src_rscale"))
         tgt_rscale = centers.dtype.type(kwargs.pop("tgt_rscale"))
 
-        knl = self.get_cached_kernel_executor()
+        knl = self.get_cached_kernel()
         result = actx.call_loopy(
             knl,
             centers=centers,
@@ -527,7 +527,7 @@ def __call__(self, actx: PyOpenCLArrayContext, **kwargs):
         tgt_rscale = centers.dtype.type(kwargs.pop("tgt_rscale"))
         src_expansions = kwargs.pop("src_expansions")
 
-        knl = self.get_cached_kernel_executor(result_dtype=src_expansions.dtype)
+        knl = self.get_cached_kernel(result_dtype=src_expansions.dtype)
         result = actx.call_loopy(
             knl,
             src_expansions=src_expansions,
@@ -641,7 +641,7 @@ def __call__(self, actx: PyOpenCLArrayContext, **kwargs):
                 "m2l_translation_classes_dependent_data")
         result_dtype = m2l_translation_classes_dependent_data.dtype
 
-        knl = self.get_cached_kernel_executor(result_dtype=result_dtype)
+        knl = self.get_cached_kernel(result_dtype=result_dtype)
         result = actx.call_loopy(
             knl,
             src_rscale=src_rscale,
@@ -737,7 +737,7 @@ def __call__(self, actx: PyOpenCLArrayContext, **kwargs):
         preprocessed_src_expansions = kwargs.pop("preprocessed_src_expansions")
         result_dtype = preprocessed_src_expansions.dtype
 
-        knl = self.get_cached_kernel_executor(result_dtype=result_dtype)
+        knl = self.get_cached_kernel(result_dtype=result_dtype)
         result = actx.call_loopy(
             knl,
             preprocessed_src_expansions=preprocessed_src_expansions,
@@ -838,7 +838,7 @@ def __call__(self, actx: PyOpenCLArrayContext, **kwargs):
         tgt_expansions = kwargs.pop("tgt_expansions")
         result_dtype = tgt_expansions.dtype
 
-        knl = self.get_cached_kernel_executor(result_dtype=result_dtype)
+        knl = self.get_cached_kernel(result_dtype=result_dtype)
         result = actx.call_loopy(
             knl,
             tgt_expansions=tgt_expansions,
@@ -958,7 +958,7 @@ def __call__(self, actx: PyOpenCLArrayContext, **kwargs):
         src_rscale = centers.dtype.type(kwargs.pop("src_rscale"))
         tgt_rscale = centers.dtype.type(kwargs.pop("tgt_rscale"))
 
-        knl = self.get_cached_kernel_executor()
+        knl = self.get_cached_kernel()
         result = actx.call_loopy(
             knl,
             centers=centers,
@@ -1065,7 +1065,7 @@ def __call__(self, actx: PyOpenCLArrayContext, **kwargs):
         src_rscale = centers.dtype.type(kwargs.pop("src_rscale"))
         tgt_rscale = centers.dtype.type(kwargs.pop("tgt_rscale"))
 
-        knl = self.get_cached_kernel_executor()
+        knl = self.get_cached_kernel()
         result = actx.call_loopy(
             knl,
             centers=centers,

sumpy/e2p.py

@@ -218,7 +218,7 @@ def __call__(self, actx: PyOpenCLArrayContext, **kwargs):
         # meaningfully inferred. Make the type of rscale explicit.
         rscale = centers.dtype.type(kwargs.pop("rscale"))
 
-        knl = self.get_cached_kernel_executor()
+        knl = self.get_cached_kernel()
         result = actx.call_loopy(
             knl,
             centers=centers, rscale=rscale, **kwargs)
@@ -337,7 +337,7 @@ def __call__(self, actx: PyOpenCLArrayContext, **kwargs):
         # meaningfully inferred. Make the type of rscale explicit.
         rscale = centers.dtype.type(kwargs.pop("rscale"))
 
-        knl = self.get_cached_kernel_executor()
+        knl = self.get_cached_kernel()
         result = actx.call_loopy(
             knl,
             centers=centers,

sumpy/p2e.py

@@ -133,7 +133,7 @@ def __call__(self, actx: PyOpenCLArrayContext, **kwargs):
         dtype = centers[0].dtype if is_obj_array_like(centers) else centers.dtype
         rscale = dtype.type(kwargs.pop("rscale"))
 
-        knl = self.get_cached_kernel_executor(
+        knl = self.get_cached_kernel(
                 sources_is_obj_array=is_obj_array_like(sources),
                 centers_is_obj_array=is_obj_array_like(centers))
 

sumpy/p2p.py

@@ -272,7 +272,7 @@ def __call__(self,
             strength: Sequence[Array],
             **kwargs: Any,
         ) -> tuple[cl.Event, Sequence[Array]]:
-        knl = self.get_cached_kernel_executor(
+        knl = self.get_cached_kernel(
                 targets_is_obj_array=is_obj_array_like(targets),
                 sources_is_obj_array=is_obj_array_like(sources))
 
@@ -345,7 +345,7 @@ def __call__(self,
             sources: ObjectArray1D[Array] | Array,
             **kwargs: Any,
         ) -> Sequence[Array]:
-        knl = self.get_cached_kernel_executor(
+        knl = self.get_cached_kernel(
                 targets_is_obj_array=is_obj_array_like(targets),
                 sources_is_obj_array=is_obj_array_like(sources))
 
@@ -439,7 +439,6 @@ def get_optimized_kernel(self, targets_is_obj_array, sources_is_obj_array):
         knl = self._allow_redundant_execution_of_knl_scaling(knl)
         knl = lp.set_options(knl,
                 enforce_variable_access_ordered="no_check")
-        knl = register_optimization_preambles(knl, self.device)
 
         return knl
 
@@ -467,7 +466,7 @@ def __call__(self,
         :returns: a one-dimensional array of interactions, for each index pair
             in (*srcindices*, *tgtindices*)
         """
-        knl = self.get_cached_kernel_executor(
+        knl = self.get_cached_kernel(
                 targets_is_obj_array=is_obj_array_like(targets),
                 sources_is_obj_array=is_obj_array_like(sources))
 
@@ -824,7 +823,7 @@ def __call__(self,
             source_dtype = None
             strength_dtype = None
 
-        knl = self.get_cached_kernel_executor(
+        knl = self.get_cached_kernel(
                 max_nsources_in_one_box=max_nsources_in_one_box,
                 max_ntargets_in_one_box=max_ntargets_in_one_box,
                 local_mem_size=actx.queue.device.local_mem_size,

sumpy/qbx.py

@@ -317,7 +317,7 @@ def __call__(self, actx: PyOpenCLArrayContext,
             already multiplied in.
         """
 
-        knl = self.get_cached_kernel_executor(
+        knl = self.get_cached_kernel(
                 is_cpu=is_cl_cpu(actx),
                 targets_is_obj_array=is_obj_array_like(targets),
                 sources_is_obj_array=is_obj_array_like(sources),
@@ -396,7 +396,7 @@ def get_kernel(self):
 
     def __call__(self, actx: PyOpenCLArrayContext,
             targets, sources, centers, expansion_radii, **kwargs):
-        knl = self.get_cached_kernel_executor(
+        knl = self.get_cached_kernel(
                 is_cpu=is_cl_cpu(actx),
                 targets_is_obj_array=is_obj_array_like(targets),
                 sources_is_obj_array=is_obj_array_like(sources),
@@ -525,7 +525,7 @@ def __call__(self, actx: PyOpenCLArrayContext,
             in (*srcindices*, *tgtindices*)
         """
 
-        knl = self.get_cached_kernel_executor(
+        knl = self.get_cached_kernel(
                 targets_is_obj_array=is_obj_array_like(targets),
                 sources_is_obj_array=is_obj_array_like(sources),
                 centers_is_obj_array=is_obj_array_like(centers))

sumpy/tools.py

@@ -454,7 +454,7 @@ def get_optimized_kernel(self, **kwargs: Any) -> lp.TranslationUnit:
         ...
 
     @memoize_method
-    def get_cached_kernel_executor(self, **kwargs) -> lp.ExecutorBase:
+    def get_cached_kernel(self, **kwargs) -> lp.TranslationUnit:
         from sumpy import CACHING_ENABLED, NO_CACHE_KERNELS, OPT_ENABLED, code_cache
 
         if CACHING_ENABLED and not (
@@ -472,7 +472,7 @@ def get_cached_kernel_executor(self, **kwargs) -> lp.ExecutorBase:
             try:
                 result = code_cache[cache_key]
                 logger.debug("%s: kernel cache hit [key=%s]", self.name, cache_key)
-                return result.executor(self.context)
+                return result
             except KeyError:
                 pass
 
@@ -493,7 +493,7 @@ def get_cached_kernel_executor(self, **kwargs) -> lp.ExecutorBase:
                 NO_CACHE_KERNELS and self.name in NO_CACHE_KERNELS):
             code_cache.store_if_not_present(cache_key, knl)
 
-        return knl.executor(self.context)
+        return knl
 
     @staticmethod
     def _allow_redundant_execution_of_knl_scaling(