[inductor][choices] rename get_mm_configs to get_template_configs (#162293)

# why

- eventually we want all templates to go through this
- we're exposing this through diode as a sort of interface/API
- avoid later renaming

# what

- rename get_mm_configs to get_template_configs
- rename _finalize_mm_configs to _finalize_template_configs

# testing

- lintrunner
- ci

Differential Revision: [D81820641](https://our.internmc.facebook.com/intern/diff/D81820641)
Pull Request resolved: #162293
Approved by: https://github.com/eellison
ghstack dependencies: #161351, #161350

Author: coconutruben

torch/_inductor/choices.py

@@ -106,7 +106,7 @@ def get_flex_decode_configs(
         flex_heuristics = self.get_config_heuristics(device_type)
         return flex_heuristics.get_flex_decode_configs(head_dim, dtype)
 
-    def _finalize_mm_configs(
+    def _finalize_template_configs(
         self,
         template_choices: dict[str, Generator[KernelTemplateChoice, None, None]],
         kernel_inputs: KernelInputs,
@@ -148,12 +148,12 @@ def get_ktc(
         """
         Utility to get the KernelTemplateChoice generator for a specific input.
 
-        This is a per template/op call, whereas get_mm_configs is an op wide call (all templates).
+        This is a per template/op call, whereas get_template_configs is an op wide call (all templates).
         Consider when overriding/using at which level you need to make decisions
         """
         # Extract device_type from kernel_inputs
         device_type = kernel_inputs.device_type
-        assert device_type is not None, "get_mm_configs requires a valid device type"
+        assert device_type is not None, "get_ktc requires a valid device type"
         # Extract template_name from the template object
         template_name = template.uid
 
@@ -233,7 +233,7 @@ def _need_to_fix_layout(
             not isinstance(ktc.template, ExternKernelChoice) for ktc in adjusted_choices
         )
 
-    def get_mm_configs(
+    def get_template_configs(
         self,
         kernel_inputs: KernelInputs,
         templates: list[Union[KernelTemplate, ExternKernelChoice]],
@@ -270,7 +270,7 @@ def get_mm_configs(
             )
 
         # Second pass: Adjust the template choices
-        adjusted_choices = self._finalize_mm_configs(
+        adjusted_choices = self._finalize_template_configs(
             template_choices,
             kernel_inputs,
             templates,

torch/_inductor/kernel/bmm.py

@@ -214,7 +214,7 @@ def may_require_contiguous(t, meta_t):
 
     # Single unified call for all templates
     choices.extend(
-        V.choices.get_mm_configs(
+        V.choices.get_template_configs(
             kernel_inputs,
             templates_to_use,
             name,
@@ -290,6 +290,8 @@ def tuned_baddbmm(inp, mat1, mat2, *, alpha=1, beta=1, layout=None):
         templates_to_use.append(bmm_template)
 
     # Single unified call for all templates
-    choices.extend(V.choices.get_mm_configs(kernel_inputs, templates_to_use, name))
+    choices.extend(
+        V.choices.get_template_configs(kernel_inputs, templates_to_use, name)
+    )
 
     return autotune_select_algorithm(name, choices, kernel_inputs.nodes(), layout)

torch/_inductor/kernel/mm.py

@@ -770,7 +770,9 @@ def tuned_mm(mat1, mat2, *, layout=None):
         templates_to_use.append(mm_contiguous_subgraph_template)
 
     # Single unified call for all non-autoheuristic templates
-    choices.extend(V.choices.get_mm_configs(kernel_inputs, templates_to_use, "mm"))
+    choices.extend(
+        V.choices.get_template_configs(kernel_inputs, templates_to_use, "mm")
+    )
 
     if (
         is_nonzero
@@ -805,7 +807,7 @@ def tuned_mm(mat1, mat2, *, layout=None):
             always_included.append("extern_mm")
         num_choices_before_extra_configs = len(choices)
         choices.extend(
-            V.choices.get_mm_configs(
+            V.choices.get_template_configs(
                 # TODO(coconutruben): remove once we deprecate ah
                 # mm-extra is a hack to keep the ah functionality alive
                 # while we transition to the unified kwargs retrieval
@@ -898,7 +900,9 @@ def tuned_int_mm(mat1, mat2, *, layout=None):
         templates_to_use.append(mm_template)
 
     # Single unified call for all templates
-    choices.extend(V.choices.get_mm_configs(kernel_inputs, templates_to_use, name))
+    choices.extend(
+        V.choices.get_template_configs(kernel_inputs, templates_to_use, name)
+    )
 
     if use_cutlass and _use_cutlass_for_op(name):
         CUTLASS3xGemmTemplate.add_cutlass_gemm_choices(
@@ -944,7 +948,7 @@ def tuned_addmm(inp, mat1, mat2, *, alpha=1, beta=1, layout=None):
             [inp, mat1, mat2], scalars=dict(alpha=alpha, beta=beta)
         )
         choices.extend(
-            V.choices.get_mm_configs(
+            V.choices.get_template_configs(
                 kernel_inputs,
                 [aten_addmm],
                 name,
@@ -966,7 +970,9 @@ def tuned_addmm(inp, mat1, mat2, *, alpha=1, beta=1, layout=None):
         templates_to_use.append(addmm_contiguous_subgraph_template)
 
     # Single unified call for all templates
-    choices.extend(V.choices.get_mm_configs(kernel_inputs, templates_to_use, name))
+    choices.extend(
+        V.choices.get_template_configs(kernel_inputs, templates_to_use, name)
+    )
 
     if (
         is_nonzero
@@ -1153,7 +1159,7 @@ def tuned_scaled_mm(
 
     # Single unified call for all templates
     choices.extend(
-        V.choices.get_mm_configs(
+        V.choices.get_template_configs(
             kernel_inputs,
             templates_to_use,
             name,

torch/_inductor/kernel/mm_plus_mm.py

@@ -167,7 +167,7 @@ def tuned_mm_plus_mm(mat1, mat2, mat3, mat4, *, layout=None):
 
     # Single unified call for all templates
     choices.extend(
-        V.choices.get_mm_configs(kernel_inputs, templates_to_use, "mm_plus_mm")
+        V.choices.get_template_configs(kernel_inputs, templates_to_use, "mm_plus_mm")
     )
 
     return autotune_select_algorithm(