Support outer reduction scheduler with SOL autotuning

csrc/python_frontend/python_bindings.cpp

@@ -30,6 +30,7 @@
 #include <scheduler/utils.h>
 #include <torch/csrc/jit/python/pybind_utils.h>
 #include <transform_replay.h>
+#include <utils.h>
 #include <iostream>
 #include <optional>
 #include <tuple>
@@ -761,6 +762,32 @@ void initNvFuserPythonBindings(PyObject* module) {
 
   nvfuser.def("clone", clone);
 
+  nvfuser.def(
+      "get_registers_per_thread",
+      getRegPerThreadGivenThreadsPerSM,
+      py::arg("threads_per_sm"),
+      R"(
+  Estimate the number of registers per thread using cuda occupancy API.
+
+  Parameters
+  ----------
+  threads_per_sm : int
+      The number of threads per SM.
+  )");
+
+  nvfuser.def(
+      "get_threads_per_sm",
+      getThreadsPerSMGivenRegPerThread,
+      py::arg("reg_per_thread"),
+      R"(
+  Get number of threads per sm using cuda occupancy API.
+
+  Parameters
+  ----------
+  reg_per_thread : int
+      The number of registers per thread.
+  )");
+
   //! DataTypes supported by nvFuser in the FusionDefinition
   py::enum_<PrimDataType>(nvfuser, "DataType")
       .value("Double", DataType::Double)

doc/dev/python_scheduling/autotune_inner_reduction.py

@@ -254,7 +254,7 @@ def inner_fn(num_inputs):
             assert False
 
     # A decorator to create a reduction fusion given some input arguments.
-    def create_fusion_func(self, inputs):
+    def create_fusion_func(self):
         def sum_fusion(fd: FusionDefinition) -> None:
             T0 = fd.define_tensor(
                 shape=[-1, -1],