Fix "set but not used" warnings/errors (halide#6683)

* Fix "set but not used" warnings/errors

Apparently XCode 13.3 has smarter warnings about unused code and emits warnings/errors for these, so let's clean them up.

* Also fix missing `ssize_t` usage

python_bindings/src/PyBuffer.cpp

@@ -313,10 +313,10 @@ void define_buffer(py::module &m) {
 
                 const int d = b.dimensions();
                 const int bytes = b.type().bytes();
-                std::vector<ssize_t> shape, strides;
+                std::vector<Py_ssize_t> shape, strides;
                 for (int i = 0; i < d; i++) {
-                    shape.push_back((ssize_t)b.raw_buffer()->dim[i].extent);
-                    strides.push_back((ssize_t)(b.raw_buffer()->dim[i].stride * bytes));
+                    shape.push_back((Py_ssize_t)b.raw_buffer()->dim[i].extent);
+                    strides.push_back((Py_ssize_t)(b.raw_buffer()->dim[i].stride * bytes));
                 }
 
                 return py::buffer_info(

src/autoschedulers/adams2019/LoopNest.cpp

@@ -729,7 +729,6 @@ void LoopNest::compute_features(const FunctionDAG &dag,
                 int64_t footprint = e->producer->bytes_per_point;
                 int64_t compute_footprint = footprint;
                 int64_t store_footprint = footprint;
-                int64_t task_footprint = footprint;
                 int64_t line_footprint = 1;
                 int64_t compute_line_footprint = 1;
                 int64_t store_line_footprint = 1;
@@ -890,7 +889,6 @@ void LoopNest::compute_features(const FunctionDAG &dag,
                     footprint *= extent;
                     compute_footprint *= compute_extent;
                     store_footprint *= store_extent;
-                    task_footprint *= task_extent;
 
                     bool dense = ((e->producer->is_input && i == 0) ||
                                   (site.produce != nullptr && i == site.produce->vector_dim));
@@ -1295,13 +1293,11 @@ void LoopNest::compute_here(const FunctionDAG::Node *f, bool tileable, int v) {
         size_t loop_dim = f->stages[s].loop.size();
         node->size.resize(loop_dim);
 
-        int64_t total_extent = 1;
         int64_t vector_size = 1;
         for (size_t i = 0; i < loop_dim; i++) {
             const auto &l = bounds->loops(s, i);
             // Initialize the loop nest
             node->size[i] = l.extent();
-            total_extent *= node->size[i];
 
             // Use the first loop iteration to represent the inner
             // loop. We'll shift it to a later one once we decide

src/autoschedulers/adams2019/State.cpp

@@ -594,7 +594,6 @@ void State::apply_schedule(const FunctionDAG &dag, const MachineParams &params)
         // Do all the reorders and pick which vars to
         // parallelize.
         vector<VarOrRVar> vars;
-        int64_t parallel_tasks = 1;
         vector<VarOrRVar> parallel_vars;
         bool any_parallel_vars = false, any_parallel_rvars = false;
         for (auto it = p.second->vars.rbegin(); it != p.second->vars.rend(); it++) {
@@ -606,7 +605,6 @@ void State::apply_schedule(const FunctionDAG &dag, const MachineParams &params)
             }
             any_parallel_rvars |= it->var.is_rvar;
             any_parallel_vars |= !it->var.is_rvar;
-            parallel_tasks *= it->extent;
             parallel_vars.push_back(it->var);
         }
 

src/autoschedulers/li2018/GradientAutoscheduler.cpp

@@ -606,10 +606,6 @@ void apply_schedule(const MachineParams &params,
         for (const ReductionVariable &r : reduction_vars) {
             rvars.emplace_back(r.var);
         }
-        int rdomain_size = 1;
-        for (int b : rvar_bounds) {
-            rdomain_size *= b;
-        }
         // Define the thresholds for the pure domain.
         // For CPU we want at least params.parallelism number of elements
         // to launch threads. For GPU we want to launch at least 64 GPU blocks.