Remove 'enable_constant_folding' setting and move into optimize method

Author: lachlansneff

lib/codegen/meta-python/base/settings.py

@@ -103,10 +103,6 @@
         """Enable the use of atomic instructions""",
         default=True)
 
-enable_constant_folding = BoolSetting(
-        """Enable the constant folding pass. This may be resource intensive""",
-        default=False)
-
 #
 # Settings specific to the `baldrdash` calling convention.
 #

lib/codegen/src/context.rs

@@ -113,6 +113,16 @@ impl Context {
         Ok(())
     }
 
+    /// Optimize the function with available optimizations.
+    /// 
+    /// Since this can be resource intensive (and code-size inflating),
+    /// it is separated from `Context::compile` to allow DCE to remove it
+    /// if it's not used.
+    pub fn optimize(&mut self, isa: &TargetIsa) -> CodegenResult<()> {
+        self.verify_if(isa)?;
+        self.fold_constants(isa)
+    }
+
     /// Compile the function.
     ///
     /// Run the function through all the passes necessary to generate code for the target ISA
@@ -124,10 +134,6 @@ impl Context {
         let _tt = timing::compile();
         self.verify_if(isa)?;
 
-        if isa.flags().enable_constant_folding() {
-            self.fold_constants(isa)?;
-        }
-
         self.compute_cfg();
         if isa.flags().opt_level() != OptLevel::Fastest {
             self.preopt(isa)?;

lib/codegen/src/settings.rs

@@ -380,7 +380,6 @@ mod tests {
              enable_nan_canonicalization = false\n\
              enable_simd = true\n\
              enable_atomics = true\n\
-             enable_constant_folding = false\n\
              baldrdash_prologue_words = 0\n\
              allones_funcaddrs = false\n\
              probestack_enabled = true\n\