builder: run function passes per package

This should result in a small compile time reduction for incremental
builds, somewhere around 5-9%.

This commit, while small, required many previous commits to not regress
binary size. Right now binary size is basically identical with very few
changes in size (the only baremetal program that changed in size did so
with a 4 byte increase).

This commit is one extra step towards doing as much work as possible in
the parallel and cached package build step, out of the serial LTO phase.
Later improvements in this area have this change as a prerequisite.

transform/optimizer.go

@@ -50,16 +50,6 @@ func Optimize(mod llvm.Module, config *compileopts.Config, optLevel, sizeLevel i
 		}
 	}
 
-	// Run function passes for each function.
-	funcPasses := llvm.NewFunctionPassManagerForModule(mod)
-	defer funcPasses.Dispose()
-	builder.PopulateFunc(funcPasses)
-	funcPasses.InitializeFunc()
-	for fn := mod.FirstFunction(); !fn.IsNil(); fn = llvm.NextFunction(fn) {
-		funcPasses.RunFunc(fn)
-	}
-	funcPasses.FinalizeFunc()
-
 	if optLevel > 0 {
 		// Run some preparatory passes for the Go optimizer.
 		goPasses := llvm.NewPassManager()
@@ -164,6 +154,10 @@ func Optimize(mod llvm.Module, config *compileopts.Config, optLevel, sizeLevel i
 
 	// Run function passes again, because without it, llvm.coro.size.i32()
 	// doesn't get lowered.
+	funcPasses := llvm.NewFunctionPassManagerForModule(mod)
+	defer funcPasses.Dispose()
+	builder.PopulateFunc(funcPasses)
+	funcPasses.InitializeFunc()
 	for fn := mod.FirstFunction(); !fn.IsNil(); fn = llvm.NextFunction(fn) {
 		funcPasses.RunFunc(fn)
 	}