Optimization refactoring:

– -Of should imply -finline
– certain optimizations should be controllable
– unused global symbols should be removed even if we're not optimizing

docs/api/command-line.md

@@ -139,35 +139,68 @@ Use a software stack for stack variables.
 
 ## Optimization options
 
-* `-O0` – Disable all optimizations.
+* `-O0` – Disable all optimizations except unused global symbol removal.
 
 * `-O`, `-O2` ... `-O8` – Optimize code, various levels. For most code, anything above `-O4` doesn't improve it anymore. 
 
-* `-O9` – Optimize code using superoptimizer (experimental). Computationally expensive, decent results.
+* `-O9` – Optimize code using superoptimizer (experimental). Computationally very expensive, decent results.
 
 * `-finline`, `-fno-inline` – Whether should inline functions automatically.
 See the [documentation about inlining](../abi/inlining.md). Computationally easy, can give decent gains.
 `.ini` equivalent: `inline`.
+Default: no
 
 * `-fipo`, `-fno-ipo` – Whether should perform interprocedural optimization.
 It enables certain optimization similar to what inlining would enable, but without actual inlining.
 `.ini` equivalent: `ipo`.
+Default: no. 
+
+* `-fregister-variables`, `-fno-register-variables` – Whether should allow moving local variables into CPU registers.
+Default: yes. 
 
 * `-foptimize-stdlib`, `-fno-optimize-stdlib` – 
 Whether should replace some standard library calls with constant parameters with more efficient variants.
 Currently affects `putstrz` and `strzlen`, but may affect more functions in the future.
 `.ini` equivalent: `optimize_stdlib`.
+Default: no.
+
+* `-ffunction-fallthrough`, `-fno-function-fallthrough` – 
+Whether should replace a tail call by simply putting one function after another.
+`.ini` equivalent: `function_fallthrough`.
+Default: yes. 
+
+* `-ffunction-deduplication`, `-fno-function-deduplication` – 
+Whether identical functions should be merged into one function.
+`.ini` equivalent: `function_deduplication`.
+Default: yes. 
+
+* `-fsubroutine-extraction`, `-fno-subroutine-extraction` – 
+Whether identical fragments of functions should be extracted into subroutines.
+`.ini` equivalent: `subroutine_extraction`.
+Default: no. 
 
 * `-Os`, `--size` – Optimize for size, sacrificing some speed (experimental).
+Also enables `-fcode-deduplication`.
 
 * `-Of`, `--fast` – Optimize for speed, even if it increases the size a bit (experimental).
+Also enables `-finline`. 
 
 * `-Ob`, `--blast-processing` – Optimize for speed, even if it increases the size a lot (experimental).
-Enables `-finline` automatically. 
+Also enables `-finline`. 
+
+* `-Og`, `--optimize-debugging` – Disables optimizations that make debugging harder.
+Sets 
+`-fno-optimize-stdlib`,
+`-fno-variable-overlap`,
+`-fno-register-variables`,
+`-fno-function-fallthrough`,
+`-fno-function-deduplication`,
+`-fno-subroutine-extraction`
+ automatically. 
 
 * `-fdangerous-optimizations` – Use dangerous optimizations (experimental).
 Dangerous optimizations are more likely to result in broken code.
-Enables `-fipo` automatically.
+Also enables `-fipo` and `-foptimize-stdlib` automatically.
 `.ini` equivalent: `dangerous_optimizations`. 
 
 Note: for compatibility with versions 0.3.0 and earlier,

docs/api/custom-platform.md

@@ -58,7 +58,7 @@ Default: the same as `encoding`.
     * `prevent_jmp_indirect_bug` – whether the compiler should try to avoid the indirect JMP bug, 
     default is `false` on 65C02-compatible or non-6502 processors and `true` elsewhere
     
-    * `compact_dispatch_params` – whether parameter values in return dispatch statements may overlap other objects, default is `true`  
+    * `compact_dispatch_params` – whether parameter values in return dispatch statements may overlap other objects, default is `true`.  
     This may cause problems if the parameter table is stored next to a hardware register that has side effects when reading.
     
     * `lunix` – generate relocatable code for LUnix/LNG, default is `false`
@@ -71,6 +71,12 @@ Default: the same as `encoding`.
     
     * `ipo` - enable interprocedural optimization, default is `false`.
     
+    * `function_fallthrough` – whether should replace a tail call by simply putting one function after another, default is `true`.
+
+    * `function_deduplication` – whether identical functions should be merged into one function, default is `true`.
+
+    * `subroutine_extraction` – whether identical fragments of functions should be extracted into subroutines, default is `false`.    
+    
     * `lenient_encoding` - allow for automatic substitution of invalid characters in string literals using the default encodings, default is `false`.
     
     * `use_shadow_registers_for_irq` – use Z80 shadow registers in interrupt routines, default is `true` for Z80 and `false` otherwise

src/main/scala/millfork/CompilationOptions.scala

@@ -218,13 +218,13 @@ object Cpu extends Enumeration {
 
   import CompilationFlag._
 
-  private val mosAlwaysDefaultFlags = Set(
+  private val alwaysDefaultFlags = Set(
-    VariableOverlap, CompactReturnDispatchParams
+    VariableOverlap, CompactReturnDispatchParams, FunctionFallthrough, RegisterVariables, FunctionDeduplication
   )
 
-  private val i80AlwaysDefaultFlags = Set(
+  private val mosAlwaysDefaultFlags = alwaysDefaultFlags
-    VariableOverlap, CompactReturnDispatchParams
+
-  )
+  private val i80AlwaysDefaultFlags = alwaysDefaultFlags
 
   def defaultFlags(x: Cpu.Value): Set[CompilationFlag.Value] = x match {
     case StrictMos =>
@@ -316,8 +316,9 @@ object CompilationFlag extends Enumeration {
   UseIntelSyntaxForInput,
   UseIntelSyntaxForOutput,
   // optimization options:
-  OptimizeForSize, OptimizeForSpeed, OptimizeForSonicSpeed,
+  OptimizeForSize, OptimizeForSpeed, OptimizeForSonicSpeed, OptimizeForDebugging,
   DangerousOptimizations, InlineFunctions, InterproceduralOptimization,
+  FunctionFallthrough, RegisterVariables, FunctionDeduplication, SubroutineExtraction,
   OptimizeStdlib,
   // memory allocation options
   VariableOverlap, CompactReturnDispatchParams, LUnixRelocatableCode,
@@ -356,6 +357,9 @@ object CompilationFlag extends Enumeration {
     "input_intel_syntax" -> UseIntelSyntaxForInput,
     "ipo" -> InterproceduralOptimization,
     "optimize_stdlib" -> OptimizeStdlib,
+    "function_fallthrough" -> FunctionFallthrough,
+    "function_deduplication" -> FunctionDeduplication,
+    "subroutine_extraction" -> SubroutineExtraction,
     "inline" -> InlineFunctions,
     "dangerous_optimizations" -> DangerousOptimizations,
     "decimal_mode" -> DecimalMode,

src/main/scala/millfork/Context.scala

@@ -0,0 +1,55 @@
+package millfork
+
+import millfork.error.Logger
+
+/**
+  * @author Karol Stasiak
+  */
+
+case class Context(errorReporting: Logger,
+                   inputFileNames: List[String],
+                   outputFileName: Option[String] = None,
+                   runFileName: Option[String] = None,
+                   optimizationLevel: Option[Int] = None,
+                   zpRegisterSize: Option[Int] = None,
+                   platform: Option[String] = None,
+                   outputAssembly: Boolean = false,
+                   outputLabels: Boolean = false,
+                   includePath: List[String] = Nil,
+                   flags: Map[CompilationFlag.Value, Boolean] = Map(),
+                   features: Map[String, Long] = Map(),
+                   verbosity: Option[Int] = None) {
+  def changeFlag(f: CompilationFlag.Value, b: Boolean): Context = {
+    if (flags.contains(f)) {
+      if (flags(f) != b) {
+        errorReporting.error("Conflicting flags")
+      }
+      this
+    } else {
+      copy(flags = this.flags + (f -> b))
+    }
+  }
+
+  def filloutFlags(): Context = {
+    var addons = Map[CompilationFlag.Value, Boolean]()
+    if (flags.contains(CompilationFlag.OptimizeForSpeed)) {
+      addons += CompilationFlag.InlineFunctions -> true
+    }
+    if (flags.contains(CompilationFlag.OptimizeForSize)) {
+      addons += CompilationFlag.SubroutineExtraction -> true
+    }
+    if (flags.contains(CompilationFlag.DangerousOptimizations)) {
+      addons += CompilationFlag.InterproceduralOptimization -> true
+      addons += CompilationFlag.OptimizeStdlib -> true
+    }
+    if (flags.contains(CompilationFlag.OptimizeForDebugging)) {
+      addons += CompilationFlag.VariableOverlap -> false
+      addons += CompilationFlag.RegisterVariables -> false
+      addons += CompilationFlag.FunctionDeduplication -> false
+      addons += CompilationFlag.SubroutineExtraction -> false
+      addons += CompilationFlag.FunctionFallthrough -> false
+      addons += CompilationFlag.OptimizeStdlib -> false
+    }
+    copy(flags = flags ++ addons.filterKeys(k => !flags.contains(k)))
+  }
+}

src/main/scala/millfork/Main.scala

@@ -18,34 +18,7 @@ import millfork.node.StandardCallGraph
 import millfork.output._
 import millfork.parser.{MosSourceLoadingQueue, ZSourceLoadingQueue}
 
-/**
-  * @author Karol Stasiak
-  */
 
-case class Context(errorReporting: Logger,
-                   inputFileNames: List[String],
-                   outputFileName: Option[String] = None,
-                   runFileName: Option[String] = None,
-                   optimizationLevel: Option[Int] = None,
-                   zpRegisterSize: Option[Int] = None,
-                   platform: Option[String] = None,
-                   outputAssembly: Boolean = false,
-                   outputLabels: Boolean = false,
-                   includePath: List[String] = Nil,
-                   flags: Map[CompilationFlag.Value, Boolean] = Map(),
-                   features: Map[String, Long] = Map(),
-                   verbosity: Option[Int] = None) {
-  def changeFlag(f: CompilationFlag.Value, b: Boolean): Context = {
-    if (flags.contains(f)) {
-      if (flags(f) != b) {
-        errorReporting.error("Conflicting flags")
-      }
-      this
-    } else {
-      copy(flags = this.flags + (f -> b))
-    }
-  }
-}
 
 object Main {
 
@@ -77,7 +50,7 @@ object Main {
     errorReporting.trace("This program comes with ABSOLUTELY NO WARRANTY.")
     errorReporting.trace("This is free software, and you are welcome to redistribute it under certain conditions")
     errorReporting.trace("You should have received a copy of the GNU General Public License along with this program. If not, see https://www.gnu.org/licenses/")
-    val c = fixMissingIncludePath(c0)
+    val c = fixMissingIncludePath(c0).filloutFlags()
     if (c.includePath.isEmpty) {
       errorReporting.warn("Failed to detect the default include directory, consider using the -I option")
     }
@@ -213,7 +186,7 @@ object Main {
     val program = if (optLevel > 0) {
       OptimizationPresets.NodeOpt.foldLeft(unoptimized)((p, opt) => p.applyNodeOptimization(opt, options))
     } else {
-      unoptimized
+      OptimizationPresets.NodeOpt0.foldLeft(unoptimized)((p, opt) => p.applyNodeOptimization(opt, options))
     }
     val callGraph = new StandardCallGraph(program, options.log)
 
@@ -265,7 +238,7 @@ object Main {
     val program = if (optLevel > 0) {
       OptimizationPresets.NodeOpt.foldLeft(unoptimized)((p, opt) => p.applyNodeOptimization(opt, options))
     } else {
-      unoptimized
+      OptimizationPresets.NodeOpt0.foldLeft(unoptimized)((p, opt) => p.applyNodeOptimization(opt, options))
     }
     val callGraph = new StandardCallGraph(program, options.log)
 
@@ -430,10 +403,10 @@ object Main {
     }.description("Whether hardware decimal mode should be used (6502 only).")
     boolean("-fvariable-overlap", "-fno-variable-overlap").action { (c, v) =>
       c.changeFlag(CompilationFlag.VariableOverlap, v)
-    }.description("Whether variables should overlap if their scopes do not intersect.")
+    }.description("Whether variables should overlap if their scopes do not intersect. Enabled by default.")
     boolean("-fcompact-dispatch-params", "-fno-compact-dispatch-params").action { (c, v) =>
       c.changeFlag(CompilationFlag.CompactReturnDispatchParams, v)
-    }.description("Whether parameter values in return dispatch statements may overlap other objects.")
+    }.description("Whether parameter values in return dispatch statements may overlap other objects. Enabled by default.")
     boolean("-fbounds-checking", "-fno-bounds-checking").action { (c, v) =>
       c.changeFlag(CompilationFlag.VariableOverlap, v)
     }.description("Whether should insert bounds checking on array access.")
@@ -506,28 +479,42 @@ object Main {
     boolean("-foptimize-stdlib", "-fno-optimize-stdlib").action { (c, v) =>
       c.changeFlag(CompilationFlag.OptimizeStdlib, v)
     }.description("Optimize standard library calls.")
+    boolean("-fsubroutine-extraction", "-fno-subroutine-extraction").action { (c, v) =>
+      c.changeFlag(CompilationFlag.SubroutineExtraction, v)
+    }.description("Extract identical code fragments into subroutines.")
+    boolean("-ffunction-fallthrough", "-fno-function-fallthrough").action { (c, v) =>
+      c.changeFlag(CompilationFlag.FunctionFallthrough, v)
+    }.description("Replace tail calls by simply putting one function after another. Enabled by default.")
+    boolean("-ffunction-deduplication", "-fno-function-deduplication").action { (c, v) =>
+      c.changeFlag(CompilationFlag.FunctionDeduplication, v)
+    }.description("Merge identical functions into one function. Enabled by default.")
+    boolean("-fregister-variables", "-fno-register-variables").action { (c, v) =>
+      c.changeFlag(CompilationFlag.RegisterVariables, v)
+    }.description("Allow moving local variables into CPU registers. Enabled by default.")
     flag("-Os", "--size").action { c =>
       c.changeFlag(CompilationFlag.OptimizeForSize, true).
         changeFlag(CompilationFlag.OptimizeForSpeed, false).
         changeFlag(CompilationFlag.OptimizeForSonicSpeed, false)
-    }.description("Prefer smaller code even if it is slightly slower (experimental).")
+    }.description("Prefer smaller code even if it is slightly slower (experimental). Implies -fsubroutine-extraction.")
     flag("-Of", "--fast").action { c =>
       c.changeFlag(CompilationFlag.OptimizeForSize, false).
         changeFlag(CompilationFlag.OptimizeForSpeed, true).
         changeFlag(CompilationFlag.OptimizeForSonicSpeed, false)
-    }.description("Prefer faster code even if it is slightly bigger (experimental).")
+    }.description("Prefer faster code even if it is slightly bigger (experimental). Implies -finline.")
     flag("-Ob", "--blast-processing").action { c =>
       c.changeFlag(CompilationFlag.OptimizeForSize, false).
         changeFlag(CompilationFlag.OptimizeForSpeed, true).
-        changeFlag(CompilationFlag.OptimizeForSonicSpeed, true).
+        changeFlag(CompilationFlag.OptimizeForSonicSpeed, true)
-        changeFlag(CompilationFlag.InlineFunctions, true)
     }.description("Prefer faster code even if it is much bigger (experimental). Implies -finline.")
     flag("--dangerous-optimizations").action { c =>
       c.changeFlag(CompilationFlag.DangerousOptimizations, true)
     }.description("Use dangerous optimizations (experimental).").hidden()
     boolean("-fdangerous-optimizations", "-fno-dangerous-optimizations").action { (c, v) =>
       c.changeFlag(CompilationFlag.DangerousOptimizations, v)
-    }.description("Use dangerous optimizations (experimental). Implies -fipo.")
+    }.description("Use dangerous optimizations (experimental). Implies -fipo and -foptimize-stdlib.")
+    flag("-Og", "--optimize-debugging").action { c =>
+      c.changeFlag(CompilationFlag.OptimizeForDebugging, true)
+    }.description("Disable optimizations that make debugging harder (experimental).")
 
     fluff("", "Warning options:", "")
 

src/main/scala/millfork/OptimizationPresets.scala

@@ -15,6 +15,10 @@ object OptimizationPresets {
     UnusedLocalVariables,
     UnusedGlobalVariables,
   )
+  val NodeOpt0 = List(
+    UnusedFunctions,
+    UnusedGlobalVariables,
+  )
   val AssOpt: List[AssemblyOptimization[AssemblyLine]] = List[AssemblyOptimization[AssemblyLine]](
     UnusedLabelRemoval,
     AlwaysGoodOptimizations.NonetAddition,

src/main/scala/millfork/assembly/AssemblyOptimization.scala

@@ -1,6 +1,6 @@
 package millfork.assembly
 
-import millfork.CompilationOptions
+import millfork.{CompilationFlag, CompilationOptions}
 import millfork.compiler.LabelGenerator
 import millfork.env.{NormalFunction, ThingInMemory}
 import millfork.error.Logger
@@ -23,4 +23,6 @@ trait AssemblyOptimization[T <: AbstractCode] {
   def name: String
 
   def optimize(f: NormalFunction, code: List[T], context: OptimizationContext): List[T]
+
+  def requiredFlags: Set[CompilationFlag.Value] = Set.empty
 }

src/main/scala/millfork/assembly/mos/opt/VariableToRegisterOptimization.scala

@@ -17,6 +17,8 @@ import scala.util.control.TailCalls.{TailRec, done, tailcall}
   */
 object VariableToRegisterOptimization extends AssemblyOptimization[AssemblyLine] {
 
+  override def requiredFlags: Set[CompilationFlag.Value] = Set(CompilationFlag.RegisterVariables)
+
   object CyclesAndBytes {
     val Zero = CyclesAndBytes(0, 0)
   }

src/main/scala/millfork/assembly/z80/opt/ByteVariableToRegisterOptimization.scala

@@ -13,6 +13,8 @@ import scala.collection.mutable.ListBuffer
   */
 object ByteVariableToRegisterOptimization extends AssemblyOptimization[ZLine] {
 
+  override def requiredFlags: Set[CompilationFlag.Value] = Set(CompilationFlag.RegisterVariables)
+
   override def name = "Allocating variables to single registers"
 
   object CyclesAndBytes {

src/main/scala/millfork/assembly/z80/opt/WordVariableToRegisterOptimization.scala

@@ -14,6 +14,8 @@ import scala.collection.mutable.ListBuffer
   */
 object WordVariableToRegisterOptimization extends AssemblyOptimization[ZLine] {
 
+  override def requiredFlags: Set[CompilationFlag.Value] = Set(CompilationFlag.RegisterVariables)
+
   override def name = "Allocating variables to register pairs"
 
   object CyclesAndBytes {

src/main/scala/millfork/output/AbstractAssembler.scala

@@ -163,7 +163,7 @@ abstract class AbstractAssembler[T <: AbstractCode](private val program: Program
 
   def deduplicate(options: CompilationOptions, compiledFunctions: mutable.Map[String, CompiledFunction[T]]): Unit
 
-  def assemble(callGraph: CallGraph, optimizations: Seq[AssemblyOptimization[T]], options: CompilationOptions): AssemblerOutput = {
+  def assemble(callGraph: CallGraph, unfilteredOptimizations: Seq[AssemblyOptimization[T]], options: CompilationOptions): AssemblerOutput = {
     mem.programName = options.outputFileName.getOrElse("MILLFORK")
     val platform = options.platform
     val variableAllocators = platform.variableAllocators
@@ -174,6 +174,8 @@ abstract class AbstractAssembler[T <: AbstractCode](private val program: Program
       zpOccupied(i + 1) = false
     }
 
+    val optimizations = unfilteredOptimizations.filter(_.requiredFlags.forall(options.flag))
+
     val assembly = mutable.ArrayBuffer[String]()
 
     val inliningResult = inliningCalculator.calculate(

src/main/scala/millfork/output/Deduplicate.scala

@@ -13,16 +13,20 @@ import scala.collection.mutable.ListBuffer
 abstract class Deduplicate[T <: AbstractCode](env: Environment, options: CompilationOptions) {
 
   def apply(compiledFunctions: mutable.Map[String, CompiledFunction[T]]): Unit = {
-    if (options.flag(CompilationFlag.OptimizeForSize)) {
+    if (options.flag(CompilationFlag.SubroutineExtraction)) {
       runStage(compiledFunctions, extractCommonCode)
     }
+    if (options.flag(CompilationFlag.FunctionDeduplication)) {
       runStage(compiledFunctions, deduplicateIdenticalFunctions)
+    }
+    if (options.flag(CompilationFlag.FunctionFallthrough)) {
       runStage(compiledFunctions, eliminateTailJumps)
       runStage(compiledFunctions, eliminateTailJumps)
       runStage(compiledFunctions, eliminateTailJumps)
       runStage(compiledFunctions, eliminateRemainingTrivialTailJumps)
       runStage(compiledFunctions, eliminateRemainingTrivialTailJumps)
       runStage(compiledFunctions, eliminateRemainingTrivialTailJumps)
+    }
     fixDoubleRedirects(compiledFunctions)
 //    println(compiledFunctions.map {
 //      case (k, v) => k + " " + (v match {

src/test/scala/millfork/test/emu/EmuRun.scala

@@ -144,6 +144,7 @@ class EmuRun(cpu: millfork.Cpu.Value, nodeOptimizations: List[NodeOptimization],
       CompilationFlag.EmitNative65816Opcodes -> (platform.cpu == millfork.Cpu.Sixteen && native16),
       CompilationFlag.Emit65CE02Opcodes -> (platform.cpu == millfork.Cpu.CE02),
       CompilationFlag.EmitHudsonOpcodes -> (platform.cpu == millfork.Cpu.HuC6280),
+      CompilationFlag.SubroutineExtraction -> optimizeForSize,
       CompilationFlag.OptimizeForSize -> optimizeForSize,
       CompilationFlag.OptimizeForSpeed -> blastProcessing,
       CompilationFlag.OptimizeForSonicSpeed -> blastProcessing