Add fast handlers

fast_call.wat

@@ -0,0 +1,10 @@
+(module
+  (import "wizeng" "puti" (func $puti (param i32)))
+  (func $f (result i32)
+	i32.const 10)
+  (func (export "main") (result i32)
+        call $f
+	call $puti
+	i32.const 0
+  )
+)

fast_call2.wat

@@ -0,0 +1,7 @@
+(module
+  (func $f (result i32)
+	i32.const 10)
+  (func (export "main") (result i32)
+        call $f
+  )
+)

fast_call_export.wat

@@ -0,0 +1,10 @@
+;; export name holds fast information, we don't modify binary ahead of time
+
+(module
+  (func $fast (export "fast:foo") (result i32)
+    (i32.const 2)
+  )
+  (func (export "main") (result i32)
+    (call $fast)
+  )
+)

fast_call_nop.wat

@@ -0,0 +1,10 @@
+(module
+  (func $f)
+  (func $g)
+  (func (export "main") (result i32)
+  	i64.const 11
+	drop
+        call $g
+	i32.const 0
+  )
+)

fast_call_param.wat

@@ -0,0 +1,18 @@
+(module
+  (import "wizeng" "puti" (func $puti (param i32)))
+  (func $f (param i32) (result i32)
+    local.get 0
+    if (result i32)
+      i32.const 999
+    else
+      i32.const -216
+    end
+  )
+  (func (export "main") (result i32)
+        (call $f (i32.const 1))
+	call $puti
+        (call $f (i32.const 0))
+	call $puti
+	i32.const 0
+  )
+)

slow_call_nop.wat

@@ -0,0 +1,7 @@
+(module
+  (func $f)
+  (func (export "main") (result i32)
+        call $f
+	i32.const 0
+  )
+)

src/engine/BytecodeIterator.v3

@@ -779,6 +779,48 @@ class BytecodeIterator {
 			RESUME_THROW => 		v.visit_RESUME_THROW(read_CONT(), read_TAG(), read_HANDLERS());
 			RESUME_THROW_REF => 		v.visit_RESUME_THROW_REF(read_CONT(), read_HANDLERS());
 			SWITCH => 			v.visit_SWITCH(read_CONT(), read_TAG());
+
+			// replacing CALL with FAST_CALL does not touch the operand so that the original function can still be recovered
+			FAST_CALL0  => v.visit_FAST_CALL(Opcodes.fastCallToIndex(opcode), read_FUNC());
+			FAST_CALL1  => v.visit_FAST_CALL(Opcodes.fastCallToIndex(opcode), read_FUNC());
+			FAST_CALL2  => v.visit_FAST_CALL(Opcodes.fastCallToIndex(opcode), read_FUNC());
+			FAST_CALL3  => v.visit_FAST_CALL(Opcodes.fastCallToIndex(opcode), read_FUNC());
+			FAST_CALL4  => v.visit_FAST_CALL(Opcodes.fastCallToIndex(opcode), read_FUNC());
+			FAST_CALL5  => v.visit_FAST_CALL(Opcodes.fastCallToIndex(opcode), read_FUNC());
+			FAST_CALL6  => v.visit_FAST_CALL(Opcodes.fastCallToIndex(opcode), read_FUNC());
+			FAST_CALL7  => v.visit_FAST_CALL(Opcodes.fastCallToIndex(opcode), read_FUNC());
+			FAST_CALL8  => v.visit_FAST_CALL(Opcodes.fastCallToIndex(opcode), read_FUNC());
+			FAST_CALL9  => v.visit_FAST_CALL(Opcodes.fastCallToIndex(opcode), read_FUNC());
+			FAST_CALL10 => v.visit_FAST_CALL(Opcodes.fastCallToIndex(opcode), read_FUNC());
+			FAST_CALL11 => v.visit_FAST_CALL(Opcodes.fastCallToIndex(opcode), read_FUNC());
+			FAST_CALL12 => v.visit_FAST_CALL(Opcodes.fastCallToIndex(opcode), read_FUNC());
+			FAST_CALL13 => v.visit_FAST_CALL(Opcodes.fastCallToIndex(opcode), read_FUNC());
+			FAST_CALL14 => v.visit_FAST_CALL(Opcodes.fastCallToIndex(opcode), read_FUNC());
+			FAST_CALL15 => v.visit_FAST_CALL(Opcodes.fastCallToIndex(opcode), read_FUNC());
+			FAST_CALL16 => v.visit_FAST_CALL(Opcodes.fastCallToIndex(opcode), read_FUNC());
+			FAST_CALL17 => v.visit_FAST_CALL(Opcodes.fastCallToIndex(opcode), read_FUNC());
+			FAST_CALL18 => v.visit_FAST_CALL(Opcodes.fastCallToIndex(opcode), read_FUNC());
+			FAST_CALL19 => v.visit_FAST_CALL(Opcodes.fastCallToIndex(opcode), read_FUNC());
+			FAST_CALL20 => v.visit_FAST_CALL(Opcodes.fastCallToIndex(opcode), read_FUNC());
+			FAST_CALL21 => v.visit_FAST_CALL(Opcodes.fastCallToIndex(opcode), read_FUNC());
+			FAST_CALL22 => v.visit_FAST_CALL(Opcodes.fastCallToIndex(opcode), read_FUNC());
+			FAST_CALL23 => v.visit_FAST_CALL(Opcodes.fastCallToIndex(opcode), read_FUNC());
+			FAST_CALL24 => v.visit_FAST_CALL(Opcodes.fastCallToIndex(opcode), read_FUNC());
+			FAST_CALL25 => v.visit_FAST_CALL(Opcodes.fastCallToIndex(opcode), read_FUNC());
+			FAST_CALL26 => v.visit_FAST_CALL(Opcodes.fastCallToIndex(opcode), read_FUNC());
+			FAST_CALL27 => v.visit_FAST_CALL(Opcodes.fastCallToIndex(opcode), read_FUNC());
+			FAST_CALL28 => v.visit_FAST_CALL(Opcodes.fastCallToIndex(opcode), read_FUNC());
+			FAST_CALL29 => v.visit_FAST_CALL(Opcodes.fastCallToIndex(opcode), read_FUNC());
+			FAST_CALL30 => v.visit_FAST_CALL(Opcodes.fastCallToIndex(opcode), read_FUNC());
+			FAST_CALL31 => v.visit_FAST_CALL(Opcodes.fastCallToIndex(opcode), read_FUNC());
+			FAST_CALL32 => v.visit_FAST_CALL(Opcodes.fastCallToIndex(opcode), read_FUNC());
+			FAST_CALL33 => v.visit_FAST_CALL(Opcodes.fastCallToIndex(opcode), read_FUNC());
+			FAST_CALL34 => v.visit_FAST_CALL(Opcodes.fastCallToIndex(opcode), read_FUNC());
+			FAST_CALL35 => v.visit_FAST_CALL(Opcodes.fastCallToIndex(opcode), read_FUNC());
+			FAST_CALL36 => v.visit_FAST_CALL(Opcodes.fastCallToIndex(opcode), read_FUNC());
+			FAST_CALL37 => v.visit_FAST_CALL(Opcodes.fastCallToIndex(opcode), read_FUNC());
+			FAST_CALL38 => v.visit_FAST_CALL(Opcodes.fastCallToIndex(opcode), read_FUNC());
+			FAST_CALL39 => v.visit_FAST_CALL(Opcodes.fastCallToIndex(opcode), read_FUNC());
 		}
 	}
 	def trace(out: StringBuilder, module: Module, tracer: InstrTracer) {

src/engine/CodeValidator.v3

@@ -420,7 +420,53 @@ class CodeValidator(extensions: Extension.set, limits: Limits, module: Module, e
 					var func = parser.readFuncRef();
 					if (func == null) return;
 					checkSignature(func.sig);
+
+					// fast call: if function is exported with fast name, replace the bytecode with FAST_CALL
+					if (FastIntTuning.useFastFunctions && func.fast_call_idx >= 0) {
+						if (Trace.validation) Trace.OUT.put1("    replacing with FAST_CALL%d\n", func.fast_call_idx);
+						this.func.replaceCall(opcode_pos, func.fast_call_idx);
+					}
 				}
+				FAST_CALL0  => if (FastIntTuning.useFastFunctions) System.error("validation error", "trying to validate FAST_CALL internal opcode");
+				FAST_CALL1  => if (FastIntTuning.useFastFunctions) System.error("validation error", "trying to validate FAST_CALL internal opcode");
+				FAST_CALL2  => if (FastIntTuning.useFastFunctions) System.error("validation error", "trying to validate FAST_CALL internal opcode");
+				FAST_CALL3  => if (FastIntTuning.useFastFunctions) System.error("validation error", "trying to validate FAST_CALL internal opcode");
+				FAST_CALL4  => if (FastIntTuning.useFastFunctions) System.error("validation error", "trying to validate FAST_CALL internal opcode");
+				FAST_CALL5  => if (FastIntTuning.useFastFunctions) System.error("validation error", "trying to validate FAST_CALL internal opcode");
+				FAST_CALL6  => if (FastIntTuning.useFastFunctions) System.error("validation error", "trying to validate FAST_CALL internal opcode");
+				FAST_CALL7  => if (FastIntTuning.useFastFunctions) System.error("validation error", "trying to validate FAST_CALL internal opcode");
+				FAST_CALL8  => if (FastIntTuning.useFastFunctions) System.error("validation error", "trying to validate FAST_CALL internal opcode");
+				FAST_CALL9  => if (FastIntTuning.useFastFunctions) System.error("validation error", "trying to validate FAST_CALL internal opcode");
+				FAST_CALL10 => if (FastIntTuning.useFastFunctions) System.error("validation error", "trying to validate FAST_CALL internal opcode");
+				FAST_CALL11 => if (FastIntTuning.useFastFunctions) System.error("validation error", "trying to validate FAST_CALL internal opcode");
+				FAST_CALL12 => if (FastIntTuning.useFastFunctions) System.error("validation error", "trying to validate FAST_CALL internal opcode");
+				FAST_CALL13 => if (FastIntTuning.useFastFunctions) System.error("validation error", "trying to validate FAST_CALL internal opcode");
+				FAST_CALL14 => if (FastIntTuning.useFastFunctions) System.error("validation error", "trying to validate FAST_CALL internal opcode");
+				FAST_CALL15 => if (FastIntTuning.useFastFunctions) System.error("validation error", "trying to validate FAST_CALL internal opcode");
+				FAST_CALL16 => if (FastIntTuning.useFastFunctions) System.error("validation error", "trying to validate FAST_CALL internal opcode");
+				FAST_CALL17 => if (FastIntTuning.useFastFunctions) System.error("validation error", "trying to validate FAST_CALL internal opcode");
+				FAST_CALL18 => if (FastIntTuning.useFastFunctions) System.error("validation error", "trying to validate FAST_CALL internal opcode");
+				FAST_CALL19 => if (FastIntTuning.useFastFunctions) System.error("validation error", "trying to validate FAST_CALL internal opcode");
+				FAST_CALL20 => if (FastIntTuning.useFastFunctions) System.error("validation error", "trying to validate FAST_CALL internal opcode");
+				FAST_CALL21 => if (FastIntTuning.useFastFunctions) System.error("validation error", "trying to validate FAST_CALL internal opcode");
+				FAST_CALL22 => if (FastIntTuning.useFastFunctions) System.error("validation error", "trying to validate FAST_CALL internal opcode");
+				FAST_CALL23 => if (FastIntTuning.useFastFunctions) System.error("validation error", "trying to validate FAST_CALL internal opcode");
+				FAST_CALL24 => if (FastIntTuning.useFastFunctions) System.error("validation error", "trying to validate FAST_CALL internal opcode");
+				FAST_CALL25 => if (FastIntTuning.useFastFunctions) System.error("validation error", "trying to validate FAST_CALL internal opcode");
+				FAST_CALL26 => if (FastIntTuning.useFastFunctions) System.error("validation error", "trying to validate FAST_CALL internal opcode");
+				FAST_CALL27 => if (FastIntTuning.useFastFunctions) System.error("validation error", "trying to validate FAST_CALL internal opcode");
+				FAST_CALL28 => if (FastIntTuning.useFastFunctions) System.error("validation error", "trying to validate FAST_CALL internal opcode");
+				FAST_CALL29 => if (FastIntTuning.useFastFunctions) System.error("validation error", "trying to validate FAST_CALL internal opcode");
+				FAST_CALL30 => if (FastIntTuning.useFastFunctions) System.error("validation error", "trying to validate FAST_CALL internal opcode");
+				FAST_CALL31 => if (FastIntTuning.useFastFunctions) System.error("validation error", "trying to validate FAST_CALL internal opcode");
+				FAST_CALL32 => if (FastIntTuning.useFastFunctions) System.error("validation error", "trying to validate FAST_CALL internal opcode");
+				FAST_CALL33 => if (FastIntTuning.useFastFunctions) System.error("validation error", "trying to validate FAST_CALL internal opcode");
+				FAST_CALL34 => if (FastIntTuning.useFastFunctions) System.error("validation error", "trying to validate FAST_CALL internal opcode");
+				FAST_CALL35 => if (FastIntTuning.useFastFunctions) System.error("validation error", "trying to validate FAST_CALL internal opcode");
+				FAST_CALL36 => if (FastIntTuning.useFastFunctions) System.error("validation error", "trying to validate FAST_CALL internal opcode");
+				FAST_CALL37 => if (FastIntTuning.useFastFunctions) System.error("validation error", "trying to validate FAST_CALL internal opcode");
+				FAST_CALL38 => if (FastIntTuning.useFastFunctions) System.error("validation error", "trying to validate FAST_CALL internal opcode");
+				FAST_CALL39 => if (FastIntTuning.useFastFunctions) System.error("validation error", "trying to validate FAST_CALL internal opcode");
 				CALL_INDIRECT => {
 					var sig = parser.readSigRef();
 					var table = parser.readTableRef();

src/engine/EngineOptions.v3

@@ -7,6 +7,8 @@ component EngineOptions {
 	var extensions: Extension.set = Extensions.getDefaults();
 	def DEFAULT_STACK_SIZE = 512u * 1024u;
 	def STACK_SIZE = group.newSizeOption("stack-size", DEFAULT_STACK_SIZE, "Initial stack size in bytes for Wasm execution stacks.");
+	def FAST_FUNCTIONS = group.newBoolOption("fast-functions", false, "Treat functions exported with `fast:` in the name as fast functions.")
+		.onSet(fun v => void(FastIntTuning.useFastFunctions = v));
 	def X_ = OptionsRegistry.addParseFunc(parse);
 
 	def parse(arg: string, err: ErrorGen) -> bool {

src/engine/Module.v3

@@ -17,6 +17,7 @@ class Module(filename: string) {
 	def exports = Vector<(string, Decl)>.new();
 	def elems = Vector<ElemDecl>.new();
 	def data = Vector<DataDecl>.new();
+	def fast_funcs = Vector<FuncDecl>.new();
 	def custom_sections = Vector<CustomSection>.new();
 	var probes: Array<Array<Probe>>;
 	var dyn_probes: Vector<(int, int, Probe)>;
@@ -143,6 +144,8 @@ class FuncDecl(sig_index: int) extends Decl {
 	var target_code: TargetCode;
 	var tierup_trigger: int = int.max;
 	var handlers = FuncHandlerInfo.new();
+	var fast_target_code: TargetCode;
+	var fast_call_idx: int = -1;
 
 	def render(names: NameSection, buf: StringBuilder) -> StringBuilder {
 		var name = if (names != null, names.getFuncName(func_index));
@@ -154,6 +157,7 @@ class FuncDecl(sig_index: int) extends Decl {
 		var tc: TargetCode;
 		var tr: TargetCode;
 		target_code = tc; // reset target code as well
+		fast_target_code = tc;
 		sidetable = Sidetables.NO_SIDETABLE;
 		cbd_sidetable = null;
 	}
@@ -168,6 +172,11 @@ class FuncDecl(sig_index: int) extends Decl {
 		if (cur_bytecode == orig_bytecode) return;
 		cur_bytecode[pc] = orig_bytecode[pc];
 	}
+	def replaceCall(pc: int, idx: int) {
+		// copy bytecode for modification
+		if (cur_bytecode == orig_bytecode) orig_bytecode = Arrays.dup(orig_bytecode);
+		cur_bytecode[pc] = byte.!(Opcodes.indexToFastCall(idx).code);
+	}
 	def reset() -> this {
 		if (cur_bytecode == orig_bytecode) return;
 		ArrayUtil.copyInto(cur_bytecode, 0, orig_bytecode, 0, orig_bytecode.length);
@@ -183,6 +192,7 @@ class FuncDecl(sig_index: int) extends Decl {
 		n.sidetable = this.sidetable;
 		n.num_locals = this.num_locals;
 		n.target_code = this.target_code;
+		n.fast_target_code = this.fast_target_code;
 		return n;
 	}
 	def findExHandler(instance: Instance, tag: Tag, throw_pc: int) -> ExHandler {