diff --git a/internal/encoder/alg/spec.go b/internal/encoder/alg/spec.go
index 88317496c..7f0e81002 100644
--- a/internal/encoder/alg/spec.go
+++ b/internal/encoder/alg/spec.go
@@ -1,3 +1,4 @@
+//go:build (amd64 && go1.16 && !go1.22) || (arm64 && go1.20 && !go1.22)
 // +build amd64,go1.16,!go1.22 arm64,go1.20,!go1.22
 
 /**
@@ -20,6 +21,7 @@ package alg
 
 import (
 	"runtime"
+	"strconv"
 	"unsafe"
 
 	"github.com/bytedance/sonic/internal/native"
@@ -150,6 +152,9 @@ func HtmlEscape(dst []byte, src []byte) []byte {
 }
 
 func F64toa(buf []byte, v float64) ([]byte) {
+	if v == 0 {
+		return append(buf, '0')
+	}
 	buf = rt.GuardSlice2(buf, 64)
 	ret := native.F64toa((*byte)(rt.IndexByte(buf, len(buf))), v)
 	if ret > 0 {
@@ -160,6 +165,9 @@ func F64toa(buf []byte, v float64) ([]byte) {
 }
 
 func F32toa(buf []byte, v float32) ([]byte) {
+	if v == 0 {
+		return append(buf, '0')
+	}
 	buf = rt.GuardSlice2(buf, 64)
 	ret := native.F32toa((*byte)(rt.IndexByte(buf, len(buf))), v)
 	if ret > 0 {
@@ -169,22 +177,39 @@ func F32toa(buf []byte, v float32) ([]byte) {
 	}
 }
 
+// func I64toa(buf []byte, v int64) ([]byte) {
+// 	if -10 < v && v < 10 {
+// 		buf = rt.GuardSlice2(buf, 32)
+// 		ret := native.I64toa((*byte)(rt.IndexByte(buf, len(buf))), v)
+// 		if ret > 0 {
+// 			return buf[:len(buf)+ret]
+// 		} else {
+// 			return buf
+// 		}
+// 	} else {
+// 		return strconv.AppendInt(buf, v, 10)
+// 	}
+// }
+
+// func U64toa(buf []byte, v uint64) ([]byte) {
+// 	if v < 10 {
+// 		buf = rt.GuardSlice2(buf, 32)
+// 		ret := native.U64toa((*byte)(rt.IndexByte(buf, len(buf))), v)
+// 		if ret > 0 {
+// 			return buf[:len(buf)+ret]
+// 		} else {
+// 			return buf
+// 		}
+// 	} else {
+// 		return strconv.AppendInt(buf, int64(v), 10)
+// 	}
+// }
+
 func I64toa(buf []byte, v int64) ([]byte) {
-	buf = rt.GuardSlice2(buf, 32)
-	ret := native.I64toa((*byte)(rt.IndexByte(buf, len(buf))), v)
-	if ret > 0 {
-		return buf[:len(buf)+ret]
-	} else {
-		return buf
-	}
+	return strconv.AppendInt(buf, int64(v), 10)
 }
 
 func U64toa(buf []byte, v uint64) ([]byte) {
-	buf = rt.GuardSlice2(buf, 32)
-	ret := native.U64toa((*byte)(rt.IndexByte(buf, len(buf))), v)
-	if ret > 0 {
-		return buf[:len(buf)+ret]
-	} else {
-		return buf
-	}
+	return strconv.AppendUint(buf, v, 10)
 }
+
diff --git a/internal/encoder/compiler.go b/internal/encoder/compiler.go
index 4ad33edd3..edd40d8bc 100644
--- a/internal/encoder/compiler.go
+++ b/internal/encoder/compiler.go
@@ -244,6 +244,8 @@ func (self *Compiler) compileMap(p *ir.Program, sp int, vt reflect.Type) {
 func (self *Compiler) compileMapBody(p *ir.Program, sp int, vt reflect.Type) {
 	p.Tag(sp + 1)
 	p.Int(ir.OP_byte, '{')
+	e := p.PC()
+	p.Add(ir.OP_is_zero_map)
 	p.Add(ir.OP_save)
 	p.Rtt(ir.OP_map_iter, vt)
 	p.Add(ir.OP_save)
@@ -271,6 +273,7 @@ func (self *Compiler) compileMapBody(p *ir.Program, sp int, vt reflect.Type) {
 	p.Pin(j)
 	p.Add(ir.OP_map_stop)
 	p.Add(ir.OP_drop_2)
+	p.Pin(e)
 	p.Int(ir.OP_byte, '}')
 }
 
@@ -355,6 +358,8 @@ func (self *Compiler) compileSliceBody(p *ir.Program, sp int, vt reflect.Type) {
 func (self *Compiler) compileSliceArray(p *ir.Program, sp int, vt reflect.Type) {
 	p.Tag(sp)
 	p.Int(ir.OP_byte, '[')
+	e := p.PC()
+	p.Add(ir.OP_is_nil)
 	p.Add(ir.OP_save)
 	p.Add(ir.OP_slice_len)
 	i := p.PC()
@@ -368,6 +373,7 @@ func (self *Compiler) compileSliceArray(p *ir.Program, sp int, vt reflect.Type)
 	p.Pin(i)
 	p.Pin(j)
 	p.Add(ir.OP_drop)
+	p.Pin(e)
 	p.Int(ir.OP_byte, ']')
 }
 
diff --git a/internal/encoder/vm/vm.go b/internal/encoder/vm/vm.go
index 731ae70ab..2311214a1 100644
--- a/internal/encoder/vm/vm.go
+++ b/internal/encoder/vm/vm.go
@@ -65,9 +65,9 @@ func Execute(b *[]byte, p unsafe.Pointer, s *vars.Stack, flags uint64, prog *ir.
 		pc++
 		op := ins.Op()
 
-		if vars.DebugSyncGC {
-			print_instr(buf, pc, op, ins)
-		}
+		// if vars.DebugSyncGC {
+		// 	print_instr(buf, pc, op, ins)
+		// }
 		
 		switch op {
 		case ir.OP_goto:
@@ -124,15 +124,15 @@ func Execute(b *[]byte, p unsafe.Pointer, s *vars.Stack, flags uint64, prog *ir.
 				continue
 			}
 		case ir.OP_null:
-			buf = append(buf, "null"...)
+			buf = append(buf, 'n', 'u', 'l', 'l')
 		case ir.OP_str:
 			v := *(*string)(p)
 			buf = alg.Quote(buf, v, false)
 		case ir.OP_bool:
 			if *(*bool)(p) {
-				buf = append(buf, "true"...)
+				buf = append(buf, 't', 'r', 'u', 'e')
 			} else {
-				buf = append(buf, "false"...)
+				buf = append(buf, 'f', 'a', 'l', 's', 'e')
 			}
 		case ir.OP_i8:
 			v := *(*int8)(p)
@@ -226,6 +226,28 @@ func Execute(b *[]byte, p unsafe.Pointer, s *vars.Stack, flags uint64, prog *ir.
 				continue
 			}
 			p = it.It.K
+		case ir.OP_marshal_text:
+			vt, itab := ins.Vtab()
+			var it rt.GoIface
+			switch vt.Kind() {
+				case reflect.Interface        : 
+				if is_nil(p) {
+					buf = append(buf, 'n', 'u', 'l', 'l')
+					continue
+				}
+				it = rt.AssertI2I(_T_encoding_TextMarshaler, *(*rt.GoIface)(p))
+				case reflect.Ptr, reflect.Map : it = convT2I(p, true, itab)
+				default                       : it = convT2I(p, !vt.Indirect(), itab)
+			}
+			if err := alg.EncodeTextMarshaler(&buf, *(*encoding.TextMarshaler)(unsafe.Pointer(&it)), (flags)); err != nil {
+				return err
+			}
+		case ir.OP_marshal_text_p:
+			_, itab := ins.Vtab()
+			it := convT2I(p, false, itab)
+			if err := alg.EncodeTextMarshaler(&buf, *(*encoding.TextMarshaler)(unsafe.Pointer(&it)), (flags)); err != nil {
+				return err
+			}
 		case ir.OP_map_write_key:
 			if has_opts(flags, alg.BitSortMapKeys) {
 				v := *(*string)(p)
@@ -280,38 +302,36 @@ func Execute(b *[]byte, p unsafe.Pointer, s *vars.Stack, flags uint64, prog *ir.
 			}
 		case ir.OP_empty_arr:
 			if has_opts(flags, alg.BitNoNullSliceOrMap) {
-				buf = append(buf, "[]"...)
+				buf = append(buf, '[', ']')
 			} else {
-				buf = append(buf, "null"...)
+				buf = append(buf, 'n', 'u', 'l', 'l')
 			}
 		case ir.OP_empty_obj:
 			if has_opts(flags, alg.BitNoNullSliceOrMap) {
-				buf = append(buf, "{}"...)
+				buf = append(buf, '{', '}')
 			} else {
-				buf = append(buf, "null"...)
+				buf = append(buf, 'n', 'u', 'l', 'l')
 			}
 		case ir.OP_marshal:
 			vt, itab := ins.Vtab()
-			var err error
-			if buf, err = call_json_marshaler(buf, vt, itab, p, flags, false); err != nil {
-				return err
-			}
-		case ir.OP_marshal_p:
-			vt, itab := ins.Vtab()
-			var err error
-			if buf, err = call_json_marshaler(buf, vt, itab, p, flags, true); err != nil {
-				return err
+			var it rt.GoIface
+			switch vt.Kind() {
+				case reflect.Interface        : 
+				if is_nil(p) {
+					buf = append(buf, 'n', 'u', 'l', 'l')
+					continue
+				}
+				it = rt.AssertI2I(_T_json_Marshaler, *(*rt.GoIface)(p))
+				case reflect.Ptr, reflect.Map : it = convT2I(p, true, itab)
+				default                       : it = convT2I(p, !vt.Indirect(), itab)
 			}
-		case ir.OP_marshal_text:
-			vt, itab := ins.Vtab()
-			var err error
-			if buf, err = call_text_marshaler(buf, vt, itab, p, flags, false); err != nil {
+			if err := alg.EncodeJsonMarshaler(&buf, *(*json.Marshaler)(unsafe.Pointer(&it)), (flags)); err != nil {
 				return err
 			}
-		case ir.OP_marshal_text_p:
-			vt, itab := ins.Vtab()
-			var err error
-			if buf, err = call_text_marshaler(buf, vt, itab, p, flags, true); err != nil {
+		case ir.OP_marshal_p:
+			_, itab := ins.Vtab()
+			it := convT2I(p, false, itab)
+			if err := alg.EncodeJsonMarshaler(&buf, *(*json.Marshaler)(unsafe.Pointer(&it)), (flags)); err != nil {
 				return err
 			}
 		default:
@@ -348,49 +368,3 @@ func convT2I(ptr unsafe.Pointer, deref bool, itab *rt.GoItab) (rt.GoIface) {
 		Value: ptr,
 	}
 }
-
-func call_text_marshaler(buf []byte, vt *rt.GoType, itab *rt.GoItab, p unsafe.Pointer, flags uint64, pointer bool) ([]byte, error) {
-	var it rt.GoIface
-	if !pointer {
-		switch vt.Kind() {
-			case reflect.Interface        : 
-			if is_nil(p) {
-				buf = append(buf, "null"...)
-				return buf, nil
-			}
-			it = rt.AssertI2I(_T_encoding_TextMarshaler, *(*rt.GoIface)(p))
-			case reflect.Ptr, reflect.Map : it = convT2I(p, true, itab)
-			default                       : it = convT2I(p, !vt.Indirect(), itab)
-		}
-
-	} else {
-		it = convT2I(p, false, itab)
-	}
-	if err := alg.EncodeTextMarshaler(&buf, *(*encoding.TextMarshaler)(unsafe.Pointer(&it)), (flags)); err != nil {
-		return buf, err
-	}
-	return buf, nil
-}
-
-func call_json_marshaler(buf []byte, vt *rt.GoType, itab *rt.GoItab, p unsafe.Pointer, flags uint64, pointer bool) ([]byte, error) {
-	var it rt.GoIface
-	if !pointer {
-		switch vt.Kind() {
-			case reflect.Interface        : 
-			if is_nil(p) {
-				buf = append(buf, "null"...)
-				return buf, nil
-			}
-			it = rt.AssertI2I(_T_json_Marshaler, *(*rt.GoIface)(p))
-			case reflect.Ptr, reflect.Map : it = convT2I(p, true, itab)
-			default                       : it = convT2I(p, !vt.Indirect(), itab)
-		}
-
-	} else {
-		it = convT2I(p, false, itab)
-	}
-	if err := alg.EncodeJsonMarshaler(&buf, *(*json.Marshaler)(unsafe.Pointer(&it)), (flags)); err != nil {
-		return buf, err
-	}
-	return buf, nil
-}
diff --git a/internal/native/neon/f32toa_arm64.s b/internal/native/neon/f32toa_arm64.s
index cc6abecd7..de4aa9da7 100644
--- a/internal/native/neon/f32toa_arm64.s
+++ b/internal/native/neon/f32toa_arm64.s
@@ -7,8 +7,9 @@
 
 TEXT ·__f32toa_entry__(SB), NOSPLIT, $16
 	NO_LOCAL_POINTERS
-	WORD $0x910083ff  // add sp, sp, #32
-	JMP _f32toa
+	WORD $0x100000a0 // adr x0, .+20
+	MOVD R0, ret(FP)
+	RET
 	  // .p2align 2, 0x00
 _f32toa:
 	WORD $0xa9be7bfd  // stp	fp, lr, [sp, #-32]!
@@ -905,7 +906,7 @@ _Digits:
 	WORD $0x37393639  // .ascii 4, '96979899'
 	WORD $0x39393839  // .ascii 4, '9899'
 	WORD $0x00000000  // .p2align 3, 0x00
-_LB_e80ea3d0: // _pow10_ceil_sig_f32.g
+_LB_8f8a73de: // _pow10_ceil_sig_f32.g
 	WORD $0x4b43fcf5; WORD $0x81ceb32c  // .quad -9093133594791772939
 	WORD $0x5e14fc32; WORD $0xa2425ff7  // .quad -6754730975062328270
 	WORD $0x359a3b3f; WORD $0xcad2f7f5  // .quad -3831727700400522433
@@ -997,7 +998,9 @@ _f32toa:
 	MOVD out+0(FP), R0
 	FMOVD val+8(FP), F0
 	WORD $0xf90007fc // str x28, [sp, #8]
-	CALL ·__f32toa_entry__(SB)  // _f32toa
+	MOVD ·_subr__f32toa(SB), R11
+	WORD $0x1000005e // adr x30, .+8
+	JMP (R11)
 	WORD $0xf94007fc // ldr x28, [sp, #8]
 	MOVD R0, ret+16(FP)
 	RET
diff --git a/internal/native/neon/f32toa_subr_arm64.go b/internal/native/neon/f32toa_subr_arm64.go
index df71102ac..1e624c14b 100644
--- a/internal/native/neon/f32toa_subr_arm64.go
+++ b/internal/native/neon/f32toa_subr_arm64.go
@@ -9,7 +9,7 @@ package neon
 func __f32toa_entry__() uintptr
 
 var (
-    _subr__f32toa uintptr = 0
+    _subr__f32toa uintptr = __f32toa_entry__() + 0
 )
 
 const (
diff --git a/internal/native/neon/f64toa_arm64.s b/internal/native/neon/f64toa_arm64.s
index 92ce3d329..6a99ed78a 100644
--- a/internal/native/neon/f64toa_arm64.s
+++ b/internal/native/neon/f64toa_arm64.s
@@ -7,8 +7,9 @@
 
 TEXT ·__f64toa_entry__(SB), NOSPLIT, $16
 	NO_LOCAL_POINTERS
-	WORD $0x910083ff  // add sp, sp, #32
-	JMP _f64toa
+	WORD $0x100000a0 // adr x0, .+20
+	MOVD R0, ret(FP)
+	RET
 	  // .p2align 2, 0x00
 _f64toa:
 	WORD $0xa9be7bfd  // stp	fp, lr, [sp, #-32]!
@@ -1221,7 +1222,7 @@ _Digits:
 	WORD $0x37393639  // .ascii 4, '96979899'
 	WORD $0x39393839  // .ascii 4, '9899'
 	  // .p2align 3, 0x00
-_LB_40c6cc40: // _pow10_ceil_sig.g
+_LB_1c9f0ea5: // _pow10_ceil_sig.g
 	WORD $0xbebcdc4f; WORD $0xff77b1fc  // .quad -38366372719436721
 	WORD $0x13bb0f7b; WORD $0x25e8e89c  // .quad 2731688931043774331
 	WORD $0xf73609b1; WORD $0x9faacf3d  // .quad -6941508010590729807
@@ -2470,7 +2471,9 @@ _f64toa:
 	MOVD out+0(FP), R0
 	FMOVD val+8(FP), F0
 	WORD $0xf90007fc // str x28, [sp, #8]
-	CALL ·__f64toa_entry__(SB)  // _f64toa
+	MOVD ·_subr__f64toa(SB), R11
+	WORD $0x1000005e // adr x30, .+8
+	JMP (R11)
 	WORD $0xf94007fc // ldr x28, [sp, #8]
 	MOVD R0, ret+16(FP)
 	RET
diff --git a/internal/native/neon/f64toa_subr_arm64.go b/internal/native/neon/f64toa_subr_arm64.go
index f0c980647..e09d800c9 100644
--- a/internal/native/neon/f64toa_subr_arm64.go
+++ b/internal/native/neon/f64toa_subr_arm64.go
@@ -9,7 +9,7 @@ package neon
 func __f64toa_entry__() uintptr
 
 var (
-    _subr__f64toa uintptr = 0
+    _subr__f64toa uintptr = __f64toa_entry__() + 0
 )
 
 const (
diff --git a/internal/native/neon/fastfloat_arm64_test.go b/internal/native/neon/fastfloat_arm64_test.go
index f3c35f4fd..dc00ec967 100644
--- a/internal/native/neon/fastfloat_arm64_test.go
+++ b/internal/native/neon/fastfloat_arm64_test.go
@@ -84,10 +84,14 @@ func BenchmarkParseFloat64(b *testing.B) {
         float   float64
     }{
         {"Zero", 0},
-        {"Decimal", 33909},
-        {"Float", 339.7784},
-        {"Exp", -5.09e75},
-        {"NegExp", -5.11e-95},
+        {"Decimal1", 39},
+        {"Decimal2", 33909},
+        {"Float1", 3.77},
+        {"Float2", 339.778442},
+        {"Exp1", 5.9e5},
+        {"Exp2", 521.091e25},
+        {"NegExp1", -5.1e-5},
+        {"NegExp2", -52.132e-35},
         {"LongExp", 1.234567890123456e-78},
         {"Big", 123456789123456789123456789},
     
@@ -116,11 +120,14 @@ func BenchmarkParseFloat32(b *testing.B) {
         float   float32
     }{
         {"Zero", 0},
-        {"Integer", 33909},
-        {"ExactFraction", 3.375},
-        {"Point", 339.7784},
-        {"Exp", -5.09e25},
-        {"NegExp", -5.11e-25},
+        {"Decimal1", 39},
+        {"Decimal2", 33909},
+        {"Float1", 3.77},
+        {"Float2", 339.778442},
+        {"Exp1", 5.9e5},
+        {"Exp2", 521.091e19},
+        {"NegExp1", -5.1e-5},
+        {"NegExp2", -52.132e-19},
         {"Shortest", 1.234567e-8},
     }
     for _, c := range f32toaBenches {
diff --git a/internal/native/neon/get_by_path_arm64.s b/internal/native/neon/get_by_path_arm64.s
index 1a1769f12..b7f1406af 100644
--- a/internal/native/neon/get_by_path_arm64.s
+++ b/internal/native/neon/get_by_path_arm64.s
@@ -7,8 +7,9 @@
 
 TEXT ·__get_by_path_entry__(SB), NOSPLIT, $192
 	NO_LOCAL_POINTERS
-	WORD $0x910343ff  // add sp, sp, #208
-	JMP _get_by_path
+	WORD $0x100000a0 // adr x0, .+20
+	MOVD R0, ret(FP)
+	RET
 	  // .p2align 4, 0x00
 lCPI0_0:
 	WORD $0x08040201
@@ -4567,7 +4568,9 @@ _get_by_path:
 	MOVD path+16(FP), R2
 	MOVD m+24(FP), R3
 	WORD $0xf90007fc // str x28, [sp, #8]
-	CALL ·__get_by_path_entry__(SB)  // _get_by_path
+	MOVD ·_subr__get_by_path(SB), R11
+	WORD $0x1000005e // adr x30, .+8
+	JMP (R11)
 	WORD $0xf94007fc // ldr x28, [sp, #8]
 	MOVD R0, ret+32(FP)
 	RET
diff --git a/internal/native/neon/get_by_path_subr_arm64.go b/internal/native/neon/get_by_path_subr_arm64.go
index c2aca952e..2d0172721 100644
--- a/internal/native/neon/get_by_path_subr_arm64.go
+++ b/internal/native/neon/get_by_path_subr_arm64.go
@@ -9,7 +9,7 @@ package neon
 func __get_by_path_entry__() uintptr
 
 var (
-    _subr__get_by_path uintptr = 48
+    _subr__get_by_path uintptr = __get_by_path_entry__() + 48
 )
 
 const (
diff --git a/internal/native/neon/html_escape_arm64.s b/internal/native/neon/html_escape_arm64.s
index 7f44ff66e..238562637 100644
--- a/internal/native/neon/html_escape_arm64.s
+++ b/internal/native/neon/html_escape_arm64.s
@@ -7,8 +7,9 @@
 
 TEXT ·__html_escape_entry__(SB), NOSPLIT, $16
 	NO_LOCAL_POINTERS
-	WORD $0x910083ff  // add sp, sp, #32
-	JMP _html_escape
+	WORD $0x100000a0 // adr x0, .+20
+	MOVD R0, ret(FP)
+	RET
 	  // .p2align 4, 0x00
 lCPI0_0:
 	WORD $0x08040201
@@ -1390,7 +1391,9 @@ _html_escape:
 	MOVD dp+16(FP), R2
 	MOVD dn+24(FP), R3
 	WORD $0xf90007fc // str x28, [sp, #8]
-	CALL ·__html_escape_entry__(SB)  // _html_escape
+	MOVD ·_subr__html_escape(SB), R11
+	WORD $0x1000005e // adr x30, .+8
+	JMP (R11)
 	WORD $0xf94007fc // ldr x28, [sp, #8]
 	MOVD R0, ret+32(FP)
 	RET
diff --git a/internal/native/neon/html_escape_subr_arm64.go b/internal/native/neon/html_escape_subr_arm64.go
index 9882fb6ad..a0a86e057 100644
--- a/internal/native/neon/html_escape_subr_arm64.go
+++ b/internal/native/neon/html_escape_subr_arm64.go
@@ -9,7 +9,7 @@ package neon
 func __html_escape_entry__() uintptr
 
 var (
-    _subr__html_escape uintptr = 32
+    _subr__html_escape uintptr = __html_escape_entry__() + 32
 )
 
 const (
diff --git a/internal/native/neon/i64toa_arm64.s b/internal/native/neon/i64toa_arm64.s
index 828e7f73b..8d13a066b 100644
--- a/internal/native/neon/i64toa_arm64.s
+++ b/internal/native/neon/i64toa_arm64.s
@@ -7,8 +7,9 @@
 
 TEXT ·__i64toa_entry__(SB), NOSPLIT, $16
 	NO_LOCAL_POINTERS
-	WORD $0x910083ff  // add sp, sp, #32
-	JMP _i64toa
+	WORD $0x100000a0 // adr x0, .+20
+	MOVD R0, ret(FP)
+	RET
 	  // .p2align 3, 0x00
 lCPI0_0:
 	WORD $0x147b20c5
@@ -964,7 +965,9 @@ _i64toa:
 	MOVD out+0(FP), R0
 	MOVD val+8(FP), R1
 	WORD $0xf90007fc // str x28, [sp, #8]
-	CALL ·__i64toa_entry__(SB)  // _i64toa
+	MOVD ·_subr__i64toa(SB), R11
+	WORD $0x1000005e // adr x30, .+8
+	JMP (R11)
 	WORD $0xf94007fc // ldr x28, [sp, #8]
 	MOVD R0, ret+16(FP)
 	RET
diff --git a/internal/native/neon/i64toa_subr_arm64.go b/internal/native/neon/i64toa_subr_arm64.go
index 442305e36..7c73cb380 100644
--- a/internal/native/neon/i64toa_subr_arm64.go
+++ b/internal/native/neon/i64toa_subr_arm64.go
@@ -9,7 +9,7 @@ package neon
 func __i64toa_entry__() uintptr
 
 var (
-    _subr__i64toa uintptr = 48
+    _subr__i64toa uintptr = __i64toa_entry__() + 48
 )
 
 const (
diff --git a/internal/native/neon/lspace_arm64.s b/internal/native/neon/lspace_arm64.s
index c62c6f615..e4e2cd2f8 100644
--- a/internal/native/neon/lspace_arm64.s
+++ b/internal/native/neon/lspace_arm64.s
@@ -7,8 +7,9 @@
 
 TEXT ·__lspace_entry__(SB), NOSPLIT, $16
 	NO_LOCAL_POINTERS
-	WORD $0x910083ff  // add sp, sp, #32
-	JMP _lspace
+	WORD $0x100000a0 // adr x0, .+20
+	MOVD R0, ret(FP)
+	RET
 	  // .p2align 2, 0x00
 _lspace:
 	WORD $0xa9be7bfd  // stp	fp, lr, [sp, #-32]!
@@ -54,7 +55,9 @@ _lspace:
 	MOVD nb+8(FP), R1
 	MOVD off+16(FP), R2
 	WORD $0xf90007fc // str x28, [sp, #8]
-	CALL ·__lspace_entry__(SB)  // _lspace
+	MOVD ·_subr__lspace(SB), R11
+	WORD $0x1000005e // adr x30, .+8
+	JMP (R11)
 	WORD $0xf94007fc // ldr x28, [sp, #8]
 	MOVD R0, ret+24(FP)
 	RET
diff --git a/internal/native/neon/lspace_subr_arm64.go b/internal/native/neon/lspace_subr_arm64.go
index 89fbf2d7c..e8dc6d56f 100644
--- a/internal/native/neon/lspace_subr_arm64.go
+++ b/internal/native/neon/lspace_subr_arm64.go
@@ -9,7 +9,7 @@ package neon
 func __lspace_entry__() uintptr
 
 var (
-    _subr__lspace uintptr = 0
+    _subr__lspace uintptr = __lspace_entry__() + 0
 )
 
 const (
diff --git a/internal/native/neon/native_arm64_test.go b/internal/native/neon/native_arm64_test.go
index 278063b4e..07644b519 100644
--- a/internal/native/neon/native_arm64_test.go
+++ b/internal/native/neon/native_arm64_test.go
@@ -36,17 +36,11 @@ import (
 	"github.com/stretchr/testify/require"
 )
 
-//go:noinline
-func BREAK() bool {
-    return true
-}
-
 func TestNative_Value(t *testing.T) {
     runtime.GC()
     var v types.JsonState
     s := `   -12345`
     p := (*rt.GoString)(unsafe.Pointer(&s))
-    BREAK()
     x := value(p.Ptr, p.Len, 0, &v, 0)
     assert.Equal(t, 9, x)
     assert.Equal(t, types.V_INTEGER, v.Vt)
diff --git a/internal/native/neon/quote_arm64.s b/internal/native/neon/quote_arm64.s
index 61cd7dd52..0092813b4 100644
--- a/internal/native/neon/quote_arm64.s
+++ b/internal/native/neon/quote_arm64.s
@@ -7,8 +7,9 @@
 
 TEXT ·__quote_entry__(SB), NOSPLIT, $16
 	NO_LOCAL_POINTERS
-	WORD $0x910083ff  // add sp, sp, #32
-	JMP _quote
+	WORD $0x100000a0 // adr x0, .+20
+	MOVD R0, ret(FP)
+	RET
 	  // .p2align 4, 0x00
 lCPI0_0:
 	WORD $0x08040201
@@ -2546,7 +2547,9 @@ _quote:
 	MOVD dn+24(FP), R3
 	MOVD flags+32(FP), R4
 	WORD $0xf90007fc // str x28, [sp, #8]
-	CALL ·__quote_entry__(SB)  // _quote
+	MOVD ·_subr__quote(SB), R11
+	WORD $0x1000005e // adr x30, .+8
+	JMP (R11)
 	WORD $0xf94007fc // ldr x28, [sp, #8]
 	MOVD R0, ret+40(FP)
 	RET
diff --git a/internal/native/neon/quote_subr_arm64.go b/internal/native/neon/quote_subr_arm64.go
index 6f7a8162d..e9d942880 100644
--- a/internal/native/neon/quote_subr_arm64.go
+++ b/internal/native/neon/quote_subr_arm64.go
@@ -9,7 +9,7 @@ package neon
 func __quote_entry__() uintptr
 
 var (
-    _subr__quote uintptr = 32
+    _subr__quote uintptr = __quote_entry__() + 32
 )
 
 const (
diff --git a/internal/native/neon/skip_array_arm64.s b/internal/native/neon/skip_array_arm64.s
index 9e258625a..21a9afcdd 100644
--- a/internal/native/neon/skip_array_arm64.s
+++ b/internal/native/neon/skip_array_arm64.s
@@ -7,8 +7,9 @@
 
 TEXT ·__skip_array_entry__(SB), NOSPLIT, $96
 	NO_LOCAL_POINTERS
-	WORD $0x9101c3ff  // add sp, sp, #112
-	JMP _skip_array
+	WORD $0x100000a0 // adr x0, .+20
+	MOVD R0, ret(FP)
+	RET
 	  // .p2align 4, 0x00
 lCPI0_0:
 	WORD $0x00000001; WORD $0x00000000  // .quad 1
@@ -2370,7 +2371,9 @@ _skip_array:
 	MOVD m+16(FP), R2
 	MOVD flags+24(FP), R3
 	WORD $0xf90007fc // str x28, [sp, #8]
-	CALL ·__skip_array_entry__(SB)  // _skip_array
+	MOVD ·_subr__skip_array(SB), R11
+	WORD $0x1000005e // adr x30, .+8
+	JMP (R11)
 	WORD $0xf94007fc // ldr x28, [sp, #8]
 	MOVD R0, ret+32(FP)
 	RET
diff --git a/internal/native/neon/skip_array_subr_arm64.go b/internal/native/neon/skip_array_subr_arm64.go
index 96ce7fdfc..c011f6ffd 100644
--- a/internal/native/neon/skip_array_subr_arm64.go
+++ b/internal/native/neon/skip_array_subr_arm64.go
@@ -9,7 +9,7 @@ package neon
 func __skip_array_entry__() uintptr
 
 var (
-    _subr__skip_array uintptr = 48
+    _subr__skip_array uintptr = __skip_array_entry__() + 48
 )
 
 const (
diff --git a/internal/native/neon/skip_number_arm64.s b/internal/native/neon/skip_number_arm64.s
index af72ff40c..bb590ca3d 100644
--- a/internal/native/neon/skip_number_arm64.s
+++ b/internal/native/neon/skip_number_arm64.s
@@ -7,8 +7,9 @@
 
 TEXT ·__skip_number_entry__(SB), NOSPLIT, $32
 	NO_LOCAL_POINTERS
-	WORD $0x9100c3ff  // add sp, sp, #48
-	JMP _skip_number
+	WORD $0x100000a0 // adr x0, .+20
+	MOVD R0, ret(FP)
+	RET
 	  // .p2align 4, 0x00
 lCPI0_0:
 	WORD $0x08040201
@@ -361,7 +362,9 @@ _skip_number:
 	MOVD s+0(FP), R0
 	MOVD p+8(FP), R1
 	WORD $0xf90007fc // str x28, [sp, #8]
-	CALL ·__skip_number_entry__(SB)  // _skip_number
+	MOVD ·_subr__skip_number(SB), R11
+	WORD $0x1000005e // adr x30, .+8
+	JMP (R11)
 	WORD $0xf94007fc // ldr x28, [sp, #8]
 	MOVD R0, ret+16(FP)
 	RET
diff --git a/internal/native/neon/skip_number_subr_arm64.go b/internal/native/neon/skip_number_subr_arm64.go
index e0ff6f5b1..d2ba7d934 100644
--- a/internal/native/neon/skip_number_subr_arm64.go
+++ b/internal/native/neon/skip_number_subr_arm64.go
@@ -9,7 +9,7 @@ package neon
 func __skip_number_entry__() uintptr
 
 var (
-    _subr__skip_number uintptr = 32
+    _subr__skip_number uintptr = __skip_number_entry__() + 32
 )
 
 const (
diff --git a/internal/native/neon/skip_object_arm64.s b/internal/native/neon/skip_object_arm64.s
index eb3cd036d..8de33e475 100644
--- a/internal/native/neon/skip_object_arm64.s
+++ b/internal/native/neon/skip_object_arm64.s
@@ -7,8 +7,9 @@
 
 TEXT ·__skip_object_entry__(SB), NOSPLIT, $96
 	NO_LOCAL_POINTERS
-	WORD $0x9101c3ff  // add sp, sp, #112
-	JMP _skip_object
+	WORD $0x100000a0 // adr x0, .+20
+	MOVD R0, ret(FP)
+	RET
 	  // .p2align 4, 0x00
 lCPI0_0:
 	WORD $0x00000001; WORD $0x00000000  // .quad 1
@@ -2370,7 +2371,9 @@ _skip_object:
 	MOVD m+16(FP), R2
 	MOVD flags+24(FP), R3
 	WORD $0xf90007fc // str x28, [sp, #8]
-	CALL ·__skip_object_entry__(SB)  // _skip_object
+	MOVD ·_subr__skip_object(SB), R11
+	WORD $0x1000005e // adr x30, .+8
+	JMP (R11)
 	WORD $0xf94007fc // ldr x28, [sp, #8]
 	MOVD R0, ret+32(FP)
 	RET
diff --git a/internal/native/neon/skip_object_subr_arm64.go b/internal/native/neon/skip_object_subr_arm64.go
index e526aeb93..ab7ab2034 100644
--- a/internal/native/neon/skip_object_subr_arm64.go
+++ b/internal/native/neon/skip_object_subr_arm64.go
@@ -9,7 +9,7 @@ package neon
 func __skip_object_entry__() uintptr
 
 var (
-    _subr__skip_object uintptr = 48
+    _subr__skip_object uintptr = __skip_object_entry__() + 48
 )
 
 const (
diff --git a/internal/native/neon/skip_one_arm64.s b/internal/native/neon/skip_one_arm64.s
index 694c38ec4..3b65cba3f 100644
--- a/internal/native/neon/skip_one_arm64.s
+++ b/internal/native/neon/skip_one_arm64.s
@@ -7,8 +7,9 @@
 
 TEXT ·__skip_one_entry__(SB), NOSPLIT, $96
 	NO_LOCAL_POINTERS
-	WORD $0x9101c3ff  // add sp, sp, #112
-	JMP _skip_one
+	WORD $0x100000a0 // adr x0, .+20
+	MOVD R0, ret(FP)
+	RET
 	  // .p2align 4, 0x00
 lCPI0_0:
 	WORD $0x00000001; WORD $0x00000000  // .quad 1
@@ -2370,7 +2371,9 @@ _skip_one:
 	MOVD m+16(FP), R2
 	MOVD flags+24(FP), R3
 	WORD $0xf90007fc // str x28, [sp, #8]
-	CALL ·__skip_one_entry__(SB)  // _skip_one
+	MOVD ·_subr__skip_one(SB), R11
+	WORD $0x1000005e // adr x30, .+8
+	JMP (R11)
 	WORD $0xf94007fc // ldr x28, [sp, #8]
 	MOVD R0, ret+32(FP)
 	RET
diff --git a/internal/native/neon/skip_one_fast_arm64.s b/internal/native/neon/skip_one_fast_arm64.s
index 47f883d50..550166c1f 100644
--- a/internal/native/neon/skip_one_fast_arm64.s
+++ b/internal/native/neon/skip_one_fast_arm64.s
@@ -7,8 +7,9 @@
 
 TEXT ·__skip_one_fast_entry__(SB), NOSPLIT, $176
 	NO_LOCAL_POINTERS
-	WORD $0x910303ff  // add sp, sp, #192
-	JMP _skip_one_fast
+	WORD $0x100000a0 // adr x0, .+20
+	MOVD R0, ret(FP)
+	RET
 	  // .p2align 4, 0x00
 lCPI0_0:
 	WORD $0x08040201
@@ -986,7 +987,9 @@ _skip_one_fast:
 	MOVD s+0(FP), R0
 	MOVD p+8(FP), R1
 	WORD $0xf90007fc // str x28, [sp, #8]
-	CALL ·__skip_one_fast_entry__(SB)  // _skip_one_fast
+	MOVD ·_subr__skip_one_fast(SB), R11
+	WORD $0x1000005e // adr x30, .+8
+	JMP (R11)
 	WORD $0xf94007fc // ldr x28, [sp, #8]
 	MOVD R0, ret+16(FP)
 	RET
diff --git a/internal/native/neon/skip_one_fast_subr_arm64.go b/internal/native/neon/skip_one_fast_subr_arm64.go
index b4c2f5d07..4801fa11d 100644
--- a/internal/native/neon/skip_one_fast_subr_arm64.go
+++ b/internal/native/neon/skip_one_fast_subr_arm64.go
@@ -9,7 +9,7 @@ package neon
 func __skip_one_fast_entry__() uintptr
 
 var (
-    _subr__skip_one_fast uintptr = 32
+    _subr__skip_one_fast uintptr = __skip_one_fast_entry__() + 32
 )
 
 const (
diff --git a/internal/native/neon/skip_one_subr_arm64.go b/internal/native/neon/skip_one_subr_arm64.go
index 87c40c784..01b275681 100644
--- a/internal/native/neon/skip_one_subr_arm64.go
+++ b/internal/native/neon/skip_one_subr_arm64.go
@@ -9,7 +9,7 @@ package neon
 func __skip_one_entry__() uintptr
 
 var (
-    _subr__skip_one uintptr = 48
+    _subr__skip_one uintptr = __skip_one_entry__() + 48
 )
 
 const (
diff --git a/internal/native/neon/u64toa_arm64.s b/internal/native/neon/u64toa_arm64.s
index da0f7018c..1b280da77 100644
--- a/internal/native/neon/u64toa_arm64.s
+++ b/internal/native/neon/u64toa_arm64.s
@@ -7,8 +7,9 @@
 
 TEXT ·__u64toa_entry__(SB), NOSPLIT, $16
 	NO_LOCAL_POINTERS
-	WORD $0x910083ff  // add sp, sp, #32
-	JMP _u64toa
+	WORD $0x100000a0 // adr x0, .+20
+	MOVD R0, ret(FP)
+	RET
 	  // .p2align 3, 0x00
 lCPI0_0:
 	WORD $0x147b20c5
@@ -581,7 +582,9 @@ _u64toa:
 	MOVD out+0(FP), R0
 	MOVD val+8(FP), R1
 	WORD $0xf90007fc // str x28, [sp, #8]
-	CALL ·__u64toa_entry__(SB)  // _u64toa
+	MOVD ·_subr__u64toa(SB), R11
+	WORD $0x1000005e // adr x30, .+8
+	JMP (R11)
 	WORD $0xf94007fc // ldr x28, [sp, #8]
 	MOVD R0, ret+16(FP)
 	RET
diff --git a/internal/native/neon/u64toa_subr_arm64.go b/internal/native/neon/u64toa_subr_arm64.go
index 3cd0f3d7d..151794d52 100644
--- a/internal/native/neon/u64toa_subr_arm64.go
+++ b/internal/native/neon/u64toa_subr_arm64.go
@@ -9,7 +9,7 @@ package neon
 func __u64toa_entry__() uintptr
 
 var (
-    _subr__u64toa uintptr = 48
+    _subr__u64toa uintptr = __u64toa_entry__() + 48
 )
 
 const (
diff --git a/internal/native/neon/unquote_arm64.s b/internal/native/neon/unquote_arm64.s
index c58cf0052..e43c4de31 100644
--- a/internal/native/neon/unquote_arm64.s
+++ b/internal/native/neon/unquote_arm64.s
@@ -7,8 +7,9 @@
 
 TEXT ·__unquote_entry__(SB), NOSPLIT, $96
 	NO_LOCAL_POINTERS
-	WORD $0x9101c3ff  // add sp, sp, #112
-	JMP _unquote
+	WORD $0x100000a0 // adr x0, .+20
+	MOVD R0, ret(FP)
+	RET
 	  // .p2align 4, 0x00
 lCPI0_0:
 	WORD $0x08040201
@@ -601,7 +602,9 @@ _unquote:
 	MOVD ep+24(FP), R3
 	MOVD flags+32(FP), R4
 	WORD $0xf90007fc // str x28, [sp, #8]
-	CALL ·__unquote_entry__(SB)  // _unquote
+	MOVD ·_subr__unquote(SB), R11
+	WORD $0x1000005e // adr x30, .+8
+	JMP (R11)
 	WORD $0xf94007fc // ldr x28, [sp, #8]
 	MOVD R0, ret+40(FP)
 	RET
diff --git a/internal/native/neon/unquote_subr_arm64.go b/internal/native/neon/unquote_subr_arm64.go
index 7f5c3e37b..040e15a3d 100644
--- a/internal/native/neon/unquote_subr_arm64.go
+++ b/internal/native/neon/unquote_subr_arm64.go
@@ -9,7 +9,7 @@ package neon
 func __unquote_entry__() uintptr
 
 var (
-    _subr__unquote uintptr = 32
+    _subr__unquote uintptr = __unquote_entry__() + 32
 )
 
 const (
diff --git a/internal/native/neon/validate_one_arm64.s b/internal/native/neon/validate_one_arm64.s
index 088b02e90..42a34ace3 100644
--- a/internal/native/neon/validate_one_arm64.s
+++ b/internal/native/neon/validate_one_arm64.s
@@ -7,8 +7,9 @@
 
 TEXT ·__validate_one_entry__(SB), NOSPLIT, $96
 	NO_LOCAL_POINTERS
-	WORD $0x9101c3ff  // add sp, sp, #112
-	JMP _validate_one
+	WORD $0x100000a0 // adr x0, .+20
+	MOVD R0, ret(FP)
+	RET
 	  // .p2align 4, 0x00
 lCPI0_0:
 	WORD $0x00000001; WORD $0x00000000  // .quad 1
@@ -1846,7 +1847,9 @@ _validate_one:
 	MOVD p+8(FP), R1
 	MOVD m+16(FP), R2
 	WORD $0xf90007fc // str x28, [sp, #8]
-	CALL ·__validate_one_entry__(SB)  // _validate_one
+	MOVD ·_subr__validate_one(SB), R11
+	WORD $0x1000005e // adr x30, .+8
+	JMP (R11)
 	WORD $0xf94007fc // ldr x28, [sp, #8]
 	MOVD R0, ret+24(FP)
 	RET
diff --git a/internal/native/neon/validate_one_subr_arm64.go b/internal/native/neon/validate_one_subr_arm64.go
index 2fa1a5215..cbebeb08b 100644
--- a/internal/native/neon/validate_one_subr_arm64.go
+++ b/internal/native/neon/validate_one_subr_arm64.go
@@ -9,7 +9,7 @@ package neon
 func __validate_one_entry__() uintptr
 
 var (
-    _subr__validate_one uintptr = 48
+    _subr__validate_one uintptr = __validate_one_entry__() + 48
 )
 
 const (
diff --git a/internal/native/neon/validate_utf8_arm64.s b/internal/native/neon/validate_utf8_arm64.s
index f87c1eaac..9398beaa7 100644
--- a/internal/native/neon/validate_utf8_arm64.s
+++ b/internal/native/neon/validate_utf8_arm64.s
@@ -7,8 +7,9 @@
 
 TEXT ·__validate_utf8_entry__(SB), NOSPLIT, $48
 	NO_LOCAL_POINTERS
-	WORD $0x910103ff  // add sp, sp, #64
-	JMP _validate_utf8
+	WORD $0x100000a0 // adr x0, .+20
+	MOVD R0, ret(FP)
+	RET
 	  // .p2align 2, 0x00
 _validate_utf8:
 	WORD $0xd10103ff  // sub	sp, sp, #64
@@ -210,7 +211,9 @@ _validate_utf8:
 	MOVD p+8(FP), R1
 	MOVD m+16(FP), R2
 	WORD $0xf90007fc // str x28, [sp, #8]
-	CALL ·__validate_utf8_entry__(SB)  // _validate_utf8
+	MOVD ·_subr__validate_utf8(SB), R11
+	WORD $0x1000005e // adr x30, .+8
+	JMP (R11)
 	WORD $0xf94007fc // ldr x28, [sp, #8]
 	MOVD R0, ret+24(FP)
 	RET
diff --git a/internal/native/neon/validate_utf8_fast_arm64.s b/internal/native/neon/validate_utf8_fast_arm64.s
index 95da166eb..a73dd7660 100644
--- a/internal/native/neon/validate_utf8_fast_arm64.s
+++ b/internal/native/neon/validate_utf8_fast_arm64.s
@@ -7,8 +7,9 @@
 
 TEXT ·__validate_utf8_fast_entry__(SB), NOSPLIT, $32
 	NO_LOCAL_POINTERS
-	WORD $0x9100c3ff  // add sp, sp, #48
-	JMP _validate_utf8_fast
+	WORD $0x100000a0 // adr x0, .+20
+	MOVD R0, ret(FP)
+	RET
 	  // .p2align 2, 0x00
 _validate_utf8_fast:
 	WORD $0xd100c3ff  // sub	sp, sp, #48
@@ -170,7 +171,9 @@ _entry:
 _validate_utf8_fast:
 	MOVD s+0(FP), R0
 	WORD $0xf90007fc // str x28, [sp, #8]
-	CALL ·__validate_utf8_fast_entry__(SB)  // _validate_utf8_fast
+	MOVD ·_subr__validate_utf8_fast(SB), R11
+	WORD $0x1000005e // adr x30, .+8
+	JMP (R11)
 	WORD $0xf94007fc // ldr x28, [sp, #8]
 	MOVD R0, ret+8(FP)
 	RET
diff --git a/internal/native/neon/validate_utf8_fast_subr_arm64.go b/internal/native/neon/validate_utf8_fast_subr_arm64.go
index a127e874a..c5fd3ba5d 100644
--- a/internal/native/neon/validate_utf8_fast_subr_arm64.go
+++ b/internal/native/neon/validate_utf8_fast_subr_arm64.go
@@ -9,7 +9,7 @@ package neon
 func __validate_utf8_fast_entry__() uintptr
 
 var (
-    _subr__validate_utf8_fast uintptr = 0
+    _subr__validate_utf8_fast uintptr = __validate_utf8_fast_entry__() + 0
 )
 
 const (
diff --git a/internal/native/neon/validate_utf8_subr_arm64.go b/internal/native/neon/validate_utf8_subr_arm64.go
index ac8a68285..b79f6e852 100644
--- a/internal/native/neon/validate_utf8_subr_arm64.go
+++ b/internal/native/neon/validate_utf8_subr_arm64.go
@@ -9,7 +9,7 @@ package neon
 func __validate_utf8_entry__() uintptr
 
 var (
-    _subr__validate_utf8 uintptr = 0
+    _subr__validate_utf8 uintptr = __validate_utf8_entry__() + 0
 )
 
 const (
diff --git a/internal/native/neon/value_arm64.s b/internal/native/neon/value_arm64.s
index a6cf69959..ea75ef717 100644
--- a/internal/native/neon/value_arm64.s
+++ b/internal/native/neon/value_arm64.s
@@ -7,8 +7,9 @@
 
 TEXT ·__value_entry__(SB), NOSPLIT, $96
 	NO_LOCAL_POINTERS
-	WORD $0x9101c3ff  // add sp, sp, #112
-	JMP _value
+	WORD $0x100000a0 // adr x0, .+20
+	MOVD R0, ret(FP)
+	RET
 	  // .p2align 4, 0x00
 lCPI0_0:
 	WORD $0x08040201
@@ -5954,7 +5955,9 @@ _value:
 	MOVD v+24(FP), R3
 	MOVD flags+32(FP), R4
 	WORD $0xf90007fc // str x28, [sp, #8]
-	CALL ·__value_entry__(SB)  // _value
+	MOVD ·_subr__value(SB), R11
+	WORD $0x1000005e // adr x30, .+8
+	JMP (R11)
 	WORD $0xf94007fc // ldr x28, [sp, #8]
 	MOVD R0, ret+40(FP)
 	RET
diff --git a/internal/native/neon/value_subr_arm64.go b/internal/native/neon/value_subr_arm64.go
index 238bcf742..df885edce 100644
--- a/internal/native/neon/value_subr_arm64.go
+++ b/internal/native/neon/value_subr_arm64.go
@@ -9,7 +9,7 @@ package neon
 func __value_entry__() uintptr
 
 var (
-    _subr__value uintptr = 32
+    _subr__value uintptr = __value_entry__() + 32
 )
 
 const (
diff --git a/internal/native/neon/vnumber_arm64.s b/internal/native/neon/vnumber_arm64.s
index a531a49ad..6d0863820 100644
--- a/internal/native/neon/vnumber_arm64.s
+++ b/internal/native/neon/vnumber_arm64.s
@@ -7,8 +7,9 @@
 
 TEXT ·__vnumber_entry__(SB), NOSPLIT, $96
 	NO_LOCAL_POINTERS
-	WORD $0x9101c3ff  // add sp, sp, #112
-	JMP _vnumber
+	WORD $0x100000a0 // adr x0, .+20
+	MOVD R0, ret(FP)
+	RET
 	  // .p2align 2, 0x00
 _vnumber:
 	WORD $0xa9b96ffc  // stp	x28, x27, [sp, #-112]!
@@ -4677,7 +4678,9 @@ _vnumber:
 	MOVD p+8(FP), R1
 	MOVD v+16(FP), R2
 	WORD $0xf90007fc // str x28, [sp, #8]
-	CALL ·__vnumber_entry__(SB)  // _vnumber
+	MOVD ·_subr__vnumber(SB), R11
+	WORD $0x1000005e // adr x30, .+8
+	JMP (R11)
 	WORD $0xf94007fc // ldr x28, [sp, #8]
 	RET
 
diff --git a/internal/native/neon/vnumber_subr_arm64.go b/internal/native/neon/vnumber_subr_arm64.go
index 2f8086dc1..78dd2c313 100644
--- a/internal/native/neon/vnumber_subr_arm64.go
+++ b/internal/native/neon/vnumber_subr_arm64.go
@@ -9,7 +9,7 @@ package neon
 func __vnumber_entry__() uintptr
 
 var (
-    _subr__vnumber uintptr = 0
+    _subr__vnumber uintptr = __vnumber_entry__() + 0
 )
 
 const (
diff --git a/internal/native/neon/vsigned_arm64.s b/internal/native/neon/vsigned_arm64.s
index 8b8acd0c4..db1603647 100644
--- a/internal/native/neon/vsigned_arm64.s
+++ b/internal/native/neon/vsigned_arm64.s
@@ -7,8 +7,9 @@
 
 TEXT ·__vsigned_entry__(SB), NOSPLIT, $16
 	NO_LOCAL_POINTERS
-	WORD $0x910083ff  // add sp, sp, #32
-	JMP _vsigned
+	WORD $0x100000a0 // adr x0, .+20
+	MOVD R0, ret(FP)
+	RET
 	  // .p2align 2, 0x00
 _vsigned:
 	WORD $0xa9be7bfd  // stp	fp, lr, [sp, #-32]!
@@ -138,7 +139,9 @@ _vsigned:
 	MOVD p+8(FP), R1
 	MOVD v+16(FP), R2
 	WORD $0xf90007fc // str x28, [sp, #8]
-	CALL ·__vsigned_entry__(SB)  // _vsigned
+	MOVD ·_subr__vsigned(SB), R11
+	WORD $0x1000005e // adr x30, .+8
+	JMP (R11)
 	WORD $0xf94007fc // ldr x28, [sp, #8]
 	RET
 
diff --git a/internal/native/neon/vsigned_subr_arm64.go b/internal/native/neon/vsigned_subr_arm64.go
index b436af3da..902bd4dc8 100644
--- a/internal/native/neon/vsigned_subr_arm64.go
+++ b/internal/native/neon/vsigned_subr_arm64.go
@@ -9,7 +9,7 @@ package neon
 func __vsigned_entry__() uintptr
 
 var (
-    _subr__vsigned uintptr = 0
+    _subr__vsigned uintptr = __vsigned_entry__() + 0
 )
 
 const (
diff --git a/internal/native/neon/vstring_arm64.s b/internal/native/neon/vstring_arm64.s
index 9c259c448..f94d5b2de 100644
--- a/internal/native/neon/vstring_arm64.s
+++ b/internal/native/neon/vstring_arm64.s
@@ -7,8 +7,9 @@
 
 TEXT ·__vstring_entry__(SB), NOSPLIT, $32
 	NO_LOCAL_POINTERS
-	WORD $0x9100c3ff  // add sp, sp, #48
-	JMP _vstring
+	WORD $0x100000a0 // adr x0, .+20
+	MOVD R0, ret(FP)
+	RET
 	  // .p2align 4, 0x00
 lCPI0_0:
 	WORD $0x08040201
@@ -632,7 +633,9 @@ _vstring:
 	MOVD v+16(FP), R2
 	MOVD flags+24(FP), R3
 	WORD $0xf90007fc // str x28, [sp, #8]
-	CALL ·__vstring_entry__(SB)  // _vstring
+	MOVD ·_subr__vstring(SB), R11
+	WORD $0x1000005e // adr x30, .+8
+	JMP (R11)
 	WORD $0xf94007fc // ldr x28, [sp, #8]
 	RET
 
diff --git a/internal/native/neon/vstring_subr_arm64.go b/internal/native/neon/vstring_subr_arm64.go
index aa3e88709..0b4a1cf7d 100644
--- a/internal/native/neon/vstring_subr_arm64.go
+++ b/internal/native/neon/vstring_subr_arm64.go
@@ -9,7 +9,7 @@ package neon
 func __vstring_entry__() uintptr
 
 var (
-    _subr__vstring uintptr = 32
+    _subr__vstring uintptr = __vstring_entry__() + 32
 )
 
 const (
diff --git a/internal/native/neon/vunsigned_arm64.s b/internal/native/neon/vunsigned_arm64.s
index b3c1f3fb3..a51083e98 100644
--- a/internal/native/neon/vunsigned_arm64.s
+++ b/internal/native/neon/vunsigned_arm64.s
@@ -7,8 +7,9 @@
 
 TEXT ·__vunsigned_entry__(SB), NOSPLIT, $16
 	NO_LOCAL_POINTERS
-	WORD $0x910083ff  // add sp, sp, #32
-	JMP _vunsigned
+	WORD $0x100000a0 // adr x0, .+20
+	MOVD R0, ret(FP)
+	RET
 	  // .p2align 2, 0x00
 _vunsigned:
 	WORD $0xa9be7bfd  // stp	fp, lr, [sp, #-32]!
@@ -132,7 +133,9 @@ _vunsigned:
 	MOVD p+8(FP), R1
 	MOVD v+16(FP), R2
 	WORD $0xf90007fc // str x28, [sp, #8]
-	CALL ·__vunsigned_entry__(SB)  // _vunsigned
+	MOVD ·_subr__vunsigned(SB), R11
+	WORD $0x1000005e // adr x30, .+8
+	JMP (R11)
 	WORD $0xf94007fc // ldr x28, [sp, #8]
 	RET
 
diff --git a/internal/native/neon/vunsigned_subr_arm64.go b/internal/native/neon/vunsigned_subr_arm64.go
index f32d6af5a..a78f3f961 100644
--- a/internal/native/neon/vunsigned_subr_arm64.go
+++ b/internal/native/neon/vunsigned_subr_arm64.go
@@ -9,7 +9,7 @@ package neon
 func __vunsigned_entry__() uintptr
 
 var (
-    _subr__vunsigned uintptr = 0
+    _subr__vunsigned uintptr = __vunsigned_entry__() + 0
 )
 
 const (
diff --git a/scripts/build-arm.sh b/scripts/build-arm.sh
index 4f3553940..753a4ffe1 100644
--- a/scripts/build-arm.sh
+++ b/scripts/build-arm.sh
@@ -5,6 +5,11 @@ SRC_DIR="native/arm"
 TMP_DIR="output/arm"
 OUT_DIR="internal/native/neon"
 TOOL_DIR="tools"
+CC=clang
+if [ "$1" != "" ]; then
+    CC=$1
+fi
+echo $CC
 
 # Create the output directory if it doesn't exist
 mkdir -p "$TMP_DIR"
@@ -19,7 +24,7 @@ for src_file in "$SRC_DIR"/*.c; do
     asm_file="$TMP_DIR/${base_name}.s"
 
     # Compile the source file into an assembly file
-    clang -Wno-error -Wno-nullability-completeness -mllvm=--go-frame -mllvm=-enable-shrink-wrap=0 -target aarch64-apple-macos11 -march=armv8-a+simd -Itools/simde/simde -mno-red-zone -fno-asynchronous-unwind-tables -fno-builtin -fno-exceptions -fno-rtti -fno-stack-protector -nostdlib -O3 -mno-red-zone -fno-asynchronous-unwind-tables -fno-builtin -fno-exceptions -fno-rtti -fno-stack-protector -nostdlib -S -o "$asm_file" "$src_file" 
+    $CC -Wno-error -Wno-nullability-completeness -mllvm=--go-frame -mllvm=-enable-shrink-wrap=0 -target aarch64-apple-macos11 -march=armv8-a+simd -Itools/simde/simde -mno-red-zone -fno-asynchronous-unwind-tables -fno-builtin -fno-exceptions -fno-rtti -fno-stack-protector -nostdlib -O3 -mno-red-zone -fno-asynchronous-unwind-tables -fno-builtin -fno-exceptions -fno-rtti -fno-stack-protector -nostdlib -S -o "$asm_file" "$src_file" 
 
     # Execute asm2asm tool
     python3 ${TOOL_DIR}/asm2arm/arm.py ${OUT_DIR}/${base_name}_arm64.go $asm_file
diff --git a/tools/asm2arm/arm.py b/tools/asm2arm/arm.py
index e4a5ea8d7..32caa957c 100644
--- a/tools/asm2arm/arm.py
+++ b/tools/asm2arm/arm.py
@@ -2018,8 +2018,11 @@ def _declare_body(self, name: str):
         gosize = 0 if size < 16 else size-16
         self.out.append('TEXT ·_%s_entry__(SB), NOSPLIT, $%d' % (name, gosize))
         self.out.append('\tNO_LOCAL_POINTERS')
-        self.out.append('\t'+Instruction('add sp, sp, #%d' % size).encoded)
-        self.out.append('\tJMP %s' % name)
+        # get current PC
+        self.out.append('\tWORD $0x100000a0 // adr x0, .+20')
+        # self.out.append('\t'+Instruction('add sp, sp, #%d' % size).encoded)
+        self.out.append('\tMOVD R0, ret(FP)')
+        self.out.append('\tRET')
         self._LE_4bytes_IntIntr_2_RawIntr()
         self._reloc()
 
@@ -2049,11 +2052,6 @@ def _declare_function(self, name: str, proto: Prototype):
         addr = self.code.get(subr)
         self.subr[subr] = addr
         size = self.code.stacksize(subr)
-        
-        # Notice: golang will insert 3 instructions when stacksize > 0.
-        # And since we use PCALIGN 16, thus we add 16 bytes PC
-        if size > 0:
-            addr += 16
 
         m_size = size + 64
         # rsp_sub_size = size + 16
@@ -2100,33 +2098,18 @@ def _declare_function(self, name: str, proto: Prototype):
             op, reg = REG_MAP[arg.creg.reg]
             self.out.append('\t%s %s+%d(FP), %s' % (op, arg.name, offs - arg.size, reg))
 
-        # the function starts at zero
-        if addr == 0 and proto.retv is None:
-            raise RuntimeError("UNIMPLEMENT FUNC: %s" % name)
-            self.out.append('\tJMP ·%s(SB)  // %s' % (STUB_NAME, subr))
 
         # Go ASM completely ignores the offset of the JMP instruction,
         # so we need to use indirect jumps instead for tail-call elimination
-        # raise RuntimeError("UNIMPLEMENT FUNC: %s" % name)
-        # self.out.append('\tLEAQ ·%s+%d(SB), AX  // %s' % (STUB_NAME, addr, subr))
-        # self.out.append('\tJMP AX')
-
         # save LR(x30) and Frame Pointer(x29)
-
-        # self.out.append('\tADD $%d, RSP, RSP' % size)
-        # self.out.append('\tSTP.W (R29, R30), -16(RSP)')
-        # self.out.append('\tMOVD RSP, R29')
-        # self.out.append('\tSUB $16, RSP')
-        # self.out.append('\tADD $%d, RSP' % (size + 32))
         self.out.append('\tWORD $0xf90007fc // str x28, [sp, #8]')
-        self.out.append('\tCALL ·_%s_entry__(SB)  // %s' % (subr, subr))
+        self.out.append('\tMOVD ·_subr_%s(SB), R11' % (subr))
+        self.out.append('\tWORD $0x1000005e // adr x30, .+8')
+        self.out.append('\tJMP (R11)')
+        # self.out.append('\tCALL ·_%s_entry__(SB)  // %s' % (subr, subr))
         self.out.append('\tWORD $0xf94007fc // ldr x28, [sp, #8]')
-        # self.out.append('\tSUB $%d, RSP' % (size + 32))
-        # self.out.append('\tADD $16, RSP')
-        # self.out.append('\tLDP -8(RSP), (R29, R30)')
-        # self.out.append('\tADD $16, RSP')
-        # self.out.append('\tLDP.P 16(RSP), (R29, R30)')
-        # self.out.append('\tSUB $%d, RSP, RSP' % size)
+        
+        
 
         # Restore LR and Frame Pointer
 
@@ -2363,8 +2346,8 @@ def main():
             print('var (', file = fp)
             mlen = max(len(s) for s in asm.subr)
             for name, entry in asm.subr.items():
-                # print('    _subr_%s uintptr = _%s_entry__() + %d' % (name.ljust(mlen, ' '), name, entry), file = fp)
-                print('    _subr_%s uintptr = %d' % (name.ljust(mlen, ' '), entry), file = fp)
+                print('    _subr_%s uintptr = _%s_entry__() + %d' % (name.ljust(mlen, ' '), name, entry), file = fp)
+                # print('    _subr_%s uintptr = %d' % (name.ljust(mlen, ' '), entry), file = fp)
             print(')', file = fp)
 
             # dump max stack depth for exported functions