Update SIMD codes

internal/cmd/generator/simd/string.c

@@ -3,7 +3,7 @@
 #include <string.h>
 #include <immintrin.h>
 
-uint64_t findEscapeIndex64(char *buf, int len) {
+uint64_t findHTMLEscapeIndex64(char *buf, int len) {
   static const uint64_t lsb = 0x0101010101010101;
   static const uint64_t msb = 0x8080808080808080;
 
@@ -29,7 +29,7 @@ uint64_t findEscapeIndex64(char *buf, int len) {
   return 8 * chunkLen;
 }
 
-uint64_t findEscapeIndex128(char *buf, int len) {
+uint64_t findHTMLEscapeIndex128(char *buf, int len) {
   static const uint64_t lsb = 0x0101010101010101;
   static const uint64_t msb = 0x8080808080808080;
 
@@ -71,12 +71,12 @@ uint64_t findEscapeIndex128(char *buf, int len) {
   }
   int idx = 16 * chunkLen;
   if (len - idx >= 8) {
-    return idx + findEscapeIndex64(sp, len - idx);
+    return idx + findHTMLEscapeIndex64(sp, len - idx);
   }
   return idx;
 }
 
-uint64_t findEscapeIndex256(char *buf, int len) {
+uint64_t findHTMLEscapeIndex256(char *buf, int len) {
   static const uint64_t lsb = 0x0101010101010101;
   static const uint64_t msb = 0x8080808080808080;
 
@@ -118,6 +118,108 @@ uint64_t findEscapeIndex256(char *buf, int len) {
   }
   int idx = 32 * chunkLen;
   int remainLen = len - idx;
+  if (remainLen >= 16) {
+    return idx + findHTMLEscapeIndex128(sp, remainLen);
+  } else if (remainLen >= 8) {
+    return idx + findHTMLEscapeIndex64(sp, remainLen);
+  }
+  return idx;
+}
+
+uint64_t findEscapeIndex64(char *buf, int len) {
+  static const uint64_t lsb = 0x0101010101010101;
+  static const uint64_t msb = 0x8080808080808080;
+
+  static const uint64_t space  = lsb * 0x20;
+  static const uint64_t quote  = lsb * '"';
+  static const uint64_t escape = lsb * '\\';
+
+  char *sp = buf;
+  size_t chunkLen = len / 8;
+  int chunkIdx = 0;
+  for (; chunkIdx < chunkLen; chunkIdx++) {
+    uint64_t n    = *(uint64_t *)sp;
+    uint64_t mask = n | (n - space) | ((n ^ quote) - lsb) | ((n ^ escape) - lsb);
+    uint64_t masked = mask & msb;
+    if (masked != 0) {
+      return __builtin_ctz(masked) / 8;
+    }
+    sp += 8;
+  }
+  return 8 * chunkLen;
+}
+
+uint64_t findEscapeIndex128(char *buf, int len) {
+  static const uint64_t lsb = 0x0101010101010101;
+  static const uint64_t msb = 0x8080808080808080;
+
+  static const __m64 space  = (__m64)(lsb * 0x20);
+  static const __m64 quote  = (__m64)(lsb * '"');
+  static const __m64 escape = (__m64)(lsb * '\\');
+
+  __m128i zeroV   = _mm_setzero_si128();
+  __m128i msbV    = _mm_set_epi64((__m64)(msb), (__m64)(msb));
+  __m128i lsbV    = _mm_set_epi64((__m64)(lsb), (__m64)(lsb));
+  __m128i spaceV  = _mm_set_epi64(space, space);
+  __m128i quoteV  = _mm_set_epi64(quote, quote);
+  __m128i escapeV = _mm_set_epi64(escape, escape);
+
+  char *sp = buf;
+  size_t chunkLen = len / 16;
+  int chunkIdx = 0;
+  for (; chunkIdx < chunkLen; chunkIdx++) {
+    __m128i n       = _mm_loadu_si128((const void *)sp);
+    __m128i spaceN  = _mm_sub_epi64(n, spaceV);
+    __m128i quoteN  = _mm_sub_epi64(_mm_xor_si128(n, quoteV), lsbV);
+    __m128i escapeN = _mm_sub_epi64(_mm_xor_si128(n, escapeV), lsbV);
+
+    __m128i mask = _mm_or_si128(_mm_or_si128(_mm_or_si128(n, spaceN), quoteN), escapeN);
+    int movemask = _mm_movemask_epi8(_mm_and_si128(mask, msbV));
+    if (movemask != 0) {
+      return __builtin_ctz(movemask);
+    }
+    sp += 16;
+  }
+  int idx = 16 * chunkLen;
+  if (len - idx >= 8) {
+    return idx + findEscapeIndex64(sp, len - idx);
+  }
+  return idx;
+}
+
+uint64_t findEscapeIndex256(char *buf, int len) {
+  static const uint64_t lsb = 0x0101010101010101;
+  static const uint64_t msb = 0x8080808080808080;
+
+  static const __m64 space  = (__m64)(lsb * 0x20);
+  static const __m64 quote  = (__m64)(lsb * '"');
+  static const __m64 escape = (__m64)(lsb * '\\');
+
+  __m256i zeroV   = _mm256_setzero_si256();
+  __m256i msbV    = _mm256_set1_epi64x(msb);
+  __m256i lsbV    = _mm256_set1_epi64x(lsb);
+  __m256i spaceV  = _mm256_set1_epi64x(space);
+  __m256i quoteV  = _mm256_set1_epi64x(quote);
+  __m256i escapeV = _mm256_set1_epi64x(escape);
+
+  char *sp = buf;
+  size_t chunkLen = len / 32;
+  int chunkIdx = 0;
+  for (; chunkIdx < chunkLen; chunkIdx++) {
+    __m256i n       = _mm256_loadu_si256((const void *)sp);
+    __m256i spaceN  = _mm256_sub_epi64(n, spaceV);
+    __m256i quoteN  = _mm256_sub_epi64(_mm256_xor_si256(n, quoteV), lsbV);
+    __m256i escapeN = _mm256_sub_epi64(_mm256_xor_si256(n, escapeV), lsbV);
+
+    __m256i mask = _mm256_or_si256(_mm256_or_si256(_mm256_or_si256(n, spaceN), quoteN), escapeN);
+    int movemask = _mm256_movemask_epi8(_mm256_and_si256(mask, msbV));
+    if (movemask != 0) {
+      return __builtin_ctz(movemask);
+    }
+    sp += 32;
+  }
+  int idx = 32 * chunkLen;
+  int remainLen = len - idx;
   if (remainLen >= 16) {
     return idx + findEscapeIndex128(sp, remainLen);
   } else if (remainLen >= 8) {

internal/cmd/generator/simd/string_avx.go

@@ -2,6 +2,18 @@ package encoder
 
 import "unsafe"
 
+//go:nosplit
+//go:noescape
+func _findHTMLEscapeIndex64(buf unsafe.Pointer, len int) (ret int)
+
+//go:nosplit
+//go:noescape
+func _findHTMLEscapeIndex128(buf unsafe.Pointer, len int) (ret int)
+
+//go:nosplit
+//go:noescape
+func _findHTMLEscapeIndex256(buf unsafe.Pointer, len int) (ret int)
+
 //go:nosplit
 //go:noescape
 func _findEscapeIndex64(buf unsafe.Pointer, len int) (ret int)

internal/cmd/generator/vm.go.tmpl

@@ -402,12 +402,13 @@ func Run(ctx *encoder.RuntimeContext, b []byte, codeSet *encoder.OpcodeSet) ([]b
 				code = code.End.Next
 				break
 			}
-			b = appendStructHead(ctx, b)
+			unorderedMap := (ctx.Option.Flag & encoder.UnorderedMapOption) != 0
-			mapCtx := encoder.NewMapContext(mlen)
+			mapCtx := encoder.NewMapContext(mlen, unorderedMap)
 			mapiterinit(code.Type, uptr, &mapCtx.Iter)
 			store(ctxptr, code.Idx, uintptr(unsafe.Pointer(mapCtx)))
 			ctx.KeepRefs = append(ctx.KeepRefs, unsafe.Pointer(mapCtx))
-			if (ctx.Option.Flag & encoder.UnorderedMapOption) != 0 {
+			b = appendStructHead(ctx, b)
+			if unorderedMap {
 				b = appendMapKeyIndent(ctx, code.Next, b)
 			} else {
 				mapCtx.Start = len(b)

internal/encoder/encoder.go

@@ -259,12 +259,14 @@ var mapContextPool = sync.Pool{
 	},
 }
 
-func NewMapContext(mapLen int) *MapContext {
+func NewMapContext(mapLen int, unorderedMap bool) *MapContext {
 	ctx := mapContextPool.Get().(*MapContext)
-	if len(ctx.Slice.Items) < mapLen {
+	if !unorderedMap {
-		ctx.Slice.Items = make([]MapItem, mapLen)
+		if len(ctx.Slice.Items) < mapLen {
-	} else {
+			ctx.Slice.Items = make([]MapItem, mapLen)
-		ctx.Slice.Items = ctx.Slice.Items[:mapLen]
+		} else {
+			ctx.Slice.Items = ctx.Slice.Items[:mapLen]
+		}
 	}
 	ctx.Buf = ctx.Buf[:0]
 	ctx.Iter = mapIter{}

internal/encoder/string.go

@@ -1,7 +1,6 @@
 package encoder
 
 import (
-	"math/bits"
 	"reflect"
 	"unsafe"
 
@@ -374,11 +373,11 @@ func AppendString(ctx *RuntimeContext, buf []byte, s string) []byte {
 	switch valLen {
 	case 1, 2, 3, 4, 5, 6, 7:
 	case 8, 9, 10, 11, 12, 13, 14, 15:
-		j = _findEscapeIndex64((*runtime.SliceHeader)(unsafe.Pointer(&s)).Data, len(s))
+		j = _findHTMLEscapeIndex64((*runtime.SliceHeader)(unsafe.Pointer(&s)).Data, len(s))
 	case 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31:
-		j = _findEscapeIndex128((*runtime.SliceHeader)(unsafe.Pointer(&s)).Data, len(s))
+		j = _findHTMLEscapeIndex128((*runtime.SliceHeader)(unsafe.Pointer(&s)).Data, len(s))
 	default:
-		j = _findEscapeIndex256((*runtime.SliceHeader)(unsafe.Pointer(&s)).Data, len(s))
+		j = _findHTMLEscapeIndex256((*runtime.SliceHeader)(unsafe.Pointer(&s)).Data, len(s))
 	}
 	for j < valLen {
 		c := s[j]
@@ -479,30 +478,15 @@ func appendString(buf []byte, s string) []byte {
 	var (
 		i, j int
 	)
-	if valLen >= 8 {
+	switch valLen {
-		chunks := stringToUint64Slice(s)
+	case 1, 2, 3, 4, 5, 6, 7:
-		for _, n := range chunks {
+	case 8, 9, 10, 11, 12, 13, 14, 15:
-			// combine masks before checking for the MSB of each byte. We include
+		j = _findEscapeIndex64((*runtime.SliceHeader)(unsafe.Pointer(&s)).Data, len(s))
-			// `n` in the mask to check whether any of the *input* byte MSBs were
+	case 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31:
-			// set (i.e. the byte was outside the ASCII range).
+		j = _findEscapeIndex128((*runtime.SliceHeader)(unsafe.Pointer(&s)).Data, len(s))
-			mask := n | (n - (lsb * 0x20)) |
+	default:
-				((n ^ (lsb * '"')) - lsb) |
+		j = _findEscapeIndex256((*runtime.SliceHeader)(unsafe.Pointer(&s)).Data, len(s))
-				((n ^ (lsb * '\\')) - lsb)
-			if (mask & msb) != 0 {
-				j = bits.TrailingZeros64(mask&msb) / 8
-				goto ESCAPE_END
-			}
-		}
-		valLen := len(s)
-		for i := len(chunks) * 8; i < valLen; i++ {
-			if needEscape[s[i]] {
-				j = i
-				goto ESCAPE_END
-			}
-		}
-		return append(append(buf, s...), '"')
 	}
-ESCAPE_END:
 	for j < valLen {
 		c := s[j]
 

internal/encoder/string_avx.go

@@ -2,6 +2,18 @@ package encoder
 
 import "unsafe"
 
+//go:nosplit
+//go:noescape
+func _findHTMLEscapeIndex64(buf unsafe.Pointer, len int) (ret int)
+
+//go:nosplit
+//go:noescape
+func _findHTMLEscapeIndex128(buf unsafe.Pointer, len int) (ret int)
+
+//go:nosplit
+//go:noescape
+func _findHTMLEscapeIndex256(buf unsafe.Pointer, len int) (ret int)
+
 //go:nosplit
 //go:noescape
 func _findEscapeIndex64(buf unsafe.Pointer, len int) (ret int)

internal/encoder/string_avx.s

@@ -1,7 +1,7 @@
 //+build !noasm !appengine
 // AUTO-GENERATED BY C2GOASM -- DO NOT EDIT
 
-TEXT ·_findEscapeIndex64(SB), $0-24
+TEXT ·_findHTMLEscapeIndex64(SB), $0-24
 
 	MOVQ buf+0(FP), DI
 	MOVQ len+8(FP), SI
@@ -84,7 +84,7 @@ DATA LCDATA1<>+0x070(SB)/8, $0x8080808080808080
 DATA LCDATA1<>+0x078(SB)/8, $0x8080808080808080
 GLOBL LCDATA1<>(SB), 8, $128
 
-TEXT ·_findEscapeIndex128(SB), $16-24
+TEXT ·_findHTMLEscapeIndex128(SB), $16-24
 
 	MOVQ buf+0(FP), DI
 	MOVQ len+8(FP), SI
@@ -237,7 +237,7 @@ DATA LCDATA2<>+0x0d0(SB)/8, $0x8080808080808080
 DATA LCDATA2<>+0x0d8(SB)/8, $0x8080808080808080
 GLOBL LCDATA2<>(SB), 8, $224
 
-TEXT ·_findEscapeIndex256(SB), $16-24
+TEXT ·_findHTMLEscapeIndex256(SB), $16-24
 
 	MOVQ buf+0(FP), DI
 	MOVQ len+8(FP), SI
@@ -470,3 +470,349 @@ LBB2_25:
 	VZEROUPPER
 	MOVQ AX, ret+16(FP)
 	RET
+
+TEXT ·_findEscapeIndex64(SB), $0-24
+
+	MOVQ buf+0(FP), DI
+	MOVQ len+8(FP), SI
+
+	WORD $0xf089                           // mov    eax, esi
+	WORD $0xf8c1; BYTE $0x1f               // sar    eax, 31
+	WORD $0xe8c1; BYTE $0x1d               // shr    eax, 29
+	WORD $0xf001                           // add    eax, esi
+	WORD $0xf8c1; BYTE $0x03               // sar    eax, 3
+	WORD $0x9848                           // cdqe
+	WORD $0xc683; BYTE $0x07               // add    esi, 7
+	WORD $0xfe83; BYTE $0x0f               // cmp    esi, 15
+	JB   LBB3_5
+	QUAD $0xfefefefefeffbe49; WORD $0xfefe // mov    r14, -72340172838076673
+	WORD $0xd231                           // xor    edx, edx
+	QUAD $0xdfdfdfdfdfe0b849; WORD $0xdfdf // mov    r8, -2314885530818453536
+	QUAD $0x222222222222b949; WORD $0x2222 // mov    r9, 2459565876494606882
+	QUAD $0x5c5c5c5c5c5cba49; WORD $0x5c5c // mov    r10, 6655295901103053916
+	QUAD $0x808080808080bb49; WORD $0x8080 // mov    r11, -9187201950435737472
+
+LBB3_2:
+	LONG $0xd7348b48         // mov    rsi, qword [rdi + 8*rdx]
+	LONG $0x061c8d4a         // lea    rbx, [rsi + r8]
+	WORD $0x0948; BYTE $0xf3 // or    rbx, rsi
+	WORD $0x8948; BYTE $0xf1 // mov    rcx, rsi
+	WORD $0x314c; BYTE $0xc9 // xor    rcx, r9
+	WORD $0x014c; BYTE $0xf1 // add    rcx, r14
+	WORD $0x0948; BYTE $0xd9 // or    rcx, rbx
+	WORD $0x314c; BYTE $0xd6 // xor    rsi, r10
+	WORD $0x014c; BYTE $0xf6 // add    rsi, r14
+	WORD $0x0948; BYTE $0xce // or    rsi, rcx
+	WORD $0x214c; BYTE $0xde // and    rsi, r11
+	JNE  LBB3_3
+	LONG $0x01c28348         // add    rdx, 1
+	WORD $0x3948; BYTE $0xc2 // cmp    rdx, rax
+	JB   LBB3_2
+
+LBB3_5:
+	LONG $0x03e0c148 // shl    rax, 3
+	JMP  LBB3_6
+
+LBB3_3:
+	WORD $0xbc0f; BYTE $0xc6 // bsf    eax, esi
+	WORD $0xe8c1; BYTE $0x03 // shr    eax, 3
+
+LBB3_6:
+	MOVQ AX, ret+16(FP)
+	RET
+
+DATA LCDATA3<>+0x000(SB)/8, $0xdfdfdfdfdfdfdfe0
+DATA LCDATA3<>+0x008(SB)/8, $0xdfdfdfdfdfdfdfe0
+DATA LCDATA3<>+0x010(SB)/8, $0x2222222222222222
+DATA LCDATA3<>+0x018(SB)/8, $0x2222222222222222
+DATA LCDATA3<>+0x020(SB)/8, $0xfefefefefefefeff
+DATA LCDATA3<>+0x028(SB)/8, $0xfefefefefefefeff
+DATA LCDATA3<>+0x030(SB)/8, $0x5c5c5c5c5c5c5c5c
+DATA LCDATA3<>+0x038(SB)/8, $0x5c5c5c5c5c5c5c5c
+DATA LCDATA3<>+0x040(SB)/8, $0x8080808080808080
+DATA LCDATA3<>+0x048(SB)/8, $0x8080808080808080
+GLOBL LCDATA3<>(SB), 8, $80
+
+TEXT ·_findEscapeIndex128(SB), $0-24
+
+	MOVQ buf+0(FP), DI
+	MOVQ len+8(FP), SI
+	LEAQ LCDATA3<>(SB), BP
+
+	WORD $0xf089                 // mov    eax, esi
+	WORD $0xf8c1; BYTE $0x1f     // sar    eax, 31
+	WORD $0xe8c1; BYTE $0x1c     // shr    eax, 28
+	WORD $0xf001                 // add    eax, esi
+	WORD $0xf8c1; BYTE $0x04     // sar    eax, 4
+	WORD $0x4e8d; BYTE $0x0f     // lea    ecx, [rsi + 15]
+	WORD $0xf983; BYTE $0x1f     // cmp    ecx, 31
+	JB   LBB4_5
+	WORD $0x6348; BYTE $0xc8     // movsxd    rcx, eax
+	WORD $0xd231                 // xor    edx, edx
+	LONG $0x456f79c5; BYTE $0x00 // vmovdqa    xmm8, oword 0[rbp] /* [rip + .LCPI4_0] */
+	LONG $0x4d6ff9c5; BYTE $0x10 // vmovdqa    xmm1, oword 16[rbp] /* [rip + .LCPI4_1] */
+	LONG $0x556ff9c5; BYTE $0x20 // vmovdqa    xmm2, oword 32[rbp] /* [rip + .LCPI4_2] */
+	LONG $0x5d6ff9c5; BYTE $0x30 // vmovdqa    xmm3, oword 48[rbp] /* [rip + .LCPI4_3] */
+	LONG $0x656ff9c5; BYTE $0x40 // vmovdqa    xmm4, oword 64[rbp] /* [rip + .LCPI4_4] */
+
+LBB4_2:
+	LONG $0x2f6ffac5             // vmovdqu    xmm5, oword [rdi]
+	LONG $0xd451c1c4; BYTE $0xf0 // vpaddq    xmm6, xmm5, xmm8
+	LONG $0xf9efd1c5             // vpxor    xmm7, xmm5, xmm1
+	LONG $0xfad4c1c5             // vpaddq    xmm7, xmm7, xmm2
+	LONG $0xc3efd1c5             // vpxor    xmm0, xmm5, xmm3
+	LONG $0xc2d4f9c5             // vpaddq    xmm0, xmm0, xmm2
+	LONG $0xedebc9c5             // vpor    xmm5, xmm6, xmm5
+	LONG $0xefebd1c5             // vpor    xmm5, xmm5, xmm7
+	LONG $0xc0ebd1c5             // vpor    xmm0, xmm5, xmm0
+	LONG $0xc4dbf9c5             // vpand    xmm0, xmm0, xmm4
+	LONG $0xd8d7f9c5             // vpmovmskb    ebx, xmm0
+	WORD $0xdb85                 // test    ebx, ebx
+	JNE  LBB4_3
+	LONG $0x10c78348             // add    rdi, 16
+	LONG $0x01c28348             // add    rdx, 1
+	WORD $0x3948; BYTE $0xca     // cmp    rdx, rcx
+	JB   LBB4_2
+
+LBB4_5:
+	WORD $0xe0c1; BYTE $0x04               // shl    eax, 4
+	WORD $0xc629                           // sub    esi, eax
+	WORD $0x9848                           // cdqe
+	WORD $0xfe83; BYTE $0x08               // cmp    esi, 8
+	JL   LBB4_12
+	QUAD $0xfefefefefeffb849; WORD $0xfefe // mov    r8, -72340172838076673
+	WORD $0x8941; BYTE $0xf7               // mov    r15d, esi
+	LONG $0x03efc141                       // shr    r15d, 3
+	WORD $0xc931                           // xor    ecx, ecx
+	QUAD $0xdfdfdfdfdfe0b949; WORD $0xdfdf // mov    r9, -2314885530818453536
+	QUAD $0x222222222222ba49; WORD $0x2222 // mov    r10, 2459565876494606882
+	QUAD $0x5c5c5c5c5c5cbb49; WORD $0x5c5c // mov    r11, 6655295901103053916
+	QUAD $0x808080808080be49; WORD $0x8080 // mov    r14, -9187201950435737472
+
+LBB4_7:
+	LONG $0xcf1c8b48         // mov    rbx, qword [rdi + 8*rcx]
+	LONG $0x0b248d4e         // lea    r12, [rbx + r9]
+	WORD $0x0949; BYTE $0xdc // or    r12, rbx
+	WORD $0x8948; BYTE $0xda // mov    rdx, rbx
+	WORD $0x314c; BYTE $0xd2 // xor    rdx, r10
+	WORD $0x014c; BYTE $0xc2 // add    rdx, r8
+	WORD $0x094c; BYTE $0xe2 // or    rdx, r12
+	WORD $0x314c; BYTE $0xdb // xor    rbx, r11
+	WORD $0x014c; BYTE $0xc3 // add    rbx, r8
+	WORD $0x0948; BYTE $0xd3 // or    rbx, rdx
+	WORD $0x214c; BYTE $0xf3 // and    rbx, r14
+	JNE  LBB4_8
+	LONG $0x01c18348         // add    rcx, 1
+	WORD $0x394c; BYTE $0xf9 // cmp    rcx, r15
+	JB   LBB4_7
+	WORD $0xe683; BYTE $0xf8 // and    esi, -8
+	JMP  LBB4_11
+
+LBB4_3:
+	WORD $0xbc0f; BYTE $0xc3 // bsf    eax, ebx
+	JMP  LBB4_12
+
+LBB4_8:
+	WORD $0xbc0f; BYTE $0xf3 // bsf    esi, ebx
+	WORD $0xeec1; BYTE $0x03 // shr    esi, 3
+
+LBB4_11:
+	WORD $0xf189             // mov    ecx, esi
+	WORD $0x0148; BYTE $0xc8 // add    rax, rcx
+
+LBB4_12:
+	MOVQ AX, ret+16(FP)
+	RET
+
+DATA LCDATA4<>+0x000(SB)/8, $0xdfdfdfdfdfdfdfe0
+DATA LCDATA4<>+0x008(SB)/8, $0x2222222222222222
+DATA LCDATA4<>+0x010(SB)/8, $0xfefefefefefefeff
+DATA LCDATA4<>+0x018(SB)/8, $0x5c5c5c5c5c5c5c5c
+DATA LCDATA4<>+0x020(SB)/8, $0x8080808080808080
+DATA LCDATA4<>+0x028(SB)/8, $0x8080808080808080
+DATA LCDATA4<>+0x030(SB)/8, $0x8080808080808080
+DATA LCDATA4<>+0x038(SB)/8, $0x8080808080808080
+DATA LCDATA4<>+0x040(SB)/8, $0xdfdfdfdfdfdfdfe0
+DATA LCDATA4<>+0x048(SB)/8, $0xdfdfdfdfdfdfdfe0
+DATA LCDATA4<>+0x050(SB)/8, $0x2222222222222222
+DATA LCDATA4<>+0x058(SB)/8, $0x2222222222222222
+DATA LCDATA4<>+0x060(SB)/8, $0xfefefefefefefeff
+DATA LCDATA4<>+0x068(SB)/8, $0xfefefefefefefeff
+DATA LCDATA4<>+0x070(SB)/8, $0x5c5c5c5c5c5c5c5c
+DATA LCDATA4<>+0x078(SB)/8, $0x5c5c5c5c5c5c5c5c
+DATA LCDATA4<>+0x080(SB)/8, $0x8080808080808080
+DATA LCDATA4<>+0x088(SB)/8, $0x8080808080808080
+GLOBL LCDATA4<>(SB), 8, $144
+
+TEXT ·_findEscapeIndex256(SB), $0-24
+
+	MOVQ buf+0(FP), DI
+	MOVQ len+8(FP), SI
+	LEAQ LCDATA4<>(SB), BP
+
+	WORD $0xf089                   // mov    eax, esi
+	WORD $0xf8c1; BYTE $0x1f       // sar    eax, 31
+	WORD $0xe8c1; BYTE $0x1b       // shr    eax, 27
+	WORD $0xf001                   // add    eax, esi
+	WORD $0xf8c1; BYTE $0x05       // sar    eax, 5
+	WORD $0x4e8d; BYTE $0x1f       // lea    ecx, [rsi + 31]
+	WORD $0xf983; BYTE $0x3f       // cmp    ecx, 63
+	JB   LBB5_4
+	WORD $0x6348; BYTE $0xc8       // movsxd    rcx, eax
+	LONG $0x597de2c4; WORD $0x0045 // vpbroadcastq    ymm0, qword 0[rbp] /* [rip + .LCPI5_0] */
+	LONG $0x597de2c4; WORD $0x084d // vpbroadcastq    ymm1, qword 8[rbp] /* [rip + .LCPI5_1] */
+	LONG $0x597de2c4; WORD $0x1055 // vpbroadcastq    ymm2, qword 16[rbp] /* [rip + .LCPI5_2] */
+	WORD $0xd231                   // xor    edx, edx
+	LONG $0x597de2c4; WORD $0x185d // vpbroadcastq    ymm3, qword 24[rbp] /* [rip + .LCPI5_3] */
+	LONG $0x656ffdc5; BYTE $0x20   // vmovdqa    ymm4, yword 32[rbp] /* [rip + .LCPI5_4] */
+
+LBB5_2:
+	LONG $0x2f6ffec5             // vmovdqu    ymm5, yword [rdi]
+	LONG $0xf0d4d5c5             // vpaddq    ymm6, ymm5, ymm0
+	LONG $0xf9efd5c5             // vpxor    ymm7, ymm5, ymm1
+	LONG $0xfad4c5c5             // vpaddq    ymm7, ymm7, ymm2
+	LONG $0xc3ef55c5             // vpxor    ymm8, ymm5, ymm3
+	LONG $0xc2d43dc5             // vpaddq    ymm8, ymm8, ymm2
+	LONG $0xedebcdc5             // vpor    ymm5, ymm6, ymm5
+	LONG $0xefebd5c5             // vpor    ymm5, ymm5, ymm7
+	LONG $0xeb55c1c4; BYTE $0xe8 // vpor    ymm5, ymm5, ymm8
+	LONG $0xecdbd5c5             // vpand    ymm5, ymm5, ymm4
+	LONG $0xddd7fdc5             // vpmovmskb    ebx, ymm5
+	WORD $0xdb85                 // test    ebx, ebx
+	JNE  LBB5_18
+	LONG $0x20c78348             // add    rdi, 32
+	LONG $0x01c28348             // add    rdx, 1
+	WORD $0x3948; BYTE $0xca     // cmp    rdx, rcx
+	JB   LBB5_2
+
+LBB5_4:
+	WORD $0xe0c1; BYTE $0x05     // shl    eax, 5
+	WORD $0xc629                 // sub    esi, eax
+	WORD $0xfe83; BYTE $0x10     // cmp    esi, 16
+	JL   LBB5_13
+	WORD $0x634c; BYTE $0xc0     // movsxd    r8, eax
+	WORD $0xf089                 // mov    eax, esi
+	WORD $0xe8c1; BYTE $0x04     // shr    eax, 4
+	WORD $0xc931                 // xor    ecx, ecx
+	LONG $0x456f79c5; BYTE $0x40 // vmovdqa    xmm8, oword 64[rbp] /* [rip + .LCPI5_5] */
+	LONG $0x4d6ff9c5; BYTE $0x50 // vmovdqa    xmm1, oword 80[rbp] /* [rip + .LCPI5_6] */
+	LONG $0x556ff9c5; BYTE $0x60 // vmovdqa    xmm2, oword 96[rbp] /* [rip + .LCPI5_7] */
+	LONG $0x5d6ff9c5; BYTE $0x70 // vmovdqa    xmm3, oword 112[rbp] /* [rip + .LCPI5_8] */
+	QUAD $0x00000080a56ff9c5     // vmovdqa    xmm4, oword 128[rbp] /* [rip + .LCPI5_9] */
+
+LBB5_6:
+	LONG $0x2f6ffac5                       // vmovdqu    xmm5, oword [rdi]
+	LONG $0xd451c1c4; BYTE $0xf0           // vpaddq    xmm6, xmm5, xmm8
+	LONG $0xf9efd1c5                       // vpxor    xmm7, xmm5, xmm1
+	LONG $0xfad4c1c5                       // vpaddq    xmm7, xmm7, xmm2
+	LONG $0xc3efd1c5                       // vpxor    xmm0, xmm5, xmm3
+	LONG $0xc2d4f9c5                       // vpaddq    xmm0, xmm0, xmm2
+	LONG $0xedebc9c5                       // vpor    xmm5, xmm6, xmm5
+	LONG $0xefebd1c5                       // vpor    xmm5, xmm5, xmm7
+	LONG $0xc0ebd1c5                       // vpor    xmm0, xmm5, xmm0
+	LONG $0xc4dbf9c5                       // vpand    xmm0, xmm0, xmm4
+	LONG $0xd0d7f9c5                       // vpmovmskb    edx, xmm0
+	WORD $0xd285                           // test    edx, edx
+	JNE  LBB5_19
+	LONG $0x10c78348                       // add    rdi, 16
+	LONG $0x01c18348                       // add    rcx, 1
+	WORD $0x3948; BYTE $0xc1               // cmp    rcx, rax
+	JB   LBB5_6
+	WORD $0xf089                           // mov    eax, esi
+	WORD $0xe083; BYTE $0xf0               // and    eax, -16
+	WORD $0xc629                           // sub    esi, eax
+	WORD $0x9848                           // cdqe
+	WORD $0xfe83; BYTE $0x08               // cmp    esi, 8
+	JL   LBB5_24
+	QUAD $0xfefefefefeffbf49; WORD $0xfefe // mov    r15, -72340172838076673
+	QUAD $0xdfdfdfdfdfe0b949; WORD $0xdfdf // mov    r9, -2314885530818453536
+	QUAD $0x808080808080ba49; WORD $0x8080 // mov    r10, -9187201950435737472
+	QUAD $0x5c5c5c5c5c5cbb49; WORD $0x5c5c // mov    r11, 6655295901103053916
+	QUAD $0x222222222222be49; WORD $0x2222 // mov    r14, 2459565876494606882
+	WORD $0x8941; BYTE $0xf4               // mov    r12d, esi
+	LONG $0x03ecc141                       // shr    r12d, 3
+	WORD $0xc931                           // xor    ecx, ecx
+
+LBB5_10:
+	LONG $0xcf148b48         // mov    rdx, qword [rdi + 8*rcx]
+	LONG $0x0a2c8d4e         // lea    r13, [rdx + r9]
+	WORD $0x0949; BYTE $0xd5 // or    r13, rdx
+	WORD $0x8948; BYTE $0xd3 // mov    rbx, rdx
+	WORD $0x314c; BYTE $0xf3 // xor    rbx, r14
+	WORD $0x014c; BYTE $0xfb // add    rbx, r15
+	WORD $0x094c; BYTE $0xeb // or    rbx, r13
+	WORD $0x314c; BYTE $0xda // xor    rdx, r11
+	WORD $0x014c; BYTE $0xfa // add    rdx, r15
+	WORD $0x0948; BYTE $0xda // or    rdx, rbx
+	WORD $0x214c; BYTE $0xd2 // and    rdx, r10
+	JNE  LBB5_22
+	LONG $0x01c18348         // add    rcx, 1
+	WORD $0x394c; BYTE $0xe1 // cmp    rcx, r12
+	JB   LBB5_10
+	WORD $0xe683; BYTE $0xf8 // and    esi, -8
+	JMP  LBB5_23
+
+LBB5_13:
+	WORD $0x9848                           // cdqe
+	WORD $0xfe83; BYTE $0x08               // cmp    esi, 8
+	JL   LBB5_25
+	QUAD $0xfefefefefeffbe49; WORD $0xfefe // mov    r14, -72340172838076673
+	QUAD $0xdfdfdfdfdfe0b849; WORD $0xdfdf // mov    r8, -2314885530818453536
+	QUAD $0x808080808080b949; WORD $0x8080 // mov    r9, -9187201950435737472
+	QUAD $0x5c5c5c5c5c5cba49; WORD $0x5c5c // mov    r10, 6655295901103053916
+	QUAD $0x222222222222bb49; WORD $0x2222 // mov    r11, 2459565876494606882
+	WORD $0xeec1; BYTE $0x03               // shr    esi, 3
+	WORD $0xe683; BYTE $0x1f               // and    esi, 31
+	WORD $0xdb31                           // xor    ebx, ebx
+
+LBB5_15:
+	LONG $0xdf148b48         // mov    rdx, qword [rdi + 8*rbx]
+	LONG $0x023c8d4e         // lea    r15, [rdx + r8]
+	WORD $0x0949; BYTE $0xd7 // or    r15, rdx
+	WORD $0x8948; BYTE $0xd1 // mov    rcx, rdx
+	WORD $0x314c; BYTE $0xd9 // xor    rcx, r11
+	WORD $0x014c; BYTE $0xf1 // add    rcx, r14
+	WORD $0x094c; BYTE $0xf9 // or    rcx, r15
+	WORD $0x314c; BYTE $0xd2 // xor    rdx, r10
+	WORD $0x014c; BYTE $0xf2 // add    rdx, r14
+	WORD $0x0948; BYTE $0xca // or    rdx, rcx
+	WORD $0x214c; BYTE $0xca // and    rdx, r9
+	JNE  LBB5_20
+	LONG $0x01c38348         // add    rbx, 1
+	WORD $0x3948; BYTE $0xf3 // cmp    rbx, rsi
+	JB   LBB5_15
+	WORD $0xe6c1; BYTE $0x03 // shl    esi, 3
+	JMP  LBB5_21
+
+LBB5_18:
+	WORD $0xbc0f; BYTE $0xc3 // bsf    eax, ebx
+	JMP  LBB5_25
+
+LBB5_19:
+	WORD $0xbc0f; BYTE $0xc2 // bsf    eax, edx
+	JMP  LBB5_24
+
+LBB5_20:
+	WORD $0xbc0f; BYTE $0xf2 // bsf    esi, edx
+	WORD $0xeec1; BYTE $0x03 // shr    esi, 3
+
+LBB5_21:
+	WORD $0xf189             // mov    ecx, esi
+	WORD $0x0148; BYTE $0xc8 // add    rax, rcx
+	JMP  LBB5_25
+
+LBB5_22:
+	WORD $0xbc0f; BYTE $0xf2 // bsf    esi, edx
+	WORD $0xeec1; BYTE $0x03 // shr    esi, 3
+
+LBB5_23:
+	WORD $0xf189             // mov    ecx, esi
+	WORD $0x0148; BYTE $0xc8 // add    rax, rcx
+
+LBB5_24:
+	WORD $0x014c; BYTE $0xc0 // add    rax, r8
+
+LBB5_25:
+	VZEROUPPER
+	MOVQ AX, ret+16(FP)
+	RET

internal/encoder/vm/vm.go

@@ -402,12 +402,13 @@ func Run(ctx *encoder.RuntimeContext, b []byte, codeSet *encoder.OpcodeSet) ([]b
 				code = code.End.Next
 				break
 			}
-			b = appendStructHead(ctx, b)
+			unorderedMap := (ctx.Option.Flag & encoder.UnorderedMapOption) != 0
-			mapCtx := encoder.NewMapContext(mlen)
+			mapCtx := encoder.NewMapContext(mlen, unorderedMap)
 			mapiterinit(code.Type, uptr, &mapCtx.Iter)
 			store(ctxptr, code.Idx, uintptr(unsafe.Pointer(mapCtx)))
 			ctx.KeepRefs = append(ctx.KeepRefs, unsafe.Pointer(mapCtx))
-			if (ctx.Option.Flag & encoder.UnorderedMapOption) != 0 {
+			b = appendStructHead(ctx, b)
+			if unorderedMap {
 				b = appendMapKeyIndent(ctx, code.Next, b)
 			} else {
 				mapCtx.Start = len(b)

internal/encoder/vm_color/vm.go

@@ -402,12 +402,13 @@ func Run(ctx *encoder.RuntimeContext, b []byte, codeSet *encoder.OpcodeSet) ([]b
 				code = code.End.Next
 				break
 			}
-			b = appendStructHead(ctx, b)
+			unorderedMap := (ctx.Option.Flag & encoder.UnorderedMapOption) != 0
-			mapCtx := encoder.NewMapContext(mlen)
+			mapCtx := encoder.NewMapContext(mlen, unorderedMap)
 			mapiterinit(code.Type, uptr, &mapCtx.Iter)
 			store(ctxptr, code.Idx, uintptr(unsafe.Pointer(mapCtx)))
 			ctx.KeepRefs = append(ctx.KeepRefs, unsafe.Pointer(mapCtx))
-			if (ctx.Option.Flag & encoder.UnorderedMapOption) != 0 {
+			b = appendStructHead(ctx, b)
+			if unorderedMap {
 				b = appendMapKeyIndent(ctx, code.Next, b)
 			} else {
 				mapCtx.Start = len(b)

internal/encoder/vm_color_indent/vm.go

@@ -402,12 +402,13 @@ func Run(ctx *encoder.RuntimeContext, b []byte, codeSet *encoder.OpcodeSet) ([]b
 				code = code.End.Next
 				break
 			}
-			b = appendStructHead(ctx, b)
+			unorderedMap := (ctx.Option.Flag & encoder.UnorderedMapOption) != 0
-			mapCtx := encoder.NewMapContext(mlen)
+			mapCtx := encoder.NewMapContext(mlen, unorderedMap)
 			mapiterinit(code.Type, uptr, &mapCtx.Iter)
 			store(ctxptr, code.Idx, uintptr(unsafe.Pointer(mapCtx)))
 			ctx.KeepRefs = append(ctx.KeepRefs, unsafe.Pointer(mapCtx))
-			if (ctx.Option.Flag & encoder.UnorderedMapOption) != 0 {
+			b = appendStructHead(ctx, b)
+			if unorderedMap {
 				b = appendMapKeyIndent(ctx, code.Next, b)
 			} else {
 				mapCtx.Start = len(b)

internal/encoder/vm_indent/vm.go

@@ -402,12 +402,13 @@ func Run(ctx *encoder.RuntimeContext, b []byte, codeSet *encoder.OpcodeSet) ([]b
 				code = code.End.Next
 				break
 			}
-			b = appendStructHead(ctx, b)
+			unorderedMap := (ctx.Option.Flag & encoder.UnorderedMapOption) != 0
-			mapCtx := encoder.NewMapContext(mlen)
+			mapCtx := encoder.NewMapContext(mlen, unorderedMap)
 			mapiterinit(code.Type, uptr, &mapCtx.Iter)
 			store(ctxptr, code.Idx, uintptr(unsafe.Pointer(mapCtx)))
 			ctx.KeepRefs = append(ctx.KeepRefs, unsafe.Pointer(mapCtx))
-			if (ctx.Option.Flag & encoder.UnorderedMapOption) != 0 {
+			b = appendStructHead(ctx, b)
+			if unorderedMap {
 				b = appendMapKeyIndent(ctx, code.Next, b)
 			} else {
 				mapCtx.Start = len(b)