diff --git a/base64x_error_native_amd64_test.go b/base64x_error_native_amd64_test.go
index 8b62c03..16026f2 100644
--- a/base64x_error_native_amd64_test.go
+++ b/base64x_error_native_amd64_test.go
@@ -12,8 +12,8 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-//go:build amd64 && !noasm && !appengine
-// +build amd64,!noasm,!appengine
+//go:build !goexperiment.simd && amd64 && !noasm && !appengine
+// +build !goexperiment.simd,amd64,!noasm,!appengine
 
 package base64x
 
diff --git a/base64x_fallback_test.go b/base64x_fallback_test.go
index 1447270..7a445ff 100644
--- a/base64x_fallback_test.go
+++ b/base64x_fallback_test.go
@@ -1,5 +1,5 @@
-//go:build !amd64 || noasm || appengine
-// +build !amd64 noasm appengine
+//go:build goexperiment.simd || !amd64 || noasm || appengine
+// +build goexperiment.simd !amd64 noasm appengine
 
 /*
  * Copyright 2024 CloudWeGo Authors
diff --git a/base64x_native.go b/base64x_native.go
index 5643729..dd5c4f3 100644
--- a/base64x_native.go
+++ b/base64x_native.go
@@ -1,5 +1,5 @@
-//go:build amd64 && !noasm && !appengine
-// +build amd64,!noasm,!appengine
+//go:build !goexperiment.simd && amd64 && !noasm && !appengine
+// +build !goexperiment.simd,amd64,!noasm,!appengine
 
 /*
  * Copyright 2024 CloudWeGo Authors
diff --git a/base64x_simd_amd64.go b/base64x_simd_amd64.go
new file mode 100644
index 0000000..9d73099
--- /dev/null
+++ b/base64x_simd_amd64.go
@@ -0,0 +1,158 @@
+//go:build goexperiment.simd && amd64 && !noasm && !appengine
+
+/*
+ * Copyright 2024 CloudWeGo Authors
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package base64x
+
+import (
+	"encoding/base64"
+	"encoding/json"
+	"simd/archsimd"
+
+	"github.com/cloudwego/base64x/internal/simd"
+)
+
+func encode(out *[]byte, src []byte, mode int) {
+	if !archsimd.X86.AVX2() {
+		encodeStdlib(out, src, mode)
+		return
+	}
+	url := 0
+	if mode&_MODE_URL != 0 {
+		url = 1
+	}
+	pad := 1
+	if mode&_MODE_RAW != 0 {
+		pad = 0
+	}
+	enc := encodingForModeSimd(mode)
+	dst := *out
+	start := len(dst)
+	need := enc.EncodedLen(len(src))
+	dst = dst[:start+need]
+	simd.B64_encode(src, uint64(len(src)), dst[start:start+need], uint64(need), url, pad)
+	*out = dst[:start+need]
+}
+
+func decode(out *[]byte, src []byte, mode int) (int, error) {
+	input := src
+	if mode&_MODE_JSON != 0 {
+		unquoted, err := unquoteJSONBase64(src)
+		if err != nil {
+			return 0, err
+		}
+		input = unquoted
+	}
+	if !archsimd.X86.AVX2() {
+		return decodeStdlib(out, input, mode)
+	}
+
+	url := 0
+	if mode&_MODE_URL != 0 {
+		url = 1
+	}
+	pad := 1
+	if mode&_MODE_RAW != 0 {
+		pad = 0
+	}
+	enc := encodingForModeSimd(mode)
+	dst := *out
+	start := len(dst)
+	need := enc.DecodedLen(len(input))
+
+	// The vector decode loop writes in 32-byte chunks.
+	// If the destination slice capacity cannot safely hold 32 bytes from start,
+	// decode into a stack scratch buffer and copy back to avoid out-of-bounds writes.
+	if cap(dst)-start < 32 || cap(dst)-start < need+8 {
+		var tmp [256]byte
+		var buf []byte
+		allocCap := need + 32
+		if allocCap <= len(tmp) {
+			buf = tmp[:allocCap]
+		} else {
+			buf = make([]byte, allocCap)
+		}
+		var posErr uint64
+		w := simd.B64_decode(input, uint64(len(input)), buf, uint64(len(buf)), url, pad, &posErr)
+		if posErr != ^uint64(0) {
+			return 0, base64.CorruptInputError(posErr)
+		}
+		dst = dst[:start+int(w)]
+		copy(dst[start:], buf[:w])
+		*out = dst
+		return int(w), nil
+	}
+
+	dst = dst[:start+need]
+	var posErr uint64
+	availCap := uint64(cap(dst) - start)
+	w := simd.B64_decode(input, uint64(len(input)), dst[start:cap(dst)], availCap, url, pad, &posErr)
+	if posErr != ^uint64(0) {
+		return 0, base64.CorruptInputError(posErr)
+	}
+	*out = dst[:start+int(w)]
+	return int(w), nil
+}
+
+func encodingForModeSimd(mode int) *base64.Encoding {
+	switch mode & (_MODE_URL | _MODE_RAW) {
+	case _MODE_URL:
+		return base64.URLEncoding
+	case _MODE_RAW:
+		return base64.RawStdEncoding
+	case _MODE_URL | _MODE_RAW:
+		return base64.RawURLEncoding
+	default:
+		return base64.StdEncoding
+	}
+}
+
+func encodeStdlib(out *[]byte, src []byte, mode int) {
+	enc := encodingForModeSimd(mode)
+	dst := *out
+	start := len(dst)
+	need := enc.EncodedLen(len(src))
+	dst = dst[:start+need]
+	enc.Encode(dst[start:], src)
+	*out = dst
+}
+
+func decodeStdlib(out *[]byte, input []byte, mode int) (int, error) {
+	enc := encodingForModeSimd(mode)
+	dst := *out
+	start := len(dst)
+	buf := dst[start:cap(dst)]
+	n, err := enc.Decode(buf, input)
+	if err != nil {
+		return 0, err
+	}
+	*out = dst[:start+n]
+	return n, nil
+}
+
+func unquoteJSONBase64(src []byte) ([]byte, error) {
+	quoted := make([]byte, 0, len(src)+2)
+	quoted = append(quoted, '"')
+	quoted = append(quoted, src...)
+	quoted = append(quoted, '"')
+
+	var decoded string
+	if err := json.Unmarshal(quoted, &decoded); err != nil {
+		return nil, err
+	}
+	return []byte(decoded), nil
+}
diff --git a/base64x_subr_native_amd64.go b/base64x_subr_native_amd64.go
index 094ad84..b092604 100644
--- a/base64x_subr_native_amd64.go
+++ b/base64x_subr_native_amd64.go
@@ -12,8 +12,8 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-//go:build amd64 && !noasm && !appengine
-// +build amd64,!noasm,!appengine
+//go:build !goexperiment.simd && amd64 && !noasm && !appengine
+// +build !goexperiment.simd,amd64,!noasm,!appengine
 
 package base64x
 
diff --git a/internal/simd/b64_gen.go b/internal/simd/b64_gen.go
new file mode 100644
index 0000000..2c87e37
--- /dev/null
+++ b/internal/simd/b64_gen.go
@@ -0,0 +1,311 @@
+//go:build goexperiment.simd && amd64
+
+// Code generated by sgoiter emit go127; DO NOT EDIT.
+// sgoiter-stamp: sha256=8c4229c7b8ba80231b22b3f6f7e3bf620480aace1357e583e0881d0026d0bfa4
+package simd
+
+import (
+	"math/bits"
+	"simd/archsimd"
+)
+
+var (
+	TabEncodeCharsetStd = [64]byte{'A', 'B', 'C', 'D', 'E', 'F', 'G', 'H', 'I', 'J', 'K', 'L', 'M', 'N', 'O', 'P', 'Q', 'R', 'S', 'T', 'U', 'V', 'W', 'X', 'Y', 'Z', 'a', 'b', 'c', 'd', 'e', 'f', 'g', 'h', 'i', 'j', 'k', 'l', 'm', 'n', 'o', 'p', 'q', 'r', 's', 't', 'u', 'v', 'w', 'x', 'y', 'z', '0', '1', '2', '3', '4', '5', '6', '7', '8', '9', '+', '/'}
+	TabEncodeCharsetURL = [64]byte{'A', 'B', 'C', 'D', 'E', 'F', 'G', 'H', 'I', 'J', 'K', 'L', 'M', 'N', 'O', 'P', 'Q', 'R', 'S', 'T', 'U', 'V', 'W', 'X', 'Y', 'Z', 'a', 'b', 'c', 'd', 'e', 'f', 'g', 'h', 'i', 'j', 'k', 'l', 'm', 'n', 'o', 'p', 'q', 'r', 's', 't', 'u', 'v', 'w', 'x', 'y', 'z', '0', '1', '2', '3', '4', '5', '6', '7', '8', '9', '-', '_'}
+	VecEncodeShuffles   = [32]byte{1, 0, 2, 1, 4, 3, 5, 4, 7, 6, 8, 7, 10, 9, 11, 10, 1, 0, 2, 1, 4, 3, 5, 4, 7, 6, 8, 7, 10, 9, 11, 10}
+	VecEncodeCharsetStd = [32]byte{71, 252, 252, 252, 252, 252, 252, 252, 252, 252, 252, 237, 240, 'A', 0, 0, 71, 252, 252, 252, 252, 252, 252, 252, 252, 252, 252, 237, 240, 'A', 0, 0}
+	VecEncodeCharsetURL = [32]byte{71, 252, 252, 252, 252, 252, 252, 252, 252, 252, 252, 239, 32, 'A', 0, 0, 71, 252, 252, 252, 252, 252, 252, 252, 252, 252, 252, 239, 32, 'A', 0, 0}
+	VecPacking          = [32]byte{2, 1, 0, 6, 5, 4, 10, 9, 8, 14, 13, 12, 128, 128, 128, 128, 2, 1, 0, 6, 5, 4, 10, 9, 8, 14, 13, 12, 128, 128, 128, 128}
+	VecDecodeBits       = [32]byte{0x01, 0x02, 0x04, 0x08, 0x10, 0x20, 0x40, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x02, 0x04, 0x08, 0x10, 0x20, 0x40, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00}
+	VecDecodeTableStd   = [128]byte{0x00, 0x00, 0x13, 0x04, 0xbf, 0xbf, 0xb9, 0xb9, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x13, 0x04, 0xbf, 0xbf, 0xb9, 0xb9, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xa8, 0xf8, 0xf8, 0xf8, 0xf8, 0xf8, 0xf8, 0xf8, 0xf8, 0xf8, 0xf0, 0x54, 0x50, 0x50, 0x50, 0x54, 0xa8, 0xf8, 0xf8, 0xf8, 0xf8, 0xf8, 0xf8, 0xf8, 0xf8, 0xf8, 0xf0, 0x54, 0x50, 0x50, 0x50, 0x54, 0x2f, 0x2f, 0x2f, 0x2f, 0x2f, 0x2f, 0x2f, 0x2f, 0x2f, 0x2f, 0x2f, 0x2f, 0x2f, 0x2f, 0x2f, 0x2f, 0x2f, 0x2f, 0x2f, 0x2f, 0x2f, 0x2f, 0x2f, 0x2f, 0x2f, 0x2f, 0x2f, 0x2f, 0x2f, 0x2f, 0x2f, 0x2f, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10}
+	VecDecodeTableURL   = [128]byte{0x00, 0x00, 0x11, 0x04, 0xbf, 0xbf, 0xb9, 0xb9, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x11, 0x04, 0xbf, 0xbf, 0xb9, 0xb9, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xa8, 0xf8, 0xf8, 0xf8, 0xf8, 0xf8, 0xf8, 0xf8, 0xf8, 0xf8, 0xf0, 0x50, 0x50, 0x54, 0x50, 0x70, 0xa8, 0xf8, 0xf8, 0xf8, 0xf8, 0xf8, 0xf8, 0xf8, 0xf8, 0xf8, 0xf0, 0x50, 0x50, 0x54, 0x50, 0x70, 0x5f, 0x5f, 0x5f, 0x5f, 0x5f, 0x5f, 0x5f, 0x5f, 0x5f, 0x5f, 0x5f, 0x5f, 0x5f, 0x5f, 0x5f, 0x5f, 0x5f, 0x5f, 0x5f, 0x5f, 0x5f, 0x5f, 0x5f, 0x5f, 0x5f, 0x5f, 0x5f, 0x5f, 0x5f, 0x5f, 0x5f, 0x5f, 0xe0, 0xe0, 0xe0, 0xe0, 0xe0, 0xe0, 0xe0, 0xe0, 0xe0, 0xe0, 0xe0, 0xe0, 0xe0, 0xe0, 0xe0, 0xe0, 0xe0, 0xe0, 0xe0, 0xe0, 0xe0, 0xe0, 0xe0, 0xe0, 0xe0, 0xe0, 0xe0, 0xe0, 0xe0, 0xe0, 0xe0, 0xe0}
+	TabDecodeCharsetStd = [256]byte{0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 62, 0xff, 0xff, 0xff, 63, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff}
+	TabDecodeCharsetURL = [256]byte{0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 62, 0xff, 0xff, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 0xff, 0xff, 0xff, 0xff, 63, 0xff, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff}
+)
+
+func __avx2() int {
+	v4 := func() int {
+		if archsimd.X86.AVX2() {
+			return 1
+		}
+		return 0
+	}()
+	return v4
+}
+
+func b64_encode_tail(src []byte, src_len uint64, i uint64, dst []byte, o uint64, url_mode int, pad_mode int) uint64 {
+	var v9 uint32
+	var v11 []byte
+	var v91 uint64
+	if uint64(len(src)) < src_len {
+		src_len = uint64(len(src))
+	}
+	v11 = TabEncodeCharsetStd[:]
+	if url_mode != 0 {
+		v11 = TabEncodeCharsetURL[:]
+	}
+	for (src_len - i) >= 3 {
+		v9 = uint32((uint64(uint64(uint32(src[i]))<<16) | uint64(uint64(uint32(src[i+1]))<<8)) | uint64(uint32(src[i+2])))
+		v34 := uint64(uint64(v9)>>18) & 63
+		dst[o] = v11[v34]
+		v39 := uint64(uint64(v9)>>12) & 63
+		dst[o+1] = v11[v39]
+		v46 := uint64(uint64(v9)>>6) & 63
+		dst[o+2] = v11[v46]
+		v51 := v9 & 63
+		dst[o+3] = v11[v51]
+		i += 3
+		o += 4
+	}
+	v59 := src_len - i
+	if v59 == 0 {
+		return o
+	}
+	v9 = uint32(uint64(uint64(uint32(src[i])) << 16))
+	if v59 == 2 {
+		v9 = uint32(uint64(v9) | uint64(uint64(uint32(src[i+1]))<<8))
+	}
+	v76 := uint64(uint64(v9)>>18) & 63
+	dst[o] = v11[v76]
+	v81 := uint64(uint64(v9)>>12) & 63
+	dst[o+1] = v11[v81]
+	o += 2
+	if v59 == 2 {
+		v91 = uint64(uint64(v9)>>6) & 63
+		dst[o] = v11[v91]
+		o++
+		if pad_mode != 0 {
+			dst[o] = '='
+			o++
+		}
+	} else {
+		if pad_mode != 0 {
+			dst[o] = '='
+			dst[o+1] = '='
+			o += 2
+		}
+	}
+	return o
+}
+
+func B64_encode_scalar(src []byte, src_len uint64, dst []byte, dst_cap uint64, url_mode int, pad_mode int) uint64 {
+	if uint64(len(src)) < src_len {
+		src_len = uint64(len(src))
+	}
+	return uint64(b64_encode_tail(src[:], src_len, uint64(0), dst[:], uint64(0), url_mode, pad_mode))
+}
+
+func B64_encode_avx2(src []byte, src_len uint64, dst []byte, dst_cap uint64, url_mode int, pad_mode int) uint64 {
+	var v9 uint64
+	var v11 uint64
+	var v13 []byte
+	if uint64(len(src)) < src_len {
+		src_len = uint64(len(src))
+	}
+	v9 = 0
+	v11 = v9
+	v13 = VecEncodeCharsetStd[:]
+	if url_mode != 0 {
+		v13 = VecEncodeCharsetURL[:]
+	}
+	for {
+		v21 := ((src_len - v9) >= 28) && ((dst_cap - v11) >= 32)
+		if !v21 {
+			break
+		}
+		v41 := archsimd.BroadcastUint8x32(0).SetLo(archsimd.LoadUint8x16(src[v9:][:16])).SetHi(archsimd.LoadUint8x16(src[int(v9+12):][:16])).PermuteOrZeroGrouped(archsimd.LoadUint8x32(VecEncodeShuffles[0:32]).BitsToInt8())
+		v59 := v41.And(archsimd.BroadcastUint32x8(uint32(264305664)).ReshapeToUint8s()).ReshapeToUint16s().MulHigh(archsimd.BroadcastUint32x8(uint32(67108928)).ReshapeToUint8s().ReshapeToUint16s()).ReshapeToUint8s().Or(v41.And(archsimd.BroadcastUint32x8(uint32(4129776)).ReshapeToUint8s()).ReshapeToUint16s().BitsToInt16().Mul(archsimd.BroadcastUint32x8(uint32(16777232)).ReshapeToUint8s().ReshapeToUint16s().BitsToInt16()).ToBits().ReshapeToUint8s())
+		v59.Add(archsimd.LoadUint8x32(v13[0:32]).PermuteOrZeroGrouped(v59.SubSaturated(archsimd.BroadcastUint8x32(byte(51))).Or(archsimd.BroadcastUint8x32(byte(13)).And(archsimd.BroadcastUint8x32(byte(26)).BitsToInt8().Greater(v59.BitsToInt8()).ToInt8x32().ToBits())).BitsToInt8())).Store(dst[v11:][:32])
+		v9 += 24
+		v11 += 32
+	}
+	return uint64(b64_encode_tail(src[:], src_len, v9, dst[:], v11, url_mode, pad_mode))
+}
+
+func B64_encode(src []byte, src_len uint64, dst []byte, dst_cap uint64, url_mode int, pad_mode int) uint64 {
+	if uint64(len(src)) < src_len {
+		src_len = uint64(len(src))
+	}
+	if archsimd.X86.AVX2() {
+		return uint64(B64_encode_avx2(src[:], src_len, dst[:], dst_cap, url_mode, pad_mode))
+	}
+	return uint64(B64_encode_scalar(src[:], src_len, dst[:], dst_cap, url_mode, pad_mode))
+}
+
+func b64_decode_tail(src []byte, src_len uint64, si uint64, dst []byte, o uint64, url_mode int, pad_mode int, pos_err *uint64) uint64 {
+	var v10 []byte
+	var v11 uint32
+	var v14 int
+	var v15 int
+	var v16 uint64
+	var v24 int
+	if uint64(len(src)) < src_len {
+		src_len = uint64(len(src))
+	}
+	v10 = TabDecodeCharsetStd[:]
+	if url_mode != 0 {
+		v10 = TabDecodeCharsetURL[:]
+	}
+	for si < src_len {
+		v11 = 0
+		v15 = 4
+		v16 = 0xffffffffffffffff
+		v14 = 0
+		for v14 < 4 {
+			if si == src_len {
+				if v14 == 0 {
+					*pos_err = 0xffffffffffffffff
+					return o
+				}
+				if (v14 == 1) || (pad_mode != 0) {
+					*pos_err = si - uint64(v14)
+					return o
+				}
+				v15 = v14
+				break
+			}
+			v34 := src[si]
+			si++
+			v37 := v10[v34]
+			if v37 != 0xff {
+				v11 = uint32(uint64(v11) | uint64(uint64(uint32(v37))<<byte(18-(6*v14))))
+				v24 = v14 + 1
+				v14 = v24
+				continue
+			}
+			if (v34 == 10) || (v34 == 13) {
+				v14--
+				v24 = v14 + 1
+				v14 = v24
+				continue
+			}
+			if (pad_mode == 0) || (v34 != '=') {
+				*pos_err = si - 1
+				return o
+			}
+			if v14 < 2 {
+				*pos_err = si - 1
+				return o
+			}
+			if v14 == 2 {
+				for si < src_len {
+					var v71 bool
+					if src[si] == 10 {
+						v71 = true
+					} else {
+						v71 = src[si] == 13
+					}
+					if !v71 {
+						break
+					}
+					si++
+				}
+				if si == src_len {
+					*pos_err = src_len
+					return o
+				}
+				if src[si] != '=' {
+					*pos_err = si - 1
+					return o
+				}
+				si++
+			}
+			for si < src_len {
+				var v88 bool
+				if src[si] == 10 {
+					v88 = true
+				} else {
+					v88 = src[si] == 13
+				}
+				if !v88 {
+					break
+				}
+				si++
+			}
+			if si < src_len {
+				v16 = si
+			}
+			v15 = v14
+			break
+			v24 = v14 + 1
+			v14 = v24
+		}
+		dst[o] = byte(uint64(uint64(v11) >> 16))
+		if v15 >= 3 {
+			dst[o+1] = byte(uint64(uint64(v11) >> 8))
+		}
+		if v15 == 4 {
+			dst[o+2] = byte(v11)
+		}
+		o += uint64(v15 - 1)
+		if v16 != 0xffffffffffffffff {
+			*pos_err = v16
+			return o
+		}
+	}
+	*pos_err = 0xffffffffffffffff
+	return o
+}
+
+func B64_decode_scalar(src []byte, src_len uint64, dst []byte, dst_cap uint64, url_mode int, pad_mode int, pos_err *uint64) uint64 {
+	if uint64(len(src)) < src_len {
+		src_len = uint64(len(src))
+	}
+	return uint64(b64_decode_tail(src[:], src_len, uint64(0), dst[:], uint64(0), url_mode, pad_mode, pos_err))
+}
+
+func B64_decode_avx2(src []byte, src_len uint64, dst []byte, dst_cap uint64, url_mode int, pad_mode int, pos_err *uint64) uint64 {
+	var v11 uint64
+	var v13 uint64
+	var v17 []byte
+	if uint64(len(src)) < src_len {
+		src_len = uint64(len(src))
+	}
+	v11 = 0
+	v13 = v11
+	v17 = VecDecodeTableStd[:]
+	if url_mode != 0 {
+		v17 = VecDecodeTableURL[:]
+	}
+	for {
+		v25 := ((src_len - v11) >= 32) && ((dst_cap - v13) >= 32)
+		if !v25 {
+			break
+		}
+		v30 := archsimd.LoadUint8x32(src[v11:][:32])
+		v36 := archsimd.BroadcastUint8x32(byte(15))
+		v41 := v30.ReshapeToUint32s().ShiftAllRight(4).ReshapeToUint8s().And(v36)
+		v109 := archsimd.LoadUint32x8Array(&[8]uint32{uint32(0), uint32(1), uint32(2), uint32(4), uint32(5), uint32(6), uint32(3), uint32(7)}).ReshapeToUint8s()
+		v110 := v30.Add(archsimd.LoadUint8x32(v17[96:128]).IfElse(v30.Equal(archsimd.LoadUint8x32(v17[64:96])).ToInt8x32().ToBits().BitsToInt8().Less(archsimd.BroadcastInt8x32(0)), archsimd.LoadUint8x32(v17[0:32]).PermuteOrZeroGrouped(v41.BitsToInt8()))).And(archsimd.BroadcastUint8x32(byte(63))).DotProductPairsSaturated(archsimd.BroadcastUint32x8(uint32(20971840)).ReshapeToUint8s().BitsToInt8()).ToBits().ReshapeToUint8s().ReshapeToUint16s().BitsToInt16().DotProductPairs(archsimd.BroadcastUint32x8(uint32(69632)).ReshapeToUint8s().ReshapeToUint16s().BitsToInt16()).ToBits().ReshapeToUint8s().PermuteOrZeroGrouped(archsimd.LoadUint8x32(VecPacking[0:32]).BitsToInt8()).ReshapeToUint32s().Permute(v109.ReshapeToUint32s()).ReshapeToUint8s()
+		if int32(bits.TrailingZeros64(uint64(int64(int(int32(archsimd.LoadUint8x32(v17[32:64]).PermuteOrZeroGrouped(v30.And(v36).BitsToInt8()).And(archsimd.LoadUint8x32(VecDecodeBits[0:32]).PermuteOrZeroGrouped(v41.BitsToInt8())).Equal(archsimd.BroadcastUint8x32(byte(0))).ToBits())))|int64(-4294967296)))) < 32 {
+			break
+		}
+		v110.Store(dst[v13:][:32])
+		v11 += 32
+		v13 += 24
+	}
+	return uint64(b64_decode_tail(src[:], src_len, v11, dst[:], v13, url_mode, pad_mode, pos_err))
+}
+
+func B64_decode(src []byte, src_len uint64, dst []byte, dst_cap uint64, url_mode int, pad_mode int, pos_err *uint64) uint64 {
+	if uint64(len(src)) < src_len {
+		src_len = uint64(len(src))
+	}
+	if archsimd.X86.AVX2() {
+		return uint64(B64_decode_avx2(src[:], src_len, dst[:], dst_cap, url_mode, pad_mode, pos_err))
+	}
+	return uint64(B64_decode_scalar(src[:], src_len, dst[:], dst_cap, url_mode, pad_mode, pos_err))
+}
+
+// IRCosts counted from IR opcodes via rodata.IROpCost; not a kernel-name table.
+const (
+	__avx2_IRCost            = uint64(6)
+	B64_encode_tail_IRCost   = uint64(122)
+	B64_encode_scalar_IRCost = uint64(4)
+	B64_encode_avx2_IRCost   = uint64(88)
+	B64_encode_IRCost        = uint64(5)
+	B64_decode_tail_IRCost   = uint64(144)
+	B64_decode_scalar_IRCost = uint64(4)
+	B64_decode_avx2_IRCost   = uint64(125)
+	B64_decode_IRCost        = uint64(5)
+)
diff --git a/internal/simd/b64_nosimd_gen.go b/internal/simd/b64_nosimd_gen.go
new file mode 100644
index 0000000..acd21ac
--- /dev/null
+++ b/internal/simd/b64_nosimd_gen.go
@@ -0,0 +1,227 @@
+//go:build !goexperiment.simd || !amd64
+
+// Code generated by sgoiter emit go127; DO NOT EDIT.
+// sgoiter-stamp: sha256=047ae148fa4614d63ab852dbd74f52171de727ab3e8e0fd127dfe8d2b6611cfb
+package simd
+
+var (
+	TabEncodeCharsetStd = [64]byte{'A', 'B', 'C', 'D', 'E', 'F', 'G', 'H', 'I', 'J', 'K', 'L', 'M', 'N', 'O', 'P', 'Q', 'R', 'S', 'T', 'U', 'V', 'W', 'X', 'Y', 'Z', 'a', 'b', 'c', 'd', 'e', 'f', 'g', 'h', 'i', 'j', 'k', 'l', 'm', 'n', 'o', 'p', 'q', 'r', 's', 't', 'u', 'v', 'w', 'x', 'y', 'z', '0', '1', '2', '3', '4', '5', '6', '7', '8', '9', '+', '/'}
+	TabEncodeCharsetURL = [64]byte{'A', 'B', 'C', 'D', 'E', 'F', 'G', 'H', 'I', 'J', 'K', 'L', 'M', 'N', 'O', 'P', 'Q', 'R', 'S', 'T', 'U', 'V', 'W', 'X', 'Y', 'Z', 'a', 'b', 'c', 'd', 'e', 'f', 'g', 'h', 'i', 'j', 'k', 'l', 'm', 'n', 'o', 'p', 'q', 'r', 's', 't', 'u', 'v', 'w', 'x', 'y', 'z', '0', '1', '2', '3', '4', '5', '6', '7', '8', '9', '-', '_'}
+	VecEncodeShuffles   = [32]byte{1, 0, 2, 1, 4, 3, 5, 4, 7, 6, 8, 7, 10, 9, 11, 10, 1, 0, 2, 1, 4, 3, 5, 4, 7, 6, 8, 7, 10, 9, 11, 10}
+	VecEncodeCharsetStd = [32]byte{71, 252, 252, 252, 252, 252, 252, 252, 252, 252, 252, 237, 240, 'A', 0, 0, 71, 252, 252, 252, 252, 252, 252, 252, 252, 252, 252, 237, 240, 'A', 0, 0}
+	VecEncodeCharsetURL = [32]byte{71, 252, 252, 252, 252, 252, 252, 252, 252, 252, 252, 239, 32, 'A', 0, 0, 71, 252, 252, 252, 252, 252, 252, 252, 252, 252, 252, 239, 32, 'A', 0, 0}
+	VecPacking          = [32]byte{2, 1, 0, 6, 5, 4, 10, 9, 8, 14, 13, 12, 128, 128, 128, 128, 2, 1, 0, 6, 5, 4, 10, 9, 8, 14, 13, 12, 128, 128, 128, 128}
+	VecDecodeBits       = [32]byte{0x01, 0x02, 0x04, 0x08, 0x10, 0x20, 0x40, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x02, 0x04, 0x08, 0x10, 0x20, 0x40, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00}
+	VecDecodeTableStd   = [128]byte{0x00, 0x00, 0x13, 0x04, 0xbf, 0xbf, 0xb9, 0xb9, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x13, 0x04, 0xbf, 0xbf, 0xb9, 0xb9, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xa8, 0xf8, 0xf8, 0xf8, 0xf8, 0xf8, 0xf8, 0xf8, 0xf8, 0xf8, 0xf0, 0x54, 0x50, 0x50, 0x50, 0x54, 0xa8, 0xf8, 0xf8, 0xf8, 0xf8, 0xf8, 0xf8, 0xf8, 0xf8, 0xf8, 0xf0, 0x54, 0x50, 0x50, 0x50, 0x54, 0x2f, 0x2f, 0x2f, 0x2f, 0x2f, 0x2f, 0x2f, 0x2f, 0x2f, 0x2f, 0x2f, 0x2f, 0x2f, 0x2f, 0x2f, 0x2f, 0x2f, 0x2f, 0x2f, 0x2f, 0x2f, 0x2f, 0x2f, 0x2f, 0x2f, 0x2f, 0x2f, 0x2f, 0x2f, 0x2f, 0x2f, 0x2f, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10}
+	VecDecodeTableURL   = [128]byte{0x00, 0x00, 0x11, 0x04, 0xbf, 0xbf, 0xb9, 0xb9, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x11, 0x04, 0xbf, 0xbf, 0xb9, 0xb9, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xa8, 0xf8, 0xf8, 0xf8, 0xf8, 0xf8, 0xf8, 0xf8, 0xf8, 0xf8, 0xf0, 0x50, 0x50, 0x54, 0x50, 0x70, 0xa8, 0xf8, 0xf8, 0xf8, 0xf8, 0xf8, 0xf8, 0xf8, 0xf8, 0xf8, 0xf0, 0x50, 0x50, 0x54, 0x50, 0x70, 0x5f, 0x5f, 0x5f, 0x5f, 0x5f, 0x5f, 0x5f, 0x5f, 0x5f, 0x5f, 0x5f, 0x5f, 0x5f, 0x5f, 0x5f, 0x5f, 0x5f, 0x5f, 0x5f, 0x5f, 0x5f, 0x5f, 0x5f, 0x5f, 0x5f, 0x5f, 0x5f, 0x5f, 0x5f, 0x5f, 0x5f, 0x5f, 0xe0, 0xe0, 0xe0, 0xe0, 0xe0, 0xe0, 0xe0, 0xe0, 0xe0, 0xe0, 0xe0, 0xe0, 0xe0, 0xe0, 0xe0, 0xe0, 0xe0, 0xe0, 0xe0, 0xe0, 0xe0, 0xe0, 0xe0, 0xe0, 0xe0, 0xe0, 0xe0, 0xe0, 0xe0, 0xe0, 0xe0, 0xe0}
+	TabDecodeCharsetStd = [256]byte{0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 62, 0xff, 0xff, 0xff, 63, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff}
+	TabDecodeCharsetURL = [256]byte{0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 62, 0xff, 0xff, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 0xff, 0xff, 0xff, 0xff, 63, 0xff, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff}
+)
+
+func b64_encode_tail(src []byte, src_len uint64, i uint64, dst []byte, o uint64, url_mode int, pad_mode int) uint64 {
+	var v9 uint32
+	var v11 []byte
+	var v91 uint64
+	if uint64(len(src)) < src_len {
+		src_len = uint64(len(src))
+	}
+	v11 = TabEncodeCharsetStd[:]
+	if url_mode != 0 {
+		v11 = TabEncodeCharsetURL[:]
+	}
+	for (src_len - i) >= 3 {
+		v9 = uint32((uint64(uint64(uint32(src[i]))<<16) | uint64(uint64(uint32(src[i+1]))<<8)) | uint64(uint32(src[i+2])))
+		v34 := uint64(uint64(v9)>>18) & 63
+		dst[o] = v11[v34]
+		v39 := uint64(uint64(v9)>>12) & 63
+		dst[o+1] = v11[v39]
+		v46 := uint64(uint64(v9)>>6) & 63
+		dst[o+2] = v11[v46]
+		v51 := v9 & 63
+		dst[o+3] = v11[v51]
+		i += 3
+		o += 4
+	}
+	v59 := src_len - i
+	if v59 == 0 {
+		return o
+	}
+	v9 = uint32(uint64(uint64(uint32(src[i])) << 16))
+	if v59 == 2 {
+		v9 = uint32(uint64(v9) | uint64(uint64(uint32(src[i+1]))<<8))
+	}
+	v76 := uint64(uint64(v9)>>18) & 63
+	dst[o] = v11[v76]
+	v81 := uint64(uint64(v9)>>12) & 63
+	dst[o+1] = v11[v81]
+	o += 2
+	if v59 == 2 {
+		v91 = uint64(uint64(v9)>>6) & 63
+		dst[o] = v11[v91]
+		o++
+		if pad_mode != 0 {
+			dst[o] = '='
+			o++
+		}
+	} else {
+		if pad_mode != 0 {
+			dst[o] = '='
+			dst[o+1] = '='
+			o += 2
+		}
+	}
+	return o
+}
+
+func B64_encode_scalar(src []byte, src_len uint64, dst []byte, dst_cap uint64, url_mode int, pad_mode int) uint64 {
+	if uint64(len(src)) < src_len {
+		src_len = uint64(len(src))
+	}
+	return uint64(b64_encode_tail(src[:], src_len, uint64(0), dst[:], uint64(0), url_mode, pad_mode))
+}
+
+func B64_encode(src []byte, src_len uint64, dst []byte, dst_cap uint64, url_mode int, pad_mode int) uint64 {
+	if uint64(len(src)) < src_len {
+		src_len = uint64(len(src))
+	}
+	return uint64(B64_encode_scalar(src[:], src_len, dst[:], dst_cap, url_mode, pad_mode))
+}
+
+func b64_decode_tail(src []byte, src_len uint64, si uint64, dst []byte, o uint64, url_mode int, pad_mode int, pos_err *uint64) uint64 {
+	var v10 []byte
+	var v11 uint32
+	var v14 int
+	var v15 int
+	var v16 uint64
+	var v24 int
+	if uint64(len(src)) < src_len {
+		src_len = uint64(len(src))
+	}
+	v10 = TabDecodeCharsetStd[:]
+	if url_mode != 0 {
+		v10 = TabDecodeCharsetURL[:]
+	}
+	for si < src_len {
+		v11 = 0
+		v15 = 4
+		v16 = 0xffffffffffffffff
+		v14 = 0
+		for v14 < 4 {
+			if si == src_len {
+				if v14 == 0 {
+					*pos_err = 0xffffffffffffffff
+					return o
+				}
+				if (v14 == 1) || (pad_mode != 0) {
+					*pos_err = si - uint64(v14)
+					return o
+				}
+				v15 = v14
+				break
+			}
+			v34 := src[si]
+			si++
+			v37 := v10[v34]
+			if v37 != 0xff {
+				v11 = uint32(uint64(v11) | uint64(uint64(uint32(v37))<<byte(18-(6*v14))))
+				v24 = v14 + 1
+				v14 = v24
+				continue
+			}
+			if (v34 == 10) || (v34 == 13) {
+				v14--
+				v24 = v14 + 1
+				v14 = v24
+				continue
+			}
+			if (pad_mode == 0) || (v34 != '=') {
+				*pos_err = si - 1
+				return o
+			}
+			if v14 < 2 {
+				*pos_err = si - 1
+				return o
+			}
+			if v14 == 2 {
+				for si < src_len {
+					var v71 bool
+					if src[si] == 10 {
+						v71 = true
+					} else {
+						v71 = src[si] == 13
+					}
+					if !v71 {
+						break
+					}
+					si++
+				}
+				if si == src_len {
+					*pos_err = src_len
+					return o
+				}
+				if src[si] != '=' {
+					*pos_err = si - 1
+					return o
+				}
+				si++
+			}
+			for si < src_len {
+				var v88 bool
+				if src[si] == 10 {
+					v88 = true
+				} else {
+					v88 = src[si] == 13
+				}
+				if !v88 {
+					break
+				}
+				si++
+			}
+			if si < src_len {
+				v16 = si
+			}
+			v15 = v14
+			break
+			v24 = v14 + 1
+			v14 = v24
+		}
+		dst[o] = byte(uint64(uint64(v11) >> 16))
+		if v15 >= 3 {
+			dst[o+1] = byte(uint64(uint64(v11) >> 8))
+		}
+		if v15 == 4 {
+			dst[o+2] = byte(v11)
+		}
+		o += uint64(v15 - 1)
+		if v16 != 0xffffffffffffffff {
+			*pos_err = v16
+			return o
+		}
+	}
+	*pos_err = 0xffffffffffffffff
+	return o
+}
+
+func B64_decode_scalar(src []byte, src_len uint64, dst []byte, dst_cap uint64, url_mode int, pad_mode int, pos_err *uint64) uint64 {
+	if uint64(len(src)) < src_len {
+		src_len = uint64(len(src))
+	}
+	return uint64(b64_decode_tail(src[:], src_len, uint64(0), dst[:], uint64(0), url_mode, pad_mode, pos_err))
+}
+
+func B64_decode(src []byte, src_len uint64, dst []byte, dst_cap uint64, url_mode int, pad_mode int, pos_err *uint64) uint64 {
+	if uint64(len(src)) < src_len {
+		src_len = uint64(len(src))
+	}
+	return uint64(B64_decode_scalar(src[:], src_len, dst[:], dst_cap, url_mode, pad_mode, pos_err))
+}
+
+// IRCosts counted from IR opcodes via rodata.IROpCost; not a kernel-name table.
+const (
+	B64_encode_tail_IRCost   = uint64(122)
+	B64_encode_scalar_IRCost = uint64(4)
+	B64_encode_IRCost        = uint64(2)
+	B64_decode_tail_IRCost   = uint64(144)
+	B64_decode_scalar_IRCost = uint64(4)
+	B64_decode_IRCost        = uint64(2)
+)