diff --git a/Cargo.lock b/Cargo.lock
index 6785db8d..829472f0 100644
--- a/Cargo.lock
+++ b/Cargo.lock
@@ -207,6 +207,7 @@ dependencies = [
  "itoa",
  "itoap",
  "once_cell",
+ "page_size",
  "pyo3-build-config",
  "pyo3-ffi",
  "ryu",
@@ -217,6 +218,16 @@ dependencies = [
  "version_check",
 ]
 
+[[package]]
+name = "page_size"
+version = "0.6.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "30d5b2194ed13191c1999ae0704b7839fb18384fa22e49b57eeaa97d79ce40da"
+dependencies = [
+ "libc",
+ "winapi",
+]
+
 [[package]]
 name = "proc-macro2"
 version = "1.0.78"
@@ -363,6 +374,28 @@ version = "0.11.0+wasi-snapshot-preview1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "9c8d87e72b64a3b4db28d11ce29237c246188f4f51057d65a7eab63b7987e423"
 
+[[package]]
+name = "winapi"
+version = "0.3.9"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "5c839a674fcd7a98952e593242ea400abe93992746761e38641405d28b00f419"
+dependencies = [
+ "winapi-i686-pc-windows-gnu",
+ "winapi-x86_64-pc-windows-gnu",
+]
+
+[[package]]
+name = "winapi-i686-pc-windows-gnu"
+version = "0.4.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "ac3b87c63620426dd9b991e5ce0329eff545bccbbb34f3be09ff6fb6ab51b7b6"
+
+[[package]]
+name = "winapi-x86_64-pc-windows-gnu"
+version = "0.4.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "712e227841d057c1ee1cd2fb22fa7e5a5461ae8e48fa2ca79ec42cfc1931183f"
+
 [[package]]
 name = "zerocopy"
 version = "0.7.32"
diff --git a/Cargo.toml b/Cargo.toml
index 526ec5fe..68fae48c 100644
--- a/Cargo.toml
+++ b/Cargo.toml
@@ -56,6 +56,7 @@ encoding_rs = { version = "0.8", default_features = false }
 itoa = { version = "1", default_features = false }
 itoap = { version = "1", features = ["std", "simd"] }
 once_cell = { version = "1", default_features = false, features = ["race"] }
+page_size = { version = "0.6" }
 pyo3-ffi = { version = "^0.20.2", default_features = false, features = ["extension-module"]}
 ryu = { version = "1", default_features = false }
 serde = { version = "1", default_features = false }
diff --git a/src/serialize/writer/simd.rs b/src/serialize/writer/simd.rs
index fe4a0dcf..94a42ac8 100644
--- a/src/serialize/writer/simd.rs
+++ b/src/serialize/writer/simd.rs
@@ -2,6 +2,7 @@
 // Copyright 2023-2024 liuq19, ijl
 // adapted from sonic-rs' src/util/string.rs
 
+use crate::typeref::PAGE_SIZE;
 use core::simd::cmp::{SimdPartialEq, SimdPartialOrd};
 
 macro_rules! impl_escape_unchecked {
@@ -9,60 +10,29 @@ macro_rules! impl_escape_unchecked {
         $nb -= $cn;
         $dst = $dst.add($cn);
         $src = $src.add($cn);
-        let mut mask = $omask << $cn;
+        $omask >>= $cn;
         loop {
             $nb -= 1;
-            mask = mask << 1;
-            let replacement = if *($src) == b'"' {
-                (*b"\\\"\0\0\0\0\0\0", 2)
+            $omask = $omask >> 1;
+
+            if *($src) == b'"' {
+                core::ptr::copy_nonoverlapping(b"\\\"".as_ptr(), $dst, 2);
+                $dst = $dst.add(2);
             } else if *($src) == b'\\' {
-                (*b"\\\\\0\0\0\0\0\0", 2)
+                core::ptr::copy_nonoverlapping(b"\\\\".as_ptr(), $dst, 2);
+                $dst = $dst.add(2);
             } else {
-                match *($src) {
-                    0 => (*b"\\u0000\0\0", 6),
-                    1 => (*b"\\u0001\0\0", 6),
-                    2 => (*b"\\u0002\0\0", 6),
-                    3 => (*b"\\u0003\0\0", 6),
-                    4 => (*b"\\u0004\0\0", 6),
-                    5 => (*b"\\u0005\0\0", 6),
-                    6 => (*b"\\u0006\0\0", 6),
-                    7 => (*b"\\u0007\0\0", 6),
-                    8 => (*b"\\b\0\0\0\0\0\0", 2),
-                    9 => (*b"\\t\0\0\0\0\0\0", 2),
-                    10 => (*b"\\n\0\0\0\0\0\0", 2),
-                    11 => (*b"\\u000b\0\0", 6),
-                    12 => (*b"\\f\0\0\0\0\0\0", 2),
-                    13 => (*b"\\r\0\0\0\0\0\0", 2),
-                    14 => (*b"\\u000e\0\0", 6),
-                    15 => (*b"\\u000f\0\0", 6),
-                    16 => (*b"\\u0010\0\0", 6),
-                    17 => (*b"\\u0011\0\0", 6),
-                    18 => (*b"\\u0012\0\0", 6),
-                    19 => (*b"\\u0013\0\0", 6),
-                    20 => (*b"\\u0014\0\0", 6),
-                    21 => (*b"\\u0015\0\0", 6),
-                    22 => (*b"\\u0016\0\0", 6),
-                    23 => (*b"\\u0017\0\0", 6),
-                    24 => (*b"\\u0018\0\0", 6),
-                    25 => (*b"\\u0019\0\0", 6),
-                    26 => (*b"\\u001a\0\0", 6),
-                    27 => (*b"\\u001b\0\0", 6),
-                    28 => (*b"\\u001c\0\0", 6),
-                    29 => (*b"\\u001d\0\0", 6),
-                    30 => (*b"\\u001e\0\0", 6),
-                    31 => (*b"\\u001f\0\0", 6),
-                    _ => unreachable!(),
-                }
+                $dst = write_unusual_escape($src, $dst);
             };
-            core::ptr::copy_nonoverlapping(replacement.0.as_ptr(), $dst, 8);
-            $dst = $dst.add(replacement.1 as usize);
+
             $src = $src.add(1);
-            if likely!(mask & (1 << (STRIDE - 1)) != 1) {
+            if likely!($omask & 1 != 1) {
                 break;
             }
         }
     };
 }
+
 macro_rules! impl_format_simd {
     ($odptr:expr, $value_ptr:expr, $value_len:expr) => {
         let mut dptr = $odptr;
@@ -81,7 +51,7 @@ macro_rules! impl_format_simd {
             while nb >= STRIDE {
                 let v = StrVector::from_slice(core::slice::from_raw_parts(sptr, STRIDE));
                 v.copy_to_slice(core::slice::from_raw_parts_mut(dptr, STRIDE));
-                let mask =
+                let mut mask =
                     (v.simd_eq(blash) | v.simd_eq(quote) | v.simd_lt(x20)).to_bitmask() as u32;
 
                 if likely!(mask == 0) {
@@ -95,12 +65,18 @@ macro_rules! impl_format_simd {
             }
 
             while nb > 0 {
-                let mut v = StrVector::default();
-                v.as_mut_array()[..nb].copy_from_slice(core::slice::from_raw_parts(sptr, nb));
+                let v = if unlikely!(is_cross_page!(sptr)) {
+                    let mut v = StrVector::default();
+                    v.as_mut_array()[..nb].copy_from_slice(core::slice::from_raw_parts(sptr, nb));
+                    v
+                } else {
+                    StrVector::from_slice(core::slice::from_raw_parts(sptr, STRIDE))
+                };
                 v.copy_to_slice(core::slice::from_raw_parts_mut(dptr, STRIDE));
-                let mask = (v.simd_eq(blash) | v.simd_eq(quote) | v.simd_lt(x20)).to_bitmask()
+                let mut mask = (v.simd_eq(blash) | v.simd_eq(quote) | v.simd_lt(x20)).to_bitmask()
                     as u32
-                    & (STRIDE_SATURATION >> (STRIDE - nb));
+                    & (STRIDE_SATURATION >> (32 - STRIDE - nb));
+
                 if likely!(mask == 0) {
                     dptr = dptr.add(nb);
                     break;
@@ -118,6 +94,57 @@ macro_rules! impl_format_simd {
     };
 }
 
+macro_rules! is_cross_page {
+    ($src:expr) => {
+        unsafe { (($src as usize & (PAGE_SIZE - 1)) + STRIDE) > PAGE_SIZE }
+    };
+}
+
+#[cold]
+#[inline(never)]
+fn write_unusual_escape(sptr: *const u8, dptr: *mut u8) -> *mut u8 {
+    unsafe {
+        debug_assert!(*sptr < 32);
+        let replacement = match *(sptr) {
+            0 => (*b"\\u0000\0\0", 6),
+            1 => (*b"\\u0001\0\0", 6),
+            2 => (*b"\\u0002\0\0", 6),
+            3 => (*b"\\u0003\0\0", 6),
+            4 => (*b"\\u0004\0\0", 6),
+            5 => (*b"\\u0005\0\0", 6),
+            6 => (*b"\\u0006\0\0", 6),
+            7 => (*b"\\u0007\0\0", 6),
+            8 => (*b"\\b\0\0\0\0\0\0", 2),
+            9 => (*b"\\t\0\0\0\0\0\0", 2),
+            10 => (*b"\\n\0\0\0\0\0\0", 2),
+            11 => (*b"\\u000b\0\0", 6),
+            12 => (*b"\\f\0\0\0\0\0\0", 2),
+            13 => (*b"\\r\0\0\0\0\0\0", 2),
+            14 => (*b"\\u000e\0\0", 6),
+            15 => (*b"\\u000f\0\0", 6),
+            16 => (*b"\\u0010\0\0", 6),
+            17 => (*b"\\u0011\0\0", 6),
+            18 => (*b"\\u0012\0\0", 6),
+            19 => (*b"\\u0013\0\0", 6),
+            20 => (*b"\\u0014\0\0", 6),
+            21 => (*b"\\u0015\0\0", 6),
+            22 => (*b"\\u0016\0\0", 6),
+            23 => (*b"\\u0017\0\0", 6),
+            24 => (*b"\\u0018\0\0", 6),
+            25 => (*b"\\u0019\0\0", 6),
+            26 => (*b"\\u001a\0\0", 6),
+            27 => (*b"\\u001b\0\0", 6),
+            28 => (*b"\\u001c\0\0", 6),
+            29 => (*b"\\u001d\0\0", 6),
+            30 => (*b"\\u001e\0\0", 6),
+            31 => (*b"\\u001f\0\0", 6),
+            _ => unreachable!(),
+        };
+        core::ptr::copy_nonoverlapping(replacement.0.as_ptr(), dptr, 8);
+        dptr.add(replacement.1 as usize)
+    }
+}
+
 #[inline(never)]
 pub unsafe fn format_escaped_str_impl_128(
     odptr: *mut u8,
diff --git a/src/typeref.rs b/src/typeref.rs
index 486122bb..3689bbb7 100644
--- a/src/typeref.rs
+++ b/src/typeref.rs
@@ -75,6 +75,9 @@ pub static mut DESCR_STR: *mut PyObject = null_mut();
 pub static mut VALUE_STR: *mut PyObject = null_mut();
 pub static mut INT_ATTR_STR: *mut PyObject = null_mut();
 
+#[cfg(feature = "unstable-simd")]
+pub static mut PAGE_SIZE: usize = 0;
+
 #[cfg(feature = "yyjson")]
 pub const YYJSON_BUFFER_SIZE: usize = 1024 * 1024 * 8;
 
@@ -135,6 +138,11 @@ pub fn init_typerefs() {
 fn _init_typerefs_impl() -> bool {
     unsafe {
         debug_assert!(crate::opt::MAX_OPT < u16::MAX as i32);
+
+        #[cfg(feature = "unstable-simd")]
+        {
+            PAGE_SIZE = page_size::get();
+        }
         assert!(crate::deserialize::KEY_MAP
             .set(crate::deserialize::KeyMap::default())
             .is_ok());