ydblib: add simdjson

commit_hash:a86d76c6047d0b7407a8f1e013ccecf0e3c886eb

96 files changed, 8538 insertions, 0 deletions

diff --git a/contrib/libs/simdjson/include/simdjson/arm64.h b/contrib/libs/simdjson/include/simdjson/arm64.h
new file mode 100644
index 00000000000..1493c3562a5
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/arm64.h
@@ -0,0 +1,8 @@
+#ifndef SIMDJSON_ARM64_H
+#define SIMDJSON_ARM64_H
+
+#include "simdjson/arm64/begin.h"
+#include "simdjson/generic/amalgamated.h"
+#include "simdjson/arm64/end.h"
+
+#endif // SIMDJSON_ARM64_H
\ No newline at end of file
diff --git a/contrib/libs/simdjson/include/simdjson/arm64/base.h b/contrib/libs/simdjson/include/simdjson/arm64/base.h
new file mode 100644
index 00000000000..8f23d18d80e
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/arm64/base.h
@@ -0,0 +1,26 @@
+#ifndef SIMDJSON_ARM64_BASE_H
+#define SIMDJSON_ARM64_BASE_H
+
+#ifndef SIMDJSON_CONDITIONAL_INCLUDE
+#include "simdjson/base.h"
+#endif // SIMDJSON_CONDITIONAL_INCLUDE
+
+namespace simdjson {
+/**
+ * Implementation for NEON (ARMv8).
+ */
+namespace arm64 {
+
+class implementation;
+
+namespace {
+namespace simd {
+template <typename T> struct simd8;
+template <typename T> struct simd8x64;
+} // namespace simd
+} // unnamed namespace
+
+} // namespace arm64
+} // namespace simdjson
+
+#endif // SIMDJSON_ARM64_BASE_H
diff --git a/contrib/libs/simdjson/include/simdjson/arm64/begin.h b/contrib/libs/simdjson/include/simdjson/arm64/begin.h
new file mode 100644
index 00000000000..ee48cec0515
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/arm64/begin.h
@@ -0,0 +1,10 @@
+#define SIMDJSON_IMPLEMENTATION arm64
+#include "simdjson/arm64/base.h"
+#include "simdjson/arm64/intrinsics.h"
+#include "simdjson/arm64/bitmanipulation.h"
+#include "simdjson/arm64/bitmask.h"
+#include "simdjson/arm64/numberparsing_defs.h"
+#include "simdjson/arm64/simd.h"
+#include "simdjson/arm64/stringparsing_defs.h"
+
+#define SIMDJSON_SKIP_BACKSLASH_SHORT_CIRCUIT 1
\ No newline at end of file
diff --git a/contrib/libs/simdjson/include/simdjson/arm64/bitmanipulation.h b/contrib/libs/simdjson/include/simdjson/arm64/bitmanipulation.h
new file mode 100644
index 00000000000..019869b1b0b
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/arm64/bitmanipulation.h
@@ -0,0 +1,106 @@
+#ifndef SIMDJSON_ARM64_BITMANIPULATION_H
+#define SIMDJSON_ARM64_BITMANIPULATION_H
+
+#ifndef SIMDJSON_CONDITIONAL_INCLUDE
+#include "simdjson/arm64/base.h"
+#include "simdjson/arm64/intrinsics.h"
+#endif // SIMDJSON_CONDITIONAL_INCLUDE
+
+namespace simdjson {
+namespace arm64 {
+namespace {
+
+// We sometimes call trailing_zero on inputs that are zero,
+// but the algorithms do not end up using the returned value.
+// Sadly, sanitizers are not smart enough to figure it out.
+SIMDJSON_NO_SANITIZE_UNDEFINED
+// This function can be used safely even if not all bytes have been
+// initialized.
+// See issue https://github.com/simdjson/simdjson/issues/1965
+SIMDJSON_NO_SANITIZE_MEMORY
+simdjson_inline int trailing_zeroes(uint64_t input_num) {
+#ifdef SIMDJSON_REGULAR_VISUAL_STUDIO
+  unsigned long ret;
+  // Search the mask data from least significant bit (LSB)
+  // to the most significant bit (MSB) for a set bit (1).
+  _BitScanForward64(&ret, input_num);
+  return (int)ret;
+#else // SIMDJSON_REGULAR_VISUAL_STUDIO
+  return __builtin_ctzll(input_num);
+#endif // SIMDJSON_REGULAR_VISUAL_STUDIO
+}
+
+/* result might be undefined when input_num is zero */
+simdjson_inline uint64_t clear_lowest_bit(uint64_t input_num) {
+  return input_num & (input_num-1);
+}
+
+/* result might be undefined when input_num is zero */
+simdjson_inline int leading_zeroes(uint64_t input_num) {
+#ifdef SIMDJSON_REGULAR_VISUAL_STUDIO
+  unsigned long leading_zero = 0;
+  // Search the mask data from most significant bit (MSB)
+  // to least significant bit (LSB) for a set bit (1).
+  if (_BitScanReverse64(&leading_zero, input_num))
+    return (int)(63 - leading_zero);
+  else
+    return 64;
+#else
+  return __builtin_clzll(input_num);
+#endif// SIMDJSON_REGULAR_VISUAL_STUDIO
+}
+
+/* result might be undefined when input_num is zero */
+simdjson_inline int count_ones(uint64_t input_num) {
+   return vaddv_u8(vcnt_u8(vcreate_u8(input_num)));
+}
+
+
+#if defined(__GNUC__) // catches clang and gcc
+/**
+ * ARM has a fast 64-bit "bit reversal function" that is handy. However,
+ * it is not generally available as an intrinsic function under Visual
+ * Studio (though this might be changing). Even under clang/gcc, we
+ * apparently need to invoke inline assembly.
+ */
+/*
+ * We use SIMDJSON_PREFER_REVERSE_BITS as a hint that algorithms that
+ * work well with bit reversal may use it.
+ */
+#define SIMDJSON_PREFER_REVERSE_BITS 1
+
+/* reverse the bits */
+simdjson_inline uint64_t reverse_bits(uint64_t input_num) {
+  uint64_t rev_bits;
+  __asm("rbit %0, %1" : "=r"(rev_bits) : "r"(input_num));
+  return rev_bits;
+}
+
+/**
+ * Flips bit at index 63 - lz. Thus if you have 'leading_zeroes' leading zeroes,
+ * then this will set to zero the leading bit. It is possible for leading_zeroes to be
+ * greating or equal to 63 in which case we trigger undefined behavior, but the output
+ * of such undefined behavior is never used.
+ **/
+SIMDJSON_NO_SANITIZE_UNDEFINED
+simdjson_inline uint64_t zero_leading_bit(uint64_t rev_bits, int leading_zeroes) {
+  return rev_bits ^ (uint64_t(0x8000000000000000) >> leading_zeroes);
+}
+
+#endif
+
+simdjson_inline bool add_overflow(uint64_t value1, uint64_t value2, uint64_t *result) {
+#ifdef SIMDJSON_REGULAR_VISUAL_STUDIO
+  *result = value1 + value2;
+  return *result < value1;
+#else
+  return __builtin_uaddll_overflow(value1, value2,
+                                   reinterpret_cast<unsigned long long *>(result));
+#endif
+}
+
+} // unnamed namespace
+} // namespace arm64
+} // namespace simdjson
+
+#endif // SIMDJSON_ARM64_BITMANIPULATION_H
diff --git a/contrib/libs/simdjson/include/simdjson/arm64/bitmask.h b/contrib/libs/simdjson/include/simdjson/arm64/bitmask.h
new file mode 100644
index 00000000000..5d6121bcc7b
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/arm64/bitmask.h
@@ -0,0 +1,44 @@
+#ifndef SIMDJSON_ARM64_BITMASK_H
+#define SIMDJSON_ARM64_BITMASK_H
+
+#ifndef SIMDJSON_CONDITIONAL_INCLUDE
+#include "simdjson/arm64/base.h"
+#endif // SIMDJSON_CONDITIONAL_INCLUDE
+
+namespace simdjson {
+namespace arm64 {
+namespace {
+
+//
+// Perform a "cumulative bitwise xor," flipping bits each time a 1 is encountered.
+//
+// For example, prefix_xor(00100100) == 00011100
+//
+simdjson_inline uint64_t prefix_xor(uint64_t bitmask) {
+  /////////////
+  // We could do this with PMULL, but it is apparently slow.
+  //
+  //#ifdef __ARM_FEATURE_CRYPTO // some ARM processors lack this extension
+  //return vmull_p64(-1ULL, bitmask);
+  //#else
+  // Analysis by @sebpop:
+  // When diffing the assembly for src/stage1_find_marks.cpp I see that the eors are all spread out
+  // in between other vector code, so effectively the extra cycles of the sequence do not matter
+  // because the GPR units are idle otherwise and the critical path is on the FP side.
+  // Also the PMULL requires two extra fmovs: GPR->FP (3 cycles in N1, 5 cycles in A72 )
+  // and FP->GPR (2 cycles on N1 and 5 cycles on A72.)
+  ///////////
+  bitmask ^= bitmask << 1;
+  bitmask ^= bitmask << 2;
+  bitmask ^= bitmask << 4;
+  bitmask ^= bitmask << 8;
+  bitmask ^= bitmask << 16;
+  bitmask ^= bitmask << 32;
+  return bitmask;
+}
+
+} // unnamed namespace
+} // namespace arm64
+} // namespace simdjson
+
+#endif
diff --git a/contrib/libs/simdjson/include/simdjson/arm64/end.h b/contrib/libs/simdjson/include/simdjson/arm64/end.h
new file mode 100644
index 00000000000..b92378df97f
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/arm64/end.h
@@ -0,0 +1,6 @@
+#ifndef SIMDJSON_CONDITIONAL_INCLUDE
+#include "simdjson/arm64/base.h"
+#endif // SIMDJSON_CONDITIONAL_INCLUDE
+
+#undef SIMDJSON_SKIP_BACKSLASH_SHORT_CIRCUIT
+#undef SIMDJSON_IMPLEMENTATION
diff --git a/contrib/libs/simdjson/include/simdjson/arm64/implementation.h b/contrib/libs/simdjson/include/simdjson/arm64/implementation.h
new file mode 100644
index 00000000000..c9b7dd75350
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/arm64/implementation.h
@@ -0,0 +1,31 @@
+#ifndef SIMDJSON_ARM64_IMPLEMENTATION_H
+#define SIMDJSON_ARM64_IMPLEMENTATION_H
+
+#ifndef SIMDJSON_CONDITIONAL_INCLUDE
+#include "simdjson/base.h"
+#include "simdjson/implementation.h"
+#include "simdjson/internal/instruction_set.h"
+#endif // SIMDJSON_CONDITIONAL_INCLUDE
+
+namespace simdjson {
+namespace arm64 {
+
+/**
+ * @private
+ */
+class implementation final : public simdjson::implementation {
+public:
+  simdjson_inline implementation() : simdjson::implementation("arm64", "ARM NEON", internal::instruction_set::NEON) {}
+  simdjson_warn_unused error_code create_dom_parser_implementation(
+    size_t capacity,
+    size_t max_length,
+    std::unique_ptr<internal::dom_parser_implementation>& dst
+  ) const noexcept final;
+  simdjson_warn_unused error_code minify(const uint8_t *buf, size_t len, uint8_t *dst, size_t &dst_len) const noexcept final;
+  simdjson_warn_unused bool validate_utf8(const char *buf, size_t len) const noexcept final;
+};
+
+} // namespace arm64
+} // namespace simdjson
+
+#endif // SIMDJSON_ARM64_IMPLEMENTATION_H
diff --git a/contrib/libs/simdjson/include/simdjson/arm64/intrinsics.h b/contrib/libs/simdjson/include/simdjson/arm64/intrinsics.h
new file mode 100644
index 00000000000..049d9986102
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/arm64/intrinsics.h
@@ -0,0 +1,14 @@
+#ifndef SIMDJSON_ARM64_INTRINSICS_H
+#define SIMDJSON_ARM64_INTRINSICS_H
+
+#ifndef SIMDJSON_CONDITIONAL_INCLUDE
+#include "simdjson/arm64/base.h"
+#endif // SIMDJSON_CONDITIONAL_INCLUDE
+
+// This should be the correct header whether
+// you use visual studio or other compilers.
+#include <arm_neon.h>
+
+static_assert(sizeof(uint8x16_t) <= simdjson::SIMDJSON_PADDING, "insufficient padding for arm64");
+
+#endif //  SIMDJSON_ARM64_INTRINSICS_H
diff --git a/contrib/libs/simdjson/include/simdjson/arm64/numberparsing_defs.h b/contrib/libs/simdjson/include/simdjson/arm64/numberparsing_defs.h
new file mode 100644
index 00000000000..f5e85b74779
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/arm64/numberparsing_defs.h
@@ -0,0 +1,56 @@
+#ifndef SIMDJSON_ARM64_NUMBERPARSING_DEFS_H
+#define SIMDJSON_ARM64_NUMBERPARSING_DEFS_H
+
+#ifndef SIMDJSON_CONDITIONAL_INCLUDE
+#include "simdjson/arm64/base.h"
+#include "simdjson/arm64/intrinsics.h"
+#include "simdjson/internal/numberparsing_tables.h"
+#endif // SIMDJSON_CONDITIONAL_INCLUDE
+
+#include <cstring>
+
+#if SIMDJSON_REGULAR_VISUAL_STUDIO && SIMDJSON_IS_ARM64
+// __umulh requires intrin.h
+#include <intrin.h>
+#endif // SIMDJSON_REGULAR_VISUAL_STUDIO && SIMDJSON_IS_ARM64
+
+namespace simdjson {
+namespace arm64 {
+namespace numberparsing {
+
+// we don't have SSE, so let us use a scalar function
+// credit: https://johnnylee-sde.github.io/Fast-numeric-string-to-int/
+/** @private */
+static simdjson_inline uint32_t parse_eight_digits_unrolled(const uint8_t *chars) {
+  uint64_t val;
+  std::memcpy(&val, chars, sizeof(uint64_t));
+  val = (val & 0x0F0F0F0F0F0F0F0F) * 2561 >> 8;
+  val = (val & 0x00FF00FF00FF00FF) * 6553601 >> 16;
+  return uint32_t((val & 0x0000FFFF0000FFFF) * 42949672960001 >> 32);
+}
+
+simdjson_inline internal::value128 full_multiplication(uint64_t value1, uint64_t value2) {
+  internal::value128 answer;
+#if SIMDJSON_REGULAR_VISUAL_STUDIO || SIMDJSON_IS_32BITS
+#if SIMDJSON_IS_ARM64
+  // ARM64 has native support for 64-bit multiplications, no need to emultate
+  answer.high = __umulh(value1, value2);
+  answer.low = value1 * value2;
+#else
+  answer.low = _umul128(value1, value2, &answer.high); // _umul128 not available on ARM64
+#endif // SIMDJSON_IS_ARM64
+#else // SIMDJSON_REGULAR_VISUAL_STUDIO || SIMDJSON_IS_32BITS
+  __uint128_t r = (static_cast<__uint128_t>(value1)) * value2;
+  answer.low = uint64_t(r);
+  answer.high = uint64_t(r >> 64);
+#endif
+  return answer;
+}
+
+} // namespace numberparsing
+} // namespace arm64
+} // namespace simdjson
+
+#define SIMDJSON_SWAR_NUMBER_PARSING 1
+
+#endif // SIMDJSON_ARM64_NUMBERPARSING_DEFS_H
diff --git a/contrib/libs/simdjson/include/simdjson/arm64/simd.h b/contrib/libs/simdjson/include/simdjson/arm64/simd.h
new file mode 100644
index 00000000000..3b0fa844f49
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/arm64/simd.h
@@ -0,0 +1,497 @@
+#ifndef SIMDJSON_ARM64_SIMD_H
+#define SIMDJSON_ARM64_SIMD_H
+
+#ifndef SIMDJSON_CONDITIONAL_INCLUDE
+#include "simdjson/arm64/base.h"
+#include "simdjson/arm64/bitmanipulation.h"
+#include "simdjson/internal/simdprune_tables.h"
+#endif // SIMDJSON_CONDITIONAL_INCLUDE
+
+namespace simdjson {
+namespace arm64 {
+namespace {
+namespace simd {
+
+#ifdef SIMDJSON_REGULAR_VISUAL_STUDIO
+namespace {
+// Start of private section with Visual Studio workaround
+
+
+#ifndef simdjson_make_uint8x16_t
+#define simdjson_make_uint8x16_t(x1, x2, x3, x4, x5, x6, x7, x8, x9, x10, x11, x12, \
+                             x13, x14, x15, x16)                                   \
+   ([=]() {                                                                        \
+     uint8_t array[16] = {x1, x2,  x3,  x4,  x5,  x6,  x7,  x8,                    \
+                                 x9, x10, x11, x12, x13, x14, x15, x16};           \
+     return vld1q_u8(array);                                                       \
+   }())
+#endif
+#ifndef simdjson_make_int8x16_t
+#define simdjson_make_int8x16_t(x1, x2, x3, x4, x5, x6, x7, x8, x9, x10, x11, x12, \
+                             x13, x14, x15, x16)                                  \
+   ([=]() {                                                                       \
+     int8_t array[16] = {x1, x2,  x3,  x4,  x5,  x6,  x7,  x8,                    \
+                                 x9, x10, x11, x12, x13, x14, x15, x16};          \
+     return vld1q_s8(array);                                                      \
+   }())
+#endif
+
+#ifndef simdjson_make_uint8x8_t
+#define simdjson_make_uint8x8_t(x1, x2, x3, x4, x5, x6, x7, x8)                \
+   ([=]() {                                                                    \
+     uint8_t array[8] = {x1, x2,  x3,  x4,  x5,  x6,  x7,  x8};                \
+     return vld1_u8(array);                                                    \
+   }())
+#endif
+#ifndef simdjson_make_int8x8_t
+#define simdjson_make_int8x8_t(x1, x2, x3, x4, x5, x6, x7, x8)                 \
+   ([=]() {                                                                    \
+     int8_t array[8] = {x1, x2,  x3,  x4,  x5,  x6,  x7,  x8};                 \
+     return vld1_s8(array);                                                    \
+   }())
+#endif
+#ifndef simdjson_make_uint16x8_t
+#define simdjson_make_uint16x8_t(x1, x2, x3, x4, x5, x6, x7, x8)               \
+   ([=]() {                                                                    \
+     uint16_t array[8] = {x1, x2,  x3,  x4,  x5,  x6,  x7,  x8};               \
+     return vld1q_u16(array);                                                  \
+   }())
+#endif
+#ifndef simdjson_make_int16x8_t
+#define simdjson_make_int16x8_t(x1, x2, x3, x4, x5, x6, x7, x8)                \
+   ([=]() {                                                                    \
+     int16_t array[8] = {x1, x2,  x3,  x4,  x5,  x6,  x7,  x8};                \
+     return vld1q_s16(array);                                                  \
+   }())
+#endif
+
+// End of private section with Visual Studio workaround
+} // namespace
+#endif // SIMDJSON_REGULAR_VISUAL_STUDIO
+
+
+  template<typename T>
+  struct simd8;
+
+  //
+  // Base class of simd8<uint8_t> and simd8<bool>, both of which use uint8x16_t internally.
+  //
+  template<typename T, typename Mask=simd8<bool>>
+  struct base_u8 {
+    uint8x16_t value;
+    static const int SIZE = sizeof(value);
+
+    // Conversion from/to SIMD register
+    simdjson_inline base_u8(const uint8x16_t _value) : value(_value) {}
+    simdjson_inline operator const uint8x16_t&() const { return this->value; }
+    simdjson_inline operator uint8x16_t&() { return this->value; }
+
+    // Bit operations
+    simdjson_inline simd8<T> operator|(const simd8<T> other) const { return vorrq_u8(*this, other); }
+    simdjson_inline simd8<T> operator&(const simd8<T> other) const { return vandq_u8(*this, other); }
+    simdjson_inline simd8<T> operator^(const simd8<T> other) const { return veorq_u8(*this, other); }
+    simdjson_inline simd8<T> bit_andnot(const simd8<T> other) const { return vbicq_u8(*this, other); }
+    simdjson_inline simd8<T> operator~() const { return *this ^ 0xFFu; }
+    simdjson_inline simd8<T>& operator|=(const simd8<T> other) { auto this_cast = static_cast<simd8<T>*>(this); *this_cast = *this_cast | other; return *this_cast; }
+    simdjson_inline simd8<T>& operator&=(const simd8<T> other) { auto this_cast = static_cast<simd8<T>*>(this); *this_cast = *this_cast & other; return *this_cast; }
+    simdjson_inline simd8<T>& operator^=(const simd8<T> other) { auto this_cast = static_cast<simd8<T>*>(this); *this_cast = *this_cast ^ other; return *this_cast; }
+
+    friend simdjson_inline Mask operator==(const simd8<T> lhs, const simd8<T> rhs) { return vceqq_u8(lhs, rhs); }
+
+    template<int N=1>
+    simdjson_inline simd8<T> prev(const simd8<T> prev_chunk) const {
+      return vextq_u8(prev_chunk, *this, 16 - N);
+    }
+  };
+
+  // SIMD byte mask type (returned by things like eq and gt)
+  template<>
+  struct simd8<bool>: base_u8<bool> {
+    typedef uint16_t bitmask_t;
+    typedef uint32_t bitmask2_t;
+
+    static simdjson_inline simd8<bool> splat(bool _value) { return vmovq_n_u8(uint8_t(-(!!_value))); }
+
+    simdjson_inline simd8(const uint8x16_t _value) : base_u8<bool>(_value) {}
+    // False constructor
+    simdjson_inline simd8() : simd8(vdupq_n_u8(0)) {}
+    // Splat constructor
+    simdjson_inline simd8(bool _value) : simd8(splat(_value)) {}
+
+    // We return uint32_t instead of uint16_t because that seems to be more efficient for most
+    // purposes (cutting it down to uint16_t costs performance in some compilers).
+    simdjson_inline uint32_t to_bitmask() const {
+#ifdef SIMDJSON_REGULAR_VISUAL_STUDIO
+      const uint8x16_t bit_mask =  simdjson_make_uint8x16_t(0x01, 0x02, 0x4, 0x8, 0x10, 0x20, 0x40, 0x80,
+                                                   0x01, 0x02, 0x4, 0x8, 0x10, 0x20, 0x40, 0x80);
+#else
+      const uint8x16_t bit_mask =  {0x01, 0x02, 0x4, 0x8, 0x10, 0x20, 0x40, 0x80,
+                                    0x01, 0x02, 0x4, 0x8, 0x10, 0x20, 0x40, 0x80};
+#endif
+      auto minput = *this & bit_mask;
+      uint8x16_t tmp = vpaddq_u8(minput, minput);
+      tmp = vpaddq_u8(tmp, tmp);
+      tmp = vpaddq_u8(tmp, tmp);
+      return vgetq_lane_u16(vreinterpretq_u16_u8(tmp), 0);
+    }
+    simdjson_inline bool any() const { return vmaxvq_u32(vreinterpretq_u32_u8(*this)) != 0; }
+  };
+
+  // Unsigned bytes
+  template<>
+  struct simd8<uint8_t>: base_u8<uint8_t> {
+    static simdjson_inline uint8x16_t splat(uint8_t _value) { return vmovq_n_u8(_value); }
+    static simdjson_inline uint8x16_t zero() { return vdupq_n_u8(0); }
+    static simdjson_inline uint8x16_t load(const uint8_t* values) { return vld1q_u8(values); }
+
+    simdjson_inline simd8(const uint8x16_t _value) : base_u8<uint8_t>(_value) {}
+    // Zero constructor
+    simdjson_inline simd8() : simd8(zero()) {}
+    // Array constructor
+    simdjson_inline simd8(const uint8_t values[16]) : simd8(load(values)) {}
+    // Splat constructor
+    simdjson_inline simd8(uint8_t _value) : simd8(splat(_value)) {}
+    // Member-by-member initialization
+#ifdef SIMDJSON_REGULAR_VISUAL_STUDIO
+    simdjson_inline simd8(
+      uint8_t v0,  uint8_t v1,  uint8_t v2,  uint8_t v3,  uint8_t v4,  uint8_t v5,  uint8_t v6,  uint8_t v7,
+      uint8_t v8,  uint8_t v9,  uint8_t v10, uint8_t v11, uint8_t v12, uint8_t v13, uint8_t v14, uint8_t v15
+    ) : simd8(simdjson_make_uint8x16_t(
+      v0, v1, v2, v3, v4, v5, v6, v7,
+      v8, v9, v10,v11,v12,v13,v14,v15
+    )) {}
+#else
+    simdjson_inline simd8(
+      uint8_t v0,  uint8_t v1,  uint8_t v2,  uint8_t v3,  uint8_t v4,  uint8_t v5,  uint8_t v6,  uint8_t v7,
+      uint8_t v8,  uint8_t v9,  uint8_t v10, uint8_t v11, uint8_t v12, uint8_t v13, uint8_t v14, uint8_t v15
+    ) : simd8(uint8x16_t{
+      v0, v1, v2, v3, v4, v5, v6, v7,
+      v8, v9, v10,v11,v12,v13,v14,v15
+    }) {}
+#endif
+
+    // Repeat 16 values as many times as necessary (usually for lookup tables)
+    simdjson_inline static simd8<uint8_t> repeat_16(
+      uint8_t v0,  uint8_t v1,  uint8_t v2,  uint8_t v3,  uint8_t v4,  uint8_t v5,  uint8_t v6,  uint8_t v7,
+      uint8_t v8,  uint8_t v9,  uint8_t v10, uint8_t v11, uint8_t v12, uint8_t v13, uint8_t v14, uint8_t v15
+    ) {
+      return simd8<uint8_t>(
+        v0, v1, v2, v3, v4, v5, v6, v7,
+        v8, v9, v10,v11,v12,v13,v14,v15
+      );
+    }
+
+    // Store to array
+    simdjson_inline void store(uint8_t dst[16]) const { return vst1q_u8(dst, *this); }
+
+    // Saturated math
+    simdjson_inline simd8<uint8_t> saturating_add(const simd8<uint8_t> other) const { return vqaddq_u8(*this, other); }
+    simdjson_inline simd8<uint8_t> saturating_sub(const simd8<uint8_t> other) const { return vqsubq_u8(*this, other); }
+
+    // Addition/subtraction are the same for signed and unsigned
+    simdjson_inline simd8<uint8_t> operator+(const simd8<uint8_t> other) const { return vaddq_u8(*this, other); }
+    simdjson_inline simd8<uint8_t> operator-(const simd8<uint8_t> other) const { return vsubq_u8(*this, other); }
+    simdjson_inline simd8<uint8_t>& operator+=(const simd8<uint8_t> other) { *this = *this + other; return *this; }
+    simdjson_inline simd8<uint8_t>& operator-=(const simd8<uint8_t> other) { *this = *this - other; return *this; }
+
+    // Order-specific operations
+    simdjson_inline uint8_t max_val() const { return vmaxvq_u8(*this); }
+    simdjson_inline uint8_t min_val() const { return vminvq_u8(*this); }
+    simdjson_inline simd8<uint8_t> max_val(const simd8<uint8_t> other) const { return vmaxq_u8(*this, other); }
+    simdjson_inline simd8<uint8_t> min_val(const simd8<uint8_t> other) const { return vminq_u8(*this, other); }
+    simdjson_inline simd8<bool> operator<=(const simd8<uint8_t> other) const { return vcleq_u8(*this, other); }
+    simdjson_inline simd8<bool> operator>=(const simd8<uint8_t> other) const { return vcgeq_u8(*this, other); }
+    simdjson_inline simd8<bool> operator<(const simd8<uint8_t> other) const { return vcltq_u8(*this, other); }
+    simdjson_inline simd8<bool> operator>(const simd8<uint8_t> other) const { return vcgtq_u8(*this, other); }
+    // Same as >, but instead of guaranteeing all 1's == true, false = 0 and true = nonzero. For ARM, returns all 1's.
+    simdjson_inline simd8<uint8_t> gt_bits(const simd8<uint8_t> other) const { return simd8<uint8_t>(*this > other); }
+    // Same as <, but instead of guaranteeing all 1's == true, false = 0 and true = nonzero. For ARM, returns all 1's.
+    simdjson_inline simd8<uint8_t> lt_bits(const simd8<uint8_t> other) const { return simd8<uint8_t>(*this < other); }
+
+    // Bit-specific operations
+    simdjson_inline simd8<bool> any_bits_set(simd8<uint8_t> bits) const { return vtstq_u8(*this, bits); }
+    simdjson_inline bool any_bits_set_anywhere() const { return this->max_val() != 0; }
+    simdjson_inline bool any_bits_set_anywhere(simd8<uint8_t> bits) const { return (*this & bits).any_bits_set_anywhere(); }
+    template<int N>
+    simdjson_inline simd8<uint8_t> shr() const { return vshrq_n_u8(*this, N); }
+    template<int N>
+    simdjson_inline simd8<uint8_t> shl() const { return vshlq_n_u8(*this, N); }
+
+    // Perform a lookup assuming the value is between 0 and 16 (undefined behavior for out of range values)
+    template<typename L>
+    simdjson_inline simd8<L> lookup_16(simd8<L> lookup_table) const {
+      return lookup_table.apply_lookup_16_to(*this);
+    }
+
+
+    // Copies to 'output" all bytes corresponding to a 0 in the mask (interpreted as a bitset).
+    // Passing a 0 value for mask would be equivalent to writing out every byte to output.
+    // Only the first 16 - count_ones(mask) bytes of the result are significant but 16 bytes
+    // get written.
+    // Design consideration: it seems like a function with the
+    // signature simd8<L> compress(uint16_t mask) would be
+    // sensible, but the AVX ISA makes this kind of approach difficult.
+    template<typename L>
+    simdjson_inline void compress(uint16_t mask, L * output) const {
+      using internal::thintable_epi8;
+      using internal::BitsSetTable256mul2;
+      using internal::pshufb_combine_table;
+      // this particular implementation was inspired by work done by @animetosho
+      // we do it in two steps, first 8 bytes and then second 8 bytes
+      uint8_t mask1 = uint8_t(mask); // least significant 8 bits
+      uint8_t mask2 = uint8_t(mask >> 8); // most significant 8 bits
+      // next line just loads the 64-bit values thintable_epi8[mask1] and
+      // thintable_epi8[mask2] into a 128-bit register, using only
+      // two instructions on most compilers.
+      uint64x2_t shufmask64 = {thintable_epi8[mask1], thintable_epi8[mask2]};
+      uint8x16_t shufmask = vreinterpretq_u8_u64(shufmask64);
+      // we increment by 0x08 the second half of the mask
+#ifdef SIMDJSON_REGULAR_VISUAL_STUDIO
+      uint8x16_t inc = simdjson_make_uint8x16_t(0, 0, 0, 0, 0, 0, 0, 0, 0x08, 0x08, 0x08, 0x08, 0x08, 0x08, 0x08, 0x08);
+#else
+      uint8x16_t inc = {0, 0, 0, 0, 0, 0, 0, 0, 0x08, 0x08, 0x08, 0x08, 0x08, 0x08, 0x08, 0x08};
+#endif
+      shufmask = vaddq_u8(shufmask, inc);
+      // this is the version "nearly pruned"
+      uint8x16_t pruned = vqtbl1q_u8(*this, shufmask);
+      // we still need to put the two halves together.
+      // we compute the popcount of the first half:
+      int pop1 = BitsSetTable256mul2[mask1];
+      // then load the corresponding mask, what it does is to write
+      // only the first pop1 bytes from the first 8 bytes, and then
+      // it fills in with the bytes from the second 8 bytes + some filling
+      // at the end.
+      uint8x16_t compactmask = vld1q_u8(reinterpret_cast<const uint8_t *>(pshufb_combine_table + pop1 * 8));
+      uint8x16_t answer = vqtbl1q_u8(pruned, compactmask);
+      vst1q_u8(reinterpret_cast<uint8_t*>(output), answer);
+    }
+
+    // Copies all bytes corresponding to a 0 in the low half of the mask (interpreted as a
+    // bitset) to output1, then those corresponding to a 0 in the high half to output2.
+    template<typename L>
+    simdjson_inline void compress_halves(uint16_t mask, L *output1, L *output2) const {
+      using internal::thintable_epi8;
+      uint8_t mask1 = uint8_t(mask); // least significant 8 bits
+      uint8_t mask2 = uint8_t(mask >> 8); // most significant 8 bits
+      uint8x8_t compactmask1 = vcreate_u8(thintable_epi8[mask1]);
+      uint8x8_t compactmask2 = vcreate_u8(thintable_epi8[mask2]);
+      // we increment by 0x08 the second half of the mask
+#ifdef SIMDJSON_REGULAR_VISUAL_STUDIO
+      uint8x8_t inc = simdjson_make_uint8x8_t(0x08, 0x08, 0x08, 0x08, 0x08, 0x08, 0x08, 0x08);
+#else
+      uint8x8_t inc = {0x08, 0x08, 0x08, 0x08, 0x08, 0x08, 0x08, 0x08};
+#endif
+      compactmask2 = vadd_u8(compactmask2, inc);
+      // store each result (with the second store possibly overlapping the first)
+      vst1_u8((uint8_t*)output1, vqtbl1_u8(*this, compactmask1));
+      vst1_u8((uint8_t*)output2, vqtbl1_u8(*this, compactmask2));
+    }
+
+    template<typename L>
+    simdjson_inline simd8<L> lookup_16(
+        L replace0,  L replace1,  L replace2,  L replace3,
+        L replace4,  L replace5,  L replace6,  L replace7,
+        L replace8,  L replace9,  L replace10, L replace11,
+        L replace12, L replace13, L replace14, L replace15) const {
+      return lookup_16(simd8<L>::repeat_16(
+        replace0,  replace1,  replace2,  replace3,
+        replace4,  replace5,  replace6,  replace7,
+        replace8,  replace9,  replace10, replace11,
+        replace12, replace13, replace14, replace15
+      ));
+    }
+
+    template<typename T>
+    simdjson_inline simd8<uint8_t> apply_lookup_16_to(const simd8<T> original) {
+      return vqtbl1q_u8(*this, simd8<uint8_t>(original));
+    }
+  };
+
+  // Signed bytes
+  template<>
+  struct simd8<int8_t> {
+    int8x16_t value;
+
+    static simdjson_inline simd8<int8_t> splat(int8_t _value) { return vmovq_n_s8(_value); }
+    static simdjson_inline simd8<int8_t> zero() { return vdupq_n_s8(0); }
+    static simdjson_inline simd8<int8_t> load(const int8_t values[16]) { return vld1q_s8(values); }
+
+    // Conversion from/to SIMD register
+    simdjson_inline simd8(const int8x16_t _value) : value{_value} {}
+    simdjson_inline operator const int8x16_t&() const { return this->value; }
+    simdjson_inline operator int8x16_t&() { return this->value; }
+
+    // Zero constructor
+    simdjson_inline simd8() : simd8(zero()) {}
+    // Splat constructor
+    simdjson_inline simd8(int8_t _value) : simd8(splat(_value)) {}
+    // Array constructor
+    simdjson_inline simd8(const int8_t* values) : simd8(load(values)) {}
+    // Member-by-member initialization
+#ifdef SIMDJSON_REGULAR_VISUAL_STUDIO
+    simdjson_inline simd8(
+      int8_t v0,  int8_t v1,  int8_t v2,  int8_t v3, int8_t v4,  int8_t v5,  int8_t v6,  int8_t v7,
+      int8_t v8,  int8_t v9,  int8_t v10, int8_t v11, int8_t v12, int8_t v13, int8_t v14, int8_t v15
+    ) : simd8(simdjson_make_int8x16_t(
+      v0, v1, v2, v3, v4, v5, v6, v7,
+      v8, v9, v10,v11,v12,v13,v14,v15
+    )) {}
+#else
+    simdjson_inline simd8(
+      int8_t v0,  int8_t v1,  int8_t v2,  int8_t v3, int8_t v4,  int8_t v5,  int8_t v6,  int8_t v7,
+      int8_t v8,  int8_t v9,  int8_t v10, int8_t v11, int8_t v12, int8_t v13, int8_t v14, int8_t v15
+    ) : simd8(int8x16_t{
+      v0, v1, v2, v3, v4, v5, v6, v7,
+      v8, v9, v10,v11,v12,v13,v14,v15
+    }) {}
+#endif
+    // Repeat 16 values as many times as necessary (usually for lookup tables)
+    simdjson_inline static simd8<int8_t> repeat_16(
+      int8_t v0,  int8_t v1,  int8_t v2,  int8_t v3,  int8_t v4,  int8_t v5,  int8_t v6,  int8_t v7,
+      int8_t v8,  int8_t v9,  int8_t v10, int8_t v11, int8_t v12, int8_t v13, int8_t v14, int8_t v15
+    ) {
+      return simd8<int8_t>(
+        v0, v1, v2, v3, v4, v5, v6, v7,
+        v8, v9, v10,v11,v12,v13,v14,v15
+      );
+    }
+
+    // Store to array
+    simdjson_inline void store(int8_t dst[16]) const { return vst1q_s8(dst, *this); }
+
+    // Explicit conversion to/from unsigned
+    //
+    // Under Visual Studio/ARM64 uint8x16_t and int8x16_t are apparently the same type.
+    // In theory, we could check this occurrence with std::same_as and std::enabled_if but it is C++14
+    // and relatively ugly and hard to read.
+#ifndef SIMDJSON_REGULAR_VISUAL_STUDIO
+    simdjson_inline explicit simd8(const uint8x16_t other): simd8(vreinterpretq_s8_u8(other)) {}
+#endif
+    simdjson_inline explicit operator simd8<uint8_t>() const { return vreinterpretq_u8_s8(this->value); }
+
+    // Math
+    simdjson_inline simd8<int8_t> operator+(const simd8<int8_t> other) const { return vaddq_s8(*this, other); }
+    simdjson_inline simd8<int8_t> operator-(const simd8<int8_t> other) const { return vsubq_s8(*this, other); }
+    simdjson_inline simd8<int8_t>& operator+=(const simd8<int8_t> other) { *this = *this + other; return *this; }
+    simdjson_inline simd8<int8_t>& operator-=(const simd8<int8_t> other) { *this = *this - other; return *this; }
+
+    // Order-sensitive comparisons
+    simdjson_inline simd8<int8_t> max_val(const simd8<int8_t> other) const { return vmaxq_s8(*this, other); }
+    simdjson_inline simd8<int8_t> min_val(const simd8<int8_t> other) const { return vminq_s8(*this, other); }
+    simdjson_inline simd8<bool> operator>(const simd8<int8_t> other) const { return vcgtq_s8(*this, other); }
+    simdjson_inline simd8<bool> operator<(const simd8<int8_t> other) const { return vcltq_s8(*this, other); }
+    simdjson_inline simd8<bool> operator==(const simd8<int8_t> other) const { return vceqq_s8(*this, other); }
+
+    template<int N=1>
+    simdjson_inline simd8<int8_t> prev(const simd8<int8_t> prev_chunk) const {
+      return vextq_s8(prev_chunk, *this, 16 - N);
+    }
+
+    // Perform a lookup assuming no value is larger than 16
+    template<typename L>
+    simdjson_inline simd8<L> lookup_16(simd8<L> lookup_table) const {
+      return lookup_table.apply_lookup_16_to(*this);
+    }
+    template<typename L>
+    simdjson_inline simd8<L> lookup_16(
+        L replace0,  L replace1,  L replace2,  L replace3,
+        L replace4,  L replace5,  L replace6,  L replace7,
+        L replace8,  L replace9,  L replace10, L replace11,
+        L replace12, L replace13, L replace14, L replace15) const {
+      return lookup_16(simd8<L>::repeat_16(
+        replace0,  replace1,  replace2,  replace3,
+        replace4,  replace5,  replace6,  replace7,
+        replace8,  replace9,  replace10, replace11,
+        replace12, replace13, replace14, replace15
+      ));
+    }
+
+    template<typename T>
+    simdjson_inline simd8<int8_t> apply_lookup_16_to(const simd8<T> original) {
+      return vqtbl1q_s8(*this, simd8<uint8_t>(original));
+    }
+  };
+
+  template<typename T>
+  struct simd8x64 {
+    static constexpr int NUM_CHUNKS = 64 / sizeof(simd8<T>);
+    static_assert(NUM_CHUNKS == 4, "ARM kernel should use four registers per 64-byte block.");
+    const simd8<T> chunks[NUM_CHUNKS];
+
+    simd8x64(const simd8x64<T>& o) = delete; // no copy allowed
+    simd8x64<T>& operator=(const simd8<T>& other) = delete; // no assignment allowed
+    simd8x64() = delete; // no default constructor allowed
+
+    simdjson_inline simd8x64(const simd8<T> chunk0, const simd8<T> chunk1, const simd8<T> chunk2, const simd8<T> chunk3) : chunks{chunk0, chunk1, chunk2, chunk3} {}
+    simdjson_inline simd8x64(const T ptr[64]) : chunks{simd8<T>::load(ptr), simd8<T>::load(ptr+16), simd8<T>::load(ptr+32), simd8<T>::load(ptr+48)} {}
+
+    simdjson_inline void store(T ptr[64]) const {
+      this->chunks[0].store(ptr+sizeof(simd8<T>)*0);
+      this->chunks[1].store(ptr+sizeof(simd8<T>)*1);
+      this->chunks[2].store(ptr+sizeof(simd8<T>)*2);
+      this->chunks[3].store(ptr+sizeof(simd8<T>)*3);
+    }
+
+    simdjson_inline simd8<T> reduce_or() const {
+      return (this->chunks[0] | this->chunks[1]) | (this->chunks[2] | this->chunks[3]);
+    }
+
+
+    simdjson_inline uint64_t compress(uint64_t mask, T * output) const {
+      uint64_t popcounts = vget_lane_u64(vreinterpret_u64_u8(vcnt_u8(vcreate_u8(~mask))), 0);
+      // compute the prefix sum of the popcounts of each byte
+      uint64_t offsets = popcounts * 0x0101010101010101;
+      this->chunks[0].compress_halves(uint16_t(mask), output, &output[popcounts & 0xFF]);
+      this->chunks[1].compress_halves(uint16_t(mask >> 16), &output[(offsets >> 8) & 0xFF], &output[(offsets >> 16) & 0xFF]);
+      this->chunks[2].compress_halves(uint16_t(mask >> 32), &output[(offsets >> 24) & 0xFF], &output[(offsets >> 32) & 0xFF]);
+      this->chunks[3].compress_halves(uint16_t(mask >> 48), &output[(offsets >> 40) & 0xFF], &output[(offsets >> 48) & 0xFF]);
+      return offsets >> 56;
+    }
+
+    simdjson_inline uint64_t to_bitmask() const {
+#ifdef SIMDJSON_REGULAR_VISUAL_STUDIO
+      const uint8x16_t bit_mask = simdjson_make_uint8x16_t(
+        0x01, 0x02, 0x4, 0x8, 0x10, 0x20, 0x40, 0x80,
+        0x01, 0x02, 0x4, 0x8, 0x10, 0x20, 0x40, 0x80
+      );
+#else
+      const uint8x16_t bit_mask = {
+        0x01, 0x02, 0x4, 0x8, 0x10, 0x20, 0x40, 0x80,
+        0x01, 0x02, 0x4, 0x8, 0x10, 0x20, 0x40, 0x80
+      };
+#endif
+      // Add each of the elements next to each other, successively, to stuff each 8 byte mask into one.
+      uint8x16_t sum0 = vpaddq_u8(this->chunks[0] & bit_mask, this->chunks[1] & bit_mask);
+      uint8x16_t sum1 = vpaddq_u8(this->chunks[2] & bit_mask, this->chunks[3] & bit_mask);
+      sum0 = vpaddq_u8(sum0, sum1);
+      sum0 = vpaddq_u8(sum0, sum0);
+      return vgetq_lane_u64(vreinterpretq_u64_u8(sum0), 0);
+    }
+
+    simdjson_inline uint64_t eq(const T m) const {
+      const simd8<T> mask = simd8<T>::splat(m);
+      return  simd8x64<bool>(
+        this->chunks[0] == mask,
+        this->chunks[1] == mask,
+        this->chunks[2] == mask,
+        this->chunks[3] == mask
+      ).to_bitmask();
+    }
+
+    simdjson_inline uint64_t lteq(const T m) const {
+      const simd8<T> mask = simd8<T>::splat(m);
+      return  simd8x64<bool>(
+        this->chunks[0] <= mask,
+        this->chunks[1] <= mask,
+        this->chunks[2] <= mask,
+        this->chunks[3] <= mask
+      ).to_bitmask();
+    }
+  }; // struct simd8x64<T>
+
+} // namespace simd
+} // unnamed namespace
+} // namespace arm64
+} // namespace simdjson
+
+#endif // SIMDJSON_ARM64_SIMD_H
diff --git a/contrib/libs/simdjson/include/simdjson/arm64/stringparsing_defs.h b/contrib/libs/simdjson/include/simdjson/arm64/stringparsing_defs.h
new file mode 100644
index 00000000000..30d02faff3f
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/arm64/stringparsing_defs.h
@@ -0,0 +1,53 @@
+#ifndef SIMDJSON_ARM64_STRINGPARSING_DEFS_H
+#define SIMDJSON_ARM64_STRINGPARSING_DEFS_H
+
+#ifndef SIMDJSON_CONDITIONAL_INCLUDE
+#include "simdjson/arm64/base.h"
+#include "simdjson/arm64/simd.h"
+#include "simdjson/arm64/bitmanipulation.h"
+#endif // SIMDJSON_CONDITIONAL_INCLUDE
+
+namespace simdjson {
+namespace arm64 {
+namespace {
+
+using namespace simd;
+
+// Holds backslashes and quotes locations.
+struct backslash_and_quote {
+public:
+  static constexpr uint32_t BYTES_PROCESSED = 32;
+  simdjson_inline static backslash_and_quote copy_and_find(const uint8_t *src, uint8_t *dst);
+
+  simdjson_inline bool has_quote_first() { return ((bs_bits - 1) & quote_bits) != 0; }
+  simdjson_inline bool has_backslash() { return bs_bits != 0; }
+  simdjson_inline int quote_index() { return trailing_zeroes(quote_bits); }
+  simdjson_inline int backslash_index() { return trailing_zeroes(bs_bits); }
+
+  uint32_t bs_bits;
+  uint32_t quote_bits;
+}; // struct backslash_and_quote
+
+simdjson_inline backslash_and_quote backslash_and_quote::copy_and_find(const uint8_t *src, uint8_t *dst) {
+  // this can read up to 31 bytes beyond the buffer size, but we require
+  // SIMDJSON_PADDING of padding
+  static_assert(SIMDJSON_PADDING >= (BYTES_PROCESSED - 1), "backslash and quote finder must process fewer than SIMDJSON_PADDING bytes");
+  simd8<uint8_t> v0(src);
+  simd8<uint8_t> v1(src + sizeof(v0));
+  v0.store(dst);
+  v1.store(dst + sizeof(v0));
+
+  // Getting a 64-bit bitmask is much cheaper than multiple 16-bit bitmasks on ARM; therefore, we
+  // smash them together into a 64-byte mask and get the bitmask from there.
+  uint64_t bs_and_quote = simd8x64<bool>(v0 == '\\', v1 == '\\', v0 == '"', v1 == '"').to_bitmask();
+  return {
+    uint32_t(bs_and_quote),      // bs_bits
+    uint32_t(bs_and_quote >> 32) // quote_bits
+  };
+}
+
+} // unnamed namespace
+} // namespace arm64
+} // namespace simdjson
+
+#endif // SIMDJSON_ARM64_STRINGPARSING_DEFS_H
diff --git a/contrib/libs/simdjson/include/simdjson/base.h b/contrib/libs/simdjson/include/simdjson/base.h
new file mode 100644
index 00000000000..a73b84bd082
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/base.h
@@ -0,0 +1,60 @@
+/**
+ * @file Base declarations for all simdjson headers
+ * @private
+ */
+#ifndef SIMDJSON_BASE_H
+#define SIMDJSON_BASE_H
+
+#include "simdjson/common_defs.h"
+#include "simdjson/compiler_check.h"
+#include "simdjson/error.h"
+#include "simdjson/portability.h"
+
+/**
+ * @brief The top level simdjson namespace, containing everything the library provides.
+ */
+namespace simdjson {
+
+SIMDJSON_PUSH_DISABLE_UNUSED_WARNINGS
+
+/** The maximum document size supported by simdjson. */
+constexpr size_t SIMDJSON_MAXSIZE_BYTES = 0xFFFFFFFF;
+
+/**
+ * The amount of padding needed in a buffer to parse JSON.
+ *
+ * The input buf should be readable up to buf + SIMDJSON_PADDING
+ * this is a stopgap; there should be a better description of the
+ * main loop and its behavior that abstracts over this
+ * See https://github.com/simdjson/simdjson/issues/174
+ */
+constexpr size_t SIMDJSON_PADDING = 64;
+
+/**
+ * By default, simdjson supports this many nested objects and arrays.
+ *
+ * This is the default for parser::max_depth().
+ */
+constexpr size_t DEFAULT_MAX_DEPTH = 1024;
+
+SIMDJSON_POP_DISABLE_UNUSED_WARNINGS
+
+class implementation;
+struct padded_string;
+class padded_string_view;
+enum class stage1_mode;
+
+namespace internal {
+
+template<typename T>
+class atomic_ptr;
+class dom_parser_implementation;
+class escape_json_string;
+class tape_ref;
+struct value128;
+enum class tape_type;
+
+} // namespace internal
+} // namespace simdjson
+
+#endif // SIMDJSON_BASE_H
diff --git a/contrib/libs/simdjson/include/simdjson/common_defs.h b/contrib/libs/simdjson/include/simdjson/common_defs.h
new file mode 100644
index 00000000000..d0ae083ecba
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/common_defs.h
@@ -0,0 +1,347 @@
+#ifndef SIMDJSON_COMMON_DEFS_H
+#define SIMDJSON_COMMON_DEFS_H
+
+#include <cassert>
+#include "simdjson/compiler_check.h"
+#include "simdjson/portability.h"
+
+namespace simdjson {
+namespace internal {
+/**
+ * @private
+ * Our own implementation of the C++17 to_chars function.
+ * Defined in src/to_chars
+ */
+char *to_chars(char *first, const char *last, double value);
+/**
+ * @private
+ * A number parsing routine.
+ * Defined in src/from_chars
+ */
+double from_chars(const char *first) noexcept;
+double from_chars(const char *first, const char* end) noexcept;
+}
+
+#ifndef SIMDJSON_EXCEPTIONS
+#if __cpp_exceptions
+#define SIMDJSON_EXCEPTIONS 1
+#else
+#define SIMDJSON_EXCEPTIONS 0
+#endif
+#endif
+
+} // namespace simdjson
+
+#if defined(__GNUC__)
+  // Marks a block with a name so that MCA analysis can see it.
+  #define SIMDJSON_BEGIN_DEBUG_BLOCK(name) __asm volatile("# LLVM-MCA-BEGIN " #name);
+  #define SIMDJSON_END_DEBUG_BLOCK(name) __asm volatile("# LLVM-MCA-END " #name);
+  #define SIMDJSON_DEBUG_BLOCK(name, block) BEGIN_DEBUG_BLOCK(name); block; END_DEBUG_BLOCK(name);
+#else
+  #define SIMDJSON_BEGIN_DEBUG_BLOCK(name)
+  #define SIMDJSON_END_DEBUG_BLOCK(name)
+  #define SIMDJSON_DEBUG_BLOCK(name, block)
+#endif
+
+// Align to N-byte boundary
+#define SIMDJSON_ROUNDUP_N(a, n) (((a) + ((n)-1)) & ~((n)-1))
+#define SIMDJSON_ROUNDDOWN_N(a, n) ((a) & ~((n)-1))
+
+#define SIMDJSON_ISALIGNED_N(ptr, n) (((uintptr_t)(ptr) & ((n)-1)) == 0)
+
+#if SIMDJSON_REGULAR_VISUAL_STUDIO
+  // We could use [[deprecated]] but it requires C++14
+  #define simdjson_deprecated __declspec(deprecated)
+
+  #define simdjson_really_inline __forceinline
+  #define simdjson_never_inline __declspec(noinline)
+
+  #define simdjson_unused
+  #define simdjson_warn_unused
+
+  #ifndef simdjson_likely
+  #define simdjson_likely(x) x
+  #endif
+  #ifndef simdjson_unlikely
+  #define simdjson_unlikely(x) x
+  #endif
+
+  #define SIMDJSON_PUSH_DISABLE_WARNINGS __pragma(warning( push ))
+  #define SIMDJSON_PUSH_DISABLE_ALL_WARNINGS __pragma(warning( push, 0 ))
+  #define SIMDJSON_DISABLE_VS_WARNING(WARNING_NUMBER) __pragma(warning( disable : WARNING_NUMBER ))
+  // Get rid of Intellisense-only warnings (Code Analysis)
+  // Though __has_include is C++17, it is supported in Visual Studio 2017 or better (_MSC_VER>=1910).
+  #ifdef __has_include
+  #if __has_include(<CppCoreCheck\Warnings.h>)
+  #error #include <CppCoreCheck\Warnings.h>
+  #define SIMDJSON_DISABLE_UNDESIRED_WARNINGS SIMDJSON_DISABLE_VS_WARNING(ALL_CPPCORECHECK_WARNINGS)
+  #endif
+  #endif
+
+  #ifndef SIMDJSON_DISABLE_UNDESIRED_WARNINGS
+  #define SIMDJSON_DISABLE_UNDESIRED_WARNINGS
+  #endif
+
+  #define SIMDJSON_DISABLE_DEPRECATED_WARNING SIMDJSON_DISABLE_VS_WARNING(4996)
+  #define SIMDJSON_DISABLE_STRICT_OVERFLOW_WARNING
+  #define SIMDJSON_POP_DISABLE_WARNINGS __pragma(warning( pop ))
+
+  #define SIMDJSON_PUSH_DISABLE_UNUSED_WARNINGS
+  #define SIMDJSON_POP_DISABLE_UNUSED_WARNINGS
+
+#else // SIMDJSON_REGULAR_VISUAL_STUDIO
+  // We could use [[deprecated]] but it requires C++14
+  #define simdjson_deprecated __attribute__((deprecated))
+
+  #define simdjson_really_inline inline __attribute__((always_inline))
+  #define simdjson_never_inline inline __attribute__((noinline))
+
+  #define simdjson_unused __attribute__((unused))
+  #define simdjson_warn_unused __attribute__((warn_unused_result))
+
+  #ifndef simdjson_likely
+  #define simdjson_likely(x) __builtin_expect(!!(x), 1)
+  #endif
+  #ifndef simdjson_unlikely
+  #define simdjson_unlikely(x) __builtin_expect(!!(x), 0)
+  #endif
+
+  #define SIMDJSON_PUSH_DISABLE_WARNINGS _Pragma("GCC diagnostic push")
+  // gcc doesn't seem to disable all warnings with all and extra, add warnings here as necessary
+  // We do it separately for clang since it has different warnings.
+  #ifdef __clang__
+  // clang is missing -Wmaybe-uninitialized.
+  #define SIMDJSON_PUSH_DISABLE_ALL_WARNINGS SIMDJSON_PUSH_DISABLE_WARNINGS \
+    SIMDJSON_DISABLE_GCC_WARNING(-Weffc++) \
+    SIMDJSON_DISABLE_GCC_WARNING(-Wall) \
+    SIMDJSON_DISABLE_GCC_WARNING(-Wconversion) \
+    SIMDJSON_DISABLE_GCC_WARNING(-Wextra) \
+    SIMDJSON_DISABLE_GCC_WARNING(-Wattributes) \
+    SIMDJSON_DISABLE_GCC_WARNING(-Wimplicit-fallthrough) \
+    SIMDJSON_DISABLE_GCC_WARNING(-Wnon-virtual-dtor) \
+    SIMDJSON_DISABLE_GCC_WARNING(-Wreturn-type) \
+    SIMDJSON_DISABLE_GCC_WARNING(-Wshadow) \
+    SIMDJSON_DISABLE_GCC_WARNING(-Wunused-parameter) \
+    SIMDJSON_DISABLE_GCC_WARNING(-Wunused-variable)
+  #else // __clang__
+  #define SIMDJSON_PUSH_DISABLE_ALL_WARNINGS SIMDJSON_PUSH_DISABLE_WARNINGS \
+    SIMDJSON_DISABLE_GCC_WARNING(-Weffc++) \
+    SIMDJSON_DISABLE_GCC_WARNING(-Wall) \
+    SIMDJSON_DISABLE_GCC_WARNING(-Wconversion) \
+    SIMDJSON_DISABLE_GCC_WARNING(-Wextra) \
+    SIMDJSON_DISABLE_GCC_WARNING(-Wattributes) \
+    SIMDJSON_DISABLE_GCC_WARNING(-Wimplicit-fallthrough) \
+    SIMDJSON_DISABLE_GCC_WARNING(-Wnon-virtual-dtor) \
+    SIMDJSON_DISABLE_GCC_WARNING(-Wreturn-type) \
+    SIMDJSON_DISABLE_GCC_WARNING(-Wshadow) \
+    SIMDJSON_DISABLE_GCC_WARNING(-Wunused-parameter) \
+    SIMDJSON_DISABLE_GCC_WARNING(-Wunused-variable) \
+    SIMDJSON_DISABLE_GCC_WARNING(-Wmaybe-uninitialized) \
+    SIMDJSON_DISABLE_GCC_WARNING(-Wformat-security)
+  #endif // __clang__
+
+  #define SIMDJSON_PRAGMA(P) _Pragma(#P)
+  #define SIMDJSON_DISABLE_GCC_WARNING(WARNING) SIMDJSON_PRAGMA(GCC diagnostic ignored #WARNING)
+  #if SIMDJSON_CLANG_VISUAL_STUDIO
+  #define SIMDJSON_DISABLE_UNDESIRED_WARNINGS SIMDJSON_DISABLE_GCC_WARNING(-Wmicrosoft-include)
+  #else
+  #define SIMDJSON_DISABLE_UNDESIRED_WARNINGS
+  #endif
+  #define SIMDJSON_DISABLE_DEPRECATED_WARNING SIMDJSON_DISABLE_GCC_WARNING(-Wdeprecated-declarations)
+  #define SIMDJSON_DISABLE_STRICT_OVERFLOW_WARNING SIMDJSON_DISABLE_GCC_WARNING(-Wstrict-overflow)
+  #define SIMDJSON_POP_DISABLE_WARNINGS _Pragma("GCC diagnostic pop")
+
+  #define SIMDJSON_PUSH_DISABLE_UNUSED_WARNINGS SIMDJSON_PUSH_DISABLE_WARNINGS \
+    SIMDJSON_DISABLE_GCC_WARNING(-Wunused)
+  #define SIMDJSON_POP_DISABLE_UNUSED_WARNINGS SIMDJSON_POP_DISABLE_WARNINGS
+
+
+
+#endif // MSC_VER
+
+#if defined(simdjson_inline)
+  // Prefer the user's definition of simdjson_inline; don't define it ourselves.
+#elif defined(__GNUC__) && !defined(__OPTIMIZE__)
+  // If optimizations are disabled, forcing inlining can lead to significant
+  // code bloat and high compile times. Don't use simdjson_really_inline for
+  // unoptimized builds.
+  #define simdjson_inline inline
+#else
+  // Force inlining for most simdjson functions.
+  #define simdjson_inline simdjson_really_inline
+#endif
+
+#if SIMDJSON_VISUAL_STUDIO
+    /**
+     * Windows users need to do some extra work when building
+     * or using a dynamic library (DLL). When building, we need
+     * to set SIMDJSON_DLLIMPORTEXPORT to __declspec(dllexport).
+     * When *using* the DLL, the user needs to set
+     * SIMDJSON_DLLIMPORTEXPORT __declspec(dllimport).
+     *
+     * Static libraries not need require such work.
+     *
+     * It does not matter here whether you are using
+     * the regular visual studio or clang under visual
+     * studio, you still need to handle these issues.
+     *
+     * Non-Windows systems do not have this complexity.
+     */
+    #if SIMDJSON_BUILDING_WINDOWS_DYNAMIC_LIBRARY
+    // We set SIMDJSON_BUILDING_WINDOWS_DYNAMIC_LIBRARY when we build a DLL under Windows.
+    // It should never happen that both SIMDJSON_BUILDING_WINDOWS_DYNAMIC_LIBRARY and
+    // SIMDJSON_USING_WINDOWS_DYNAMIC_LIBRARY are set.
+    #define SIMDJSON_DLLIMPORTEXPORT __declspec(dllexport)
+    #elif SIMDJSON_USING_WINDOWS_DYNAMIC_LIBRARY
+    // Windows user who call a dynamic library should set SIMDJSON_USING_WINDOWS_DYNAMIC_LIBRARY to 1.
+    #define SIMDJSON_DLLIMPORTEXPORT __declspec(dllimport)
+    #else
+    // We assume by default static linkage
+    #define SIMDJSON_DLLIMPORTEXPORT
+    #endif
+
+/**
+ * Workaround for the vcpkg package manager. Only vcpkg should
+ * ever touch the next line. The SIMDJSON_USING_LIBRARY macro is otherwise unused.
+ */
+#if SIMDJSON_USING_LIBRARY
+#define SIMDJSON_DLLIMPORTEXPORT __declspec(dllimport)
+#endif
+/**
+ * End of workaround for the vcpkg package manager.
+ */
+#else
+    #define SIMDJSON_DLLIMPORTEXPORT
+#endif
+
+// C++17 requires string_view.
+#if SIMDJSON_CPLUSPLUS17
+#define SIMDJSON_HAS_STRING_VIEW
+#include <string_view> // by the standard, this has to be safe.
+#endif
+
+// This macro (__cpp_lib_string_view) has to be defined
+// for C++17 and better, but if it is otherwise defined,
+// we are going to assume that string_view is available
+// even if we do not have C++17 support.
+#ifdef __cpp_lib_string_view
+#define SIMDJSON_HAS_STRING_VIEW
+#endif
+
+// Some systems have string_view even if we do not have C++17 support,
+// and even if __cpp_lib_string_view is undefined, it is the case
+// with Apple clang version 11.
+// We must handle it. *This is important.*
+#ifndef SIMDJSON_HAS_STRING_VIEW
+#if defined __has_include
+// do not combine the next #if with the previous one (unsafe)
+#if __has_include (<string_view>)
+// now it is safe to trigger the include
+#include <string_view> // though the file is there, it does not follow that we got the implementation
+#if defined(_LIBCPP_STRING_VIEW)
+// Ah! So we under libc++ which under its Library Fundamentals Technical Specification, which preceded C++17,
+// included string_view.
+// This means that we have string_view *even though* we may not have C++17.
+#define SIMDJSON_HAS_STRING_VIEW
+#endif // _LIBCPP_STRING_VIEW
+#endif // __has_include (<string_view>)
+#endif // defined __has_include
+#endif // def SIMDJSON_HAS_STRING_VIEW
+// end of complicated but important routine to try to detect string_view.
+
+//
+// Backfill std::string_view using nonstd::string_view on systems where
+// we expect that string_view is missing. Important: if we get this wrong,
+// we will end up with two string_view definitions and potential trouble.
+// That is why we work so hard above to avoid it.
+//
+#ifndef SIMDJSON_HAS_STRING_VIEW
+SIMDJSON_PUSH_DISABLE_ALL_WARNINGS
+#error #include "simdjson/nonstd/string_view.hpp"
+SIMDJSON_POP_DISABLE_WARNINGS
+
+namespace std {
+  using string_view = nonstd::string_view;
+}
+#endif // SIMDJSON_HAS_STRING_VIEW
+#undef SIMDJSON_HAS_STRING_VIEW // We are not going to need this macro anymore.
+
+/// If EXPR is an error, returns it.
+#define SIMDJSON_TRY(EXPR) { auto _err = (EXPR); if (_err) { return _err; } }
+
+// Unless the programmer has already set SIMDJSON_DEVELOPMENT_CHECKS,
+// we want to set it under debug builds. We detect a debug build
+// under Visual Studio when the _DEBUG macro is set. Under the other
+// compilers, we use the fact that they define __OPTIMIZE__ whenever
+// they allow optimizations.
+// It is possible that this could miss some cases where SIMDJSON_DEVELOPMENT_CHECKS
+// is helpful, but the programmer can set the macro SIMDJSON_DEVELOPMENT_CHECKS.
+// It could also wrongly set SIMDJSON_DEVELOPMENT_CHECKS (e.g., if the programmer
+// sets _DEBUG in a release build under Visual Studio, or if some compiler fails to
+// set the __OPTIMIZE__ macro).
+#ifndef SIMDJSON_DEVELOPMENT_CHECKS
+#ifdef _MSC_VER
+// Visual Studio seems to set _DEBUG for debug builds.
+#ifdef _DEBUG
+#define SIMDJSON_DEVELOPMENT_CHECKS 1
+#endif // _DEBUG
+#else // _MSC_VER
+// All other compilers appear to set __OPTIMIZE__ to a positive integer
+// when the compiler is optimizing.
+#ifndef __OPTIMIZE__
+#define SIMDJSON_DEVELOPMENT_CHECKS 1
+#endif // __OPTIMIZE__
+#endif // _MSC_VER
+#endif // SIMDJSON_DEVELOPMENT_CHECKS
+
+// The SIMDJSON_CHECK_EOF macro is a feature flag for the "don't require padding"
+// feature.
+
+#if SIMDJSON_CPLUSPLUS17
+// if we have C++, then fallthrough is a default attribute
+# define simdjson_fallthrough [[fallthrough]]
+// check if we have __attribute__ support
+#elif defined(__has_attribute)
+// check if we have the __fallthrough__ attribute
+#if __has_attribute(__fallthrough__)
+// we are good to go:
+# define simdjson_fallthrough                    __attribute__((__fallthrough__))
+#endif // __has_attribute(__fallthrough__)
+#endif // SIMDJSON_CPLUSPLUS17
+// on some systems, we simply do not have support for fallthrough, so use a default:
+#ifndef simdjson_fallthrough
+# define simdjson_fallthrough do {} while (0)  /* fallthrough */
+#endif // simdjson_fallthrough
+
+#if SIMDJSON_DEVELOPMENT_CHECKS
+#define SIMDJSON_DEVELOPMENT_ASSERT(expr) do { assert ((expr)); } while (0)
+#else
+#define SIMDJSON_DEVELOPMENT_ASSERT(expr) do { } while (0)
+#endif
+
+#ifndef SIMDJSON_UTF8VALIDATION
+#define SIMDJSON_UTF8VALIDATION 1
+#endif
+
+#ifdef __has_include
+// How do we detect that a compiler supports vbmi2?
+// For sure if the following header is found, we are ok?
+#if __has_include(<avx512vbmi2intrin.h>)
+#define SIMDJSON_COMPILER_SUPPORTS_VBMI2 1
+#endif
+#endif
+
+#ifdef _MSC_VER
+#if _MSC_VER >= 1920
+// Visual Studio 2019 and up support VBMI2 under x64 even if the header
+// avx512vbmi2intrin.h is not found.
+#define SIMDJSON_COMPILER_SUPPORTS_VBMI2 1
+#endif
+#endif
+
+// By default, we allow AVX512.
+#ifndef SIMDJSON_AVX512_ALLOWED
+#define SIMDJSON_AVX512_ALLOWED 1
+#endif
+
+#endif // SIMDJSON_COMMON_DEFS_H
diff --git a/contrib/libs/simdjson/include/simdjson/compiler_check.h b/contrib/libs/simdjson/include/simdjson/compiler_check.h
new file mode 100644
index 00000000000..1d0d03d1545
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/compiler_check.h
@@ -0,0 +1,53 @@
+#ifndef SIMDJSON_COMPILER_CHECK_H
+#define SIMDJSON_COMPILER_CHECK_H
+
+#ifndef __cplusplus
+#error simdjson requires a C++ compiler
+#endif
+
+#ifndef SIMDJSON_CPLUSPLUS
+#if defined(_MSVC_LANG) && !defined(__clang__)
+#define SIMDJSON_CPLUSPLUS (_MSC_VER == 1900 ? 201103L : _MSVC_LANG)
+#else
+#define SIMDJSON_CPLUSPLUS __cplusplus
+#endif
+#endif
+
+// C++ 23
+#if !defined(SIMDJSON_CPLUSPLUS23) && (SIMDJSON_CPLUSPLUS >= 202302L)
+#define SIMDJSON_CPLUSPLUS23 1
+#endif
+
+// C++ 20
+#if !defined(SIMDJSON_CPLUSPLUS20) && (SIMDJSON_CPLUSPLUS >= 202002L)
+#define SIMDJSON_CPLUSPLUS20 1
+#endif
+
+// C++ 17
+#if !defined(SIMDJSON_CPLUSPLUS17) && (SIMDJSON_CPLUSPLUS >= 201703L)
+#define SIMDJSON_CPLUSPLUS17 1
+#endif
+
+// C++ 14
+#if !defined(SIMDJSON_CPLUSPLUS14) && (SIMDJSON_CPLUSPLUS >= 201402L)
+#define SIMDJSON_CPLUSPLUS14 1
+#endif
+
+// C++ 11
+#if !defined(SIMDJSON_CPLUSPLUS11) && (SIMDJSON_CPLUSPLUS >= 201103L)
+#define SIMDJSON_CPLUSPLUS11 1
+#endif
+
+#ifndef SIMDJSON_CPLUSPLUS11
+#error simdjson requires a compiler compliant with the C++11 standard
+#endif
+
+#ifndef SIMDJSON_IF_CONSTEXPR
+#if SIMDJSON_CPLUSPLUS17
+#define SIMDJSON_IF_CONSTEXPR if constexpr
+#else
+#define SIMDJSON_IF_CONSTEXPR if
+#endif
+#endif
+
+#endif // SIMDJSON_COMPILER_CHECK_H
diff --git a/contrib/libs/simdjson/include/simdjson/dom/base.h b/contrib/libs/simdjson/include/simdjson/dom/base.h
new file mode 100644
index 00000000000..b862277a211
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/dom/base.h
@@ -0,0 +1,54 @@
+#ifndef SIMDJSON_DOM_BASE_H
+#define SIMDJSON_DOM_BASE_H
+
+#include "simdjson/base.h"
+
+namespace simdjson {
+
+/**
+ * @brief A DOM API on top of the simdjson parser.
+ */
+namespace dom {
+
+/** The default batch size for parser.parse_many() and parser.load_many() */
+static constexpr size_t DEFAULT_BATCH_SIZE = 1000000;
+/**
+ * Some adversary might try to set the batch size to 0 or 1, which might cause problems.
+ * We set a minimum of 32B since anything else is highly likely to be an error. In practice,
+ * most users will want a much larger batch size.
+ *
+ * All non-negative MINIMAL_BATCH_SIZE values should be 'safe' except that, obviously, no JSON
+ * document can ever span 0 or 1 byte and that very large values would create memory allocation issues.
+ */
+static constexpr size_t MINIMAL_BATCH_SIZE = 32;
+
+/**
+ * It is wasteful to allocate memory for tiny documents (e.g., 4 bytes).
+ */
+static constexpr size_t MINIMAL_DOCUMENT_CAPACITY = 32;
+
+class array;
+class document;
+class document_stream;
+class element;
+class key_value_pair;
+class object;
+class parser;
+
+#ifdef SIMDJSON_THREADS_ENABLED
+struct stage1_worker;
+#endif // SIMDJSON_THREADS_ENABLED
+
+} // namespace dom
+
+namespace internal {
+
+template<typename T>
+class string_builder;
+class tape_ref;
+
+} // namespace internal
+
+} // namespace simdjson
+
+#endif // SIMDJSON_DOM_BASE_H
\ No newline at end of file
diff --git a/contrib/libs/simdjson/include/simdjson/dom/document.h b/contrib/libs/simdjson/include/simdjson/dom/document.h
new file mode 100644
index 00000000000..6c5e284bb47
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/dom/document.h
@@ -0,0 +1,91 @@
+#ifndef SIMDJSON_DOM_DOCUMENT_H
+#define SIMDJSON_DOM_DOCUMENT_H
+
+#include "simdjson/dom/base.h"
+
+#include <memory>
+
+namespace simdjson {
+namespace dom {
+
+/**
+ * A parsed JSON document.
+ *
+ * This class cannot be copied, only moved, to avoid unintended allocations.
+ */
+class document {
+public:
+  /**
+   * Create a document container with zero capacity.
+   *
+   * The parser will allocate capacity as needed.
+   */
+  document() noexcept = default;
+  ~document() noexcept = default;
+
+  /**
+   * Take another document's buffers.
+   *
+   * @param other The document to take. Its capacity is zeroed and it is invalidated.
+   */
+  document(document &&other) noexcept = default;
+  /** @private */
+  document(const document &) = delete; // Disallow copying
+  /**
+   * Take another document's buffers.
+   *
+   * @param other The document to take. Its capacity is zeroed.
+   */
+  document &operator=(document &&other) noexcept = default;
+  /** @private */
+  document &operator=(const document &) = delete; // Disallow copying
+
+  /**
+   * Get the root element of this document as a JSON array.
+   */
+  element root() const noexcept;
+
+  /**
+   * @private Dump the raw tape for debugging.
+   *
+   * @param os the stream to output to.
+   * @return false if the tape is likely wrong (e.g., you did not parse a valid JSON).
+   */
+  bool dump_raw_tape(std::ostream &os) const noexcept;
+
+  /** @private Structural values. */
+  std::unique_ptr<uint64_t[]> tape{};
+
+  /** @private String values.
+   *
+   * Should be at least byte_capacity.
+   */
+  std::unique_ptr<uint8_t[]> string_buf{};
+  /** @private Allocate memory to support
+   * input JSON documents of up to len bytes.
+   *
+   * When calling this function, you lose
+   * all the data.
+   *
+   * The memory allocation is strict: you
+   * can you use this function to increase
+   * or lower the amount of allocated memory.
+   * Passsing zero clears the memory.
+   */
+  error_code allocate(size_t len) noexcept;
+  /** @private Capacity in bytes, in terms
+   * of how many bytes of input JSON we can
+   * support.
+   */
+  size_t capacity() const noexcept;
+
+
+private:
+  size_t allocated_capacity{0};
+  friend class parser;
+}; // class document
+
+} // namespace dom
+} // namespace simdjson
+
+#endif // SIMDJSON_DOM_DOCUMENT_H
diff --git a/contrib/libs/simdjson/include/simdjson/error-inl.h b/contrib/libs/simdjson/include/simdjson/error-inl.h
new file mode 100644
index 00000000000..5cf61d9c307
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/error-inl.h
@@ -0,0 +1,184 @@
+#ifndef SIMDJSON_ERROR_INL_H
+#define SIMDJSON_ERROR_INL_H
+
+#include "simdjson/error.h"
+
+#include <iostream>
+
+namespace simdjson {
+namespace internal {
+  // We store the error code so we can validate the error message is associated with the right code
+  struct error_code_info {
+    error_code code;
+    const char* message; // do not use a fancy std::string where a simple C string will do (no alloc, no destructor)
+  };
+  // These MUST match the codes in error_code. We check this constraint in basictests.
+  extern SIMDJSON_DLLIMPORTEXPORT const error_code_info error_codes[];
+} // namespace internal
+
+
+inline const char *error_message(error_code error) noexcept {
+  // If you're using error_code, we're trusting you got it from the enum.
+  return internal::error_codes[int(error)].message;
+}
+
+// deprecated function
+#ifndef SIMDJSON_DISABLE_DEPRECATED_API
+inline const std::string error_message(int error) noexcept {
+  if (error < 0 || error >= error_code::NUM_ERROR_CODES) {
+    return internal::error_codes[UNEXPECTED_ERROR].message;
+  }
+  return internal::error_codes[error].message;
+}
+#endif // SIMDJSON_DISABLE_DEPRECATED_API
+
+inline std::ostream& operator<<(std::ostream& out, error_code error) noexcept {
+  return out << error_message(error);
+}
+
+namespace internal {
+
+//
+// internal::simdjson_result_base<T> inline implementation
+//
+
+template<typename T>
+simdjson_inline void simdjson_result_base<T>::tie(T &value, error_code &error) && noexcept {
+  error = this->second;
+  if (!error) {
+    value = std::forward<simdjson_result_base<T>>(*this).first;
+  }
+}
+
+template<typename T>
+simdjson_warn_unused simdjson_inline error_code simdjson_result_base<T>::get(T &value) && noexcept {
+  error_code error;
+  std::forward<simdjson_result_base<T>>(*this).tie(value, error);
+  return error;
+}
+
+template<typename T>
+simdjson_inline error_code simdjson_result_base<T>::error() const noexcept {
+  return this->second;
+}
+
+#if SIMDJSON_EXCEPTIONS
+
+template<typename T>
+simdjson_inline T& simdjson_result_base<T>::value() & noexcept(false) {
+  if (error()) { throw simdjson_error(error()); }
+  return this->first;
+}
+
+template<typename T>
+simdjson_inline T&& simdjson_result_base<T>::value() && noexcept(false) {
+  return std::forward<simdjson_result_base<T>>(*this).take_value();
+}
+
+template<typename T>
+simdjson_inline T&& simdjson_result_base<T>::take_value() && noexcept(false) {
+  if (error()) { throw simdjson_error(error()); }
+  return std::forward<T>(this->first);
+}
+
+template<typename T>
+simdjson_inline simdjson_result_base<T>::operator T&&() && noexcept(false) {
+  return std::forward<simdjson_result_base<T>>(*this).take_value();
+}
+
+#endif // SIMDJSON_EXCEPTIONS
+
+template<typename T>
+simdjson_inline const T& simdjson_result_base<T>::value_unsafe() const& noexcept {
+  return this->first;
+}
+
+template<typename T>
+simdjson_inline T&& simdjson_result_base<T>::value_unsafe() && noexcept {
+  return std::forward<T>(this->first);
+}
+
+template<typename T>
+simdjson_inline simdjson_result_base<T>::simdjson_result_base(T &&value, error_code error) noexcept
+    : std::pair<T, error_code>(std::forward<T>(value), error) {}
+template<typename T>
+simdjson_inline simdjson_result_base<T>::simdjson_result_base(error_code error) noexcept
+    : simdjson_result_base(T{}, error) {}
+template<typename T>
+simdjson_inline simdjson_result_base<T>::simdjson_result_base(T &&value) noexcept
+    : simdjson_result_base(std::forward<T>(value), SUCCESS) {}
+template<typename T>
+simdjson_inline simdjson_result_base<T>::simdjson_result_base() noexcept
+    : simdjson_result_base(T{}, UNINITIALIZED) {}
+
+} // namespace internal
+
+///
+/// simdjson_result<T> inline implementation
+///
+
+template<typename T>
+simdjson_inline void simdjson_result<T>::tie(T &value, error_code &error) && noexcept {
+  std::forward<internal::simdjson_result_base<T>>(*this).tie(value, error);
+}
+
+template<typename T>
+simdjson_warn_unused simdjson_inline error_code simdjson_result<T>::get(T &value) && noexcept {
+  return std::forward<internal::simdjson_result_base<T>>(*this).get(value);
+}
+
+template<typename T>
+simdjson_inline error_code simdjson_result<T>::error() const noexcept {
+  return internal::simdjson_result_base<T>::error();
+}
+
+#if SIMDJSON_EXCEPTIONS
+
+template<typename T>
+simdjson_inline T& simdjson_result<T>::value() & noexcept(false) {
+  return internal::simdjson_result_base<T>::value();
+}
+
+template<typename T>
+simdjson_inline T&& simdjson_result<T>::value() && noexcept(false) {
+  return std::forward<internal::simdjson_result_base<T>>(*this).value();
+}
+
+template<typename T>
+simdjson_inline T&& simdjson_result<T>::take_value() && noexcept(false) {
+  return std::forward<internal::simdjson_result_base<T>>(*this).take_value();
+}
+
+template<typename T>
+simdjson_inline simdjson_result<T>::operator T&&() && noexcept(false) {
+  return std::forward<internal::simdjson_result_base<T>>(*this).take_value();
+}
+
+#endif // SIMDJSON_EXCEPTIONS
+
+template<typename T>
+simdjson_inline const T& simdjson_result<T>::value_unsafe() const& noexcept {
+  return internal::simdjson_result_base<T>::value_unsafe();
+}
+
+template<typename T>
+simdjson_inline T&& simdjson_result<T>::value_unsafe() && noexcept {
+  return std::forward<internal::simdjson_result_base<T>>(*this).value_unsafe();
+}
+
+template<typename T>
+simdjson_inline simdjson_result<T>::simdjson_result(T &&value, error_code error) noexcept
+    : internal::simdjson_result_base<T>(std::forward<T>(value), error) {}
+template<typename T>
+simdjson_inline simdjson_result<T>::simdjson_result(error_code error) noexcept
+    : internal::simdjson_result_base<T>(error) {}
+template<typename T>
+simdjson_inline simdjson_result<T>::simdjson_result(T &&value) noexcept
+    : internal::simdjson_result_base<T>(std::forward<T>(value)) {}
+template<typename T>
+simdjson_inline simdjson_result<T>::simdjson_result() noexcept
+    : internal::simdjson_result_base<T>() {}
+
+} // namespace simdjson
+
+#endif // SIMDJSON_ERROR_INL_H
diff --git a/contrib/libs/simdjson/include/simdjson/error.h b/contrib/libs/simdjson/include/simdjson/error.h
new file mode 100644
index 00000000000..4848f6ef912
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/error.h
@@ -0,0 +1,318 @@
+#ifndef SIMDJSON_ERROR_H
+#define SIMDJSON_ERROR_H
+
+#include "simdjson/base.h"
+
+#include <string>
+#include <ostream>
+
+namespace simdjson {
+
+/**
+ * All possible errors returned by simdjson. These error codes are subject to change
+ * and not all simdjson kernel returns the same error code given the same input: it is not
+ * well defined which error a given input should produce.
+ *
+ * Only SUCCESS evaluates to false as a Boolean. All other error codes will evaluate
+ * to true as a Boolean.
+ */
+enum error_code {
+  SUCCESS = 0,                ///< No error
+  CAPACITY,                   ///< This parser can't support a document that big
+  MEMALLOC,                   ///< Error allocating memory, most likely out of memory
+  TAPE_ERROR,                 ///< Something went wrong, this is a generic error
+  DEPTH_ERROR,                ///< Your document exceeds the user-specified depth limitation
+  STRING_ERROR,               ///< Problem while parsing a string
+  T_ATOM_ERROR,               ///< Problem while parsing an atom starting with the letter 't'
+  F_ATOM_ERROR,               ///< Problem while parsing an atom starting with the letter 'f'
+  N_ATOM_ERROR,               ///< Problem while parsing an atom starting with the letter 'n'
+  NUMBER_ERROR,               ///< Problem while parsing a number
+  BIGINT_ERROR,               ///< The integer value exceeds 64 bits
+  UTF8_ERROR,                 ///< the input is not valid UTF-8
+  UNINITIALIZED,              ///< unknown error, or uninitialized document
+  EMPTY,                      ///< no structural element found
+  UNESCAPED_CHARS,            ///< found unescaped characters in a string.
+  UNCLOSED_STRING,            ///< missing quote at the end
+  UNSUPPORTED_ARCHITECTURE,   ///< unsupported architecture
+  INCORRECT_TYPE,             ///< JSON element has a different type than user expected
+  NUMBER_OUT_OF_RANGE,        ///< JSON number does not fit in 64 bits
+  INDEX_OUT_OF_BOUNDS,        ///< JSON array index too large
+  NO_SUCH_FIELD,              ///< JSON field not found in object
+  IO_ERROR,                   ///< Error reading a file
+  INVALID_JSON_POINTER,       ///< Invalid JSON pointer syntax
+  INVALID_URI_FRAGMENT,       ///< Invalid URI fragment
+  UNEXPECTED_ERROR,           ///< indicative of a bug in simdjson
+  PARSER_IN_USE,              ///< parser is already in use.
+  OUT_OF_ORDER_ITERATION,     ///< tried to iterate an array or object out of order (checked when SIMDJSON_DEVELOPMENT_CHECKS=1)
+  INSUFFICIENT_PADDING,       ///< The JSON doesn't have enough padding for simdjson to safely parse it.
+  INCOMPLETE_ARRAY_OR_OBJECT, ///< The document ends early.
+  SCALAR_DOCUMENT_AS_VALUE,   ///< A scalar document is treated as a value.
+  OUT_OF_BOUNDS,              ///< Attempted to access location outside of document.
+  TRAILING_CONTENT,           ///< Unexpected trailing content in the JSON input
+  NUM_ERROR_CODES
+};
+
+/**
+ * It is the convention throughout the code that  the macro SIMDJSON_DEVELOPMENT_CHECKS determines whether
+ * we check for OUT_OF_ORDER_ITERATION. The logic behind it is that these errors only occurs when the code
+ * that was written while breaking some simdjson::ondemand requirement. They should not occur in released
+ * code after these issues were fixed.
+ */
+
+/**
+ * Get the error message for the given error code.
+ *
+ *   dom::parser parser;
+ *   dom::element doc;
+ *   auto error = parser.parse("foo",3).get(doc);
+ *   if (error) { printf("Error: %s\n", error_message(error)); }
+ *
+ * @return The error message.
+ */
+inline const char *error_message(error_code error) noexcept;
+
+/**
+ * Write the error message to the output stream
+ */
+inline std::ostream& operator<<(std::ostream& out, error_code error) noexcept;
+
+/**
+ * Exception thrown when an exception-supporting simdjson method is called
+ */
+struct simdjson_error : public std::exception {
+  /**
+   * Create an exception from a simdjson error code.
+   * @param error The error code
+   */
+  simdjson_error(error_code error) noexcept : _error{error} { }
+  /** The error message */
+  const char *what() const noexcept { return error_message(error()); }
+  /** The error code */
+  error_code error() const noexcept { return _error; }
+private:
+  /** The error code that was used */
+  error_code _error;
+};
+
+namespace internal {
+
+/**
+ * The result of a simdjson operation that could fail.
+ *
+ * Gives the option of reading error codes, or throwing an exception by casting to the desired result.
+ *
+ * This is a base class for implementations that want to add functions to the result type for
+ * chaining.
+ *
+ * Override like:
+ *
+ *   struct simdjson_result<T> : public internal::simdjson_result_base<T> {
+ *     simdjson_result() noexcept : internal::simdjson_result_base<T>() {}
+ *     simdjson_result(error_code error) noexcept : internal::simdjson_result_base<T>(error) {}
+ *     simdjson_result(T &&value) noexcept : internal::simdjson_result_base<T>(std::forward(value)) {}
+ *     simdjson_result(T &&value, error_code error) noexcept : internal::simdjson_result_base<T>(value, error) {}
+ *     // Your extra methods here
+ *   }
+ *
+ * Then any method returning simdjson_result<T> will be chainable with your methods.
+ */
+template<typename T>
+struct simdjson_result_base : protected std::pair<T, error_code> {
+
+  /**
+   * Create a new empty result with error = UNINITIALIZED.
+   */
+  simdjson_inline simdjson_result_base() noexcept;
+
+  /**
+   * Create a new error result.
+   */
+  simdjson_inline simdjson_result_base(error_code error) noexcept;
+
+  /**
+   * Create a new successful result.
+   */
+  simdjson_inline simdjson_result_base(T &&value) noexcept;
+
+  /**
+   * Create a new result with both things (use if you don't want to branch when creating the result).
+   */
+  simdjson_inline simdjson_result_base(T &&value, error_code error) noexcept;
+
+  /**
+   * Move the value and the error to the provided variables.
+   *
+   * @param value The variable to assign the value to. May not be set if there is an error.
+   * @param error The variable to assign the error to. Set to SUCCESS if there is no error.
+   */
+  simdjson_inline void tie(T &value, error_code &error) && noexcept;
+
+  /**
+   * Move the value to the provided variable.
+   *
+   * @param value The variable to assign the value to. May not be set if there is an error.
+   */
+  simdjson_inline error_code get(T &value) && noexcept;
+
+  /**
+   * The error.
+   */
+  simdjson_inline error_code error() const noexcept;
+
+#if SIMDJSON_EXCEPTIONS
+
+  /**
+   * Get the result value.
+   *
+   * @throw simdjson_error if there was an error.
+   */
+  simdjson_inline T& value() & noexcept(false);
+
+  /**
+   * Take the result value (move it).
+   *
+   * @throw simdjson_error if there was an error.
+   */
+  simdjson_inline T&& value() && noexcept(false);
+
+  /**
+   * Take the result value (move it).
+   *
+   * @throw simdjson_error if there was an error.
+   */
+  simdjson_inline T&& take_value() && noexcept(false);
+
+  /**
+   * Cast to the value (will throw on error).
+   *
+   * @throw simdjson_error if there was an error.
+   */
+  simdjson_inline operator T&&() && noexcept(false);
+#endif // SIMDJSON_EXCEPTIONS
+
+  /**
+   * Get the result value. This function is safe if and only
+   * the error() method returns a value that evaluates to false.
+   */
+  simdjson_inline const T& value_unsafe() const& noexcept;
+
+  /**
+   * Take the result value (move it). This function is safe if and only
+   * the error() method returns a value that evaluates to false.
+   */
+  simdjson_inline T&& value_unsafe() && noexcept;
+
+}; // struct simdjson_result_base
+
+} // namespace internal
+
+/**
+ * The result of a simdjson operation that could fail.
+ *
+ * Gives the option of reading error codes, or throwing an exception by casting to the desired result.
+ */
+template<typename T>
+struct simdjson_result : public internal::simdjson_result_base<T> {
+  /**
+   * @private Create a new empty result with error = UNINITIALIZED.
+   */
+  simdjson_inline simdjson_result() noexcept;
+  /**
+   * @private Create a new successful result.
+   */
+  simdjson_inline simdjson_result(T &&value) noexcept;
+  /**
+   * @private Create a new error result.
+   */
+  simdjson_inline simdjson_result(error_code error_code) noexcept;
+  /**
+   * @private Create a new result with both things (use if you don't want to branch when creating the result).
+   */
+  simdjson_inline simdjson_result(T &&value, error_code error) noexcept;
+
+  /**
+   * Move the value and the error to the provided variables.
+   *
+   * @param value The variable to assign the value to. May not be set if there is an error.
+   * @param error The variable to assign the error to. Set to SUCCESS if there is no error.
+   */
+  simdjson_inline void tie(T &value, error_code &error) && noexcept;
+
+  /**
+   * Move the value to the provided variable.
+   *
+   * @param value The variable to assign the value to. May not be set if there is an error.
+   */
+  simdjson_warn_unused simdjson_inline error_code get(T &value) && noexcept;
+
+  /**
+   * The error.
+   */
+  simdjson_inline error_code error() const noexcept;
+
+#if SIMDJSON_EXCEPTIONS
+
+  /**
+   * Get the result value.
+   *
+   * @throw simdjson_error if there was an error.
+   */
+  simdjson_inline T& value() & noexcept(false);
+
+  /**
+   * Take the result value (move it).
+   *
+   * @throw simdjson_error if there was an error.
+   */
+  simdjson_inline T&& value() && noexcept(false);
+
+  /**
+   * Take the result value (move it).
+   *
+   * @throw simdjson_error if there was an error.
+   */
+  simdjson_inline T&& take_value() && noexcept(false);
+
+  /**
+   * Cast to the value (will throw on error).
+   *
+   * @throw simdjson_error if there was an error.
+   */
+  simdjson_inline operator T&&() && noexcept(false);
+#endif // SIMDJSON_EXCEPTIONS
+
+  /**
+   * Get the result value. This function is safe if and only
+   * the error() method returns a value that evaluates to false.
+   */
+  simdjson_inline const T& value_unsafe() const& noexcept;
+
+  /**
+   * Take the result value (move it). This function is safe if and only
+   * the error() method returns a value that evaluates to false.
+   */
+  simdjson_inline T&& value_unsafe() && noexcept;
+
+}; // struct simdjson_result
+
+#if SIMDJSON_EXCEPTIONS
+
+template<typename T>
+inline std::ostream& operator<<(std::ostream& out, simdjson_result<T> value) { return out << value.value(); }
+#endif // SIMDJSON_EXCEPTIONS
+
+#ifndef SIMDJSON_DISABLE_DEPRECATED_API
+/**
+ * @deprecated This is an alias and will be removed, use error_code instead
+ */
+using ErrorValues [[deprecated("This is an alias and will be removed, use error_code instead")]] = error_code;
+
+/**
+ * @deprecated Error codes should be stored and returned as `error_code`, use `error_message()` instead.
+ */
+[[deprecated("Error codes should be stored and returned as `error_code`, use `error_message()` instead.")]]
+inline const std::string error_message(int error) noexcept;
+#endif // SIMDJSON_DISABLE_DEPRECATED_API
+} // namespace simdjson
+
+#endif // SIMDJSON_ERROR_H
diff --git a/contrib/libs/simdjson/include/simdjson/fallback.h b/contrib/libs/simdjson/include/simdjson/fallback.h
new file mode 100644
index 00000000000..4588cdc00f0
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/fallback.h
@@ -0,0 +1,8 @@
+#ifndef SIMDJSON_FALLBACK_H
+#define SIMDJSON_FALLBACK_H
+
+#include "simdjson/fallback/begin.h"
+#include "simdjson/generic/amalgamated.h"
+#include "simdjson/fallback/end.h"
+
+#endif // SIMDJSON_FALLBACK_H
\ No newline at end of file
diff --git a/contrib/libs/simdjson/include/simdjson/fallback/base.h b/contrib/libs/simdjson/include/simdjson/fallback/base.h
new file mode 100644
index 00000000000..99cb37423d7
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/fallback/base.h
@@ -0,0 +1,19 @@
+#ifndef SIMDJSON_FALLBACK_BASE_H
+#define SIMDJSON_FALLBACK_BASE_H
+
+#ifndef SIMDJSON_CONDITIONAL_INCLUDE
+#include "simdjson/base.h"
+#endif // SIMDJSON_CONDITIONAL_INCLUDE
+
+namespace simdjson {
+/**
+ * Fallback implementation (runs on any machine).
+ */
+namespace fallback {
+
+class implementation;
+
+} // namespace fallback
+} // namespace simdjson
+
+#endif // SIMDJSON_FALLBACK_BASE_H
diff --git a/contrib/libs/simdjson/include/simdjson/fallback/begin.h b/contrib/libs/simdjson/include/simdjson/fallback/begin.h
new file mode 100644
index 00000000000..74749f6d456
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/fallback/begin.h
@@ -0,0 +1,5 @@
+#define SIMDJSON_IMPLEMENTATION fallback
+#include "simdjson/fallback/base.h"
+#include "simdjson/fallback/bitmanipulation.h"
+#include "simdjson/fallback/stringparsing_defs.h"
+#include "simdjson/fallback/numberparsing_defs.h"
diff --git a/contrib/libs/simdjson/include/simdjson/fallback/bitmanipulation.h b/contrib/libs/simdjson/include/simdjson/fallback/bitmanipulation.h
new file mode 100644
index 00000000000..ba47dcccfbf
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/fallback/bitmanipulation.h
@@ -0,0 +1,48 @@
+#ifndef SIMDJSON_FALLBACK_BITMANIPULATION_H
+#define SIMDJSON_FALLBACK_BITMANIPULATION_H
+
+#ifndef SIMDJSON_CONDITIONAL_INCLUDE
+#include "simdjson/fallback/base.h"
+#endif // SIMDJSON_CONDITIONAL_INCLUDE
+
+namespace simdjson {
+namespace fallback {
+namespace {
+
+#if defined(_MSC_VER) && !defined(_M_ARM64) && !defined(_M_X64)
+static inline unsigned char _BitScanForward64(unsigned long* ret, uint64_t x) {
+  unsigned long x0 = (unsigned long)x, top, bottom;
+  _BitScanForward(&top, (unsigned long)(x >> 32));
+  _BitScanForward(&bottom, x0);
+  *ret = x0 ? bottom : 32 + top;
+  return x != 0;
+}
+static unsigned char _BitScanReverse64(unsigned long* ret, uint64_t x) {
+  unsigned long x1 = (unsigned long)(x >> 32), top, bottom;
+  _BitScanReverse(&top, x1);
+  _BitScanReverse(&bottom, (unsigned long)x);
+  *ret = x1 ? top + 32 : bottom;
+  return x != 0;
+}
+#endif
+
+/* result might be undefined when input_num is zero */
+simdjson_inline int leading_zeroes(uint64_t input_num) {
+#ifdef _MSC_VER
+  unsigned long leading_zero = 0;
+  // Search the mask data from most significant bit (MSB)
+  // to least significant bit (LSB) for a set bit (1).
+  if (_BitScanReverse64(&leading_zero, input_num))
+    return (int)(63 - leading_zero);
+  else
+    return 64;
+#else
+  return __builtin_clzll(input_num);
+#endif// _MSC_VER
+}
+
+} // unnamed namespace
+} // namespace fallback
+} // namespace simdjson
+
+#endif // SIMDJSON_FALLBACK_BITMANIPULATION_H
diff --git a/contrib/libs/simdjson/include/simdjson/fallback/end.h b/contrib/libs/simdjson/include/simdjson/fallback/end.h
new file mode 100644
index 00000000000..fbd14132b13
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/fallback/end.h
@@ -0,0 +1,5 @@
+#ifndef SIMDJSON_CONDITIONAL_INCLUDE
+#include "simdjson/fallback/base.h"
+#endif // SIMDJSON_CONDITIONAL_INCLUDE
+
+#undef SIMDJSON_IMPLEMENTATION
diff --git a/contrib/libs/simdjson/include/simdjson/fallback/implementation.h b/contrib/libs/simdjson/include/simdjson/fallback/implementation.h
new file mode 100644
index 00000000000..523f06d2e92
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/fallback/implementation.h
@@ -0,0 +1,34 @@
+#ifndef SIMDJSON_FALLBACK_IMPLEMENTATION_H
+#define SIMDJSON_FALLBACK_IMPLEMENTATION_H
+
+#ifndef SIMDJSON_CONDITIONAL_INCLUDE
+#include "simdjson/fallback/base.h"
+#include "simdjson/implementation.h"
+#endif // SIMDJSON_CONDITIONAL_INCLUDE
+
+namespace simdjson {
+namespace fallback {
+
+/**
+ * @private
+ */
+class implementation final : public simdjson::implementation {
+public:
+  simdjson_inline implementation() : simdjson::implementation(
+      "fallback",
+      "Generic fallback implementation",
+      0
+  ) {}
+  simdjson_warn_unused error_code create_dom_parser_implementation(
+    size_t capacity,
+    size_t max_length,
+    std::unique_ptr<simdjson::internal::dom_parser_implementation>& dst
+  ) const noexcept final;
+  simdjson_warn_unused error_code minify(const uint8_t *buf, size_t len, uint8_t *dst, size_t &dst_len) const noexcept final;
+  simdjson_warn_unused bool validate_utf8(const char *buf, size_t len) const noexcept final;
+};
+
+} // namespace fallback
+} // namespace simdjson
+
+#endif // SIMDJSON_FALLBACK_IMPLEMENTATION_H
diff --git a/contrib/libs/simdjson/include/simdjson/fallback/numberparsing_defs.h b/contrib/libs/simdjson/include/simdjson/fallback/numberparsing_defs.h
new file mode 100644
index 00000000000..e807423e6f8
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/fallback/numberparsing_defs.h
@@ -0,0 +1,80 @@
+#ifndef SIMDJSON_FALLBACK_NUMBERPARSING_DEFS_H
+#define SIMDJSON_FALLBACK_NUMBERPARSING_DEFS_H
+
+#ifndef SIMDJSON_CONDITIONAL_INCLUDE
+#include "simdjson/fallback/base.h"
+#include "simdjson/internal/numberparsing_tables.h"
+#endif // SIMDJSON_CONDITIONAL_INCLUDE
+
+#include <cstring>
+
+#ifdef JSON_TEST_NUMBERS // for unit testing
+void found_invalid_number(const uint8_t *buf);
+void found_integer(int64_t result, const uint8_t *buf);
+void found_unsigned_integer(uint64_t result, const uint8_t *buf);
+void found_float(double result, const uint8_t *buf);
+#endif
+
+namespace simdjson {
+namespace fallback {
+namespace numberparsing {
+
+// credit: https://johnnylee-sde.github.io/Fast-numeric-string-to-int/
+/** @private */
+static simdjson_inline uint32_t parse_eight_digits_unrolled(const char *chars) {
+  uint64_t val;
+  memcpy(&val, chars, sizeof(uint64_t));
+  val = (val & 0x0F0F0F0F0F0F0F0F) * 2561 >> 8;
+  val = (val & 0x00FF00FF00FF00FF) * 6553601 >> 16;
+  return uint32_t((val & 0x0000FFFF0000FFFF) * 42949672960001 >> 32);
+}
+
+/** @private */
+static simdjson_inline uint32_t parse_eight_digits_unrolled(const uint8_t *chars) {
+  return parse_eight_digits_unrolled(reinterpret_cast<const char *>(chars));
+}
+
+#if SIMDJSON_IS_32BITS // _umul128 for x86, arm
+// this is a slow emulation routine for 32-bit
+//
+static simdjson_inline uint64_t __emulu(uint32_t x, uint32_t y) {
+  return x * (uint64_t)y;
+}
+static simdjson_inline uint64_t _umul128(uint64_t ab, uint64_t cd, uint64_t *hi) {
+  uint64_t ad = __emulu((uint32_t)(ab >> 32), (uint32_t)cd);
+  uint64_t bd = __emulu((uint32_t)ab, (uint32_t)cd);
+  uint64_t adbc = ad + __emulu((uint32_t)ab, (uint32_t)(cd >> 32));
+  uint64_t adbc_carry = !!(adbc < ad);
+  uint64_t lo = bd + (adbc << 32);
+  *hi = __emulu((uint32_t)(ab >> 32), (uint32_t)(cd >> 32)) + (adbc >> 32) +
+        (adbc_carry << 32) + !!(lo < bd);
+  return lo;
+}
+#endif
+
+/** @private */
+simdjson_inline internal::value128 full_multiplication(uint64_t value1, uint64_t value2) {
+  internal::value128 answer;
+#if SIMDJSON_REGULAR_VISUAL_STUDIO || SIMDJSON_IS_32BITS
+#if SIMDJSON_IS_ARM64
+  // ARM64 has native support for 64-bit multiplications, no need to emultate
+  answer.high = __umulh(value1, value2);
+  answer.low = value1 * value2;
+#else
+  answer.low = _umul128(value1, value2, &answer.high); // _umul128 not available on ARM64
+#endif // SIMDJSON_IS_ARM64
+#else // SIMDJSON_REGULAR_VISUAL_STUDIO || SIMDJSON_IS_32BITS
+  __uint128_t r = (static_cast<__uint128_t>(value1)) * value2;
+  answer.low = uint64_t(r);
+  answer.high = uint64_t(r >> 64);
+#endif
+  return answer;
+}
+
+} // namespace numberparsing
+} // namespace fallback
+} // namespace simdjson
+
+#define SIMDJSON_SWAR_NUMBER_PARSING 1
+
+#endif // SIMDJSON_FALLBACK_NUMBERPARSING_DEFS_H
diff --git a/contrib/libs/simdjson/include/simdjson/fallback/stringparsing_defs.h b/contrib/libs/simdjson/include/simdjson/fallback/stringparsing_defs.h
new file mode 100644
index 00000000000..64f23c4b03d
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/fallback/stringparsing_defs.h
@@ -0,0 +1,36 @@
+#ifndef SIMDJSON_FALLBACK_STRINGPARSING_DEFS_H
+#define SIMDJSON_FALLBACK_STRINGPARSING_DEFS_H
+
+#ifndef SIMDJSON_CONDITIONAL_INCLUDE
+#include "simdjson/fallback/base.h"
+#endif // SIMDJSON_CONDITIONAL_INCLUDE
+
+namespace simdjson {
+namespace fallback {
+namespace {
+
+// Holds backslashes and quotes locations.
+struct backslash_and_quote {
+public:
+  static constexpr uint32_t BYTES_PROCESSED = 1;
+  simdjson_inline static backslash_and_quote copy_and_find(const uint8_t *src, uint8_t *dst);
+
+  simdjson_inline bool has_quote_first() { return c == '"'; }
+  simdjson_inline bool has_backslash() { return c == '\\'; }
+  simdjson_inline int quote_index() { return c == '"' ? 0 : 1; }
+  simdjson_inline int backslash_index() { return c == '\\' ? 0 : 1; }
+
+  uint8_t c;
+}; // struct backslash_and_quote
+
+simdjson_inline backslash_and_quote backslash_and_quote::copy_and_find(const uint8_t *src, uint8_t *dst) {
+  // store to dest unconditionally - we can overwrite the bits we don't like later
+  dst[0] = src[0];
+  return { src[0] };
+}
+
+} // unnamed namespace
+} // namespace fallback
+} // namespace simdjson
+
+#endif // SIMDJSON_FALLBACK_STRINGPARSING_DEFS_H
diff --git a/contrib/libs/simdjson/include/simdjson/generic/amalgamated.h b/contrib/libs/simdjson/include/simdjson/generic/amalgamated.h
new file mode 100644
index 00000000000..4b235fa9f7b
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/generic/amalgamated.h
@@ -0,0 +1,12 @@
+#if defined(SIMDJSON_CONDITIONAL_INCLUDE) && !defined(SIMDJSON_GENERIC_DEPENDENCIES_H)
+#error simdjson/generic/dependencies.h must be included before simdjson/generic/amalgamated.h!
+#endif
+
+#include "simdjson/generic/base.h"
+#include "simdjson/generic/jsoncharutils.h"
+#include "simdjson/generic/atomparsing.h"
+#include "simdjson/generic/dom_parser_implementation.h"
+#include "simdjson/generic/implementation_simdjson_result_base.h"
+#include "simdjson/generic/numberparsing.h"
+
+#include "simdjson/generic/implementation_simdjson_result_base-inl.h"
diff --git a/contrib/libs/simdjson/include/simdjson/generic/atomparsing.h b/contrib/libs/simdjson/include/simdjson/generic/atomparsing.h
new file mode 100644
index 00000000000..2eeb7b81a26
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/generic/atomparsing.h
@@ -0,0 +1,77 @@
+#ifndef SIMDJSON_GENERIC_ATOMPARSING_H
+
+#ifndef SIMDJSON_CONDITIONAL_INCLUDE
+#define SIMDJSON_GENERIC_ATOMPARSING_H
+#include "simdjson/generic/base.h"
+#include "simdjson/generic/jsoncharutils.h"
+#endif // SIMDJSON_CONDITIONAL_INCLUDE
+
+#include <cstring>
+
+namespace simdjson {
+namespace SIMDJSON_IMPLEMENTATION {
+namespace {
+/// @private
+namespace atomparsing {
+
+// The string_to_uint32 is exclusively used to map literal strings to 32-bit values.
+// We use memcpy instead of a pointer cast to avoid undefined behaviors since we cannot
+// be certain that the character pointer will be properly aligned.
+// You might think that using memcpy makes this function expensive, but you'd be wrong.
+// All decent optimizing compilers (GCC, clang, Visual Studio) will compile string_to_uint32("false");
+// to the compile-time constant 1936482662.
+simdjson_inline uint32_t string_to_uint32(const char* str) { uint32_t val; std::memcpy(&val, str, sizeof(uint32_t)); return val; }
+
+
+// Again in str4ncmp we use a memcpy to avoid undefined behavior. The memcpy may appear expensive.
+// Yet all decent optimizing compilers will compile memcpy to a single instruction, just about.
+simdjson_warn_unused
+simdjson_inline uint32_t str4ncmp(const uint8_t *src, const char* atom) {
+  uint32_t srcval; // we want to avoid unaligned 32-bit loads (undefined in C/C++)
+  static_assert(sizeof(uint32_t) <= SIMDJSON_PADDING, "SIMDJSON_PADDING must be larger than 4 bytes");
+  std::memcpy(&srcval, src, sizeof(uint32_t));
+  return srcval ^ string_to_uint32(atom);
+}
+
+simdjson_warn_unused
+simdjson_inline bool is_valid_true_atom(const uint8_t *src) {
+  return (str4ncmp(src, "true") | jsoncharutils::is_not_structural_or_whitespace(src[4])) == 0;
+}
+
+simdjson_warn_unused
+simdjson_inline bool is_valid_true_atom(const uint8_t *src, size_t len) {
+  if (len > 4) { return is_valid_true_atom(src); }
+  else if (len == 4) { return !str4ncmp(src, "true"); }
+  else { return false; }
+}
+
+simdjson_warn_unused
+simdjson_inline bool is_valid_false_atom(const uint8_t *src) {
+  return (str4ncmp(src+1, "alse") | jsoncharutils::is_not_structural_or_whitespace(src[5])) == 0;
+}
+
+simdjson_warn_unused
+simdjson_inline bool is_valid_false_atom(const uint8_t *src, size_t len) {
+  if (len > 5) { return is_valid_false_atom(src); }
+  else if (len == 5) { return !str4ncmp(src+1, "alse"); }
+  else { return false; }
+}
+
+simdjson_warn_unused
+simdjson_inline bool is_valid_null_atom(const uint8_t *src) {
+  return (str4ncmp(src, "null") | jsoncharutils::is_not_structural_or_whitespace(src[4])) == 0;
+}
+
+simdjson_warn_unused
+simdjson_inline bool is_valid_null_atom(const uint8_t *src, size_t len) {
+  if (len > 4) { return is_valid_null_atom(src); }
+  else if (len == 4) { return !str4ncmp(src, "null"); }
+  else { return false; }
+}
+
+} // namespace atomparsing
+} // unnamed namespace
+} // namespace SIMDJSON_IMPLEMENTATION
+} // namespace simdjson
+
+#endif // SIMDJSON_GENERIC_ATOMPARSING_H
\ No newline at end of file
diff --git a/contrib/libs/simdjson/include/simdjson/generic/base.h b/contrib/libs/simdjson/include/simdjson/generic/base.h
new file mode 100644
index 00000000000..65180837229
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/generic/base.h
@@ -0,0 +1,51 @@
+#ifndef SIMDJSON_GENERIC_BASE_H
+
+#ifndef SIMDJSON_CONDITIONAL_INCLUDE
+#define SIMDJSON_GENERIC_BASE_H
+#include "simdjson/base.h"
+// If we haven't got an implementation yet, we're in the editor, editing a generic file! Just
+// use the most advanced one we can so the most possible stuff can be tested.
+#ifndef SIMDJSON_IMPLEMENTATION
+#include "simdjson/implementation_detection.h"
+#if SIMDJSON_IMPLEMENTATION_ICELAKE
+#include "simdjson/icelake/begin.h"
+#elif SIMDJSON_IMPLEMENTATION_HASWELL
+#include "simdjson/haswell/begin.h"
+#elif SIMDJSON_IMPLEMENTATION_WESTMERE
+#include "simdjson/westmere/begin.h"
+#elif SIMDJSON_IMPLEMENTATION_ARM64
+#include "simdjson/arm64/begin.h"
+#elif SIMDJSON_IMPLEMENTATION_PPC64
+#error #include "simdjson/ppc64/begin.h"
+#elif SIMDJSON_IMPLEMENTATION_LSX
+#include "simdjson/lsx/begin.h"
+#elif SIMDJSON_IMPLEMENTATION_LASX
+#include "simdjson/lasx/begin.h"
+#elif SIMDJSON_IMPLEMENTATION_FALLBACK
+#include "simdjson/fallback/begin.h"
+#else
+#error "All possible implementations (including fallback) have been disabled! simdjson will not run."
+#endif
+#endif // SIMDJSON_IMPLEMENTATION
+#endif // SIMDJSON_CONDITIONAL_INCLUDE
+
+namespace simdjson {
+namespace SIMDJSON_IMPLEMENTATION {
+
+struct open_container;
+class dom_parser_implementation;
+
+/**
+ * The type of a JSON number
+ */
+enum class number_type {
+    floating_point_number=1, /// a binary64 number
+    signed_integer,          /// a signed integer that fits in a 64-bit word using two's complement
+    unsigned_integer,        /// a positive integer larger or equal to 1<<63
+    big_integer              /// a big integer that does not fit in a 64-bit word
+};
+
+} // namespace SIMDJSON_IMPLEMENTATION
+} // namespace simdjson
+
+#endif // SIMDJSON_GENERIC_BASE_H
\ No newline at end of file
diff --git a/contrib/libs/simdjson/include/simdjson/generic/dependencies.h b/contrib/libs/simdjson/include/simdjson/generic/dependencies.h
new file mode 100644
index 00000000000..28d9749f276
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/generic/dependencies.h
@@ -0,0 +1,20 @@
+#ifdef SIMDJSON_CONDITIONAL_INCLUDE
+#error simdjson/generic/dependencies.h must be included before defining SIMDJSON_CONDITIONAL_INCLUDE!
+#endif
+
+#ifndef SIMDJSON_GENERIC_DEPENDENCIES_H
+#define SIMDJSON_GENERIC_DEPENDENCIES_H
+
+// Internal headers needed for generics.
+// All includes referencing simdjson headers *not* under simdjson/generic must be here!
+// Otherwise, amalgamation will fail.
+#include "simdjson/base.h"
+#include "simdjson/implementation.h"
+#include "simdjson/implementation_detection.h"
+#include "simdjson/internal/instruction_set.h"
+#include "simdjson/internal/dom_parser_implementation.h"
+#include "simdjson/internal/jsoncharutils_tables.h"
+#include "simdjson/internal/numberparsing_tables.h"
+#include "simdjson/internal/simdprune_tables.h"
+
+#endif // SIMDJSON_GENERIC_DEPENDENCIES_H
\ No newline at end of file
diff --git a/contrib/libs/simdjson/include/simdjson/generic/dom_parser_implementation.h b/contrib/libs/simdjson/include/simdjson/generic/dom_parser_implementation.h
new file mode 100644
index 00000000000..e51d2c1279b
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/generic/dom_parser_implementation.h
@@ -0,0 +1,89 @@
+#ifndef SIMDJSON_GENERIC_DOM_PARSER_IMPLEMENTATION_H
+
+#ifndef SIMDJSON_CONDITIONAL_INCLUDE
+#define SIMDJSON_GENERIC_DOM_PARSER_IMPLEMENTATION_H
+#include "simdjson/generic/base.h"
+#include "simdjson/internal/dom_parser_implementation.h"
+#endif // SIMDJSON_CONDITIONAL_INCLUDE
+
+namespace simdjson {
+namespace SIMDJSON_IMPLEMENTATION {
+
+// expectation: sizeof(open_container) = 64/8.
+struct open_container {
+  uint32_t tape_index; // where, on the tape, does the scope ([,{) begins
+  uint32_t count; // how many elements in the scope
+}; // struct open_container
+
+static_assert(sizeof(open_container) == 64/8, "Open container must be 64 bits");
+
+class dom_parser_implementation final : public internal::dom_parser_implementation {
+public:
+  /** Tape location of each open { or [ */
+  std::unique_ptr<open_container[]> open_containers{};
+  /** Whether each open container is a [ or { */
+  std::unique_ptr<bool[]> is_array{};
+  /** Buffer passed to stage 1 */
+  const uint8_t *buf{};
+  /** Length passed to stage 1 */
+  size_t len{0};
+  /** Document passed to stage 2 */
+  dom::document *doc{};
+
+  inline dom_parser_implementation() noexcept;
+  inline dom_parser_implementation(dom_parser_implementation &&other) noexcept;
+  inline dom_parser_implementation &operator=(dom_parser_implementation &&other) noexcept;
+  dom_parser_implementation(const dom_parser_implementation &) = delete;
+  dom_parser_implementation &operator=(const dom_parser_implementation &) = delete;
+
+  simdjson_warn_unused error_code parse(const uint8_t *buf, size_t len, dom::document &doc) noexcept final;
+  simdjson_warn_unused error_code stage1(const uint8_t *buf, size_t len, stage1_mode partial) noexcept final;
+  simdjson_warn_unused error_code stage2(dom::document &doc) noexcept final;
+  simdjson_warn_unused error_code stage2_next(dom::document &doc) noexcept final;
+  simdjson_warn_unused uint8_t *parse_string(const uint8_t *src, uint8_t *dst, bool allow_replacement) const noexcept final;
+  simdjson_warn_unused uint8_t *parse_wobbly_string(const uint8_t *src, uint8_t *dst) const noexcept final;
+  inline simdjson_warn_unused error_code set_capacity(size_t capacity) noexcept final;
+  inline simdjson_warn_unused error_code set_max_depth(size_t max_depth) noexcept final;
+private:
+  simdjson_inline simdjson_warn_unused error_code set_capacity_stage1(size_t capacity);
+
+};
+
+} // namespace SIMDJSON_IMPLEMENTATION
+} // namespace simdjson
+
+namespace simdjson {
+namespace SIMDJSON_IMPLEMENTATION {
+
+inline dom_parser_implementation::dom_parser_implementation() noexcept = default;
+inline dom_parser_implementation::dom_parser_implementation(dom_parser_implementation &&other) noexcept = default;
+inline dom_parser_implementation &dom_parser_implementation::operator=(dom_parser_implementation &&other) noexcept = default;
+
+// Leaving these here so they can be inlined if so desired
+inline simdjson_warn_unused error_code dom_parser_implementation::set_capacity(size_t capacity) noexcept {
+  if(capacity > SIMDJSON_MAXSIZE_BYTES) { return CAPACITY; }
+  // Stage 1 index output
+  size_t max_structures = SIMDJSON_ROUNDUP_N(capacity, 64) + 2 + 7;
+  structural_indexes.reset( new (std::nothrow) uint32_t[max_structures] );
+  if (!structural_indexes) { _capacity = 0; return MEMALLOC; }
+  structural_indexes[0] = 0;
+  n_structural_indexes = 0;
+
+  _capacity = capacity;
+  return SUCCESS;
+}
+
+inline simdjson_warn_unused error_code dom_parser_implementation::set_max_depth(size_t max_depth) noexcept {
+  // Stage 2 stacks
+  open_containers.reset(new (std::nothrow) open_container[max_depth]);
+  is_array.reset(new (std::nothrow) bool[max_depth]);
+  if (!is_array || !open_containers) { _max_depth = 0; return MEMALLOC; }
+
+  _max_depth = max_depth;
+  return SUCCESS;
+}
+
+} // namespace SIMDJSON_IMPLEMENTATION
+} // namespace simdjson
+
+#endif // SIMDJSON_GENERIC_DOM_PARSER_IMPLEMENTATION_H
\ No newline at end of file
diff --git a/contrib/libs/simdjson/include/simdjson/generic/implementation_simdjson_result_base-inl.h b/contrib/libs/simdjson/include/simdjson/generic/implementation_simdjson_result_base-inl.h
new file mode 100644
index 00000000000..4990ad7ea15
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/generic/implementation_simdjson_result_base-inl.h
@@ -0,0 +1,90 @@
+#ifndef SIMDJSON_GENERIC_IMPLEMENTATION_SIMDJSON_RESULT_BASE_INL_H
+
+#ifndef SIMDJSON_CONDITIONAL_INCLUDE
+#define SIMDJSON_GENERIC_IMPLEMENTATION_SIMDJSON_RESULT_BASE_INL_H
+#include "simdjson/generic/base.h"
+#include "simdjson/generic/implementation_simdjson_result_base.h"
+#endif // SIMDJSON_CONDITIONAL_INCLUDE
+
+namespace simdjson {
+namespace SIMDJSON_IMPLEMENTATION {
+
+//
+// internal::implementation_simdjson_result_base<T> inline implementation
+//
+
+template<typename T>
+simdjson_inline void implementation_simdjson_result_base<T>::tie(T &value, error_code &error) && noexcept {
+  error = this->second;
+  if (!error) {
+    value = std::forward<implementation_simdjson_result_base<T>>(*this).first;
+  }
+}
+
+template<typename T>
+simdjson_warn_unused simdjson_inline error_code implementation_simdjson_result_base<T>::get(T &value) && noexcept {
+  error_code error;
+  std::forward<implementation_simdjson_result_base<T>>(*this).tie(value, error);
+  return error;
+}
+
+template<typename T>
+simdjson_inline error_code implementation_simdjson_result_base<T>::error() const noexcept {
+  return this->second;
+}
+
+#if SIMDJSON_EXCEPTIONS
+
+template<typename T>
+simdjson_inline T& implementation_simdjson_result_base<T>::value() & noexcept(false) {
+  if (error()) { throw simdjson_error(error()); }
+  return this->first;
+}
+
+template<typename T>
+simdjson_inline T&& implementation_simdjson_result_base<T>::value() && noexcept(false) {
+  return std::forward<implementation_simdjson_result_base<T>>(*this).take_value();
+}
+
+template<typename T>
+simdjson_inline T&& implementation_simdjson_result_base<T>::take_value() && noexcept(false) {
+  if (error()) { throw simdjson_error(error()); }
+  return std::forward<T>(this->first);
+}
+
+template<typename T>
+simdjson_inline implementation_simdjson_result_base<T>::operator T&&() && noexcept(false) {
+  return std::forward<implementation_simdjson_result_base<T>>(*this).take_value();
+}
+
+#endif // SIMDJSON_EXCEPTIONS
+
+template<typename T>
+simdjson_inline const T& implementation_simdjson_result_base<T>::value_unsafe() const& noexcept {
+  return this->first;
+}
+
+template<typename T>
+simdjson_inline T& implementation_simdjson_result_base<T>::value_unsafe() & noexcept {
+  return this->first;
+}
+
+template<typename T>
+simdjson_inline T&& implementation_simdjson_result_base<T>::value_unsafe() && noexcept {
+  return std::forward<T>(this->first);
+}
+
+template<typename T>
+simdjson_inline implementation_simdjson_result_base<T>::implementation_simdjson_result_base(T &&value, error_code error) noexcept
+    : first{std::forward<T>(value)}, second{error} {}
+template<typename T>
+simdjson_inline implementation_simdjson_result_base<T>::implementation_simdjson_result_base(error_code error) noexcept
+    : implementation_simdjson_result_base(T{}, error) {}
+template<typename T>
+simdjson_inline implementation_simdjson_result_base<T>::implementation_simdjson_result_base(T &&value) noexcept
+    : implementation_simdjson_result_base(std::forward<T>(value), SUCCESS) {}
+
+} // namespace SIMDJSON_IMPLEMENTATION
+} // namespace simdjson
+
+#endif // SIMDJSON_GENERIC_IMPLEMENTATION_SIMDJSON_RESULT_BASE_INL_H
\ No newline at end of file
diff --git a/contrib/libs/simdjson/include/simdjson/generic/implementation_simdjson_result_base.h b/contrib/libs/simdjson/include/simdjson/generic/implementation_simdjson_result_base.h
new file mode 100644
index 00000000000..aaf2bce1267
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/generic/implementation_simdjson_result_base.h
@@ -0,0 +1,134 @@
+#ifndef SIMDJSON_GENERIC_IMPLEMENTATION_SIMDJSON_RESULT_BASE_H
+
+#ifndef SIMDJSON_CONDITIONAL_INCLUDE
+#define SIMDJSON_GENERIC_IMPLEMENTATION_SIMDJSON_RESULT_BASE_H
+#include "simdjson/generic/base.h"
+#endif // SIMDJSON_CONDITIONAL_INCLUDE
+
+namespace simdjson {
+namespace SIMDJSON_IMPLEMENTATION {
+
+// This is a near copy of include/error.h's implementation_simdjson_result_base, except it doesn't use std::pair
+// so we can avoid inlining errors
+// TODO reconcile these!
+/**
+ * The result of a simdjson operation that could fail.
+ *
+ * Gives the option of reading error codes, or throwing an exception by casting to the desired result.
+ *
+ * This is a base class for implementations that want to add functions to the result type for
+ * chaining.
+ *
+ * Override like:
+ *
+ *   struct simdjson_result<T> : public internal::implementation_simdjson_result_base<T> {
+ *     simdjson_result() noexcept : internal::implementation_simdjson_result_base<T>() {}
+ *     simdjson_result(error_code error) noexcept : internal::implementation_simdjson_result_base<T>(error) {}
+ *     simdjson_result(T &&value) noexcept : internal::implementation_simdjson_result_base<T>(std::forward(value)) {}
+ *     simdjson_result(T &&value, error_code error) noexcept : internal::implementation_simdjson_result_base<T>(value, error) {}
+ *     // Your extra methods here
+ *   }
+ *
+ * Then any method returning simdjson_result<T> will be chainable with your methods.
+ */
+template<typename T>
+struct implementation_simdjson_result_base {
+
+  /**
+   * Create a new empty result with error = UNINITIALIZED.
+   */
+  simdjson_inline implementation_simdjson_result_base() noexcept = default;
+
+  /**
+   * Create a new error result.
+   */
+  simdjson_inline implementation_simdjson_result_base(error_code error) noexcept;
+
+  /**
+   * Create a new successful result.
+   */
+  simdjson_inline implementation_simdjson_result_base(T &&value) noexcept;
+
+  /**
+   * Create a new result with both things (use if you don't want to branch when creating the result).
+   */
+  simdjson_inline implementation_simdjson_result_base(T &&value, error_code error) noexcept;
+
+  /**
+   * Move the value and the error to the provided variables.
+   *
+   * @param value The variable to assign the value to. May not be set if there is an error.
+   * @param error The variable to assign the error to. Set to SUCCESS if there is no error.
+   */
+  simdjson_inline void tie(T &value, error_code &error) && noexcept;
+
+  /**
+   * Move the value to the provided variable.
+   *
+   * @param value The variable to assign the value to. May not be set if there is an error.
+   */
+  simdjson_inline error_code get(T &value) && noexcept;
+
+  /**
+   * The error.
+   */
+  simdjson_inline error_code error() const noexcept;
+
+#if SIMDJSON_EXCEPTIONS
+
+  /**
+   * Get the result value.
+   *
+   * @throw simdjson_error if there was an error.
+   */
+  simdjson_inline T& value() & noexcept(false);
+
+  /**
+   * Take the result value (move it).
+   *
+   * @throw simdjson_error if there was an error.
+   */
+  simdjson_inline T&& value() && noexcept(false);
+
+  /**
+   * Take the result value (move it).
+   *
+   * @throw simdjson_error if there was an error.
+   */
+  simdjson_inline T&& take_value() && noexcept(false);
+
+  /**
+   * Cast to the value (will throw on error).
+   *
+   * @throw simdjson_error if there was an error.
+   */
+  simdjson_inline operator T&&() && noexcept(false);
+
+
+#endif // SIMDJSON_EXCEPTIONS
+
+  /**
+   * Get the result value. This function is safe if and only
+   * the error() method returns a value that evaluates to false.
+   */
+  simdjson_inline const T& value_unsafe() const& noexcept;
+  /**
+   * Get the result value. This function is safe if and only
+   * the error() method returns a value that evaluates to false.
+   */
+  simdjson_inline T& value_unsafe() & noexcept;
+  /**
+   * Take the result value (move it). This function is safe if and only
+   * the error() method returns a value that evaluates to false.
+   */
+  simdjson_inline T&& value_unsafe() && noexcept;
+protected:
+  /** users should never directly access first and second. **/
+  T first{}; /** Users should never directly access 'first'. **/
+  error_code second{UNINITIALIZED}; /** Users should never directly access 'second'. **/
+}; // struct implementation_simdjson_result_base
+
+} // namespace SIMDJSON_IMPLEMENTATION
+} // namespace simdjson
+
+#endif // SIMDJSON_GENERIC_IMPLEMENTATION_SIMDJSON_RESULT_BASE_H
\ No newline at end of file
diff --git a/contrib/libs/simdjson/include/simdjson/generic/jsoncharutils.h b/contrib/libs/simdjson/include/simdjson/generic/jsoncharutils.h
new file mode 100644
index 00000000000..a79b1f8507d
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/generic/jsoncharutils.h
@@ -0,0 +1,104 @@
+#ifndef SIMDJSON_GENERIC_JSONCHARUTILS_H
+
+#ifndef SIMDJSON_CONDITIONAL_INCLUDE
+#define SIMDJSON_GENERIC_JSONCHARUTILS_H
+#include "simdjson/generic/base.h"
+#include "simdjson/internal/jsoncharutils_tables.h"
+#include "simdjson/internal/numberparsing_tables.h"
+#endif // SIMDJSON_CONDITIONAL_INCLUDE
+
+namespace simdjson {
+namespace SIMDJSON_IMPLEMENTATION {
+namespace {
+namespace jsoncharutils {
+
+// return non-zero if not a structural or whitespace char
+// zero otherwise
+simdjson_inline uint32_t is_not_structural_or_whitespace(uint8_t c) {
+  return internal::structural_or_whitespace_negated[c];
+}
+
+simdjson_inline uint32_t is_structural_or_whitespace(uint8_t c) {
+  return internal::structural_or_whitespace[c];
+}
+
+// returns a value with the high 16 bits set if not valid
+// otherwise returns the conversion of the 4 hex digits at src into the bottom
+// 16 bits of the 32-bit return register
+//
+// see
+// https://lemire.me/blog/2019/04/17/parsing-short-hexadecimal-strings-efficiently/
+static inline uint32_t hex_to_u32_nocheck(
+    const uint8_t *src) { // strictly speaking, static inline is a C-ism
+  uint32_t v1 = internal::digit_to_val32[630 + src[0]];
+  uint32_t v2 = internal::digit_to_val32[420 + src[1]];
+  uint32_t v3 = internal::digit_to_val32[210 + src[2]];
+  uint32_t v4 = internal::digit_to_val32[0 + src[3]];
+  return v1 | v2 | v3 | v4;
+}
+
+// given a code point cp, writes to c
+// the utf-8 code, outputting the length in
+// bytes, if the length is zero, the code point
+// is invalid
+//
+// This can possibly be made faster using pdep
+// and clz and table lookups, but JSON documents
+// have few escaped code points, and the following
+// function looks cheap.
+//
+// Note: we assume that surrogates are treated separately
+//
+simdjson_inline size_t codepoint_to_utf8(uint32_t cp, uint8_t *c) {
+  if (cp <= 0x7F) {
+    c[0] = uint8_t(cp);
+    return 1; // ascii
+  }
+  if (cp <= 0x7FF) {
+    c[0] = uint8_t((cp >> 6) + 192);
+    c[1] = uint8_t((cp & 63) + 128);
+    return 2; // universal plane
+    //  Surrogates are treated elsewhere...
+    //} //else if (0xd800 <= cp && cp <= 0xdfff) {
+    //  return 0; // surrogates // could put assert here
+  } else if (cp <= 0xFFFF) {
+    c[0] = uint8_t((cp >> 12) + 224);
+    c[1] = uint8_t(((cp >> 6) & 63) + 128);
+    c[2] = uint8_t((cp & 63) + 128);
+    return 3;
+  } else if (cp <= 0x10FFFF) { // if you know you have a valid code point, this
+                               // is not needed
+    c[0] = uint8_t((cp >> 18) + 240);
+    c[1] = uint8_t(((cp >> 12) & 63) + 128);
+    c[2] = uint8_t(((cp >> 6) & 63) + 128);
+    c[3] = uint8_t((cp & 63) + 128);
+    return 4;
+  }
+  // will return 0 when the code point was too large.
+  return 0; // bad r
+}
+
+#if SIMDJSON_IS_32BITS // _umul128 for x86, arm
+// this is a slow emulation routine for 32-bit
+//
+static simdjson_inline uint64_t __emulu(uint32_t x, uint32_t y) {
+  return x * (uint64_t)y;
+}
+static simdjson_inline uint64_t _umul128(uint64_t ab, uint64_t cd, uint64_t *hi) {
+  uint64_t ad = __emulu((uint32_t)(ab >> 32), (uint32_t)cd);
+  uint64_t bd = __emulu((uint32_t)ab, (uint32_t)cd);
+  uint64_t adbc = ad + __emulu((uint32_t)ab, (uint32_t)(cd >> 32));
+  uint64_t adbc_carry = !!(adbc < ad);
+  uint64_t lo = bd + (adbc << 32);
+  *hi = __emulu((uint32_t)(ab >> 32), (uint32_t)(cd >> 32)) + (adbc >> 32) +
+        (adbc_carry << 32) + !!(lo < bd);
+  return lo;
+}
+#endif
+
+} // namespace jsoncharutils
+} // unnamed namespace
+} // namespace SIMDJSON_IMPLEMENTATION
+} // namespace simdjson
+
+#endif // SIMDJSON_GENERIC_JSONCHARUTILS_H
\ No newline at end of file
diff --git a/contrib/libs/simdjson/include/simdjson/generic/numberparsing.h b/contrib/libs/simdjson/include/simdjson/generic/numberparsing.h
new file mode 100644
index 00000000000..facc2acc6f9
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/generic/numberparsing.h
@@ -0,0 +1,1310 @@
+#ifndef SIMDJSON_GENERIC_NUMBERPARSING_H
+
+#ifndef SIMDJSON_CONDITIONAL_INCLUDE
+#define SIMDJSON_GENERIC_NUMBERPARSING_H
+#include "simdjson/generic/base.h"
+#include "simdjson/generic/jsoncharutils.h"
+#include "simdjson/internal/numberparsing_tables.h"
+#endif // SIMDJSON_CONDITIONAL_INCLUDE
+
+#include <limits>
+#include <ostream>
+#include <cstring>
+
+namespace simdjson {
+namespace SIMDJSON_IMPLEMENTATION {
+namespace numberparsing {
+
+#ifdef JSON_TEST_NUMBERS
+#define INVALID_NUMBER(SRC) (found_invalid_number((SRC)), NUMBER_ERROR)
+#define WRITE_INTEGER(VALUE, SRC, WRITER) (found_integer((VALUE), (SRC)), (WRITER).append_s64((VALUE)))
+#define WRITE_UNSIGNED(VALUE, SRC, WRITER) (found_unsigned_integer((VALUE), (SRC)), (WRITER).append_u64((VALUE)))
+#define WRITE_DOUBLE(VALUE, SRC, WRITER) (found_float((VALUE), (SRC)), (WRITER).append_double((VALUE)))
+#define BIGINT_NUMBER(SRC) (found_invalid_number((SRC)), BIGINT_ERROR)
+#else
+#define INVALID_NUMBER(SRC) (NUMBER_ERROR)
+#define WRITE_INTEGER(VALUE, SRC, WRITER) (WRITER).append_s64((VALUE))
+#define WRITE_UNSIGNED(VALUE, SRC, WRITER) (WRITER).append_u64((VALUE))
+#define WRITE_DOUBLE(VALUE, SRC, WRITER) (WRITER).append_double((VALUE))
+#define BIGINT_NUMBER(SRC) (BIGINT_ERROR)
+#endif
+
+namespace {
+
+// Convert a mantissa, an exponent and a sign bit into an ieee64 double.
+// The real_exponent needs to be in [0, 2046] (technically real_exponent = 2047 would be acceptable).
+// The mantissa should be in [0,1<<53). The bit at index (1ULL << 52) while be zeroed.
+simdjson_inline double to_double(uint64_t mantissa, uint64_t real_exponent, bool negative) {
+    double d;
+    mantissa &= ~(1ULL << 52);
+    mantissa |= real_exponent << 52;
+    mantissa |= ((static_cast<uint64_t>(negative)) << 63);
+    std::memcpy(&d, &mantissa, sizeof(d));
+    return d;
+}
+
+// Attempts to compute i * 10^(power) exactly; and if "negative" is
+// true, negate the result.
+// This function will only work in some cases, when it does not work, success is
+// set to false. This should work *most of the time* (like 99% of the time).
+// We assume that power is in the [smallest_power,
+// largest_power] interval: the caller is responsible for this check.
+simdjson_inline bool compute_float_64(int64_t power, uint64_t i, bool negative, double &d) {
+  // we start with a fast path
+  // It was described in
+  // Clinger WD. How to read floating point numbers accurately.
+  // ACM SIGPLAN Notices. 1990
+#ifndef FLT_EVAL_METHOD
+#error "FLT_EVAL_METHOD should be defined, please include cfloat."
+#endif
+#if (FLT_EVAL_METHOD != 1) && (FLT_EVAL_METHOD != 0)
+  // We cannot be certain that x/y is rounded to nearest.
+  if (0 <= power && power <= 22 && i <= 9007199254740991)
+#else
+  if (-22 <= power && power <= 22 && i <= 9007199254740991)
+#endif
+  {
+    // convert the integer into a double. This is lossless since
+    // 0 <= i <= 2^53 - 1.
+    d = double(i);
+    //
+    // The general idea is as follows.
+    // If 0 <= s < 2^53 and if 10^0 <= p <= 10^22 then
+    // 1) Both s and p can be represented exactly as 64-bit floating-point
+    // values
+    // (binary64).
+    // 2) Because s and p can be represented exactly as floating-point values,
+    // then s * p
+    // and s / p will produce correctly rounded values.
+    //
+    if (power < 0) {
+      d = d / simdjson::internal::power_of_ten[-power];
+    } else {
+      d = d * simdjson::internal::power_of_ten[power];
+    }
+    if (negative) {
+      d = -d;
+    }
+    return true;
+  }
+  // When 22 < power && power <  22 + 16, we could
+  // hope for another, secondary fast path.  It was
+  // described by David M. Gay in  "Correctly rounded
+  // binary-decimal and decimal-binary conversions." (1990)
+  // If you need to compute i * 10^(22 + x) for x < 16,
+  // first compute i * 10^x, if you know that result is exact
+  // (e.g., when i * 10^x < 2^53),
+  // then you can still proceed and do (i * 10^x) * 10^22.
+  // Is this worth your time?
+  // You need  22 < power *and* power <  22 + 16 *and* (i * 10^(x-22) < 2^53)
+  // for this second fast path to work.
+  // If you you have 22 < power *and* power <  22 + 16, and then you
+  // optimistically compute "i * 10^(x-22)", there is still a chance that you
+  // have wasted your time if i * 10^(x-22) >= 2^53. It makes the use cases of
+  // this optimization maybe less common than we would like. Source:
+  // http://www.exploringbinary.com/fast-path-decimal-to-floating-point-conversion/
+  // also used in RapidJSON: https://rapidjson.org/strtod_8h_source.html
+
+  // The fast path has now failed, so we are failing back on the slower path.
+
+  // In the slow path, we need to adjust i so that it is > 1<<63 which is always
+  // possible, except if i == 0, so we handle i == 0 separately.
+  if(i == 0) {
+    d = negative ? -0.0 : 0.0;
+    return true;
+  }
+
+
+  // The exponent is 1024 + 63 + power
+  //     + floor(log(5**power)/log(2)).
+  // The 1024 comes from the ieee64 standard.
+  // The 63 comes from the fact that we use a 64-bit word.
+  //
+  // Computing floor(log(5**power)/log(2)) could be
+  // slow. Instead we use a fast function.
+  //
+  // For power in (-400,350), we have that
+  // (((152170 + 65536) * power ) >> 16);
+  // is equal to
+  //  floor(log(5**power)/log(2)) + power when power >= 0
+  // and it is equal to
+  //  ceil(log(5**-power)/log(2)) + power when power < 0
+  //
+  // The 65536 is (1<<16) and corresponds to
+  // (65536 * power) >> 16 ---> power
+  //
+  // ((152170 * power ) >> 16) is equal to
+  // floor(log(5**power)/log(2))
+  //
+  // Note that this is not magic: 152170/(1<<16) is
+  // approximatively equal to log(5)/log(2).
+  // The 1<<16 value is a power of two; we could use a
+  // larger power of 2 if we wanted to.
+  //
+  int64_t exponent = (((152170 + 65536) * power) >> 16) + 1024 + 63;
+
+
+  // We want the most significant bit of i to be 1. Shift if needed.
+  int lz = leading_zeroes(i);
+  i <<= lz;
+
+
+  // We are going to need to do some 64-bit arithmetic to get a precise product.
+  // We use a table lookup approach.
+  // It is safe because
+  // power >= smallest_power
+  // and power <= largest_power
+  // We recover the mantissa of the power, it has a leading 1. It is always
+  // rounded down.
+  //
+  // We want the most significant 64 bits of the product. We know
+  // this will be non-zero because the most significant bit of i is
+  // 1.
+  const uint32_t index = 2 * uint32_t(power - simdjson::internal::smallest_power);
+  // Optimization: It may be that materializing the index as a variable might confuse some compilers and prevent effective complex-addressing loads. (Done for code clarity.)
+  //
+  // The full_multiplication function computes the 128-bit product of two 64-bit words
+  // with a returned value of type value128 with a "low component" corresponding to the
+  // 64-bit least significant bits of the product and with a "high component" corresponding
+  // to the 64-bit most significant bits of the product.
+  simdjson::internal::value128 firstproduct = full_multiplication(i, simdjson::internal::power_of_five_128[index]);
+  // Both i and power_of_five_128[index] have their most significant bit set to 1 which
+  // implies that the either the most or the second most significant bit of the product
+  // is 1. We pack values in this manner for efficiency reasons: it maximizes the use
+  // we make of the product. It also makes it easy to reason about the product: there
+  // is 0 or 1 leading zero in the product.
+
+  // Unless the least significant 9 bits of the high (64-bit) part of the full
+  // product are all 1s, then we know that the most significant 55 bits are
+  // exact and no further work is needed. Having 55 bits is necessary because
+  // we need 53 bits for the mantissa but we have to have one rounding bit and
+  // we can waste a bit if the most significant bit of the product is zero.
+  if((firstproduct.high & 0x1FF) == 0x1FF) {
+    // We want to compute i * 5^q, but only care about the top 55 bits at most.
+    // Consider the scenario where q>=0. Then 5^q may not fit in 64-bits. Doing
+    // the full computation is wasteful. So we do what is called a "truncated
+    // multiplication".
+    // We take the most significant 64-bits, and we put them in
+    // power_of_five_128[index]. Usually, that's good enough to approximate i * 5^q
+    // to the desired approximation using one multiplication. Sometimes it does not suffice.
+    // Then we store the next most significant 64 bits in power_of_five_128[index + 1], and
+    // then we get a better approximation to i * 5^q.
+    //
+    // That's for when q>=0. The logic for q<0 is somewhat similar but it is somewhat
+    // more complicated.
+    //
+    // There is an extra layer of complexity in that we need more than 55 bits of
+    // accuracy in the round-to-even scenario.
+    //
+    // The full_multiplication function computes the 128-bit product of two 64-bit words
+    // with a returned value of type value128 with a "low component" corresponding to the
+    // 64-bit least significant bits of the product and with a "high component" corresponding
+    // to the 64-bit most significant bits of the product.
+    simdjson::internal::value128 secondproduct = full_multiplication(i, simdjson::internal::power_of_five_128[index + 1]);
+    firstproduct.low += secondproduct.high;
+    if(secondproduct.high > firstproduct.low) { firstproduct.high++; }
+    // As it has been proven by Noble Mushtak and Daniel Lemire in "Fast Number Parsing Without
+    // Fallback" (https://arxiv.org/abs/2212.06644), at this point we are sure that the product
+    // is sufficiently accurate, and more computation is not needed.
+  }
+  uint64_t lower = firstproduct.low;
+  uint64_t upper = firstproduct.high;
+  // The final mantissa should be 53 bits with a leading 1.
+  // We shift it so that it occupies 54 bits with a leading 1.
+  ///////
+  uint64_t upperbit = upper >> 63;
+  uint64_t mantissa = upper >> (upperbit + 9);
+  lz += int(1 ^ upperbit);
+
+  // Here we have mantissa < (1<<54).
+  int64_t real_exponent = exponent - lz;
+  if (simdjson_unlikely(real_exponent <= 0)) { // we have a subnormal?
+    // Here have that real_exponent <= 0 so -real_exponent >= 0
+    if(-real_exponent + 1 >= 64) { // if we have more than 64 bits below the minimum exponent, you have a zero for sure.
+      d = negative ? -0.0 : 0.0;
+      return true;
+    }
+    // next line is safe because -real_exponent + 1 < 0
+    mantissa >>= -real_exponent + 1;
+    // Thankfully, we can't have both "round-to-even" and subnormals because
+    // "round-to-even" only occurs for powers close to 0.
+    mantissa += (mantissa & 1); // round up
+    mantissa >>= 1;
+    // There is a weird scenario where we don't have a subnormal but just.
+    // Suppose we start with 2.2250738585072013e-308, we end up
+    // with 0x3fffffffffffff x 2^-1023-53 which is technically subnormal
+    // whereas 0x40000000000000 x 2^-1023-53  is normal. Now, we need to round
+    // up 0x3fffffffffffff x 2^-1023-53  and once we do, we are no longer
+    // subnormal, but we can only know this after rounding.
+    // So we only declare a subnormal if we are smaller than the threshold.
+    real_exponent = (mantissa < (uint64_t(1) << 52)) ? 0 : 1;
+    d = to_double(mantissa, real_exponent, negative);
+    return true;
+  }
+  // We have to round to even. The "to even" part
+  // is only a problem when we are right in between two floats
+  // which we guard against.
+  // If we have lots of trailing zeros, we may fall right between two
+  // floating-point values.
+  //
+  // The round-to-even cases take the form of a number 2m+1 which is in (2^53,2^54]
+  // times a power of two. That is, it is right between a number with binary significand
+  // m and another number with binary significand m+1; and it must be the case
+  // that it cannot be represented by a float itself.
+  //
+  // We must have that w * 10 ^q == (2m+1) * 2^p for some power of two 2^p.
+  // Recall that 10^q = 5^q * 2^q.
+  // When q >= 0, we must have that (2m+1) is divible by 5^q, so 5^q <= 2^54. We have that
+  //  5^23 <=  2^54 and it is the last power of five to qualify, so q <= 23.
+  // When q<0, we have  w  >=  (2m+1) x 5^{-q}.  We must have that w<2^{64} so
+  // (2m+1) x 5^{-q} < 2^{64}. We have that 2m+1>2^{53}. Hence, we must have
+  // 2^{53} x 5^{-q} < 2^{64}.
+  // Hence we have 5^{-q} < 2^{11}$ or q>= -4.
+  //
+  // We require lower <= 1 and not lower == 0 because we could not prove that
+  // that lower == 0 is implied; but we could prove that lower <= 1 is a necessary and sufficient test.
+  if (simdjson_unlikely((lower <= 1) && (power >= -4) && (power <= 23) && ((mantissa & 3) == 1))) {
+    if((mantissa  << (upperbit + 64 - 53 - 2)) ==  upper) {
+      mantissa &= ~1;             // flip it so that we do not round up
+    }
+  }
+
+  mantissa += mantissa & 1;
+  mantissa >>= 1;
+
+  // Here we have mantissa < (1<<53), unless there was an overflow
+  if (mantissa >= (1ULL << 53)) {
+    //////////
+    // This will happen when parsing values such as 7.2057594037927933e+16
+    ////////
+    mantissa = (1ULL << 52);
+    real_exponent++;
+  }
+  mantissa &= ~(1ULL << 52);
+  // we have to check that real_exponent is in range, otherwise we bail out
+  if (simdjson_unlikely(real_exponent > 2046)) {
+    // We have an infinite value!!! We could actually throw an error here if we could.
+    return false;
+  }
+  d = to_double(mantissa, real_exponent, negative);
+  return true;
+}
+
+// We call a fallback floating-point parser that might be slow. Note
+// it will accept JSON numbers, but the JSON spec. is more restrictive so
+// before you call parse_float_fallback, you need to have validated the input
+// string with the JSON grammar.
+// It will return an error (false) if the parsed number is infinite.
+// The string parsing itself always succeeds. We know that there is at least
+// one digit.
+static bool parse_float_fallback(const uint8_t *ptr, double *outDouble) {
+  *outDouble = simdjson::internal::from_chars(reinterpret_cast<const char *>(ptr));
+  // We do not accept infinite values.
+
+  // Detecting finite values in a portable manner is ridiculously hard, ideally
+  // we would want to do:
+  // return !std::isfinite(*outDouble);
+  // but that mysteriously fails under legacy/old libc++ libraries, see
+  // https://github.com/simdjson/simdjson/issues/1286
+  //
+  // Therefore, fall back to this solution (the extra parens are there
+  // to handle that max may be a macro on windows).
+  return !(*outDouble > (std::numeric_limits<double>::max)() || *outDouble < std::numeric_limits<double>::lowest());
+}
+
+static bool parse_float_fallback(const uint8_t *ptr, const uint8_t *end_ptr, double *outDouble) {
+  *outDouble = simdjson::internal::from_chars(reinterpret_cast<const char *>(ptr), reinterpret_cast<const char *>(end_ptr));
+  // We do not accept infinite values.
+
+  // Detecting finite values in a portable manner is ridiculously hard, ideally
+  // we would want to do:
+  // return !std::isfinite(*outDouble);
+  // but that mysteriously fails under legacy/old libc++ libraries, see
+  // https://github.com/simdjson/simdjson/issues/1286
+  //
+  // Therefore, fall back to this solution (the extra parens are there
+  // to handle that max may be a macro on windows).
+  return !(*outDouble > (std::numeric_limits<double>::max)() || *outDouble < std::numeric_limits<double>::lowest());
+}
+
+// check quickly whether the next 8 chars are made of digits
+// at a glance, it looks better than Mula's
+// http://0x80.pl/articles/swar-digits-validate.html
+simdjson_inline bool is_made_of_eight_digits_fast(const uint8_t *chars) {
+  uint64_t val;
+  // this can read up to 7 bytes beyond the buffer size, but we require
+  // SIMDJSON_PADDING of padding
+  static_assert(7 <= SIMDJSON_PADDING, "SIMDJSON_PADDING must be bigger than 7");
+  std::memcpy(&val, chars, 8);
+  // a branchy method might be faster:
+  // return (( val & 0xF0F0F0F0F0F0F0F0 ) == 0x3030303030303030)
+  //  && (( (val + 0x0606060606060606) & 0xF0F0F0F0F0F0F0F0 ) ==
+  //  0x3030303030303030);
+  return (((val & 0xF0F0F0F0F0F0F0F0) |
+           (((val + 0x0606060606060606) & 0xF0F0F0F0F0F0F0F0) >> 4)) ==
+          0x3333333333333333);
+}
+
+template<typename I>
+SIMDJSON_NO_SANITIZE_UNDEFINED // We deliberately allow overflow here and check later
+simdjson_inline bool parse_digit(const uint8_t c, I &i) {
+  const uint8_t digit = static_cast<uint8_t>(c - '0');
+  if (digit > 9) {
+    return false;
+  }
+  // PERF NOTE: multiplication by 10 is cheaper than arbitrary integer multiplication
+  i = 10 * i + digit; // might overflow, we will handle the overflow later
+  return true;
+}
+
+simdjson_inline bool is_digit(const uint8_t c) {
+  return static_cast<uint8_t>(c - '0') <= 9;
+}
+
+simdjson_inline error_code parse_decimal_after_separator(simdjson_unused const uint8_t *const src, const uint8_t *&p, uint64_t &i, int64_t &exponent) {
+  // we continue with the fiction that we have an integer. If the
+  // floating point number is representable as x * 10^z for some integer
+  // z that fits in 53 bits, then we will be able to convert back the
+  // the integer into a float in a lossless manner.
+  const uint8_t *const first_after_period = p;
+
+#ifdef SIMDJSON_SWAR_NUMBER_PARSING
+#if SIMDJSON_SWAR_NUMBER_PARSING
+  // this helps if we have lots of decimals!
+  // this turns out to be frequent enough.
+  if (is_made_of_eight_digits_fast(p)) {
+    i = i * 100000000 + parse_eight_digits_unrolled(p);
+    p += 8;
+  }
+#endif // SIMDJSON_SWAR_NUMBER_PARSING
+#endif // #ifdef SIMDJSON_SWAR_NUMBER_PARSING
+  // Unrolling the first digit makes a small difference on some implementations (e.g. westmere)
+  if (parse_digit(*p, i)) { ++p; }
+  while (parse_digit(*p, i)) { p++; }
+  exponent = first_after_period - p;
+  // Decimal without digits (123.) is illegal
+  if (exponent == 0) {
+    return INVALID_NUMBER(src);
+  }
+  return SUCCESS;
+}
+
+simdjson_inline error_code parse_exponent(simdjson_unused const uint8_t *const src, const uint8_t *&p, int64_t &exponent) {
+  // Exp Sign: -123.456e[-]78
+  bool neg_exp = ('-' == *p);
+  if (neg_exp || '+' == *p) { p++; } // Skip + as well
+
+  // Exponent: -123.456e-[78]
+  auto start_exp = p;
+  int64_t exp_number = 0;
+  while (parse_digit(*p, exp_number)) { ++p; }
+  // It is possible for parse_digit to overflow.
+  // In particular, it could overflow to INT64_MIN, and we cannot do - INT64_MIN.
+  // Thus we *must* check for possible overflow before we negate exp_number.
+
+  // Performance notes: it may seem like combining the two "simdjson_unlikely checks" below into
+  // a single simdjson_unlikely path would be faster. The reasoning is sound, but the compiler may
+  // not oblige and may, in fact, generate two distinct paths in any case. It might be
+  // possible to do uint64_t(p - start_exp - 1) >= 18 but it could end up trading off
+  // instructions for a simdjson_likely branch, an unconclusive gain.
+
+  // If there were no digits, it's an error.
+  if (simdjson_unlikely(p == start_exp)) {
+    return INVALID_NUMBER(src);
+  }
+  // We have a valid positive exponent in exp_number at this point, except that
+  // it may have overflowed.
+
+  // If there were more than 18 digits, we may have overflowed the integer. We have to do
+  // something!!!!
+  if (simdjson_unlikely(p > start_exp+18)) {
+    // Skip leading zeroes: 1e000000000000000000001 is technically valid and does not overflow
+    while (*start_exp == '0') { start_exp++; }
+    // 19 digits could overflow int64_t and is kind of absurd anyway. We don't
+    // support exponents smaller than -999,999,999,999,999,999 and bigger
+    // than 999,999,999,999,999,999.
+    // We can truncate.
+    // Note that 999999999999999999 is assuredly too large. The maximal ieee64 value before
+    // infinity is ~1.8e308. The smallest subnormal is ~5e-324. So, actually, we could
+    // truncate at 324.
+    // Note that there is no reason to fail per se at this point in time.
+    // E.g., 0e999999999999999999999 is a fine number.
+    if (p > start_exp+18) { exp_number = 999999999999999999; }
+  }
+  // At this point, we know that exp_number is a sane, positive, signed integer.
+  // It is <= 999,999,999,999,999,999. As long as 'exponent' is in
+  // [-8223372036854775808, 8223372036854775808], we won't overflow. Because 'exponent'
+  // is bounded in magnitude by the size of the JSON input, we are fine in this universe.
+  // To sum it up: the next line should never overflow.
+  exponent += (neg_exp ? -exp_number : exp_number);
+  return SUCCESS;
+}
+
+simdjson_inline bool check_if_integer(const uint8_t *const src, size_t max_length) {
+  const uint8_t *const srcend = src + max_length;
+  bool negative = (*src == '-'); // we can always read at least one character after the '-'
+  const uint8_t *p = src + uint8_t(negative);
+  if(p == srcend) { return false; }
+  if(*p == '0') {
+    ++p;
+    if(p == srcend) { return true; }
+    if(jsoncharutils::is_not_structural_or_whitespace(*p)) { return false; }
+    return true;
+  }
+  while(p != srcend && is_digit(*p)) { ++p; }
+  if(p == srcend) { return true; }
+  if(jsoncharutils::is_not_structural_or_whitespace(*p)) { return false; }
+  return true;
+}
+
+simdjson_inline size_t significant_digits(const uint8_t * start_digits, size_t digit_count) {
+  // It is possible that the integer had an overflow.
+  // We have to handle the case where we have 0.0000somenumber.
+  const uint8_t *start = start_digits;
+  while ((*start == '0') || (*start == '.')) { ++start; }
+  // we over-decrement by one when there is a '.'
+  return digit_count - size_t(start - start_digits);
+}
+
+} // unnamed namespace
+
+/** @private */
+static error_code slow_float_parsing(simdjson_unused const uint8_t * src, double* answer) {
+  if (parse_float_fallback(src, answer)) {
+    return SUCCESS;
+  }
+  return INVALID_NUMBER(src);
+}
+
+/** @private */
+template<typename W>
+simdjson_inline error_code write_float(const uint8_t *const src, bool negative, uint64_t i, const uint8_t * start_digits, size_t digit_count, int64_t exponent, W &writer) {
+  // If we frequently had to deal with long strings of digits,
+  // we could extend our code by using a 128-bit integer instead
+  // of a 64-bit integer. However, this is uncommon in practice.
+  //
+  // 9999999999999999999 < 2**64 so we can accommodate 19 digits.
+  // If we have a decimal separator, then digit_count - 1 is the number of digits, but we
+  // may not have a decimal separator!
+  if (simdjson_unlikely(digit_count > 19 && significant_digits(start_digits, digit_count) > 19)) {
+    // Ok, chances are good that we had an overflow!
+    // this is almost never going to get called!!!
+    // we start anew, going slowly!!!
+    // This will happen in the following examples:
+    // 10000000000000000000000000000000000000000000e+308
+    // 3.1415926535897932384626433832795028841971693993751
+    //
+    // NOTE: We do not pass a reference to the to slow_float_parsing. If we passed our writer
+    // reference to it, it would force it to be stored in memory, preventing the compiler from
+    // picking it apart and putting into registers. i.e. if we pass it as reference,
+    // it gets slow.
+    double d;
+    error_code error = slow_float_parsing(src, &d);
+    writer.append_double(d);
+    return error;
+  }
+  // NOTE: it's weird that the simdjson_unlikely() only wraps half the if, but it seems to get slower any other
+  // way we've tried: https://github.com/simdjson/simdjson/pull/990#discussion_r448497331
+  // To future reader: we'd love if someone found a better way, or at least could explain this result!
+  if (simdjson_unlikely(exponent < simdjson::internal::smallest_power) || (exponent > simdjson::internal::largest_power)) {
+    //
+    // Important: smallest_power is such that it leads to a zero value.
+    // Observe that 18446744073709551615e-343 == 0, i.e. (2**64 - 1) e -343 is zero
+    // so something x 10^-343 goes to zero, but not so with  something x 10^-342.
+    static_assert(simdjson::internal::smallest_power <= -342, "smallest_power is not small enough");
+    //
+    if((exponent < simdjson::internal::smallest_power) || (i == 0)) {
+      // E.g. Parse "-0.0e-999" into the same value as "-0.0". See https://en.wikipedia.org/wiki/Signed_zero
+      WRITE_DOUBLE(negative ? -0.0 : 0.0, src, writer);
+      return SUCCESS;
+    } else { // (exponent > largest_power) and (i != 0)
+      // We have, for sure, an infinite value and simdjson refuses to parse infinite values.
+      return INVALID_NUMBER(src);
+    }
+  }
+  double d;
+  if (!compute_float_64(exponent, i, negative, d)) {
+    // we are almost never going to get here.
+    if (!parse_float_fallback(src, &d)) { return INVALID_NUMBER(src); }
+  }
+  WRITE_DOUBLE(d, src, writer);
+  return SUCCESS;
+}
+
+// parse the number at src
+// define JSON_TEST_NUMBERS for unit testing
+//
+// It is assumed that the number is followed by a structural ({,},],[) character
+// or a white space character. If that is not the case (e.g., when the JSON
+// document is made of a single number), then it is necessary to copy the
+// content and append a space before calling this function.
+//
+// Our objective is accurate parsing (ULP of 0) at high speed.
+template<typename W>
+simdjson_inline error_code parse_number(const uint8_t *const src, W &writer);
+
+// for performance analysis, it is sometimes  useful to skip parsing
+#ifdef SIMDJSON_SKIPNUMBERPARSING
+
+template<typename W>
+simdjson_inline error_code parse_number(const uint8_t *const, W &writer) {
+  writer.append_s64(0);        // always write zero
+  return SUCCESS;              // always succeeds
+}
+
+simdjson_unused simdjson_inline simdjson_result<uint64_t> parse_unsigned(const uint8_t * const src) noexcept { return 0; }
+simdjson_unused simdjson_inline simdjson_result<int64_t> parse_integer(const uint8_t * const src) noexcept { return 0; }
+simdjson_unused simdjson_inline simdjson_result<double> parse_double(const uint8_t * const src) noexcept { return 0; }
+simdjson_unused simdjson_inline simdjson_result<uint64_t> parse_unsigned_in_string(const uint8_t * const src) noexcept { return 0; }
+simdjson_unused simdjson_inline simdjson_result<int64_t> parse_integer_in_string(const uint8_t * const src) noexcept { return 0; }
+simdjson_unused simdjson_inline simdjson_result<double> parse_double_in_string(const uint8_t * const src) noexcept { return 0; }
+simdjson_unused simdjson_inline bool is_negative(const uint8_t * src) noexcept  { return false; }
+simdjson_unused simdjson_inline simdjson_result<bool> is_integer(const uint8_t * src) noexcept  { return false; }
+simdjson_unused simdjson_inline simdjson_result<number_type> get_number_type(const uint8_t * src) noexcept { return number_type::signed_integer; }
+#else
+
+// parse the number at src
+// define JSON_TEST_NUMBERS for unit testing
+//
+// It is assumed that the number is followed by a structural ({,},],[) character
+// or a white space character. If that is not the case (e.g., when the JSON
+// document is made of a single number), then it is necessary to copy the
+// content and append a space before calling this function.
+//
+// Our objective is accurate parsing (ULP of 0) at high speed.
+template<typename W>
+simdjson_inline error_code parse_number(const uint8_t *const src, W &writer) {
+
+  //
+  // Check for minus sign
+  //
+  bool negative = (*src == '-');
+  const uint8_t *p = src + uint8_t(negative);
+
+  //
+  // Parse the integer part.
+  //
+  // PERF NOTE: we don't use is_made_of_eight_digits_fast because large integers like 123456789 are rare
+  const uint8_t *const start_digits = p;
+  uint64_t i = 0;
+  while (parse_digit(*p, i)) { p++; }
+
+  // If there were no digits, or if the integer starts with 0 and has more than one digit, it's an error.
+  // Optimization note: size_t is expected to be unsigned.
+  size_t digit_count = size_t(p - start_digits);
+  if (digit_count == 0 || ('0' == *start_digits && digit_count > 1)) { return INVALID_NUMBER(src); }
+
+  //
+  // Handle floats if there is a . or e (or both)
+  //
+  int64_t exponent = 0;
+  bool is_float = false;
+  if ('.' == *p) {
+    is_float = true;
+    ++p;
+    SIMDJSON_TRY( parse_decimal_after_separator(src, p, i, exponent) );
+    digit_count = int(p - start_digits); // used later to guard against overflows
+  }
+  if (('e' == *p) || ('E' == *p)) {
+    is_float = true;
+    ++p;
+    SIMDJSON_TRY( parse_exponent(src, p, exponent) );
+  }
+  if (is_float) {
+    const bool dirty_end = jsoncharutils::is_not_structural_or_whitespace(*p);
+    SIMDJSON_TRY( write_float(src, negative, i, start_digits, digit_count, exponent, writer) );
+    if (dirty_end) { return INVALID_NUMBER(src); }
+    return SUCCESS;
+  }
+
+  // The longest negative 64-bit number is 19 digits.
+  // The longest positive 64-bit number is 20 digits.
+  // We do it this way so we don't trigger this branch unless we must.
+  size_t longest_digit_count = negative ? 19 : 20;
+  if (digit_count > longest_digit_count) { return BIGINT_NUMBER(src); }
+  if (digit_count == longest_digit_count) {
+    if (negative) {
+      // Anything negative above INT64_MAX+1 is invalid
+      if (i > uint64_t(INT64_MAX)+1) { return BIGINT_NUMBER(src);  }
+      WRITE_INTEGER(~i+1, src, writer);
+      if (jsoncharutils::is_not_structural_or_whitespace(*p)) { return INVALID_NUMBER(src); }
+      return SUCCESS;
+    // Positive overflow check:
+    // - A 20 digit number starting with 2-9 is overflow, because 18,446,744,073,709,551,615 is the
+    //   biggest uint64_t.
+    // - A 20 digit number starting with 1 is overflow if it is less than INT64_MAX.
+    //   If we got here, it's a 20 digit number starting with the digit "1".
+    // - If a 20 digit number starting with 1 overflowed (i*10+digit), the result will be smaller
+    //   than 1,553,255,926,290,448,384.
+    // - That is smaller than the smallest possible 20-digit number the user could write:
+    //   10,000,000,000,000,000,000.
+    // - Therefore, if the number is positive and lower than that, it's overflow.
+    // - The value we are looking at is less than or equal to INT64_MAX.
+    //
+    }  else if (src[0] != uint8_t('1') || i <= uint64_t(INT64_MAX)) { return INVALID_NUMBER(src); }
+  }
+
+  // Write unsigned if it does not fit in a signed integer.
+  if (i > uint64_t(INT64_MAX)) {
+    WRITE_UNSIGNED(i, src, writer);
+  } else {
+    WRITE_INTEGER(negative ? (~i+1) : i, src, writer);
+  }
+  if (jsoncharutils::is_not_structural_or_whitespace(*p)) { return INVALID_NUMBER(src); }
+  return SUCCESS;
+}
+
+// Inlineable functions
+namespace {
+
+// This table can be used to characterize the final character of an integer
+// string. For JSON structural character and allowable white space characters,
+// we return SUCCESS. For 'e', '.' and 'E', we return INCORRECT_TYPE. Otherwise
+// we return NUMBER_ERROR.
+// Optimization note: we could easily reduce the size of the table by half (to 128)
+// at the cost of an extra branch.
+// Optimization note: we want the values to use at most 8 bits (not, e.g., 32 bits):
+static_assert(error_code(uint8_t(NUMBER_ERROR))== NUMBER_ERROR, "bad NUMBER_ERROR cast");
+static_assert(error_code(uint8_t(SUCCESS))== SUCCESS, "bad NUMBER_ERROR cast");
+static_assert(error_code(uint8_t(INCORRECT_TYPE))== INCORRECT_TYPE, "bad NUMBER_ERROR cast");
+
+const uint8_t integer_string_finisher[256] = {
+    NUMBER_ERROR, NUMBER_ERROR,   NUMBER_ERROR, NUMBER_ERROR, NUMBER_ERROR,
+    NUMBER_ERROR, NUMBER_ERROR,   NUMBER_ERROR, NUMBER_ERROR, SUCCESS,
+    SUCCESS,      NUMBER_ERROR,   NUMBER_ERROR, SUCCESS,      NUMBER_ERROR,
+    NUMBER_ERROR, NUMBER_ERROR,   NUMBER_ERROR, NUMBER_ERROR, NUMBER_ERROR,
+    NUMBER_ERROR, NUMBER_ERROR,   NUMBER_ERROR, NUMBER_ERROR, NUMBER_ERROR,
+    NUMBER_ERROR, NUMBER_ERROR,   NUMBER_ERROR, NUMBER_ERROR, NUMBER_ERROR,
+    NUMBER_ERROR, NUMBER_ERROR,   SUCCESS,      NUMBER_ERROR, NUMBER_ERROR,
+    NUMBER_ERROR, NUMBER_ERROR,   NUMBER_ERROR, NUMBER_ERROR, NUMBER_ERROR,
+    NUMBER_ERROR, NUMBER_ERROR,   NUMBER_ERROR, NUMBER_ERROR, SUCCESS,
+    NUMBER_ERROR, INCORRECT_TYPE, NUMBER_ERROR, NUMBER_ERROR, NUMBER_ERROR,
+    NUMBER_ERROR, NUMBER_ERROR,   NUMBER_ERROR, NUMBER_ERROR, NUMBER_ERROR,
+    NUMBER_ERROR, NUMBER_ERROR,   NUMBER_ERROR, SUCCESS,      NUMBER_ERROR,
+    NUMBER_ERROR, NUMBER_ERROR,   NUMBER_ERROR, NUMBER_ERROR, NUMBER_ERROR,
+    NUMBER_ERROR, NUMBER_ERROR,   NUMBER_ERROR, NUMBER_ERROR, INCORRECT_TYPE,
+    NUMBER_ERROR, NUMBER_ERROR,   NUMBER_ERROR, NUMBER_ERROR, NUMBER_ERROR,
+    NUMBER_ERROR, NUMBER_ERROR,   NUMBER_ERROR, NUMBER_ERROR, NUMBER_ERROR,
+    NUMBER_ERROR, NUMBER_ERROR,   NUMBER_ERROR, NUMBER_ERROR, NUMBER_ERROR,
+    NUMBER_ERROR, NUMBER_ERROR,   NUMBER_ERROR, NUMBER_ERROR, NUMBER_ERROR,
+    NUMBER_ERROR, SUCCESS,        NUMBER_ERROR, SUCCESS,      NUMBER_ERROR,
+    NUMBER_ERROR, NUMBER_ERROR,   NUMBER_ERROR, NUMBER_ERROR, NUMBER_ERROR,
+    NUMBER_ERROR, INCORRECT_TYPE, NUMBER_ERROR, NUMBER_ERROR, NUMBER_ERROR,
+    NUMBER_ERROR, NUMBER_ERROR,   NUMBER_ERROR, NUMBER_ERROR, NUMBER_ERROR,
+    NUMBER_ERROR, NUMBER_ERROR,   NUMBER_ERROR, NUMBER_ERROR, NUMBER_ERROR,
+    NUMBER_ERROR, NUMBER_ERROR,   NUMBER_ERROR, NUMBER_ERROR, NUMBER_ERROR,
+    NUMBER_ERROR, NUMBER_ERROR,   NUMBER_ERROR, SUCCESS,      NUMBER_ERROR,
+    SUCCESS,      NUMBER_ERROR,   NUMBER_ERROR, NUMBER_ERROR, NUMBER_ERROR,
+    NUMBER_ERROR, NUMBER_ERROR,   NUMBER_ERROR, NUMBER_ERROR, NUMBER_ERROR,
+    NUMBER_ERROR, NUMBER_ERROR,   NUMBER_ERROR, NUMBER_ERROR, NUMBER_ERROR,
+    NUMBER_ERROR, NUMBER_ERROR,   NUMBER_ERROR, NUMBER_ERROR, NUMBER_ERROR,
+    NUMBER_ERROR, NUMBER_ERROR,   NUMBER_ERROR, NUMBER_ERROR, NUMBER_ERROR,
+    NUMBER_ERROR, NUMBER_ERROR,   NUMBER_ERROR, NUMBER_ERROR, NUMBER_ERROR,
+    NUMBER_ERROR, NUMBER_ERROR,   NUMBER_ERROR, NUMBER_ERROR, NUMBER_ERROR,
+    NUMBER_ERROR, NUMBER_ERROR,   NUMBER_ERROR, NUMBER_ERROR, NUMBER_ERROR,
+    NUMBER_ERROR, NUMBER_ERROR,   NUMBER_ERROR, NUMBER_ERROR, NUMBER_ERROR,
+    NUMBER_ERROR, NUMBER_ERROR,   NUMBER_ERROR, NUMBER_ERROR, NUMBER_ERROR,
+    NUMBER_ERROR, NUMBER_ERROR,   NUMBER_ERROR, NUMBER_ERROR, NUMBER_ERROR,
+    NUMBER_ERROR, NUMBER_ERROR,   NUMBER_ERROR, NUMBER_ERROR, NUMBER_ERROR,
+    NUMBER_ERROR, NUMBER_ERROR,   NUMBER_ERROR, NUMBER_ERROR, NUMBER_ERROR,
+    NUMBER_ERROR, NUMBER_ERROR,   NUMBER_ERROR, NUMBER_ERROR, NUMBER_ERROR,
+    NUMBER_ERROR, NUMBER_ERROR,   NUMBER_ERROR, NUMBER_ERROR, NUMBER_ERROR,
+    NUMBER_ERROR, NUMBER_ERROR,   NUMBER_ERROR, NUMBER_ERROR, NUMBER_ERROR,
+    NUMBER_ERROR, NUMBER_ERROR,   NUMBER_ERROR, NUMBER_ERROR, NUMBER_ERROR,
+    NUMBER_ERROR, NUMBER_ERROR,   NUMBER_ERROR, NUMBER_ERROR, NUMBER_ERROR,
+    NUMBER_ERROR, NUMBER_ERROR,   NUMBER_ERROR, NUMBER_ERROR, NUMBER_ERROR,
+    NUMBER_ERROR, NUMBER_ERROR,   NUMBER_ERROR, NUMBER_ERROR, NUMBER_ERROR,
+    NUMBER_ERROR, NUMBER_ERROR,   NUMBER_ERROR, NUMBER_ERROR, NUMBER_ERROR,
+    NUMBER_ERROR, NUMBER_ERROR,   NUMBER_ERROR, NUMBER_ERROR, NUMBER_ERROR,
+    NUMBER_ERROR, NUMBER_ERROR,   NUMBER_ERROR, NUMBER_ERROR, NUMBER_ERROR,
+    NUMBER_ERROR, NUMBER_ERROR,   NUMBER_ERROR, NUMBER_ERROR, NUMBER_ERROR,
+    NUMBER_ERROR, NUMBER_ERROR,   NUMBER_ERROR, NUMBER_ERROR, NUMBER_ERROR,
+    NUMBER_ERROR, NUMBER_ERROR,   NUMBER_ERROR, NUMBER_ERROR, NUMBER_ERROR,
+    NUMBER_ERROR};
+
+// Parse any number from 0 to 18,446,744,073,709,551,615
+simdjson_unused simdjson_inline simdjson_result<uint64_t> parse_unsigned(const uint8_t * const src) noexcept {
+  const uint8_t *p = src;
+  //
+  // Parse the integer part.
+  //
+  // PERF NOTE: we don't use is_made_of_eight_digits_fast because large integers like 123456789 are rare
+  const uint8_t *const start_digits = p;
+  uint64_t i = 0;
+  while (parse_digit(*p, i)) { p++; }
+
+  // If there were no digits, or if the integer starts with 0 and has more than one digit, it's an error.
+  // Optimization note: size_t is expected to be unsigned.
+  size_t digit_count = size_t(p - start_digits);
+  // The longest positive 64-bit number is 20 digits.
+  // We do it this way so we don't trigger this branch unless we must.
+  // Optimization note: the compiler can probably merge
+  // ((digit_count == 0) || (digit_count > 20))
+  // into a single  branch since digit_count is unsigned.
+  if ((digit_count == 0) || (digit_count > 20)) { return INCORRECT_TYPE; }
+  // Here digit_count > 0.
+  if (('0' == *start_digits) && (digit_count > 1)) { return NUMBER_ERROR; }
+  // We can do the following...
+  // if (!jsoncharutils::is_structural_or_whitespace(*p)) {
+  //  return (*p == '.' || *p == 'e' || *p == 'E') ? INCORRECT_TYPE : NUMBER_ERROR;
+  // }
+  // as a single table lookup:
+  if (integer_string_finisher[*p] != SUCCESS) { return error_code(integer_string_finisher[*p]); }
+
+  if (digit_count == 20) {
+    // Positive overflow check:
+    // - A 20 digit number starting with 2-9 is overflow, because 18,446,744,073,709,551,615 is the
+    //   biggest uint64_t.
+    // - A 20 digit number starting with 1 is overflow if it is less than INT64_MAX.
+    //   If we got here, it's a 20 digit number starting with the digit "1".
+    // - If a 20 digit number starting with 1 overflowed (i*10+digit), the result will be smaller
+    //   than 1,553,255,926,290,448,384.
+    // - That is smaller than the smallest possible 20-digit number the user could write:
+    //   10,000,000,000,000,000,000.
+    // - Therefore, if the number is positive and lower than that, it's overflow.
+    // - The value we are looking at is less than or equal to INT64_MAX.
+    //
+    if (src[0] != uint8_t('1') || i <= uint64_t(INT64_MAX)) { return INCORRECT_TYPE; }
+  }
+
+  return i;
+}
+
+
+// Parse any number from 0 to 18,446,744,073,709,551,615
+// Never read at src_end or beyond
+simdjson_unused simdjson_inline simdjson_result<uint64_t> parse_unsigned(const uint8_t * const src, const uint8_t * const src_end) noexcept {
+  const uint8_t *p = src;
+  //
+  // Parse the integer part.
+  //
+  // PERF NOTE: we don't use is_made_of_eight_digits_fast because large integers like 123456789 are rare
+  const uint8_t *const start_digits = p;
+  uint64_t i = 0;
+  while ((p != src_end) && parse_digit(*p, i)) { p++; }
+
+  // If there were no digits, or if the integer starts with 0 and has more than one digit, it's an error.
+  // Optimization note: size_t is expected to be unsigned.
+  size_t digit_count = size_t(p - start_digits);
+  // The longest positive 64-bit number is 20 digits.
+  // We do it this way so we don't trigger this branch unless we must.
+  // Optimization note: the compiler can probably merge
+  // ((digit_count == 0) || (digit_count > 20))
+  // into a single  branch since digit_count is unsigned.
+  if ((digit_count == 0) || (digit_count > 20)) { return INCORRECT_TYPE; }
+  // Here digit_count > 0.
+  if (('0' == *start_digits) && (digit_count > 1)) { return NUMBER_ERROR; }
+  // We can do the following...
+  // if (!jsoncharutils::is_structural_or_whitespace(*p)) {
+  //  return (*p == '.' || *p == 'e' || *p == 'E') ? INCORRECT_TYPE : NUMBER_ERROR;
+  // }
+  // as a single table lookup:
+  if ((p != src_end) && integer_string_finisher[*p] != SUCCESS) { return error_code(integer_string_finisher[*p]); }
+
+  if (digit_count == 20) {
+    // Positive overflow check:
+    // - A 20 digit number starting with 2-9 is overflow, because 18,446,744,073,709,551,615 is the
+    //   biggest uint64_t.
+    // - A 20 digit number starting with 1 is overflow if it is less than INT64_MAX.
+    //   If we got here, it's a 20 digit number starting with the digit "1".
+    // - If a 20 digit number starting with 1 overflowed (i*10+digit), the result will be smaller
+    //   than 1,553,255,926,290,448,384.
+    // - That is smaller than the smallest possible 20-digit number the user could write:
+    //   10,000,000,000,000,000,000.
+    // - Therefore, if the number is positive and lower than that, it's overflow.
+    // - The value we are looking at is less than or equal to INT64_MAX.
+    //
+    if (src[0] != uint8_t('1') || i <= uint64_t(INT64_MAX)) { return INCORRECT_TYPE; }
+  }
+
+  return i;
+}
+
+// Parse any number from 0 to 18,446,744,073,709,551,615
+simdjson_unused simdjson_inline simdjson_result<uint64_t> parse_unsigned_in_string(const uint8_t * const src) noexcept {
+  const uint8_t *p = src + 1;
+  //
+  // Parse the integer part.
+  //
+  // PERF NOTE: we don't use is_made_of_eight_digits_fast because large integers like 123456789 are rare
+  const uint8_t *const start_digits = p;
+  uint64_t i = 0;
+  while (parse_digit(*p, i)) { p++; }
+
+  // If there were no digits, or if the integer starts with 0 and has more than one digit, it's an error.
+  // Optimization note: size_t is expected to be unsigned.
+  size_t digit_count = size_t(p - start_digits);
+  // The longest positive 64-bit number is 20 digits.
+  // We do it this way so we don't trigger this branch unless we must.
+  // Optimization note: the compiler can probably merge
+  // ((digit_count == 0) || (digit_count > 20))
+  // into a single  branch since digit_count is unsigned.
+  if ((digit_count == 0) || (digit_count > 20)) { return INCORRECT_TYPE; }
+  // Here digit_count > 0.
+  if (('0' == *start_digits) && (digit_count > 1)) { return NUMBER_ERROR; }
+  // We can do the following...
+  // if (!jsoncharutils::is_structural_or_whitespace(*p)) {
+  //  return (*p == '.' || *p == 'e' || *p == 'E') ? INCORRECT_TYPE : NUMBER_ERROR;
+  // }
+  // as a single table lookup:
+  if (*p != '"') { return NUMBER_ERROR; }
+
+  if (digit_count == 20) {
+    // Positive overflow check:
+    // - A 20 digit number starting with 2-9 is overflow, because 18,446,744,073,709,551,615 is the
+    //   biggest uint64_t.
+    // - A 20 digit number starting with 1 is overflow if it is less than INT64_MAX.
+    //   If we got here, it's a 20 digit number starting with the digit "1".
+    // - If a 20 digit number starting with 1 overflowed (i*10+digit), the result will be smaller
+    //   than 1,553,255,926,290,448,384.
+    // - That is smaller than the smallest possible 20-digit number the user could write:
+    //   10,000,000,000,000,000,000.
+    // - Therefore, if the number is positive and lower than that, it's overflow.
+    // - The value we are looking at is less than or equal to INT64_MAX.
+    //
+    // Note: we use src[1] and not src[0] because src[0] is the quote character in this
+    // instance.
+    if (src[1] != uint8_t('1') || i <= uint64_t(INT64_MAX)) { return INCORRECT_TYPE; }
+  }
+
+  return i;
+}
+
+// Parse any number from  -9,223,372,036,854,775,808 to 9,223,372,036,854,775,807
+simdjson_unused simdjson_inline simdjson_result<int64_t> parse_integer(const uint8_t *src) noexcept {
+  //
+  // Check for minus sign
+  //
+  bool negative = (*src == '-');
+  const uint8_t *p = src + uint8_t(negative);
+
+  //
+  // Parse the integer part.
+  //
+  // PERF NOTE: we don't use is_made_of_eight_digits_fast because large integers like 123456789 are rare
+  const uint8_t *const start_digits = p;
+  uint64_t i = 0;
+  while (parse_digit(*p, i)) { p++; }
+
+  // If there were no digits, or if the integer starts with 0 and has more than one digit, it's an error.
+  // Optimization note: size_t is expected to be unsigned.
+  size_t digit_count = size_t(p - start_digits);
+  // We go from
+  // -9,223,372,036,854,775,808 to 9,223,372,036,854,775,807
+  // so we can never represent numbers that have more than 19 digits.
+  size_t longest_digit_count = 19;
+  // Optimization note: the compiler can probably merge
+  // ((digit_count == 0) || (digit_count > longest_digit_count))
+  // into a single  branch since digit_count is unsigned.
+  if ((digit_count == 0) || (digit_count > longest_digit_count)) { return INCORRECT_TYPE; }
+  // Here digit_count > 0.
+  if (('0' == *start_digits) && (digit_count > 1)) { return NUMBER_ERROR; }
+  // We can do the following...
+  // if (!jsoncharutils::is_structural_or_whitespace(*p)) {
+  //  return (*p == '.' || *p == 'e' || *p == 'E') ? INCORRECT_TYPE : NUMBER_ERROR;
+  // }
+  // as a single table lookup:
+  if(integer_string_finisher[*p] != SUCCESS) { return error_code(integer_string_finisher[*p]); }
+  // Negative numbers have can go down to - INT64_MAX - 1 whereas positive numbers are limited to INT64_MAX.
+  // Performance note: This check is only needed when digit_count == longest_digit_count but it is
+  // so cheap that we might as well always make it.
+  if(i > uint64_t(INT64_MAX) + uint64_t(negative)) { return INCORRECT_TYPE; }
+  return negative ? (~i+1) : i;
+}
+
+// Parse any number from  -9,223,372,036,854,775,808 to 9,223,372,036,854,775,807
+// Never read at src_end or beyond
+simdjson_unused simdjson_inline simdjson_result<int64_t> parse_integer(const uint8_t * const src, const uint8_t * const src_end) noexcept {
+  //
+  // Check for minus sign
+  //
+  if(src == src_end) { return NUMBER_ERROR; }
+  bool negative = (*src == '-');
+  const uint8_t *p = src + uint8_t(negative);
+
+  //
+  // Parse the integer part.
+  //
+  // PERF NOTE: we don't use is_made_of_eight_digits_fast because large integers like 123456789 are rare
+  const uint8_t *const start_digits = p;
+  uint64_t i = 0;
+  while ((p != src_end) && parse_digit(*p, i)) { p++; }
+
+  // If there were no digits, or if the integer starts with 0 and has more than one digit, it's an error.
+  // Optimization note: size_t is expected to be unsigned.
+  size_t digit_count = size_t(p - start_digits);
+  // We go from
+  // -9,223,372,036,854,775,808 to 9,223,372,036,854,775,807
+  // so we can never represent numbers that have more than 19 digits.
+  size_t longest_digit_count = 19;
+  // Optimization note: the compiler can probably merge
+  // ((digit_count == 0) || (digit_count > longest_digit_count))
+  // into a single  branch since digit_count is unsigned.
+  if ((digit_count == 0) || (digit_count > longest_digit_count)) { return INCORRECT_TYPE; }
+  // Here digit_count > 0.
+  if (('0' == *start_digits) && (digit_count > 1)) { return NUMBER_ERROR; }
+  // We can do the following...
+  // if (!jsoncharutils::is_structural_or_whitespace(*p)) {
+  //  return (*p == '.' || *p == 'e' || *p == 'E') ? INCORRECT_TYPE : NUMBER_ERROR;
+  // }
+  // as a single table lookup:
+  if((p != src_end) && integer_string_finisher[*p] != SUCCESS) { return error_code(integer_string_finisher[*p]); }
+  // Negative numbers have can go down to - INT64_MAX - 1 whereas positive numbers are limited to INT64_MAX.
+  // Performance note: This check is only needed when digit_count == longest_digit_count but it is
+  // so cheap that we might as well always make it.
+  if(i > uint64_t(INT64_MAX) + uint64_t(negative)) { return INCORRECT_TYPE; }
+  return negative ? (~i+1) : i;
+}
+
+// Parse any number from  -9,223,372,036,854,775,808 to 9,223,372,036,854,775,807
+simdjson_unused simdjson_inline simdjson_result<int64_t> parse_integer_in_string(const uint8_t *src) noexcept {
+  //
+  // Check for minus sign
+  //
+  bool negative = (*(src + 1) == '-');
+  src += uint8_t(negative) + 1;
+
+  //
+  // Parse the integer part.
+  //
+  // PERF NOTE: we don't use is_made_of_eight_digits_fast because large integers like 123456789 are rare
+  const uint8_t *const start_digits = src;
+  uint64_t i = 0;
+  while (parse_digit(*src, i)) { src++; }
+
+  // If there were no digits, or if the integer starts with 0 and has more than one digit, it's an error.
+  // Optimization note: size_t is expected to be unsigned.
+  size_t digit_count = size_t(src - start_digits);
+  // We go from
+  // -9,223,372,036,854,775,808 to 9,223,372,036,854,775,807
+  // so we can never represent numbers that have more than 19 digits.
+  size_t longest_digit_count = 19;
+  // Optimization note: the compiler can probably merge
+  // ((digit_count == 0) || (digit_count > longest_digit_count))
+  // into a single  branch since digit_count is unsigned.
+  if ((digit_count == 0) || (digit_count > longest_digit_count)) { return INCORRECT_TYPE; }
+  // Here digit_count > 0.
+  if (('0' == *start_digits) && (digit_count > 1)) { return NUMBER_ERROR; }
+  // We can do the following...
+  // if (!jsoncharutils::is_structural_or_whitespace(*src)) {
+  //  return (*src == '.' || *src == 'e' || *src == 'E') ? INCORRECT_TYPE : NUMBER_ERROR;
+  // }
+  // as a single table lookup:
+  if(*src != '"') { return NUMBER_ERROR; }
+  // Negative numbers have can go down to - INT64_MAX - 1 whereas positive numbers are limited to INT64_MAX.
+  // Performance note: This check is only needed when digit_count == longest_digit_count but it is
+  // so cheap that we might as well always make it.
+  if(i > uint64_t(INT64_MAX) + uint64_t(negative)) { return INCORRECT_TYPE; }
+  return negative ? (~i+1) : i;
+}
+
+simdjson_unused simdjson_inline simdjson_result<double> parse_double(const uint8_t * src) noexcept {
+  //
+  // Check for minus sign
+  //
+  bool negative = (*src == '-');
+  src += uint8_t(negative);
+
+  //
+  // Parse the integer part.
+  //
+  uint64_t i = 0;
+  const uint8_t *p = src;
+  p += parse_digit(*p, i);
+  bool leading_zero = (i == 0);
+  while (parse_digit(*p, i)) { p++; }
+  // no integer digits, or 0123 (zero must be solo)
+  if ( p == src ) { return INCORRECT_TYPE; }
+  if ( (leading_zero && p != src+1)) { return NUMBER_ERROR; }
+
+  //
+  // Parse the decimal part.
+  //
+  int64_t exponent = 0;
+  bool overflow;
+  if (simdjson_likely(*p == '.')) {
+    p++;
+    const uint8_t *start_decimal_digits = p;
+    if (!parse_digit(*p, i)) { return NUMBER_ERROR; } // no decimal digits
+    p++;
+    while (parse_digit(*p, i)) { p++; }
+    exponent = -(p - start_decimal_digits);
+
+    // Overflow check. More than 19 digits (minus the decimal) may be overflow.
+    overflow = p-src-1 > 19;
+    if (simdjson_unlikely(overflow && leading_zero)) {
+      // Skip leading 0.00000 and see if it still overflows
+      const uint8_t *start_digits = src + 2;
+      while (*start_digits == '0') { start_digits++; }
+      overflow = p-start_digits > 19;
+    }
+  } else {
+    overflow = p-src > 19;
+  }
+
+  //
+  // Parse the exponent
+  //
+  if (*p == 'e' || *p == 'E') {
+    p++;
+    bool exp_neg = *p == '-';
+    p += exp_neg || *p == '+';
+
+    uint64_t exp = 0;
+    const uint8_t *start_exp_digits = p;
+    while (parse_digit(*p, exp)) { p++; }
+    // no exp digits, or 20+ exp digits
+    if (p-start_exp_digits == 0 || p-start_exp_digits > 19) { return NUMBER_ERROR; }
+
+    exponent += exp_neg ? 0-exp : exp;
+  }
+
+  if (jsoncharutils::is_not_structural_or_whitespace(*p)) { return NUMBER_ERROR; }
+
+  overflow = overflow || exponent < simdjson::internal::smallest_power || exponent > simdjson::internal::largest_power;
+
+  //
+  // Assemble (or slow-parse) the float
+  //
+  double d;
+  if (simdjson_likely(!overflow)) {
+    if (compute_float_64(exponent, i, negative, d)) { return d; }
+  }
+  if (!parse_float_fallback(src - uint8_t(negative), &d)) {
+    return NUMBER_ERROR;
+  }
+  return d;
+}
+
+simdjson_unused simdjson_inline bool is_negative(const uint8_t * src) noexcept {
+  return (*src == '-');
+}
+
+simdjson_unused simdjson_inline simdjson_result<bool> is_integer(const uint8_t * src) noexcept {
+  bool negative = (*src == '-');
+  src += uint8_t(negative);
+  const uint8_t *p = src;
+  while(static_cast<uint8_t>(*p - '0') <= 9) { p++; }
+  if ( p == src ) { return NUMBER_ERROR; }
+  if (jsoncharutils::is_structural_or_whitespace(*p)) { return true; }
+  return false;
+}
+
+simdjson_unused simdjson_inline simdjson_result<number_type> get_number_type(const uint8_t * src) noexcept {
+  bool negative = (*src == '-');
+  src += uint8_t(negative);
+  const uint8_t *p = src;
+  while(static_cast<uint8_t>(*p - '0') <= 9) { p++; }
+  size_t digit_count = size_t(p - src);
+  if ( p == src ) { return NUMBER_ERROR; }
+  if (jsoncharutils::is_structural_or_whitespace(*p)) {
+    static const uint8_t * smaller_big_integer = reinterpret_cast<const uint8_t *>("9223372036854775808");
+    // We have an integer.
+    if(simdjson_unlikely(digit_count > 20)) {
+      return number_type::big_integer;
+    }
+    // If the number is negative and valid, it must be a signed integer.
+    if(negative) {
+      if (simdjson_unlikely(digit_count > 19)) return number_type::big_integer;
+      if (simdjson_unlikely(digit_count == 19 && memcmp(src, smaller_big_integer, 19) > 0)) {
+        return number_type::big_integer;
+      }
+      return number_type::signed_integer;
+    }
+    // Let us check if we have a big integer (>=2**64).
+    static const uint8_t * two_to_sixtyfour = reinterpret_cast<const uint8_t *>("18446744073709551616");
+    if((digit_count > 20) || (digit_count == 20 && memcmp(src, two_to_sixtyfour, 20) >= 0)) {
+      return number_type::big_integer;
+    }
+    // The number is positive and smaller than 18446744073709551616 (or 2**64).
+    // We want values larger or equal to 9223372036854775808 to be unsigned
+    // integers, and the other values to be signed integers.
+    if((digit_count == 20) || (digit_count >= 19 && memcmp(src, smaller_big_integer, 19) >= 0)) {
+      return number_type::unsigned_integer;
+    }
+    return number_type::signed_integer;
+  }
+  // Hopefully, we have 'e' or 'E' or '.'.
+  return number_type::floating_point_number;
+}
+
+// Never read at src_end or beyond
+simdjson_unused simdjson_inline simdjson_result<double> parse_double(const uint8_t * src, const uint8_t * const src_end) noexcept {
+  if(src == src_end) { return NUMBER_ERROR; }
+  //
+  // Check for minus sign
+  //
+  bool negative = (*src == '-');
+  src += uint8_t(negative);
+
+  //
+  // Parse the integer part.
+  //
+  uint64_t i = 0;
+  const uint8_t *p = src;
+  if(p == src_end) { return NUMBER_ERROR; }
+  p += parse_digit(*p, i);
+  bool leading_zero = (i == 0);
+  while ((p != src_end) && parse_digit(*p, i)) { p++; }
+  // no integer digits, or 0123 (zero must be solo)
+  if ( p == src ) { return INCORRECT_TYPE; }
+  if ( (leading_zero && p != src+1)) { return NUMBER_ERROR; }
+
+  //
+  // Parse the decimal part.
+  //
+  int64_t exponent = 0;
+  bool overflow;
+  if (simdjson_likely((p != src_end) && (*p == '.'))) {
+    p++;
+    const uint8_t *start_decimal_digits = p;
+    if ((p == src_end) || !parse_digit(*p, i)) { return NUMBER_ERROR; } // no decimal digits
+    p++;
+    while ((p != src_end) && parse_digit(*p, i)) { p++; }
+    exponent = -(p - start_decimal_digits);
+
+    // Overflow check. More than 19 digits (minus the decimal) may be overflow.
+    overflow = p-src-1 > 19;
+    if (simdjson_unlikely(overflow && leading_zero)) {
+      // Skip leading 0.00000 and see if it still overflows
+      const uint8_t *start_digits = src + 2;
+      while (*start_digits == '0') { start_digits++; }
+      overflow = start_digits-src > 19;
+    }
+  } else {
+    overflow = p-src > 19;
+  }
+
+  //
+  // Parse the exponent
+  //
+  if ((p != src_end) && (*p == 'e' || *p == 'E')) {
+    p++;
+    if(p == src_end) { return NUMBER_ERROR; }
+    bool exp_neg = *p == '-';
+    p += exp_neg || *p == '+';
+
+    uint64_t exp = 0;
+    const uint8_t *start_exp_digits = p;
+    while ((p != src_end) && parse_digit(*p, exp)) { p++; }
+    // no exp digits, or 20+ exp digits
+    if (p-start_exp_digits == 0 || p-start_exp_digits > 19) { return NUMBER_ERROR; }
+
+    exponent += exp_neg ? 0-exp : exp;
+  }
+
+  if ((p != src_end) && jsoncharutils::is_not_structural_or_whitespace(*p)) { return NUMBER_ERROR; }
+
+  overflow = overflow || exponent < simdjson::internal::smallest_power || exponent > simdjson::internal::largest_power;
+
+  //
+  // Assemble (or slow-parse) the float
+  //
+  double d;
+  if (simdjson_likely(!overflow)) {
+    if (compute_float_64(exponent, i, negative, d)) { return d; }
+  }
+  if (!parse_float_fallback(src - uint8_t(negative), src_end, &d)) {
+    return NUMBER_ERROR;
+  }
+  return d;
+}
+
+simdjson_unused simdjson_inline simdjson_result<double> parse_double_in_string(const uint8_t * src) noexcept {
+  //
+  // Check for minus sign
+  //
+  bool negative = (*(src + 1) == '-');
+  src += uint8_t(negative) + 1;
+
+  //
+  // Parse the integer part.
+  //
+  uint64_t i = 0;
+  const uint8_t *p = src;
+  p += parse_digit(*p, i);
+  bool leading_zero = (i == 0);
+  while (parse_digit(*p, i)) { p++; }
+  // no integer digits, or 0123 (zero must be solo)
+  if ( p == src ) { return INCORRECT_TYPE; }
+  if ( (leading_zero && p != src+1)) { return NUMBER_ERROR; }
+
+  //
+  // Parse the decimal part.
+  //
+  int64_t exponent = 0;
+  bool overflow;
+  if (simdjson_likely(*p == '.')) {
+    p++;
+    const uint8_t *start_decimal_digits = p;
+    if (!parse_digit(*p, i)) { return NUMBER_ERROR; } // no decimal digits
+    p++;
+    while (parse_digit(*p, i)) { p++; }
+    exponent = -(p - start_decimal_digits);
+
+    // Overflow check. More than 19 digits (minus the decimal) may be overflow.
+    overflow = p-src-1 > 19;
+    if (simdjson_unlikely(overflow && leading_zero)) {
+      // Skip leading 0.00000 and see if it still overflows
+      const uint8_t *start_digits = src + 2;
+      while (*start_digits == '0') { start_digits++; }
+      overflow = p-start_digits > 19;
+    }
+  } else {
+    overflow = p-src > 19;
+  }
+
+  //
+  // Parse the exponent
+  //
+  if (*p == 'e' || *p == 'E') {
+    p++;
+    bool exp_neg = *p == '-';
+    p += exp_neg || *p == '+';
+
+    uint64_t exp = 0;
+    const uint8_t *start_exp_digits = p;
+    while (parse_digit(*p, exp)) { p++; }
+    // no exp digits, or 20+ exp digits
+    if (p-start_exp_digits == 0 || p-start_exp_digits > 19) { return NUMBER_ERROR; }
+
+    exponent += exp_neg ? 0-exp : exp;
+  }
+
+  if (*p != '"') { return NUMBER_ERROR; }
+
+  overflow = overflow || exponent < simdjson::internal::smallest_power || exponent > simdjson::internal::largest_power;
+
+  //
+  // Assemble (or slow-parse) the float
+  //
+  double d;
+  if (simdjson_likely(!overflow)) {
+    if (compute_float_64(exponent, i, negative, d)) { return d; }
+  }
+  if (!parse_float_fallback(src - uint8_t(negative), &d)) {
+    return NUMBER_ERROR;
+  }
+  return d;
+}
+
+} // unnamed namespace
+#endif // SIMDJSON_SKIPNUMBERPARSING
+
+} // namespace numberparsing
+
+inline std::ostream& operator<<(std::ostream& out, number_type type) noexcept {
+    switch (type) {
+        case number_type::signed_integer: out << "integer in [-9223372036854775808,9223372036854775808)"; break;
+        case number_type::unsigned_integer: out << "unsigned integer in [9223372036854775808,18446744073709551616)"; break;
+        case number_type::floating_point_number: out << "floating-point number (binary64)"; break;
+        case number_type::big_integer: out << "big integer"; break;
+        default: SIMDJSON_UNREACHABLE();
+    }
+    return out;
+}
+
+} // namespace SIMDJSON_IMPLEMENTATION
+} // namespace simdjson
+
+#endif // SIMDJSON_GENERIC_NUMBERPARSING_H
diff --git a/contrib/libs/simdjson/include/simdjson/haswell.h b/contrib/libs/simdjson/include/simdjson/haswell.h
new file mode 100644
index 00000000000..867b7a44911
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/haswell.h
@@ -0,0 +1,8 @@
+#ifndef SIMDJSON_HASWELL_H
+#define SIMDJSON_HASWELL_H
+
+#include "simdjson/haswell/begin.h"
+#include "simdjson/generic/amalgamated.h"
+#include "simdjson/haswell/end.h"
+
+#endif // SIMDJSON_HASWELL_H
\ No newline at end of file
diff --git a/contrib/libs/simdjson/include/simdjson/haswell/base.h b/contrib/libs/simdjson/include/simdjson/haswell/base.h
new file mode 100644
index 00000000000..73cc3ef1b4a
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/haswell/base.h
@@ -0,0 +1,27 @@
+#ifndef SIMDJSON_HASWELL_BASE_H
+#define SIMDJSON_HASWELL_BASE_H
+
+#ifndef SIMDJSON_CONDITIONAL_INCLUDE
+#include "simdjson/base.h"
+#endif // SIMDJSON_CONDITIONAL_INCLUDE
+
+// The constructor may be executed on any host, so we take care not to use SIMDJSON_TARGET_HASWELL
+namespace simdjson {
+/**
+ * Implementation for Haswell (Intel AVX2).
+ */
+namespace haswell {
+
+class implementation;
+
+namespace {
+namespace simd {
+template <typename T> struct simd8;
+template <typename T> struct simd8x64;
+} // namespace simd
+} // unnamed namespace
+
+} // namespace haswell
+} // namespace simdjson
+
+#endif // SIMDJSON_HASWELL_BASE_H
diff --git a/contrib/libs/simdjson/include/simdjson/haswell/begin.h b/contrib/libs/simdjson/include/simdjson/haswell/begin.h
new file mode 100644
index 00000000000..1be8c192c57
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/haswell/begin.h
@@ -0,0 +1,14 @@
+#define SIMDJSON_IMPLEMENTATION haswell
+
+#include "simdjson/haswell/base.h"
+#include "simdjson/haswell/intrinsics.h"
+
+#if !SIMDJSON_CAN_ALWAYS_RUN_HASWELL
+SIMDJSON_TARGET_REGION("avx2,bmi,pclmul,lzcnt,popcnt")
+#endif
+
+#include "simdjson/haswell/bitmanipulation.h"
+#include "simdjson/haswell/bitmask.h"
+#include "simdjson/haswell/numberparsing_defs.h"
+#include "simdjson/haswell/simd.h"
+#include "simdjson/haswell/stringparsing_defs.h"
diff --git a/contrib/libs/simdjson/include/simdjson/haswell/bitmanipulation.h b/contrib/libs/simdjson/include/simdjson/haswell/bitmanipulation.h
new file mode 100644
index 00000000000..9b0e59905c8
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/haswell/bitmanipulation.h
@@ -0,0 +1,71 @@
+#ifndef SIMDJSON_HASWELL_BITMANIPULATION_H
+#define SIMDJSON_HASWELL_BITMANIPULATION_H
+
+#ifndef SIMDJSON_CONDITIONAL_INCLUDE
+#include "simdjson/haswell/base.h"
+#include "simdjson/haswell/intrinsics.h"
+#include "simdjson/haswell/bitmask.h"
+#endif // SIMDJSON_CONDITIONAL_INCLUDE
+
+namespace simdjson {
+namespace haswell {
+namespace {
+
+// We sometimes call trailing_zero on inputs that are zero,
+// but the algorithms do not end up using the returned value.
+// Sadly, sanitizers are not smart enough to figure it out.
+SIMDJSON_NO_SANITIZE_UNDEFINED
+// This function can be used safely even if not all bytes have been
+// initialized.
+// See issue https://github.com/simdjson/simdjson/issues/1965
+SIMDJSON_NO_SANITIZE_MEMORY
+simdjson_inline int trailing_zeroes(uint64_t input_num) {
+#if SIMDJSON_REGULAR_VISUAL_STUDIO
+  return (int)_tzcnt_u64(input_num);
+#else // SIMDJSON_REGULAR_VISUAL_STUDIO
+  ////////
+  // You might expect the next line to be equivalent to
+  // return (int)_tzcnt_u64(input_num);
+  // but the generated code differs and might be less efficient?
+  ////////
+  return __builtin_ctzll(input_num);
+#endif // SIMDJSON_REGULAR_VISUAL_STUDIO
+}
+
+/* result might be undefined when input_num is zero */
+simdjson_inline uint64_t clear_lowest_bit(uint64_t input_num) {
+  return _blsr_u64(input_num);
+}
+
+/* result might be undefined when input_num is zero */
+simdjson_inline int leading_zeroes(uint64_t input_num) {
+  return int(_lzcnt_u64(input_num));
+}
+
+#if SIMDJSON_REGULAR_VISUAL_STUDIO
+simdjson_inline unsigned __int64 count_ones(uint64_t input_num) {
+  // note: we do not support legacy 32-bit Windows in this kernel
+  return __popcnt64(input_num);// Visual Studio wants two underscores
+}
+#else
+simdjson_inline long long int count_ones(uint64_t input_num) {
+  return _popcnt64(input_num);
+}
+#endif
+
+simdjson_inline bool add_overflow(uint64_t value1, uint64_t value2,
+                                uint64_t *result) {
+#if SIMDJSON_REGULAR_VISUAL_STUDIO
+  return _addcarry_u64(0, value1, value2,
+                       reinterpret_cast<unsigned __int64 *>(result));
+#else
+  return __builtin_uaddll_overflow(value1, value2,
+                                   reinterpret_cast<unsigned long long *>(result));
+#endif
+}
+
+} // unnamed namespace
+} // namespace haswell
+} // namespace simdjson
+
+#endif // SIMDJSON_HASWELL_BITMANIPULATION_H
diff --git a/contrib/libs/simdjson/include/simdjson/haswell/bitmask.h b/contrib/libs/simdjson/include/simdjson/haswell/bitmask.h
new file mode 100644
index 00000000000..310c3280bec
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/haswell/bitmask.h
@@ -0,0 +1,30 @@
+#ifndef SIMDJSON_HASWELL_BITMASK_H
+#define SIMDJSON_HASWELL_BITMASK_H
+
+#ifndef SIMDJSON_CONDITIONAL_INCLUDE
+#include "simdjson/haswell/base.h"
+#include "simdjson/haswell/intrinsics.h"
+#endif // SIMDJSON_CONDITIONAL_INCLUDE
+
+namespace simdjson {
+namespace haswell {
+namespace {
+
+//
+// Perform a "cumulative bitwise xor," flipping bits each time a 1 is encountered.
+//
+// For example, prefix_xor(00100100) == 00011100
+//
+simdjson_inline uint64_t prefix_xor(const uint64_t bitmask) {
+  // There should be no such thing with a processor supporting avx2
+  // but not clmul.
+  __m128i all_ones = _mm_set1_epi8('\xFF');
+  __m128i result = _mm_clmulepi64_si128(_mm_set_epi64x(0ULL, bitmask), all_ones, 0);
+  return _mm_cvtsi128_si64(result);
+}
+
+} // unnamed namespace
+} // namespace haswell
+} // namespace simdjson
+
+#endif // SIMDJSON_HASWELL_BITMASK_H
diff --git a/contrib/libs/simdjson/include/simdjson/haswell/end.h b/contrib/libs/simdjson/include/simdjson/haswell/end.h
new file mode 100644
index 00000000000..421df3653c9
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/haswell/end.h
@@ -0,0 +1,9 @@
+#ifndef SIMDJSON_CONDITIONAL_INCLUDE
+#include "simdjson/haswell/base.h"
+#endif // SIMDJSON_CONDITIONAL_INCLUDE
+
+#if !SIMDJSON_CAN_ALWAYS_RUN_HASWELL
+SIMDJSON_UNTARGET_REGION
+#endif
+
+#undef SIMDJSON_IMPLEMENTATION
diff --git a/contrib/libs/simdjson/include/simdjson/haswell/implementation.h b/contrib/libs/simdjson/include/simdjson/haswell/implementation.h
new file mode 100644
index 00000000000..6861e42983c
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/haswell/implementation.h
@@ -0,0 +1,36 @@
+#ifndef SIMDJSON_HASWELL_IMPLEMENTATION_H
+#define SIMDJSON_HASWELL_IMPLEMENTATION_H
+
+#ifndef SIMDJSON_CONDITIONAL_INCLUDE
+#include "simdjson/haswell/base.h"
+#include "simdjson/implementation.h"
+#include "simdjson/internal/instruction_set.h"
+#endif // SIMDJSON_CONDITIONAL_INCLUDE
+
+// The constructor may be executed on any host, so we take care not to use SIMDJSON_TARGET_HASWELL
+namespace simdjson {
+namespace haswell {
+
+/**
+ * @private
+ */
+class implementation final : public simdjson::implementation {
+public:
+  simdjson_inline implementation() : simdjson::implementation(
+      "haswell",
+      "Intel/AMD AVX2",
+      internal::instruction_set::AVX2 | internal::instruction_set::PCLMULQDQ | internal::instruction_set::BMI1 | internal::instruction_set::BMI2
+  ) {}
+  simdjson_warn_unused error_code create_dom_parser_implementation(
+    size_t capacity,
+    size_t max_length,
+    std::unique_ptr<internal::dom_parser_implementation>& dst
+  ) const noexcept final;
+  simdjson_warn_unused error_code minify(const uint8_t *buf, size_t len, uint8_t *dst, size_t &dst_len) const noexcept final;
+  simdjson_warn_unused bool validate_utf8(const char *buf, size_t len) const noexcept final;
+};
+
+} // namespace haswell
+} // namespace simdjson
+
+#endif // SIMDJSON_HASWELL_IMPLEMENTATION_H
diff --git a/contrib/libs/simdjson/include/simdjson/haswell/intrinsics.h b/contrib/libs/simdjson/include/simdjson/haswell/intrinsics.h
new file mode 100644
index 00000000000..a4593dbb698
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/haswell/intrinsics.h
@@ -0,0 +1,52 @@
+#ifndef SIMDJSON_HASWELL_INTRINSICS_H
+#define SIMDJSON_HASWELL_INTRINSICS_H
+
+#ifndef SIMDJSON_CONDITIONAL_INCLUDE
+#include "simdjson/haswell/base.h"
+#endif // SIMDJSON_CONDITIONAL_INCLUDE
+
+#if SIMDJSON_VISUAL_STUDIO
+// under clang within visual studio, this will include <x86intrin.h>
+#include <intrin.h>  // visual studio or clang
+#else
+#include <x86intrin.h> // elsewhere
+#endif // SIMDJSON_VISUAL_STUDIO
+
+#if SIMDJSON_CLANG_VISUAL_STUDIO
+/**
+ * You are not supposed, normally, to include these
+ * headers directly. Instead you should either include intrin.h
+ * or x86intrin.h. However, when compiling with clang
+ * under Windows (i.e., when _MSC_VER is set), these headers
+ * only get included *if* the corresponding features are detected
+ * from macros:
+ * e.g., if __AVX2__ is set... in turn,  we normally set these
+ * macros by compiling against the corresponding architecture
+ * (e.g., arch:AVX2, -mavx2, etc.) which compiles the whole
+ * software with these advanced instructions. In simdjson, we
+ * want to compile the whole program for a generic target,
+ * and only target our specific kernels. As a workaround,
+ * we directly include the needed headers. These headers would
+ * normally guard against such usage, but we carefully included
+ * <x86intrin.h>  (or <intrin.h>) before, so the headers
+ * are fooled.
+ */
+#include <bmiintrin.h>   // for _blsr_u64
+#include <lzcntintrin.h> // for  __lzcnt64
+#include <immintrin.h>   // for most things (AVX2, AVX512, _popcnt64)
+#include <smmintrin.h>
+#include <tmmintrin.h>
+#include <avxintrin.h>
+#include <avx2intrin.h>
+#include <wmmintrin.h>   // for  _mm_clmulepi64_si128
+// unfortunately, we may not get _blsr_u64, but, thankfully, clang
+// has it as a macro.
+#ifndef _blsr_u64
+// we roll our own
+#define _blsr_u64(n) ((n - 1) & n)
+#endif //  _blsr_u64
+#endif // SIMDJSON_CLANG_VISUAL_STUDIO
+
+static_assert(sizeof(__m256i) <= simdjson::SIMDJSON_PADDING, "insufficient padding for haswell kernel.");
+
+#endif // SIMDJSON_HASWELL_INTRINSICS_H
diff --git a/contrib/libs/simdjson/include/simdjson/haswell/numberparsing_defs.h b/contrib/libs/simdjson/include/simdjson/haswell/numberparsing_defs.h
new file mode 100644
index 00000000000..5673e5e4f81
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/haswell/numberparsing_defs.h
@@ -0,0 +1,61 @@
+#ifndef SIMDJSON_HASWELL_NUMBERPARSING_DEFS_H
+#define SIMDJSON_HASWELL_NUMBERPARSING_DEFS_H
+
+#ifndef SIMDJSON_CONDITIONAL_INCLUDE
+#include "simdjson/haswell/base.h"
+#include "simdjson/haswell/intrinsics.h"
+#endif // SIMDJSON_CONDITIONAL_INCLUDE
+
+#ifndef SIMDJSON_CONDITIONAL_INCLUDE
+#include "simdjson/internal/numberparsing_tables.h"
+#endif // SIMDJSON_CONDITIONAL_INCLUDE
+
+namespace simdjson {
+namespace haswell {
+namespace numberparsing {
+
+/** @private */
+static simdjson_inline uint32_t parse_eight_digits_unrolled(const uint8_t *chars) {
+  // this actually computes *16* values so we are being wasteful.
+  const __m128i ascii0 = _mm_set1_epi8('0');
+  const __m128i mul_1_10 =
+      _mm_setr_epi8(10, 1, 10, 1, 10, 1, 10, 1, 10, 1, 10, 1, 10, 1, 10, 1);
+  const __m128i mul_1_100 = _mm_setr_epi16(100, 1, 100, 1, 100, 1, 100, 1);
+  const __m128i mul_1_10000 =
+      _mm_setr_epi16(10000, 1, 10000, 1, 10000, 1, 10000, 1);
+  const __m128i input = _mm_sub_epi8(
+      _mm_loadu_si128(reinterpret_cast<const __m128i *>(chars)), ascii0);
+  const __m128i t1 = _mm_maddubs_epi16(input, mul_1_10);
+  const __m128i t2 = _mm_madd_epi16(t1, mul_1_100);
+  const __m128i t3 = _mm_packus_epi32(t2, t2);
+  const __m128i t4 = _mm_madd_epi16(t3, mul_1_10000);
+  return _mm_cvtsi128_si32(
+      t4); // only captures the sum of the first 8 digits, drop the rest
+}
+
+/** @private */
+simdjson_inline internal::value128 full_multiplication(uint64_t value1, uint64_t value2) {
+  internal::value128 answer;
+#if SIMDJSON_REGULAR_VISUAL_STUDIO || SIMDJSON_IS_32BITS
+#if SIMDJSON_IS_ARM64
+  // ARM64 has native support for 64-bit multiplications, no need to emultate
+  answer.high = __umulh(value1, value2);
+  answer.low = value1 * value2;
+#else
+  answer.low = _umul128(value1, value2, &answer.high); // _umul128 not available on ARM64
+#endif // SIMDJSON_IS_ARM64
+#else // SIMDJSON_REGULAR_VISUAL_STUDIO || SIMDJSON_IS_32BITS
+  __uint128_t r = (static_cast<__uint128_t>(value1)) * value2;
+  answer.low = uint64_t(r);
+  answer.high = uint64_t(r >> 64);
+#endif
+  return answer;
+}
+
+} // namespace numberparsing
+} // namespace haswell
+} // namespace simdjson
+
+#define SIMDJSON_SWAR_NUMBER_PARSING 1
+
+#endif // SIMDJSON_HASWELL_NUMBERPARSING_DEFS_H
diff --git a/contrib/libs/simdjson/include/simdjson/haswell/simd.h b/contrib/libs/simdjson/include/simdjson/haswell/simd.h
new file mode 100644
index 00000000000..9c40f4f4fdd
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/haswell/simd.h
@@ -0,0 +1,372 @@
+#ifndef SIMDJSON_HASWELL_SIMD_H
+#define SIMDJSON_HASWELL_SIMD_H
+
+#ifndef SIMDJSON_CONDITIONAL_INCLUDE
+#include "simdjson/haswell/base.h"
+#include "simdjson/haswell/intrinsics.h"
+#include "simdjson/haswell/bitmanipulation.h"
+#include "simdjson/internal/simdprune_tables.h"
+#endif // SIMDJSON_CONDITIONAL_INCLUDE
+
+namespace simdjson {
+namespace haswell {
+namespace {
+namespace simd {
+
+  // Forward-declared so they can be used by splat and friends.
+  template<typename Child>
+  struct base {
+    __m256i value;
+
+    // Zero constructor
+    simdjson_inline base() : value{__m256i()} {}
+
+    // Conversion from SIMD register
+    simdjson_inline base(const __m256i _value) : value(_value) {}
+
+    // Conversion to SIMD register
+    simdjson_inline operator const __m256i&() const { return this->value; }
+    simdjson_inline operator __m256i&() { return this->value; }
+
+    // Bit operations
+    simdjson_inline Child operator|(const Child other) const { return _mm256_or_si256(*this, other); }
+    simdjson_inline Child operator&(const Child other) const { return _mm256_and_si256(*this, other); }
+    simdjson_inline Child operator^(const Child other) const { return _mm256_xor_si256(*this, other); }
+    simdjson_inline Child bit_andnot(const Child other) const { return _mm256_andnot_si256(other, *this); }
+    simdjson_inline Child& operator|=(const Child other) { auto this_cast = static_cast<Child*>(this); *this_cast = *this_cast | other; return *this_cast; }
+    simdjson_inline Child& operator&=(const Child other) { auto this_cast = static_cast<Child*>(this); *this_cast = *this_cast & other; return *this_cast; }
+    simdjson_inline Child& operator^=(const Child other) { auto this_cast = static_cast<Child*>(this); *this_cast = *this_cast ^ other; return *this_cast; }
+  };
+
+  // Forward-declared so they can be used by splat and friends.
+  template<typename T>
+  struct simd8;
+
+  template<typename T, typename Mask=simd8<bool>>
+  struct base8: base<simd8<T>> {
+    typedef uint32_t bitmask_t;
+    typedef uint64_t bitmask2_t;
+
+    simdjson_inline base8() : base<simd8<T>>() {}
+    simdjson_inline base8(const __m256i _value) : base<simd8<T>>(_value) {}
+
+    friend simdjson_really_inline Mask operator==(const simd8<T> lhs, const simd8<T> rhs) { return _mm256_cmpeq_epi8(lhs, rhs); }
+
+    static const int SIZE = sizeof(base<T>::value);
+
+    template<int N=1>
+    simdjson_inline simd8<T> prev(const simd8<T> prev_chunk) const {
+      return _mm256_alignr_epi8(*this, _mm256_permute2x128_si256(prev_chunk, *this, 0x21), 16 - N);
+    }
+  };
+
+  // SIMD byte mask type (returned by things like eq and gt)
+  template<>
+  struct simd8<bool>: base8<bool> {
+    static simdjson_inline simd8<bool> splat(bool _value) { return _mm256_set1_epi8(uint8_t(-(!!_value))); }
+
+    simdjson_inline simd8() : base8() {}
+    simdjson_inline simd8(const __m256i _value) : base8<bool>(_value) {}
+    // Splat constructor
+    simdjson_inline simd8(bool _value) : base8<bool>(splat(_value)) {}
+
+    simdjson_inline int to_bitmask() const { return _mm256_movemask_epi8(*this); }
+    simdjson_inline bool any() const { return !_mm256_testz_si256(*this, *this); }
+    simdjson_inline simd8<bool> operator~() const { return *this ^ true; }
+  };
+
+  template<typename T>
+  struct base8_numeric: base8<T> {
+    static simdjson_inline simd8<T> splat(T _value) { return _mm256_set1_epi8(_value); }
+    static simdjson_inline simd8<T> zero() { return _mm256_setzero_si256(); }
+    static simdjson_inline simd8<T> load(const T values[32]) {
+      return _mm256_loadu_si256(reinterpret_cast<const __m256i *>(values));
+    }
+    // Repeat 16 values as many times as necessary (usually for lookup tables)
+    static simdjson_inline simd8<T> repeat_16(
+      T v0,  T v1,  T v2,  T v3,  T v4,  T v5,  T v6,  T v7,
+      T v8,  T v9,  T v10, T v11, T v12, T v13, T v14, T v15
+    ) {
+      return simd8<T>(
+        v0, v1, v2, v3, v4, v5, v6, v7,
+        v8, v9, v10,v11,v12,v13,v14,v15,
+        v0, v1, v2, v3, v4, v5, v6, v7,
+        v8, v9, v10,v11,v12,v13,v14,v15
+      );
+    }
+
+    simdjson_inline base8_numeric() : base8<T>() {}
+    simdjson_inline base8_numeric(const __m256i _value) : base8<T>(_value) {}
+
+    // Store to array
+    simdjson_inline void store(T dst[32]) const { return _mm256_storeu_si256(reinterpret_cast<__m256i *>(dst), *this); }
+
+    // Addition/subtraction are the same for signed and unsigned
+    simdjson_inline simd8<T> operator+(const simd8<T> other) const { return _mm256_add_epi8(*this, other); }
+    simdjson_inline simd8<T> operator-(const simd8<T> other) const { return _mm256_sub_epi8(*this, other); }
+    simdjson_inline simd8<T>& operator+=(const simd8<T> other) { *this = *this + other; return *static_cast<simd8<T>*>(this); }
+    simdjson_inline simd8<T>& operator-=(const simd8<T> other) { *this = *this - other; return *static_cast<simd8<T>*>(this); }
+
+    // Override to distinguish from bool version
+    simdjson_inline simd8<T> operator~() const { return *this ^ 0xFFu; }
+
+    // Perform a lookup assuming the value is between 0 and 16 (undefined behavior for out of range values)
+    template<typename L>
+    simdjson_inline simd8<L> lookup_16(simd8<L> lookup_table) const {
+      return _mm256_shuffle_epi8(lookup_table, *this);
+    }
+
+    // Copies to 'output" all bytes corresponding to a 0 in the mask (interpreted as a bitset).
+    // Passing a 0 value for mask would be equivalent to writing out every byte to output.
+    // Only the first 32 - count_ones(mask) bytes of the result are significant but 32 bytes
+    // get written.
+    // Design consideration: it seems like a function with the
+    // signature simd8<L> compress(uint32_t mask) would be
+    // sensible, but the AVX ISA makes this kind of approach difficult.
+    template<typename L>
+    simdjson_inline void compress(uint32_t mask, L * output) const {
+      using internal::thintable_epi8;
+      using internal::BitsSetTable256mul2;
+      using internal::pshufb_combine_table;
+      // this particular implementation was inspired by work done by @animetosho
+      // we do it in four steps, first 8 bytes and then second 8 bytes...
+      uint8_t mask1 = uint8_t(mask); // least significant 8 bits
+      uint8_t mask2 = uint8_t(mask >> 8); // second least significant 8 bits
+      uint8_t mask3 = uint8_t(mask >> 16); // ...
+      uint8_t mask4 = uint8_t(mask >> 24); // ...
+      // next line just loads the 64-bit values thintable_epi8[mask1] and
+      // thintable_epi8[mask2] into a 128-bit register, using only
+      // two instructions on most compilers.
+      __m256i shufmask =  _mm256_set_epi64x(thintable_epi8[mask4], thintable_epi8[mask3],
+        thintable_epi8[mask2], thintable_epi8[mask1]);
+      // we increment by 0x08 the second half of the mask and so forth
+      shufmask =
+      _mm256_add_epi8(shufmask, _mm256_set_epi32(0x18181818, 0x18181818,
+         0x10101010, 0x10101010, 0x08080808, 0x08080808, 0, 0));
+      // this is the version "nearly pruned"
+      __m256i pruned = _mm256_shuffle_epi8(*this, shufmask);
+      // we still need to put the  pieces back together.
+      // we compute the popcount of the first words:
+      int pop1 = BitsSetTable256mul2[mask1];
+      int pop3 = BitsSetTable256mul2[mask3];
+
+      // then load the corresponding mask
+      // could be done with _mm256_loadu2_m128i but many standard libraries omit this intrinsic.
+      __m256i v256 = _mm256_castsi128_si256(
+        _mm_loadu_si128(reinterpret_cast<const __m128i *>(pshufb_combine_table + pop1 * 8)));
+      __m256i compactmask = _mm256_insertf128_si256(v256,
+         _mm_loadu_si128(reinterpret_cast<const __m128i *>(pshufb_combine_table + pop3 * 8)), 1);
+      __m256i almostthere =  _mm256_shuffle_epi8(pruned, compactmask);
+      // We just need to write out the result.
+      // This is the tricky bit that is hard to do
+      // if we want to return a SIMD register, since there
+      // is no single-instruction approach to recombine
+      // the two 128-bit lanes with an offset.
+      __m128i v128;
+      v128 = _mm256_castsi256_si128(almostthere);
+      _mm_storeu_si128( reinterpret_cast<__m128i *>(output), v128);
+      v128 = _mm256_extractf128_si256(almostthere, 1);
+      _mm_storeu_si128( reinterpret_cast<__m128i *>(output + 16 - count_ones(mask & 0xFFFF)), v128);
+    }
+
+    template<typename L>
+    simdjson_inline simd8<L> lookup_16(
+        L replace0,  L replace1,  L replace2,  L replace3,
+        L replace4,  L replace5,  L replace6,  L replace7,
+        L replace8,  L replace9,  L replace10, L replace11,
+        L replace12, L replace13, L replace14, L replace15) const {
+      return lookup_16(simd8<L>::repeat_16(
+        replace0,  replace1,  replace2,  replace3,
+        replace4,  replace5,  replace6,  replace7,
+        replace8,  replace9,  replace10, replace11,
+        replace12, replace13, replace14, replace15
+      ));
+    }
+  };
+
+  // Signed bytes
+  template<>
+  struct simd8<int8_t> : base8_numeric<int8_t> {
+    simdjson_inline simd8() : base8_numeric<int8_t>() {}
+    simdjson_inline simd8(const __m256i _value) : base8_numeric<int8_t>(_value) {}
+    // Splat constructor
+    simdjson_inline simd8(int8_t _value) : simd8(splat(_value)) {}
+    // Array constructor
+    simdjson_inline simd8(const int8_t values[32]) : simd8(load(values)) {}
+    // Member-by-member initialization
+    simdjson_inline simd8(
+      int8_t v0,  int8_t v1,  int8_t v2,  int8_t v3,  int8_t v4,  int8_t v5,  int8_t v6,  int8_t v7,
+      int8_t v8,  int8_t v9,  int8_t v10, int8_t v11, int8_t v12, int8_t v13, int8_t v14, int8_t v15,
+      int8_t v16, int8_t v17, int8_t v18, int8_t v19, int8_t v20, int8_t v21, int8_t v22, int8_t v23,
+      int8_t v24, int8_t v25, int8_t v26, int8_t v27, int8_t v28, int8_t v29, int8_t v30, int8_t v31
+    ) : simd8(_mm256_setr_epi8(
+      v0, v1, v2, v3, v4, v5, v6, v7,
+      v8, v9, v10,v11,v12,v13,v14,v15,
+      v16,v17,v18,v19,v20,v21,v22,v23,
+      v24,v25,v26,v27,v28,v29,v30,v31
+    )) {}
+    // Repeat 16 values as many times as necessary (usually for lookup tables)
+    simdjson_inline static simd8<int8_t> repeat_16(
+      int8_t v0,  int8_t v1,  int8_t v2,  int8_t v3,  int8_t v4,  int8_t v5,  int8_t v6,  int8_t v7,
+      int8_t v8,  int8_t v9,  int8_t v10, int8_t v11, int8_t v12, int8_t v13, int8_t v14, int8_t v15
+    ) {
+      return simd8<int8_t>(
+        v0, v1, v2, v3, v4, v5, v6, v7,
+        v8, v9, v10,v11,v12,v13,v14,v15,
+        v0, v1, v2, v3, v4, v5, v6, v7,
+        v8, v9, v10,v11,v12,v13,v14,v15
+      );
+    }
+
+    // Order-sensitive comparisons
+    simdjson_inline simd8<int8_t> max_val(const simd8<int8_t> other) const { return _mm256_max_epi8(*this, other); }
+    simdjson_inline simd8<int8_t> min_val(const simd8<int8_t> other) const { return _mm256_min_epi8(*this, other); }
+    simdjson_inline simd8<bool> operator>(const simd8<int8_t> other) const { return _mm256_cmpgt_epi8(*this, other); }
+    simdjson_inline simd8<bool> operator<(const simd8<int8_t> other) const { return _mm256_cmpgt_epi8(other, *this); }
+  };
+
+  // Unsigned bytes
+  template<>
+  struct simd8<uint8_t>: base8_numeric<uint8_t> {
+    simdjson_inline simd8() : base8_numeric<uint8_t>() {}
+    simdjson_inline simd8(const __m256i _value) : base8_numeric<uint8_t>(_value) {}
+    // Splat constructor
+    simdjson_inline simd8(uint8_t _value) : simd8(splat(_value)) {}
+    // Array constructor
+    simdjson_inline simd8(const uint8_t values[32]) : simd8(load(values)) {}
+    // Member-by-member initialization
+    simdjson_inline simd8(
+      uint8_t v0,  uint8_t v1,  uint8_t v2,  uint8_t v3,  uint8_t v4,  uint8_t v5,  uint8_t v6,  uint8_t v7,
+      uint8_t v8,  uint8_t v9,  uint8_t v10, uint8_t v11, uint8_t v12, uint8_t v13, uint8_t v14, uint8_t v15,
+      uint8_t v16, uint8_t v17, uint8_t v18, uint8_t v19, uint8_t v20, uint8_t v21, uint8_t v22, uint8_t v23,
+      uint8_t v24, uint8_t v25, uint8_t v26, uint8_t v27, uint8_t v28, uint8_t v29, uint8_t v30, uint8_t v31
+    ) : simd8(_mm256_setr_epi8(
+      v0, v1, v2, v3, v4, v5, v6, v7,
+      v8, v9, v10,v11,v12,v13,v14,v15,
+      v16,v17,v18,v19,v20,v21,v22,v23,
+      v24,v25,v26,v27,v28,v29,v30,v31
+    )) {}
+    // Repeat 16 values as many times as necessary (usually for lookup tables)
+    simdjson_inline static simd8<uint8_t> repeat_16(
+      uint8_t v0,  uint8_t v1,  uint8_t v2,  uint8_t v3,  uint8_t v4,  uint8_t v5,  uint8_t v6,  uint8_t v7,
+      uint8_t v8,  uint8_t v9,  uint8_t v10, uint8_t v11, uint8_t v12, uint8_t v13, uint8_t v14, uint8_t v15
+    ) {
+      return simd8<uint8_t>(
+        v0, v1, v2, v3, v4, v5, v6, v7,
+        v8, v9, v10,v11,v12,v13,v14,v15,
+        v0, v1, v2, v3, v4, v5, v6, v7,
+        v8, v9, v10,v11,v12,v13,v14,v15
+      );
+    }
+
+    // Saturated math
+    simdjson_inline simd8<uint8_t> saturating_add(const simd8<uint8_t> other) const { return _mm256_adds_epu8(*this, other); }
+    simdjson_inline simd8<uint8_t> saturating_sub(const simd8<uint8_t> other) const { return _mm256_subs_epu8(*this, other); }
+
+    // Order-specific operations
+    simdjson_inline simd8<uint8_t> max_val(const simd8<uint8_t> other) const { return _mm256_max_epu8(*this, other); }
+    simdjson_inline simd8<uint8_t> min_val(const simd8<uint8_t> other) const { return _mm256_min_epu8(other, *this); }
+    // Same as >, but only guarantees true is nonzero (< guarantees true = -1)
+    simdjson_inline simd8<uint8_t> gt_bits(const simd8<uint8_t> other) const { return this->saturating_sub(other); }
+    // Same as <, but only guarantees true is nonzero (< guarantees true = -1)
+    simdjson_inline simd8<uint8_t> lt_bits(const simd8<uint8_t> other) const { return other.saturating_sub(*this); }
+    simdjson_inline simd8<bool> operator<=(const simd8<uint8_t> other) const { return other.max_val(*this) == other; }
+    simdjson_inline simd8<bool> operator>=(const simd8<uint8_t> other) const { return other.min_val(*this) == other; }
+    simdjson_inline simd8<bool> operator>(const simd8<uint8_t> other) const { return this->gt_bits(other).any_bits_set(); }
+    simdjson_inline simd8<bool> operator<(const simd8<uint8_t> other) const { return this->lt_bits(other).any_bits_set(); }
+
+    // Bit-specific operations
+    simdjson_inline simd8<bool> bits_not_set() const { return *this == uint8_t(0); }
+    simdjson_inline simd8<bool> bits_not_set(simd8<uint8_t> bits) const { return (*this & bits).bits_not_set(); }
+    simdjson_inline simd8<bool> any_bits_set() const { return ~this->bits_not_set(); }
+    simdjson_inline simd8<bool> any_bits_set(simd8<uint8_t> bits) const { return ~this->bits_not_set(bits); }
+    simdjson_inline bool is_ascii() const { return _mm256_movemask_epi8(*this) == 0; }
+    simdjson_inline bool bits_not_set_anywhere() const { return _mm256_testz_si256(*this, *this); }
+    simdjson_inline bool any_bits_set_anywhere() const { return !bits_not_set_anywhere(); }
+    simdjson_inline bool bits_not_set_anywhere(simd8<uint8_t> bits) const { return _mm256_testz_si256(*this, bits); }
+    simdjson_inline bool any_bits_set_anywhere(simd8<uint8_t> bits) const { return !bits_not_set_anywhere(bits); }
+    template<int N>
+    simdjson_inline simd8<uint8_t> shr() const { return simd8<uint8_t>(_mm256_srli_epi16(*this, N)) & uint8_t(0xFFu >> N); }
+    template<int N>
+    simdjson_inline simd8<uint8_t> shl() const { return simd8<uint8_t>(_mm256_slli_epi16(*this, N)) & uint8_t(0xFFu << N); }
+    // Get one of the bits and make a bitmask out of it.
+    // e.g. value.get_bit<7>() gets the high bit
+    template<int N>
+    simdjson_inline int get_bit() const { return _mm256_movemask_epi8(_mm256_slli_epi16(*this, 7-N)); }
+  };
+
+  template<typename T>
+  struct simd8x64 {
+    static constexpr int NUM_CHUNKS = 64 / sizeof(simd8<T>);
+    static_assert(NUM_CHUNKS == 2, "Haswell kernel should use two registers per 64-byte block.");
+    const simd8<T> chunks[NUM_CHUNKS];
+
+    simd8x64(const simd8x64<T>& o) = delete; // no copy allowed
+    simd8x64<T>& operator=(const simd8<T>& other) = delete; // no assignment allowed
+    simd8x64() = delete; // no default constructor allowed
+
+    simdjson_inline simd8x64(const simd8<T> chunk0, const simd8<T> chunk1) : chunks{chunk0, chunk1} {}
+    simdjson_inline simd8x64(const T ptr[64]) : chunks{simd8<T>::load(ptr), simd8<T>::load(ptr+32)} {}
+
+    simdjson_inline uint64_t compress(uint64_t mask, T * output) const {
+      uint32_t mask1 = uint32_t(mask);
+      uint32_t mask2 = uint32_t(mask >> 32);
+      this->chunks[0].compress(mask1, output);
+      this->chunks[1].compress(mask2, output + 32 - count_ones(mask1));
+      return 64 - count_ones(mask);
+    }
+
+    simdjson_inline void store(T ptr[64]) const {
+      this->chunks[0].store(ptr+sizeof(simd8<T>)*0);
+      this->chunks[1].store(ptr+sizeof(simd8<T>)*1);
+    }
+
+    simdjson_inline uint64_t to_bitmask() const {
+      uint64_t r_lo = uint32_t(this->chunks[0].to_bitmask());
+      uint64_t r_hi =                       this->chunks[1].to_bitmask();
+      return r_lo | (r_hi << 32);
+    }
+
+    simdjson_inline simd8<T> reduce_or() const {
+      return this->chunks[0] | this->chunks[1];
+    }
+
+    simdjson_inline simd8x64<T> bit_or(const T m) const {
+      const simd8<T> mask = simd8<T>::splat(m);
+      return simd8x64<T>(
+        this->chunks[0] | mask,
+        this->chunks[1] | mask
+      );
+    }
+
+    simdjson_inline uint64_t eq(const T m) const {
+      const simd8<T> mask = simd8<T>::splat(m);
+      return  simd8x64<bool>(
+        this->chunks[0] == mask,
+        this->chunks[1] == mask
+      ).to_bitmask();
+    }
+
+    simdjson_inline uint64_t eq(const simd8x64<uint8_t> &other) const {
+      return  simd8x64<bool>(
+        this->chunks[0] == other.chunks[0],
+        this->chunks[1] == other.chunks[1]
+      ).to_bitmask();
+    }
+
+    simdjson_inline uint64_t lteq(const T m) const {
+      const simd8<T> mask = simd8<T>::splat(m);
+      return  simd8x64<bool>(
+        this->chunks[0] <= mask,
+        this->chunks[1] <= mask
+      ).to_bitmask();
+    }
+  }; // struct simd8x64<T>
+
+} // namespace simd
+
+} // unnamed namespace
+} // namespace haswell
+} // namespace simdjson
+
+#endif // SIMDJSON_HASWELL_SIMD_H
diff --git a/contrib/libs/simdjson/include/simdjson/haswell/stringparsing_defs.h b/contrib/libs/simdjson/include/simdjson/haswell/stringparsing_defs.h
new file mode 100644
index 00000000000..f896a10e2c1
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/haswell/stringparsing_defs.h
@@ -0,0 +1,48 @@
+#ifndef SIMDJSON_HASWELL_STRINGPARSING_DEFS_H
+#define SIMDJSON_HASWELL_STRINGPARSING_DEFS_H
+
+#ifndef SIMDJSON_CONDITIONAL_INCLUDE
+#include "simdjson/haswell/base.h"
+#include "simdjson/haswell/simd.h"
+#include "simdjson/haswell/bitmanipulation.h"
+#endif // SIMDJSON_CONDITIONAL_INCLUDE
+
+namespace simdjson {
+namespace haswell {
+namespace {
+
+using namespace simd;
+
+// Holds backslashes and quotes locations.
+struct backslash_and_quote {
+public:
+  static constexpr uint32_t BYTES_PROCESSED = 32;
+  simdjson_inline static backslash_and_quote copy_and_find(const uint8_t *src, uint8_t *dst);
+
+  simdjson_inline bool has_quote_first() { return ((bs_bits - 1) & quote_bits) != 0; }
+  simdjson_inline bool has_backslash() { return ((quote_bits - 1) & bs_bits) != 0; }
+  simdjson_inline int quote_index() { return trailing_zeroes(quote_bits); }
+  simdjson_inline int backslash_index() { return trailing_zeroes(bs_bits); }
+
+  uint32_t bs_bits;
+  uint32_t quote_bits;
+}; // struct backslash_and_quote
+
+simdjson_inline backslash_and_quote backslash_and_quote::copy_and_find(const uint8_t *src, uint8_t *dst) {
+  // this can read up to 15 bytes beyond the buffer size, but we require
+  // SIMDJSON_PADDING of padding
+  static_assert(SIMDJSON_PADDING >= (BYTES_PROCESSED - 1), "backslash and quote finder must process fewer than SIMDJSON_PADDING bytes");
+  simd8<uint8_t> v(src);
+  // store to dest unconditionally - we can overwrite the bits we don't like later
+  v.store(dst);
+  return {
+      static_cast<uint32_t>((v == '\\').to_bitmask()),     // bs_bits
+      static_cast<uint32_t>((v == '"').to_bitmask()), // quote_bits
+  };
+}
+
+} // unnamed namespace
+} // namespace haswell
+} // namespace simdjson
+
+#endif // SIMDJSON_HASWELL_STRINGPARSING_DEFS_H
diff --git a/contrib/libs/simdjson/include/simdjson/icelake.h b/contrib/libs/simdjson/include/simdjson/icelake.h
new file mode 100644
index 00000000000..96429603406
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/icelake.h
@@ -0,0 +1,8 @@
+#ifndef SIMDJSON_ICELAKE_H
+#define SIMDJSON_ICELAKE_H
+
+#include "simdjson/icelake/begin.h"
+#include "simdjson/generic/amalgamated.h"
+#include "simdjson/icelake/end.h"
+
+#endif // SIMDJSON_ICELAKE_H
\ No newline at end of file
diff --git a/contrib/libs/simdjson/include/simdjson/icelake/base.h b/contrib/libs/simdjson/include/simdjson/icelake/base.h
new file mode 100644
index 00000000000..f44c0d32a1f
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/icelake/base.h
@@ -0,0 +1,20 @@
+#ifndef SIMDJSON_ICELAKE_BASE_H
+#define SIMDJSON_ICELAKE_BASE_H
+
+#ifndef SIMDJSON_CONDITIONAL_INCLUDE
+#include "simdjson/base.h"
+#endif // SIMDJSON_CONDITIONAL_INCLUDE
+
+// The constructor may be executed on any host, so we take care not to use SIMDJSON_TARGET_ICELAKE
+namespace simdjson {
+/**
+ * Implementation for Icelake (Intel AVX512).
+ */
+namespace icelake {
+
+class implementation;
+
+} // namespace icelake
+} // namespace simdjson
+
+#endif // SIMDJSON_ICELAKE_BASE_H
diff --git a/contrib/libs/simdjson/include/simdjson/icelake/begin.h b/contrib/libs/simdjson/include/simdjson/icelake/begin.h
new file mode 100644
index 00000000000..60fe2cd6bf9
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/icelake/begin.h
@@ -0,0 +1,13 @@
+#define SIMDJSON_IMPLEMENTATION icelake
+#include "simdjson/icelake/base.h"
+#include "simdjson/icelake/intrinsics.h"
+
+#if !SIMDJSON_CAN_ALWAYS_RUN_ICELAKE
+SIMDJSON_TARGET_REGION("avx512f,avx512dq,avx512cd,avx512bw,avx512vbmi,avx512vbmi2,avx512vl,avx2,bmi,pclmul,lzcnt,popcnt")
+#endif
+
+#include "simdjson/icelake/bitmanipulation.h"
+#include "simdjson/icelake/bitmask.h"
+#include "simdjson/icelake/simd.h"
+#include "simdjson/icelake/stringparsing_defs.h"
+#include "simdjson/icelake/numberparsing_defs.h"
diff --git a/contrib/libs/simdjson/include/simdjson/icelake/bitmanipulation.h b/contrib/libs/simdjson/include/simdjson/icelake/bitmanipulation.h
new file mode 100644
index 00000000000..5bcf7116015
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/icelake/bitmanipulation.h
@@ -0,0 +1,70 @@
+#ifndef SIMDJSON_ICELAKE_BITMANIPULATION_H
+#define SIMDJSON_ICELAKE_BITMANIPULATION_H
+
+#ifndef SIMDJSON_CONDITIONAL_INCLUDE
+#include "simdjson/icelake/base.h"
+#include "simdjson/icelake/intrinsics.h"
+#endif // SIMDJSON_CONDITIONAL_INCLUDE
+
+namespace simdjson {
+namespace icelake {
+namespace {
+
+// We sometimes call trailing_zero on inputs that are zero,
+// but the algorithms do not end up using the returned value.
+// Sadly, sanitizers are not smart enough to figure it out.
+SIMDJSON_NO_SANITIZE_UNDEFINED
+// This function can be used safely even if not all bytes have been
+// initialized.
+// See issue https://github.com/simdjson/simdjson/issues/1965
+SIMDJSON_NO_SANITIZE_MEMORY
+simdjson_inline int trailing_zeroes(uint64_t input_num) {
+#if SIMDJSON_REGULAR_VISUAL_STUDIO
+  return (int)_tzcnt_u64(input_num);
+#else // SIMDJSON_REGULAR_VISUAL_STUDIO
+  ////////
+  // You might expect the next line to be equivalent to
+  // return (int)_tzcnt_u64(input_num);
+  // but the generated code differs and might be less efficient?
+  ////////
+  return __builtin_ctzll(input_num);
+#endif // SIMDJSON_REGULAR_VISUAL_STUDIO
+}
+
+/* result might be undefined when input_num is zero */
+simdjson_inline uint64_t clear_lowest_bit(uint64_t input_num) {
+  return _blsr_u64(input_num);
+}
+
+/* result might be undefined when input_num is zero */
+simdjson_inline int leading_zeroes(uint64_t input_num) {
+  return int(_lzcnt_u64(input_num));
+}
+
+#if SIMDJSON_REGULAR_VISUAL_STUDIO
+simdjson_inline unsigned __int64 count_ones(uint64_t input_num) {
+  // note: we do not support legacy 32-bit Windows
+  return __popcnt64(input_num);// Visual Studio wants two underscores
+}
+#else
+simdjson_inline long long int count_ones(uint64_t input_num) {
+  return _popcnt64(input_num);
+}
+#endif
+
+simdjson_inline bool add_overflow(uint64_t value1, uint64_t value2,
+                                uint64_t *result) {
+#if SIMDJSON_REGULAR_VISUAL_STUDIO
+  return _addcarry_u64(0, value1, value2,
+                       reinterpret_cast<unsigned __int64 *>(result));
+#else
+  return __builtin_uaddll_overflow(value1, value2,
+                                   reinterpret_cast<unsigned long long *>(result));
+#endif
+}
+
+} // unnamed namespace
+} // namespace icelake
+} // namespace simdjson
+
+#endif // SIMDJSON_ICELAKE_BITMANIPULATION_H
diff --git a/contrib/libs/simdjson/include/simdjson/icelake/bitmask.h b/contrib/libs/simdjson/include/simdjson/icelake/bitmask.h
new file mode 100644
index 00000000000..ed55962fa49
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/icelake/bitmask.h
@@ -0,0 +1,30 @@
+#ifndef SIMDJSON_ICELAKE_BITMASK_H
+#define SIMDJSON_ICELAKE_BITMASK_H
+
+#ifndef SIMDJSON_CONDITIONAL_INCLUDE
+#include "simdjson/icelake/base.h"
+#include "simdjson/icelake/intrinsics.h"
+#endif // SIMDJSON_CONDITIONAL_INCLUDE
+
+namespace simdjson {
+namespace icelake {
+namespace {
+
+//
+// Perform a "cumulative bitwise xor," flipping bits each time a 1 is encountered.
+//
+// For example, prefix_xor(00100100) == 00011100
+//
+simdjson_inline uint64_t prefix_xor(const uint64_t bitmask) {
+  // There should be no such thing with a processor supporting avx2
+  // but not clmul.
+  __m128i all_ones = _mm_set1_epi8('\xFF');
+  __m128i result = _mm_clmulepi64_si128(_mm_set_epi64x(0ULL, bitmask), all_ones, 0);
+  return _mm_cvtsi128_si64(result);
+}
+
+} // unnamed namespace
+} // namespace icelake
+} // namespace simdjson
+
+#endif // SIMDJSON_ICELAKE_BITMASK_H
diff --git a/contrib/libs/simdjson/include/simdjson/icelake/end.h b/contrib/libs/simdjson/include/simdjson/icelake/end.h
new file mode 100644
index 00000000000..2accd5fb0d5
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/icelake/end.h
@@ -0,0 +1,9 @@
+#ifndef SIMDJSON_CONDITIONAL_INCLUDE
+#include "simdjson/icelake/base.h"
+#endif // SIMDJSON_CONDITIONAL_INCLUDE
+
+#if !SIMDJSON_CAN_ALWAYS_RUN_ICELAKE
+SIMDJSON_UNTARGET_REGION
+#endif
+
+#undef SIMDJSON_IMPLEMENTATION
diff --git a/contrib/libs/simdjson/include/simdjson/icelake/implementation.h b/contrib/libs/simdjson/include/simdjson/icelake/implementation.h
new file mode 100644
index 00000000000..940c5f9923c
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/icelake/implementation.h
@@ -0,0 +1,36 @@
+#ifndef SIMDJSON_ICELAKE_IMPLEMENTATION_H
+#define SIMDJSON_ICELAKE_IMPLEMENTATION_H
+
+#ifndef SIMDJSON_CONDITIONAL_INCLUDE
+#include "simdjson/icelake/base.h"
+#include "simdjson/implementation.h"
+#include "simdjson/internal/instruction_set.h"
+#endif // SIMDJSON_CONDITIONAL_INCLUDE
+
+// The constructor may be executed on any host, so we take care not to use SIMDJSON_TARGET_ICELAKE
+namespace simdjson {
+namespace icelake {
+
+/**
+ * @private
+ */
+class implementation final : public simdjson::implementation {
+public:
+  simdjson_inline implementation() : simdjson::implementation(
+      "icelake",
+      "Intel/AMD AVX512",
+      internal::instruction_set::AVX2 | internal::instruction_set::PCLMULQDQ | internal::instruction_set::BMI1 | internal::instruction_set::BMI2 | internal::instruction_set::AVX512F | internal::instruction_set::AVX512DQ | internal::instruction_set::AVX512CD | internal::instruction_set::AVX512BW | internal::instruction_set::AVX512VL | internal::instruction_set::AVX512VBMI2
+  ) {}
+  simdjson_warn_unused error_code create_dom_parser_implementation(
+    size_t capacity,
+    size_t max_length,
+    std::unique_ptr<internal::dom_parser_implementation>& dst
+  ) const noexcept final;
+  simdjson_warn_unused error_code minify(const uint8_t *buf, size_t len, uint8_t *dst, size_t &dst_len) const noexcept final;
+  simdjson_warn_unused bool validate_utf8(const char *buf, size_t len) const noexcept final;
+};
+
+} // namespace icelake
+} // namespace simdjson
+
+#endif // SIMDJSON_ICELAKE_IMPLEMENTATION_H
diff --git a/contrib/libs/simdjson/include/simdjson/icelake/intrinsics.h b/contrib/libs/simdjson/include/simdjson/icelake/intrinsics.h
new file mode 100644
index 00000000000..4c68c290c1e
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/icelake/intrinsics.h
@@ -0,0 +1,60 @@
+#ifndef SIMDJSON_ICELAKE_INTRINSICS_H
+#define SIMDJSON_ICELAKE_INTRINSICS_H
+
+#ifndef SIMDJSON_CONDITIONAL_INCLUDE
+#include "simdjson/icelake/base.h"
+#endif // SIMDJSON_CONDITIONAL_INCLUDE
+
+#if SIMDJSON_VISUAL_STUDIO
+// under clang within visual studio, this will include <x86intrin.h>
+#include <intrin.h>  // visual studio or clang
+#else
+#include <x86intrin.h> // elsewhere
+#endif // SIMDJSON_VISUAL_STUDIO
+
+#if SIMDJSON_CLANG_VISUAL_STUDIO
+/**
+ * You are not supposed, normally, to include these
+ * headers directly. Instead you should either include intrin.h
+ * or x86intrin.h. However, when compiling with clang
+ * under Windows (i.e., when _MSC_VER is set), these headers
+ * only get included *if* the corresponding features are detected
+ * from macros:
+ * e.g., if __AVX2__ is set... in turn,  we normally set these
+ * macros by compiling against the corresponding architecture
+ * (e.g., arch:AVX2, -mavx2, etc.) which compiles the whole
+ * software with these advanced instructions. In simdjson, we
+ * want to compile the whole program for a generic target,
+ * and only target our specific kernels. As a workaround,
+ * we directly include the needed headers. These headers would
+ * normally guard against such usage, but we carefully included
+ * <x86intrin.h>  (or <intrin.h>) before, so the headers
+ * are fooled.
+ */
+#include <bmiintrin.h>   // for _blsr_u64
+#include <lzcntintrin.h> // for  __lzcnt64
+#include <immintrin.h>   // for most things (AVX2, AVX512, _popcnt64)
+#include <smmintrin.h>
+#include <tmmintrin.h>
+#include <avxintrin.h>
+#include <avx2intrin.h>
+#include <wmmintrin.h>   // for  _mm_clmulepi64_si128
+// Important: we need the AVX-512 headers:
+#include <avx512fintrin.h>
+#include <avx512dqintrin.h>
+#include <avx512cdintrin.h>
+#include <avx512bwintrin.h>
+#include <avx512vlintrin.h>
+#include <avx512vbmiintrin.h>
+#include <avx512vbmi2intrin.h>
+// unfortunately, we may not get _blsr_u64, but, thankfully, clang
+// has it as a macro.
+#ifndef _blsr_u64
+// we roll our own
+#define _blsr_u64(n) ((n - 1) & n)
+#endif //  _blsr_u64
+#endif // SIMDJSON_CLANG_VISUAL_STUDIO
+
+static_assert(sizeof(__m512i) <= simdjson::SIMDJSON_PADDING, "insufficient padding for icelake");
+
+#endif // SIMDJSON_ICELAKE_INTRINSICS_H
diff --git a/contrib/libs/simdjson/include/simdjson/icelake/numberparsing_defs.h b/contrib/libs/simdjson/include/simdjson/icelake/numberparsing_defs.h
new file mode 100644
index 00000000000..b095cf6c66c
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/icelake/numberparsing_defs.h
@@ -0,0 +1,57 @@
+#ifndef SIMDJSON_ICELAKE_NUMBERPARSING_DEFS_H
+#define SIMDJSON_ICELAKE_NUMBERPARSING_DEFS_H
+
+#ifndef SIMDJSON_CONDITIONAL_INCLUDE
+#include "simdjson/icelake/base.h"
+#include "simdjson/icelake/intrinsics.h"
+#include "simdjson/internal/numberparsing_tables.h"
+#endif // SIMDJSON_CONDITIONAL_INCLUDE
+
+namespace simdjson {
+namespace icelake {
+namespace numberparsing {
+
+static simdjson_inline uint32_t parse_eight_digits_unrolled(const uint8_t *chars) {
+  // this actually computes *16* values so we are being wasteful.
+  const __m128i ascii0 = _mm_set1_epi8('0');
+  const __m128i mul_1_10 =
+      _mm_setr_epi8(10, 1, 10, 1, 10, 1, 10, 1, 10, 1, 10, 1, 10, 1, 10, 1);
+  const __m128i mul_1_100 = _mm_setr_epi16(100, 1, 100, 1, 100, 1, 100, 1);
+  const __m128i mul_1_10000 =
+      _mm_setr_epi16(10000, 1, 10000, 1, 10000, 1, 10000, 1);
+  const __m128i input = _mm_sub_epi8(
+      _mm_loadu_si128(reinterpret_cast<const __m128i *>(chars)), ascii0);
+  const __m128i t1 = _mm_maddubs_epi16(input, mul_1_10);
+  const __m128i t2 = _mm_madd_epi16(t1, mul_1_100);
+  const __m128i t3 = _mm_packus_epi32(t2, t2);
+  const __m128i t4 = _mm_madd_epi16(t3, mul_1_10000);
+  return _mm_cvtsi128_si32(
+      t4); // only captures the sum of the first 8 digits, drop the rest
+}
+
+/** @private */
+simdjson_inline internal::value128 full_multiplication(uint64_t value1, uint64_t value2) {
+  internal::value128 answer;
+#if SIMDJSON_REGULAR_VISUAL_STUDIO || SIMDJSON_IS_32BITS
+#if SIMDJSON_IS_ARM64
+  // ARM64 has native support for 64-bit multiplications, no need to emultate
+  answer.high = __umulh(value1, value2);
+  answer.low = value1 * value2;
+#else
+  answer.low = _umul128(value1, value2, &answer.high); // _umul128 not available on ARM64
+#endif // SIMDJSON_IS_ARM64
+#else // SIMDJSON_REGULAR_VISUAL_STUDIO || SIMDJSON_IS_32BITS
+  __uint128_t r = (static_cast<__uint128_t>(value1)) * value2;
+  answer.low = uint64_t(r);
+  answer.high = uint64_t(r >> 64);
+#endif
+  return answer;
+}
+
+} // namespace numberparsing
+} // namespace icelake
+} // namespace simdjson
+
+#define SIMDJSON_SWAR_NUMBER_PARSING 1
+
+#endif // SIMDJSON_ICELAKE_NUMBERPARSING_DEFS_H
diff --git a/contrib/libs/simdjson/include/simdjson/icelake/simd.h b/contrib/libs/simdjson/include/simdjson/icelake/simd.h
new file mode 100644
index 00000000000..04203f4b9a5
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/icelake/simd.h
@@ -0,0 +1,372 @@
+#ifndef SIMDJSON_ICELAKE_SIMD_H
+#define SIMDJSON_ICELAKE_SIMD_H
+
+#ifndef SIMDJSON_CONDITIONAL_INCLUDE
+#include "simdjson/icelake/base.h"
+#include "simdjson/icelake/intrinsics.h"
+#include "simdjson/icelake/bitmanipulation.h"
+#include "simdjson/internal/simdprune_tables.h"
+#endif // SIMDJSON_CONDITIONAL_INCLUDE
+
+#if defined(__GNUC__) && !defined(__clang__)
+#if __GNUC__ == 8
+#define SIMDJSON_GCC8 1
+#endif //  __GNUC__ == 8
+#endif // defined(__GNUC__) && !defined(__clang__)
+
+#if SIMDJSON_GCC8
+/**
+ * GCC 8 fails to provide _mm512_set_epi8. We roll our own.
+ */
+inline __m512i _mm512_set_epi8(uint8_t a0, uint8_t a1, uint8_t a2, uint8_t a3, uint8_t a4, uint8_t a5, uint8_t a6, uint8_t a7, uint8_t a8, uint8_t a9, uint8_t a10, uint8_t a11, uint8_t a12, uint8_t a13, uint8_t a14, uint8_t a15, uint8_t a16, uint8_t a17, uint8_t a18, uint8_t a19, uint8_t a20, uint8_t a21, uint8_t a22, uint8_t a23, uint8_t a24, uint8_t a25, uint8_t a26, uint8_t a27, uint8_t a28, uint8_t a29, uint8_t a30, uint8_t a31, uint8_t a32, uint8_t a33, uint8_t a34, uint8_t a35, uint8_t a36, uint8_t a37, uint8_t a38, uint8_t a39, uint8_t a40, uint8_t a41, uint8_t a42, uint8_t a43, uint8_t a44, uint8_t a45, uint8_t a46, uint8_t a47, uint8_t a48, uint8_t a49, uint8_t a50, uint8_t a51, uint8_t a52, uint8_t a53, uint8_t a54, uint8_t a55, uint8_t a56, uint8_t a57, uint8_t a58, uint8_t a59, uint8_t a60, uint8_t a61, uint8_t a62, uint8_t a63) {
+  return _mm512_set_epi64(uint64_t(a7) + (uint64_t(a6) << 8) + (uint64_t(a5) << 16) + (uint64_t(a4) << 24) + (uint64_t(a3) << 32) + (uint64_t(a2) << 40) + (uint64_t(a1) << 48) + (uint64_t(a0) << 56),
+                          uint64_t(a15) + (uint64_t(a14) << 8) + (uint64_t(a13) << 16) + (uint64_t(a12) << 24) + (uint64_t(a11) << 32) + (uint64_t(a10) << 40) + (uint64_t(a9) << 48) + (uint64_t(a8) << 56),
+                          uint64_t(a23) + (uint64_t(a22) << 8) + (uint64_t(a21) << 16) + (uint64_t(a20) << 24) + (uint64_t(a19) << 32) + (uint64_t(a18) << 40) + (uint64_t(a17) << 48) + (uint64_t(a16) << 56),
+                          uint64_t(a31) + (uint64_t(a30) << 8) + (uint64_t(a29) << 16) + (uint64_t(a28) << 24) + (uint64_t(a27) << 32) + (uint64_t(a26) << 40) + (uint64_t(a25) << 48) + (uint64_t(a24) << 56),
+                          uint64_t(a39) + (uint64_t(a38) << 8) + (uint64_t(a37) << 16) + (uint64_t(a36) << 24) + (uint64_t(a35) << 32) + (uint64_t(a34) << 40) + (uint64_t(a33) << 48) + (uint64_t(a32) << 56),
+                          uint64_t(a47) + (uint64_t(a46) << 8) + (uint64_t(a45) << 16) + (uint64_t(a44) << 24) + (uint64_t(a43) << 32) + (uint64_t(a42) << 40) + (uint64_t(a41) << 48) + (uint64_t(a40) << 56),
+                          uint64_t(a55) + (uint64_t(a54) << 8) + (uint64_t(a53) << 16) + (uint64_t(a52) << 24) + (uint64_t(a51) << 32) + (uint64_t(a50) << 40) + (uint64_t(a49) << 48) + (uint64_t(a48) << 56),
+                          uint64_t(a63) + (uint64_t(a62) << 8) + (uint64_t(a61) << 16) + (uint64_t(a60) << 24) + (uint64_t(a59) << 32) + (uint64_t(a58) << 40) + (uint64_t(a57) << 48) + (uint64_t(a56) << 56));
+}
+#endif // SIMDJSON_GCC8
+
+
+
+namespace simdjson {
+namespace icelake {
+namespace {
+namespace simd {
+
+  // Forward-declared so they can be used by splat and friends.
+  template<typename Child>
+  struct base {
+    __m512i value;
+
+    // Zero constructor
+    simdjson_inline base() : value{__m512i()} {}
+
+    // Conversion from SIMD register
+    simdjson_inline base(const __m512i _value) : value(_value) {}
+
+    // Conversion to SIMD register
+    simdjson_inline operator const __m512i&() const { return this->value; }
+    simdjson_inline operator __m512i&() { return this->value; }
+
+    // Bit operations
+    simdjson_inline Child operator|(const Child other) const { return _mm512_or_si512(*this, other); }
+    simdjson_inline Child operator&(const Child other) const { return _mm512_and_si512(*this, other); }
+    simdjson_inline Child operator^(const Child other) const { return _mm512_xor_si512(*this, other); }
+    simdjson_inline Child bit_andnot(const Child other) const { return _mm512_andnot_si512(other, *this); }
+    simdjson_inline Child& operator|=(const Child other) { auto this_cast = static_cast<Child*>(this); *this_cast = *this_cast | other; return *this_cast; }
+    simdjson_inline Child& operator&=(const Child other) { auto this_cast = static_cast<Child*>(this); *this_cast = *this_cast & other; return *this_cast; }
+    simdjson_inline Child& operator^=(const Child other) { auto this_cast = static_cast<Child*>(this); *this_cast = *this_cast ^ other; return *this_cast; }
+  };
+
+  // Forward-declared so they can be used by splat and friends.
+  template<typename T>
+  struct simd8;
+
+  template<typename T, typename Mask=simd8<bool>>
+  struct base8: base<simd8<T>> {
+    typedef uint32_t bitmask_t;
+    typedef uint64_t bitmask2_t;
+
+    simdjson_inline base8() : base<simd8<T>>() {}
+    simdjson_inline base8(const __m512i _value) : base<simd8<T>>(_value) {}
+
+    friend simdjson_really_inline uint64_t operator==(const simd8<T> lhs, const simd8<T> rhs) {
+      return _mm512_cmpeq_epi8_mask(lhs, rhs);
+    }
+
+    static const int SIZE = sizeof(base<T>::value);
+
+    template<int N=1>
+    simdjson_inline simd8<T> prev(const simd8<T> prev_chunk) const {
+     // workaround for compilers unable to figure out that 16 - N is a constant (GCC 8)
+      constexpr int shift = 16 - N;
+      return _mm512_alignr_epi8(*this, _mm512_permutex2var_epi64(prev_chunk, _mm512_set_epi64(13, 12, 11, 10, 9, 8, 7, 6), *this), shift);
+    }
+  };
+
+  // SIMD byte mask type (returned by things like eq and gt)
+  template<>
+  struct simd8<bool>: base8<bool> {
+    static simdjson_inline simd8<bool> splat(bool _value) { return _mm512_set1_epi8(uint8_t(-(!!_value))); }
+
+    simdjson_inline simd8() : base8() {}
+    simdjson_inline simd8(const __m512i _value) : base8<bool>(_value) {}
+    // Splat constructor
+    simdjson_inline simd8(bool _value) : base8<bool>(splat(_value)) {}
+    simdjson_inline bool any() const { return !!_mm512_test_epi8_mask (*this, *this); }
+    simdjson_inline simd8<bool> operator~() const { return *this ^ true; }
+  };
+
+  template<typename T>
+  struct base8_numeric: base8<T> {
+    static simdjson_inline simd8<T> splat(T _value) { return _mm512_set1_epi8(_value); }
+    static simdjson_inline simd8<T> zero() { return _mm512_setzero_si512(); }
+    static simdjson_inline simd8<T> load(const T values[64]) {
+      return _mm512_loadu_si512(reinterpret_cast<const __m512i *>(values));
+    }
+    // Repeat 16 values as many times as necessary (usually for lookup tables)
+    static simdjson_inline simd8<T> repeat_16(
+      T v0,  T v1,  T v2,  T v3,  T v4,  T v5,  T v6,  T v7,
+      T v8,  T v9,  T v10, T v11, T v12, T v13, T v14, T v15
+    ) {
+      return simd8<T>(
+        v0, v1, v2, v3, v4, v5, v6, v7,
+        v8, v9, v10,v11,v12,v13,v14,v15,
+        v0, v1, v2, v3, v4, v5, v6, v7,
+        v8, v9, v10,v11,v12,v13,v14,v15,
+        v0, v1, v2, v3, v4, v5, v6, v7,
+        v8, v9, v10,v11,v12,v13,v14,v15,
+        v0, v1, v2, v3, v4, v5, v6, v7,
+        v8, v9, v10,v11,v12,v13,v14,v15
+      );
+    }
+
+    simdjson_inline base8_numeric() : base8<T>() {}
+    simdjson_inline base8_numeric(const __m512i _value) : base8<T>(_value) {}
+
+    // Store to array
+    simdjson_inline void store(T dst[64]) const { return _mm512_storeu_si512(reinterpret_cast<__m512i *>(dst), *this); }
+
+    // Addition/subtraction are the same for signed and unsigned
+    simdjson_inline simd8<T> operator+(const simd8<T> other) const { return _mm512_add_epi8(*this, other); }
+    simdjson_inline simd8<T> operator-(const simd8<T> other) const { return _mm512_sub_epi8(*this, other); }
+    simdjson_inline simd8<T>& operator+=(const simd8<T> other) { *this = *this + other; return *static_cast<simd8<T>*>(this); }
+    simdjson_inline simd8<T>& operator-=(const simd8<T> other) { *this = *this - other; return *static_cast<simd8<T>*>(this); }
+
+    // Override to distinguish from bool version
+    simdjson_inline simd8<T> operator~() const { return *this ^ 0xFFu; }
+
+    // Perform a lookup assuming the value is between 0 and 16 (undefined behavior for out of range values)
+    template<typename L>
+    simdjson_inline simd8<L> lookup_16(simd8<L> lookup_table) const {
+      return _mm512_shuffle_epi8(lookup_table, *this);
+    }
+
+    // Copies to 'output" all bytes corresponding to a 0 in the mask (interpreted as a bitset).
+    // Passing a 0 value for mask would be equivalent to writing out every byte to output.
+    // Only the first 32 - count_ones(mask) bytes of the result are significant but 32 bytes
+    // get written.
+    // Design consideration: it seems like a function with the
+    // signature simd8<L> compress(uint32_t mask) would be
+    // sensible, but the AVX ISA makes this kind of approach difficult.
+    template<typename L>
+    simdjson_inline void compress(uint64_t mask, L * output) const {
+      _mm512_mask_compressstoreu_epi8 (output,~mask,*this);
+    }
+
+    template<typename L>
+    simdjson_inline simd8<L> lookup_16(
+        L replace0,  L replace1,  L replace2,  L replace3,
+        L replace4,  L replace5,  L replace6,  L replace7,
+        L replace8,  L replace9,  L replace10, L replace11,
+        L replace12, L replace13, L replace14, L replace15) const {
+      return lookup_16(simd8<L>::repeat_16(
+        replace0,  replace1,  replace2,  replace3,
+        replace4,  replace5,  replace6,  replace7,
+        replace8,  replace9,  replace10, replace11,
+        replace12, replace13, replace14, replace15
+      ));
+    }
+  };
+
+  // Signed bytes
+  template<>
+  struct simd8<int8_t> : base8_numeric<int8_t> {
+    simdjson_inline simd8() : base8_numeric<int8_t>() {}
+    simdjson_inline simd8(const __m512i _value) : base8_numeric<int8_t>(_value) {}
+    // Splat constructor
+    simdjson_inline simd8(int8_t _value) : simd8(splat(_value)) {}
+    // Array constructor
+    simdjson_inline simd8(const int8_t values[64]) : simd8(load(values)) {}
+    // Member-by-member initialization
+    simdjson_inline simd8(
+      int8_t v0,  int8_t v1,  int8_t v2,  int8_t v3,  int8_t v4,  int8_t v5,  int8_t v6,  int8_t v7,
+      int8_t v8,  int8_t v9,  int8_t v10, int8_t v11, int8_t v12, int8_t v13, int8_t v14, int8_t v15,
+      int8_t v16, int8_t v17, int8_t v18, int8_t v19, int8_t v20, int8_t v21, int8_t v22, int8_t v23,
+      int8_t v24, int8_t v25, int8_t v26, int8_t v27, int8_t v28, int8_t v29, int8_t v30, int8_t v31,
+      int8_t v32, int8_t v33, int8_t v34, int8_t v35, int8_t v36, int8_t v37, int8_t v38, int8_t v39,
+      int8_t v40, int8_t v41, int8_t v42, int8_t v43, int8_t v44, int8_t v45, int8_t v46, int8_t v47,
+      int8_t v48, int8_t v49, int8_t v50, int8_t v51, int8_t v52, int8_t v53, int8_t v54, int8_t v55,
+      int8_t v56, int8_t v57, int8_t v58, int8_t v59, int8_t v60, int8_t v61, int8_t v62, int8_t v63
+    ) : simd8(_mm512_set_epi8(
+      v63, v62, v61, v60, v59, v58, v57, v56,
+      v55, v54, v53, v52, v51, v50, v49, v48,
+      v47, v46, v45, v44, v43, v42, v41, v40,
+      v39, v38, v37, v36, v35, v34, v33, v32,
+      v31, v30, v29, v28, v27, v26, v25, v24,
+      v23, v22, v21, v20, v19, v18, v17, v16,
+      v15, v14, v13, v12, v11, v10,  v9,  v8,
+       v7,  v6,  v5,  v4,  v3,  v2,  v1,  v0
+    )) {}
+
+    // Repeat 16 values as many times as necessary (usually for lookup tables)
+    simdjson_inline static simd8<int8_t> repeat_16(
+      int8_t v0,  int8_t v1,  int8_t v2,  int8_t v3,  int8_t v4,  int8_t v5,  int8_t v6,  int8_t v7,
+      int8_t v8,  int8_t v9,  int8_t v10, int8_t v11, int8_t v12, int8_t v13, int8_t v14, int8_t v15
+    ) {
+      return simd8<int8_t>(
+        v0, v1, v2, v3, v4, v5, v6, v7,
+        v8, v9, v10,v11,v12,v13,v14,v15,
+        v0, v1, v2, v3, v4, v5, v6, v7,
+        v8, v9, v10,v11,v12,v13,v14,v15,
+        v0, v1, v2, v3, v4, v5, v6, v7,
+        v8, v9, v10,v11,v12,v13,v14,v15,
+        v0, v1, v2, v3, v4, v5, v6, v7,
+        v8, v9, v10,v11,v12,v13,v14,v15
+      );
+    }
+
+    // Order-sensitive comparisons
+    simdjson_inline simd8<int8_t> max_val(const simd8<int8_t> other) const { return _mm512_max_epi8(*this, other); }
+    simdjson_inline simd8<int8_t> min_val(const simd8<int8_t> other) const { return _mm512_min_epi8(*this, other); }
+
+    simdjson_inline simd8<bool> operator>(const simd8<int8_t> other) const { return _mm512_maskz_abs_epi8(_mm512_cmpgt_epi8_mask(*this, other),_mm512_set1_epi8(uint8_t(0x80))); }
+    simdjson_inline simd8<bool> operator<(const simd8<int8_t> other) const { return _mm512_maskz_abs_epi8(_mm512_cmpgt_epi8_mask(other, *this),_mm512_set1_epi8(uint8_t(0x80))); }
+  };
+
+  // Unsigned bytes
+  template<>
+  struct simd8<uint8_t>: base8_numeric<uint8_t> {
+    simdjson_inline simd8() : base8_numeric<uint8_t>() {}
+    simdjson_inline simd8(const __m512i _value) : base8_numeric<uint8_t>(_value) {}
+    // Splat constructor
+    simdjson_inline simd8(uint8_t _value) : simd8(splat(_value)) {}
+    // Array constructor
+    simdjson_inline simd8(const uint8_t values[64]) : simd8(load(values)) {}
+    // Member-by-member initialization
+    simdjson_inline simd8(
+      uint8_t v0,  uint8_t v1,  uint8_t v2,  uint8_t v3,  uint8_t v4,  uint8_t v5,  uint8_t v6,  uint8_t v7,
+      uint8_t v8,  uint8_t v9,  uint8_t v10, uint8_t v11, uint8_t v12, uint8_t v13, uint8_t v14, uint8_t v15,
+      uint8_t v16, uint8_t v17, uint8_t v18, uint8_t v19, uint8_t v20, uint8_t v21, uint8_t v22, uint8_t v23,
+      uint8_t v24, uint8_t v25, uint8_t v26, uint8_t v27, uint8_t v28, uint8_t v29, uint8_t v30, uint8_t v31,
+      uint8_t v32, uint8_t v33, uint8_t v34, uint8_t v35, uint8_t v36, uint8_t v37, uint8_t v38, uint8_t v39,
+      uint8_t v40, uint8_t v41, uint8_t v42, uint8_t v43, uint8_t v44, uint8_t v45, uint8_t v46, uint8_t v47,
+      uint8_t v48, uint8_t v49, uint8_t v50, uint8_t v51, uint8_t v52, uint8_t v53, uint8_t v54, uint8_t v55,
+      uint8_t v56, uint8_t v57, uint8_t v58, uint8_t v59, uint8_t v60, uint8_t v61, uint8_t v62, uint8_t v63
+    ) : simd8(_mm512_set_epi8(
+      v63, v62, v61, v60, v59, v58, v57, v56,
+      v55, v54, v53, v52, v51, v50, v49, v48,
+      v47, v46, v45, v44, v43, v42, v41, v40,
+      v39, v38, v37, v36, v35, v34, v33, v32,
+      v31, v30, v29, v28, v27, v26, v25, v24,
+      v23, v22, v21, v20, v19, v18, v17, v16,
+      v15, v14, v13, v12, v11, v10,  v9,  v8,
+       v7,  v6,  v5,  v4,  v3,  v2,  v1,  v0
+    )) {}
+
+    // Repeat 16 values as many times as necessary (usually for lookup tables)
+    simdjson_inline static simd8<uint8_t> repeat_16(
+      uint8_t v0,  uint8_t v1,  uint8_t v2,  uint8_t v3,  uint8_t v4,  uint8_t v5,  uint8_t v6,  uint8_t v7,
+      uint8_t v8,  uint8_t v9,  uint8_t v10, uint8_t v11, uint8_t v12, uint8_t v13, uint8_t v14, uint8_t v15
+    ) {
+      return simd8<uint8_t>(
+        v0, v1, v2, v3, v4, v5, v6, v7,
+        v8, v9, v10,v11,v12,v13,v14,v15,
+        v0, v1, v2, v3, v4, v5, v6, v7,
+        v8, v9, v10,v11,v12,v13,v14,v15,
+        v0, v1, v2, v3, v4, v5, v6, v7,
+        v8, v9, v10,v11,v12,v13,v14,v15,
+        v0, v1, v2, v3, v4, v5, v6, v7,
+        v8, v9, v10,v11,v12,v13,v14,v15
+      );
+    }
+
+    // Saturated math
+    simdjson_inline simd8<uint8_t> saturating_add(const simd8<uint8_t> other) const { return _mm512_adds_epu8(*this, other); }
+    simdjson_inline simd8<uint8_t> saturating_sub(const simd8<uint8_t> other) const { return _mm512_subs_epu8(*this, other); }
+
+    // Order-specific operations
+    simdjson_inline simd8<uint8_t> max_val(const simd8<uint8_t> other) const { return _mm512_max_epu8(*this, other); }
+    simdjson_inline simd8<uint8_t> min_val(const simd8<uint8_t> other) const { return _mm512_min_epu8(other, *this); }
+    // Same as >, but only guarantees true is nonzero (< guarantees true = -1)
+    simdjson_inline simd8<uint8_t> gt_bits(const simd8<uint8_t> other) const { return this->saturating_sub(other); }
+    // Same as <, but only guarantees true is nonzero (< guarantees true = -1)
+    simdjson_inline simd8<uint8_t> lt_bits(const simd8<uint8_t> other) const { return other.saturating_sub(*this); }
+    simdjson_inline uint64_t operator<=(const simd8<uint8_t> other) const { return other.max_val(*this) == other; }
+    simdjson_inline uint64_t operator>=(const simd8<uint8_t> other) const { return other.min_val(*this) == other; }
+    simdjson_inline simd8<bool> operator>(const simd8<uint8_t> other) const { return this->gt_bits(other).any_bits_set(); }
+    simdjson_inline simd8<bool> operator<(const simd8<uint8_t> other) const { return this->lt_bits(other).any_bits_set(); }
+
+    // Bit-specific operations
+    simdjson_inline simd8<bool> bits_not_set() const { return _mm512_mask_blend_epi8(*this == uint8_t(0), _mm512_set1_epi8(0), _mm512_set1_epi8(-1)); }
+    simdjson_inline simd8<bool> bits_not_set(simd8<uint8_t> bits) const { return (*this & bits).bits_not_set(); }
+    simdjson_inline simd8<bool> any_bits_set() const { return ~this->bits_not_set(); }
+    simdjson_inline simd8<bool> any_bits_set(simd8<uint8_t> bits) const { return ~this->bits_not_set(bits); }
+
+    simdjson_inline bool is_ascii() const { return _mm512_movepi8_mask(*this) == 0; }
+    simdjson_inline bool bits_not_set_anywhere() const {
+      return !_mm512_test_epi8_mask(*this, *this);
+    }
+    simdjson_inline bool any_bits_set_anywhere() const { return !bits_not_set_anywhere(); }
+    simdjson_inline bool bits_not_set_anywhere(simd8<uint8_t> bits) const { return !_mm512_test_epi8_mask(*this, bits); }
+    simdjson_inline bool any_bits_set_anywhere(simd8<uint8_t> bits) const { return !bits_not_set_anywhere(bits); }
+    template<int N>
+    simdjson_inline simd8<uint8_t> shr() const { return simd8<uint8_t>(_mm512_srli_epi16(*this, N)) & uint8_t(0xFFu >> N); }
+    template<int N>
+    simdjson_inline simd8<uint8_t> shl() const { return simd8<uint8_t>(_mm512_slli_epi16(*this, N)) & uint8_t(0xFFu << N); }
+    // Get one of the bits and make a bitmask out of it.
+    // e.g. value.get_bit<7>() gets the high bit
+    template<int N>
+    simdjson_inline uint64_t get_bit() const { return _mm512_movepi8_mask(_mm512_slli_epi16(*this, 7-N)); }
+  };
+
+  template<typename T>
+  struct simd8x64 {
+    static constexpr int NUM_CHUNKS = 64 / sizeof(simd8<T>);
+    static_assert(NUM_CHUNKS == 1, "Icelake kernel should use one register per 64-byte block.");
+    const simd8<T> chunks[NUM_CHUNKS];
+
+    simd8x64(const simd8x64<T>& o) = delete; // no copy allowed
+    simd8x64<T>& operator=(const simd8<T>& other) = delete; // no assignment allowed
+    simd8x64() = delete; // no default constructor allowed
+
+    simdjson_inline simd8x64(const simd8<T> chunk0, const simd8<T> chunk1) : chunks{chunk0, chunk1} {}
+    simdjson_inline simd8x64(const simd8<T> chunk0) : chunks{chunk0} {}
+    simdjson_inline simd8x64(const T ptr[64]) : chunks{simd8<T>::load(ptr)} {}
+
+    simdjson_inline uint64_t compress(uint64_t mask, T * output) const {
+      this->chunks[0].compress(mask, output);
+      return 64 - count_ones(mask);
+    }
+
+    simdjson_inline void store(T ptr[64]) const {
+      this->chunks[0].store(ptr+sizeof(simd8<T>)*0);
+    }
+
+    simdjson_inline simd8<T> reduce_or() const {
+      return this->chunks[0];
+    }
+
+    simdjson_inline simd8x64<T> bit_or(const T m) const {
+      const simd8<T> mask = simd8<T>::splat(m);
+      return simd8x64<T>(
+        this->chunks[0] | mask
+      );
+    }
+
+    simdjson_inline uint64_t eq(const T m) const {
+      const simd8<T> mask = simd8<T>::splat(m);
+      return this->chunks[0] == mask;
+    }
+
+    simdjson_inline uint64_t eq(const simd8x64<uint8_t> &other) const {
+      return this->chunks[0] == other.chunks[0];
+    }
+
+    simdjson_inline uint64_t lteq(const T m) const {
+      const simd8<T> mask = simd8<T>::splat(m);
+      return this->chunks[0] <= mask;
+    }
+  }; // struct simd8x64<T>
+
+} // namespace simd
+
+} // unnamed namespace
+} // namespace icelake
+} // namespace simdjson
+
+#endif // SIMDJSON_ICELAKE_SIMD_H
diff --git a/contrib/libs/simdjson/include/simdjson/icelake/stringparsing_defs.h b/contrib/libs/simdjson/include/simdjson/icelake/stringparsing_defs.h
new file mode 100644
index 00000000000..4cc582737f2
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/icelake/stringparsing_defs.h
@@ -0,0 +1,48 @@
+#ifndef SIMDJSON_ICELAKE_STRINGPARSING_DEFS_H
+#define SIMDJSON_ICELAKE_STRINGPARSING_DEFS_H
+
+#ifndef SIMDJSON_CONDITIONAL_INCLUDE
+#include "simdjson/icelake/base.h"
+#include "simdjson/icelake/simd.h"
+#include "simdjson/icelake/bitmanipulation.h"
+#endif // SIMDJSON_CONDITIONAL_INCLUDE
+
+namespace simdjson {
+namespace icelake {
+namespace {
+
+using namespace simd;
+
+// Holds backslashes and quotes locations.
+struct backslash_and_quote {
+public:
+  static constexpr uint32_t BYTES_PROCESSED = 64;
+  simdjson_inline static backslash_and_quote copy_and_find(const uint8_t *src, uint8_t *dst);
+
+  simdjson_inline bool has_quote_first() { return ((bs_bits - 1) & quote_bits) != 0; }
+  simdjson_inline bool has_backslash() { return ((quote_bits - 1) & bs_bits) != 0; }
+  simdjson_inline int quote_index() { return trailing_zeroes(quote_bits); }
+  simdjson_inline int backslash_index() { return trailing_zeroes(bs_bits); }
+
+  uint64_t bs_bits;
+  uint64_t quote_bits;
+}; // struct backslash_and_quote
+
+simdjson_inline backslash_and_quote backslash_and_quote::copy_and_find(const uint8_t *src, uint8_t *dst) {
+  // this can read up to 15 bytes beyond the buffer size, but we require
+  // SIMDJSON_PADDING of padding
+  static_assert(SIMDJSON_PADDING >= (BYTES_PROCESSED - 1), "backslash and quote finder must process fewer than SIMDJSON_PADDING bytes");
+  simd8<uint8_t> v(src);
+  // store to dest unconditionally - we can overwrite the bits we don't like later
+  v.store(dst);
+  return {
+      static_cast<uint64_t>(v == '\\'), // bs_bits
+      static_cast<uint64_t>(v == '"'), // quote_bits
+  };
+}
+
+} // unnamed namespace
+} // namespace icelake
+} // namespace simdjson
+
+#endif // SIMDJSON_ICELAKE_STRINGPARSING_DEFS_H
diff --git a/contrib/libs/simdjson/include/simdjson/implementation.h b/contrib/libs/simdjson/include/simdjson/implementation.h
new file mode 100644
index 00000000000..19cb3716297
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/implementation.h
@@ -0,0 +1,230 @@
+#ifndef SIMDJSON_IMPLEMENTATION_H
+#define SIMDJSON_IMPLEMENTATION_H
+
+#include "simdjson/internal/atomic_ptr.h"
+#include "simdjson/internal/dom_parser_implementation.h"
+
+#include <memory>
+
+namespace simdjson {
+
+/**
+ * Validate the UTF-8 string.
+ *
+ * @param buf the string to validate.
+ * @param len the length of the string in bytes.
+ * @return true if the string is valid UTF-8.
+ */
+simdjson_warn_unused bool validate_utf8(const char * buf, size_t len) noexcept;
+/**
+ * Validate the UTF-8 string.
+ *
+ * @param sv the string_view to validate.
+ * @return true if the string is valid UTF-8.
+ */
+simdjson_inline simdjson_warn_unused bool validate_utf8(const std::string_view sv) noexcept {
+  return validate_utf8(sv.data(), sv.size());
+}
+
+/**
+ * Validate the UTF-8 string.
+ *
+ * @param p the string to validate.
+ * @return true if the string is valid UTF-8.
+ */
+simdjson_inline simdjson_warn_unused bool validate_utf8(const std::string& s) noexcept {
+  return validate_utf8(s.data(), s.size());
+}
+
+/**
+ * An implementation of simdjson for a particular CPU architecture.
+ *
+ * Also used to maintain the currently active implementation. The active implementation is
+ * automatically initialized on first use to the most advanced implementation supported by the host.
+ */
+class implementation {
+public:
+
+  /**
+   * The name of this implementation.
+   *
+   *     const implementation *impl = simdjson::get_active_implementation();
+   *     cout << "simdjson is optimized for " << impl->name() << "(" << impl->description() << ")" << endl;
+   *
+   * @return the name of the implementation, e.g. "haswell", "westmere", "arm64".
+   */
+  virtual std::string name() const { return std::string(_name); }
+
+  /**
+   * The description of this implementation.
+   *
+   *     const implementation *impl = simdjson::get_active_implementation();
+   *     cout << "simdjson is optimized for " << impl->name() << "(" << impl->description() << ")" << endl;
+   *
+   * @return the description of the implementation, e.g. "Intel/AMD AVX2", "Intel/AMD SSE4.2", "ARM NEON".
+   */
+  virtual std::string description() const { return std::string(_description); }
+
+  /**
+   * The instruction sets this implementation is compiled against
+   * and the current CPU match. This function may poll the current CPU/system
+   * and should therefore not be called too often if performance is a concern.
+   *
+   * @return true if the implementation can be safely used on the current system (determined at runtime).
+   */
+  bool supported_by_runtime_system() const;
+
+  /**
+   * @private For internal implementation use
+   *
+   * The instruction sets this implementation is compiled against.
+   *
+   * @return a mask of all required `internal::instruction_set::` values.
+   */
+  virtual uint32_t required_instruction_sets() const { return _required_instruction_sets; }
+
+  /**
+   * @private For internal implementation use
+   *
+   *     const implementation *impl = simdjson::get_active_implementation();
+   *     cout << "simdjson is optimized for " << impl->name() << "(" << impl->description() << ")" << endl;
+   *
+   * @param capacity The largest document that will be passed to the parser.
+   * @param max_depth The maximum JSON object/array nesting this parser is expected to handle.
+   * @param dst The place to put the resulting parser implementation.
+   * @return the error code, or SUCCESS if there was no error.
+   */
+  virtual error_code create_dom_parser_implementation(
+    size_t capacity,
+    size_t max_depth,
+    std::unique_ptr<internal::dom_parser_implementation> &dst
+  ) const noexcept = 0;
+
+  /**
+   * @private For internal implementation use
+   *
+   * Minify the input string assuming that it represents a JSON string, does not parse or validate.
+   *
+   * Overridden by each implementation.
+   *
+   * @param buf the json document to minify.
+   * @param len the length of the json document.
+   * @param dst the buffer to write the minified document to. *MUST* be allocated up to len + SIMDJSON_PADDING bytes.
+   * @param dst_len the number of bytes written. Output only.
+   * @return the error code, or SUCCESS if there was no error.
+   */
+  simdjson_warn_unused virtual error_code minify(const uint8_t *buf, size_t len, uint8_t *dst, size_t &dst_len) const noexcept = 0;
+
+
+  /**
+   * Validate the UTF-8 string.
+   *
+   * Overridden by each implementation.
+   *
+   * @param buf the string to validate.
+   * @param len the length of the string in bytes.
+   * @return true if and only if the string is valid UTF-8.
+   */
+  simdjson_warn_unused virtual bool validate_utf8(const char *buf, size_t len) const noexcept = 0;
+
+protected:
+  /** @private Construct an implementation with the given name and description. For subclasses. */
+  simdjson_inline implementation(
+    std::string_view name,
+    std::string_view description,
+    uint32_t required_instruction_sets
+  ) :
+    _name(name),
+    _description(description),
+    _required_instruction_sets(required_instruction_sets)
+  {
+  }
+protected:
+  ~implementation() = default;
+
+private:
+  /**
+   * The name of this implementation.
+   */
+  std::string_view _name;
+
+  /**
+   * The description of this implementation.
+   */
+  std::string_view _description;
+
+  /**
+   * Instruction sets required for this implementation.
+   */
+  const uint32_t _required_instruction_sets;
+};
+
+/** @private */
+namespace internal {
+
+/**
+ * The list of available implementations compiled into simdjson.
+ */
+class available_implementation_list {
+public:
+  /** Get the list of available implementations compiled into simdjson */
+  simdjson_inline available_implementation_list() {}
+  /** Number of implementations */
+  size_t size() const noexcept;
+  /** STL const begin() iterator */
+  const implementation * const *begin() const noexcept;
+  /** STL const end() iterator */
+  const implementation * const *end() const noexcept;
+
+  /**
+   * Get the implementation with the given name.
+   *
+   * Case sensitive.
+   *
+   *     const implementation *impl = simdjson::get_available_implementations()["westmere"];
+   *     if (!impl) { exit(1); }
+   *     if (!imp->supported_by_runtime_system()) { exit(1); }
+   *     simdjson::get_active_implementation() = impl;
+   *
+   * @param name the implementation to find, e.g. "westmere", "haswell", "arm64"
+   * @return the implementation, or nullptr if the parse failed.
+   */
+  const implementation * operator[](const std::string_view &name) const noexcept {
+    for (const implementation * impl : *this) {
+      if (impl->name() == name) { return impl; }
+    }
+    return nullptr;
+  }
+
+  /**
+   * Detect the most advanced implementation supported by the current host.
+   *
+   * This is used to initialize the implementation on startup.
+   *
+   *     const implementation *impl = simdjson::available_implementation::detect_best_supported();
+   *     simdjson::get_active_implementation() = impl;
+   *
+   * @return the most advanced supported implementation for the current host, or an
+   *         implementation that returns UNSUPPORTED_ARCHITECTURE if there is no supported
+   *         implementation. Will never return nullptr.
+   */
+  const implementation *detect_best_supported() const noexcept;
+};
+
+} // namespace internal
+
+/**
+ * The list of available implementations compiled into simdjson.
+ */
+extern SIMDJSON_DLLIMPORTEXPORT const internal::available_implementation_list& get_available_implementations();
+
+/**
+  * The active implementation.
+  *
+  * Automatically initialized on first use to the most advanced implementation supported by this hardware.
+  */
+extern SIMDJSON_DLLIMPORTEXPORT internal::atomic_ptr<const implementation>& get_active_implementation();
+
+} // namespace simdjson
+
+#endif // SIMDJSON_IMPLEMENTATION_H
diff --git a/contrib/libs/simdjson/include/simdjson/implementation_detection.h b/contrib/libs/simdjson/include/simdjson/implementation_detection.h
new file mode 100644
index 00000000000..0ff315b7adb
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/implementation_detection.h
@@ -0,0 +1,168 @@
+#ifndef SIMDJSON_IMPLEMENTATION_DETECTION_H
+#define SIMDJSON_IMPLEMENTATION_DETECTION_H
+
+#include "simdjson/base.h"
+
+// 0 is reserved, because undefined SIMDJSON_IMPLEMENTATION equals 0 in preprocessor macros.
+#define SIMDJSON_IMPLEMENTATION_ID_arm64 1
+#define SIMDJSON_IMPLEMENTATION_ID_fallback 2
+#define SIMDJSON_IMPLEMENTATION_ID_haswell 3
+#define SIMDJSON_IMPLEMENTATION_ID_icelake 4
+#define SIMDJSON_IMPLEMENTATION_ID_ppc64 5
+#define SIMDJSON_IMPLEMENTATION_ID_westmere 6
+#define SIMDJSON_IMPLEMENTATION_ID_lsx 7
+#define SIMDJSON_IMPLEMENTATION_ID_lasx 8
+
+#define SIMDJSON_IMPLEMENTATION_ID_FOR(IMPL) SIMDJSON_CAT(SIMDJSON_IMPLEMENTATION_ID_, IMPL)
+#define SIMDJSON_IMPLEMENTATION_ID SIMDJSON_IMPLEMENTATION_ID_FOR(SIMDJSON_IMPLEMENTATION)
+
+#define SIMDJSON_IMPLEMENTATION_IS(IMPL) SIMDJSON_IMPLEMENTATION_ID == SIMDJSON_IMPLEMENTATION_ID_FOR(IMPL)
+
+//
+// First, figure out which implementations can be run. Doing it here makes it so we don't have to worry about the order
+// in which we include them.
+//
+
+#ifndef SIMDJSON_IMPLEMENTATION_ARM64
+#define SIMDJSON_IMPLEMENTATION_ARM64 (SIMDJSON_IS_ARM64)
+#endif
+#if SIMDJSON_IMPLEMENTATION_ARM64 && SIMDJSON_IS_ARM64
+#define SIMDJSON_CAN_ALWAYS_RUN_ARM64 1
+#else
+#define SIMDJSON_CAN_ALWAYS_RUN_ARM64 0
+#endif
+
+// Default Icelake to on if this is x86-64. Even if we're not compiled for it, it could be selected
+// at runtime.
+#ifndef SIMDJSON_IMPLEMENTATION_ICELAKE
+#define SIMDJSON_IMPLEMENTATION_ICELAKE ((SIMDJSON_IS_X86_64) && (SIMDJSON_AVX512_ALLOWED) && (SIMDJSON_COMPILER_SUPPORTS_VBMI2))
+#endif
+
+#ifdef _MSC_VER
+// To see why  (__BMI__) && (__PCLMUL__) && (__LZCNT__) are not part of this next line, see
+// https://github.com/simdjson/simdjson/issues/1247
+#if ((SIMDJSON_IMPLEMENTATION_ICELAKE) && (__AVX2__) && (__AVX512F__) && (__AVX512DQ__) && (__AVX512CD__) && (__AVX512BW__) && (__AVX512VL__) && (__AVX512VBMI2__))
+#define SIMDJSON_CAN_ALWAYS_RUN_ICELAKE 1
+#else
+#define SIMDJSON_CAN_ALWAYS_RUN_ICELAKE 0
+#endif
+
+#else
+
+#if ((SIMDJSON_IMPLEMENTATION_ICELAKE) && (__AVX2__) && (__BMI__) && (__PCLMUL__) && (__LZCNT__) && (__AVX512F__) && (__AVX512DQ__) && (__AVX512CD__) && (__AVX512BW__) && (__AVX512VL__) && (__AVX512VBMI2__))
+#define SIMDJSON_CAN_ALWAYS_RUN_ICELAKE 1
+#else
+#define SIMDJSON_CAN_ALWAYS_RUN_ICELAKE 0
+#endif
+
+#endif
+
+// Default Haswell to on if this is x86-64. Even if we're not compiled for it, it could be selected
+// at runtime.
+#ifndef SIMDJSON_IMPLEMENTATION_HASWELL
+#if SIMDJSON_CAN_ALWAYS_RUN_ICELAKE
+// if icelake is always available, never enable haswell.
+#define SIMDJSON_IMPLEMENTATION_HASWELL 0
+#else
+#define SIMDJSON_IMPLEMENTATION_HASWELL SIMDJSON_IS_X86_64
+#endif
+#endif
+#ifdef _MSC_VER
+// To see why  (__BMI__) && (__PCLMUL__) && (__LZCNT__) are not part of this next line, see
+// https://github.com/simdjson/simdjson/issues/1247
+#if ((SIMDJSON_IMPLEMENTATION_HASWELL) && (SIMDJSON_IS_X86_64) && (__AVX2__))
+#define SIMDJSON_CAN_ALWAYS_RUN_HASWELL 1
+#else
+#define SIMDJSON_CAN_ALWAYS_RUN_HASWELL 0
+#endif
+
+#else
+
+#if ((SIMDJSON_IMPLEMENTATION_HASWELL) && (SIMDJSON_IS_X86_64) && (__AVX2__) && (__BMI__) && (__PCLMUL__) && (__LZCNT__))
+#define SIMDJSON_CAN_ALWAYS_RUN_HASWELL 1
+#else
+#define SIMDJSON_CAN_ALWAYS_RUN_HASWELL 0
+#endif
+
+#endif
+
+// Default Westmere to on if this is x86-64.
+#ifndef SIMDJSON_IMPLEMENTATION_WESTMERE
+#if SIMDJSON_CAN_ALWAYS_RUN_ICELAKE || SIMDJSON_CAN_ALWAYS_RUN_HASWELL
+// if icelake or haswell are always available, never enable westmere.
+#define SIMDJSON_IMPLEMENTATION_WESTMERE 0
+#else
+#define SIMDJSON_IMPLEMENTATION_WESTMERE SIMDJSON_IS_X86_64
+#endif
+#endif
+
+#if (SIMDJSON_IMPLEMENTATION_WESTMERE && SIMDJSON_IS_X86_64 && __SSE4_2__ && __PCLMUL__)
+#define SIMDJSON_CAN_ALWAYS_RUN_WESTMERE 1
+#else
+#define SIMDJSON_CAN_ALWAYS_RUN_WESTMERE 0
+#endif
+
+
+#ifndef SIMDJSON_IMPLEMENTATION_PPC64
+#define SIMDJSON_IMPLEMENTATION_PPC64 (SIMDJSON_IS_PPC64 && SIMDJSON_IS_PPC64_VMX)
+#endif
+#if SIMDJSON_IMPLEMENTATION_PPC64 && SIMDJSON_IS_PPC64 && SIMDJSON_IS_PPC64_VMX
+#define SIMDJSON_CAN_ALWAYS_RUN_PPC64 1
+#else
+#define SIMDJSON_CAN_ALWAYS_RUN_PPC64 0
+#endif
+
+#ifndef SIMDJSON_IMPLEMENTATION_LASX
+#define SIMDJSON_IMPLEMENTATION_LASX (SIMDJSON_IS_LOONGARCH64 && __loongarch_asx)
+#endif
+#define SIMDJSON_CAN_ALWAYS_RUN_LASX (SIMDJSON_IMPLEMENTATION_LASX)
+
+#ifndef SIMDJSON_IMPLEMENTATION_LSX
+#if SIMDJSON_CAN_ALWAYS_RUN_LASX
+#define SIMDJSON_IMPLEMENTATION_LSX 0
+#else
+#define SIMDJSON_IMPLEMENTATION_LSX (SIMDJSON_IS_LOONGARCH64 && __loongarch_sx)
+#endif
+#endif
+#define SIMDJSON_CAN_ALWAYS_RUN_LSX (SIMDJSON_IMPLEMENTATION_LSX)
+
+// Default Fallback to on unless a builtin implementation has already been selected.
+#ifndef SIMDJSON_IMPLEMENTATION_FALLBACK
+#if SIMDJSON_CAN_ALWAYS_RUN_ARM64 || SIMDJSON_CAN_ALWAYS_RUN_ICELAKE || SIMDJSON_CAN_ALWAYS_RUN_HASWELL || SIMDJSON_CAN_ALWAYS_RUN_WESTMERE || SIMDJSON_CAN_ALWAYS_RUN_PPC64 || SIMDJSON_CAN_ALWAYS_RUN_LSX || SIMDJSON_CAN_ALWAYS_RUN_LASX
+// if anything at all except fallback can always run, then disable fallback.
+#define SIMDJSON_IMPLEMENTATION_FALLBACK 0
+#else
+#define SIMDJSON_IMPLEMENTATION_FALLBACK 1
+#endif
+#endif
+#define SIMDJSON_CAN_ALWAYS_RUN_FALLBACK SIMDJSON_IMPLEMENTATION_FALLBACK
+
+// Determine the best builtin implementation
+#ifndef SIMDJSON_BUILTIN_IMPLEMENTATION
+
+#if SIMDJSON_CAN_ALWAYS_RUN_ICELAKE
+#define SIMDJSON_BUILTIN_IMPLEMENTATION icelake
+#elif SIMDJSON_CAN_ALWAYS_RUN_HASWELL
+#define SIMDJSON_BUILTIN_IMPLEMENTATION haswell
+#elif SIMDJSON_CAN_ALWAYS_RUN_WESTMERE
+#define SIMDJSON_BUILTIN_IMPLEMENTATION westmere
+#elif SIMDJSON_CAN_ALWAYS_RUN_ARM64
+#define SIMDJSON_BUILTIN_IMPLEMENTATION arm64
+#elif SIMDJSON_CAN_ALWAYS_RUN_PPC64
+#define SIMDJSON_BUILTIN_IMPLEMENTATION ppc64
+#elif SIMDJSON_CAN_ALWAYS_RUN_LSX
+#define SIMDJSON_BUILTIN_IMPLEMENTATION lsx
+#elif SIMDJSON_CAN_ALWAYS_RUN_LASX
+#define SIMDJSON_BUILTIN_IMPLEMENTATION lasx
+#elif SIMDJSON_CAN_ALWAYS_RUN_FALLBACK
+#define SIMDJSON_BUILTIN_IMPLEMENTATION fallback
+#else
+#error "All possible implementations (including fallback) have been disabled! simdjson will not run."
+#endif
+
+#endif // SIMDJSON_BUILTIN_IMPLEMENTATION
+
+#define SIMDJSON_BUILTIN_IMPLEMENTATION_ID SIMDJSON_IMPLEMENTATION_ID_FOR(SIMDJSON_BUILTIN_IMPLEMENTATION)
+#define SIMDJSON_BUILTIN_IMPLEMENTATION_IS(IMPL) SIMDJSON_BUILTIN_IMPLEMENTATION_ID == SIMDJSON_IMPLEMENTATION_ID_FOR(IMPL)
+
+#endif // SIMDJSON_IMPLEMENTATION_DETECTION_H
\ No newline at end of file
diff --git a/contrib/libs/simdjson/include/simdjson/internal/atomic_ptr.h b/contrib/libs/simdjson/include/simdjson/internal/atomic_ptr.h
new file mode 100644
index 00000000000..c4fe41b05a6
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/internal/atomic_ptr.h
@@ -0,0 +1,31 @@
+#ifndef SIMDJSON_INTERNAL_ATOMIC_PTR_H
+#define SIMDJSON_INTERNAL_ATOMIC_PTR_H
+
+#include "simdjson/base.h"
+#include <atomic>
+
+namespace simdjson {
+namespace internal {
+
+template<typename T>
+class atomic_ptr {
+public:
+  atomic_ptr(T *_ptr) : ptr{_ptr} {}
+
+  operator const T*() const { return ptr.load(); }
+  const T& operator*() const { return *ptr; }
+  const T* operator->() const { return ptr.load(); }
+
+  operator T*() { return ptr.load(); }
+  T& operator*() { return *ptr; }
+  T* operator->() { return ptr.load(); }
+  atomic_ptr& operator=(T *_ptr) { ptr = _ptr; return *this; }
+
+private:
+  std::atomic<T*> ptr;
+};
+
+} // namespace internal
+} // namespace simdjson
+
+#endif // SIMDJSON_INTERNAL_ATOMIC_PTR_H
diff --git a/contrib/libs/simdjson/include/simdjson/internal/dom_parser_implementation.h b/contrib/libs/simdjson/include/simdjson/internal/dom_parser_implementation.h
new file mode 100644
index 00000000000..a93fe38ff9e
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/internal/dom_parser_implementation.h
@@ -0,0 +1,252 @@
+#ifndef SIMDJSON_INTERNAL_DOM_PARSER_IMPLEMENTATION_H
+#define SIMDJSON_INTERNAL_DOM_PARSER_IMPLEMENTATION_H
+
+#include "simdjson/base.h"
+#include "simdjson/error.h"
+#include <memory>
+
+namespace simdjson {
+
+namespace dom {
+class document;
+} // namespace dom
+
+/**
+* This enum is used with the dom_parser_implementation::stage1 function.
+* 1) The regular mode expects a fully formed JSON document.
+* 2) The streaming_partial mode expects a possibly truncated
+* input within a stream on JSON documents.
+* 3) The stream_final mode allows us to truncate final
+* unterminated strings. It is useful in conjunction with streaming_partial.
+*/
+enum class stage1_mode { regular, streaming_partial, streaming_final};
+
+/**
+ * Returns true if mode == streaming_partial or mode == streaming_final
+ */
+inline bool is_streaming(stage1_mode mode) {
+  // performance note: it is probably faster to check that mode is different
+  // from regular than checking that it is either streaming_partial or streaming_final.
+  return (mode != stage1_mode::regular);
+  // return (mode == stage1_mode::streaming_partial || mode == stage1_mode::streaming_final);
+}
+
+
+namespace internal {
+
+
+/**
+ * An implementation of simdjson's DOM parser for a particular CPU architecture.
+ *
+ * This class is expected to be accessed only by pointer, and never move in memory (though the
+ * pointer can move).
+ */
+class dom_parser_implementation {
+public:
+
+  /**
+   * @private For internal implementation use
+   *
+   * Run a full JSON parse on a single document (stage1 + stage2).
+   *
+   * Guaranteed only to be called when capacity > document length.
+   *
+   * Overridden by each implementation.
+   *
+   * @param buf The json document to parse. *MUST* be allocated up to len + SIMDJSON_PADDING bytes.
+   * @param len The length of the json document.
+   * @return The error code, or SUCCESS if there was no error.
+   */
+  simdjson_warn_unused virtual error_code parse(const uint8_t *buf, size_t len, dom::document &doc) noexcept = 0;
+
+  /**
+   * @private For internal implementation use
+   *
+   * Stage 1 of the document parser.
+   *
+   * Guaranteed only to be called when capacity > document length.
+   *
+   * Overridden by each implementation.
+   *
+   * @param buf The json document to parse.
+   * @param len The length of the json document.
+   * @param streaming Whether this is being called by parser::parse_many.
+   * @return The error code, or SUCCESS if there was no error.
+   */
+  simdjson_warn_unused virtual error_code stage1(const uint8_t *buf, size_t len, stage1_mode streaming) noexcept = 0;
+
+  /**
+   * @private For internal implementation use
+   *
+   * Stage 2 of the document parser.
+   *
+   * Called after stage1().
+   *
+   * Overridden by each implementation.
+   *
+   * @param doc The document to output to.
+   * @return The error code, or SUCCESS if there was no error.
+   */
+  simdjson_warn_unused virtual error_code stage2(dom::document &doc) noexcept = 0;
+
+  /**
+   * @private For internal implementation use
+   *
+   * Stage 2 of the document parser for parser::parse_many.
+   *
+   * Guaranteed only to be called after stage1().
+   * Overridden by each implementation.
+   *
+   * @param doc The document to output to.
+   * @return The error code, SUCCESS if there was no error, or EMPTY if all documents have been parsed.
+   */
+  simdjson_warn_unused virtual error_code stage2_next(dom::document &doc) noexcept = 0;
+
+  /**
+   * Unescape a valid UTF-8 string from src to dst, stopping at a final unescaped quote. There
+   * must be an unescaped quote terminating the string. It returns the final output
+   * position as pointer. In case of error (e.g., the string has bad escaped codes),
+   * then null_ptr is returned. It is assumed that the output buffer is large
+   * enough. E.g., if src points at 'joe"', then dst needs to have four free bytes +
+   * SIMDJSON_PADDING bytes.
+   *
+   * Overridden by each implementation.
+   *
+   * @param str pointer to the beginning of a valid UTF-8 JSON string, must end with an unescaped quote.
+   * @param dst pointer to a destination buffer, it must point a region in memory of sufficient size.
+   * @param allow_replacement whether we allow a replacement character when the UTF-8 contains unmatched surrogate pairs.
+   * @return end of the of the written region (exclusive) or nullptr in case of error.
+   */
+  simdjson_warn_unused virtual uint8_t *parse_string(const uint8_t *src, uint8_t *dst, bool allow_replacement) const noexcept = 0;
+
+  /**
+   * Unescape a NON-valid UTF-8 string from src to dst, stopping at a final unescaped quote. There
+   * must be an unescaped quote terminating the string. It returns the final output
+   * position as pointer. In case of error (e.g., the string has bad escaped codes),
+   * then null_ptr is returned. It is assumed that the output buffer is large
+   * enough. E.g., if src points at 'joe"', then dst needs to have four free bytes +
+   * SIMDJSON_PADDING bytes.
+   *
+   * Overridden by each implementation.
+   *
+   * @param str pointer to the beginning of a possibly invalid UTF-8 JSON string, must end with an unescaped quote.
+   * @param dst pointer to a destination buffer, it must point a region in memory of sufficient size.
+   * @return end of the of the written region (exclusive) or nullptr in case of error.
+   */
+  simdjson_warn_unused virtual uint8_t *parse_wobbly_string(const uint8_t *src, uint8_t *dst) const noexcept = 0;
+
+  /**
+   * Change the capacity of this parser.
+   *
+   * The capacity can never exceed SIMDJSON_MAXSIZE_BYTES (e.g., 4 GB)
+   * and an CAPACITY error is returned if it is attempted.
+   *
+   * Generally used for reallocation.
+   *
+   * @param capacity The new capacity.
+   * @param max_depth The new max_depth.
+   * @return The error code, or SUCCESS if there was no error.
+   */
+  virtual error_code set_capacity(size_t capacity) noexcept = 0;
+
+  /**
+   * Change the max depth of this parser.
+   *
+   * Generally used for reallocation.
+   *
+   * @param capacity The new capacity.
+   * @param max_depth The new max_depth.
+   * @return The error code, or SUCCESS if there was no error.
+   */
+  virtual error_code set_max_depth(size_t max_depth) noexcept = 0;
+
+  /**
+   * Deallocate this parser.
+   */
+  virtual ~dom_parser_implementation() = default;
+
+  /** Number of structural indices passed from stage 1 to stage 2 */
+  uint32_t n_structural_indexes{0};
+  /** Structural indices passed from stage 1 to stage 2 */
+  std::unique_ptr<uint32_t[]> structural_indexes{};
+  /** Next structural index to parse */
+  uint32_t next_structural_index{0};
+
+  /**
+   * The largest document this parser can support without reallocating.
+   *
+   * @return Current capacity, in bytes.
+   */
+  simdjson_pure simdjson_inline size_t capacity() const noexcept;
+
+  /**
+   * The maximum level of nested object and arrays supported by this parser.
+   *
+   * @return Maximum depth, in bytes.
+   */
+  simdjson_pure simdjson_inline size_t max_depth() const noexcept;
+
+  /**
+   * Ensure this parser has enough memory to process JSON documents up to `capacity` bytes in length
+   * and `max_depth` depth.
+   *
+   * @param capacity The new capacity.
+   * @param max_depth The new max_depth. Defaults to DEFAULT_MAX_DEPTH.
+   * @return The error, if there is one.
+   */
+  simdjson_warn_unused inline error_code allocate(size_t capacity, size_t max_depth) noexcept;
+
+
+protected:
+  /**
+   * The maximum document length this parser supports.
+   *
+   * Buffers are large enough to handle any document up to this length.
+   */
+  size_t _capacity{0};
+
+  /**
+   * The maximum depth (number of nested objects and arrays) supported by this parser.
+   *
+   * Defaults to DEFAULT_MAX_DEPTH.
+   */
+  size_t _max_depth{0};
+
+  // Declaring these so that subclasses can use them to implement their constructors.
+  simdjson_inline dom_parser_implementation() noexcept;
+  simdjson_inline dom_parser_implementation(dom_parser_implementation &&other) noexcept;
+  simdjson_inline dom_parser_implementation &operator=(dom_parser_implementation &&other) noexcept;
+
+  simdjson_inline dom_parser_implementation(const dom_parser_implementation &) noexcept = delete;
+  simdjson_inline dom_parser_implementation &operator=(const dom_parser_implementation &other) noexcept = delete;
+}; // class dom_parser_implementation
+
+simdjson_inline dom_parser_implementation::dom_parser_implementation() noexcept = default;
+simdjson_inline dom_parser_implementation::dom_parser_implementation(dom_parser_implementation &&other) noexcept = default;
+simdjson_inline dom_parser_implementation &dom_parser_implementation::operator=(dom_parser_implementation &&other) noexcept = default;
+
+simdjson_pure simdjson_inline size_t dom_parser_implementation::capacity() const noexcept {
+  return _capacity;
+}
+
+simdjson_pure simdjson_inline size_t dom_parser_implementation::max_depth() const noexcept {
+  return _max_depth;
+}
+
+simdjson_warn_unused
+inline error_code dom_parser_implementation::allocate(size_t capacity, size_t max_depth) noexcept {
+  if (this->max_depth() != max_depth) {
+    error_code err = set_max_depth(max_depth);
+    if (err) { return err; }
+  }
+  if (_capacity != capacity) {
+    error_code err = set_capacity(capacity);
+    if (err) { return err; }
+  }
+  return SUCCESS;
+}
+
+} // namespace internal
+} // namespace simdjson
+
+#endif // SIMDJSON_INTERNAL_DOM_PARSER_IMPLEMENTATION_H
diff --git a/contrib/libs/simdjson/include/simdjson/internal/instruction_set.h b/contrib/libs/simdjson/include/simdjson/internal/instruction_set.h
new file mode 100644
index 00000000000..1dc0a81fb3c
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/internal/instruction_set.h
@@ -0,0 +1,77 @@
+/* From
+https://github.com/endorno/pytorch/blob/master/torch/lib/TH/generic/simd/simd.h
+Highly modified.
+
+Copyright (c) 2016-     Facebook, Inc            (Adam Paszke)
+Copyright (c) 2014-     Facebook, Inc            (Soumith Chintala)
+Copyright (c) 2011-2014 Idiap Research Institute (Ronan Collobert)
+Copyright (c) 2012-2014 Deepmind Technologies    (Koray Kavukcuoglu)
+Copyright (c) 2011-2012 NEC Laboratories America (Koray Kavukcuoglu)
+Copyright (c) 2011-2013 NYU                      (Clement Farabet)
+Copyright (c) 2006-2010 NEC Laboratories America (Ronan Collobert, Leon Bottou,
+Iain Melvin, Jason Weston) Copyright (c) 2006      Idiap Research Institute
+(Samy Bengio) Copyright (c) 2001-2004 Idiap Research Institute (Ronan Collobert,
+Samy Bengio, Johnny Mariethoz)
+
+All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+
+1. Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+
+2. Redistributions in binary form must reproduce the above copyright
+   notice, this list of conditions and the following disclaimer in the
+   documentation and/or other materials provided with the distribution.
+
+3. Neither the names of Facebook, Deepmind Technologies, NYU, NEC Laboratories
+America and IDIAP Research Institute nor the names of its contributors may be
+   used to endorse or promote products derived from this software without
+   specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+POSSIBILITY OF SUCH DAMAGE.
+*/
+
+#ifndef SIMDJSON_INTERNAL_INSTRUCTION_SET_H
+#define SIMDJSON_INTERNAL_INSTRUCTION_SET_H
+
+namespace simdjson {
+namespace internal {
+
+enum instruction_set {
+  DEFAULT = 0x0,
+  NEON = 0x1,
+  AVX2 = 0x4,
+  SSE42 = 0x8,
+  PCLMULQDQ = 0x10,
+  BMI1 = 0x20,
+  BMI2 = 0x40,
+  ALTIVEC = 0x80,
+  AVX512F = 0x100,
+  AVX512DQ = 0x200,
+  AVX512IFMA = 0x400,
+  AVX512PF = 0x800,
+  AVX512ER = 0x1000,
+  AVX512CD = 0x2000,
+  AVX512BW = 0x4000,
+  AVX512VL = 0x8000,
+  AVX512VBMI2 = 0x10000,
+  LSX = 0x20000,
+  LASX = 0x40000,
+};
+
+} // namespace internal
+} // namespace simdjson
+
+#endif // SIMDJSON_INTERNAL_INSTRUCTION_SET_H
diff --git a/contrib/libs/simdjson/include/simdjson/internal/jsoncharutils_tables.h b/contrib/libs/simdjson/include/simdjson/internal/jsoncharutils_tables.h
new file mode 100644
index 00000000000..d72bce12262
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/internal/jsoncharutils_tables.h
@@ -0,0 +1,26 @@
+#ifndef SIMDJSON_INTERNAL_JSONCHARUTILS_TABLES_H
+#define SIMDJSON_INTERNAL_JSONCHARUTILS_TABLES_H
+
+#include "simdjson/base.h"
+
+#ifdef JSON_TEST_STRINGS
+void found_string(const uint8_t *buf, const uint8_t *parsed_begin,
+                  const uint8_t *parsed_end);
+void found_bad_string(const uint8_t *buf);
+#endif
+
+namespace simdjson {
+namespace internal {
+// structural chars here are
+// they are { 0x7b } 0x7d : 0x3a [ 0x5b ] 0x5d , 0x2c (and NULL)
+// we are also interested in the four whitespace characters
+// space 0x20, linefeed 0x0a, horizontal tab 0x09 and carriage return 0x0d
+
+extern SIMDJSON_DLLIMPORTEXPORT const bool structural_or_whitespace_negated[256];
+extern SIMDJSON_DLLIMPORTEXPORT const bool structural_or_whitespace[256];
+extern SIMDJSON_DLLIMPORTEXPORT const uint32_t digit_to_val32[886];
+
+} // namespace internal
+} // namespace simdjson
+
+#endif // SIMDJSON_INTERNAL_JSONCHARUTILS_TABLES_H
diff --git a/contrib/libs/simdjson/include/simdjson/internal/numberparsing_tables.h b/contrib/libs/simdjson/include/simdjson/internal/numberparsing_tables.h
new file mode 100644
index 00000000000..1762056f750
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/internal/numberparsing_tables.h
@@ -0,0 +1,59 @@
+#ifndef SIMDJSON_INTERNAL_NUMBERPARSING_TABLES_H
+#define SIMDJSON_INTERNAL_NUMBERPARSING_TABLES_H
+
+#include "simdjson/base.h"
+
+namespace simdjson {
+namespace internal {
+/**
+ * The smallest non-zero float (binary64) is 2^-1074.
+ * We take as input numbers of the form w x 10^q where w < 2^64.
+ * We have that w * 10^-343  <  2^(64-344) 5^-343 < 2^-1076.
+ * However, we have that
+ * (2^64-1) * 10^-342 =  (2^64-1) * 2^-342 * 5^-342 > 2^-1074.
+ * Thus it is possible for a number of the form w * 10^-342 where
+ * w is a 64-bit value to be a non-zero floating-point number.
+ *********
+ * Any number of form w * 10^309 where w>= 1 is going to be
+ * infinite in binary64 so we never need to worry about powers
+ * of 5 greater than 308.
+ */
+constexpr int smallest_power = -342;
+constexpr int largest_power = 308;
+
+/**
+ * Represents a 128-bit value.
+ * low: least significant 64 bits.
+ * high: most significant 64 bits.
+ */
+struct value128 {
+  uint64_t low;
+  uint64_t high;
+};
+
+
+// Precomputed powers of ten from 10^0 to 10^22. These
+// can be represented exactly using the double type.
+extern SIMDJSON_DLLIMPORTEXPORT const double power_of_ten[];
+
+
+/**
+ * When mapping numbers from decimal to binary,
+ * we go from w * 10^q to m * 2^p but we have
+ * 10^q = 5^q * 2^q, so effectively
+ * we are trying to match
+ * w * 2^q * 5^q to m * 2^p. Thus the powers of two
+ * are not a concern since they can be represented
+ * exactly using the binary notation, only the powers of five
+ * affect the binary significand.
+ */
+
+
+// The truncated powers of five from 5^-342 all the way to 5^308
+// The mantissa is truncated to 128 bits, and
+// never rounded up. Uses about 10KB.
+extern SIMDJSON_DLLIMPORTEXPORT const uint64_t power_of_five_128[];
+} // namespace internal
+} // namespace simdjson
+
+#endif // SIMDJSON_INTERNAL_NUMBERPARSING_TABLES_H
diff --git a/contrib/libs/simdjson/include/simdjson/internal/simdprune_tables.h b/contrib/libs/simdjson/include/simdjson/internal/simdprune_tables.h
new file mode 100644
index 00000000000..7b8f4650c15
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/internal/simdprune_tables.h
@@ -0,0 +1,21 @@
+#ifndef SIMDJSON_INTERNAL_SIMDPRUNE_TABLES_H
+#define SIMDJSON_INTERNAL_SIMDPRUNE_TABLES_H
+
+#include "simdjson/base.h"
+
+#include <cstdint>
+
+namespace simdjson { // table modified and copied from
+namespace internal { // http://graphics.stanford.edu/~seander/bithacks.html#CountBitsSetTable
+
+extern SIMDJSON_DLLIMPORTEXPORT const unsigned char BitsSetTable256mul2[256];
+
+extern SIMDJSON_DLLIMPORTEXPORT const uint8_t pshufb_combine_table[272];
+
+// 256 * 8 bytes = 2kB, easily fits in cache.
+extern SIMDJSON_DLLIMPORTEXPORT const uint64_t thintable_epi8[256];
+
+} // namespace internal
+} // namespace simdjson
+
+#endif // SIMDJSON_INTERNAL_SIMDPRUNE_TABLES_H
diff --git a/contrib/libs/simdjson/include/simdjson/internal/tape_type.h b/contrib/libs/simdjson/include/simdjson/internal/tape_type.h
new file mode 100644
index 00000000000..d43c57c7a49
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/internal/tape_type.h
@@ -0,0 +1,28 @@
+#ifndef SIMDJSON_INTERNAL_TAPE_TYPE_H
+#define SIMDJSON_INTERNAL_TAPE_TYPE_H
+
+namespace simdjson {
+namespace internal {
+
+/**
+ * The possible types in the tape.
+ */
+enum class tape_type {
+  ROOT = 'r',
+  START_ARRAY = '[',
+  START_OBJECT = '{',
+  END_ARRAY = ']',
+  END_OBJECT = '}',
+  STRING = '"',
+  INT64 = 'l',
+  UINT64 = 'u',
+  DOUBLE = 'd',
+  TRUE_VALUE = 't',
+  FALSE_VALUE = 'f',
+  NULL_VALUE = 'n'
+}; // enum class tape_type
+
+} // namespace internal
+} // namespace simdjson
+
+#endif // SIMDJSON_INTERNAL_TAPE_TYPE_H
diff --git a/contrib/libs/simdjson/include/simdjson/lasx/base.h b/contrib/libs/simdjson/include/simdjson/lasx/base.h
new file mode 100644
index 00000000000..9d9a866c375
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/lasx/base.h
@@ -0,0 +1,26 @@
+#ifndef SIMDJSON_LASX_BASE_H
+#define SIMDJSON_LASX_BASE_H
+
+#ifndef SIMDJSON_CONDITIONAL_INCLUDE
+#include "simdjson/base.h"
+#endif // SIMDJSON_CONDITIONAL_INCLUDE
+
+namespace simdjson {
+/**
+ * Implementation for LASX.
+ */
+namespace lasx {
+
+class implementation;
+
+namespace {
+namespace simd {
+template <typename T> struct simd8;
+template <typename T> struct simd8x64;
+} // namespace simd
+} // unnamed namespace
+
+} // namespace lasx
+} // namespace simdjson
+
+#endif // SIMDJSON_LASX_BASE_H
diff --git a/contrib/libs/simdjson/include/simdjson/lasx/begin.h b/contrib/libs/simdjson/include/simdjson/lasx/begin.h
new file mode 100644
index 00000000000..560eba737ce
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/lasx/begin.h
@@ -0,0 +1,10 @@
+#define SIMDJSON_IMPLEMENTATION lasx
+#include "simdjson/lasx/base.h"
+#include "simdjson/lasx/intrinsics.h"
+#include "simdjson/lasx/bitmanipulation.h"
+#include "simdjson/lasx/bitmask.h"
+#include "simdjson/lasx/numberparsing_defs.h"
+#include "simdjson/lasx/simd.h"
+#include "simdjson/lasx/stringparsing_defs.h"
+
+#define SIMDJSON_SKIP_BACKSLASH_SHORT_CIRCUIT 1
diff --git a/contrib/libs/simdjson/include/simdjson/lasx/bitmanipulation.h b/contrib/libs/simdjson/include/simdjson/lasx/bitmanipulation.h
new file mode 100644
index 00000000000..962ddbf56bc
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/lasx/bitmanipulation.h
@@ -0,0 +1,50 @@
+#ifndef SIMDJSON_LASX_BITMANIPULATION_H
+#define SIMDJSON_LASX_BITMANIPULATION_H
+
+#ifndef SIMDJSON_CONDITIONAL_INCLUDE
+#include "simdjson/lasx/base.h"
+#include "simdjson/lasx/intrinsics.h"
+#include "simdjson/lasx/bitmask.h"
+#endif // SIMDJSON_CONDITIONAL_INCLUDE
+
+namespace simdjson {
+namespace lasx {
+namespace {
+
+// We sometimes call trailing_zero on inputs that are zero,
+// but the algorithms do not end up using the returned value.
+// Sadly, sanitizers are not smart enough to figure it out.
+SIMDJSON_NO_SANITIZE_UNDEFINED
+// This function can be used safely even if not all bytes have been
+// initialized.
+// See issue https://github.com/simdjson/simdjson/issues/1965
+SIMDJSON_NO_SANITIZE_MEMORY
+simdjson_inline int trailing_zeroes(uint64_t input_num) {
+  return __builtin_ctzll(input_num);
+}
+
+/* result might be undefined when input_num is zero */
+simdjson_inline uint64_t clear_lowest_bit(uint64_t input_num) {
+  return input_num & (input_num-1);
+}
+
+/* result might be undefined when input_num is zero */
+simdjson_inline int leading_zeroes(uint64_t input_num) {
+  return __builtin_clzll(input_num);
+}
+
+/* result might be undefined when input_num is zero */
+simdjson_inline int count_ones(uint64_t input_num) {
+  return __lasx_xvpickve2gr_w(__lasx_xvpcnt_d(__m256i(v4u64{input_num, 0, 0, 0})), 0);
+}
+
+simdjson_inline bool add_overflow(uint64_t value1, uint64_t value2, uint64_t *result) {
+  return __builtin_uaddll_overflow(value1, value2,
+                                   reinterpret_cast<unsigned long long *>(result));
+}
+
+} // unnamed namespace
+} // namespace lasx
+} // namespace simdjson
+
+#endif // SIMDJSON_LASX_BITMANIPULATION_H
diff --git a/contrib/libs/simdjson/include/simdjson/lasx/bitmask.h b/contrib/libs/simdjson/include/simdjson/lasx/bitmask.h
new file mode 100644
index 00000000000..e847c1d7100
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/lasx/bitmask.h
@@ -0,0 +1,31 @@
+#ifndef SIMDJSON_LASX_BITMASK_H
+#define SIMDJSON_LASX_BITMASK_H
+
+#ifndef SIMDJSON_CONDITIONAL_INCLUDE
+#include "simdjson/lasx/base.h"
+#endif // SIMDJSON_CONDITIONAL_INCLUDE
+
+namespace simdjson {
+namespace lasx {
+namespace {
+
+//
+// Perform a "cumulative bitwise xor," flipping bits each time a 1 is encountered.
+//
+// For example, prefix_xor(00100100) == 00011100
+//
+simdjson_inline uint64_t prefix_xor(uint64_t bitmask) {
+  bitmask ^= bitmask << 1;
+  bitmask ^= bitmask << 2;
+  bitmask ^= bitmask << 4;
+  bitmask ^= bitmask << 8;
+  bitmask ^= bitmask << 16;
+  bitmask ^= bitmask << 32;
+  return bitmask;
+}
+
+} // unnamed namespace
+} // namespace lasx
+} // namespace simdjson
+
+#endif
diff --git a/contrib/libs/simdjson/include/simdjson/lasx/implementation.h b/contrib/libs/simdjson/include/simdjson/lasx/implementation.h
new file mode 100644
index 00000000000..8aafbb8b88b
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/lasx/implementation.h
@@ -0,0 +1,31 @@
+#ifndef SIMDJSON_LASX_IMPLEMENTATION_H
+#define SIMDJSON_LASX_IMPLEMENTATION_H
+
+#ifndef SIMDJSON_CONDITIONAL_INCLUDE
+#include "simdjson/base.h"
+#include "simdjson/implementation.h"
+#include "simdjson/internal/instruction_set.h"
+#endif // SIMDJSON_CONDITIONAL_INCLUDE
+
+namespace simdjson {
+namespace lasx {
+
+/**
+ * @private
+ */
+class implementation final : public simdjson::implementation {
+public:
+  simdjson_inline implementation() : simdjson::implementation("lasx", "LoongArch ASX", internal::instruction_set::LASX) {}
+  simdjson_warn_unused error_code create_dom_parser_implementation(
+    size_t capacity,
+    size_t max_length,
+    std::unique_ptr<internal::dom_parser_implementation>& dst
+  ) const noexcept final;
+  simdjson_warn_unused error_code minify(const uint8_t *buf, size_t len, uint8_t *dst, size_t &dst_len) const noexcept final;
+  simdjson_warn_unused bool validate_utf8(const char *buf, size_t len) const noexcept final;
+};
+
+} // namespace lasx
+} // namespace simdjson
+
+#endif // SIMDJSON_LASX_IMPLEMENTATION_H
diff --git a/contrib/libs/simdjson/include/simdjson/lasx/intrinsics.h b/contrib/libs/simdjson/include/simdjson/lasx/intrinsics.h
new file mode 100644
index 00000000000..47b152d9a4d
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/lasx/intrinsics.h
@@ -0,0 +1,14 @@
+#ifndef SIMDJSON_LASX_INTRINSICS_H
+#define SIMDJSON_LASX_INTRINSICS_H
+
+#ifndef SIMDJSON_CONDITIONAL_INCLUDE
+#include "simdjson/lasx/base.h"
+#endif // SIMDJSON_CONDITIONAL_INCLUDE
+
+// This should be the correct header whether
+// you use visual studio or other compilers.
+#error #include <lasxintrin.h>
+
+static_assert(sizeof(__m256i) <= simdjson::SIMDJSON_PADDING, "insufficient padding for LoongArch ASX");
+
+#endif //  SIMDJSON_LASX_INTRINSICS_H
diff --git a/contrib/libs/simdjson/include/simdjson/lasx/numberparsing_defs.h b/contrib/libs/simdjson/include/simdjson/lasx/numberparsing_defs.h
new file mode 100644
index 00000000000..cb9278261c6
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/lasx/numberparsing_defs.h
@@ -0,0 +1,41 @@
+#ifndef SIMDJSON_LASX_NUMBERPARSING_DEFS_H
+#define SIMDJSON_LASX_NUMBERPARSING_DEFS_H
+
+#ifndef SIMDJSON_CONDITIONAL_INCLUDE
+#include "simdjson/lasx/base.h"
+#include "simdjson/lasx/intrinsics.h"
+#include "simdjson/internal/numberparsing_tables.h"
+#endif // SIMDJSON_CONDITIONAL_INCLUDE
+
+#include <cstring>
+
+namespace simdjson {
+namespace lasx {
+namespace numberparsing {
+
+// we don't have appropriate instructions, so let us use a scalar function
+// credit: https://johnnylee-sde.github.io/Fast-numeric-string-to-int/
+/** @private */
+static simdjson_inline uint32_t parse_eight_digits_unrolled(const uint8_t *chars) {
+  uint64_t val;
+  std::memcpy(&val, chars, sizeof(uint64_t));
+  val = (val & 0x0F0F0F0F0F0F0F0F) * 2561 >> 8;
+  val = (val & 0x00FF00FF00FF00FF) * 6553601 >> 16;
+  return uint32_t((val & 0x0000FFFF0000FFFF) * 42949672960001 >> 32);
+}
+
+simdjson_inline internal::value128 full_multiplication(uint64_t value1, uint64_t value2) {
+  internal::value128 answer;
+  __uint128_t r = (static_cast<__uint128_t>(value1)) * value2;
+  answer.low = uint64_t(r);
+  answer.high = uint64_t(r >> 64);
+  return answer;
+}
+
+} // namespace numberparsing
+} // namespace lasx
+} // namespace simdjson
+
+#define SIMDJSON_SWAR_NUMBER_PARSING 1
+
+#endif // SIMDJSON_LASX_NUMBERPARSING_DEFS_H
diff --git a/contrib/libs/simdjson/include/simdjson/lasx/simd.h b/contrib/libs/simdjson/include/simdjson/lasx/simd.h
new file mode 100644
index 00000000000..907a8acb9b3
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/lasx/simd.h
@@ -0,0 +1,376 @@
+#ifndef SIMDJSON_LASX_SIMD_H
+#define SIMDJSON_LASX_SIMD_H
+
+#ifndef SIMDJSON_CONDITIONAL_INCLUDE
+#include "simdjson/lasx/base.h"
+#include "simdjson/lasx/bitmanipulation.h"
+#include "simdjson/internal/simdprune_tables.h"
+#endif // SIMDJSON_CONDITIONAL_INCLUDE
+
+namespace simdjson {
+namespace lasx {
+namespace {
+namespace simd {
+
+  // Forward-declared so they can be used by splat and friends.
+  template<typename Child>
+  struct base {
+    __m256i value;
+
+    // Zero constructor
+    simdjson_inline base() : value{__m256i()} {}
+
+    // Conversion from SIMD register
+    simdjson_inline base(const __m256i _value) : value(_value) {}
+
+    // Conversion to SIMD register
+    simdjson_inline operator const __m256i&() const { return this->value; }
+    simdjson_inline operator __m256i&() { return this->value; }
+    simdjson_inline operator const v32i8&() const { return (v32i8&)this->value; }
+    simdjson_inline operator v32i8&() { return (v32i8&)this->value; }
+
+    // Bit operations
+    simdjson_inline Child operator|(const Child other) const { return __lasx_xvor_v(*this, other); }
+    simdjson_inline Child operator&(const Child other) const { return __lasx_xvand_v(*this, other); }
+    simdjson_inline Child operator^(const Child other) const { return __lasx_xvxor_v(*this, other); }
+    simdjson_inline Child bit_andnot(const Child other) const { return __lasx_xvandn_v(other, *this); }
+    simdjson_inline Child& operator|=(const Child other) { auto this_cast = static_cast<Child*>(this); *this_cast = *this_cast | other; return *this_cast; }
+    simdjson_inline Child& operator&=(const Child other) { auto this_cast = static_cast<Child*>(this); *this_cast = *this_cast & other; return *this_cast; }
+    simdjson_inline Child& operator^=(const Child other) { auto this_cast = static_cast<Child*>(this); *this_cast = *this_cast ^ other; return *this_cast; }
+  };
+
+  // Forward-declared so they can be used by splat and friends.
+  template<typename T>
+  struct simd8;
+
+  template<typename T, typename Mask=simd8<bool>>
+  struct base8: base<simd8<T>> {
+    simdjson_inline base8() : base<simd8<T>>() {}
+    simdjson_inline base8(const __m256i _value) : base<simd8<T>>(_value) {}
+
+    friend simdjson_really_inline Mask operator==(const simd8<T> lhs, const simd8<T> rhs) { return __lasx_xvseq_b(lhs, rhs); }
+
+    static const int SIZE = sizeof(base<simd8<T>>::value);
+
+    template<int N=1>
+    simdjson_inline simd8<T> prev(const simd8<T> prev_chunk) const {
+        __m256i hi = __lasx_xvbsll_v(*this, N);
+        __m256i lo = __lasx_xvbsrl_v(*this, 16 - N);
+        __m256i tmp = __lasx_xvbsrl_v(prev_chunk, 16 - N);
+        lo = __lasx_xvpermi_q(lo, tmp, 0x21);
+        return __lasx_xvor_v(hi, lo);
+    }
+  };
+
+  // SIMD byte mask type (returned by things like eq and gt)
+  template<>
+  struct simd8<bool>: base8<bool> {
+    static simdjson_inline simd8<bool> splat(bool _value) { return __lasx_xvreplgr2vr_b(uint8_t(-(!!_value))); }
+
+    simdjson_inline simd8() : base8() {}
+    simdjson_inline simd8(const __m256i _value) : base8<bool>(_value) {}
+    // Splat constructor
+    simdjson_inline simd8(bool _value) : base8<bool>(splat(_value)) {}
+
+    simdjson_inline int to_bitmask() const {
+      __m256i mask = __lasx_xvmskltz_b(*this);
+      return (__lasx_xvpickve2gr_w(mask, 4) << 16) | (__lasx_xvpickve2gr_w(mask, 0));
+    }
+   simdjson_inline bool any() const {
+      __m256i v = __lasx_xvmsknz_b(*this);
+      return (0 == __lasx_xvpickve2gr_w(v, 0)) && (0 == __lasx_xvpickve2gr_w(v, 4));
+    }
+    simdjson_inline simd8<bool> operator~() const { return *this ^ true; }
+  };
+
+  template<typename T>
+  struct base8_numeric: base8<T> {
+    static simdjson_inline simd8<T> splat(T _value) {
+      return __lasx_xvreplgr2vr_b(_value);
+    }
+    static simdjson_inline simd8<T> zero() { return __lasx_xvldi(0); }
+    static simdjson_inline simd8<T> load(const T values[32]) {
+      return __lasx_xvld(reinterpret_cast<const __m256i *>(values), 0);
+    }
+    // Repeat 16 values as many times as necessary (usually for lookup tables)
+    static simdjson_inline simd8<T> repeat_16(
+      T v0,  T v1,  T v2,  T v3,  T v4,  T v5,  T v6,  T v7,
+      T v8,  T v9,  T v10, T v11, T v12, T v13, T v14, T v15
+    ) {
+      return simd8<T>(
+        v0, v1, v2, v3, v4, v5, v6, v7,
+        v8, v9, v10,v11,v12,v13,v14,v15,
+        v0, v1, v2, v3, v4, v5, v6, v7,
+        v8, v9, v10,v11,v12,v13,v14,v15
+      );
+    }
+
+    simdjson_inline base8_numeric() : base8<T>() {}
+    simdjson_inline base8_numeric(const __m256i _value) : base8<T>(_value) {}
+
+    // Store to array
+    simdjson_inline void store(T dst[32]) const {
+      return __lasx_xvst(*this, reinterpret_cast<__m256i *>(dst), 0);
+    }
+
+    // Addition/subtraction are the same for signed and unsigned
+    simdjson_inline simd8<T> operator+(const simd8<T> other) const { return __lasx_xvadd_b(*this, other); }
+    simdjson_inline simd8<T> operator-(const simd8<T> other) const { return __lasx_xvsub_b(*this, other); }
+    simdjson_inline simd8<T>& operator+=(const simd8<T> other) { *this = *this + other; return *static_cast<simd8<T>*>(this); }
+    simdjson_inline simd8<T>& operator-=(const simd8<T> other) { *this = *this - other; return *static_cast<simd8<T>*>(this); }
+
+    // Override to distinguish from bool version
+    simdjson_inline simd8<T> operator~() const { return *this ^ 0xFFu; }
+
+    // Perform a lookup assuming the value is between 0 and 16 (undefined behavior for out of range values)
+    template<typename L>
+    simdjson_inline simd8<L> lookup_16(simd8<L> lookup_table) const {
+      return __lasx_xvshuf_b(lookup_table, lookup_table, *this);
+    }
+
+    // Copies to 'output" all bytes corresponding to a 0 in the mask (interpreted as a bitset).
+    // Passing a 0 value for mask would be equivalent to writing out every byte to output.
+    // Only the first 16 - count_ones(mask) bytes of the result are significant but 16 bytes
+    // get written.
+    template<typename L>
+    simdjson_inline void compress(uint32_t mask, L * output) const {
+      using internal::thintable_epi8;
+      using internal::BitsSetTable256mul2;
+      using internal::pshufb_combine_table;
+      // this particular implementation was inspired by haswell
+      // lasx do it in 4 steps, first 8 bytes and then second 8 bytes...
+      uint8_t mask1 = uint8_t(mask); // least significant 8 bits
+      uint8_t mask2 = uint8_t(mask >> 8); // second significant 8 bits
+      uint8_t mask3 = uint8_t(mask >> 16); // ...
+      uint8_t mask4 = uint8_t(mask >> 24); // ...
+      // next line just loads the 64-bit values thintable_epi8[mask{1,2,3,4}]
+      // into a 256-bit register.
+      __m256i shufmask = {int64_t(thintable_epi8[mask1]), int64_t(thintable_epi8[mask2]) + 0x0808080808080808, int64_t(thintable_epi8[mask3]), int64_t(thintable_epi8[mask4]) + 0x0808080808080808};
+      // this is the version "nearly pruned"
+      __m256i pruned = __lasx_xvshuf_b(*this, *this, shufmask);
+      // we still need to put the  pieces back together.
+      // we compute the popcount of the first words:
+      int pop1 = BitsSetTable256mul2[mask1];
+      int pop2 = BitsSetTable256mul2[mask2];
+      int pop3 = BitsSetTable256mul2[mask3];
+
+      // then load the corresponding mask
+      __m256i masklo = __lasx_xvldx(reinterpret_cast<void*>(reinterpret_cast<unsigned long>(pshufb_combine_table)), pop1 * 8);
+      __m256i maskhi = __lasx_xvldx(reinterpret_cast<void*>(reinterpret_cast<unsigned long>(pshufb_combine_table)), pop3 * 8);
+      __m256i compactmask = __lasx_xvpermi_q(maskhi, masklo, 0x20);
+      __m256i answer = __lasx_xvshuf_b(pruned, pruned, compactmask);
+      __lasx_xvst(answer, reinterpret_cast<uint8_t*>(output), 0);
+      uint64_t value3 = __lasx_xvpickve2gr_du(answer, 2);
+      uint64_t value4 = __lasx_xvpickve2gr_du(answer, 3);
+      uint64_t *pos = reinterpret_cast<uint64_t*>(reinterpret_cast<uint8_t*>(output) + 16 - (pop1 + pop2) / 2);
+      pos[0] = value3;
+      pos[1] = value4;
+    }
+
+    template<typename L>
+    simdjson_inline simd8<L> lookup_16(
+        L replace0,  L replace1,  L replace2,  L replace3,
+        L replace4,  L replace5,  L replace6,  L replace7,
+        L replace8,  L replace9,  L replace10, L replace11,
+        L replace12, L replace13, L replace14, L replace15) const {
+      return lookup_16(simd8<L>::repeat_16(
+        replace0,  replace1,  replace2,  replace3,
+        replace4,  replace5,  replace6,  replace7,
+        replace8,  replace9,  replace10, replace11,
+        replace12, replace13, replace14, replace15
+      ));
+    }
+  };
+
+  // Signed bytes
+  template<>
+  struct simd8<int8_t> : base8_numeric<int8_t> {
+    simdjson_inline simd8() : base8_numeric<int8_t>() {}
+    simdjson_inline simd8(const __m256i _value) : base8_numeric<int8_t>(_value) {}
+    // Splat constructor
+    simdjson_inline simd8(int8_t _value) : simd8(splat(_value)) {}
+    // Array constructor
+    simdjson_inline simd8(const int8_t values[32]) : simd8(load(values)) {}
+    // Member-by-member initialization
+    simdjson_inline simd8(
+      int8_t v0,  int8_t v1,  int8_t v2,  int8_t v3,  int8_t v4,  int8_t v5,  int8_t v6,  int8_t v7,
+      int8_t v8,  int8_t v9,  int8_t v10, int8_t v11, int8_t v12, int8_t v13, int8_t v14, int8_t v15,
+      int8_t v16, int8_t v17, int8_t v18, int8_t v19, int8_t v20, int8_t v21, int8_t v22, int8_t v23,
+      int8_t v24, int8_t v25, int8_t v26, int8_t v27, int8_t v28, int8_t v29, int8_t v30, int8_t v31
+    ) : simd8({
+      v0, v1, v2, v3, v4, v5, v6, v7,
+      v8, v9, v10,v11,v12,v13,v14,v15,
+      v16,v17,v18,v19,v20,v21,v22,v23,
+      v24,v25,v26,v27,v28,v29,v30,v31
+      }) {}
+    // Repeat 16 values as many times as necessary (usually for lookup tables)
+    simdjson_inline static simd8<int8_t> repeat_16(
+      int8_t v0,  int8_t v1,  int8_t v2,  int8_t v3,  int8_t v4,  int8_t v5,  int8_t v6,  int8_t v7,
+      int8_t v8,  int8_t v9,  int8_t v10, int8_t v11, int8_t v12, int8_t v13, int8_t v14, int8_t v15
+    ) {
+      return simd8<int8_t>(
+        v0, v1, v2, v3, v4, v5, v6, v7,
+        v8, v9, v10,v11,v12,v13,v14,v15,
+        v0, v1, v2, v3, v4, v5, v6, v7,
+        v8, v9, v10,v11,v12,v13,v14,v15
+      );
+    }
+
+    // Order-sensitive comparisons
+    simdjson_inline simd8<int8_t> max_val(const simd8<int8_t> other) const { return __lasx_xvmax_b(*this, other); }
+    simdjson_inline simd8<int8_t> min_val(const simd8<int8_t> other) const { return __lasx_xvmin_b(*this, other); }
+    simdjson_inline simd8<bool> operator>(const simd8<int8_t> other) const { return __lasx_xvslt_b(other, *this); }
+    simdjson_inline simd8<bool> operator<(const simd8<int8_t> other) const { return __lasx_xvslt_b(*this, other); }
+  };
+
+  // Unsigned bytes
+  template<>
+  struct simd8<uint8_t>: base8_numeric<uint8_t> {
+    simdjson_inline simd8() : base8_numeric<uint8_t>() {}
+    simdjson_inline simd8(const __m256i _value) : base8_numeric<uint8_t>(_value) {}
+    // Splat constructor
+    simdjson_inline simd8(uint8_t _value) : simd8(splat(_value)) {}
+    // Array constructor
+    simdjson_inline simd8(const uint8_t values[32]) : simd8(load(values)) {}
+    // Member-by-member initialization
+    simdjson_inline simd8(
+      uint8_t v0,  uint8_t v1,  uint8_t v2,  uint8_t v3,  uint8_t v4,  uint8_t v5,  uint8_t v6,  uint8_t v7,
+      uint8_t v8,  uint8_t v9,  uint8_t v10, uint8_t v11, uint8_t v12, uint8_t v13, uint8_t v14, uint8_t v15,
+      uint8_t v16, uint8_t v17, uint8_t v18, uint8_t v19, uint8_t v20, uint8_t v21, uint8_t v22, uint8_t v23,
+      uint8_t v24, uint8_t v25, uint8_t v26, uint8_t v27, uint8_t v28, uint8_t v29, uint8_t v30, uint8_t v31
+    ) : simd8(__m256i(v32u8{
+      v0, v1, v2, v3, v4, v5, v6, v7,
+      v8, v9, v10,v11,v12,v13,v14,v15,
+      v16,v17,v18,v19,v20,v21,v22,v23,
+      v24,v25,v26,v27,v28,v29,v30,v31
+    })) {}
+    // Repeat 16 values as many times as necessary (usually for lookup tables)
+    simdjson_inline static simd8<uint8_t> repeat_16(
+      uint8_t v0,  uint8_t v1,  uint8_t v2,  uint8_t v3,  uint8_t v4,  uint8_t v5,  uint8_t v6,  uint8_t v7,
+      uint8_t v8,  uint8_t v9,  uint8_t v10, uint8_t v11, uint8_t v12, uint8_t v13, uint8_t v14, uint8_t v15
+    ) {
+      return simd8<uint8_t>(
+        v0, v1, v2, v3, v4, v5, v6, v7,
+        v8, v9, v10,v11,v12,v13,v14,v15,
+        v0, v1, v2, v3, v4, v5, v6, v7,
+        v8, v9, v10,v11,v12,v13,v14,v15
+      );
+    }
+
+    // Saturated math
+    simdjson_inline simd8<uint8_t> saturating_add(const simd8<uint8_t> other) const { return __lasx_xvsadd_bu(*this, other); }
+    simdjson_inline simd8<uint8_t> saturating_sub(const simd8<uint8_t> other) const { return __lasx_xvssub_bu(*this, other); }
+
+    // Order-specific operations
+    simdjson_inline simd8<uint8_t> max_val(const simd8<uint8_t> other) const { return __lasx_xvmax_bu(*this, other); }
+    simdjson_inline simd8<uint8_t> min_val(const simd8<uint8_t> other) const { return __lasx_xvmin_bu(other, *this); }
+    // Same as >, but only guarantees true is nonzero (< guarantees true = -1)
+    simdjson_inline simd8<uint8_t> gt_bits(const simd8<uint8_t> other) const { return this->saturating_sub(other); }
+    // Same as <, but only guarantees true is nonzero (< guarantees true = -1)
+    simdjson_inline simd8<uint8_t> lt_bits(const simd8<uint8_t> other) const { return other.saturating_sub(*this); }
+    simdjson_inline simd8<bool> operator<=(const simd8<uint8_t> other) const { return other.max_val(*this) == other; }
+    simdjson_inline simd8<bool> operator>=(const simd8<uint8_t> other) const { return other.min_val(*this) == other; }
+    simdjson_inline simd8<bool> operator>(const simd8<uint8_t> other) const { return this->gt_bits(other).any_bits_set(); }
+    simdjson_inline simd8<bool> operator<(const simd8<uint8_t> other) const { return this->lt_bits(other).any_bits_set(); }
+
+    // Bit-specific operations
+    simdjson_inline simd8<bool> bits_not_set() const { return *this == uint8_t(0); }
+    simdjson_inline simd8<bool> bits_not_set(simd8<uint8_t> bits) const { return (*this & bits).bits_not_set(); }
+    simdjson_inline simd8<bool> any_bits_set() const { return ~this->bits_not_set(); }
+    simdjson_inline simd8<bool> any_bits_set(simd8<uint8_t> bits) const { return ~this->bits_not_set(bits); }
+    simdjson_inline bool is_ascii() const {
+      __m256i mask = __lasx_xvmskltz_b(*this);
+      return (0 == __lasx_xvpickve2gr_w(mask, 0)) && (0 == __lasx_xvpickve2gr_w(mask, 4));
+    }
+    simdjson_inline bool bits_not_set_anywhere() const {
+      __m256i v = __lasx_xvmsknz_b(*this);
+      return (0 == __lasx_xvpickve2gr_w(v, 0)) && (0 == __lasx_xvpickve2gr_w(v, 4));
+    }
+    simdjson_inline bool any_bits_set_anywhere() const { return !bits_not_set_anywhere(); }
+    simdjson_inline bool bits_not_set_anywhere(simd8<uint8_t> bits) const {
+      __m256i v = __lasx_xvmsknz_b(__lasx_xvand_v(*this, bits));
+      return (0 == __lasx_xvpickve2gr_w(v, 0)) && (0 == __lasx_xvpickve2gr_w(v, 4));
+    }
+    simdjson_inline bool any_bits_set_anywhere(simd8<uint8_t> bits) const { return !bits_not_set_anywhere(bits); }
+    template<int N>
+    simdjson_inline simd8<uint8_t> shr() const { return simd8<uint8_t>(__lasx_xvsrli_b(*this, N)); }
+    template<int N>
+    simdjson_inline simd8<uint8_t> shl() const { return simd8<uint8_t>(__lasx_xvslli_b(*this, N)); }
+  };
+
+  template<typename T>
+  struct simd8x64 {
+    static constexpr int NUM_CHUNKS = 64 / sizeof(simd8<T>);
+    static_assert(NUM_CHUNKS == 2, "LASX kernel should use two registers per 64-byte block.");
+    const simd8<T> chunks[NUM_CHUNKS];
+
+    simd8x64(const simd8x64<T>& o) = delete; // no copy allowed
+    simd8x64<T>& operator=(const simd8<T>& other) = delete; // no assignment allowed
+    simd8x64() = delete; // no default constructor allowed
+
+    simdjson_inline simd8x64(const simd8<T> chunk0, const simd8<T> chunk1) : chunks{chunk0, chunk1} {}
+    simdjson_inline simd8x64(const T ptr[64]) : chunks{simd8<T>::load(ptr), simd8<T>::load(ptr+32)} {}
+
+    simdjson_inline uint64_t compress(uint64_t mask, T * output) const {
+      uint32_t mask1 = uint32_t(mask);
+      uint32_t mask2 = uint32_t(mask >> 32);
+      __m256i zcnt = __lasx_xvpcnt_w(__m256i(v4u64{~mask, 0, 0, 0}));
+      uint64_t zcnt1 = __lasx_xvpickve2gr_wu(zcnt, 0);
+      uint64_t zcnt2 = __lasx_xvpickve2gr_wu(zcnt, 1);
+      // There should be a critical value which processes in scaler is faster.
+      if (zcnt1)
+        this->chunks[0].compress(mask1, output);
+      if (zcnt2)
+        this->chunks[1].compress(mask2, output + zcnt1);
+      return zcnt1 + zcnt2;
+    }
+
+    simdjson_inline void store(T ptr[64]) const {
+      this->chunks[0].store(ptr+sizeof(simd8<T>)*0);
+      this->chunks[1].store(ptr+sizeof(simd8<T>)*1);
+    }
+
+    simdjson_inline uint64_t to_bitmask() const {
+      __m256i mask0 = __lasx_xvmskltz_b(this->chunks[0]);
+      __m256i mask1 = __lasx_xvmskltz_b(this->chunks[1]);
+      __m256i mask_tmp = __lasx_xvpickve_w(mask0, 4);
+      __m256i tmp = __lasx_xvpickve_w(mask1, 4);
+      mask0 = __lasx_xvinsve0_w(mask0, mask1, 1);
+      mask_tmp = __lasx_xvinsve0_w(mask_tmp, tmp, 1);
+      return __lasx_xvpickve2gr_du(__lasx_xvpackev_h(mask_tmp, mask0), 0);
+    }
+
+    simdjson_inline simd8<T> reduce_or() const {
+      return this->chunks[0] | this->chunks[1];
+    }
+
+    simdjson_inline uint64_t eq(const T m) const {
+      const simd8<T> mask = simd8<T>::splat(m);
+      return  simd8x64<bool>(
+        this->chunks[0] == mask,
+        this->chunks[1] == mask
+      ).to_bitmask();
+    }
+
+    simdjson_inline uint64_t eq(const simd8x64<uint8_t> &other) const {
+      return  simd8x64<bool>(
+        this->chunks[0] == other.chunks[0],
+        this->chunks[1] == other.chunks[1]
+      ).to_bitmask();
+    }
+
+    simdjson_inline uint64_t lteq(const T m) const {
+      const simd8<T> mask = simd8<T>::splat(m);
+      return  simd8x64<bool>(
+        this->chunks[0] <= mask,
+        this->chunks[1] <= mask
+      ).to_bitmask();
+    }
+  }; // struct simd8x64<T>
+
+} // namespace simd
+} // unnamed namespace
+} // namespace lasx
+} // namespace simdjson
+
+#endif // SIMDJSON_LASX_SIMD_H
diff --git a/contrib/libs/simdjson/include/simdjson/lasx/stringparsing_defs.h b/contrib/libs/simdjson/include/simdjson/lasx/stringparsing_defs.h
new file mode 100644
index 00000000000..fe7a7430e00
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/lasx/stringparsing_defs.h
@@ -0,0 +1,47 @@
+#ifndef SIMDJSON_LASX_STRINGPARSING_DEFS_H
+#define SIMDJSON_LASX_STRINGPARSING_DEFS_H
+
+#ifndef SIMDJSON_CONDITIONAL_INCLUDE
+#include "simdjson/lasx/base.h"
+#include "simdjson/lasx/simd.h"
+#include "simdjson/lasx/bitmanipulation.h"
+#endif // SIMDJSON_CONDITIONAL_INCLUDE
+
+namespace simdjson {
+namespace lasx {
+namespace {
+
+using namespace simd;
+
+// Holds backslashes and quotes locations.
+struct backslash_and_quote {
+public:
+  static constexpr uint32_t BYTES_PROCESSED = 32;
+  simdjson_inline static backslash_and_quote copy_and_find(const uint8_t *src, uint8_t *dst);
+
+  simdjson_inline bool has_quote_first() { return ((bs_bits - 1) & quote_bits) != 0; }
+  simdjson_inline bool has_backslash() { return bs_bits != 0; }
+  simdjson_inline int quote_index() { return trailing_zeroes(quote_bits); }
+  simdjson_inline int backslash_index() { return trailing_zeroes(bs_bits); }
+
+  uint32_t bs_bits;
+  uint32_t quote_bits;
+}; // struct backslash_and_quote
+
+simdjson_inline backslash_and_quote backslash_and_quote::copy_and_find(const uint8_t *src, uint8_t *dst) {
+  // this can read up to 31 bytes beyond the buffer size, but we require
+  // SIMDJSON_PADDING of padding
+  static_assert(SIMDJSON_PADDING >= (BYTES_PROCESSED - 1), "backslash and quote finder must process fewer than SIMDJSON_PADDING bytes");
+  simd8<uint8_t> v(src);
+  v.store(dst);
+  return {
+      static_cast<uint32_t>((v == '\\').to_bitmask()),     // bs_bits
+      static_cast<uint32_t>((v == '"').to_bitmask()), // quote_bits
+  };
+}
+
+} // unnamed namespace
+} // namespace lasx
+} // namespace simdjson
+
+#endif // SIMDJSON_LASX_STRINGPARSING_DEFS_H
diff --git a/contrib/libs/simdjson/include/simdjson/lsx/base.h b/contrib/libs/simdjson/include/simdjson/lsx/base.h
new file mode 100644
index 00000000000..ff02450184a
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/lsx/base.h
@@ -0,0 +1,26 @@
+#ifndef SIMDJSON_LSX_BASE_H
+#define SIMDJSON_LSX_BASE_H
+
+#ifndef SIMDJSON_CONDITIONAL_INCLUDE
+#include "simdjson/base.h"
+#endif // SIMDJSON_CONDITIONAL_INCLUDE
+
+namespace simdjson {
+/**
+ * Implementation for LSX.
+ */
+namespace lsx {
+
+class implementation;
+
+namespace {
+namespace simd {
+template <typename T> struct simd8;
+template <typename T> struct simd8x64;
+} // namespace simd
+} // unnamed namespace
+
+} // namespace lsx
+} // namespace simdjson
+
+#endif // SIMDJSON_LSX_BASE_H
diff --git a/contrib/libs/simdjson/include/simdjson/lsx/begin.h b/contrib/libs/simdjson/include/simdjson/lsx/begin.h
new file mode 100644
index 00000000000..78a92819a34
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/lsx/begin.h
@@ -0,0 +1,10 @@
+#define SIMDJSON_IMPLEMENTATION lsx
+#include "simdjson/lsx/base.h"
+#include "simdjson/lsx/intrinsics.h"
+#include "simdjson/lsx/bitmanipulation.h"
+#include "simdjson/lsx/bitmask.h"
+#include "simdjson/lsx/numberparsing_defs.h"
+#include "simdjson/lsx/simd.h"
+#include "simdjson/lsx/stringparsing_defs.h"
+
+#define SIMDJSON_SKIP_BACKSLASH_SHORT_CIRCUIT 1
diff --git a/contrib/libs/simdjson/include/simdjson/lsx/bitmanipulation.h b/contrib/libs/simdjson/include/simdjson/lsx/bitmanipulation.h
new file mode 100644
index 00000000000..96e1794bae5
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/lsx/bitmanipulation.h
@@ -0,0 +1,50 @@
+#ifndef SIMDJSON_LSX_BITMANIPULATION_H
+#define SIMDJSON_LSX_BITMANIPULATION_H
+
+#ifndef SIMDJSON_CONDITIONAL_INCLUDE
+#include "simdjson/lsx/base.h"
+#include "simdjson/lsx/intrinsics.h"
+#include "simdjson/lsx/bitmask.h"
+#endif // SIMDJSON_CONDITIONAL_INCLUDE
+
+namespace simdjson {
+namespace lsx {
+namespace {
+
+// We sometimes call trailing_zero on inputs that are zero,
+// but the algorithms do not end up using the returned value.
+// Sadly, sanitizers are not smart enough to figure it out.
+SIMDJSON_NO_SANITIZE_UNDEFINED
+// This function can be used safely even if not all bytes have been
+// initialized.
+// See issue https://github.com/simdjson/simdjson/issues/1965
+SIMDJSON_NO_SANITIZE_MEMORY
+simdjson_inline int trailing_zeroes(uint64_t input_num) {
+  return __builtin_ctzll(input_num);
+}
+
+/* result might be undefined when input_num is zero */
+simdjson_inline uint64_t clear_lowest_bit(uint64_t input_num) {
+  return input_num & (input_num-1);
+}
+
+/* result might be undefined when input_num is zero */
+simdjson_inline int leading_zeroes(uint64_t input_num) {
+  return __builtin_clzll(input_num);
+}
+
+/* result might be undefined when input_num is zero */
+simdjson_inline int count_ones(uint64_t input_num) {
+  return __lsx_vpickve2gr_w(__lsx_vpcnt_d(__m128i(v2u64{input_num, 0})), 0);
+}
+
+simdjson_inline bool add_overflow(uint64_t value1, uint64_t value2, uint64_t *result) {
+  return __builtin_uaddll_overflow(value1, value2,
+                                   reinterpret_cast<unsigned long long *>(result));
+}
+
+} // unnamed namespace
+} // namespace lsx
+} // namespace simdjson
+
+#endif // SIMDJSON_LSX_BITMANIPULATION_H
diff --git a/contrib/libs/simdjson/include/simdjson/lsx/bitmask.h b/contrib/libs/simdjson/include/simdjson/lsx/bitmask.h
new file mode 100644
index 00000000000..3a9f0d768c5
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/lsx/bitmask.h
@@ -0,0 +1,31 @@
+#ifndef SIMDJSON_LSX_BITMASK_H
+#define SIMDJSON_LSX_BITMASK_H
+
+#ifndef SIMDJSON_CONDITIONAL_INCLUDE
+#include "simdjson/lsx/base.h"
+#endif // SIMDJSON_CONDITIONAL_INCLUDE
+
+namespace simdjson {
+namespace lsx {
+namespace {
+
+//
+// Perform a "cumulative bitwise xor," flipping bits each time a 1 is encountered.
+//
+// For example, prefix_xor(00100100) == 00011100
+//
+simdjson_inline uint64_t prefix_xor(uint64_t bitmask) {
+  bitmask ^= bitmask << 1;
+  bitmask ^= bitmask << 2;
+  bitmask ^= bitmask << 4;
+  bitmask ^= bitmask << 8;
+  bitmask ^= bitmask << 16;
+  bitmask ^= bitmask << 32;
+  return bitmask;
+}
+
+} // unnamed namespace
+} // namespace lsx
+} // namespace simdjson
+
+#endif
diff --git a/contrib/libs/simdjson/include/simdjson/lsx/implementation.h b/contrib/libs/simdjson/include/simdjson/lsx/implementation.h
new file mode 100644
index 00000000000..14468777de7
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/lsx/implementation.h
@@ -0,0 +1,31 @@
+#ifndef SIMDJSON_LSX_IMPLEMENTATION_H
+#define SIMDJSON_LSX_IMPLEMENTATION_H
+
+#ifndef SIMDJSON_CONDITIONAL_INCLUDE
+#include "simdjson/base.h"
+#include "simdjson/implementation.h"
+#include "simdjson/internal/instruction_set.h"
+#endif // SIMDJSON_CONDITIONAL_INCLUDE
+
+namespace simdjson {
+namespace lsx {
+
+/**
+ * @private
+ */
+class implementation final : public simdjson::implementation {
+public:
+  simdjson_inline implementation() : simdjson::implementation("lsx", "LoongArch SX", internal::instruction_set::LSX) {}
+  simdjson_warn_unused error_code create_dom_parser_implementation(
+    size_t capacity,
+    size_t max_length,
+    std::unique_ptr<internal::dom_parser_implementation>& dst
+  ) const noexcept final;
+  simdjson_warn_unused error_code minify(const uint8_t *buf, size_t len, uint8_t *dst, size_t &dst_len) const noexcept final;
+  simdjson_warn_unused bool validate_utf8(const char *buf, size_t len) const noexcept final;
+};
+
+} // namespace lsx
+} // namespace simdjson
+
+#endif // SIMDJSON_LSX_IMPLEMENTATION_H
diff --git a/contrib/libs/simdjson/include/simdjson/lsx/intrinsics.h b/contrib/libs/simdjson/include/simdjson/lsx/intrinsics.h
new file mode 100644
index 00000000000..c4cf1dd6094
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/lsx/intrinsics.h
@@ -0,0 +1,14 @@
+#ifndef SIMDJSON_LSX_INTRINSICS_H
+#define SIMDJSON_LSX_INTRINSICS_H
+
+#ifndef SIMDJSON_CONDITIONAL_INCLUDE
+#include "simdjson/lsx/base.h"
+#endif // SIMDJSON_CONDITIONAL_INCLUDE
+
+// This should be the correct header whether
+// you use visual studio or other compilers.
+#error #include <lsxintrin.h>
+
+static_assert(sizeof(__m128i) <= simdjson::SIMDJSON_PADDING, "insufficient padding for LoongArch SX");
+
+#endif //  SIMDJSON_LSX_INTRINSICS_H
diff --git a/contrib/libs/simdjson/include/simdjson/lsx/numberparsing_defs.h b/contrib/libs/simdjson/include/simdjson/lsx/numberparsing_defs.h
new file mode 100644
index 00000000000..1dfe4264e92
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/lsx/numberparsing_defs.h
@@ -0,0 +1,41 @@
+#ifndef SIMDJSON_LSX_NUMBERPARSING_DEFS_H
+#define SIMDJSON_LSX_NUMBERPARSING_DEFS_H
+
+#ifndef SIMDJSON_CONDITIONAL_INCLUDE
+#include "simdjson/lsx/base.h"
+#include "simdjson/lsx/intrinsics.h"
+#include "simdjson/internal/numberparsing_tables.h"
+#endif // SIMDJSON_CONDITIONAL_INCLUDE
+
+#include <cstring>
+
+namespace simdjson {
+namespace lsx {
+namespace numberparsing {
+
+// we don't have appropriate instructions, so let us use a scalar function
+// credit: https://johnnylee-sde.github.io/Fast-numeric-string-to-int/
+/** @private */
+static simdjson_inline uint32_t parse_eight_digits_unrolled(const uint8_t *chars) {
+  uint64_t val;
+  std::memcpy(&val, chars, sizeof(uint64_t));
+  val = (val & 0x0F0F0F0F0F0F0F0F) * 2561 >> 8;
+  val = (val & 0x00FF00FF00FF00FF) * 6553601 >> 16;
+  return uint32_t((val & 0x0000FFFF0000FFFF) * 42949672960001 >> 32);
+}
+
+simdjson_inline internal::value128 full_multiplication(uint64_t value1, uint64_t value2) {
+  internal::value128 answer;
+  __uint128_t r = (static_cast<__uint128_t>(value1)) * value2;
+  answer.low = uint64_t(r);
+  answer.high = uint64_t(r >> 64);
+  return answer;
+}
+
+} // namespace numberparsing
+} // namespace lsx
+} // namespace simdjson
+
+#define SIMDJSON_SWAR_NUMBER_PARSING 1
+
+#endif // SIMDJSON_LSX_NUMBERPARSING_DEFS_H
diff --git a/contrib/libs/simdjson/include/simdjson/lsx/simd.h b/contrib/libs/simdjson/include/simdjson/lsx/simd.h
new file mode 100644
index 00000000000..3f0d66560d3
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/lsx/simd.h
@@ -0,0 +1,354 @@
+#ifndef SIMDJSON_LSX_SIMD_H
+#define SIMDJSON_LSX_SIMD_H
+
+#ifndef SIMDJSON_CONDITIONAL_INCLUDE
+#include "simdjson/lsx/base.h"
+#include "simdjson/lsx/bitmanipulation.h"
+#include "simdjson/internal/simdprune_tables.h"
+#endif // SIMDJSON_CONDITIONAL_INCLUDE
+
+namespace simdjson {
+namespace lsx {
+namespace {
+namespace simd {
+
+  // Forward-declared so they can be used by splat and friends.
+  template<typename Child>
+  struct base {
+    __m128i value;
+
+    // Zero constructor
+    simdjson_inline base() : value{__m128i()} {}
+
+    // Conversion from SIMD register
+    simdjson_inline base(const __m128i _value) : value(_value) {}
+
+    // Conversion to SIMD register
+    simdjson_inline operator const __m128i&() const { return this->value; }
+    simdjson_inline operator __m128i&() { return this->value; }
+    simdjson_inline operator const v16i8&() const { return (v16i8&)this->value; }
+    simdjson_inline operator v16i8&() { return (v16i8&)this->value; }
+
+    // Bit operations
+    simdjson_inline Child operator|(const Child other) const { return __lsx_vor_v(*this, other); }
+    simdjson_inline Child operator&(const Child other) const { return __lsx_vand_v(*this, other); }
+    simdjson_inline Child operator^(const Child other) const { return __lsx_vxor_v(*this, other); }
+    simdjson_inline Child bit_andnot(const Child other) const { return __lsx_vandn_v(other, *this); }
+    simdjson_inline Child& operator|=(const Child other) { auto this_cast = static_cast<Child*>(this); *this_cast = *this_cast | other; return *this_cast; }
+    simdjson_inline Child& operator&=(const Child other) { auto this_cast = static_cast<Child*>(this); *this_cast = *this_cast & other; return *this_cast; }
+    simdjson_inline Child& operator^=(const Child other) { auto this_cast = static_cast<Child*>(this); *this_cast = *this_cast ^ other; return *this_cast; }
+  };
+
+  // Forward-declared so they can be used by splat and friends.
+  template<typename T>
+  struct simd8;
+
+  template<typename T, typename Mask=simd8<bool>>
+  struct base8: base<simd8<T>> {
+    simdjson_inline base8() : base<simd8<T>>() {}
+    simdjson_inline base8(const __m128i _value) : base<simd8<T>>(_value) {}
+
+    friend simdjson_really_inline Mask operator==(const simd8<T> lhs, const simd8<T> rhs) { return __lsx_vseq_b(lhs, rhs); }
+
+    static const int SIZE = sizeof(base<simd8<T>>::value);
+
+    template<int N=1>
+    simdjson_inline simd8<T> prev(const simd8<T> prev_chunk) const {
+      return __lsx_vor_v(__lsx_vbsll_v(*this, N), __lsx_vbsrl_v(prev_chunk, 16 - N));
+    }
+  };
+
+  // SIMD byte mask type (returned by things like eq and gt)
+  template<>
+  struct simd8<bool>: base8<bool> {
+    static simdjson_inline simd8<bool> splat(bool _value) {
+      return __lsx_vreplgr2vr_b(uint8_t(-(!!_value)));
+    }
+
+    simdjson_inline simd8() : base8() {}
+    simdjson_inline simd8(const __m128i _value) : base8<bool>(_value) {}
+    // Splat constructor
+    simdjson_inline simd8(bool _value) : base8<bool>(splat(_value)) {}
+
+    simdjson_inline int to_bitmask() const { return __lsx_vpickve2gr_w(__lsx_vmskltz_b(*this), 0); }
+    simdjson_inline bool any() const { return 0 == __lsx_vpickve2gr_hu(__lsx_vmsknz_b(*this), 0); }
+    simdjson_inline simd8<bool> operator~() const { return *this ^ true; }
+  };
+
+  template<typename T>
+  struct base8_numeric: base8<T> {
+    static simdjson_inline simd8<T> splat(T _value) { return __lsx_vreplgr2vr_b(_value); }
+    static simdjson_inline simd8<T> zero() { return __lsx_vldi(0); }
+    static simdjson_inline simd8<T> load(const T values[16]) {
+      return __lsx_vld(reinterpret_cast<const __m128i *>(values), 0);
+    }
+    // Repeat 16 values as many times as necessary (usually for lookup tables)
+    static simdjson_inline simd8<T> repeat_16(
+      T v0,  T v1,  T v2,  T v3,  T v4,  T v5,  T v6,  T v7,
+      T v8,  T v9,  T v10, T v11, T v12, T v13, T v14, T v15
+    ) {
+      return simd8<T>(
+        v0, v1, v2, v3, v4, v5, v6, v7,
+        v8, v9, v10,v11,v12,v13,v14,v15
+      );
+    }
+
+    simdjson_inline base8_numeric() : base8<T>() {}
+    simdjson_inline base8_numeric(const __m128i _value) : base8<T>(_value) {}
+
+    // Store to array
+    simdjson_inline void store(T dst[16]) const {
+      return __lsx_vst(*this, reinterpret_cast<__m128i *>(dst), 0);
+    }
+
+    // Addition/subtraction are the same for signed and unsigned
+    simdjson_inline simd8<T> operator+(const simd8<T> other) const { return __lsx_vadd_b(*this, other); }
+    simdjson_inline simd8<T> operator-(const simd8<T> other) const { return __lsx_vsub_b(*this, other); }
+    simdjson_inline simd8<T>& operator+=(const simd8<T> other) { *this = *this + other; return *static_cast<simd8<T>*>(this); }
+    simdjson_inline simd8<T>& operator-=(const simd8<T> other) { *this = *this - other; return *static_cast<simd8<T>*>(this); }
+
+    // Override to distinguish from bool version
+    simdjson_inline simd8<T> operator~() const { return *this ^ 0xFFu; }
+
+    // Perform a lookup assuming the value is between 0 and 16 (undefined behavior for out of range values)
+    template<typename L>
+    simdjson_inline simd8<L> lookup_16(simd8<L> lookup_table) const {
+      return __lsx_vshuf_b(lookup_table, lookup_table, *this);
+    }
+
+    // Copies to 'output" all bytes corresponding to a 0 in the mask (interpreted as a bitset).
+    // Passing a 0 value for mask would be equivalent to writing out every byte to output.
+    // Only the first 16 - count_ones(mask) bytes of the result are significant but 16 bytes
+    // get written.
+    template<typename L>
+    simdjson_inline void compress(uint16_t mask, L * output) const {
+      using internal::thintable_epi8;
+      using internal::BitsSetTable256mul2;
+      using internal::pshufb_combine_table;
+      // this particular implementation was inspired by haswell
+      // lsx do it in 2 steps, first 8 bytes and then second 8 bytes...
+      uint8_t mask1 = uint8_t(mask); // least significant 8 bits
+      uint8_t mask2 = uint8_t(mask >> 8); // second least significant 8 bits
+      // next line just loads the 64-bit values thintable_epi8[mask1] and
+      // thintable_epi8[mask2] into a 128-bit register.
+      __m128i shufmask = {int64_t(thintable_epi8[mask1]), int64_t(thintable_epi8[mask2]) + 0x0808080808080808};
+      // this is the version "nearly pruned"
+      __m128i pruned = __lsx_vshuf_b(*this, *this, shufmask);
+      // we still need to put the  pieces back together.
+      // we compute the popcount of the first words:
+      int pop1 = BitsSetTable256mul2[mask1];
+      // then load the corresponding mask
+      __m128i compactmask = __lsx_vldx(reinterpret_cast<void*>(reinterpret_cast<unsigned long>(pshufb_combine_table)), pop1 * 8);
+      __m128i answer = __lsx_vshuf_b(pruned, pruned, compactmask);
+      __lsx_vst(answer, reinterpret_cast<uint8_t*>(output), 0);
+    }
+
+    template<typename L>
+    simdjson_inline simd8<L> lookup_16(
+        L replace0,  L replace1,  L replace2,  L replace3,
+        L replace4,  L replace5,  L replace6,  L replace7,
+        L replace8,  L replace9,  L replace10, L replace11,
+        L replace12, L replace13, L replace14, L replace15) const {
+      return lookup_16(simd8<L>::repeat_16(
+        replace0,  replace1,  replace2,  replace3,
+        replace4,  replace5,  replace6,  replace7,
+        replace8,  replace9,  replace10, replace11,
+        replace12, replace13, replace14, replace15
+      ));
+    }
+  };
+
+  // Signed bytes
+  template<>
+  struct simd8<int8_t> : base8_numeric<int8_t> {
+    simdjson_inline simd8() : base8_numeric<int8_t>() {}
+    simdjson_inline simd8(const __m128i _value) : base8_numeric<int8_t>(_value) {}
+    // Splat constructor
+    simdjson_inline simd8(int8_t _value) : simd8(splat(_value)) {}
+    // Array constructor
+    simdjson_inline simd8(const int8_t values[16]) : simd8(load(values)) {}
+    // Member-by-member initialization
+    simdjson_inline simd8(
+      int8_t v0,  int8_t v1,  int8_t v2,  int8_t v3,  int8_t v4,  int8_t v5,  int8_t v6,  int8_t v7,
+      int8_t v8,  int8_t v9,  int8_t v10, int8_t v11, int8_t v12, int8_t v13, int8_t v14, int8_t v15
+    ) : simd8({
+      v0, v1, v2, v3, v4, v5, v6, v7,
+      v8, v9, v10,v11,v12,v13,v14,v15
+      }) {}
+    // Repeat 16 values as many times as necessary (usually for lookup tables)
+    simdjson_inline static simd8<int8_t> repeat_16(
+      int8_t v0,  int8_t v1,  int8_t v2,  int8_t v3,  int8_t v4,  int8_t v5,  int8_t v6,  int8_t v7,
+      int8_t v8,  int8_t v9,  int8_t v10, int8_t v11, int8_t v12, int8_t v13, int8_t v14, int8_t v15
+    ) {
+      return simd8<int8_t>(
+        v0, v1, v2, v3, v4, v5, v6, v7,
+        v8, v9, v10,v11,v12,v13,v14,v15
+      );
+    }
+
+    // Order-sensitive comparisons
+    simdjson_inline simd8<int8_t> max_val(const simd8<int8_t> other) const { return __lsx_vmax_b(*this, other); }
+    simdjson_inline simd8<int8_t> min_val(const simd8<int8_t> other) const { return __lsx_vmin_b(*this, other); }
+    simdjson_inline simd8<bool> operator>(const simd8<int8_t> other) const { return __lsx_vslt_b(other, *this); }
+    simdjson_inline simd8<bool> operator<(const simd8<int8_t> other) const { return __lsx_vslt_b(*this, other); }
+  };
+
+  // Unsigned bytes
+  template<>
+  struct simd8<uint8_t>: base8_numeric<uint8_t> {
+    simdjson_inline simd8() : base8_numeric<uint8_t>() {}
+    simdjson_inline simd8(const __m128i _value) : base8_numeric<uint8_t>(_value) {}
+    // Splat constructor
+    simdjson_inline simd8(uint8_t _value) : simd8(splat(_value)) {}
+    // Array constructor
+    simdjson_inline simd8(const uint8_t values[16]) : simd8(load(values)) {}
+    // Member-by-member initialization
+    simdjson_inline simd8(
+      uint8_t v0,  uint8_t v1,  uint8_t v2,  uint8_t v3,  uint8_t v4,  uint8_t v5,  uint8_t v6,  uint8_t v7,
+      uint8_t v8,  uint8_t v9,  uint8_t v10, uint8_t v11, uint8_t v12, uint8_t v13, uint8_t v14, uint8_t v15
+    ) : simd8(__m128i(v16u8{
+      v0, v1, v2, v3, v4, v5, v6, v7,
+      v8, v9, v10,v11,v12,v13,v14,v15
+    })) {}
+    // Repeat 16 values as many times as necessary (usually for lookup tables)
+    simdjson_inline static simd8<uint8_t> repeat_16(
+      uint8_t v0,  uint8_t v1,  uint8_t v2,  uint8_t v3,  uint8_t v4,  uint8_t v5,  uint8_t v6,  uint8_t v7,
+      uint8_t v8,  uint8_t v9,  uint8_t v10, uint8_t v11, uint8_t v12, uint8_t v13, uint8_t v14, uint8_t v15
+    ) {
+      return simd8<uint8_t>(
+        v0, v1, v2, v3, v4, v5, v6, v7,
+        v8, v9, v10,v11,v12,v13,v14,v15
+      );
+    }
+
+    // Saturated math
+    simdjson_inline simd8<uint8_t> saturating_add(const simd8<uint8_t> other) const { return __lsx_vsadd_bu(*this, other); }
+    simdjson_inline simd8<uint8_t> saturating_sub(const simd8<uint8_t> other) const { return __lsx_vssub_bu(*this, other); }
+
+    // Order-specific operations
+    simdjson_inline simd8<uint8_t> max_val(const simd8<uint8_t> other) const { return __lsx_vmax_bu(*this, other); }
+    simdjson_inline simd8<uint8_t> min_val(const simd8<uint8_t> other) const { return __lsx_vmin_bu(other, *this); }
+    // Same as >, but only guarantees true is nonzero (< guarantees true = -1)
+    simdjson_inline simd8<uint8_t> gt_bits(const simd8<uint8_t> other) const { return this->saturating_sub(other); }
+    // Same as <, but only guarantees true is nonzero (< guarantees true = -1)
+    simdjson_inline simd8<uint8_t> lt_bits(const simd8<uint8_t> other) const { return other.saturating_sub(*this); }
+    simdjson_inline simd8<bool> operator<=(const simd8<uint8_t> other) const { return other.max_val(*this) == other; }
+    simdjson_inline simd8<bool> operator>=(const simd8<uint8_t> other) const { return other.min_val(*this) == other; }
+    simdjson_inline simd8<bool> operator>(const simd8<uint8_t> other) const { return this->gt_bits(other).any_bits_set(); }
+    simdjson_inline simd8<bool> operator<(const simd8<uint8_t> other) const { return this->lt_bits(other).any_bits_set(); }
+
+    // Bit-specific operations
+    simdjson_inline simd8<bool> bits_not_set() const { return *this == uint8_t(0); }
+    simdjson_inline simd8<bool> bits_not_set(simd8<uint8_t> bits) const { return (*this & bits).bits_not_set(); }
+    simdjson_inline simd8<bool> any_bits_set() const { return ~this->bits_not_set(); }
+    simdjson_inline simd8<bool> any_bits_set(simd8<uint8_t> bits) const { return ~this->bits_not_set(bits); }
+    simdjson_inline bool is_ascii() const { return 0 == __lsx_vpickve2gr_w(__lsx_vmskltz_b(*this), 0); }
+    simdjson_inline bool bits_not_set_anywhere() const { return 0 == __lsx_vpickve2gr_hu(__lsx_vmsknz_b(*this), 0); }
+    simdjson_inline bool any_bits_set_anywhere() const { return !bits_not_set_anywhere(); }
+    simdjson_inline bool bits_not_set_anywhere(simd8<uint8_t> bits) const {
+      return 0 == __lsx_vpickve2gr_hu(__lsx_vmsknz_b(__lsx_vand_v(*this, bits)), 0);
+    }
+    simdjson_inline bool any_bits_set_anywhere(simd8<uint8_t> bits) const { return !bits_not_set_anywhere(bits); }
+    template<int N>
+    simdjson_inline simd8<uint8_t> shr() const { return simd8<uint8_t>(__lsx_vsrli_b(*this, N)); }
+    template<int N>
+    simdjson_inline simd8<uint8_t> shl() const { return simd8<uint8_t>(__lsx_vslli_b(*this, N)); }
+  };
+
+  template<typename T>
+  struct simd8x64 {
+    static constexpr int NUM_CHUNKS = 64 / sizeof(simd8<T>);
+    static_assert(NUM_CHUNKS == 4, "LSX kernel should use four registers per 64-byte block.");
+    const simd8<T> chunks[NUM_CHUNKS];
+
+    simd8x64(const simd8x64<T>& o) = delete; // no copy allowed
+    simd8x64<T>& operator=(const simd8<T>& other) = delete; // no assignment allowed
+    simd8x64() = delete; // no default constructor allowed
+
+    simdjson_inline simd8x64(const simd8<T> chunk0, const simd8<T> chunk1, const simd8<T> chunk2, const simd8<T> chunk3) : chunks{chunk0, chunk1, chunk2, chunk3} {}
+    simdjson_inline simd8x64(const T ptr[64]) : chunks{simd8<T>::load(ptr), simd8<T>::load(ptr+16), simd8<T>::load(ptr+32), simd8<T>::load(ptr+48)} {}
+
+    simdjson_inline uint64_t compress(uint64_t mask, T * output) const {
+      uint16_t mask1 = uint16_t(mask);
+      uint16_t mask2 = uint16_t(mask >> 16);
+      uint16_t mask3 = uint16_t(mask >> 32);
+      uint16_t mask4 = uint16_t(mask >> 48);
+      __m128i zcnt = __lsx_vpcnt_h(__m128i(v2u64{~mask, 0}));
+      uint64_t zcnt1 = __lsx_vpickve2gr_hu(zcnt, 0);
+      uint64_t zcnt2 = __lsx_vpickve2gr_hu(zcnt, 1);
+      uint64_t zcnt3 = __lsx_vpickve2gr_hu(zcnt, 2);
+      uint64_t zcnt4 = __lsx_vpickve2gr_hu(zcnt, 3);
+      uint8_t *voutput = reinterpret_cast<uint8_t*>(output);
+      // There should be a critical value which processes in scaler is faster.
+      if (zcnt1)
+        this->chunks[0].compress(mask1, reinterpret_cast<T*>(voutput));
+      voutput += zcnt1;
+      if (zcnt2)
+        this->chunks[1].compress(mask2, reinterpret_cast<T*>(voutput));
+      voutput += zcnt2;
+      if (zcnt3)
+        this->chunks[2].compress(mask3, reinterpret_cast<T*>(voutput));
+      voutput += zcnt3;
+      if (zcnt4)
+        this->chunks[3].compress(mask4, reinterpret_cast<T*>(voutput));
+      voutput += zcnt4;
+      return reinterpret_cast<uint64_t>(voutput) - reinterpret_cast<uint64_t>(output);
+    }
+
+    simdjson_inline void store(T ptr[64]) const {
+      this->chunks[0].store(ptr+sizeof(simd8<T>)*0);
+      this->chunks[1].store(ptr+sizeof(simd8<T>)*1);
+      this->chunks[2].store(ptr+sizeof(simd8<T>)*2);
+      this->chunks[3].store(ptr+sizeof(simd8<T>)*3);
+    }
+
+    simdjson_inline uint64_t to_bitmask() const {
+      __m128i mask1 = __lsx_vmskltz_b(this->chunks[0]);
+      __m128i mask2 = __lsx_vmskltz_b(this->chunks[1]);
+      __m128i mask3 = __lsx_vmskltz_b(this->chunks[2]);
+      __m128i mask4 = __lsx_vmskltz_b(this->chunks[3]);
+      mask1 = __lsx_vilvl_h(mask2, mask1);
+      mask2 = __lsx_vilvl_h(mask4, mask3);
+      return __lsx_vpickve2gr_du(__lsx_vilvl_w(mask2, mask1), 0);
+    }
+
+    simdjson_inline simd8<T> reduce_or() const {
+      return (this->chunks[0] | this->chunks[1]) | (this->chunks[2] | this->chunks[3]);
+    }
+
+    simdjson_inline uint64_t eq(const T m) const {
+      const simd8<T> mask = simd8<T>::splat(m);
+      return  simd8x64<bool>(
+        this->chunks[0] == mask,
+        this->chunks[1] == mask,
+        this->chunks[2] == mask,
+        this->chunks[3] == mask
+      ).to_bitmask();
+    }
+
+    simdjson_inline uint64_t eq(const simd8x64<uint8_t> &other) const {
+      return  simd8x64<bool>(
+        this->chunks[0] == other.chunks[0],
+        this->chunks[1] == other.chunks[1],
+        this->chunks[2] == other.chunks[2],
+        this->chunks[3] == other.chunks[3]
+      ).to_bitmask();
+    }
+
+    simdjson_inline uint64_t lteq(const T m) const {
+      const simd8<T> mask = simd8<T>::splat(m);
+      return  simd8x64<bool>(
+        this->chunks[0] <= mask,
+        this->chunks[1] <= mask,
+        this->chunks[2] <= mask,
+        this->chunks[3] <= mask
+      ).to_bitmask();
+    }
+  }; // struct simd8x64<T>
+
+} // namespace simd
+} // unnamed namespace
+} // namespace lsx
+} // namespace simdjson
+
+#endif // SIMDJSON_LSX_SIMD_H
diff --git a/contrib/libs/simdjson/include/simdjson/lsx/stringparsing_defs.h b/contrib/libs/simdjson/include/simdjson/lsx/stringparsing_defs.h
new file mode 100644
index 00000000000..af493dc55f6
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/lsx/stringparsing_defs.h
@@ -0,0 +1,53 @@
+#ifndef SIMDJSON_LSX_STRINGPARSING_DEFS_H
+#define SIMDJSON_LSX_STRINGPARSING_DEFS_H
+
+#ifndef SIMDJSON_CONDITIONAL_INCLUDE
+#include "simdjson/lsx/base.h"
+#include "simdjson/lsx/simd.h"
+#include "simdjson/lsx/bitmanipulation.h"
+#endif // SIMDJSON_CONDITIONAL_INCLUDE
+
+namespace simdjson {
+namespace lsx {
+namespace {
+
+using namespace simd;
+
+// Holds backslashes and quotes locations.
+struct backslash_and_quote {
+public:
+  static constexpr uint32_t BYTES_PROCESSED = 32;
+  simdjson_inline static backslash_and_quote copy_and_find(const uint8_t *src, uint8_t *dst);
+
+  simdjson_inline bool has_quote_first() { return ((bs_bits - 1) & quote_bits) != 0; }
+  simdjson_inline bool has_backslash() { return bs_bits != 0; }
+  simdjson_inline int quote_index() { return trailing_zeroes(quote_bits); }
+  simdjson_inline int backslash_index() { return trailing_zeroes(bs_bits); }
+
+  uint32_t bs_bits;
+  uint32_t quote_bits;
+}; // struct backslash_and_quote
+
+simdjson_inline backslash_and_quote backslash_and_quote::copy_and_find(const uint8_t *src, uint8_t *dst) {
+  // this can read up to 31 bytes beyond the buffer size, but we require
+  // SIMDJSON_PADDING of padding
+  static_assert(SIMDJSON_PADDING >= (BYTES_PROCESSED - 1), "backslash and quote finder must process fewer than SIMDJSON_PADDING bytes");
+  simd8<uint8_t> v0(src);
+  simd8<uint8_t> v1(src + sizeof(v0));
+  v0.store(dst);
+  v1.store(dst + sizeof(v0));
+
+  // Getting a 64-bit bitmask is much cheaper than multiple 16-bit bitmasks on LSX; therefore, we
+  // smash them together into a 64-byte mask and get the bitmask from there.
+  uint64_t bs_and_quote = simd8x64<bool>(v0 == '\\', v1 == '\\', v0 == '"', v1 == '"').to_bitmask();
+  return {
+    uint32_t(bs_and_quote),      // bs_bits
+    uint32_t(bs_and_quote >> 32) // quote_bits
+  };
+}
+
+} // unnamed namespace
+} // namespace lsx
+} // namespace simdjson
+
+#endif // SIMDJSON_LSX_STRINGPARSING_DEFS_H
diff --git a/contrib/libs/simdjson/include/simdjson/portability.h b/contrib/libs/simdjson/include/simdjson/portability.h
new file mode 100644
index 00000000000..ff65aa9deb3
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/portability.h
@@ -0,0 +1,204 @@
+#ifndef SIMDJSON_PORTABILITY_H
+#define SIMDJSON_PORTABILITY_H
+
+#include <cstddef>
+#include <cstdint>
+#include <cstdlib>
+#include <cfloat>
+#include <cassert>
+#ifndef _WIN32
+// strcasecmp, strncasecmp
+#include <strings.h>
+#endif
+
+// We are using size_t without namespace std:: throughout the project
+using std::size_t;
+
+#ifdef _MSC_VER
+#define SIMDJSON_VISUAL_STUDIO 1
+/**
+ * We want to differentiate carefully between
+ * clang under visual studio and regular visual
+ * studio.
+ *
+ * Under clang for Windows, we enable:
+ *  * target pragmas so that part and only part of the
+ *     code gets compiled for advanced instructions.
+ *
+ */
+#ifdef __clang__
+// clang under visual studio
+#define SIMDJSON_CLANG_VISUAL_STUDIO 1
+#else
+// just regular visual studio (best guess)
+#define SIMDJSON_REGULAR_VISUAL_STUDIO 1
+#endif // __clang__
+#endif // _MSC_VER
+
+#if (defined(__x86_64__) || defined(_M_AMD64)) && !defined(_M_ARM64EC)
+#define SIMDJSON_IS_X86_64 1
+#elif defined(__aarch64__) || defined(_M_ARM64) || defined(_M_ARM64EC)
+#define SIMDJSON_IS_ARM64 1
+#elif defined(__riscv) && __riscv_xlen == 64
+#define SIMDJSON_IS_RISCV64 1
+#elif defined(__loongarch_lp64)
+#define SIMDJSON_IS_LOONGARCH64 1
+#elif defined(__PPC64__) || defined(_M_PPC64)
+#if defined(__ALTIVEC__)
+#define SIMDJSON_IS_PPC64_VMX 1
+#endif // defined(__ALTIVEC__)
+#else
+#define SIMDJSON_IS_32BITS 1
+
+#if defined(_M_IX86) || defined(__i386__)
+#define SIMDJSON_IS_X86_32BITS 1
+#elif defined(__arm__) || defined(_M_ARM)
+#define SIMDJSON_IS_ARM_32BITS 1
+#elif defined(__PPC__) || defined(_M_PPC)
+#define SIMDJSON_IS_PPC_32BITS 1
+#endif
+
+#endif // defined(__x86_64__) || defined(_M_AMD64)
+#ifndef SIMDJSON_IS_32BITS
+#define SIMDJSON_IS_32BITS 0
+#endif
+
+#if SIMDJSON_IS_32BITS
+#ifndef SIMDJSON_NO_PORTABILITY_WARNING
+// In the future, we should allow programmers
+// to get warning.
+#endif // SIMDJSON_NO_PORTABILITY_WARNING
+#endif // SIMDJSON_IS_32BITS
+
+#define SIMDJSON_CAT_IMPLEMENTATION_(a,...) a ## __VA_ARGS__
+#define SIMDJSON_CAT(a,...) SIMDJSON_CAT_IMPLEMENTATION_(a, __VA_ARGS__)
+
+#define SIMDJSON_STRINGIFY_IMPLEMENTATION_(a,...) #a SIMDJSON_STRINGIFY(__VA_ARGS__)
+#define SIMDJSON_STRINGIFY(a,...) SIMDJSON_CAT_IMPLEMENTATION_(a, __VA_ARGS__)
+
+// this is almost standard?
+#undef SIMDJSON_STRINGIFY_IMPLEMENTATION_
+#undef SIMDJSON_STRINGIFY
+#define SIMDJSON_STRINGIFY_IMPLEMENTATION_(a) #a
+#define SIMDJSON_STRINGIFY(a) SIMDJSON_STRINGIFY_IMPLEMENTATION_(a)
+
+// Our fast kernels require 64-bit systems.
+//
+// On 32-bit x86, we lack 64-bit popcnt, lzcnt, blsr instructions.
+// Furthermore, the number of SIMD registers is reduced.
+//
+// On 32-bit ARM, we would have smaller registers.
+//
+// The simdjson users should still have the fallback kernel. It is
+// slower, but it should run everywhere.
+
+//
+// Enable valid runtime implementations, and select SIMDJSON_BUILTIN_IMPLEMENTATION
+//
+
+// We are going to use runtime dispatch.
+#if SIMDJSON_IS_X86_64
+#ifdef __clang__
+// clang does not have GCC push pop
+// warning: clang attribute push can't be used within a namespace in clang up
+// til 8.0 so SIMDJSON_TARGET_REGION and SIMDJSON_UNTARGET_REGION must be *outside* of a
+// namespace.
+#define SIMDJSON_TARGET_REGION(T)                                                       \
+  _Pragma(SIMDJSON_STRINGIFY(                                                           \
+      clang attribute push(__attribute__((target(T))), apply_to = function)))
+#define SIMDJSON_UNTARGET_REGION _Pragma("clang attribute pop")
+#elif defined(__GNUC__)
+// GCC is easier
+#define SIMDJSON_TARGET_REGION(T)                                                       \
+  _Pragma("GCC push_options") _Pragma(SIMDJSON_STRINGIFY(GCC target(T)))
+#define SIMDJSON_UNTARGET_REGION _Pragma("GCC pop_options")
+#endif // clang then gcc
+
+#endif // x86
+
+// Default target region macros don't do anything.
+#ifndef SIMDJSON_TARGET_REGION
+#define SIMDJSON_TARGET_REGION(T)
+#define SIMDJSON_UNTARGET_REGION
+#endif
+
+// Is threading enabled?
+#if defined(_REENTRANT) || defined(_MT)
+#ifndef SIMDJSON_THREADS_ENABLED
+#define SIMDJSON_THREADS_ENABLED
+#endif
+#endif
+
+// workaround for large stack sizes under -O0.
+// https://github.com/simdjson/simdjson/issues/691
+#ifdef __APPLE__
+#ifndef __OPTIMIZE__
+// Apple systems have small stack sizes in secondary threads.
+// Lack of compiler optimization may generate high stack usage.
+// Users may want to disable threads for safety, but only when
+// in debug mode which we detect by the fact that the __OPTIMIZE__
+// macro is not defined.
+#undef SIMDJSON_THREADS_ENABLED
+#endif
+#endif
+
+
+#if defined(__clang__)
+#define SIMDJSON_NO_SANITIZE_UNDEFINED __attribute__((no_sanitize("undefined")))
+#elif defined(__GNUC__)
+#define SIMDJSON_NO_SANITIZE_UNDEFINED __attribute__((no_sanitize_undefined))
+#else
+#define SIMDJSON_NO_SANITIZE_UNDEFINED
+#endif
+
+#if defined(__clang__) || defined(__GNUC__)
+#define simdjson_pure [[gnu::pure]]
+#else
+#define simdjson_pure
+#endif
+
+#if defined(__clang__) || defined(__GNUC__)
+#if defined(__has_feature)
+#  if __has_feature(memory_sanitizer)
+#define SIMDJSON_NO_SANITIZE_MEMORY __attribute__((no_sanitize("memory")))
+#  endif // if __has_feature(memory_sanitizer)
+#endif // defined(__has_feature)
+#endif
+// make sure it is defined as 'nothing' if it is unapplicable.
+#ifndef SIMDJSON_NO_SANITIZE_MEMORY
+#define SIMDJSON_NO_SANITIZE_MEMORY
+#endif
+
+#if SIMDJSON_VISUAL_STUDIO
+// This is one case where we do not distinguish between
+// regular visual studio and clang under visual studio.
+// clang under Windows has _stricmp (like visual studio) but not strcasecmp (as clang normally has)
+#define simdjson_strcasecmp _stricmp
+#define simdjson_strncasecmp _strnicmp
+#else
+// The strcasecmp, strncasecmp, and strcasestr functions do not work with multibyte strings (e.g. UTF-8).
+// So they are only useful for ASCII in our context.
+// https://www.gnu.org/software/libunistring/manual/libunistring.html#char-_002a-strings
+#define simdjson_strcasecmp strcasecmp
+#define simdjson_strncasecmp strncasecmp
+#endif
+
+#if defined(NDEBUG) || defined(__OPTIMIZE__) || (defined(_MSC_VER) && !defined(_DEBUG))
+// If NDEBUG is set, or __OPTIMIZE__ is set, or we are under MSVC in release mode,
+// then do away with asserts and use __assume.
+#if SIMDJSON_VISUAL_STUDIO
+#define SIMDJSON_UNREACHABLE() __assume(0)
+#define SIMDJSON_ASSUME(COND) __assume(COND)
+#else
+#define SIMDJSON_UNREACHABLE() __builtin_unreachable();
+#define SIMDJSON_ASSUME(COND) do { if (!(COND)) __builtin_unreachable(); } while (0)
+#endif
+
+#else // defined(NDEBUG) || defined(__OPTIMIZE__) || (defined(_MSC_VER) && !defined(_DEBUG))
+// This should only ever be enabled in debug mode.
+#define SIMDJSON_UNREACHABLE() assert(0);
+#define SIMDJSON_ASSUME(COND) assert(COND)
+
+#endif
+
+#endif // SIMDJSON_PORTABILITY_H
diff --git a/contrib/libs/simdjson/include/simdjson/westmere.h b/contrib/libs/simdjson/include/simdjson/westmere.h
new file mode 100644
index 00000000000..f05ba1145c0
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/westmere.h
@@ -0,0 +1,8 @@
+#ifndef SIMDJSON_WESTMERE_H
+#define SIMDJSON_WESTMERE_H
+
+#include "simdjson/westmere/begin.h"
+#include "simdjson/generic/amalgamated.h"
+#include "simdjson/westmere/end.h"
+
+#endif // SIMDJSON_WESTMERE_H
\ No newline at end of file
diff --git a/contrib/libs/simdjson/include/simdjson/westmere/base.h b/contrib/libs/simdjson/include/simdjson/westmere/base.h
new file mode 100644
index 00000000000..82ad333a8dc
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/westmere/base.h
@@ -0,0 +1,29 @@
+#ifndef SIMDJSON_WESTMERE_BASE_H
+#define SIMDJSON_WESTMERE_BASE_H
+
+#ifndef SIMDJSON_CONDITIONAL_INCLUDE
+#include "simdjson/base.h"
+#endif // SIMDJSON_CONDITIONAL_INCLUDE
+
+// The constructor may be executed on any host, so we take care not to use SIMDJSON_TARGET_WESTMERE
+namespace simdjson {
+/**
+ * Implementation for Westmere (Intel SSE4.2).
+ */
+namespace westmere {
+
+class implementation;
+
+namespace {
+namespace simd {
+
+template <typename T> struct simd8;
+template <typename T> struct simd8x64;
+
+} // namespace simd
+} // unnamed namespace
+
+} // namespace westmere
+} // namespace simdjson
+
+#endif // SIMDJSON_WESTMERE_BASE_H
diff --git a/contrib/libs/simdjson/include/simdjson/westmere/begin.h b/contrib/libs/simdjson/include/simdjson/westmere/begin.h
new file mode 100644
index 00000000000..d807e6d9995
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/westmere/begin.h
@@ -0,0 +1,13 @@
+#define SIMDJSON_IMPLEMENTATION westmere
+#include "simdjson/westmere/base.h"
+#include "simdjson/westmere/intrinsics.h"
+
+#if !SIMDJSON_CAN_ALWAYS_RUN_WESTMERE
+SIMDJSON_TARGET_REGION("sse4.2,pclmul,popcnt")
+#endif
+
+#include "simdjson/westmere/bitmanipulation.h"
+#include "simdjson/westmere/bitmask.h"
+#include "simdjson/westmere/numberparsing_defs.h"
+#include "simdjson/westmere/simd.h"
+#include "simdjson/westmere/stringparsing_defs.h"
diff --git a/contrib/libs/simdjson/include/simdjson/westmere/bitmanipulation.h b/contrib/libs/simdjson/include/simdjson/westmere/bitmanipulation.h
new file mode 100644
index 00000000000..7cd29a406a7
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/westmere/bitmanipulation.h
@@ -0,0 +1,79 @@
+#ifndef SIMDJSON_WESTMERE_BITMANIPULATION_H
+#define SIMDJSON_WESTMERE_BITMANIPULATION_H
+
+#ifndef SIMDJSON_CONDITIONAL_INCLUDE
+#include "simdjson/westmere/base.h"
+#include "simdjson/westmere/intrinsics.h"
+#endif // SIMDJSON_CONDITIONAL_INCLUDE
+
+namespace simdjson {
+namespace westmere {
+namespace {
+
+// We sometimes call trailing_zero on inputs that are zero,
+// but the algorithms do not end up using the returned value.
+// Sadly, sanitizers are not smart enough to figure it out.
+SIMDJSON_NO_SANITIZE_UNDEFINED
+// This function can be used safely even if not all bytes have been
+// initialized.
+// See issue https://github.com/simdjson/simdjson/issues/1965
+SIMDJSON_NO_SANITIZE_MEMORY
+simdjson_inline int trailing_zeroes(uint64_t input_num) {
+#if SIMDJSON_REGULAR_VISUAL_STUDIO
+  unsigned long ret;
+  // Search the mask data from least significant bit (LSB)
+  // to the most significant bit (MSB) for a set bit (1).
+  _BitScanForward64(&ret, input_num);
+  return (int)ret;
+#else // SIMDJSON_REGULAR_VISUAL_STUDIO
+  return __builtin_ctzll(input_num);
+#endif // SIMDJSON_REGULAR_VISUAL_STUDIO
+}
+
+/* result might be undefined when input_num is zero */
+simdjson_inline uint64_t clear_lowest_bit(uint64_t input_num) {
+  return input_num & (input_num-1);
+}
+
+/* result might be undefined when input_num is zero */
+simdjson_inline int leading_zeroes(uint64_t input_num) {
+#if SIMDJSON_REGULAR_VISUAL_STUDIO
+  unsigned long leading_zero = 0;
+  // Search the mask data from most significant bit (MSB)
+  // to least significant bit (LSB) for a set bit (1).
+  if (_BitScanReverse64(&leading_zero, input_num))
+    return (int)(63 - leading_zero);
+  else
+    return 64;
+#else
+  return __builtin_clzll(input_num);
+#endif// SIMDJSON_REGULAR_VISUAL_STUDIO
+}
+
+#if SIMDJSON_REGULAR_VISUAL_STUDIO
+simdjson_inline unsigned __int64 count_ones(uint64_t input_num) {
+  // note: we do not support legacy 32-bit Windows in this kernel
+  return __popcnt64(input_num);// Visual Studio wants two underscores
+}
+#else
+simdjson_inline long long int count_ones(uint64_t input_num) {
+  return _popcnt64(input_num);
+}
+#endif
+
+simdjson_inline bool add_overflow(uint64_t value1, uint64_t value2,
+                                uint64_t *result) {
+#if SIMDJSON_REGULAR_VISUAL_STUDIO
+  return _addcarry_u64(0, value1, value2,
+                       reinterpret_cast<unsigned __int64 *>(result));
+#else
+  return __builtin_uaddll_overflow(value1, value2,
+                                   reinterpret_cast<unsigned long long *>(result));
+#endif
+}
+
+} // unnamed namespace
+} // namespace westmere
+} // namespace simdjson
+
+#endif // SIMDJSON_WESTMERE_BITMANIPULATION_H
diff --git a/contrib/libs/simdjson/include/simdjson/westmere/bitmask.h b/contrib/libs/simdjson/include/simdjson/westmere/bitmask.h
new file mode 100644
index 00000000000..cd79b724119
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/westmere/bitmask.h
@@ -0,0 +1,30 @@
+#ifndef SIMDJSON_WESTMERE_BITMASK_H
+#define SIMDJSON_WESTMERE_BITMASK_H
+
+#ifndef SIMDJSON_CONDITIONAL_INCLUDE
+#include "simdjson/westmere/base.h"
+#include "simdjson/westmere/intrinsics.h"
+#endif // SIMDJSON_CONDITIONAL_INCLUDE
+
+namespace simdjson {
+namespace westmere {
+namespace {
+
+//
+// Perform a "cumulative bitwise xor," flipping bits each time a 1 is encountered.
+//
+// For example, prefix_xor(00100100) == 00011100
+//
+simdjson_inline uint64_t prefix_xor(const uint64_t bitmask) {
+  // There should be no such thing with a processing supporting avx2
+  // but not clmul.
+  __m128i all_ones = _mm_set1_epi8('\xFF');
+  __m128i result = _mm_clmulepi64_si128(_mm_set_epi64x(0ULL, bitmask), all_ones, 0);
+  return _mm_cvtsi128_si64(result);
+}
+
+} // unnamed namespace
+} // namespace westmere
+} // namespace simdjson
+
+#endif // SIMDJSON_WESTMERE_BITMASK_H
diff --git a/contrib/libs/simdjson/include/simdjson/westmere/end.h b/contrib/libs/simdjson/include/simdjson/westmere/end.h
new file mode 100644
index 00000000000..bd4a9404903
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/westmere/end.h
@@ -0,0 +1,9 @@
+#ifndef SIMDJSON_CONDITIONAL_INCLUDE
+#include "simdjson/westmere/base.h"
+#endif // SIMDJSON_CONDITIONAL_INCLUDE
+
+#if !SIMDJSON_CAN_ALWAYS_RUN_WESTMERE
+SIMDJSON_UNTARGET_REGION
+#endif
+
+#undef SIMDJSON_IMPLEMENTATION
diff --git a/contrib/libs/simdjson/include/simdjson/westmere/implementation.h b/contrib/libs/simdjson/include/simdjson/westmere/implementation.h
new file mode 100644
index 00000000000..37392be2ae9
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/westmere/implementation.h
@@ -0,0 +1,32 @@
+#ifndef SIMDJSON_WESTMERE_IMPLEMENTATION_H
+#define SIMDJSON_WESTMERE_IMPLEMENTATION_H
+
+#ifndef SIMDJSON_CONDITIONAL_INCLUDE
+#include "simdjson/westmere/base.h"
+#include "simdjson/implementation.h"
+#include "simdjson/internal/instruction_set.h"
+#endif // SIMDJSON_CONDITIONAL_INCLUDE
+
+// The constructor may be executed on any host, so we take care not to use SIMDJSON_TARGET_WESTMERE
+namespace simdjson {
+namespace westmere {
+
+/**
+ * @private
+ */
+class implementation final : public simdjson::implementation {
+public:
+  simdjson_inline implementation() : simdjson::implementation("westmere", "Intel/AMD SSE4.2", internal::instruction_set::SSE42 | internal::instruction_set::PCLMULQDQ) {}
+  simdjson_warn_unused error_code create_dom_parser_implementation(
+    size_t capacity,
+    size_t max_length,
+    std::unique_ptr<internal::dom_parser_implementation>& dst
+  ) const noexcept final;
+  simdjson_warn_unused error_code minify(const uint8_t *buf, size_t len, uint8_t *dst, size_t &dst_len) const noexcept final;
+  simdjson_warn_unused bool validate_utf8(const char *buf, size_t len) const noexcept final;
+};
+
+} // namespace westmere
+} // namespace simdjson
+
+#endif // SIMDJSON_WESTMERE_IMPLEMENTATION_H
diff --git a/contrib/libs/simdjson/include/simdjson/westmere/intrinsics.h b/contrib/libs/simdjson/include/simdjson/westmere/intrinsics.h
new file mode 100644
index 00000000000..63a351c8090
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/westmere/intrinsics.h
@@ -0,0 +1,31 @@
+#ifndef SIMDJSON_WESTMERE_INTRINSICS_H
+#define SIMDJSON_WESTMERE_INTRINSICS_H
+
+#ifndef SIMDJSON_CONDITIONAL_INCLUDE
+#include "simdjson/westmere/base.h"
+#endif // SIMDJSON_CONDITIONAL_INCLUDE
+
+#if SIMDJSON_VISUAL_STUDIO
+// under clang within visual studio, this will include <x86intrin.h>
+#include <intrin.h> // visual studio or clang
+#else
+#include <x86intrin.h> // elsewhere
+#endif // SIMDJSON_VISUAL_STUDIO
+
+
+#if SIMDJSON_CLANG_VISUAL_STUDIO
+/**
+ * You are not supposed, normally, to include these
+ * headers directly. Instead you should either include intrin.h
+ * or x86intrin.h. However, when compiling with clang
+ * under Windows (i.e., when _MSC_VER is set), these headers
+ * only get included *if* the corresponding features are detected
+ * from macros:
+ */
+#include <smmintrin.h>  // for _mm_alignr_epi8
+#include <wmmintrin.h>  // for  _mm_clmulepi64_si128
+#endif
+
+static_assert(sizeof(__m128i) <= simdjson::SIMDJSON_PADDING, "insufficient padding for westmere");
+
+#endif // SIMDJSON_WESTMERE_INTRINSICS_H
diff --git a/contrib/libs/simdjson/include/simdjson/westmere/numberparsing_defs.h b/contrib/libs/simdjson/include/simdjson/westmere/numberparsing_defs.h
new file mode 100644
index 00000000000..05cfccfd10d
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/westmere/numberparsing_defs.h
@@ -0,0 +1,59 @@
+#ifndef SIMDJSON_WESTMERE_NUMBERPARSING_DEFS_H
+#define SIMDJSON_WESTMERE_NUMBERPARSING_DEFS_H
+
+#include "simdjson/westmere/base.h"
+#include "simdjson/westmere/intrinsics.h"
+
+#ifndef SIMDJSON_CONDITIONAL_INCLUDE
+#include "simdjson/internal/numberparsing_tables.h"
+#endif // SIMDJSON_CONDITIONAL_INCLUDE
+
+namespace simdjson {
+namespace westmere {
+namespace numberparsing {
+
+/** @private */
+static simdjson_inline uint32_t parse_eight_digits_unrolled(const uint8_t *chars) {
+  // this actually computes *16* values so we are being wasteful.
+  const __m128i ascii0 = _mm_set1_epi8('0');
+  const __m128i mul_1_10 =
+      _mm_setr_epi8(10, 1, 10, 1, 10, 1, 10, 1, 10, 1, 10, 1, 10, 1, 10, 1);
+  const __m128i mul_1_100 = _mm_setr_epi16(100, 1, 100, 1, 100, 1, 100, 1);
+  const __m128i mul_1_10000 =
+      _mm_setr_epi16(10000, 1, 10000, 1, 10000, 1, 10000, 1);
+  const __m128i input = _mm_sub_epi8(
+      _mm_loadu_si128(reinterpret_cast<const __m128i *>(chars)), ascii0);
+  const __m128i t1 = _mm_maddubs_epi16(input, mul_1_10);
+  const __m128i t2 = _mm_madd_epi16(t1, mul_1_100);
+  const __m128i t3 = _mm_packus_epi32(t2, t2);
+  const __m128i t4 = _mm_madd_epi16(t3, mul_1_10000);
+  return _mm_cvtsi128_si32(
+      t4); // only captures the sum of the first 8 digits, drop the rest
+}
+
+/** @private */
+simdjson_inline internal::value128 full_multiplication(uint64_t value1, uint64_t value2) {
+  internal::value128 answer;
+#if SIMDJSON_REGULAR_VISUAL_STUDIO || SIMDJSON_IS_32BITS
+#if SIMDJSON_IS_ARM64
+  // ARM64 has native support for 64-bit multiplications, no need to emultate
+  answer.high = __umulh(value1, value2);
+  answer.low = value1 * value2;
+#else
+  answer.low = _umul128(value1, value2, &answer.high); // _umul128 not available on ARM64
+#endif // SIMDJSON_IS_ARM64
+#else // SIMDJSON_REGULAR_VISUAL_STUDIO || SIMDJSON_IS_32BITS
+  __uint128_t r = (static_cast<__uint128_t>(value1)) * value2;
+  answer.low = uint64_t(r);
+  answer.high = uint64_t(r >> 64);
+#endif
+  return answer;
+}
+
+} // namespace numberparsing
+} // namespace westmere
+} // namespace simdjson
+
+#define SIMDJSON_SWAR_NUMBER_PARSING 1
+
+#endif //  SIMDJSON_WESTMERE_NUMBERPARSING_DEFS_H
diff --git a/contrib/libs/simdjson/include/simdjson/westmere/simd.h b/contrib/libs/simdjson/include/simdjson/westmere/simd.h
new file mode 100644
index 00000000000..28329d9642e
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/westmere/simd.h
@@ -0,0 +1,338 @@
+#ifndef SIMDJSON_WESTMERE_SIMD_H
+#define SIMDJSON_WESTMERE_SIMD_H
+
+#ifndef SIMDJSON_CONDITIONAL_INCLUDE
+#include "simdjson/westmere/base.h"
+#include "simdjson/westmere/bitmanipulation.h"
+#include "simdjson/internal/simdprune_tables.h"
+#endif // SIMDJSON_CONDITIONAL_INCLUDE
+
+namespace simdjson {
+namespace westmere {
+namespace {
+namespace simd {
+
+  template<typename Child>
+  struct base {
+    __m128i value;
+
+    // Zero constructor
+    simdjson_inline base() : value{__m128i()} {}
+
+    // Conversion from SIMD register
+    simdjson_inline base(const __m128i _value) : value(_value) {}
+
+    // Conversion to SIMD register
+    simdjson_inline operator const __m128i&() const { return this->value; }
+    simdjson_inline operator __m128i&() { return this->value; }
+
+    // Bit operations
+    simdjson_inline Child operator|(const Child other) const { return _mm_or_si128(*this, other); }
+    simdjson_inline Child operator&(const Child other) const { return _mm_and_si128(*this, other); }
+    simdjson_inline Child operator^(const Child other) const { return _mm_xor_si128(*this, other); }
+    simdjson_inline Child bit_andnot(const Child other) const { return _mm_andnot_si128(other, *this); }
+    simdjson_inline Child& operator|=(const Child other) { auto this_cast = static_cast<Child*>(this); *this_cast = *this_cast | other; return *this_cast; }
+    simdjson_inline Child& operator&=(const Child other) { auto this_cast = static_cast<Child*>(this); *this_cast = *this_cast & other; return *this_cast; }
+    simdjson_inline Child& operator^=(const Child other) { auto this_cast = static_cast<Child*>(this); *this_cast = *this_cast ^ other; return *this_cast; }
+  };
+
+  template<typename T, typename Mask=simd8<bool>>
+  struct base8: base<simd8<T>> {
+    typedef uint16_t bitmask_t;
+    typedef uint32_t bitmask2_t;
+
+    simdjson_inline base8() : base<simd8<T>>() {}
+    simdjson_inline base8(const __m128i _value) : base<simd8<T>>(_value) {}
+
+    friend simdjson_inline Mask operator==(const simd8<T> lhs, const simd8<T> rhs) { return _mm_cmpeq_epi8(lhs, rhs); }
+
+    static const int SIZE = sizeof(base<simd8<T>>::value);
+
+    template<int N=1>
+    simdjson_inline simd8<T> prev(const simd8<T> prev_chunk) const {
+      return _mm_alignr_epi8(*this, prev_chunk, 16 - N);
+    }
+  };
+
+  // SIMD byte mask type (returned by things like eq and gt)
+  template<>
+  struct simd8<bool>: base8<bool> {
+    static simdjson_inline simd8<bool> splat(bool _value) { return _mm_set1_epi8(uint8_t(-(!!_value))); }
+
+    simdjson_inline simd8() : base8() {}
+    simdjson_inline simd8(const __m128i _value) : base8<bool>(_value) {}
+    // Splat constructor
+    simdjson_inline simd8(bool _value) : base8<bool>(splat(_value)) {}
+
+    simdjson_inline int to_bitmask() const { return _mm_movemask_epi8(*this); }
+    simdjson_inline bool any() const { return !_mm_testz_si128(*this, *this); }
+    simdjson_inline simd8<bool> operator~() const { return *this ^ true; }
+  };
+
+  template<typename T>
+  struct base8_numeric: base8<T> {
+    static simdjson_inline simd8<T> splat(T _value) { return _mm_set1_epi8(_value); }
+    static simdjson_inline simd8<T> zero() { return _mm_setzero_si128(); }
+    static simdjson_inline simd8<T> load(const T values[16]) {
+      return _mm_loadu_si128(reinterpret_cast<const __m128i *>(values));
+    }
+    // Repeat 16 values as many times as necessary (usually for lookup tables)
+    static simdjson_inline simd8<T> repeat_16(
+      T v0,  T v1,  T v2,  T v3,  T v4,  T v5,  T v6,  T v7,
+      T v8,  T v9,  T v10, T v11, T v12, T v13, T v14, T v15
+    ) {
+      return simd8<T>(
+        v0, v1, v2, v3, v4, v5, v6, v7,
+        v8, v9, v10,v11,v12,v13,v14,v15
+      );
+    }
+
+    simdjson_inline base8_numeric() : base8<T>() {}
+    simdjson_inline base8_numeric(const __m128i _value) : base8<T>(_value) {}
+
+    // Store to array
+    simdjson_inline void store(T dst[16]) const { return _mm_storeu_si128(reinterpret_cast<__m128i *>(dst), *this); }
+
+    // Override to distinguish from bool version
+    simdjson_inline simd8<T> operator~() const { return *this ^ 0xFFu; }
+
+    // Addition/subtraction are the same for signed and unsigned
+    simdjson_inline simd8<T> operator+(const simd8<T> other) const { return _mm_add_epi8(*this, other); }
+    simdjson_inline simd8<T> operator-(const simd8<T> other) const { return _mm_sub_epi8(*this, other); }
+    simdjson_inline simd8<T>& operator+=(const simd8<T> other) { *this = *this + other; return *static_cast<simd8<T>*>(this); }
+    simdjson_inline simd8<T>& operator-=(const simd8<T> other) { *this = *this - other; return *static_cast<simd8<T>*>(this); }
+
+    // Perform a lookup assuming the value is between 0 and 16 (undefined behavior for out of range values)
+    template<typename L>
+    simdjson_inline simd8<L> lookup_16(simd8<L> lookup_table) const {
+      return _mm_shuffle_epi8(lookup_table, *this);
+    }
+
+    // Copies to 'output" all bytes corresponding to a 0 in the mask (interpreted as a bitset).
+    // Passing a 0 value for mask would be equivalent to writing out every byte to output.
+    // Only the first 16 - count_ones(mask) bytes of the result are significant but 16 bytes
+    // get written.
+    // Design consideration: it seems like a function with the
+    // signature simd8<L> compress(uint32_t mask) would be
+    // sensible, but the AVX ISA makes this kind of approach difficult.
+    template<typename L>
+    simdjson_inline void compress(uint16_t mask, L * output) const {
+      using internal::thintable_epi8;
+      using internal::BitsSetTable256mul2;
+      using internal::pshufb_combine_table;
+      // this particular implementation was inspired by work done by @animetosho
+      // we do it in two steps, first 8 bytes and then second 8 bytes
+      uint8_t mask1 = uint8_t(mask); // least significant 8 bits
+      uint8_t mask2 = uint8_t(mask >> 8); // most significant 8 bits
+      // next line just loads the 64-bit values thintable_epi8[mask1] and
+      // thintable_epi8[mask2] into a 128-bit register, using only
+      // two instructions on most compilers.
+      __m128i shufmask =  _mm_set_epi64x(thintable_epi8[mask2], thintable_epi8[mask1]);
+      // we increment by 0x08 the second half of the mask
+      shufmask =
+      _mm_add_epi8(shufmask, _mm_set_epi32(0x08080808, 0x08080808, 0, 0));
+      // this is the version "nearly pruned"
+      __m128i pruned = _mm_shuffle_epi8(*this, shufmask);
+      // we still need to put the two halves together.
+      // we compute the popcount of the first half:
+      int pop1 = BitsSetTable256mul2[mask1];
+      // then load the corresponding mask, what it does is to write
+      // only the first pop1 bytes from the first 8 bytes, and then
+      // it fills in with the bytes from the second 8 bytes + some filling
+      // at the end.
+      __m128i compactmask =
+      _mm_loadu_si128(reinterpret_cast<const __m128i *>(pshufb_combine_table + pop1 * 8));
+      __m128i answer = _mm_shuffle_epi8(pruned, compactmask);
+      _mm_storeu_si128(reinterpret_cast<__m128i *>(output), answer);
+    }
+
+    template<typename L>
+    simdjson_inline simd8<L> lookup_16(
+        L replace0,  L replace1,  L replace2,  L replace3,
+        L replace4,  L replace5,  L replace6,  L replace7,
+        L replace8,  L replace9,  L replace10, L replace11,
+        L replace12, L replace13, L replace14, L replace15) const {
+      return lookup_16(simd8<L>::repeat_16(
+        replace0,  replace1,  replace2,  replace3,
+        replace4,  replace5,  replace6,  replace7,
+        replace8,  replace9,  replace10, replace11,
+        replace12, replace13, replace14, replace15
+      ));
+    }
+  };
+
+  // Signed bytes
+  template<>
+  struct simd8<int8_t> : base8_numeric<int8_t> {
+    simdjson_inline simd8() : base8_numeric<int8_t>() {}
+    simdjson_inline simd8(const __m128i _value) : base8_numeric<int8_t>(_value) {}
+    // Splat constructor
+    simdjson_inline simd8(int8_t _value) : simd8(splat(_value)) {}
+    // Array constructor
+    simdjson_inline simd8(const int8_t* values) : simd8(load(values)) {}
+    // Member-by-member initialization
+    simdjson_inline simd8(
+      int8_t v0,  int8_t v1,  int8_t v2,  int8_t v3,  int8_t v4,  int8_t v5,  int8_t v6,  int8_t v7,
+      int8_t v8,  int8_t v9,  int8_t v10, int8_t v11, int8_t v12, int8_t v13, int8_t v14, int8_t v15
+    ) : simd8(_mm_setr_epi8(
+      v0, v1, v2, v3, v4, v5, v6, v7,
+      v8, v9, v10,v11,v12,v13,v14,v15
+    )) {}
+    // Repeat 16 values as many times as necessary (usually for lookup tables)
+    simdjson_inline static simd8<int8_t> repeat_16(
+      int8_t v0,  int8_t v1,  int8_t v2,  int8_t v3,  int8_t v4,  int8_t v5,  int8_t v6,  int8_t v7,
+      int8_t v8,  int8_t v9,  int8_t v10, int8_t v11, int8_t v12, int8_t v13, int8_t v14, int8_t v15
+    ) {
+      return simd8<int8_t>(
+        v0, v1, v2, v3, v4, v5, v6, v7,
+        v8, v9, v10,v11,v12,v13,v14,v15
+      );
+    }
+
+    // Order-sensitive comparisons
+    simdjson_inline simd8<int8_t> max_val(const simd8<int8_t> other) const { return _mm_max_epi8(*this, other); }
+    simdjson_inline simd8<int8_t> min_val(const simd8<int8_t> other) const { return _mm_min_epi8(*this, other); }
+    simdjson_inline simd8<bool> operator>(const simd8<int8_t> other) const { return _mm_cmpgt_epi8(*this, other); }
+    simdjson_inline simd8<bool> operator<(const simd8<int8_t> other) const { return _mm_cmpgt_epi8(other, *this); }
+  };
+
+  // Unsigned bytes
+  template<>
+  struct simd8<uint8_t>: base8_numeric<uint8_t> {
+    simdjson_inline simd8() : base8_numeric<uint8_t>() {}
+    simdjson_inline simd8(const __m128i _value) : base8_numeric<uint8_t>(_value) {}
+    // Splat constructor
+    simdjson_inline simd8(uint8_t _value) : simd8(splat(_value)) {}
+    // Array constructor
+    simdjson_inline simd8(const uint8_t* values) : simd8(load(values)) {}
+    // Member-by-member initialization
+    simdjson_inline simd8(
+      uint8_t v0,  uint8_t v1,  uint8_t v2,  uint8_t v3,  uint8_t v4,  uint8_t v5,  uint8_t v6,  uint8_t v7,
+      uint8_t v8,  uint8_t v9,  uint8_t v10, uint8_t v11, uint8_t v12, uint8_t v13, uint8_t v14, uint8_t v15
+    ) : simd8(_mm_setr_epi8(
+      v0, v1, v2, v3, v4, v5, v6, v7,
+      v8, v9, v10,v11,v12,v13,v14,v15
+    )) {}
+    // Repeat 16 values as many times as necessary (usually for lookup tables)
+    simdjson_inline static simd8<uint8_t> repeat_16(
+      uint8_t v0,  uint8_t v1,  uint8_t v2,  uint8_t v3,  uint8_t v4,  uint8_t v5,  uint8_t v6,  uint8_t v7,
+      uint8_t v8,  uint8_t v9,  uint8_t v10, uint8_t v11, uint8_t v12, uint8_t v13, uint8_t v14, uint8_t v15
+    ) {
+      return simd8<uint8_t>(
+        v0, v1, v2, v3, v4, v5, v6, v7,
+        v8, v9, v10,v11,v12,v13,v14,v15
+      );
+    }
+
+    // Saturated math
+    simdjson_inline simd8<uint8_t> saturating_add(const simd8<uint8_t> other) const { return _mm_adds_epu8(*this, other); }
+    simdjson_inline simd8<uint8_t> saturating_sub(const simd8<uint8_t> other) const { return _mm_subs_epu8(*this, other); }
+
+    // Order-specific operations
+    simdjson_inline simd8<uint8_t> max_val(const simd8<uint8_t> other) const { return _mm_max_epu8(*this, other); }
+    simdjson_inline simd8<uint8_t> min_val(const simd8<uint8_t> other) const { return _mm_min_epu8(*this, other); }
+    // Same as >, but only guarantees true is nonzero (< guarantees true = -1)
+    simdjson_inline simd8<uint8_t> gt_bits(const simd8<uint8_t> other) const { return this->saturating_sub(other); }
+    // Same as <, but only guarantees true is nonzero (< guarantees true = -1)
+    simdjson_inline simd8<uint8_t> lt_bits(const simd8<uint8_t> other) const { return other.saturating_sub(*this); }
+    simdjson_inline simd8<bool> operator<=(const simd8<uint8_t> other) const { return other.max_val(*this) == other; }
+    simdjson_inline simd8<bool> operator>=(const simd8<uint8_t> other) const { return other.min_val(*this) == other; }
+    simdjson_inline simd8<bool> operator>(const simd8<uint8_t> other) const { return this->gt_bits(other).any_bits_set(); }
+    simdjson_inline simd8<bool> operator<(const simd8<uint8_t> other) const { return this->gt_bits(other).any_bits_set(); }
+
+    // Bit-specific operations
+    simdjson_inline simd8<bool> bits_not_set() const { return *this == uint8_t(0); }
+    simdjson_inline simd8<bool> bits_not_set(simd8<uint8_t> bits) const { return (*this & bits).bits_not_set(); }
+    simdjson_inline simd8<bool> any_bits_set() const { return ~this->bits_not_set(); }
+    simdjson_inline simd8<bool> any_bits_set(simd8<uint8_t> bits) const { return ~this->bits_not_set(bits); }
+    simdjson_inline bool is_ascii() const { return _mm_movemask_epi8(*this) == 0; }
+    simdjson_inline bool bits_not_set_anywhere() const { return _mm_testz_si128(*this, *this); }
+    simdjson_inline bool any_bits_set_anywhere() const { return !bits_not_set_anywhere(); }
+    simdjson_inline bool bits_not_set_anywhere(simd8<uint8_t> bits) const { return _mm_testz_si128(*this, bits); }
+    simdjson_inline bool any_bits_set_anywhere(simd8<uint8_t> bits) const { return !bits_not_set_anywhere(bits); }
+    template<int N>
+    simdjson_inline simd8<uint8_t> shr() const { return simd8<uint8_t>(_mm_srli_epi16(*this, N)) & uint8_t(0xFFu >> N); }
+    template<int N>
+    simdjson_inline simd8<uint8_t> shl() const { return simd8<uint8_t>(_mm_slli_epi16(*this, N)) & uint8_t(0xFFu << N); }
+    // Get one of the bits and make a bitmask out of it.
+    // e.g. value.get_bit<7>() gets the high bit
+    template<int N>
+    simdjson_inline int get_bit() const { return _mm_movemask_epi8(_mm_slli_epi16(*this, 7-N)); }
+  };
+
+  template<typename T>
+  struct simd8x64 {
+    static constexpr int NUM_CHUNKS = 64 / sizeof(simd8<T>);
+    static_assert(NUM_CHUNKS == 4, "Westmere kernel should use four registers per 64-byte block.");
+    const simd8<T> chunks[NUM_CHUNKS];
+
+    simd8x64(const simd8x64<T>& o) = delete; // no copy allowed
+    simd8x64<T>& operator=(const simd8<T>& other) = delete; // no assignment allowed
+    simd8x64() = delete; // no default constructor allowed
+
+    simdjson_inline simd8x64(const simd8<T> chunk0, const simd8<T> chunk1, const simd8<T> chunk2, const simd8<T> chunk3) : chunks{chunk0, chunk1, chunk2, chunk3} {}
+    simdjson_inline simd8x64(const T ptr[64]) : chunks{simd8<T>::load(ptr), simd8<T>::load(ptr+16), simd8<T>::load(ptr+32), simd8<T>::load(ptr+48)} {}
+
+    simdjson_inline void store(T ptr[64]) const {
+      this->chunks[0].store(ptr+sizeof(simd8<T>)*0);
+      this->chunks[1].store(ptr+sizeof(simd8<T>)*1);
+      this->chunks[2].store(ptr+sizeof(simd8<T>)*2);
+      this->chunks[3].store(ptr+sizeof(simd8<T>)*3);
+    }
+
+    simdjson_inline simd8<T> reduce_or() const {
+      return (this->chunks[0] | this->chunks[1]) | (this->chunks[2] | this->chunks[3]);
+    }
+
+    simdjson_inline uint64_t compress(uint64_t mask, T * output) const {
+      this->chunks[0].compress(uint16_t(mask), output);
+      this->chunks[1].compress(uint16_t(mask >> 16), output + 16 - count_ones(mask & 0xFFFF));
+      this->chunks[2].compress(uint16_t(mask >> 32), output + 32 - count_ones(mask & 0xFFFFFFFF));
+      this->chunks[3].compress(uint16_t(mask >> 48), output + 48 - count_ones(mask & 0xFFFFFFFFFFFF));
+      return 64 - count_ones(mask);
+    }
+
+    simdjson_inline uint64_t to_bitmask() const {
+      uint64_t r0 = uint32_t(this->chunks[0].to_bitmask() );
+      uint64_t r1 =          this->chunks[1].to_bitmask() ;
+      uint64_t r2 =          this->chunks[2].to_bitmask() ;
+      uint64_t r3 =          this->chunks[3].to_bitmask() ;
+      return r0 | (r1 << 16) | (r2 << 32) | (r3 << 48);
+    }
+
+    simdjson_inline uint64_t eq(const T m) const {
+      const simd8<T> mask = simd8<T>::splat(m);
+      return  simd8x64<bool>(
+        this->chunks[0] == mask,
+        this->chunks[1] == mask,
+        this->chunks[2] == mask,
+        this->chunks[3] == mask
+      ).to_bitmask();
+    }
+
+    simdjson_inline uint64_t eq(const simd8x64<uint8_t> &other) const {
+      return  simd8x64<bool>(
+        this->chunks[0] == other.chunks[0],
+        this->chunks[1] == other.chunks[1],
+        this->chunks[2] == other.chunks[2],
+        this->chunks[3] == other.chunks[3]
+      ).to_bitmask();
+    }
+
+    simdjson_inline uint64_t lteq(const T m) const {
+      const simd8<T> mask = simd8<T>::splat(m);
+      return  simd8x64<bool>(
+        this->chunks[0] <= mask,
+        this->chunks[1] <= mask,
+        this->chunks[2] <= mask,
+        this->chunks[3] <= mask
+      ).to_bitmask();
+    }
+  }; // struct simd8x64<T>
+
+} // namespace simd
+} // unnamed namespace
+} // namespace westmere
+} // namespace simdjson
+
+#endif // SIMDJSON_WESTMERE_SIMD_INPUT_H
diff --git a/contrib/libs/simdjson/include/simdjson/westmere/stringparsing_defs.h b/contrib/libs/simdjson/include/simdjson/westmere/stringparsing_defs.h
new file mode 100644
index 00000000000..439f19cbc73
--- /dev/null
+++ b/contrib/libs/simdjson/include/simdjson/westmere/stringparsing_defs.h
@@ -0,0 +1,47 @@
+#ifndef SIMDJSON_WESTMERE_STRINGPARSING_DEFS_H
+#define SIMDJSON_WESTMERE_STRINGPARSING_DEFS_H
+
+#include "simdjson/westmere/bitmanipulation.h"
+#include "simdjson/westmere/simd.h"
+
+namespace simdjson {
+namespace westmere {
+namespace {
+
+using namespace simd;
+
+// Holds backslashes and quotes locations.
+struct backslash_and_quote {
+public:
+  static constexpr uint32_t BYTES_PROCESSED = 32;
+  simdjson_inline static backslash_and_quote copy_and_find(const uint8_t *src, uint8_t *dst);
+
+  simdjson_inline bool has_quote_first() { return ((bs_bits - 1) & quote_bits) != 0; }
+  simdjson_inline bool has_backslash() { return bs_bits != 0; }
+  simdjson_inline int quote_index() { return trailing_zeroes(quote_bits); }
+  simdjson_inline int backslash_index() { return trailing_zeroes(bs_bits); }
+
+  uint32_t bs_bits;
+  uint32_t quote_bits;
+}; // struct backslash_and_quote
+
+simdjson_inline backslash_and_quote backslash_and_quote::copy_and_find(const uint8_t *src, uint8_t *dst) {
+  // this can read up to 31 bytes beyond the buffer size, but we require
+  // SIMDJSON_PADDING of padding
+  static_assert(SIMDJSON_PADDING >= (BYTES_PROCESSED - 1), "backslash and quote finder must process fewer than SIMDJSON_PADDING bytes");
+  simd8<uint8_t> v0(src);
+  simd8<uint8_t> v1(src + 16);
+  v0.store(dst);
+  v1.store(dst + 16);
+  uint64_t bs_and_quote = simd8x64<bool>(v0 == '\\', v1 == '\\', v0 == '"', v1 == '"').to_bitmask();
+  return {
+    uint32_t(bs_and_quote),      // bs_bits
+    uint32_t(bs_and_quote >> 32) // quote_bits
+  };
+}
+
+} // unnamed namespace
+} // namespace westmere
+} // namespace simdjson
+
+#endif // SIMDJSON_WESTMERE_STRINGPARSING_DEFS_H