json_view: skip the digits of a fraction 16 bytes at a time (SSE2)

The digits after the decimal point were skipped by skip_digits(), 16 compare-and-branch steps in a row: about 3.5 instructions per digit, and an exit branch that is mispredicted when the lengths of the fractions vary. On x86 with SSE2, the fraction is now skipped by skip_fraction_digits(): 16 bytes are compared at once and the end of the run is taken from the bit mask, which is about a third of the instructions for a fraction of 12 digits. The last bytes of the input are skipped one at a time as before. Where there is no SSE2 (AArch64, other targets, JSON_NO_SIMD) skip_fraction_digits() is skip_digits(): nothing changes there. Integer digits keep skip_digits(), whose exit is predictable. GCC unrolls the loop over the last bytes in every inlined copy of the new function, which made the parse loop 4 KiB larger and parse() of documents with hardly any fractions about 4 % slower. For GCC that loop is therefore out of line (skip_digits_tail()); Clang is faster with it inline. json_document::parse(), A/B in one binary, 6 code layouts, x86-64 Haswell, time after / before (GCC 16.2 / Clang 23): - numbers.json (10,001 floats): 0.715 / 0.680 - canada.json: 0.778 / 0.790 - mesh.json: 0.908 / 0.883 - marine_ik.json: 0.937 / 0.926 - mesh.pretty.json: 0.923 / 0.960 Files with few or no fractions and files with fractions of a fixed length (20,000 prices with two decimals, 60,000 coordinates with six): between 0.98 and 1.014 on both compilers, except citm_catalog.json with GCC (1.05: the instruction count is +0.3 %, and that row varies by +-7 % between builds of unrelated variants on this machine). New test: fractions of every length from 0 to 40 digits followed by every kind of byte that ends a number or lies next to the digits in ASCII, in several positions, and cut in every place near the end of the text, against json::parse() (the length of the number is part of what is compared), for all option combinations and with and without a terminating NUL. Checked on x86-64 with GCC 16.2, Clang 23 and Apple clang under Rosetta, and with a deliberate off-by-one at the 16th byte. Signed-off-by: Niels Lohmann <mail@nlohmann.me>
2026-10-31 13:00:39 +00:00 · 2026-10-30 22:17:00 +01:00
9 changed files with 235 additions and 220 deletions
@@ -268,10 +268,10 @@ class lexer : public lexer_base<BasicJsonType>
  public:
    using token_type = typename lexer_base<BasicJsonType>::token_type;

-    explicit lexer(InputAdapterType&& adapter, bool ignore_comments_ = false, bool discard_values_ = false) noexcept
+    explicit lexer(InputAdapterType&& adapter, bool ignore_comments_ = false, bool discard_number_values_ = false) noexcept
        : ia(std::move(adapter))
        , ignore_comments(ignore_comments_)
-        , discard_values(discard_values_)
+        , discard_number_values(discard_number_values_)
    {
        // the token string of an error before the first token is everything
        // read so far
@@ -388,8 +388,6 @@ class lexer : public lexer_base<BasicJsonType>
    checks if it is inside the range. If a violation was detected, set up an
    error message and return false. Otherwise, return true.

-    @param[in] keep    whether the bytes are added to token_buffer (false for
-                       accept(), see scan())
    @param[in] ranges  list of integers; interpreted as list of pairs of
                       inclusive lower and upper bound, respectively

@@ -398,18 +396,17 @@ class lexer : public lexer_base<BasicJsonType>

    @return true if and only if no range violation was detected
    */
-    template<bool Keep>
-    bool next_byte_in_range(std::integral_constant<bool, Keep> keep, std::initializer_list<char_int_type> ranges)
+    bool next_byte_in_range(std::initializer_list<char_int_type> ranges)
    {
        JSON_ASSERT(ranges.size() == 2 || ranges.size() == 4 || ranges.size() == 6);
-        add_string_char(keep, current);
+        add(current);

        for (auto range = ranges.begin(); range != ranges.end(); ++range)
        {
            get();
            if (JSON_HEDLEY_LIKELY(*range <= current && current <= *(++range))) // NOLINT(bugprone-inc-dec-in-conditions)
            {
-                add_string_char(keep, current);
+                add(current);
            }
            else
            {
@@ -422,16 +419,14 @@ class lexer : public lexer_base<BasicJsonType>
    }

    /// contiguous input: bulk-append the run of ordinary characters and complete
-    /// well-formed UTF-8 sequences starting at the current read position (only
-    /// skip it if @a keep is false), then
+    /// well-formed UTF-8 sequences starting at the current read position, then
    /// read the first byte that needs individual handling (the closing quote, an
    /// escape, a control character, or an ill-formed UTF-8 byte) like get().
    /// That byte is taken from the buffer that was just scanned. Through get(),
    /// Apple Clang merged this read with the other get() calls of the scan
    /// function, which load the read position again right after it was stored
    /// (3% of json::accept() on jeopardy.json on Apple M1).
-    template<bool Keep>
-    char_int_type get_after_string_run(std::true_type /*bulk*/, std::integral_constant<bool, Keep> keep)
+    char_int_type get_after_string_run(std::true_type /*bulk*/)
    {
        // a pending unget must be consumed through the normal path first
        if (next_unget)
@@ -449,7 +444,7 @@ class lexer : public lexer_base<BasicJsonType>
        const std::size_t pos = lexer_string_bulk_run(data, remaining);
        if (pos != 0)
        {
-            append_string_run(keep, data, pos);
+            token_buffer.append(reinterpret_cast<const typename string_t::value_type*>(data), pos);
            bulk_consume(reinterpret_cast<const char*>(data), pos);
        }
        if (pos == remaining)
@@ -493,8 +488,7 @@ class lexer : public lexer_base<BasicJsonType>
    }

    /// streaming input: no bulk fast path
-    template<bool Keep>
-    char_int_type get_after_string_run(std::false_type /*bulk*/, std::integral_constant<bool, Keep> /*keep*/)
+    char_int_type get_after_string_run(std::false_type /*bulk*/)
    {
        return get();
    }
@@ -508,18 +502,13 @@ class lexer : public lexer_base<BasicJsonType>
    may contain \0 bytes), and token_buffer.size() is the number of bytes in the
    string.

-    @param[in] keep  whether the string is copied into token_buffer; if false
-                     (accept(), see scan()), it is only checked, with the same
-                     errors, and token_buffer stays empty
-
    @return token_type::value_string if string could be successfully scanned,
            token_type::parse_error otherwise

    @note In case of errors, variable error_message contains a textual
          description.
    */
-    template<bool Keep>
-    token_type scan_string(std::integral_constant<bool, Keep> keep)
+    token_type scan_string()
    {
        // reset token_buffer (ignore opening quote)
        reset();
@@ -531,7 +520,7 @@ class lexer : public lexer_base<BasicJsonType>
        {
            // bulk-consume ordinary characters from contiguous input, then
            // get the next character, the one that ends the run
-            switch (get_after_string_run(std::integral_constant<bool, bulk_scan>(), keep))
+            switch (get_after_string_run(std::integral_constant<bool, bulk_scan>()))
            {
                // end of file while parsing the string
                case char_traits<char_type>::eof():
@@ -553,35 +542,35 @@ class lexer : public lexer_base<BasicJsonType>
                    {
                        // quotation mark
                        case '\"':
-                            add_string_char(keep, '\"');
+                            add('\"');
                            break;
                        // reverse solidus
                        case '\\':
-                            add_string_char(keep, '\\');
+                            add('\\');
                            break;
                        // solidus
                        case '/':
-                            add_string_char(keep, '/');
+                            add('/');
                            break;
                        // backspace
                        case 'b':
-                            add_string_char(keep, '\b');
+                            add('\b');
                            break;
                        // form feed
                        case 'f':
-                            add_string_char(keep, '\f');
+                            add('\f');
                            break;
                        // line feed
                        case 'n':
-                            add_string_char(keep, '\n');
+                            add('\n');
                            break;
                        // carriage return
                        case 'r':
-                            add_string_char(keep, '\r');
+                            add('\r');
                            break;
                        // tab
                        case 't':
-                            add_string_char(keep, '\t');
+                            add('\t');
                            break;

                        // unicode escapes
@@ -649,7 +638,10 @@ class lexer : public lexer_base<BasicJsonType>
                            JSON_ASSERT(0x00 <= codepoint && codepoint <= 0x10FFFF);

                            // translate codepoint into bytes
-                            add_codepoint(keep, codepoint);
+                            encode_utf8(static_cast<std::uint32_t>(codepoint), [this](std::uint32_t byte)
+                            {
+                                add(static_cast<char_int_type>(byte));
+                            });

                            break;
                        }
@@ -832,7 +824,7 @@ class lexer : public lexer_base<BasicJsonType>
                case 0x7E:
                case 0x7F:
                {
-                    add_string_char(keep, current);
+                    add(current);
                    break;
                }

@@ -868,7 +860,7 @@ class lexer : public lexer_base<BasicJsonType>
                case 0xDE:
                case 0xDF:
                {
-                    if (JSON_HEDLEY_UNLIKELY(!next_byte_in_range(keep, {0x80, 0xBF})))
+                    if (JSON_HEDLEY_UNLIKELY(!next_byte_in_range({0x80, 0xBF})))
                    {
                        return token_type::parse_error;
                    }
@@ -878,7 +870,7 @@ class lexer : public lexer_base<BasicJsonType>
                // U+0800..U+0FFF: bytes E0 A0..BF 80..BF
                case 0xE0:
                {
-                    if (JSON_HEDLEY_UNLIKELY(!(next_byte_in_range(keep, {0xA0, 0xBF, 0x80, 0xBF}))))
+                    if (JSON_HEDLEY_UNLIKELY(!(next_byte_in_range({0xA0, 0xBF, 0x80, 0xBF}))))
                    {
                        return token_type::parse_error;
                    }
@@ -902,7 +894,7 @@ class lexer : public lexer_base<BasicJsonType>
                case 0xEE:
                case 0xEF:
                {
-                    if (JSON_HEDLEY_UNLIKELY(!(next_byte_in_range(keep, {0x80, 0xBF, 0x80, 0xBF}))))
+                    if (JSON_HEDLEY_UNLIKELY(!(next_byte_in_range({0x80, 0xBF, 0x80, 0xBF}))))
                    {
                        return token_type::parse_error;
                    }
@@ -912,7 +904,7 @@ class lexer : public lexer_base<BasicJsonType>
                // U+D000..U+D7FF: bytes ED 80..9F 80..BF
                case 0xED:
                {
-                    if (JSON_HEDLEY_UNLIKELY(!(next_byte_in_range(keep, {0x80, 0x9F, 0x80, 0xBF}))))
+                    if (JSON_HEDLEY_UNLIKELY(!(next_byte_in_range({0x80, 0x9F, 0x80, 0xBF}))))
                    {
                        return token_type::parse_error;
                    }
@@ -922,7 +914,7 @@ class lexer : public lexer_base<BasicJsonType>
                // U+10000..U+3FFFF F0 90..BF 80..BF 80..BF
                case 0xF0:
                {
-                    if (JSON_HEDLEY_UNLIKELY(!(next_byte_in_range(keep, {0x90, 0xBF, 0x80, 0xBF, 0x80, 0xBF}))))
+                    if (JSON_HEDLEY_UNLIKELY(!(next_byte_in_range({0x90, 0xBF, 0x80, 0xBF, 0x80, 0xBF}))))
                    {
                        return token_type::parse_error;
                    }
@@ -934,7 +926,7 @@ class lexer : public lexer_base<BasicJsonType>
                case 0xF2:
                case 0xF3:
                {
-                    if (JSON_HEDLEY_UNLIKELY(!(next_byte_in_range(keep, {0x80, 0xBF, 0x80, 0xBF, 0x80, 0xBF}))))
+                    if (JSON_HEDLEY_UNLIKELY(!(next_byte_in_range({0x80, 0xBF, 0x80, 0xBF, 0x80, 0xBF}))))
                    {
                        return token_type::parse_error;
                    }
@@ -944,7 +936,7 @@ class lexer : public lexer_base<BasicJsonType>
                // U+100000..U+10FFFF F4 80..8F 80..BF 80..BF
                case 0xF4:
                {
-                    if (JSON_HEDLEY_UNLIKELY(!(next_byte_in_range(keep, {0x80, 0x8F, 0x80, 0xBF, 0x80, 0xBF}))))
+                    if (JSON_HEDLEY_UNLIKELY(!(next_byte_in_range({0x80, 0x8F, 0x80, 0xBF, 0x80, 0xBF}))))
                    {
                        return token_type::parse_error;
                    }
@@ -1435,7 +1427,7 @@ scan_number_done:
    token_type convert_number(token_type number_type, std::size_t mantissa_end)
    {
        // accept() only needs to know whether the input is valid, so it sets
-        // discard_values (see json.hpp), and an integer token whose
+        // discard_number_values (see json.hpp), and an integer token whose
        // digit count shows that it fits is reported without calling
        // convert_integer(). A number with up to 18 digits always fits into
        // both std::uint64_t and std::int64_t (18 nines is about 1e18, below
@@ -1450,7 +1442,7 @@ scan_number_done:
        // json_sax_acceptor, whose number callbacks discard their argument and
        // return true, and the parser rejects neither integers nor finite
        // floats. value_unsigned/value_integer are left unset here, so a caller
-        // that reads the converted value must not set discard_values.
+        // that reads the converted value must not set discard_number_values.
        //
        // On contiguous input, scan_number_bulk_contiguous() converts integer
        // tokens itself and does not pass them to this function, unless
@@ -1458,7 +1450,7 @@ scan_number_done:
        // reached for input without bulk access (e.g. streams), with
        // JSON_DIAGNOSTIC_POSITIONS, or when scan_number_bulk_contiguous()
        // falls back to scan_number().
-        if (discard_values)
+        if (discard_number_values)
        {
            constexpr std::size_t safe_digit_count = 18;
            if (number_type == token_type::value_unsigned && token_buffer.size() <= safe_digit_count)
@@ -1709,7 +1701,7 @@ scan_number_done:
        // w < 10^19, so a value below 10^max_exponent10 cannot overflow, and a
        // value that underflows is a finite number as well. Only a token that
        // may overflow has to be converted to find out.
-        if (JSON_HEDLEY_UNLIKELY(discard_values && s.exponent + 19 <= std::numeric_limits<number_float_t>::max_exponent10))
+        if (JSON_HEDLEY_UNLIKELY(discard_number_values && s.exponent + 19 <= std::numeric_limits<number_float_t>::max_exponent10))
        {
            value_float = 0;
            return true;
@@ -2041,34 +2033,6 @@ scan_number_done:
        token_buffer.push_back(static_cast<typename string_t::value_type>(c));
    }

-    /// add a character of a string token; accept() checks the string but does
-    /// not keep its value (@a keep is false)
-    void add_string_char(std::true_type /*keep*/, char_int_type c)
-    {
-        token_buffer.push_back(static_cast<typename string_t::value_type>(c));
-    }
-
-    static void add_string_char(std::false_type /*keep*/, char_int_type /*c*/) noexcept {}
-
-    /// add the UTF-8 bytes of a code point from a \\u escape to a string token
-    void add_codepoint(std::true_type /*keep*/, int codepoint)
-    {
-        encode_utf8(static_cast<std::uint32_t>(codepoint), [this](std::uint32_t byte)
-        {
-            add(static_cast<char_int_type>(byte));
-        });
-    }
-
-    static void add_codepoint(std::false_type /*keep*/, int /*codepoint*/) noexcept {}
-
-    /// append a run of ordinary string characters (see get_after_string_run())
-    void append_string_run(std::true_type /*keep*/, const unsigned char* data, std::size_t n)
-    {
-        token_buffer.append(reinterpret_cast<const typename string_t::value_type*>(data), n);
-    }
-
-    static void append_string_run(std::false_type /*keep*/, const unsigned char* /*data*/, std::size_t /*n*/) noexcept {}
-
  public:
    /////////////////////
    // value getters
@@ -2463,11 +2427,7 @@ scan_number_done:

            // string
            case '\"':
-                // accept() (discard_values) checks strings without copying them
-                // into token_buffer. The choice is made once per string: a
-                // check inside the string loop made number-heavy documents up
-                // to 3 % slower (mesh.json, Apple M1)
-                return discard_values ? scan_string(std::false_type {}) : scan_string(std::true_type {});
+                return scan_string();

            // number
            case '-':
@@ -2552,13 +2512,11 @@ scan_number_done:
    std::size_t decimal_point_position = std::string::npos;

    /// whether the caller only needs the token types and never looks at the
-    /// values of number and string tokens; set only by accept(), which parses
-    /// through json_sax_acceptor. When set, convert_number() skips converting
-    /// integer tokens whose digit count guarantees that they fit into 64 bits
-    /// (see there), floats are converted only when they may overflow (see
-    /// convert_accumulated_float()), and strings are checked without being
-    /// copied into token_buffer (see scan())
-    const bool discard_values = false;
+    /// converted numeric values; set only by accept(), which parses through
+    /// json_sax_acceptor. When set, convert_number() skips converting integer
+    /// tokens whose digit count guarantees that they fit into 64 bits (see
+    /// there)
+    const bool discard_number_values = false;
 };

 }  // namespace detail
@@ -89,9 +89,9 @@ class parser
                    const bool allow_exceptions_ = true,
                    const bool ignore_comments = false,
                    const bool ignore_trailing_commas_ = false,
-                    const bool discard_values_ = false)
+                    const bool discard_number_values_ = false)
        : callback(std::move(cb))
-        , m_lexer(std::move(adapter), ignore_comments, discard_values_)
+        , m_lexer(std::move(adapter), ignore_comments, discard_number_values_)
        , allow_exceptions(allow_exceptions_)
        , ignore_trailing_commas(ignore_trailing_commas_)
    {
@@ -1082,7 +1082,7 @@ indent_done:
            {
                ++p;
                const unsigned char* const f0 = p;
-                p = skip_digits(p, e);
+                p = skip_fraction_digits(p, e);
                if (NLOHMANN_VIEW_UNLIKELY(p == f0))
                {
                    return fail(error_code::number_after_dot);
@@ -198,6 +198,57 @@ inline const unsigned char* find_block_comment_end(const unsigned char* p, const
    return p;
 }

+/// The digits of a fraction, whose number varies from one number to the next: skip_digits() with 16 bytes compared at
+/// once and the end of the run taken from a bit mask (SSE2), instead of a compare and a branch for each digit. That
+/// is about a third of the instructions for a fraction of 12 digits, and no branch that is mispredicted when the
+/// lengths vary. Where there is no SSE2, skip_digits() is used.
+#if NLOHMANN_VIEW_SSE2
+// The last bytes of the input, one at a time. GCC unrolls this loop in every inlined copy of the function below, which
+// makes the parse loop 4 KiB larger and slows down documents that have hardly any fractions; out of line, it does
+// not. Clang does better with the loop inline (a call in the parse loop costs it more than the unrolling).
+#if defined(__GNUC__) && !defined(__clang__)
+inline NLOHMANN_VIEW_NOINLINE const unsigned char* skip_digits_tail(const unsigned char* p, const unsigned char* e) noexcept
+{
+    while (p != e && is_digit(*p))
+    {
+        ++p;
+    }
+    return p;
+}
+#else
+NLOHMANN_VIEW_ALWAYS_INLINE const unsigned char* skip_digits_tail(const unsigned char* p, const unsigned char* e) noexcept
+{
+    while (p != e && is_digit(*p))
+    {
+        ++p;
+    }
+    return p;
+}
+#endif
+
+NLOHMANN_VIEW_ALWAYS_INLINE const unsigned char* skip_fraction_digits(const unsigned char* p, const unsigned char* e) noexcept
+{
+    while (e - p >= 16)
+    {
+        const __m128i in = _mm_loadu_si128(static_cast<const __m128i*>(static_cast<const void*>(p))); // NOLINT(bugprone-casting-through-void)
+        // a digit is above '/' and below ':' (signed compares: a byte with the top bit set is negative, and neither)
+        const __m128i digits = _mm_and_si128(_mm_cmpgt_epi8(in, _mm_set1_epi8('/')), _mm_cmplt_epi8(in, _mm_set1_epi8(':')));
+        const auto not_digit = static_cast<std::uint64_t>(~static_cast<unsigned>(_mm_movemask_epi8(digits)) & 0xFFFFu);
+        if (not_digit != 0)
+        {
+            return p + count_trailing_zeros(not_digit);
+        }
+        p += 16;
+    }
+    return skip_digits_tail(p, e);
+}
+#else
+NLOHMANN_VIEW_ALWAYS_INLINE const unsigned char* skip_fraction_digits(const unsigned char* p, const unsigned char* e) noexcept
+{
+    return skip_digits(p, e);
+}
+#endif
+
 /// powers of ten up to 10^19 as integers
 inline std::uint64_t int_pow10(unsigned k) noexcept
 {
@@ -184,11 +184,11 @@ class basic_json // NOLINT(cppcoreguidelines-special-member-functions,hicpp-spec
        const bool allow_exceptions = true,
        const bool ignore_comments = false,
        const bool ignore_trailing_commas = false,
-        const bool discard_values = false
+        const bool discard_number_values = false
    )
    {
        return ::nlohmann::detail::parser<basic_json, InputAdapterType>(std::move(adapter),
-               std::move(cb), allow_exceptions, ignore_comments, ignore_trailing_commas, discard_values);
+               std::move(cb), allow_exceptions, ignore_comments, ignore_trailing_commas, discard_number_values);
    }

  private:
@@ -13124,10 +13124,10 @@ class lexer : public lexer_base<BasicJsonType>
  public:
    using token_type = typename lexer_base<BasicJsonType>::token_type;

-    explicit lexer(InputAdapterType&& adapter, bool ignore_comments_ = false, bool discard_values_ = false) noexcept
+    explicit lexer(InputAdapterType&& adapter, bool ignore_comments_ = false, bool discard_number_values_ = false) noexcept
        : ia(std::move(adapter))
        , ignore_comments(ignore_comments_)
-        , discard_values(discard_values_)
+        , discard_number_values(discard_number_values_)
    {
        // the token string of an error before the first token is everything
        // read so far
@@ -13244,8 +13244,6 @@ class lexer : public lexer_base<BasicJsonType>
    checks if it is inside the range. If a violation was detected, set up an
    error message and return false. Otherwise, return true.

-    @param[in] keep    whether the bytes are added to token_buffer (false for
-                       accept(), see scan())
    @param[in] ranges  list of integers; interpreted as list of pairs of
                       inclusive lower and upper bound, respectively

@@ -13254,18 +13252,17 @@ class lexer : public lexer_base<BasicJsonType>

    @return true if and only if no range violation was detected
    */
-    template<bool Keep>
-    bool next_byte_in_range(std::integral_constant<bool, Keep> keep, std::initializer_list<char_int_type> ranges)
+    bool next_byte_in_range(std::initializer_list<char_int_type> ranges)
    {
        JSON_ASSERT(ranges.size() == 2 || ranges.size() == 4 || ranges.size() == 6);
-        add_string_char(keep, current);
+        add(current);

        for (auto range = ranges.begin(); range != ranges.end(); ++range)
        {
            get();
            if (JSON_HEDLEY_LIKELY(*range <= current && current <= *(++range))) // NOLINT(bugprone-inc-dec-in-conditions)
            {
-                add_string_char(keep, current);
+                add(current);
            }
            else
            {
@@ -13278,16 +13275,14 @@ class lexer : public lexer_base<BasicJsonType>
    }

    /// contiguous input: bulk-append the run of ordinary characters and complete
-    /// well-formed UTF-8 sequences starting at the current read position (only
-    /// skip it if @a keep is false), then
+    /// well-formed UTF-8 sequences starting at the current read position, then
    /// read the first byte that needs individual handling (the closing quote, an
    /// escape, a control character, or an ill-formed UTF-8 byte) like get().
    /// That byte is taken from the buffer that was just scanned. Through get(),
    /// Apple Clang merged this read with the other get() calls of the scan
    /// function, which load the read position again right after it was stored
    /// (3% of json::accept() on jeopardy.json on Apple M1).
-    template<bool Keep>
-    char_int_type get_after_string_run(std::true_type /*bulk*/, std::integral_constant<bool, Keep> keep)
+    char_int_type get_after_string_run(std::true_type /*bulk*/)
    {
        // a pending unget must be consumed through the normal path first
        if (next_unget)
@@ -13305,7 +13300,7 @@ class lexer : public lexer_base<BasicJsonType>
        const std::size_t pos = lexer_string_bulk_run(data, remaining);
        if (pos != 0)
        {
-            append_string_run(keep, data, pos);
+            token_buffer.append(reinterpret_cast<const typename string_t::value_type*>(data), pos);
            bulk_consume(reinterpret_cast<const char*>(data), pos);
        }
        if (pos == remaining)
@@ -13349,8 +13344,7 @@ class lexer : public lexer_base<BasicJsonType>
    }

    /// streaming input: no bulk fast path
-    template<bool Keep>
-    char_int_type get_after_string_run(std::false_type /*bulk*/, std::integral_constant<bool, Keep> /*keep*/)
+    char_int_type get_after_string_run(std::false_type /*bulk*/)
    {
        return get();
    }
@@ -13364,18 +13358,13 @@ class lexer : public lexer_base<BasicJsonType>
    may contain \0 bytes), and token_buffer.size() is the number of bytes in the
    string.

-    @param[in] keep  whether the string is copied into token_buffer; if false
-                     (accept(), see scan()), it is only checked, with the same
-                     errors, and token_buffer stays empty
-
    @return token_type::value_string if string could be successfully scanned,
            token_type::parse_error otherwise

    @note In case of errors, variable error_message contains a textual
          description.
    */
-    template<bool Keep>
-    token_type scan_string(std::integral_constant<bool, Keep> keep)
+    token_type scan_string()
    {
        // reset token_buffer (ignore opening quote)
        reset();
@@ -13387,7 +13376,7 @@ class lexer : public lexer_base<BasicJsonType>
        {
            // bulk-consume ordinary characters from contiguous input, then
            // get the next character, the one that ends the run
-            switch (get_after_string_run(std::integral_constant<bool, bulk_scan>(), keep))
+            switch (get_after_string_run(std::integral_constant<bool, bulk_scan>()))
            {
                // end of file while parsing the string
                case char_traits<char_type>::eof():
@@ -13409,35 +13398,35 @@ class lexer : public lexer_base<BasicJsonType>
                    {
                        // quotation mark
                        case '\"':
-                            add_string_char(keep, '\"');
+                            add('\"');
                            break;
                        // reverse solidus
                        case '\\':
-                            add_string_char(keep, '\\');
+                            add('\\');
                            break;
                        // solidus
                        case '/':
-                            add_string_char(keep, '/');
+                            add('/');
                            break;
                        // backspace
                        case 'b':
-                            add_string_char(keep, '\b');
+                            add('\b');
                            break;
                        // form feed
                        case 'f':
-                            add_string_char(keep, '\f');
+                            add('\f');
                            break;
                        // line feed
                        case 'n':
-                            add_string_char(keep, '\n');
+                            add('\n');
                            break;
                        // carriage return
                        case 'r':
-                            add_string_char(keep, '\r');
+                            add('\r');
                            break;
                        // tab
                        case 't':
-                            add_string_char(keep, '\t');
+                            add('\t');
                            break;

                        // unicode escapes
@@ -13505,7 +13494,10 @@ class lexer : public lexer_base<BasicJsonType>
                            JSON_ASSERT(0x00 <= codepoint && codepoint <= 0x10FFFF);

                            // translate codepoint into bytes
-                            add_codepoint(keep, codepoint);
+                            encode_utf8(static_cast<std::uint32_t>(codepoint), [this](std::uint32_t byte)
+                            {
+                                add(static_cast<char_int_type>(byte));
+                            });

                            break;
                        }
@@ -13688,7 +13680,7 @@ class lexer : public lexer_base<BasicJsonType>
                case 0x7E:
                case 0x7F:
                {
-                    add_string_char(keep, current);
+                    add(current);
                    break;
                }

@@ -13724,7 +13716,7 @@ class lexer : public lexer_base<BasicJsonType>
                case 0xDE:
                case 0xDF:
                {
-                    if (JSON_HEDLEY_UNLIKELY(!next_byte_in_range(keep, {0x80, 0xBF})))
+                    if (JSON_HEDLEY_UNLIKELY(!next_byte_in_range({0x80, 0xBF})))
                    {
                        return token_type::parse_error;
                    }
@@ -13734,7 +13726,7 @@ class lexer : public lexer_base<BasicJsonType>
                // U+0800..U+0FFF: bytes E0 A0..BF 80..BF
                case 0xE0:
                {
-                    if (JSON_HEDLEY_UNLIKELY(!(next_byte_in_range(keep, {0xA0, 0xBF, 0x80, 0xBF}))))
+                    if (JSON_HEDLEY_UNLIKELY(!(next_byte_in_range({0xA0, 0xBF, 0x80, 0xBF}))))
                    {
                        return token_type::parse_error;
                    }
@@ -13758,7 +13750,7 @@ class lexer : public lexer_base<BasicJsonType>
                case 0xEE:
                case 0xEF:
                {
-                    if (JSON_HEDLEY_UNLIKELY(!(next_byte_in_range(keep, {0x80, 0xBF, 0x80, 0xBF}))))
+                    if (JSON_HEDLEY_UNLIKELY(!(next_byte_in_range({0x80, 0xBF, 0x80, 0xBF}))))
                    {
                        return token_type::parse_error;
                    }
@@ -13768,7 +13760,7 @@ class lexer : public lexer_base<BasicJsonType>
                // U+D000..U+D7FF: bytes ED 80..9F 80..BF
                case 0xED:
                {
-                    if (JSON_HEDLEY_UNLIKELY(!(next_byte_in_range(keep, {0x80, 0x9F, 0x80, 0xBF}))))
+                    if (JSON_HEDLEY_UNLIKELY(!(next_byte_in_range({0x80, 0x9F, 0x80, 0xBF}))))
                    {
                        return token_type::parse_error;
                    }
@@ -13778,7 +13770,7 @@ class lexer : public lexer_base<BasicJsonType>
                // U+10000..U+3FFFF F0 90..BF 80..BF 80..BF
                case 0xF0:
                {
-                    if (JSON_HEDLEY_UNLIKELY(!(next_byte_in_range(keep, {0x90, 0xBF, 0x80, 0xBF, 0x80, 0xBF}))))
+                    if (JSON_HEDLEY_UNLIKELY(!(next_byte_in_range({0x90, 0xBF, 0x80, 0xBF, 0x80, 0xBF}))))
                    {
                        return token_type::parse_error;
                    }
@@ -13790,7 +13782,7 @@ class lexer : public lexer_base<BasicJsonType>
                case 0xF2:
                case 0xF3:
                {
-                    if (JSON_HEDLEY_UNLIKELY(!(next_byte_in_range(keep, {0x80, 0xBF, 0x80, 0xBF, 0x80, 0xBF}))))
+                    if (JSON_HEDLEY_UNLIKELY(!(next_byte_in_range({0x80, 0xBF, 0x80, 0xBF, 0x80, 0xBF}))))
                    {
                        return token_type::parse_error;
                    }
@@ -13800,7 +13792,7 @@ class lexer : public lexer_base<BasicJsonType>
                // U+100000..U+10FFFF F4 80..8F 80..BF 80..BF
                case 0xF4:
                {
-                    if (JSON_HEDLEY_UNLIKELY(!(next_byte_in_range(keep, {0x80, 0x8F, 0x80, 0xBF, 0x80, 0xBF}))))
+                    if (JSON_HEDLEY_UNLIKELY(!(next_byte_in_range({0x80, 0x8F, 0x80, 0xBF, 0x80, 0xBF}))))
                    {
                        return token_type::parse_error;
                    }
@@ -14291,7 +14283,7 @@ scan_number_done:
    token_type convert_number(token_type number_type, std::size_t mantissa_end)
    {
        // accept() only needs to know whether the input is valid, so it sets
-        // discard_values (see json.hpp), and an integer token whose
+        // discard_number_values (see json.hpp), and an integer token whose
        // digit count shows that it fits is reported without calling
        // convert_integer(). A number with up to 18 digits always fits into
        // both std::uint64_t and std::int64_t (18 nines is about 1e18, below
@@ -14306,7 +14298,7 @@ scan_number_done:
        // json_sax_acceptor, whose number callbacks discard their argument and
        // return true, and the parser rejects neither integers nor finite
        // floats. value_unsigned/value_integer are left unset here, so a caller
-        // that reads the converted value must not set discard_values.
+        // that reads the converted value must not set discard_number_values.
        //
        // On contiguous input, scan_number_bulk_contiguous() converts integer
        // tokens itself and does not pass them to this function, unless
@@ -14314,7 +14306,7 @@ scan_number_done:
        // reached for input without bulk access (e.g. streams), with
        // JSON_DIAGNOSTIC_POSITIONS, or when scan_number_bulk_contiguous()
        // falls back to scan_number().
-        if (discard_values)
+        if (discard_number_values)
        {
            constexpr std::size_t safe_digit_count = 18;
            if (number_type == token_type::value_unsigned && token_buffer.size() <= safe_digit_count)
@@ -14565,7 +14557,7 @@ scan_number_done:
        // w < 10^19, so a value below 10^max_exponent10 cannot overflow, and a
        // value that underflows is a finite number as well. Only a token that
        // may overflow has to be converted to find out.
-        if (JSON_HEDLEY_UNLIKELY(discard_values && s.exponent + 19 <= std::numeric_limits<number_float_t>::max_exponent10))
+        if (JSON_HEDLEY_UNLIKELY(discard_number_values && s.exponent + 19 <= std::numeric_limits<number_float_t>::max_exponent10))
        {
            value_float = 0;
            return true;
@@ -14897,34 +14889,6 @@ scan_number_done:
        token_buffer.push_back(static_cast<typename string_t::value_type>(c));
    }

-    /// add a character of a string token; accept() checks the string but does
-    /// not keep its value (@a keep is false)
-    void add_string_char(std::true_type /*keep*/, char_int_type c)
-    {
-        token_buffer.push_back(static_cast<typename string_t::value_type>(c));
-    }
-
-    static void add_string_char(std::false_type /*keep*/, char_int_type /*c*/) noexcept {}
-
-    /// add the UTF-8 bytes of a code point from a \\u escape to a string token
-    void add_codepoint(std::true_type /*keep*/, int codepoint)
-    {
-        encode_utf8(static_cast<std::uint32_t>(codepoint), [this](std::uint32_t byte)
-        {
-            add(static_cast<char_int_type>(byte));
-        });
-    }
-
-    static void add_codepoint(std::false_type /*keep*/, int /*codepoint*/) noexcept {}
-
-    /// append a run of ordinary string characters (see get_after_string_run())
-    void append_string_run(std::true_type /*keep*/, const unsigned char* data, std::size_t n)
-    {
-        token_buffer.append(reinterpret_cast<const typename string_t::value_type*>(data), n);
-    }
-
-    static void append_string_run(std::false_type /*keep*/, const unsigned char* /*data*/, std::size_t /*n*/) noexcept {}
-
  public:
    /////////////////////
    // value getters
@@ -15319,11 +15283,7 @@ scan_number_done:

            // string
            case '\"':
-                // accept() (discard_values) checks strings without copying them
-                // into token_buffer. The choice is made once per string: a
-                // check inside the string loop made number-heavy documents up
-                // to 3 % slower (mesh.json, Apple M1)
-                return discard_values ? scan_string(std::false_type {}) : scan_string(std::true_type {});
+                return scan_string();

            // number
            case '-':
@@ -15408,13 +15368,11 @@ scan_number_done:
    std::size_t decimal_point_position = std::string::npos;

    /// whether the caller only needs the token types and never looks at the
-    /// values of number and string tokens; set only by accept(), which parses
-    /// through json_sax_acceptor. When set, convert_number() skips converting
-    /// integer tokens whose digit count guarantees that they fit into 64 bits
-    /// (see there), floats are converted only when they may overflow (see
-    /// convert_accumulated_float()), and strings are checked without being
-    /// copied into token_buffer (see scan())
-    const bool discard_values = false;
+    /// converted numeric values; set only by accept(), which parses through
+    /// json_sax_acceptor. When set, convert_number() skips converting integer
+    /// tokens whose digit count guarantees that they fit into 64 bits (see
+    /// there)
+    const bool discard_number_values = false;
 };

 }  // namespace detail
@@ -22623,9 +22581,9 @@ class parser
                    const bool allow_exceptions_ = true,
                    const bool ignore_comments = false,
                    const bool ignore_trailing_commas_ = false,
-                    const bool discard_values_ = false)
+                    const bool discard_number_values_ = false)
        : callback(std::move(cb))
-        , m_lexer(std::move(adapter), ignore_comments, discard_values_)
+        , m_lexer(std::move(adapter), ignore_comments, discard_number_values_)
        , allow_exceptions(allow_exceptions_)
        , ignore_trailing_commas(ignore_trailing_commas_)
    {
@@ -37022,11 +36980,11 @@ class basic_json // NOLINT(cppcoreguidelines-special-member-functions,hicpp-spec
        const bool allow_exceptions = true,
        const bool ignore_comments = false,
        const bool ignore_trailing_commas = false,
-        const bool discard_values = false
+        const bool discard_number_values = false
                                 )
    {
        return ::nlohmann::detail::parser<basic_json, InputAdapterType>(std::move(adapter),
-            std::move(cb), allow_exceptions, ignore_comments, ignore_trailing_commas, discard_values);
+            std::move(cb), allow_exceptions, ignore_comments, ignore_trailing_commas, discard_number_values);
    }

  private:
@@ -1218,6 +1218,57 @@ inline const unsigned char* find_block_comment_end(const unsigned char* p, const
    return p;
 }

+/// The digits of a fraction, whose number varies from one number to the next: skip_digits() with 16 bytes compared at
+/// once and the end of the run taken from a bit mask (SSE2), instead of a compare and a branch for each digit. That
+/// is about a third of the instructions for a fraction of 12 digits, and no branch that is mispredicted when the
+/// lengths vary. Where there is no SSE2, skip_digits() is used.
+#if NLOHMANN_VIEW_SSE2
+// The last bytes of the input, one at a time. GCC unrolls this loop in every inlined copy of the function below, which
+// makes the parse loop 4 KiB larger and slows down documents that have hardly any fractions; out of line, it does
+// not. Clang does better with the loop inline (a call in the parse loop costs it more than the unrolling).
+#if defined(__GNUC__) && !defined(__clang__)
+inline NLOHMANN_VIEW_NOINLINE const unsigned char* skip_digits_tail(const unsigned char* p, const unsigned char* e) noexcept
+{
+    while (p != e && is_digit(*p))
+    {
+        ++p;
+    }
+    return p;
+}
+#else
+NLOHMANN_VIEW_ALWAYS_INLINE const unsigned char* skip_digits_tail(const unsigned char* p, const unsigned char* e) noexcept
+{
+    while (p != e && is_digit(*p))
+    {
+        ++p;
+    }
+    return p;
+}
+#endif
+
+NLOHMANN_VIEW_ALWAYS_INLINE const unsigned char* skip_fraction_digits(const unsigned char* p, const unsigned char* e) noexcept
+{
+    while (e - p >= 16)
+    {
+        const __m128i in = _mm_loadu_si128(static_cast<const __m128i*>(static_cast<const void*>(p))); // NOLINT(bugprone-casting-through-void)
+        // a digit is above '/' and below ':' (signed compares: a byte with the top bit set is negative, and neither)
+        const __m128i digits = _mm_and_si128(_mm_cmpgt_epi8(in, _mm_set1_epi8('/')), _mm_cmplt_epi8(in, _mm_set1_epi8(':')));
+        const auto not_digit = static_cast<std::uint64_t>(~static_cast<unsigned>(_mm_movemask_epi8(digits)) & 0xFFFFu);
+        if (not_digit != 0)
+        {
+            return p + count_trailing_zeros(not_digit);
+        }
+        p += 16;
+    }
+    return skip_digits_tail(p, e);
+}
+#else
+NLOHMANN_VIEW_ALWAYS_INLINE const unsigned char* skip_fraction_digits(const unsigned char* p, const unsigned char* e) noexcept
+{
+    return skip_digits(p, e);
+}
+#endif
+
 /// powers of ten up to 10^19 as integers
 inline std::uint64_t int_pow10(unsigned k) noexcept
 {
@@ -2324,7 +2375,7 @@ indent_done:
            {
                ++p;
                const unsigned char* const f0 = p;
-                p = skip_digits(p, e);
+                p = skip_fraction_digits(p, e);
                if (NLOHMANN_VIEW_UNLIKELY(p == f0))
                {
                    return fail(error_code::number_after_dot);
@@ -2982,50 +2982,6 @@ TEST_CASE("accept() agrees with parse() on floats around the overflow limit")
    }
 }

-TEST_CASE("accept() agrees with parse() on strings, which it checks without copying them")
-{
-    // accept() scans strings and object keys with a separate instantiation of
-    // the string scanner that does not copy them into the token buffer; it
-    // must accept and reject exactly what parse() does
-    const std::vector<std::string> pieces =
-    {
-        "", "abc", "\\\"\\\\\\/\\b\\f\\n\\r\\t", "\\u0041\\u00e4\\u20AC", "\\ud83d\\ude00", "\\uD83D\\uDE00",
-        "\\ud800", "\\udc00", "\\ud800\\u0041", "\\ud800x", "\\udbff\\udfff", "\\u12", "\\u12g4", "\\x", "\\",
-        "\xc3\xa4", "\xe2\x82\xac", "\xf0\x9f\x98\x80", "\xf4\x8f\xbf\xbf", "\xc0\xaf", "\xed\xa0\x80", "\xf4\x90\x80\x80",
-        "\xff", "\xe2\x82", "\xc3", "\x01", "\x1f", "\x7f", "\n", std::string(1, '\0')
-    };
-
-    for (std::size_t run = 0; run <= 70; ++run)
-    {
-        for (const std::string& piece : pieces)
-        {
-            // a run of ordinary characters puts the piece at every offset of the bulk scanners
-            const std::string str = std::string(run, 'a') + piece;
-            for (const std::string& text :
-                    {
-                        "\"" + str + "\"", "[\"" + str + "\"]", "{\"" + str + "\":1}", "{\"k\":\"" + str + "\",\"" + str + "\":[\"" + str + "\"]}",
-                        "[\"" + str
-                    })
-            {
-                CAPTURE(text)
-                for (int options = 0; options < 4; ++options)
-                {
-                    const bool ignore_comments = (options & 1) != 0;
-                    const bool ignore_trailing_commas = (options & 2) != 0;
-                    const bool parses = !json::parse(text, nullptr, false, ignore_comments, ignore_trailing_commas).is_discarded();
-                    CHECK(json::accept(text, ignore_comments, ignore_trailing_commas) == parses);
-
-                    std::istringstream stream(text);
-                    CHECK(json::accept(stream, ignore_comments, ignore_trailing_commas) == parses);
-
-                    const std::list<char> chars(text.begin(), text.end());
-                    CHECK(json::accept(chars.begin(), chars.end(), ignore_comments, ignore_trailing_commas) == parses);
-                }
-            }
-        }
-    }
-}
-
 namespace
 {
 // a DOM parser that reports parse errors without throwing, as in the sax_no_exception example
@@ -523,6 +523,47 @@ TEST_CASE("json_view builder: strings across vector blocks")
    }
 }

+TEST_CASE("json_view builder: fraction digits")
+{
+    // The digits of a fraction are skipped 16 bytes at a time where there are
+    // that many left (SSE2), and one at a time after that. Fractions of every
+    // length from 0 to 40 digits, followed by every kind of byte that ends a
+    // number or lies next to the digits in ASCII ('/' and ':'), as values, as
+    // keys' neighbours, and at the very end of the text (a vector without a
+    // terminating NUL). The length of the number is part of the value that
+    // the test compares with json::parse().
+    const std::string followers = ",]}/:.eE+-x \n\t\"09";
+    std::mt19937 rng(2028); // NOLINT(cert-msc32-c,cert-msc51-cpp,bugprone-random-generator-seed): reproducible
+    for (std::size_t digits = 0; digits <= 40; ++digits)
+    {
+        for (int i = 0; i < 3; ++i)
+        {
+            std::string fraction;
+            for (std::size_t k = 0; k < digits; ++k)
+            {
+                fraction += static_cast<char>('0' + (rng() % 10));
+            }
+            const std::string number = std::string(rng() % 2 == 0 ? "-" : "") + std::to_string(rng() % 100000) + "." + fraction;
+            check_same(number);
+            check_same("[" + number + "]");
+            check_same("[" + number + "," + number + "]");
+            check_same("{\"a\":" + number + "}");
+            for (const char follower : followers)
+            {
+                check_same(number + follower);
+                check_same("[" + number + follower + "]");
+                check_same("[" + number + follower + std::string(rng() % 20, '1'));
+            }
+            // everything near the end of the text: the number cut in every place
+            for (std::size_t cut = 1; cut <= 18 && cut <= number.size(); ++cut)
+            {
+                check_same(number.substr(0, number.size() - cut));
+                check_same("[" + number.substr(0, number.size() - cut));
+            }
+        }
+    }
+}
+
 TEST_CASE("json_view builder: escapes between runs of text")
 {
    // The text between escapes is copied 16 bytes at a time where there are