@hxa-rn/react-native-quick-crypto 1.1.0-beta.1

package/harmony/quick_crypto/deps/simdutf/src/westmere/sse_convert_latin1_to_utf32.cpp

@@ -0,0 +1,31 @@
+std::pair<const char *, char32_t *>
+sse_convert_latin1_to_utf32(const char *buf, size_t len,
+                            char32_t *utf32_output) {
+  const char *end = buf + len;
+
+  while (end - buf >= 16) {
+    // Load 16 Latin1 characters (16 bytes) into a 128-bit register
+    __m128i in = _mm_loadu_si128((__m128i *)buf);
+
+    // Shift input to process next 4 bytes
+    __m128i in_shifted1 = _mm_srli_si128(in, 4);
+    __m128i in_shifted2 = _mm_srli_si128(in, 8);
+    __m128i in_shifted3 = _mm_srli_si128(in, 12);
+
+    // expand 8-bit to 32-bit unit
+    __m128i out1 = _mm_cvtepu8_epi32(in);
+    __m128i out2 = _mm_cvtepu8_epi32(in_shifted1);
+    __m128i out3 = _mm_cvtepu8_epi32(in_shifted2);
+    __m128i out4 = _mm_cvtepu8_epi32(in_shifted3);
+
+    _mm_storeu_si128((__m128i *)utf32_output, out1);
+    _mm_storeu_si128((__m128i *)(utf32_output + 4), out2);
+    _mm_storeu_si128((__m128i *)(utf32_output + 8), out3);
+    _mm_storeu_si128((__m128i *)(utf32_output + 12), out4);
+
+    utf32_output += 16;
+    buf += 16;
+  }
+
+  return std::make_pair(buf, utf32_output);
+}

package/harmony/quick_crypto/deps/simdutf/src/westmere/sse_convert_latin1_to_utf8.cpp

@@ -0,0 +1,71 @@
+std::pair<const char *const, char *const>
+sse_convert_latin1_to_utf8(const char *latin_input,
+                           const size_t latin_input_length, char *utf8_output) {
+  const char *end = latin_input + latin_input_length;
+
+  const __m128i v_0000 = _mm_setzero_si128();
+  // 0b1000_0000
+  const __m128i v_80 = _mm_set1_epi8((uint8_t)0x80);
+  // 0b1111_1111_1000_0000
+  const __m128i v_ff80 = _mm_set1_epi16((uint16_t)0xff80);
+
+  const __m128i latin_1_half_into_u16_byte_mask =
+      _mm_setr_epi8(0, '\x80', 1, '\x80', 2, '\x80', 3, '\x80', 4, '\x80', 5,
+                    '\x80', 6, '\x80', 7, '\x80');
+
+  const __m128i latin_2_half_into_u16_byte_mask =
+      _mm_setr_epi8(8, '\x80', 9, '\x80', 10, '\x80', 11, '\x80', 12, '\x80',
+                    13, '\x80', 14, '\x80', 15, '\x80');
+
+  // each latin1 takes 1-2 utf8 bytes
+  // slow path writes useful 8-15 bytes twice (eagerly writes 16 bytes and then
+  // adjust the pointer) so the last write can exceed the utf8_output size by
+  // 8-1 bytes by reserving 8 extra input bytes, we expect the output to have
+  // 8-16 bytes free
+  while (end - latin_input >= 16 + 8) {
+    // Load 16 Latin1 characters (16 bytes) into a 128-bit register
+    __m128i v_latin = _mm_loadu_si128((__m128i *)latin_input);
+
+    if (_mm_testz_si128(v_latin, v_80)) { // ASCII fast path!!!!
+      _mm_storeu_si128((__m128i *)utf8_output, v_latin);
+      latin_input += 16;
+      utf8_output += 16;
+      continue;
+    }
+
+    // assuming a/b are bytes and A/B are uint16 of the same value
+    // aaaa_aaaa_bbbb_bbbb -> AAAA_AAAA
+    __m128i v_u16_latin_1_half =
+        _mm_shuffle_epi8(v_latin, latin_1_half_into_u16_byte_mask);
+    // aaaa_aaaa_bbbb_bbbb -> BBBB_BBBB
+    __m128i v_u16_latin_2_half =
+        _mm_shuffle_epi8(v_latin, latin_2_half_into_u16_byte_mask);
+
+    internal::westmere::write_v_u16_11bits_to_utf8(v_u16_latin_1_half,
+                                                   utf8_output, v_0000, v_ff80);
+    internal::westmere::write_v_u16_11bits_to_utf8(v_u16_latin_2_half,
+                                                   utf8_output, v_0000, v_ff80);
+    latin_input += 16;
+  }
+
+  if (end - latin_input >= 16) {
+    // Load 16 Latin1 characters (16 bytes) into a 128-bit register
+    __m128i v_latin = _mm_loadu_si128((__m128i *)latin_input);
+
+    if (_mm_testz_si128(v_latin, v_80)) { // ASCII fast path!!!!
+      _mm_storeu_si128((__m128i *)utf8_output, v_latin);
+      latin_input += 16;
+      utf8_output += 16;
+    } else {
+      // assuming a/b are bytes and A/B are uint16 of the same value
+      // aaaa_aaaa_bbbb_bbbb -> AAAA_AAAA
+      __m128i v_u16_latin_1_half =
+          _mm_shuffle_epi8(v_latin, latin_1_half_into_u16_byte_mask);
+      internal::westmere::write_v_u16_11bits_to_utf8(
+          v_u16_latin_1_half, utf8_output, v_0000, v_ff80);
+      latin_input += 8;
+    }
+  }
+
+  return std::make_pair(latin_input, utf8_output);
+}

package/harmony/quick_crypto/deps/simdutf/src/westmere/sse_convert_utf16_to_latin1.cpp

@@ -0,0 +1,70 @@
+template <endianness big_endian>
+std::pair<const char16_t *, char *>
+sse_convert_utf16_to_latin1(const char16_t *buf, size_t len,
+                            char *latin1_output) {
+  const char16_t *end = buf + len;
+  while (end - buf >= 8) {
+    // Load 8 UTF-16 characters into 128-bit SSE register
+    __m128i in = _mm_loadu_si128(reinterpret_cast<const __m128i *>(buf));
+
+    if simdutf_constexpr (!match_system(big_endian)) {
+      const __m128i swap =
+          _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14);
+      in = _mm_shuffle_epi8(in, swap);
+    }
+
+    __m128i high_byte_mask = _mm_set1_epi16((int16_t)0xFF00);
+    if (_mm_testz_si128(in, high_byte_mask)) {
+      // Pack 16-bit characters into 8-bit and store in latin1_output
+      __m128i latin1_packed = _mm_packus_epi16(in, in);
+      _mm_storel_epi64(reinterpret_cast<__m128i *>(latin1_output),
+                       latin1_packed);
+      // Adjust pointers for next iteration
+      buf += 8;
+      latin1_output += 8;
+    } else {
+      return std::make_pair(nullptr, reinterpret_cast<char *>(latin1_output));
+    }
+  } // while
+  return std::make_pair(buf, latin1_output);
+}
+
+template <endianness big_endian>
+std::pair<result, char *>
+sse_convert_utf16_to_latin1_with_errors(const char16_t *buf, size_t len,
+                                        char *latin1_output) {
+  const char16_t *start = buf;
+  const char16_t *end = buf + len;
+  while (end - buf >= 8) {
+    __m128i in = _mm_loadu_si128(reinterpret_cast<const __m128i *>(buf));
+
+    if simdutf_constexpr (!match_system(big_endian)) {
+      const __m128i swap =
+          _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14);
+      in = _mm_shuffle_epi8(in, swap);
+    }
+
+    __m128i high_byte_mask = _mm_set1_epi16((int16_t)0xFF00);
+    if (_mm_testz_si128(in, high_byte_mask)) {
+      __m128i latin1_packed = _mm_packus_epi16(in, in);
+      _mm_storel_epi64(reinterpret_cast<__m128i *>(latin1_output),
+                       latin1_packed);
+      buf += 8;
+      latin1_output += 8;
+    } else {
+      // Fallback to scalar code for handling errors
+      for (int k = 0; k < 8; k++) {
+        uint16_t word = scalar::utf16::swap_if_needed<big_endian>(buf[k]);
+        if (word <= 0xff) {
+          *latin1_output++ = char(word);
+        } else {
+          return std::make_pair(result(error_code::TOO_LARGE, buf - start + k),
+                                latin1_output);
+        }
+      }
+      buf += 8;
+    }
+  } // while
+  return std::make_pair(result(error_code::SUCCESS, buf - start),
+                        latin1_output);
+}

package/harmony/quick_crypto/deps/simdutf/src/westmere/sse_convert_utf16_to_utf32.cpp

@@ -0,0 +1,206 @@
+/*
+    The vectorized algorithm works on single SSE register i.e., it
+    loads eight 16-bit code units.
+
+    We consider three cases:
+    1. an input register contains no surrogates and each value
+       is in range 0x0000 .. 0x07ff.
+    2. an input register contains no surrogates and values are
+       is in range 0x0000 .. 0xffff.
+    3. an input register contains surrogates --- i.e. codepoints
+       can have 16 or 32 bits.
+
+    Ad 1.
+
+    When values are less than 0x0800, it means that a 16-bit code unit
+    can be converted into: 1) single UTF8 byte (when it's an ASCII
+    char) or 2) two UTF8 bytes.
+
+    For this case we do only some shuffle to obtain these 2-byte
+    codes and finally compress the whole SSE register with a single
+    shuffle.
+
+    We need 256-entry lookup table to get a compression pattern
+    and the number of output bytes in the compressed vector register.
+    Each entry occupies 17 bytes.
+
+    Ad 2.
+
+    When values fit in 16-bit code units, but are above 0x07ff, then
+    a single word may produce one, two or three UTF8 bytes.
+
+    We prepare data for all these three cases in two registers.
+    The first register contains lower two UTF8 bytes (used in all
+    cases), while the second one contains just the third byte for
+    the three-UTF8-bytes case.
+
+    Finally these two registers are interleaved forming eight-element
+    array of 32-bit values. The array spans two SSE registers.
+    The bytes from the registers are compressed using two shuffles.
+
+    We need 256-entry lookup table to get a compression pattern
+    and the number of output bytes in the compressed vector register.
+    Each entry occupies 17 bytes.
+
+
+    To summarize:
+    - We need two 256-entry tables that have 8704 bytes in total.
+*/
+
+/*
+  Returns a pair: the first unprocessed byte from buf and utf8_output
+  A scalar routine should carry on the conversion of the tail.
+*/
+template <endianness big_endian>
+std::pair<const char16_t *, char32_t *>
+sse_convert_utf16_to_utf32(const char16_t *buf, size_t len,
+                           char32_t *utf32_output) {
+  const char16_t *end = buf + len;
+
+  const __m128i v_f800 = _mm_set1_epi16((int16_t)0xf800);
+  const __m128i v_d800 = _mm_set1_epi16((int16_t)0xd800);
+
+  while (end - buf >= 8) {
+    __m128i in = _mm_loadu_si128((__m128i *)buf);
+
+    if (big_endian) {
+      const __m128i swap =
+          _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14);
+      in = _mm_shuffle_epi8(in, swap);
+    }
+
+    // 1. Check if there are any surrogate word in the input chunk.
+    //    We have also deal with situation when there is a surrogate word
+    //    at the end of a chunk.
+    const __m128i surrogates_bytemask =
+        _mm_cmpeq_epi16(_mm_and_si128(in, v_f800), v_d800);
+
+    // bitmask = 0x0000 if there are no surrogates
+    //         = 0xc000 if the last word is a surrogate
+    const uint16_t surrogates_bitmask =
+        static_cast<uint16_t>(_mm_movemask_epi8(surrogates_bytemask));
+    // It might seem like checking for surrogates_bitmask == 0xc000 could help.
+    // However, it is likely an uncommon occurrence.
+    if (surrogates_bitmask == 0x0000) {
+      // case: no surrogate pair, extend 16-bit code units to 32-bit code units
+      _mm_storeu_si128(reinterpret_cast<__m128i *>(utf32_output),
+                       _mm_cvtepu16_epi32(in));
+      _mm_storeu_si128(reinterpret_cast<__m128i *>(utf32_output + 4),
+                       _mm_cvtepu16_epi32(_mm_srli_si128(in, 8)));
+      utf32_output += 8;
+      buf += 8;
+      // surrogate pair(s) in a register
+    } else {
+      // Let us do a scalar fallback.
+      // It may seem wasteful to use scalar code, but being efficient with SIMD
+      // in the presence of surrogate pairs may require non-trivial tables.
+      size_t forward = 15;
+      size_t k = 0;
+      if (size_t(end - buf) < forward + 1) {
+        forward = size_t(end - buf - 1);
+      }
+      for (; k < forward; k++) {
+        uint16_t word = scalar::utf16::swap_if_needed<big_endian>(buf[k]);
+        if ((word & 0xF800) != 0xD800) {
+          *utf32_output++ = char32_t(word);
+        } else {
+          // must be a surrogate pair
+          uint16_t diff = uint16_t(word - 0xD800);
+          uint16_t next_word =
+              scalar::utf16::swap_if_needed<big_endian>(buf[k + 1]);
+          k++;
+          uint16_t diff2 = uint16_t(next_word - 0xDC00);
+          if ((diff | diff2) > 0x3FF) {
+            return std::make_pair(nullptr, utf32_output);
+          }
+          uint32_t value = (diff << 10) + diff2 + 0x10000;
+          *utf32_output++ = char32_t(value);
+        }
+      }
+      buf += k;
+    }
+  } // while
+  return std::make_pair(buf, utf32_output);
+}
+
+/*
+  Returns a pair: a result struct and utf8_output.
+  If there is an error, the count field of the result is the position of the
+  error. Otherwise, it is the position of the first unprocessed byte in buf
+  (even if finished). A scalar routing should carry on the conversion of the
+  tail if needed.
+*/
+template <endianness big_endian>
+std::pair<result, char32_t *>
+sse_convert_utf16_to_utf32_with_errors(const char16_t *buf, size_t len,
+                                       char32_t *utf32_output) {
+  const char16_t *start = buf;
+  const char16_t *end = buf + len;
+
+  const __m128i v_f800 = _mm_set1_epi16((int16_t)0xf800);
+  const __m128i v_d800 = _mm_set1_epi16((int16_t)0xd800);
+
+  while (end - buf >= 8) {
+    __m128i in = _mm_loadu_si128((__m128i *)buf);
+
+    if (big_endian) {
+      const __m128i swap =
+          _mm_setr_epi8(1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14);
+      in = _mm_shuffle_epi8(in, swap);
+    }
+
+    // 1. Check if there are any surrogate word in the input chunk.
+    //    We have also deal with situation when there is a surrogate word
+    //    at the end of a chunk.
+    const __m128i surrogates_bytemask =
+        _mm_cmpeq_epi16(_mm_and_si128(in, v_f800), v_d800);
+
+    // bitmask = 0x0000 if there are no surrogates
+    //         = 0xc000 if the last word is a surrogate
+    const uint16_t surrogates_bitmask =
+        static_cast<uint16_t>(_mm_movemask_epi8(surrogates_bytemask));
+    // It might seem like checking for surrogates_bitmask == 0xc000 could help.
+    // However, it is likely an uncommon occurrence.
+    if (surrogates_bitmask == 0x0000) {
+      // case: no surrogate pair, extend 16-bit code units to 32-bit code units
+      _mm_storeu_si128(reinterpret_cast<__m128i *>(utf32_output),
+                       _mm_cvtepu16_epi32(in));
+      _mm_storeu_si128(reinterpret_cast<__m128i *>(utf32_output + 4),
+                       _mm_cvtepu16_epi32(_mm_srli_si128(in, 8)));
+      utf32_output += 8;
+      buf += 8;
+      // surrogate pair(s) in a register
+    } else {
+      // Let us do a scalar fallback.
+      // It may seem wasteful to use scalar code, but being efficient with SIMD
+      // in the presence of surrogate pairs may require non-trivial tables.
+      size_t forward = 15;
+      size_t k = 0;
+      if (size_t(end - buf) < forward + 1) {
+        forward = size_t(end - buf - 1);
+      }
+      for (; k < forward; k++) {
+        uint16_t word = scalar::utf16::swap_if_needed<big_endian>(buf[k]);
+        if ((word & 0xF800) != 0xD800) {
+          *utf32_output++ = char32_t(word);
+        } else {
+          // must be a surrogate pair
+          uint16_t diff = uint16_t(word - 0xD800);
+          uint16_t next_word =
+              scalar::utf16::swap_if_needed<big_endian>(buf[k + 1]);
+          k++;
+          uint16_t diff2 = uint16_t(next_word - 0xDC00);
+          if ((diff | diff2) > 0x3FF) {
+            return std::make_pair(
+                result(error_code::SURROGATE, buf - start + k - 1),
+                utf32_output);
+          }
+          uint32_t value = (diff << 10) + diff2 + 0x10000;
+          *utf32_output++ = char32_t(value);
+        }
+      }
+      buf += k;
+    }
+  } // while
+  return std::make_pair(result(error_code::SUCCESS, buf - start), utf32_output);
+}