Skip to content

Commit

Permalink
introduce swar routines for capitalization
Browse files Browse the repository at this point in the history
  • Loading branch information
anonrig committed Sep 19, 2024
1 parent 695b0b4 commit c934972
Showing 1 changed file with 48 additions and 6 deletions.
54 changes: 48 additions & 6 deletions src/workerd/api/util.c++
Original file line number Diff line number Diff line change
Expand Up @@ -40,19 +40,61 @@ kj::ArrayPtr<const char> split(kj::ArrayPtr<const char>& text, char c) {
return result;
}

constexpr uint64_t broadcast(uint8_t v) noexcept {
return 0x101010101010101ull * v;
}

// SWAR routine designed to convert ASCII uppercase letters to lowercase.
// Let's process 8 bytes (64 bits) at a time using a single 64-bit int,
// treating as 8 parallel 8-bit values.
// PS: This will enable the use of auto-vectorization.
constexpr void toLowerAscii(char* input, size_t length) noexcept {
constexpr const uint64_t broadcast_80 = broadcast(0x80);
constexpr const uint64_t broadcast_Ap = broadcast(128 - 'A');
constexpr const uint64_t broadcast_Zp = broadcast(128 - 'Z' - 1);
size_t i = 0;
for (; i + 7 < length; i += 8) {
uint64_t word{};
memcpy(&word, input + i, sizeof(word));
word ^= (((word + broadcast_Ap) ^ (word + broadcast_Zp)) & broadcast_80) >> 2;
memcpy(input + i, &word, sizeof(word));
}
if (i < length) {
uint64_t word{};
memcpy(&word, input + i, length - i);
word ^= (((word + broadcast_Ap) ^ (word + broadcast_Zp)) & broadcast_80) >> 2;
memcpy(input + i, &word, length - i);
}
}

constexpr void toUpperAscii(char* input, size_t length) noexcept {
constexpr const uint64_t broadcast_80 = broadcast(0x80);
constexpr const uint64_t broadcast_ap = broadcast(128 - 'a');
constexpr const uint64_t broadcast_zp = broadcast(128 - 'z' - 1);
size_t i = 0;
for (; i + 7 < length; i += 8) {
uint64_t word{};
memcpy(&word, input + i, sizeof(word));
word ^= (((word + broadcast_ap) ^ (word + broadcast_zp)) & broadcast_80) >> 2;
memcpy(input + i, &word, sizeof(word));
}
if (i < length) {
uint64_t word{};
memcpy(&word, input + i, length - i);
word ^= (((word + broadcast_ap) ^ (word + broadcast_zp)) & broadcast_80) >> 2;
memcpy(input + i, &word, length - i);
}
}

} // namespace

kj::String toLower(kj::String&& str) {
for (char& c: str) {
if ('A' <= c && c <= 'Z') c += 'a' - 'A';
}
toLowerAscii(str.begin(), str.size());
return kj::mv(str);
}

kj::String toUpper(kj::String&& str) {
for (char& c: str) {
if ('a' <= c && c <= 'z') c -= 'a' - 'A';
}
toUpperAscii(str.begin(), str.size());
return kj::mv(str);
}

Expand Down

0 comments on commit c934972

Please sign in to comment.