size_t qfind_first_byte_of_sse42(const StringPiece& haystack,
const StringPiece& needles) {
if (UNLIKELY(needles.empty() || haystack.empty())) {
return StringPiece::npos;
} else if (needles.size() <= 16) {
// we can save some unnecessary load instructions by optimizing for
// the common case of needles.size() <= 16
return qfind_first_byte_of_needles16(haystack, needles);
}
if (haystack.size() < 16 &&
PAGE_FOR(haystack.end() - 1) != PAGE_FOR(haystack.data() + 16)) {
// We can't safely SSE-load haystack. Use a different approach.
if (haystack.size() <= 2) {
return qfind_first_of(haystack, needles, asciiCaseSensitive);
}
return qfind_first_byte_of_byteset(haystack, needles);
}
auto ret = scanHaystackBlock<false>(haystack, needles, 0);
if (ret != StringPiece::npos) {
return ret;
}
size_t i = nextAlignedIndex(haystack.data());
for (; i < haystack.size(); i += 16) {
auto ret = scanHaystackBlock<true>(haystack, needles, i);
if (ret != StringPiece::npos) {
return ret;
}
}
return StringPiece::npos;
}
size_t qfind_first_byte_of_sse42(const StringPiece& haystack,
const StringPiece& needles) {
if (UNLIKELY(needles.empty() || haystack.empty())) {
return StringPiece::npos;
} else if (needles.size() <= 16) {
// we can save some unnecessary load instructions by optimizing for
// the common case of needles.size() <= 16
return qfind_first_byte_of_needles16(haystack, needles);
}
size_t index = haystack.size();
for (size_t i = 0; i < haystack.size(); i += 16) {
size_t b = 16;
auto arr1 = __builtin_ia32_loaddqu(haystack.data() + i);
for (size_t j = 0; j < needles.size(); j += 16) {
auto arr2 = __builtin_ia32_loaddqu(needles.data() + j);
auto index = __builtin_ia32_pcmpestri128(arr2, needles.size() - j,
arr1, haystack.size() - i, 0);
b = std::min<size_t>(index, b);
}
if (b < 16) {
return i + b;
}
};
return StringPiece::npos;
}
