1berry/qtbase
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

QUtf8: store even mojibake in simdDecodeAscii()

Browse Source 
Like we do in simdEncodeAscii(), because there's a good chance it starts
with a sequence of ASCII-only content and thus saves us the trouble of
processing up to 15 characters in scalar code.

Change-Id: I37f4aab92036daf1cec7fffdc2313dfa19f6d2fd
Reviewed-by: Allan Sandfeld Jensen <allan.jensen@qt.io>
This commit is contained in:
Thiago Macieira 2024-10-03 22:01:19 -07:00
parent ad31a90319
commit 5ca5b36b6c
1 changed files with 20 additions and 32 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

52
src/corelib/text/qstringconverter.cpp
View File

@ -143,50 +143,38 @@ static inline bool simdDecodeAscii(char16_t *&dst, const uchar *&nextAscii, cons
#ifdef __AVX2__ #ifdef __AVX2__
// load and zero extend to an YMM register // load and zero extend to an YMM register
const __m256i extended = _mm256_cvtepu8_epi16(data); const __m256i extended = _mm256_cvtepu8_epi16(data);
if (!n) {
// store // store everything, even mojibake
_mm256_storeu_si256((__m256i*)dst, extended); _mm256_storeu_si256((__m256i*)dst, extended);
continue;
}
#else #else
if (!n) { // store everything, even mojibake
// unpack _mm_storeu_si128((__m128i*)dst, _mm_unpacklo_epi8(data, _mm_setzero_si128()));
_mm_storeu_si128((__m128i*)dst, _mm_unpacklo_epi8(data, _mm_setzero_si128())); _mm_storeu_si128(1+(__m128i*)dst, _mm_unpackhi_epi8(data, _mm_setzero_si128()));
_mm_storeu_si128(1+(__m128i*)dst, _mm_unpackhi_epi8(data, _mm_setzero_si128()));
continue;
}
#endif #endif
// copy the front part that is still ASCII
while (!(n & 1)) {
*dst++ = *src++;
n >>= 1;
}
// find the next probable ASCII character // find the next probable ASCII character
// we don't want to load 16 bytes again in this loop if we know there are non-ASCII // we don't want to load 16 bytes again in this loop if we know there are non-ASCII
// characters still coming // characters still coming
n = qBitScanReverse(n); if (n) {
nextAscii = src + n + 1; uint c = qCountTrailingZeroBits(n);
return false; n = qBitScanReverse(n);
nextAscii = src + n + 1;
src += c;
dst += c;
}
return src == end;
} }
if (end - src >= 8) { if (end - src >= 8) {
__m128i data = _mm_loadl_epi64(reinterpret_cast<const __m128i *>(src)); __m128i data = _mm_loadl_epi64(reinterpret_cast<const __m128i *>(src));
uint n = _mm_movemask_epi8(data) & 0xff; uint n = _mm_movemask_epi8(data) & 0xff;
if (!n) { // store everything, even mojibake
// unpack and store _mm_storeu_si128(reinterpret_cast<__m128i *>(dst), _mm_unpacklo_epi8(data, _mm_setzero_si128()));
_mm_storeu_si128(reinterpret_cast<__m128i *>(dst), _mm_unpacklo_epi8(data, _mm_setzero_si128())); if (n) {
} else { uint c = qCountTrailingZeroBits(n);
while (!(n & 1)) {
*dst++ = *src++;
n >>= 1;
}
n = qBitScanReverse(n); n = qBitScanReverse(n);
nextAscii = src + n + 1; nextAscii = src + n + 1;
return false; src += c;
dst += c;
} }
} }