// dofile: invoked with prepath=/Users/dlemire/CVS/github/simdutf/src, filename=simdutf.cpp

/* begin file src/simdutf.cpp */

#include "simdutf.h"

#if __GNUC__ == 8

#define SIMDUTF_GCC8 1

#endif // defined(__GNUC__) && !defined(__clang__)

__mmask64 bxorleading = _kxor_mask64(b, leading);

if (_kshiftli_mask64(m234, 1) != bxorleading) { return false; }

}

__m512i lead = _mm512_maskz_compress_epi8(leading, leading2byte); // will contain zero for ascii, and the data

lead = _mm512_cvtepu8_epi16(_mm512_castsi512_si256(lead)); // ... zero extended into words

__m512i follow = _mm512_maskz_compress_epi8(continuation_or_ascii, input); // the last bytes of each sequence

if(big_endian) { final = _mm512_shuffle_epi8(final, byteflip); }

if (tail == SIMDUTF_FULL) {

// Next part is UTF-16 specific and can be generalized to UTF-32.

} else {

int nout = int(_mm_popcnt_u64(_pdep_u64(0xFFFFFFFF, leading)));

_mm512_mask_storeu_epi16(out, __mmask32((uint64_t(1) << nout) - 1), final);

static_assert(N<=32, "N must be no larger than 32");

const __m512i movemask = _mm512_setr_epi32(28,29,30,31,0,1,2,3,4,5,6,7,8,9,10,11);

const __m512i rotated = _mm512_permutex2var_epi32(input, movemask, previous);

return _mm512_alignr_epi8(input, rotated, shift);

#else

return _mm512_alignr_epi8(input, rotated, 16-N);

}

template <unsigned idx0, unsigned idx1, unsigned idx2, unsigned idx3>