static uint32_t avx512maxbitas32int(const __m512i accumulator) {
uint32_t ans1 = maxbitas32int(_mm512_castsi512_si256(accumulator));
uint32_t ans2 = maxbitas32int(_mm512_extracti64x4_epi64(accumulator, 1));
printf("ans1 = %u ans2 = % u\n", ans1, ans2);
uint32_t ans = ans1 > ans2 ? ans1 : ans2;
return bits(ans);
}
SIMDCOMP_PURE uint32_t maxbits(const uint32_t * begin) {
const __m128i* pin = (const __m128i*)(begin);
__m128i accumulator = _mm_loadu_si128(pin);
uint32_t k = 1;
for(; 4*k < SIMDBlockSize; ++k) {
__m128i newvec = _mm_loadu_si128(pin+k);
accumulator = _mm_or_si128(accumulator,newvec);
}
return maxbitas32int(accumulator);
}
/* maxbit over 128 integers (SIMDBlockSize) with provided initial value */
uint32_t simdmaxbitsd1(uint32_t initvalue, const uint32_t * in) {
__m128i initoffset = _mm_set1_epi32 (initvalue);
const __m128i* pin = (const __m128i*)(in);
__m128i newvec = _mm_loadu_si128(pin);
__m128i accumulator = Delta(newvec , initoffset);
__m128i oldvec = newvec;
uint32_t k = 1;
for(; 4*k < SIMDBlockSize; ++k) {
newvec = _mm_loadu_si128(pin+k);
accumulator = _mm_or_si128(accumulator,Delta(newvec , oldvec));
oldvec = newvec;
}
initoffset = oldvec;
return maxbitas32int(accumulator);
}
