size_t strlenNEON(const char *p)
{
const char *const top = p;
uint8x16_t c16 = vdupq_n_u8(0);
/* 16 byte alignment */
size_t n = reinterpret_cast<size_t>(p) & 15;
if (n > 0) {
uint8x16_t x = *(const uint8x16_t*)&p[-n];
uint8x16_t a = vceqq_u8(x, c16);
unsigned long mask = GetByteMask(a) << (16 + n);
if (mask) {
return __builtin_clz(mask);
}
p += 16 - n;
}
assert((reinterpret_cast<size_t>(p) & 15) == 0);
for (;;) {
uint8x16_t x = *(const uint8x16_t*)&p[0];
uint8x16_t a = vceqq_u8(x, c16);
if (isFound(a)) {
unsigned int mask = GetByteMask(a);
return p + __builtin_clz(mask) - top;
}
p += 16;
}
}
void test_vceqQu8 (void)
{
uint8x16_t out_uint8x16_t;
uint8x16_t arg0_uint8x16_t;
uint8x16_t arg1_uint8x16_t;
out_uint8x16_t = vceqq_u8 (arg0_uint8x16_t, arg1_uint8x16_t);
}
void *memchrNEON(const void *ptr, int c, size_t len)
{
const char *p = reinterpret_cast<const char*>(ptr);
if (len >= 16) {
uint8x16_t c16 = vdupq_n_u8(static_cast<char>(c));
/* 16 byte alignment */
size_t n = reinterpret_cast<size_t>(p) & 15;
if (n > 0) {
uint8x16_t x = *(const uint8x16_t*)&p[-n];
uint8x16_t a = vceqq_u8(x, c16);
unsigned long mask = GetByteMask(a) << (16 + n);
if (mask) {
return (void*)(p + __builtin_clz(mask));
}
n = 16 - n;
len -= n;
p += n;
}
while (len >= 32) {
uint8x16_t x = *(const uint8x16_t*)&p[0];
uint8x16_t y = *(const uint8x16_t*)&p[16];
uint8x16_t a = vceqq_u8(x, c16);
uint8x16_t b = vceqq_u8(y, c16);
if (isFound2(a, b)) {
unsigned int mask = GetByteMask2(a,b);
return (void*)(p + __builtin_clz(mask));
}
len -= 32;
p += 32;
}
}
while (len > 0) {
if (*p == c) return (void*)p;
p++;
len--;
}
return 0;
}
template <bool align> void SquaredDifferenceSumMasked(
const uint8_t *a, size_t aStride, const uint8_t *b, size_t bStride,
const uint8_t *mask, size_t maskStride, uint8_t index, size_t width, size_t height, uint64_t * sum)
{
assert(width < 0x10000);
if (align)
{
assert(Aligned(a) && Aligned(aStride) && Aligned(b) && Aligned(bStride));
assert(Aligned(mask) && Aligned(maskStride));
}
size_t alignedWidth = Simd::AlignLo(width, A);
uint8x16_t tailMask = ShiftLeft(K8_FF, A - width + alignedWidth);
uint8x16_t _index = vdupq_n_u8(index);
uint64x2_t _sum = K64_0000000000000000;
for (size_t row = 0; row < height; ++row)
{
uint32x4_t rowSum = K32_00000000;
for (size_t col = 0; col < alignedWidth; col += A)
{
uint8x16_t _mask = vceqq_u8(Load<align>(mask + col), _index);
uint8x16_t _a = Load<align>(a + col);
uint8x16_t _b = Load<align>(b + col);
rowSum = vaddq_u32(rowSum, SquaredDifferenceSumMasked(_a, _b, _mask));
}
if (width - alignedWidth)
{
uint8x16_t _mask = vandq_u8(tailMask, vceqq_u8(Load<align>(mask + width - A), _index));
uint8x16_t _a = Load<align>(a + width - A);
uint8x16_t _b = Load<align>(b + width - A);
rowSum = vaddq_u32(rowSum, SquaredDifferenceSumMasked(_a, _b, _mask));
}
_sum = vaddq_u64(_sum, vpaddlq_u32(rowSum));
a += aStride;
b += bStride;
mask += maskStride;
}
*sum = ExtractSum64u(_sum);
}
inline uint8x16_t vceqq(const uint8x16_t & v0, const uint8x16_t & v1) { return vceqq_u8 (v0, v1); }
