static void
TEST (void)
{
union
{
__m128i x;
unsigned int i[4];
} val[4];
int i, j, l;
int res[32];
val[0].i[0] = 0x11111111;
val[0].i[1] = 0x00000000;
val[0].i[2] = 0x00000000;
val[0].i[3] = 0x11111111;
val[1].i[0] = 0x00000000;
val[1].i[1] = 0x11111111;
val[1].i[2] = 0x11111111;
val[1].i[3] = 0x00000000;
val[2].i[0] = 0;
val[2].i[1] = 0;
val[2].i[2] = 0;
val[2].i[3] = 0;
val[3].i[0] = 0xffffffff;
val[3].i[1] = 0xffffffff;
val[3].i[2] = 0xffffffff;
val[3].i[3] = 0xffffffff;
l = 0;
for(i = 0; i < 4; i++)
for(j = 0; j < 4; j++)
{
res[l++] = _mm_testz_si128 (val[j].x, val[i].x);
res[l++] = _mm_testc_si128 (val[j].x, val[i].x);
}
l = 0;
for(i = 0; i < 4; i++)
for(j = 0; j < 4; j++)
{
if (res[l++] != make_ptestz (val[j].x, val[i].x))
abort ();
if (res[l++] != make_ptestc (val[j].x, val[i].x))
abort ();
}
if (res[2] != _mm_testz_si128 (val[1].x, val[0].x))
abort ();
if (res[3] != _mm_testc_si128 (val[1].x, val[0].x))
abort ();
}
SIMD_INLINE bool RowHasIndex(const uint8_t * mask, size_t alignedSize, size_t fullSize, __m128i index)
{
for (size_t col = 0; col < alignedSize; col += A)
{
if(!_mm_testz_si128(_mm_cmpeq_epi8(_mm_loadu_si128((__m128i*)(mask + col)), index), K_INV_ZERO))
return true;
}
if(alignedSize != fullSize)
{
if(!_mm_testz_si128(_mm_cmpeq_epi8(_mm_loadu_si128((__m128i*)(mask + fullSize - A)), index), K_INV_ZERO))
return true;
}
return false;
}
// これはコードが見難くなるけど仕方ない。
bool andIsNot0(const Bitboard& bb) const {
#ifdef HAVE_SSE4
return !(_mm_testz_si128(this->m_, bb.m_));
#else
return (*this & bb).isNot0();
#endif
}
SIMD_INLINE bool ColsHasIndex(const uint8_t * mask, size_t stride, size_t size, __m128i index, uint8_t * cols)
{
__m128i _cols = _mm_setzero_si128();
for (size_t row = 0; row < size; ++row)
{
_cols = _mm_or_si128(_cols, _mm_cmpeq_epi8(_mm_loadu_si128((__m128i*)mask), index));
mask += stride;
}
_mm_storeu_si128((__m128i*)cols, _cols);
return !_mm_testz_si128(_cols, K_INV_ZERO);
}
/*
int32_t search_range(Rect rect, int32_t x[], int32_t y[],
int32_t w[], int32_t n) {
int32_t ret = 0;
for (int i = 0; i < n; i++) {
if (rect.lx <= x[i] && x[i] <= rect.rx &&
rect.ly <= y[i] && y[i] <= rect.ry) {
ret += w[i];
}
}
return ret;
}
*/
int32_t search_range(Rect rect, int32_t x[], int32_t y[], int32_t w[], int32_t n) {
__m128i ret = _mm_set_epi32(0, 0, 0, 0);
rect.lx--, rect.ly--;
rect.rx++, rect.ry++;
__m128i lx = _mm_broadcastd_epi32(*((__m128i *) &rect.lx));
__m128i ly = _mm_broadcastd_epi32(*((__m128i *) &rect.ly));
__m128i rx = _mm_broadcastd_epi32(*((__m128i *) &rect.rx));
__m128i ry = _mm_broadcastd_epi32(*((__m128i *) &rect.ry));
__m128i zo = _mm_set_epi32(0, 0, 0, 0);
__m128i ic = _mm_set_epi32(3, 2, 1, 0);
for (int i = 0; i+4 <= n; i += 4) {
__m128i sx = _mm_load_si128((__m128i *) (x+i));
__m128i sy = _mm_load_si128((__m128i *) (y+i));
__m128i c1 = _mm_and_si128(_mm_cmplt_epi32(lx, sx), _mm_cmplt_epi32(sx, rx));
__m128i c2 = _mm_and_si128(_mm_cmplt_epi32(ly, sy), _mm_cmplt_epi32(sy, ry));
if (_mm_testz_si128(c1, c2) == 0) {
__m128i cc = _mm_and_si128(c1, c2);
__m128i vi = _mm_add_epi32(ic, _mm_set_epi32(i, i, i, i));
__m128i rs = _mm_mask_i32gather_epi32(zo, w+i, ic, cc, 4);
ret = _mm_add_epi32(ret, rs);
}
}
int32_t sum = 0;
for (int i = (n>>2)<<2; i < n; i++) {
if (rect.lx <= x[i] && x[i] <= rect.rx &&
rect.ly <= y[i] && y[i] <= rect.ry) {
sum += w[i];
}
}
static int32_t tmp[4] __attribute__ ((aligned (16)));
_mm_store_si128((__m128i*) &tmp[0], ret);
sum += tmp[0] + tmp[1] + tmp[2] + tmp[3];
return sum;
}
KFR_SINTRIN bool bittestany(const i64sse& x) { return !_mm_testz_si128(*x, *x); }
KFR_SINTRIN bool bittestany(const f64sse& x) { return !_mm_testz_si128(*bitcast<u8>(x), *bitcast<u8>(x)); }
}bool validate_utf8_sse(const char *src, size_t len) {
const char *end = src + len;
while (src + 16 < end) {
__m128i chunk = _mm_loadu_si128((const __m128i *)(src));
int asciiMask = _mm_movemask_epi8(chunk);
if (!asciiMask) {
src += 16;
continue;
}
__m128i chunk_signed = _mm_add_epi8(chunk, _mm_set1_epi8(0x80));
__m128i cond2 =
_mm_cmplt_epi8(_mm_set1_epi8(0xc2 - 1 - 0x80), chunk_signed);
__m128i state = _mm_set1_epi8((char)(0x0 | 0x80));
state = _mm_blendv_epi8(state, _mm_set1_epi8((char)(0x2 | 0xc0)), cond2);
__m128i cond3 =
_mm_cmplt_epi8(_mm_set1_epi8(0xe0 - 1 - 0x80), chunk_signed);
state = _mm_blendv_epi8(state, _mm_set1_epi8((char)(0x3 | 0xe0)), cond3);
__m128i mask3 = _mm_slli_si128(cond3, 1);
__m128i cond4 =
_mm_cmplt_epi8(_mm_set1_epi8(0xf0 - 1 - 0x80), chunk_signed);
// Fall back to the scalar processing
if (_mm_movemask_epi8(cond4)) {
break;
}
__m128i count = _mm_and_si128(state, _mm_set1_epi8(0x7));
__m128i count_sub1 = _mm_subs_epu8(count, _mm_set1_epi8(0x1));
__m128i counts = _mm_add_epi8(count, _mm_slli_si128(count_sub1, 1));
__m128i shifts = count_sub1;
shifts = _mm_add_epi8(shifts, _mm_slli_si128(shifts, 1));
counts = _mm_add_epi8(
counts, _mm_slli_si128(_mm_subs_epu8(counts, _mm_set1_epi8(0x2)), 2));
shifts = _mm_add_epi8(shifts, _mm_slli_si128(shifts, 2));
if (asciiMask ^ _mm_movemask_epi8(_mm_cmpgt_epi8(counts, _mm_set1_epi8(0))))
return false; // error
shifts = _mm_add_epi8(shifts, _mm_slli_si128(shifts, 4));
if (_mm_movemask_epi8(_mm_cmpgt_epi8(
_mm_sub_epi8(_mm_slli_si128(counts, 1), counts), _mm_set1_epi8(1))))
return false; // error
shifts = _mm_add_epi8(shifts, _mm_slli_si128(shifts, 8));
__m128i mask = _mm_and_si128(state, _mm_set1_epi8(0xf8));
shifts =
_mm_and_si128(shifts, _mm_cmplt_epi8(counts, _mm_set1_epi8(2))); // <=1
chunk =
_mm_andnot_si128(mask, chunk); // from now on, we only have usefull bits
shifts = _mm_blendv_epi8(shifts, _mm_srli_si128(shifts, 1),
_mm_srli_si128(_mm_slli_epi16(shifts, 7), 1));
__m128i chunk_right = _mm_slli_si128(chunk, 1);
__m128i chunk_low = _mm_blendv_epi8(
chunk,
_mm_or_si128(chunk, _mm_and_si128(_mm_slli_epi16(chunk_right, 6),
_mm_set1_epi8(0xc0))),
_mm_cmpeq_epi8(counts, _mm_set1_epi8(1)));
__m128i chunk_high =
_mm_and_si128(chunk, _mm_cmpeq_epi8(counts, _mm_set1_epi8(2)));
shifts = _mm_blendv_epi8(shifts, _mm_srli_si128(shifts, 2),
_mm_srli_si128(_mm_slli_epi16(shifts, 6), 2));
chunk_high = _mm_srli_epi32(chunk_high, 2);
shifts = _mm_blendv_epi8(shifts, _mm_srli_si128(shifts, 4),
_mm_srli_si128(_mm_slli_epi16(shifts, 5), 4));
chunk_high = _mm_or_si128(
chunk_high, _mm_and_si128(_mm_and_si128(_mm_slli_epi32(chunk_right, 4),
_mm_set1_epi8(0xf0)),
mask3));
int c = _mm_extract_epi16(counts, 7);
int source_advance = !(c & 0x0200) ? 16 : !(c & 0x02) ? 15 : 14;
__m128i high_bits = _mm_and_si128(chunk_high, _mm_set1_epi8(0xf8));
if (!_mm_testz_si128(
mask3,
_mm_or_si128(_mm_cmpeq_epi8(high_bits, _mm_set1_epi8(0x00)),
_mm_cmpeq_epi8(high_bits, _mm_set1_epi8(0xd8)))))
return false;
shifts = _mm_blendv_epi8(shifts, _mm_srli_si128(shifts, 8),
_mm_srli_si128(_mm_slli_epi16(shifts, 4), 8));
chunk_high = _mm_slli_si128(chunk_high, 1);
__m128i shuf =
_mm_add_epi8(shifts, _mm_set_epi8(15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5,
4, 3, 2, 1, 0));
chunk_low = _mm_shuffle_epi8(chunk_low, shuf);
chunk_high = _mm_shuffle_epi8(chunk_high, shuf);
__m128i utf16_low = _mm_unpacklo_epi8(chunk_low, chunk_high);
__m128i utf16_high = _mm_unpackhi_epi8(chunk_low, chunk_high);
if (_mm_cmpestrc(_mm_cvtsi64_si128(0xfdeffdd0fffffffe), 4, utf16_high, 8,
_SIDD_UWORD_OPS | _SIDD_CMP_RANGES) |
_mm_cmpestrc(_mm_cvtsi64_si128(0xfdeffdd0fffffffe), 4, utf16_low, 8,
_SIDD_UWORD_OPS | _SIDD_CMP_RANGES)) {
return false;
}
src += source_advance;
}
return validate_utf8(src, end - src);
}
int test_mm_testz_si128(__m128i x, __m128i y) {
// CHECK-LABEL: test_mm_testz_si128
// CHECK: call i32 @llvm.x86.sse41.ptestz
// CHECK-ASM: ptest %xmm{{.*}}, %xmm{{.*}}
return _mm_testz_si128(x, y);
}
int test_mm_testz_si128(__m128i x, __m128i y) {
// CHECK-LABEL: test_mm_testz_si128
// CHECK: call i32 @llvm.x86.sse41.ptestz(<2 x i64> %{{.*}}, <2 x i64> %{{.*}})
return _mm_testz_si128(x, y);
}
