use std::arch::x86_64::*;
#[inline]
#[target_feature(enable = "avx2")]
fn load_u8x16_i16(src: &[u8]) -> __m256i {
unsafe { _mm256_cvtepu8_epi16(_mm_loadu_si128(src.as_ptr().cast())) }
}
#[inline]
#[target_feature(enable = "avx2")]
fn load_i16x8(src: &[i16]) -> __m128i {
unsafe { _mm_loadu_si128(src.as_ptr().cast()) }
}
#[inline]
#[target_feature(enable = "avx2")]
fn store_i16x16(dst: &mut [i16], v: __m256i) {
unsafe { _mm256_storeu_si256(dst.as_mut_ptr().cast(), v) };
}
#[inline]
#[target_feature(enable = "avx2")]
fn store_i16x8(dst: &mut [i16], v: __m256i) {
unsafe {
let lo = _mm256_castsi256_si128(v);
let hi = _mm256_extracti128_si256::<1>(v);
_mm_storeu_si128(dst.as_mut_ptr().cast(), _mm_packs_epi32(lo, hi));
}
}
#[inline]
#[target_feature(enable = "avx2")]
fn store_u8x16_from_i16(dst: &mut [u8], v: __m256i) {
let p8 = _mm256_packus_epi16(v, v);
let lo = _mm256_castsi256_si128(p8);
let hi = _mm256_extracti128_si256::<1>(p8);
unsafe {
_mm_storeu_si128(dst.as_mut_ptr().cast(), _mm_unpacklo_epi64(lo, hi));
}
}
#[inline]
#[target_feature(enable = "avx2")]
fn store_u8x16_round4_from_i16(dst: &mut [u8], v: __m256i) {
let v = _mm256_srli_epi16::<4>(_mm256_add_epi16(v, _mm256_set1_epi16(8)));
store_u8x16_from_i16(dst, v);
}
#[inline]
#[target_feature(enable = "avx2")]
fn store_u8x8_round8_from_i32(dst: &mut [u8], v: __m256i) {
let v = _mm256_srai_epi32::<8>(_mm256_add_epi32(v, _mm256_set1_epi32(128)));
let lo = _mm256_castsi256_si128(v);
let hi = _mm256_extracti128_si256::<1>(v);
let p16 = _mm_packus_epi32(lo, hi);
let p8 = _mm_packus_epi16(p16, p16);
unsafe {
_mm_storel_epi64(dst.as_mut_ptr().cast(), p8);
}
}
#[inline]
#[target_feature(enable = "avx2")]
fn store_u8x16_round8_from_i32(dst: &mut [u8], lo: __m256i, hi: __m256i) {
let lo = _mm256_srai_epi32::<8>(_mm256_add_epi32(lo, _mm256_set1_epi32(128)));
let hi = _mm256_srai_epi32::<8>(_mm256_add_epi32(hi, _mm256_set1_epi32(128)));
let lo0 = _mm256_castsi256_si128(lo);
let lo1 = _mm256_extracti128_si256::<1>(lo);
let hi0 = _mm256_castsi256_si128(hi);
let hi1 = _mm256_extracti128_si256::<1>(hi);
let p16_lo = _mm_packus_epi32(lo0, lo1);
let p16_hi = _mm_packus_epi32(hi0, hi1);
let p8 = _mm_packus_epi16(p16_lo, p16_hi);
unsafe { _mm_storeu_si128(dst.as_mut_ptr().cast(), p8) };
}
#[inline]
#[target_feature(enable = "avx2")]
fn store_i16x8_round4_from_i32(dst: &mut [i16], v: __m256i) {
let v = _mm256_srai_epi32::<4>(_mm256_add_epi32(v, _mm256_set1_epi32(8)));
store_i16x8(dst, v);
}
#[inline]
#[target_feature(enable = "avx2")]
fn store_i16x16_round4_from_i32(dst: &mut [i16], lo: __m256i, hi: __m256i) {
let lo = _mm256_srai_epi32::<4>(_mm256_add_epi32(lo, _mm256_set1_epi32(8)));
let hi = _mm256_srai_epi32::<4>(_mm256_add_epi32(hi, _mm256_set1_epi32(8)));
let lo0 = _mm256_castsi256_si128(lo);
let lo1 = _mm256_extracti128_si256::<1>(lo);
let hi0 = _mm256_castsi256_si128(hi);
let hi1 = _mm256_extracti128_si256::<1>(hi);
unsafe {
_mm256_storeu_si256(
dst.as_mut_ptr().cast(),
_mm256_setr_m128i(_mm_packs_epi32(lo0, lo1), _mm_packs_epi32(hi0, hi1)),
);
}
}
#[inline]
#[target_feature(enable = "avx2")]
fn bilin_coeff_u8(mxy: i32) -> __m256i {
_mm256_set1_epi16(((mxy as i16) << 8) | ((16 - mxy) as i16))
}
#[inline]
#[target_feature(enable = "avx2")]
fn bilin_coeff_i16(mxy: i32) -> __m256i {
_mm256_set1_epi32(((mxy as i16 as u16 as i32) << 16) | ((16 - mxy) as i16 as u16 as i32))
}
#[inline]
#[target_feature(enable = "avx2")]
fn bilin_u8x16_i16(src: &[u8], base: usize, stride: usize, mxy: i32) -> __m256i {
unsafe {
let a = _mm_loadu_si128(src.get_unchecked(base..).as_ptr().cast());
let b = _mm_loadu_si128(src.get_unchecked(base + stride..).as_ptr().cast());
let c = _mm256_castsi256_si128(bilin_coeff_u8(mxy));
let lo = _mm_maddubs_epi16(_mm_unpacklo_epi8(a, b), c);
let hi = _mm_maddubs_epi16(_mm_unpackhi_epi8(a, b), c);
_mm256_inserti128_si256::<1>(_mm256_castsi128_si256(lo), hi)
}
}
#[inline]
#[target_feature(enable = "avx2")]
fn bilin_i16x8_i32(a16: __m128i, b16: __m128i, mxy: i32) -> __m256i {
let lo = _mm_unpacklo_epi16(a16, b16);
let hi = _mm_unpackhi_epi16(a16, b16);
let pairs = _mm256_inserti128_si256::<1>(_mm256_castsi128_si256(lo), hi);
_mm256_madd_epi16(pairs, bilin_coeff_i16(mxy))
}
#[inline]
#[target_feature(enable = "avx2")]
fn bilin_i16x16_i32(
a0: __m128i,
b0: __m128i,
a1: __m128i,
b1: __m128i,
mxy: i32,
) -> (__m256i, __m256i) {
(bilin_i16x8_i32(a0, b0, mxy), bilin_i16x8_i32(a1, b1, mxy))
}
#[inline(always)]
fn bilin_scalar(a: i32, b: i32, mxy: i32) -> i32 {
16 * a + mxy * (b - a)
}
#[target_feature(enable = "avx2")]
#[allow(clippy::too_many_arguments)]
pub(crate) fn put_bilin_8bpc_avx2(
dst: &mut [u8],
dst_stride: usize,
src: &[u8],
src_stride: usize,
w: usize,
h: usize,
mx: i32,
my: i32,
mid_scratch: &mut [i16],
) {
if mx != 0 && my != 0 {
let mid_stride = w.next_multiple_of(16).max(64);
let mid = &mut mid_scratch[..mid_stride * (h + 1)];
for (y, mid_row) in mid.chunks_mut(mid_stride).take(h + 1).enumerate() {
let src_row = unsafe { src.get_unchecked(y * src_stride..) };
let (mid_chunks, mid_rem) = mid_row[..w].as_chunks_mut::<16>();
for (chunk_idx, mid_chunk) in mid_chunks.iter_mut().enumerate() {
let x = chunk_idx * 16;
store_i16x16(mid_chunk, bilin_u8x16_i16(src_row, x, 1, mx));
}
let processed = mid_chunks.len() * 16;
for (x, mid_px) in (processed..w).zip(mid_rem.iter_mut()) {
let si = x;
*mid_px = bilin_scalar(src_row[si] as i32, src_row[si + 1] as i32, mx) as i16;
}
}
for (y, dst_row) in dst.chunks_mut(dst_stride).take(h).enumerate() {
let mid_row = unsafe { mid.get_unchecked(y * mid_stride..) };
let mid_next_row = unsafe { mid.get_unchecked((y + 1) * mid_stride..) };
let (dst_chunks16, dst_rem16) = dst_row[..w].as_chunks_mut::<16>();
for (chunk_idx, dst_chunk) in dst_chunks16.iter_mut().enumerate() {
let x = chunk_idx * 16;
let a0 = load_i16x8(unsafe { mid_row.get_unchecked(x..) });
let b0 = load_i16x8(unsafe { mid_next_row.get_unchecked(x..) });
let a1 = load_i16x8(unsafe { mid_row.get_unchecked(x + 8..) });
let b1 = load_i16x8(unsafe { mid_next_row.get_unchecked(x + 8..) });
let (lo, hi) = bilin_i16x16_i32(a0, b0, a1, b1, my);
store_u8x16_round8_from_i32(dst_chunk, lo, hi);
}
let x16_done = dst_chunks16.len() * 16;
let (dst_chunks8, dst_rem) = dst_rem16.as_chunks_mut::<8>();
for (chunk_idx, dst_chunk) in dst_chunks8.iter_mut().enumerate() {
let x = x16_done + chunk_idx * 8;
let a = load_i16x8(unsafe { mid_row.get_unchecked(x..) });
let b = load_i16x8(unsafe { mid_next_row.get_unchecked(x..) });
store_u8x8_round8_from_i32(dst_chunk, bilin_i16x8_i32(a, b, my));
}
let processed = x16_done + dst_chunks8.len() * 8;
for (x, dst_px) in (processed..w).zip(dst_rem.iter_mut()) {
*dst_px = ((bilin_scalar(mid_row[x] as i32, mid_next_row[x] as i32, my) + 128) >> 8)
.clamp(0, 255) as u8;
}
}
} else if mx != 0 {
for (y, dst_row) in dst.chunks_mut(dst_stride).take(h).enumerate() {
let src_row = unsafe { src.get_unchecked(y * src_stride..) };
let (dst_chunks, dst_rem) = dst_row[..w].as_chunks_mut::<16>();
for (chunk_idx, dst_chunk) in dst_chunks.iter_mut().enumerate() {
let x = chunk_idx * 16;
store_u8x16_round4_from_i16(dst_chunk, bilin_u8x16_i16(src_row, x, 1, mx));
}
let processed = dst_chunks.len() * 16;
for (x, dst_px) in (processed..w).zip(dst_rem.iter_mut()) {
let si = x;
*dst_px =
((bilin_scalar(src_row[si] as i32, src_row[si + 1] as i32, mx) + 8) >> 4) as u8;
}
}
} else if my != 0 {
for (y, dst_row) in dst.chunks_mut(dst_stride).take(h).enumerate() {
let src_row = unsafe { src.get_unchecked(y * src_stride..) };
let (dst_chunks, dst_rem) = dst_row[..w].as_chunks_mut::<16>();
for (chunk_idx, dst_chunk) in dst_chunks.iter_mut().enumerate() {
let x = chunk_idx * 16;
store_u8x16_round4_from_i16(dst_chunk, bilin_u8x16_i16(src_row, x, src_stride, my));
}
let processed = dst_chunks.len() * 16;
for (x, dst_px) in (processed..w).zip(dst_rem.iter_mut()) {
let si = x;
*dst_px = ((bilin_scalar(src_row[si] as i32, src_row[si + src_stride] as i32, my)
+ 8)
>> 4) as u8;
}
}
} else {
for (src_row, dst_row) in src
.chunks(src_stride)
.zip(dst.chunks_mut(dst_stride))
.take(h)
{
dst_row[..w].copy_from_slice(&src_row[..w]);
}
}
}
#[target_feature(enable = "avx2")]
#[allow(clippy::too_many_arguments)]
pub(crate) fn prep_bilin_8bpc_avx2(
tmp: &mut [i16],
tmp_stride: usize,
src: &[u8],
src_stride: usize,
w: usize,
h: usize,
mx: i32,
my: i32,
mid_scratch: &mut [i16],
) {
if mx != 0 && my != 0 {
let mid_stride = w.next_multiple_of(16).max(64);
let mid = &mut mid_scratch[..mid_stride * (h + 1)];
for (y, mid_row) in mid.chunks_mut(mid_stride).take(h + 1).enumerate() {
let src_row = unsafe { src.get_unchecked(y * src_stride..) };
let (mid_chunks, mid_rem) = mid_row[..w].as_chunks_mut::<16>();
for (chunk_idx, mid_chunk) in mid_chunks.iter_mut().enumerate() {
let x = chunk_idx * 16;
store_i16x16(mid_chunk, bilin_u8x16_i16(src_row, x, 1, mx));
}
let processed = mid_chunks.len() * 16;
for (x, mid_px) in (processed..w).zip(mid_rem.iter_mut()) {
let si = x;
*mid_px = bilin_scalar(src_row[si] as i32, src_row[si + 1] as i32, mx) as i16;
}
}
for (y, tmp_row) in tmp.chunks_mut(tmp_stride).take(h).enumerate() {
let mid_row = unsafe { mid.get_unchecked(y * mid_stride..) };
let mid_next_row = unsafe { mid.get_unchecked((y + 1) * mid_stride..) };
let (tmp_chunks16, tmp_rem16) = tmp_row[..w].as_chunks_mut::<16>();
for (chunk_idx, tmp_chunk) in tmp_chunks16.iter_mut().enumerate() {
let x = chunk_idx * 16;
let a0 = load_i16x8(unsafe { mid_row.get_unchecked(x..) });
let b0 = load_i16x8(unsafe { mid_next_row.get_unchecked(x..) });
let a1 = load_i16x8(unsafe { mid_row.get_unchecked(x + 8..) });
let b1 = load_i16x8(unsafe { mid_next_row.get_unchecked(x + 8..) });
let (lo, hi) = bilin_i16x16_i32(a0, b0, a1, b1, my);
store_i16x16_round4_from_i32(tmp_chunk, lo, hi);
}
let x16_done = tmp_chunks16.len() * 16;
let (tmp_chunks8, tmp_rem) = tmp_rem16.as_chunks_mut::<8>();
for (chunk_idx, tmp_chunk) in tmp_chunks8.iter_mut().enumerate() {
let x = x16_done + chunk_idx * 8;
let a = load_i16x8(unsafe { mid_row.get_unchecked(x..) });
let b = load_i16x8(unsafe { mid_next_row.get_unchecked(x..) });
store_i16x8_round4_from_i32(tmp_chunk, bilin_i16x8_i32(a, b, my));
}
let processed = x16_done + tmp_chunks8.len() * 8;
for (x, tmp_px) in (processed..w).zip(tmp_rem.iter_mut()) {
*tmp_px =
((bilin_scalar(mid_row[x] as i32, mid_next_row[x] as i32, my) + 8) >> 4) as i16;
}
}
} else if mx != 0 {
for (y, tmp_row) in tmp.chunks_mut(tmp_stride).take(h).enumerate() {
let src_row = unsafe { src.get_unchecked(y * src_stride..) };
let (tmp_chunks, tmp_rem) = tmp_row[..w].as_chunks_mut::<16>();
for (chunk_idx, tmp_chunk) in tmp_chunks.iter_mut().enumerate() {
let x = chunk_idx * 16;
store_i16x16(tmp_chunk, bilin_u8x16_i16(src_row, x, 1, mx));
}
let processed = tmp_chunks.len() * 16;
for (x, tmp_px) in (processed..w).zip(tmp_rem.iter_mut()) {
let si = x;
*tmp_px = bilin_scalar(src_row[si] as i32, src_row[si + 1] as i32, mx) as i16;
}
}
} else if my != 0 {
for (y, tmp_row) in tmp.chunks_mut(tmp_stride).take(h).enumerate() {
let src_row = unsafe { src.get_unchecked(y * src_stride..) };
let (tmp_chunks, tmp_rem) = tmp_row[..w].as_chunks_mut::<16>();
for (chunk_idx, tmp_chunk) in tmp_chunks.iter_mut().enumerate() {
let x = chunk_idx * 16;
store_i16x16(tmp_chunk, bilin_u8x16_i16(src_row, x, src_stride, my));
}
let processed = tmp_chunks.len() * 16;
for (x, tmp_px) in (processed..w).zip(tmp_rem.iter_mut()) {
let si = x;
*tmp_px =
bilin_scalar(src_row[si] as i32, src_row[si + src_stride] as i32, my) as i16;
}
}
} else {
for (y, tmp_row) in tmp.chunks_mut(tmp_stride).take(h).enumerate() {
let src_row = unsafe { src.get_unchecked(y * src_stride..) };
let (tmp_chunks, tmp_rem) = tmp_row[..w].as_chunks_mut::<16>();
for (chunk_idx, tmp_chunk) in tmp_chunks.iter_mut().enumerate() {
let x = chunk_idx * 16;
let v =
_mm256_slli_epi16::<4>(load_u8x16_i16(unsafe { src_row.get_unchecked(x..) }));
store_i16x16(tmp_chunk, v);
}
let processed = tmp_chunks.len() * 16;
for (x, tmp_px) in (processed..w).zip(tmp_rem.iter_mut()) {
*tmp_px = (src_row[x] as i16) << 4;
}
}
}
}
#[inline]
#[target_feature(enable = "avx2")]
fn madd_i16x8_pair_s32(a: __m128i, b: __m128i, c0: i16, c1: i16) -> __m256i {
let lo = _mm_unpacklo_epi16(a, b);
let hi = _mm_unpackhi_epi16(a, b);
let pairs = _mm256_inserti128_si256::<1>(_mm256_castsi128_si256(lo), hi);
let coeff = _mm256_set1_epi32(((c1 as u16 as i32) << 16) | c0 as u16 as i32);
_mm256_madd_epi16(pairs, coeff)
}
#[inline(always)]
fn pair_coeff(a: i8, b: i8) -> i16 {
(((b as i16 as u16) << 8) | ((a as i16 as u16) & 0xff)) as i16
}
#[inline]
#[target_feature(enable = "avx2")]
fn filter_u8x8_h(src: &[u8], base: usize, f: &[i8; 8]) -> __m256i {
unsafe {
let s = _mm_loadu_si128(src.get_unchecked(base - 3..).as_ptr().cast());
let shuf01 = _mm_setr_epi8(0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8);
let shuf23 = _mm_setr_epi8(2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8, 8, 9, 9, 10);
let shuf45 = _mm_setr_epi8(4, 5, 5, 6, 6, 7, 7, 8, 8, 9, 9, 10, 10, 11, 11, 12);
let shuf67 = _mm_setr_epi8(6, 7, 7, 8, 8, 9, 9, 10, 10, 11, 11, 12, 12, 13, 13, 14);
let p01 = _mm_maddubs_epi16(
_mm_shuffle_epi8(s, shuf01),
_mm_set1_epi16(pair_coeff(f[0], f[1])),
);
let p23 = _mm_maddubs_epi16(
_mm_shuffle_epi8(s, shuf23),
_mm_set1_epi16(pair_coeff(f[2], f[3])),
);
let p45 = _mm_maddubs_epi16(
_mm_shuffle_epi8(s, shuf45),
_mm_set1_epi16(pair_coeff(f[4], f[5])),
);
let p67 = _mm_maddubs_epi16(
_mm_shuffle_epi8(s, shuf67),
_mm_set1_epi16(pair_coeff(f[6], f[7])),
);
let sum = _mm_add_epi16(_mm_add_epi16(p01, p23), _mm_add_epi16(p45, p67));
_mm256_cvtepi16_epi32(sum)
}
}
#[inline]
#[target_feature(enable = "avx2")]
#[allow(clippy::neg_multiply)]
fn filter_u8x8_v(src: &[u8], base: usize, stride: isize, f: &[i8; 8]) -> __m256i {
unsafe {
let mut sum = _mm_setzero_si128();
macro_rules! add_pair {
($a:literal, $b:literal, $oa:expr, $ob:expr) => {{
if f[$a] != 0 || f[$b] != 0 {
let ia = (base as isize + $oa * stride) as usize;
let ib = (base as isize + $ob * stride) as usize;
let a = _mm_loadl_epi64(src.get_unchecked(ia..).as_ptr().cast());
let b = _mm_loadl_epi64(src.get_unchecked(ib..).as_ptr().cast());
let pairs = _mm_unpacklo_epi8(a, b);
let coeff = _mm_set1_epi16(pair_coeff(f[$a], f[$b]));
sum = _mm_add_epi16(sum, _mm_maddubs_epi16(pairs, coeff));
}
}};
}
add_pair!(0, 1, -3isize, -2isize);
add_pair!(2, 3, -1isize, 0isize);
add_pair!(4, 5, 1isize, 2isize);
add_pair!(6, 7, 3isize, 4isize);
_mm256_cvtepi16_epi32(sum)
}
}
#[inline(always)]
fn needs_wide(f: &[i8; 8]) -> bool {
255 * f.iter().map(|&c| (c as i32).max(0)).sum::<i32>() > i16::MAX as i32
}
#[inline]
#[target_feature(enable = "avx2")]
fn filter_u8x8_wide(src: &[u8], base: usize, stride: isize, f: &[i8; 8]) -> __m256i {
static OFFSETS: [isize; 8] = [-3isize, -2, -1, 0, 1, 2, 3, 4];
let mut sum = _mm256_setzero_si256();
for k in (0..8).step_by(2) {
let (c0, c1) = (f[k], f[k + 1]);
if c0 == 0 && c1 == 0 {
continue;
}
let idx0 = (base as isize + OFFSETS[k] * stride) as usize;
let idx1 = (base as isize + OFFSETS[k + 1] * stride) as usize;
let (a, b) = unsafe {
(
_mm_cvtepu8_epi16(_mm_loadl_epi64(src.get_unchecked(idx0..).as_ptr().cast())),
_mm_cvtepu8_epi16(_mm_loadl_epi64(src.get_unchecked(idx1..).as_ptr().cast())),
)
};
sum = _mm256_add_epi32(sum, madd_i16x8_pair_s32(a, b, c0 as i16, c1 as i16));
}
sum
}
#[inline]
#[target_feature(enable = "avx2")]
fn filter_u8x8(src: &[u8], base: usize, stride: isize, f: &[i8; 8]) -> __m256i {
if needs_wide(f) {
filter_u8x8_wide(src, base, stride, f)
} else if stride == 1 {
filter_u8x8_h(src, base, f)
} else {
filter_u8x8_v(src, base, stride, f)
}
}
#[inline]
#[target_feature(enable = "avx2")]
#[allow(clippy::neg_multiply)]
fn filter_u8x16_v(src: &[u8], base: usize, stride: isize, f: &[i8; 8]) -> (__m256i, __m256i) {
unsafe {
let mut lo = _mm_setzero_si128();
let mut hi = _mm_setzero_si128();
macro_rules! add_pair {
($a:literal, $b:literal, $oa:expr, $ob:expr) => {{
if f[$a] != 0 || f[$b] != 0 {
let ia = (base as isize + $oa * stride) as usize;
let ib = (base as isize + $ob * stride) as usize;
let a = _mm_loadu_si128(src.get_unchecked(ia..).as_ptr().cast());
let b = _mm_loadu_si128(src.get_unchecked(ib..).as_ptr().cast());
let coeff = _mm_set1_epi16(pair_coeff(f[$a], f[$b]));
lo = _mm_add_epi16(lo, _mm_maddubs_epi16(_mm_unpacklo_epi8(a, b), coeff));
hi = _mm_add_epi16(hi, _mm_maddubs_epi16(_mm_unpackhi_epi8(a, b), coeff));
}
}};
}
add_pair!(0, 1, -3isize, -2isize);
add_pair!(2, 3, -1isize, 0isize);
add_pair!(4, 5, 1isize, 2isize);
add_pair!(6, 7, 3isize, 4isize);
(_mm256_cvtepi16_epi32(lo), _mm256_cvtepi16_epi32(hi))
}
}
#[inline]
#[target_feature(enable = "avx2")]
fn filter_u8x16(src: &[u8], base: usize, stride: isize, f: &[i8; 8]) -> (__m256i, __m256i) {
if needs_wide(f) {
(
filter_u8x8_wide(src, base, stride, f),
filter_u8x8_wide(src, base + 8, stride, f),
)
} else if stride == 1 {
(filter_u8x8_h(src, base, f), filter_u8x8_h(src, base + 8, f))
} else {
filter_u8x16_v(src, base, stride, f)
}
}
#[inline]
#[target_feature(enable = "avx2")]
fn filter_i16x8_8tap(src: &[i16], base: usize, stride: isize, f: &[i8; 8]) -> __m256i {
static OFFSETS: [isize; 8] = [-3isize, -2, -1, 0, 1, 2, 3, 4];
let mut sum = _mm256_setzero_si256();
for k in (0..8).step_by(2) {
let c0 = f[k];
let c1 = f[k + 1];
if c0 == 0 && c1 == 0 {
continue;
}
let idx0 = (base as isize + OFFSETS[k] * stride) as usize;
let idx1 = (base as isize + OFFSETS[k + 1] * stride) as usize;
let a = load_i16x8(unsafe { src.get_unchecked(idx0..) });
let b = load_i16x8(unsafe { src.get_unchecked(idx1..) });
sum = _mm256_add_epi32(sum, madd_i16x8_pair_s32(a, b, c0 as i16, c1 as i16));
}
sum
}
#[inline]
#[target_feature(enable = "avx2")]
fn filter_i16x16_8tap(src: &[i16], base: usize, stride: isize, f: &[i8; 8]) -> (__m256i, __m256i) {
static OFFSETS: [isize; 8] = [-3isize, -2, -1, 0, 1, 2, 3, 4];
let mut lo = _mm256_setzero_si256();
let mut hi = _mm256_setzero_si256();
for k in (0..8).step_by(2) {
let c0 = f[k];
let c1 = f[k + 1];
if c0 == 0 && c1 == 0 {
continue;
}
let idx0 = (base as isize + OFFSETS[k] * stride) as usize;
let idx1 = (base as isize + OFFSETS[k + 1] * stride) as usize;
let a0 = load_i16x8(unsafe { src.get_unchecked(idx0..) });
let b0 = load_i16x8(unsafe { src.get_unchecked(idx1..) });
let a1 = load_i16x8(unsafe { src.get_unchecked(idx0 + 8..) });
let b1 = load_i16x8(unsafe { src.get_unchecked(idx1 + 8..) });
lo = _mm256_add_epi32(lo, madd_i16x8_pair_s32(a0, b0, c0 as i16, c1 as i16));
hi = _mm256_add_epi32(hi, madd_i16x8_pair_s32(a1, b1, c0 as i16, c1 as i16));
}
(lo, hi)
}
#[inline(always)]
fn filter_u8_scalar(src: &[u8], base: usize, stride: isize, f: &[i8; 8]) -> i32 {
let c = base as isize;
f[0] as i32 * src[(c - 3 * stride) as usize] as i32
+ f[1] as i32 * src[(c - 2 * stride) as usize] as i32
+ f[2] as i32 * src[(c - stride) as usize] as i32
+ f[3] as i32 * src[base] as i32
+ f[4] as i32 * src[(c + stride) as usize] as i32
+ f[5] as i32 * src[(c + 2 * stride) as usize] as i32
+ f[6] as i32 * src[(c + 3 * stride) as usize] as i32
+ f[7] as i32 * src[(c + 4 * stride) as usize] as i32
}
#[inline(always)]
fn filter_i16_scalar(src: &[i16], base: usize, stride: isize, f: &[i8; 8]) -> i32 {
let c = base as isize;
f[0] as i32 * src[(c - 3 * stride) as usize] as i32
+ f[1] as i32 * src[(c - 2 * stride) as usize] as i32
+ f[2] as i32 * src[(c - stride) as usize] as i32
+ f[3] as i32 * src[base] as i32
+ f[4] as i32 * src[(c + stride) as usize] as i32
+ f[5] as i32 * src[(c + 2 * stride) as usize] as i32
+ f[6] as i32 * src[(c + 3 * stride) as usize] as i32
+ f[7] as i32 * src[(c + 4 * stride) as usize] as i32
}
#[inline(always)]
fn round_scalar(v: i32, rnd: i32, shift: i32) -> i32 {
if shift == 0 {
v + rnd
} else {
(v + rnd) >> shift
}
}
#[inline]
#[target_feature(enable = "avx2")]
fn round_s32(v: __m256i, rnd: i32, shift: i32) -> __m256i {
_mm256_sra_epi32(
_mm256_add_epi32(v, _mm256_set1_epi32(rnd)),
_mm_cvtsi32_si128(shift),
)
}
#[inline]
#[target_feature(enable = "avx2")]
fn store_u8x8_clip_shift(dst: &mut [u8], v: __m256i, rnd: i32, shift: i32) {
unsafe {
let v = round_s32(v, rnd, shift);
let lo = _mm256_castsi256_si128(v);
let hi = _mm256_extracti128_si256::<1>(v);
let p16 = _mm_packus_epi32(lo, hi);
let p8 = _mm_packus_epi16(p16, p16);
_mm_storel_epi64(dst.as_mut_ptr().cast(), p8);
}
}
#[inline]
#[target_feature(enable = "avx2")]
fn store_i16x8_shift(dst: &mut [i16], v: __m256i, rnd: i32, shift: i32) {
let v = round_s32(v, rnd, shift);
store_i16x8(dst, v);
}
#[inline]
#[target_feature(enable = "avx2")]
fn store_u8x16_clip_shift(dst: &mut [u8], lo: __m256i, hi: __m256i, rnd: i32, shift: i32) {
unsafe {
let lo = round_s32(lo, rnd, shift);
let hi = round_s32(hi, rnd, shift);
let lo0 = _mm256_castsi256_si128(lo);
let lo1 = _mm256_extracti128_si256::<1>(lo);
let hi0 = _mm256_castsi256_si128(hi);
let hi1 = _mm256_extracti128_si256::<1>(hi);
let lo16 = _mm_packus_epi32(lo0, lo1);
let hi16 = _mm_packus_epi32(hi0, hi1);
_mm_storeu_si128(dst.as_mut_ptr().cast(), _mm_packus_epi16(lo16, hi16));
}
}
#[inline]
#[target_feature(enable = "avx2")]
fn store_i16x16_shift(dst: &mut [i16], lo: __m256i, hi: __m256i, rnd: i32, shift: i32) {
unsafe {
let lo = round_s32(lo, rnd, shift);
let hi = round_s32(hi, rnd, shift);
let lo0 = _mm256_castsi256_si128(lo);
let lo1 = _mm256_extracti128_si256::<1>(lo);
let hi0 = _mm256_castsi256_si128(hi);
let hi1 = _mm256_extracti128_si256::<1>(hi);
_mm_storeu_si128(dst.as_mut_ptr().cast(), _mm_packs_epi32(lo0, lo1));
_mm_storeu_si128(dst.as_mut_ptr().add(8).cast(), _mm_packs_epi32(hi0, hi1));
}
}
#[target_feature(enable = "avx2")]
#[allow(clippy::too_many_arguments)]
pub(crate) fn put_8tap_8bpc_avx2(
dst: &mut [u8],
dst_stride: usize,
src: &[u8],
src_off: usize,
src_stride: usize,
w: usize,
h: usize,
mx: i32,
my: i32,
filter_type: i32,
mid_scratch: &mut [i16],
) {
let bits = 6 + (filter_type < 0) as i32;
let intermediate_rnd = ((1 << bits) >> 1) + ((1 << (bits - 4)) >> 1);
let fh = crate::mc::get_h_filter(mx, filter_type, w);
let fv = crate::mc::get_v_filter(my, filter_type, h);
match (fh, fv) {
(Some(fh), Some(fv)) => {
let tmp_h = h + 7;
let mid_stride = w.next_multiple_of(16).max(64);
let mid = &mut mid_scratch[..mid_stride * tmp_h];
let sh0 = bits - 4;
let rnd0 = (1 << sh0) >> 1;
for (y, mid_row) in mid.chunks_mut(mid_stride).take(tmp_h).enumerate() {
let base = (src_off as isize + (y as isize - 3) * src_stride as isize) as usize;
let (mid_chunks16, mid_rem16) = mid_row[..w].as_chunks_mut::<16>();
for (chunk_idx, mid_chunk) in mid_chunks16.iter_mut().enumerate() {
let x = chunk_idx * 16;
let (lo, hi) = filter_u8x16(src, base + x, 1, &fh);
store_i16x16_shift(mid_chunk, lo, hi, rnd0, sh0);
}
let x16_done = mid_chunks16.len() * 16;
let (mid_chunks8, mid_rem) = mid_rem16.as_chunks_mut::<8>();
for (chunk_idx, mid_chunk) in mid_chunks8.iter_mut().enumerate() {
let x = x16_done + chunk_idx * 8;
store_i16x8_shift(mid_chunk, filter_u8x8(src, base + x, 1, &fh), rnd0, sh0);
}
let processed = x16_done + mid_chunks8.len() * 8;
for (x, mid_px) in (processed..w).zip(mid_rem.iter_mut()) {
*mid_px =
round_scalar(filter_u8_scalar(src, base + x, 1, &fh), rnd0, sh0) as i16;
}
}
let sh1 = bits + 4;
let rnd1 = (1 << sh1) >> 1;
for (y, dst_row) in dst.chunks_mut(dst_stride).take(h).enumerate() {
let (dst_chunks16, dst_rem16) = dst_row[..w].as_chunks_mut::<16>();
for (chunk_idx, dst_chunk) in dst_chunks16.iter_mut().enumerate() {
let x = chunk_idx * 16;
let (lo, hi) = filter_i16x16_8tap(
&mid,
(y + 3) * mid_stride + x,
mid_stride as isize,
&fv,
);
store_u8x16_clip_shift(dst_chunk, lo, hi, rnd1, sh1);
}
let x16_done = dst_chunks16.len() * 16;
let (dst_chunks8, dst_rem) = dst_rem16.as_chunks_mut::<8>();
for (chunk_idx, dst_chunk) in dst_chunks8.iter_mut().enumerate() {
let x = x16_done + chunk_idx * 8;
store_u8x8_clip_shift(
dst_chunk,
filter_i16x8_8tap(&mid, (y + 3) * mid_stride + x, mid_stride as isize, &fv),
rnd1,
sh1,
);
}
let processed = x16_done + dst_chunks8.len() * 8;
for (x, dst_px) in (processed..w).zip(dst_rem.iter_mut()) {
*dst_px = round_scalar(
filter_i16_scalar(&mid, (y + 3) * mid_stride + x, mid_stride as isize, &fv),
rnd1,
sh1,
)
.clamp(0, 255) as u8;
}
}
}
(Some(fh), None) => {
for (y, dst_row) in dst.chunks_mut(dst_stride).take(h).enumerate() {
let base = src_off + y * src_stride;
let (dst_chunks16, dst_rem16) = dst_row[..w].as_chunks_mut::<16>();
for (chunk_idx, dst_chunk) in dst_chunks16.iter_mut().enumerate() {
let x = chunk_idx * 16;
let (lo, hi) = filter_u8x16(src, base + x, 1, &fh);
store_u8x16_clip_shift(dst_chunk, lo, hi, intermediate_rnd, bits);
}
let x16_done = dst_chunks16.len() * 16;
let (dst_chunks8, dst_rem) = dst_rem16.as_chunks_mut::<8>();
for (chunk_idx, dst_chunk) in dst_chunks8.iter_mut().enumerate() {
let x = x16_done + chunk_idx * 8;
store_u8x8_clip_shift(
dst_chunk,
filter_u8x8(src, base + x, 1, &fh),
intermediate_rnd,
bits,
);
}
let processed = x16_done + dst_chunks8.len() * 8;
for (x, dst_px) in (processed..w).zip(dst_rem.iter_mut()) {
*dst_px = round_scalar(
filter_u8_scalar(src, base + x, 1, &fh),
intermediate_rnd,
bits,
)
.clamp(0, 255) as u8;
}
}
}
(None, Some(fv)) => {
let ss = src_stride as isize;
for (y, dst_row) in dst.chunks_mut(dst_stride).take(h).enumerate() {
let base = src_off + y * src_stride;
let (dst_chunks16, dst_rem16) = dst_row[..w].as_chunks_mut::<16>();
for (chunk_idx, dst_chunk) in dst_chunks16.iter_mut().enumerate() {
let x = chunk_idx * 16;
let (lo, hi) = filter_u8x16(src, base + x, ss, &fv);
store_u8x16_clip_shift(dst_chunk, lo, hi, (1 << bits) >> 1, bits);
}
let x16_done = dst_chunks16.len() * 16;
let (dst_chunks8, dst_rem) = dst_rem16.as_chunks_mut::<8>();
for (chunk_idx, dst_chunk) in dst_chunks8.iter_mut().enumerate() {
let x = x16_done + chunk_idx * 8;
store_u8x8_clip_shift(
dst_chunk,
filter_u8x8(src, base + x, ss, &fv),
(1 << bits) >> 1,
bits,
);
}
let processed = x16_done + dst_chunks8.len() * 8;
for (x, dst_px) in (processed..w).zip(dst_rem.iter_mut()) {
*dst_px = round_scalar(
filter_u8_scalar(src, base + x, ss, &fv),
(1 << bits) >> 1,
bits,
)
.clamp(0, 255) as u8;
}
}
}
(None, None) => {
for (src_row, dst_row) in src[src_off..]
.chunks(src_stride)
.zip(dst.chunks_mut(dst_stride))
.take(h)
{
dst_row[..w].copy_from_slice(&src_row[..w]);
}
}
}
}
#[target_feature(enable = "avx2")]
#[allow(clippy::too_many_arguments)]
pub(crate) fn prep_8tap_8bpc_avx2(
tmp: &mut [i16],
tmp_stride: usize,
src: &[u8],
src_off: usize,
src_stride: usize,
w: usize,
h: usize,
mx: i32,
my: i32,
filter_type: i32,
mid_scratch: &mut [i16],
) {
let bits = 6 + (filter_type < 0) as i32;
let fh = crate::mc::get_h_filter(mx, filter_type, w);
let fv = crate::mc::get_v_filter(my, filter_type, h);
match (fh, fv) {
(Some(fh), Some(fv)) => {
let tmp_h = h + 7;
let mid_stride = w.next_multiple_of(16).max(64);
let mid = &mut mid_scratch[..mid_stride * tmp_h];
let sh0 = bits - 4;
let rnd0 = (1 << sh0) >> 1;
for (y, mid_row) in mid.chunks_mut(mid_stride).take(tmp_h).enumerate() {
let base = (src_off as isize + (y as isize - 3) * src_stride as isize) as usize;
let (mid_chunks16, mid_rem16) = mid_row[..w].as_chunks_mut::<16>();
for (chunk_idx, mid_chunk) in mid_chunks16.iter_mut().enumerate() {
let x = chunk_idx * 16;
let (lo, hi) = filter_u8x16(src, base + x, 1, &fh);
store_i16x16_shift(mid_chunk, lo, hi, rnd0, sh0);
}
let x16_done = mid_chunks16.len() * 16;
let (mid_chunks8, mid_rem) = mid_rem16.as_chunks_mut::<8>();
for (chunk_idx, mid_chunk) in mid_chunks8.iter_mut().enumerate() {
let x = x16_done + chunk_idx * 8;
store_i16x8_shift(mid_chunk, filter_u8x8(src, base + x, 1, &fh), rnd0, sh0);
}
let processed = x16_done + mid_chunks8.len() * 8;
for (x, mid_px) in (processed..w).zip(mid_rem.iter_mut()) {
*mid_px =
round_scalar(filter_u8_scalar(src, base + x, 1, &fh), rnd0, sh0) as i16;
}
}
let rnd1 = (1 << bits) >> 1;
for (y, tmp_row) in tmp.chunks_mut(tmp_stride).take(h).enumerate() {
let (tmp_chunks16, tmp_rem16) = tmp_row[..w].as_chunks_mut::<16>();
for (chunk_idx, tmp_chunk) in tmp_chunks16.iter_mut().enumerate() {
let x = chunk_idx * 16;
let (lo, hi) = filter_i16x16_8tap(
&mid,
(y + 3) * mid_stride + x,
mid_stride as isize,
&fv,
);
store_i16x16_shift(tmp_chunk, lo, hi, rnd1, bits);
}
let x16_done = tmp_chunks16.len() * 16;
let (tmp_chunks8, tmp_rem) = tmp_rem16.as_chunks_mut::<8>();
for (chunk_idx, tmp_chunk) in tmp_chunks8.iter_mut().enumerate() {
let x = x16_done + chunk_idx * 8;
store_i16x8_shift(
tmp_chunk,
filter_i16x8_8tap(&mid, (y + 3) * mid_stride + x, mid_stride as isize, &fv),
rnd1,
bits,
);
}
let processed = x16_done + tmp_chunks8.len() * 8;
for (x, tmp_px) in (processed..w).zip(tmp_rem.iter_mut()) {
*tmp_px = round_scalar(
filter_i16_scalar(&mid, (y + 3) * mid_stride + x, mid_stride as isize, &fv),
rnd1,
bits,
) as i16;
}
}
}
(Some(fh), None) => {
let sh0 = bits - 4;
let rnd0 = (1 << sh0) >> 1;
for (y, tmp_row) in tmp.chunks_mut(tmp_stride).take(h).enumerate() {
let base = src_off + y * src_stride;
let (tmp_chunks16, tmp_rem16) = tmp_row[..w].as_chunks_mut::<16>();
for (chunk_idx, tmp_chunk) in tmp_chunks16.iter_mut().enumerate() {
let x = chunk_idx * 16;
let (lo, hi) = filter_u8x16(src, base + x, 1, &fh);
store_i16x16_shift(tmp_chunk, lo, hi, rnd0, sh0);
}
let x16_done = tmp_chunks16.len() * 16;
let (tmp_chunks8, tmp_rem) = tmp_rem16.as_chunks_mut::<8>();
for (chunk_idx, tmp_chunk) in tmp_chunks8.iter_mut().enumerate() {
let x = x16_done + chunk_idx * 8;
store_i16x8_shift(tmp_chunk, filter_u8x8(src, base + x, 1, &fh), rnd0, sh0);
}
let processed = x16_done + tmp_chunks8.len() * 8;
for (x, tmp_px) in (processed..w).zip(tmp_rem.iter_mut()) {
*tmp_px =
round_scalar(filter_u8_scalar(src, base + x, 1, &fh), rnd0, sh0) as i16;
}
}
}
(None, Some(fv)) => {
let ss = src_stride as isize;
let sh0 = bits - 4;
let rnd0 = (1 << sh0) >> 1;
for (y, tmp_row) in tmp.chunks_mut(tmp_stride).take(h).enumerate() {
let base = src_off + y * src_stride;
let (tmp_chunks16, tmp_rem16) = tmp_row[..w].as_chunks_mut::<16>();
for (chunk_idx, tmp_chunk) in tmp_chunks16.iter_mut().enumerate() {
let x = chunk_idx * 16;
let (lo, hi) = filter_u8x16(src, base + x, ss, &fv);
store_i16x16_shift(tmp_chunk, lo, hi, rnd0, sh0);
}
let x16_done = tmp_chunks16.len() * 16;
let (tmp_chunks8, tmp_rem) = tmp_rem16.as_chunks_mut::<8>();
for (chunk_idx, tmp_chunk) in tmp_chunks8.iter_mut().enumerate() {
let x = x16_done + chunk_idx * 8;
store_i16x8_shift(tmp_chunk, filter_u8x8(src, base + x, ss, &fv), rnd0, sh0);
}
let processed = x16_done + tmp_chunks8.len() * 8;
for (x, tmp_px) in (processed..w).zip(tmp_rem.iter_mut()) {
*tmp_px =
round_scalar(filter_u8_scalar(src, base + x, ss, &fv), rnd0, sh0) as i16;
}
}
}
(None, None) => {
for (y, tmp_row) in tmp.chunks_mut(tmp_stride).take(h).enumerate() {
let base = src_off + y * src_stride;
let src_row = unsafe { src.get_unchecked(base..) };
let (tmp_chunks, tmp_rem) = tmp_row[..w].as_chunks_mut::<16>();
for (chunk_idx, tmp_chunk) in tmp_chunks.iter_mut().enumerate() {
let x = chunk_idx * 16;
let v = _mm256_slli_epi16::<4>(load_u8x16_i16(unsafe {
src_row.get_unchecked(x..)
}));
store_i16x16(tmp_chunk, v);
}
let processed = tmp_chunks.len() * 16;
for (x, tmp_px) in (processed..w).zip(tmp_rem.iter_mut()) {
*tmp_px = (src_row[x] as i16) << 4;
}
}
}
}
}
#[inline]
#[target_feature(enable = "avx2")]
fn load_u8x8_i32_warp(src: &[u8]) -> __m256i {
unsafe { _mm256_cvtepu8_epi32(_mm_loadl_epi64(src.as_ptr().cast())) }
}
#[inline]
#[target_feature(enable = "avx2")]
fn load_i16x8_i32_warp(src: &[i16]) -> __m256i {
unsafe { _mm256_cvtepi16_epi32(_mm_loadu_si128(src.as_ptr().cast())) }
}
#[inline]
#[target_feature(enable = "avx2")]
fn warp_coeff_i32x8(pos: i32, step: i32, tap: usize) -> __m256i {
let f0 = &crate::tables::MC_WARP_FILTER[(192 + ((pos + 512) >> 10)) as usize];
let f1 = &crate::tables::MC_WARP_FILTER[(192 + ((pos + step + 512) >> 10)) as usize];
let f2 = &crate::tables::MC_WARP_FILTER[(192 + ((pos + step * 2 + 512) >> 10)) as usize];
let f3 = &crate::tables::MC_WARP_FILTER[(192 + ((pos + step * 3 + 512) >> 10)) as usize];
let f4 = &crate::tables::MC_WARP_FILTER[(192 + ((pos + step * 4 + 512) >> 10)) as usize];
let f5 = &crate::tables::MC_WARP_FILTER[(192 + ((pos + step * 5 + 512) >> 10)) as usize];
let f6 = &crate::tables::MC_WARP_FILTER[(192 + ((pos + step * 6 + 512) >> 10)) as usize];
let f7 = &crate::tables::MC_WARP_FILTER[(192 + ((pos + step * 7 + 512) >> 10)) as usize];
_mm256_setr_epi32(
f0[tap] as i32,
f1[tap] as i32,
f2[tap] as i32,
f3[tap] as i32,
f4[tap] as i32,
f5[tap] as i32,
f6[tap] as i32,
f7[tap] as i32,
)
}
#[inline]
#[target_feature(enable = "avx2")]
fn warp_horz_u8x8(src: &[u8], row_base: usize, mx: i32, alpha: i32) -> __m256i {
let mut acc = _mm256_setzero_si256();
for tap in 0..8 {
let px = load_u8x8_i32_warp(unsafe { src.get_unchecked(row_base + tap..) });
let coeff = warp_coeff_i32x8(mx, alpha, tap);
acc = _mm256_add_epi32(acc, _mm256_mullo_epi32(px, coeff));
}
round_s32(acc, 4, 3)
}
#[inline]
#[target_feature(enable = "avx2")]
fn warp_vert_i16x8(mid: &[i16], base: usize, stride: usize, my: i32, gamma: i32) -> __m256i {
let mut acc = _mm256_setzero_si256();
for tap in 0..8 {
let px = load_i16x8_i32_warp(unsafe { mid.get_unchecked(base + tap * stride..) });
let coeff = warp_coeff_i32x8(my, gamma, tap);
acc = _mm256_add_epi32(acc, _mm256_mullo_epi32(px, coeff));
}
acc
}
#[inline]
#[target_feature(enable = "avx2")]
fn store_i16x8_shift_warp(dst: &mut [i16], v: __m256i, rnd: i32, shift: i32) {
store_i16x8(dst, round_s32(v, rnd, shift));
}
#[inline]
#[target_feature(enable = "avx2")]
fn store_u8x8_shift_warp(dst: &mut [u8], v: __m256i, rnd: i32, shift: i32) {
store_u8x8_clip_shift(dst, v, rnd, shift);
}
#[allow(clippy::too_many_arguments)]
#[target_feature(enable = "avx2")]
pub(crate) fn warp_affine_8x8_8bpc_avx2(
dst: &mut [u8],
dst_stride: usize,
src: &[u8],
src_stride: usize,
src_off: usize,
abcd: &[i16; 4],
mut mx: i32,
mut my: i32,
) {
let alpha = abcd[0] as i32;
let beta = abcd[1] as i32;
let gamma = abcd[2] as i32;
let delta = abcd[3] as i32;
let mut mid = [0i16; 15 * 8];
let mut row_base = src_off.wrapping_sub(3 * src_stride + 3);
for mid_row in mid.as_chunks_mut::<8>().0.iter_mut() {
store_i16x8(mid_row, warp_horz_u8x8(src, row_base, mx, alpha));
row_base += src_stride;
mx += beta;
}
for (y, dst_row) in dst.chunks_mut(dst_stride).take(8).enumerate() {
let v = warp_vert_i16x8(&mid, y * 8, 8, my, gamma);
store_u8x8_shift_warp(&mut dst_row[..8], v, 1024, 11);
my += delta;
}
}
#[allow(clippy::too_many_arguments)]
#[target_feature(enable = "avx2")]
pub(crate) fn warp_affine_8x8t_8bpc_avx2(
tmp: &mut [i16],
tmp_stride: usize,
src: &[u8],
src_stride: usize,
src_off: usize,
abcd: &[i16; 4],
mut mx: i32,
mut my: i32,
) {
let alpha = abcd[0] as i32;
let beta = abcd[1] as i32;
let gamma = abcd[2] as i32;
let delta = abcd[3] as i32;
let mut mid = [0i16; 15 * 8];
let mut row_base = src_off.wrapping_sub(3 * src_stride + 3);
for mid_row in mid.as_chunks_mut::<8>().0.iter_mut() {
store_i16x8(mid_row, warp_horz_u8x8(src, row_base, mx, alpha));
row_base += src_stride;
mx += beta;
}
for (y, tmp_row) in tmp.chunks_mut(tmp_stride).take(8).enumerate() {
let v = warp_vert_i16x8(&mid, y * 8, 8, my, gamma);
store_i16x8_shift_warp(&mut tmp_row[..8], v, 64, 7);
my += delta;
}
}