static void intra_predict_dc_tl_32x32_msa(const uint8_t *src, uint8_t *dst,
int32_t dst_stride) {
uint32_t row;
v16u8 data0, data1, out;
v8u16 sum_h, sum_data0, sum_data1;
v4u32 sum_w;
v2u64 sum_d;
LD_UB2(src, 16, data0, data1);
HADD_UB2_UH(data0, data1, sum_data0, sum_data1);
sum_h = sum_data0 + sum_data1;
sum_w = __msa_hadd_u_w(sum_h, sum_h);
sum_d = __msa_hadd_u_d(sum_w, sum_w);
sum_w = (v4u32)__msa_pckev_w((v4i32)sum_d, (v4i32)sum_d);
sum_d = __msa_hadd_u_d(sum_w, sum_w);
sum_w = (v4u32)__msa_srari_w((v4i32)sum_d, 5);
out = (v16u8)__msa_splati_b((v16i8)sum_w, 0);
for (row = 16; row--;) {
ST_UB2(out, out, dst, 16);
dst += dst_stride;
ST_UB2(out, out, dst, 16);
dst += dst_stride;
}
}
static void intra_predict_horiz_32x32_msa(const uint8_t *src, uint8_t *dst,
int32_t dst_stride) {
uint32_t row;
uint8_t inp0, inp1, inp2, inp3;
v16u8 src0, src1, src2, src3;
for (row = 8; row--;) {
inp0 = src[0];
inp1 = src[1];
inp2 = src[2];
inp3 = src[3];
src += 4;
src0 = (v16u8)__msa_fill_b(inp0);
src1 = (v16u8)__msa_fill_b(inp1);
src2 = (v16u8)__msa_fill_b(inp2);
src3 = (v16u8)__msa_fill_b(inp3);
ST_UB2(src0, src0, dst, 16);
dst += dst_stride;
ST_UB2(src1, src1, dst, 16);
dst += dst_stride;
ST_UB2(src2, src2, dst, 16);
dst += dst_stride;
ST_UB2(src3, src3, dst, 16);
dst += dst_stride;
}
}
static void intra_predict_dc_32x32_msa(const uint8_t *src_top,
const uint8_t *src_left, uint8_t *dst,
int32_t dst_stride) {
uint32_t row;
v16u8 top0, top1, left0, left1, out;
v8u16 sum_h, sum_top0, sum_top1, sum_left0, sum_left1;
v4u32 sum_w;
v2u64 sum_d;
LD_UB2(src_top, 16, top0, top1);
LD_UB2(src_left, 16, left0, left1);
HADD_UB2_UH(top0, top1, sum_top0, sum_top1);
HADD_UB2_UH(left0, left1, sum_left0, sum_left1);
sum_h = sum_top0 + sum_top1;
sum_h += sum_left0 + sum_left1;
sum_w = __msa_hadd_u_w(sum_h, sum_h);
sum_d = __msa_hadd_u_d(sum_w, sum_w);
sum_w = (v4u32)__msa_pckev_w((v4i32)sum_d, (v4i32)sum_d);
sum_d = __msa_hadd_u_d(sum_w, sum_w);
sum_w = (v4u32)__msa_srari_w((v4i32)sum_d, 6);
out = (v16u8)__msa_splati_b((v16i8)sum_w, 0);
for (row = 16; row--;) {
ST_UB2(out, out, dst, 16);
dst += dst_stride;
ST_UB2(out, out, dst, 16);
dst += dst_stride;
}
}
static void intra_predict_128dc_32x32_msa(uint8_t *dst, int32_t dst_stride) {
uint32_t row;
const v16u8 out = (v16u8)__msa_ldi_b(128);
for (row = 16; row--;) {
ST_UB2(out, out, dst, 16);
dst += dst_stride;
ST_UB2(out, out, dst, 16);
dst += dst_stride;
}
}
static WEBP_INLINE void PredictLineInverse0(const uint8_t* src,
const uint8_t* pred,
uint8_t* dst, int length) {
v16u8 src0, pred0, dst0;
assert(length >= 0);
while (length >= 32) {
v16u8 src1, pred1, dst1;
LD_UB2(src, 16, src0, src1);
LD_UB2(pred, 16, pred0, pred1);
SUB2(src0, pred0, src1, pred1, dst0, dst1);
ST_UB2(dst0, dst1, dst, 16);
src += 32;
pred += 32;
dst += 32;
length -= 32;
}
if (length > 0) {
int i;
if (length >= 16) {
src0 = LD_UB(src);
pred0 = LD_UB(pred);
dst0 = src0 - pred0;
ST_UB(dst0, dst);
src += 16;
pred += 16;
dst += 16;
length -= 16;
}
for (i = 0; i < length; i++) {
dst[i] = src[i] - pred[i];
}
}
}
void vp8_loop_filter_simple_horizontal_edge_msa(uint8_t *src, int32_t pitch,
const uint8_t *b_limit_ptr) {
v16u8 p1, p0, q1, q0;
v16u8 mask, b_limit;
b_limit = (v16u8)__msa_fill_b(*b_limit_ptr);
LD_UB4(src - (pitch << 1), pitch, p1, p0, q0, q1);
VP8_SIMPLE_MASK(p1, p0, q0, q1, b_limit, mask);
VP8_SIMPLE_FILT(p1, p0, q0, q1, mask);
ST_UB2(p0, q0, (src - pitch), pitch);
}
static void intra_predict_vert_32x32_msa(const uint8_t *src, uint8_t *dst,
int32_t dst_stride) {
uint32_t row;
v16u8 src1, src2;
src1 = LD_UB(src);
src2 = LD_UB(src + 16);
for (row = 32; row--;) {
ST_UB2(src1, src2, dst, 16);
dst += dst_stride;
}
}
void ff_vp8_v_loop_filter_simple_msa(uint8_t *src, ptrdiff_t pitch,
int b_limit_ptr)
{
v16u8 p1, p0, q1, q0;
v16u8 mask, b_limit;
b_limit = (v16u8) __msa_fill_b(b_limit_ptr);
/* load vector elements */
LD_UB4(src - (pitch << 1), pitch, p1, p0, q0, q1);
VP8_SIMPLE_MASK(p1, p0, q0, q1, b_limit, mask);
VP8_SIMPLE_FILT(p1, p0, q0, q1, mask);
ST_UB2(p0, q0, (src - pitch), pitch);
}
static void avc_sub4x4_dct_msa( uint8_t *p_src, int32_t i_src_stride,
uint8_t *p_ref, int32_t i_dst_stride,
int16_t *p_dst )
{
uint32_t i_src0, i_src1, i_src2, i_src3;
uint32_t i_ref0, i_ref1, i_ref2, i_ref3;
v16i8 src = { 0 };
v16i8 ref = { 0 };
v16u8 inp0, inp1;
v8i16 diff0, diff1, diff2, diff3;
v8i16 temp0, temp1, temp2, temp3;
LW4( p_src, i_src_stride, i_src0, i_src1, i_src2, i_src3 );
LW4( p_ref, i_dst_stride, i_ref0, i_ref1, i_ref2, i_ref3 );
INSERT_W4_SB( i_src0, i_src1, i_src2, i_src3, src );
INSERT_W4_SB( i_ref0, i_ref1, i_ref2, i_ref3, ref );
ILVRL_B2_UB( src, ref, inp0, inp1 );
HSUB_UB2_SH( inp0, inp1, diff0, diff2 );
diff1 = ( v8i16 ) __msa_ilvl_d( ( v2i64 ) diff0, ( v2i64 ) diff0 );
diff3 = ( v8i16 ) __msa_ilvl_d( ( v2i64 ) diff2, ( v2i64 ) diff2 );
BUTTERFLY_4( diff0, diff1, diff2, diff3, temp0, temp1, temp2, temp3 );
diff0 = temp0 + temp1;
diff1 = ( temp3 << 1 ) + temp2;
diff2 = temp0 - temp1;
diff3 = temp3 - ( temp2 << 1 );
TRANSPOSE4x4_SH_SH( diff0, diff1, diff2, diff3,
temp0, temp1, temp2, temp3 );
BUTTERFLY_4( temp0, temp1, temp2, temp3, diff0, diff1, diff2, diff3 );
temp0 = diff0 + diff1;
temp1 = ( diff3 << 1 ) + diff2;
temp2 = diff0 - diff1;
temp3 = diff3 - ( diff2 << 1 );
ILVR_D2_UB( temp1, temp0, temp3, temp2, inp0, inp1 );
ST_UB2( inp0, inp1, p_dst, 8 );
}
static void mbloop_filter_horizontal_edge_y_msa(uint8_t *src, int32_t pitch,
const uint8_t b_limit_in,
const uint8_t limit_in,
const uint8_t thresh_in) {
uint8_t *temp_src;
v16u8 p3, p2, p1, p0, q3, q2, q1, q0;
v16u8 mask, hev, flat, thresh, limit, b_limit;
b_limit = (v16u8)__msa_fill_b(b_limit_in);
limit = (v16u8)__msa_fill_b(limit_in);
thresh = (v16u8)__msa_fill_b(thresh_in);
temp_src = src - (pitch << 2);
LD_UB8(temp_src, pitch, p3, p2, p1, p0, q0, q1, q2, q3);
LPF_MASK_HEV(p3, p2, p1, p0, q0, q1, q2, q3, limit, b_limit, thresh, hev,
mask, flat);
VP8_MBFILTER(p2, p1, p0, q0, q1, q2, mask, hev);
temp_src = src - 3 * pitch;
ST_UB4(p2, p1, p0, q0, temp_src, pitch);
temp_src += (4 * pitch);
ST_UB2(q1, q2, temp_src, pitch);
}
void ff_vp8_v_loop_filter16_msa(uint8_t *src, ptrdiff_t pitch, int b_limit_in,
int limit_in, int thresh_in)
{
uint8_t *temp_src;
v16u8 p3, p2, p1, p0, q3, q2, q1, q0;
v16u8 mask, hev, flat, thresh, limit, b_limit;
b_limit = (v16u8) __msa_fill_b(b_limit_in);
limit = (v16u8) __msa_fill_b(limit_in);
thresh = (v16u8) __msa_fill_b(thresh_in);
/* load vector elements */
temp_src = src - (pitch << 2);
LD_UB8(temp_src, pitch, p3, p2, p1, p0, q0, q1, q2, q3);
LPF_MASK_HEV(p3, p2, p1, p0, q0, q1, q2, q3, limit, b_limit, thresh,
hev, mask, flat);
VP8_MBFILTER(p2, p1, p0, q0, q1, q2, mask, hev);
/* store vector elements */
temp_src = src - 3 * pitch;
ST_UB4(p2, p1, p0, q0, temp_src, pitch);
temp_src += (4 * pitch);
ST_UB2(q1, q2, temp_src, pitch);
}
void
yuv_abgr_convert_msa (JSAMPROW p_in_y, JSAMPROW p_in_cb, JSAMPROW p_in_cr,
JSAMPROW p_rgb, JDIMENSION out_width)
{
int y, cb, cr;
unsigned int col, num_cols_mul_16 = out_width >> 4;
unsigned int remaining_wd = out_width & 0xF;
v16i8 alpha = __msa_ldi_b(0xFF);
v16i8 const_128 = __msa_ldi_b(128);
v16u8 out0, out1, out2, out3, input_y = {0};
v16i8 input_cb, input_cr, out_rgb0, out_rgb1, out_ab0, out_ab1;
v8i16 y_h0, y_h1, cb_h0, cb_h1, cr_h0, cr_h1;
v4i32 cb_w0, cb_w1, cb_w2, cb_w3, cr_w0, cr_w1, cr_w2, cr_w3, zero = {0};
v16i8 out_r0, out_g0, out_b0;
for (col = num_cols_mul_16; col--;) {
input_y = LD_UB(p_in_y);
input_cb = LD_SB(p_in_cb);
input_cr = LD_SB(p_in_cr);
p_in_y += 16;
p_in_cb += 16;
p_in_cr += 16;
input_cb -= const_128;
input_cr -= const_128;
UNPCK_UB_SH(input_y, y_h0, y_h1);
UNPCK_SB_SH(input_cb, cb_h0, cb_h1);
UNPCK_SB_SH(input_cr, cr_h0, cr_h1);
CALC_G4_FRM_YUV(y_h0, y_h1, cb_h0, cb_h1, cr_h0, cr_h1, out_g0);
UNPCK_SH_SW(cr_h0, cr_w0, cr_w1);
UNPCK_SH_SW(cr_h1, cr_w2, cr_w3);
CALC_R4_FRM_YUV(y_h0, y_h1, cr_w0, cr_w1, cr_w2, cr_w3, out_r0);
UNPCK_SH_SW(cb_h0, cb_w0, cb_w1);
UNPCK_SH_SW(cb_h1, cb_w2, cb_w3);
CALC_B4_FRM_YUV(y_h0, y_h1, cb_w0, cb_w1, cb_w2, cb_w3, out_b0);
ILVRL_B2_SB(out_r0, out_g0, out_rgb0, out_rgb1);
ILVRL_B2_SB(out_b0, alpha, out_ab0, out_ab1);
ILVRL_H2_UB(out_rgb0, out_ab0, out0, out1);
ILVRL_H2_UB(out_rgb1, out_ab1, out2, out3);
ST_UB4(out0, out1, out2, out3, p_rgb, 16);
p_rgb += 16 * 4;
}
if (remaining_wd >= 8) {
uint64_t in_y, in_cb, in_cr;
v16i8 input_cbcr = {0};
in_y = LD(p_in_y);
in_cb = LD(p_in_cb);
in_cr = LD(p_in_cr);
p_in_y += 8;
p_in_cb += 8;
p_in_cr += 8;
input_y = (v16u8) __msa_insert_d((v2i64) input_y, 0, in_y);
input_cbcr = (v16i8) __msa_insert_d((v2i64) input_cbcr, 0, in_cb);
input_cbcr = (v16i8) __msa_insert_d((v2i64) input_cbcr, 1, in_cr);
input_cbcr -= const_128;
y_h0 = (v8i16) __msa_ilvr_b((v16i8) zero, (v16i8) input_y);
UNPCK_SB_SH(input_cbcr, cb_h0, cr_h0);
UNPCK_SH_SW(cb_h0, cb_w0, cb_w1);
UNPCK_SH_SW(cr_h0, cr_w0, cr_w1);
CALC_R2_FRM_YUV(y_h0, cr_w0, cr_w1, out_r0);
CALC_G2_FRM_YUV(y_h0, cb_h0, cr_h0, out_g0);
CALC_B2_FRM_YUV(y_h0, cb_w0, cb_w1, out_b0);
out_rgb0 = (v16i8) __msa_ilvr_b((v16i8) out_r0, (v16i8) out_g0);
out_ab0 = (v16i8) __msa_ilvr_b((v16i8) out_b0, alpha);
ILVRL_H2_UB(out_rgb0, out_ab0, out0, out1);
ST_UB2(out0, out1, p_rgb, 16);
p_rgb += 16 * 2;
remaining_wd -= 8;
}
for (col = 0; col < remaining_wd; col++) {
y = (int) (p_in_y[col]);
cb = (int) (p_in_cb[col]) - 128;
cr = (int) (p_in_cr[col]) - 128;
p_rgb[0] = 0xFF;
p_rgb[1] = clip_pixel(y + ROUND_POWER_OF_TWO(FIX_1_77200 * cb, 16));
p_rgb[2] = clip_pixel(y + ROUND_POWER_OF_TWO(((-FIX_0_34414) * cb -
FIX_0_71414 * cr), 16));
p_rgb[3] = clip_pixel(y + ROUND_POWER_OF_TWO(FIX_1_40200 * cr, 16));
p_rgb += 4;
}
}
void vpx_lpf_horizontal_8_dual_msa(uint8_t *src, int32_t pitch,
const uint8_t *b_limit0,
const uint8_t *limit0,
const uint8_t *thresh0,
const uint8_t *b_limit1,
const uint8_t *limit1,
const uint8_t *thresh1) {
v16u8 p3, p2, p1, p0, q3, q2, q1, q0;
v16u8 p2_out, p1_out, p0_out, q0_out, q1_out, q2_out;
v16u8 flat, mask, hev, tmp, thresh, b_limit, limit;
v8u16 p3_r, p2_r, p1_r, p0_r, q0_r, q1_r, q2_r, q3_r;
v8u16 p3_l, p2_l, p1_l, p0_l, q0_l, q1_l, q2_l, q3_l;
v8i16 p2_filt8_r, p1_filt8_r, p0_filt8_r, q0_filt8_r, q1_filt8_r, q2_filt8_r;
v8i16 p2_filt8_l, p1_filt8_l, p0_filt8_l, q0_filt8_l, q1_filt8_l, q2_filt8_l;
v16u8 zero = { 0 };
/* load vector elements */
LD_UB8(src - (4 * pitch), pitch, p3, p2, p1, p0, q0, q1, q2, q3);
thresh = (v16u8)__msa_fill_b(*thresh0);
tmp = (v16u8)__msa_fill_b(*thresh1);
thresh = (v16u8)__msa_ilvr_d((v2i64)tmp, (v2i64)thresh);
b_limit = (v16u8)__msa_fill_b(*b_limit0);
tmp = (v16u8)__msa_fill_b(*b_limit1);
b_limit = (v16u8)__msa_ilvr_d((v2i64)tmp, (v2i64)b_limit);
limit = (v16u8)__msa_fill_b(*limit0);
tmp = (v16u8)__msa_fill_b(*limit1);
limit = (v16u8)__msa_ilvr_d((v2i64)tmp, (v2i64)limit);
/* mask and hev */
LPF_MASK_HEV(p3, p2, p1, p0, q0, q1, q2, q3, limit, b_limit, thresh,
hev, mask, flat);
VP9_FLAT4(p3, p2, p0, q0, q2, q3, flat);
VP9_LPF_FILTER4_4W(p1, p0, q0, q1, mask, hev, p1_out, p0_out, q0_out, q1_out);
if (__msa_test_bz_v(flat)) {
ST_UB4(p1_out, p0_out, q0_out, q1_out, (src - 2 * pitch), pitch);
} else {
ILVR_B8_UH(zero, p3, zero, p2, zero, p1, zero, p0, zero, q0, zero, q1,
zero, q2, zero, q3, p3_r, p2_r, p1_r, p0_r, q0_r, q1_r,
q2_r, q3_r);
VP9_FILTER8(p3_r, p2_r, p1_r, p0_r, q0_r, q1_r, q2_r, q3_r, p2_filt8_r,
p1_filt8_r, p0_filt8_r, q0_filt8_r, q1_filt8_r, q2_filt8_r);
ILVL_B4_UH(zero, p3, zero, p2, zero, p1, zero, p0, p3_l, p2_l, p1_l, p0_l);
ILVL_B4_UH(zero, q0, zero, q1, zero, q2, zero, q3, q0_l, q1_l, q2_l, q3_l);
VP9_FILTER8(p3_l, p2_l, p1_l, p0_l, q0_l, q1_l, q2_l, q3_l, p2_filt8_l,
p1_filt8_l, p0_filt8_l, q0_filt8_l, q1_filt8_l, q2_filt8_l);
/* convert 16 bit output data into 8 bit */
PCKEV_B4_SH(p2_filt8_l, p2_filt8_r, p1_filt8_l, p1_filt8_r, p0_filt8_l,
p0_filt8_r, q0_filt8_l, q0_filt8_r, p2_filt8_r, p1_filt8_r,
p0_filt8_r, q0_filt8_r);
PCKEV_B2_SH(q1_filt8_l, q1_filt8_r, q2_filt8_l, q2_filt8_r, q1_filt8_r,
q2_filt8_r);
/* store pixel values */
p2_out = __msa_bmnz_v(p2, (v16u8)p2_filt8_r, flat);
p1_out = __msa_bmnz_v(p1_out, (v16u8)p1_filt8_r, flat);
p0_out = __msa_bmnz_v(p0_out, (v16u8)p0_filt8_r, flat);
q0_out = __msa_bmnz_v(q0_out, (v16u8)q0_filt8_r, flat);
q1_out = __msa_bmnz_v(q1_out, (v16u8)q1_filt8_r, flat);
q2_out = __msa_bmnz_v(q2, (v16u8)q2_filt8_r, flat);
src -= 3 * pitch;
ST_UB4(p2_out, p1_out, p0_out, q0_out, src, pitch);
src += (4 * pitch);
ST_UB2(q1_out, q2_out, src, pitch);
src += (2 * pitch);
}
}
static void hevc_addblk_32x32_msa(int16_t *coeffs, uint8_t *dst, int32_t stride)
{
uint8_t loop_cnt;
uint8_t *temp_dst = dst;
v16u8 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7;
v8i16 dst_r0, dst_l0, dst_r1, dst_l1, dst_r2, dst_l2, dst_r3, dst_l3;
v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
/* Pre-load for next iteration */
LD_UB2(temp_dst, 16, dst4, dst5);
temp_dst += stride;
LD_UB2(temp_dst, 16, dst6, dst7);
temp_dst += stride;
LD_SH4(coeffs, 16, in0, in2, in4, in6);
LD_SH4((coeffs + 8), 16, in1, in3, in5, in7);
coeffs += 64;
for (loop_cnt = 14; loop_cnt--;) {
UNPCK_UB_SH(dst4, dst_r0, dst_l0);
UNPCK_UB_SH(dst5, dst_r1, dst_l1);
UNPCK_UB_SH(dst6, dst_r2, dst_l2);
UNPCK_UB_SH(dst7, dst_r3, dst_l3);
dst_r0 += in0;
dst_l0 += in1;
dst_r1 += in2;
dst_l1 += in3;
dst_r2 += in4;
dst_l2 += in5;
dst_r3 += in6;
dst_l3 += in7;
/* Pre-load for next iteration */
LD_UB2(temp_dst, 16, dst4, dst5);
temp_dst += stride;
LD_UB2(temp_dst, 16, dst6, dst7);
temp_dst += stride;
LD_SH4(coeffs, 16, in0, in2, in4, in6);
LD_SH4((coeffs + 8), 16, in1, in3, in5, in7);
coeffs += 64;
CLIP_SH4_0_255(dst_r0, dst_l0, dst_r1, dst_l1);
CLIP_SH4_0_255(dst_r2, dst_l2, dst_r3, dst_l3);
PCKEV_B4_UB(dst_l0, dst_r0, dst_l1, dst_r1, dst_l2, dst_r2, dst_l3,
dst_r3, dst0, dst1, dst2, dst3);
ST_UB2(dst0, dst1, dst, 16);
dst += stride;
ST_UB2(dst2, dst3, dst, 16);
dst += stride;
}
UNPCK_UB_SH(dst4, dst_r0, dst_l0);
UNPCK_UB_SH(dst5, dst_r1, dst_l1);
UNPCK_UB_SH(dst6, dst_r2, dst_l2);
UNPCK_UB_SH(dst7, dst_r3, dst_l3);
dst_r0 += in0;
dst_l0 += in1;
dst_r1 += in2;
dst_l1 += in3;
dst_r2 += in4;
dst_l2 += in5;
dst_r3 += in6;
dst_l3 += in7;
/* Pre-load for next iteration */
LD_UB2(temp_dst, 16, dst4, dst5);
temp_dst += stride;
LD_UB2(temp_dst, 16, dst6, dst7);
temp_dst += stride;
LD_SH4(coeffs, 16, in0, in2, in4, in6);
LD_SH4((coeffs + 8), 16, in1, in3, in5, in7);
CLIP_SH4_0_255(dst_r0, dst_l0, dst_r1, dst_l1);
CLIP_SH4_0_255(dst_r2, dst_l2, dst_r3, dst_l3);
PCKEV_B4_UB(dst_l0, dst_r0, dst_l1, dst_r1, dst_l2, dst_r2, dst_l3,
dst_r3, dst0, dst1, dst2, dst3);
ST_UB2(dst0, dst1, dst, 16);
dst += stride;
ST_UB2(dst2, dst3, dst, 16);
dst += stride;
UNPCK_UB_SH(dst4, dst_r0, dst_l0);
UNPCK_UB_SH(dst5, dst_r1, dst_l1);
UNPCK_UB_SH(dst6, dst_r2, dst_l2);
UNPCK_UB_SH(dst7, dst_r3, dst_l3);
dst_r0 += in0;
dst_l0 += in1;
dst_r1 += in2;
dst_l1 += in3;
dst_r2 += in4;
dst_l2 += in5;
dst_r3 += in6;
dst_l3 += in7;
CLIP_SH4_0_255(dst_r0, dst_l0, dst_r1, dst_l1);
CLIP_SH4_0_255(dst_r2, dst_l2, dst_r3, dst_l3);
PCKEV_B4_UB(dst_l0, dst_r0, dst_l1, dst_r1, dst_l2, dst_r2, dst_l3,
dst_r3, dst0, dst1, dst2, dst3);
ST_UB2(dst0, dst1, dst, 16);
dst += stride;
ST_UB2(dst2, dst3, dst, 16);
}