static void ConvertARGBToUV_NEON(const uint32_t* argb, uint8_t* u, uint8_t* v,
int src_width, int do_store) {
int i;
for (i = 0; i + 16 <= src_width; i += 16, u += 8, v += 8) {
const uint8x16x4_t RGB = vld4q_u8((const uint8_t*)&argb[i]);
const uint16x8_t R = vpaddlq_u8(RGB.val[2]); // pair-wise adds
const uint16x8_t G = vpaddlq_u8(RGB.val[1]);
const uint16x8_t B = vpaddlq_u8(RGB.val[0]);
int16x8_t U_tmp, V_tmp;
CONVERT_RGB_TO_UV(R, G, B, 1, U_tmp, V_tmp);
{
const uint8x8_t U = vqrshrun_n_s16(U_tmp, 1);
const uint8x8_t V = vqrshrun_n_s16(V_tmp, 1);
if (do_store) {
vst1_u8(u, U);
vst1_u8(v, V);
} else {
const uint8x8_t prev_u = vld1_u8(u);
const uint8x8_t prev_v = vld1_u8(v);
vst1_u8(u, vrhadd_u8(U, prev_u));
vst1_u8(v, vrhadd_u8(V, prev_v));
}
}
}
if (i < src_width) { // left-over
WebPConvertARGBToUV_C(argb + i, u, v, src_width - i, do_store);
}
}
static void ConvertRGBA32ToUV_NEON(const uint16_t* rgb,
uint8_t* u, uint8_t* v, int width) {
int i;
for (i = 0; i + 8 <= width; i += 8, rgb += 4 * 8) {
const uint16x8x4_t RGB = vld4q_u16((const uint16_t*)rgb);
int16x8_t U, V;
CONVERT_RGB_TO_UV(RGB.val[0], RGB.val[1], RGB.val[2], 2, U, V);
vst1_u8(u + i, vqrshrun_n_s16(U, 2));
vst1_u8(v + i, vqrshrun_n_s16(V, 2));
}
for (; i < width; i += 1, rgb += 4) {
const int r = rgb[0], g = rgb[1], b = rgb[2];
u[i] = VP8RGBToU(r, g, b, YUV_HALF << 2);
v[i] = VP8RGBToV(r, g, b, YUV_HALF << 2);
}
}
void test_vqRshrun_ns16 (void)
{
uint8x8_t out_uint8x8_t;
int16x8_t arg0_int16x8_t;
out_uint8x8_t = vqrshrun_n_s16 (arg0_int16x8_t, 1);
}
static INLINE void scaledconvolve_vert_w4(
const uint8_t *src, const ptrdiff_t src_stride, uint8_t *dst,
const ptrdiff_t dst_stride, const InterpKernel *const y_filters,
const int y0_q4, const int y_step_q4, const int w, const int h) {
int y;
int y_q4 = y0_q4;
src -= src_stride * (SUBPEL_TAPS / 2 - 1);
y = h;
do {
const unsigned char *src_y = &src[(y_q4 >> SUBPEL_BITS) * src_stride];
if (y_q4 & SUBPEL_MASK) {
const int16x8_t filters = vld1q_s16(y_filters[y_q4 & SUBPEL_MASK]);
const int16x4_t filter3 = vdup_lane_s16(vget_low_s16(filters), 3);
const int16x4_t filter4 = vdup_lane_s16(vget_high_s16(filters), 0);
uint8x8_t s[8], d;
int16x4_t t[8], tt;
load_u8_8x8(src_y, src_stride, &s[0], &s[1], &s[2], &s[3], &s[4], &s[5],
&s[6], &s[7]);
t[0] = vget_low_s16(vreinterpretq_s16_u16(vmovl_u8(s[0])));
t[1] = vget_low_s16(vreinterpretq_s16_u16(vmovl_u8(s[1])));
t[2] = vget_low_s16(vreinterpretq_s16_u16(vmovl_u8(s[2])));
t[3] = vget_low_s16(vreinterpretq_s16_u16(vmovl_u8(s[3])));
t[4] = vget_low_s16(vreinterpretq_s16_u16(vmovl_u8(s[4])));
t[5] = vget_low_s16(vreinterpretq_s16_u16(vmovl_u8(s[5])));
t[6] = vget_low_s16(vreinterpretq_s16_u16(vmovl_u8(s[6])));
t[7] = vget_low_s16(vreinterpretq_s16_u16(vmovl_u8(s[7])));
tt = convolve8_4(t[0], t[1], t[2], t[3], t[4], t[5], t[6], t[7], filters,
filter3, filter4);
d = vqrshrun_n_s16(vcombine_s16(tt, tt), 7);
vst1_lane_u32((uint32_t *)dst, vreinterpret_u32_u8(d), 0);
} else {
memcpy(dst, &src_y[3 * src_stride], w);
}
dst += dst_stride;
y_q4 += y_step_q4;
} while (--y);
}
static INLINE void scaledconvolve_horiz_w4(
const uint8_t *src, const ptrdiff_t src_stride, uint8_t *dst,
const ptrdiff_t dst_stride, const InterpKernel *const x_filters,
const int x0_q4, const int x_step_q4, const int w, const int h) {
DECLARE_ALIGNED(16, uint8_t, temp[4 * 4]);
int x, y, z;
src -= SUBPEL_TAPS / 2 - 1;
y = h;
do {
int x_q4 = x0_q4;
x = 0;
do {
// process 4 src_x steps
for (z = 0; z < 4; ++z) {
const uint8_t *const src_x = &src[x_q4 >> SUBPEL_BITS];
if (x_q4 & SUBPEL_MASK) {
const int16x8_t filters = vld1q_s16(x_filters[x_q4 & SUBPEL_MASK]);
const int16x4_t filter3 = vdup_lane_s16(vget_low_s16(filters), 3);
const int16x4_t filter4 = vdup_lane_s16(vget_high_s16(filters), 0);
uint8x8_t s[8], d;
int16x8_t ss[4];
int16x4_t t[8], tt;
load_u8_8x4(src_x, src_stride, &s[0], &s[1], &s[2], &s[3]);
transpose_u8_8x4(&s[0], &s[1], &s[2], &s[3]);
ss[0] = vreinterpretq_s16_u16(vmovl_u8(s[0]));
ss[1] = vreinterpretq_s16_u16(vmovl_u8(s[1]));
ss[2] = vreinterpretq_s16_u16(vmovl_u8(s[2]));
ss[3] = vreinterpretq_s16_u16(vmovl_u8(s[3]));
t[0] = vget_low_s16(ss[0]);
t[1] = vget_low_s16(ss[1]);
t[2] = vget_low_s16(ss[2]);
t[3] = vget_low_s16(ss[3]);
t[4] = vget_high_s16(ss[0]);
t[5] = vget_high_s16(ss[1]);
t[6] = vget_high_s16(ss[2]);
t[7] = vget_high_s16(ss[3]);
tt = convolve8_4(t[0], t[1], t[2], t[3], t[4], t[5], t[6], t[7],
filters, filter3, filter4);
d = vqrshrun_n_s16(vcombine_s16(tt, tt), 7);
vst1_lane_u32((uint32_t *)&temp[4 * z], vreinterpret_u32_u8(d), 0);
} else {
int i;
for (i = 0; i < 4; ++i) {
temp[z * 4 + i] = src_x[i * src_stride + 3];
}
}
x_q4 += x_step_q4;
}
// transpose the 4x4 filters values back to dst
{
const uint8x8x4_t d4 = vld4_u8(temp);
vst1_lane_u32((uint32_t *)&dst[x + 0 * dst_stride],
vreinterpret_u32_u8(d4.val[0]), 0);
vst1_lane_u32((uint32_t *)&dst[x + 1 * dst_stride],
vreinterpret_u32_u8(d4.val[1]), 0);
vst1_lane_u32((uint32_t *)&dst[x + 2 * dst_stride],
vreinterpret_u32_u8(d4.val[2]), 0);
vst1_lane_u32((uint32_t *)&dst[x + 3 * dst_stride],
vreinterpret_u32_u8(d4.val[3]), 0);
}
x += 4;
} while (x < w);
src += src_stride * 4;
dst += dst_stride * 4;
y -= 4;
} while (y > 0);
}
void vp8_sixtap_predict16x16_neon(
unsigned char *src_ptr,
int src_pixels_per_line,
int xoffset,
int yoffset,
unsigned char *dst_ptr,
int dst_pitch) {
unsigned char *src, *src_tmp, *dst, *tmpp;
unsigned char tmp[336];
int i, j;
uint8x8_t d0u8, d1u8, d2u8, d3u8, d4u8, d5u8, d6u8, d7u8, d8u8, d9u8;
uint8x8_t d10u8, d11u8, d12u8, d13u8, d14u8, d15u8, d18u8, d19u8;
uint8x8_t d20u8, d21u8, d22u8, d23u8, d24u8, d25u8, d26u8, d27u8;
uint8x8_t d28u8, d29u8, d30u8, d31u8;
int8x8_t dtmps8, d0s8, d1s8, d2s8, d3s8, d4s8, d5s8;
uint8x16_t q3u8, q4u8;
uint16x8_t q3u16, q4u16, q5u16, q6u16, q7u16, q8u16, q9u16, q10u16;
uint16x8_t q11u16, q12u16, q13u16, q15u16;
int16x8_t q3s16, q4s16, q5s16, q6s16, q7s16, q8s16, q9s16, q10s16;
int16x8_t q11s16, q12s16, q13s16, q15s16;
if (xoffset == 0) { // secondpass_filter8x8_only
// load second_pass filter
dtmps8 = vld1_s8(vp8_sub_pel_filters[yoffset]);
d0s8 = vdup_lane_s8(dtmps8, 0);
d1s8 = vdup_lane_s8(dtmps8, 1);
d2s8 = vdup_lane_s8(dtmps8, 2);
d3s8 = vdup_lane_s8(dtmps8, 3);
d4s8 = vdup_lane_s8(dtmps8, 4);
d5s8 = vdup_lane_s8(dtmps8, 5);
d0u8 = vreinterpret_u8_s8(vabs_s8(d0s8));
d1u8 = vreinterpret_u8_s8(vabs_s8(d1s8));
d2u8 = vreinterpret_u8_s8(vabs_s8(d2s8));
d3u8 = vreinterpret_u8_s8(vabs_s8(d3s8));
d4u8 = vreinterpret_u8_s8(vabs_s8(d4s8));
d5u8 = vreinterpret_u8_s8(vabs_s8(d5s8));
// load src data
src_tmp = src_ptr - src_pixels_per_line * 2;
for (i = 0; i < 2; i++) {
src = src_tmp + i * 8;
dst = dst_ptr + i * 8;
d18u8 = vld1_u8(src);
src += src_pixels_per_line;
d19u8 = vld1_u8(src);
src += src_pixels_per_line;
d20u8 = vld1_u8(src);
src += src_pixels_per_line;
d21u8 = vld1_u8(src);
src += src_pixels_per_line;
d22u8 = vld1_u8(src);
src += src_pixels_per_line;
for (j = 0; j < 4; j++) {
d23u8 = vld1_u8(src);
src += src_pixels_per_line;
d24u8 = vld1_u8(src);
src += src_pixels_per_line;
d25u8 = vld1_u8(src);
src += src_pixels_per_line;
d26u8 = vld1_u8(src);
src += src_pixels_per_line;
q3u16 = vmull_u8(d18u8, d0u8);
q4u16 = vmull_u8(d19u8, d0u8);
q5u16 = vmull_u8(d20u8, d0u8);
q6u16 = vmull_u8(d21u8, d0u8);
q3u16 = vmlsl_u8(q3u16, d19u8, d1u8);
q4u16 = vmlsl_u8(q4u16, d20u8, d1u8);
q5u16 = vmlsl_u8(q5u16, d21u8, d1u8);
q6u16 = vmlsl_u8(q6u16, d22u8, d1u8);
q3u16 = vmlsl_u8(q3u16, d22u8, d4u8);
q4u16 = vmlsl_u8(q4u16, d23u8, d4u8);
q5u16 = vmlsl_u8(q5u16, d24u8, d4u8);
q6u16 = vmlsl_u8(q6u16, d25u8, d4u8);
q3u16 = vmlal_u8(q3u16, d20u8, d2u8);
q4u16 = vmlal_u8(q4u16, d21u8, d2u8);
q5u16 = vmlal_u8(q5u16, d22u8, d2u8);
q6u16 = vmlal_u8(q6u16, d23u8, d2u8);
q3u16 = vmlal_u8(q3u16, d23u8, d5u8);
q4u16 = vmlal_u8(q4u16, d24u8, d5u8);
q5u16 = vmlal_u8(q5u16, d25u8, d5u8);
q6u16 = vmlal_u8(q6u16, d26u8, d5u8);
q7u16 = vmull_u8(d21u8, d3u8);
q8u16 = vmull_u8(d22u8, d3u8);
q9u16 = vmull_u8(d23u8, d3u8);
q10u16 = vmull_u8(d24u8, d3u8);
q3s16 = vreinterpretq_s16_u16(q3u16);
q4s16 = vreinterpretq_s16_u16(q4u16);
q5s16 = vreinterpretq_s16_u16(q5u16);
q6s16 = vreinterpretq_s16_u16(q6u16);
q7s16 = vreinterpretq_s16_u16(q7u16);
q8s16 = vreinterpretq_s16_u16(q8u16);
q9s16 = vreinterpretq_s16_u16(q9u16);
q10s16 = vreinterpretq_s16_u16(q10u16);
q7s16 = vqaddq_s16(q7s16, q3s16);
q8s16 = vqaddq_s16(q8s16, q4s16);
q9s16 = vqaddq_s16(q9s16, q5s16);
q10s16 = vqaddq_s16(q10s16, q6s16);
d6u8 = vqrshrun_n_s16(q7s16, 7);
d7u8 = vqrshrun_n_s16(q8s16, 7);
d8u8 = vqrshrun_n_s16(q9s16, 7);
d9u8 = vqrshrun_n_s16(q10s16, 7);
d18u8 = d22u8;
d19u8 = d23u8;
d20u8 = d24u8;
d21u8 = d25u8;
d22u8 = d26u8;
vst1_u8(dst, d6u8);
dst += dst_pitch;
vst1_u8(dst, d7u8);
dst += dst_pitch;
vst1_u8(dst, d8u8);
dst += dst_pitch;
vst1_u8(dst, d9u8);
dst += dst_pitch;
}
}
return;
}
// load first_pass filter
dtmps8 = vld1_s8(vp8_sub_pel_filters[xoffset]);
d0s8 = vdup_lane_s8(dtmps8, 0);
d1s8 = vdup_lane_s8(dtmps8, 1);
d2s8 = vdup_lane_s8(dtmps8, 2);
d3s8 = vdup_lane_s8(dtmps8, 3);
d4s8 = vdup_lane_s8(dtmps8, 4);
d5s8 = vdup_lane_s8(dtmps8, 5);
d0u8 = vreinterpret_u8_s8(vabs_s8(d0s8));
d1u8 = vreinterpret_u8_s8(vabs_s8(d1s8));
d2u8 = vreinterpret_u8_s8(vabs_s8(d2s8));
d3u8 = vreinterpret_u8_s8(vabs_s8(d3s8));
d4u8 = vreinterpret_u8_s8(vabs_s8(d4s8));
d5u8 = vreinterpret_u8_s8(vabs_s8(d5s8));
// First pass: output_height lines x output_width columns (9x4)
if (yoffset == 0) { // firstpass_filter4x4_only
src = src_ptr - 2;
dst = dst_ptr;
for (i = 0; i < 8; i++) {
d6u8 = vld1_u8(src);
d7u8 = vld1_u8(src + 8);
d8u8 = vld1_u8(src + 16);
src += src_pixels_per_line;
d9u8 = vld1_u8(src);
d10u8 = vld1_u8(src + 8);
d11u8 = vld1_u8(src + 16);
src += src_pixels_per_line;
__builtin_prefetch(src);
__builtin_prefetch(src + src_pixels_per_line);
q6u16 = vmull_u8(d6u8, d0u8);
q7u16 = vmull_u8(d7u8, d0u8);
q8u16 = vmull_u8(d9u8, d0u8);
q9u16 = vmull_u8(d10u8, d0u8);
d20u8 = vext_u8(d6u8, d7u8, 1);
d21u8 = vext_u8(d9u8, d10u8, 1);
d22u8 = vext_u8(d7u8, d8u8, 1);
d23u8 = vext_u8(d10u8, d11u8, 1);
d24u8 = vext_u8(d6u8, d7u8, 4);
d25u8 = vext_u8(d9u8, d10u8, 4);
d26u8 = vext_u8(d7u8, d8u8, 4);
d27u8 = vext_u8(d10u8, d11u8, 4);
d28u8 = vext_u8(d6u8, d7u8, 5);
d29u8 = vext_u8(d9u8, d10u8, 5);
q6u16 = vmlsl_u8(q6u16, d20u8, d1u8);
q8u16 = vmlsl_u8(q8u16, d21u8, d1u8);
q7u16 = vmlsl_u8(q7u16, d22u8, d1u8);
q9u16 = vmlsl_u8(q9u16, d23u8, d1u8);
q6u16 = vmlsl_u8(q6u16, d24u8, d4u8);
q8u16 = vmlsl_u8(q8u16, d25u8, d4u8);
q7u16 = vmlsl_u8(q7u16, d26u8, d4u8);
q9u16 = vmlsl_u8(q9u16, d27u8, d4u8);
q6u16 = vmlal_u8(q6u16, d28u8, d5u8);
q8u16 = vmlal_u8(q8u16, d29u8, d5u8);
d20u8 = vext_u8(d7u8, d8u8, 5);
d21u8 = vext_u8(d10u8, d11u8, 5);
d22u8 = vext_u8(d6u8, d7u8, 2);
d23u8 = vext_u8(d9u8, d10u8, 2);
d24u8 = vext_u8(d7u8, d8u8, 2);
d25u8 = vext_u8(d10u8, d11u8, 2);
d26u8 = vext_u8(d6u8, d7u8, 3);
d27u8 = vext_u8(d9u8, d10u8, 3);
d28u8 = vext_u8(d7u8, d8u8, 3);
d29u8 = vext_u8(d10u8, d11u8, 3);
q7u16 = vmlal_u8(q7u16, d20u8, d5u8);
q9u16 = vmlal_u8(q9u16, d21u8, d5u8);
q6u16 = vmlal_u8(q6u16, d22u8, d2u8);
q8u16 = vmlal_u8(q8u16, d23u8, d2u8);
q7u16 = vmlal_u8(q7u16, d24u8, d2u8);
q9u16 = vmlal_u8(q9u16, d25u8, d2u8);
q10u16 = vmull_u8(d26u8, d3u8);
q11u16 = vmull_u8(d27u8, d3u8);
q12u16 = vmull_u8(d28u8, d3u8);
q15u16 = vmull_u8(d29u8, d3u8);
q6s16 = vreinterpretq_s16_u16(q6u16);
q7s16 = vreinterpretq_s16_u16(q7u16);
q8s16 = vreinterpretq_s16_u16(q8u16);
q9s16 = vreinterpretq_s16_u16(q9u16);
q10s16 = vreinterpretq_s16_u16(q10u16);
q11s16 = vreinterpretq_s16_u16(q11u16);
q12s16 = vreinterpretq_s16_u16(q12u16);
q15s16 = vreinterpretq_s16_u16(q15u16);
q6s16 = vqaddq_s16(q6s16, q10s16);
q8s16 = vqaddq_s16(q8s16, q11s16);
q7s16 = vqaddq_s16(q7s16, q12s16);
q9s16 = vqaddq_s16(q9s16, q15s16);
d6u8 = vqrshrun_n_s16(q6s16, 7);
d7u8 = vqrshrun_n_s16(q7s16, 7);
d8u8 = vqrshrun_n_s16(q8s16, 7);
d9u8 = vqrshrun_n_s16(q9s16, 7);
q3u8 = vcombine_u8(d6u8, d7u8);
q4u8 = vcombine_u8(d8u8, d9u8);
vst1q_u8(dst, q3u8);
dst += dst_pitch;
vst1q_u8(dst, q4u8);
dst += dst_pitch;
}
return;
}
src = src_ptr - 2 - src_pixels_per_line * 2;
tmpp = tmp;
for (i = 0; i < 7; i++) {
d6u8 = vld1_u8(src);
d7u8 = vld1_u8(src + 8);
d8u8 = vld1_u8(src + 16);
src += src_pixels_per_line;
d9u8 = vld1_u8(src);
d10u8 = vld1_u8(src + 8);
d11u8 = vld1_u8(src + 16);
src += src_pixels_per_line;
d12u8 = vld1_u8(src);
d13u8 = vld1_u8(src + 8);
d14u8 = vld1_u8(src + 16);
src += src_pixels_per_line;
__builtin_prefetch(src);
__builtin_prefetch(src + src_pixels_per_line);
__builtin_prefetch(src + src_pixels_per_line * 2);
q8u16 = vmull_u8(d6u8, d0u8);
q9u16 = vmull_u8(d7u8, d0u8);
q10u16 = vmull_u8(d9u8, d0u8);
q11u16 = vmull_u8(d10u8, d0u8);
q12u16 = vmull_u8(d12u8, d0u8);
q13u16 = vmull_u8(d13u8, d0u8);
d28u8 = vext_u8(d6u8, d7u8, 1);
d29u8 = vext_u8(d9u8, d10u8, 1);
d30u8 = vext_u8(d12u8, d13u8, 1);
q8u16 = vmlsl_u8(q8u16, d28u8, d1u8);
q10u16 = vmlsl_u8(q10u16, d29u8, d1u8);
q12u16 = vmlsl_u8(q12u16, d30u8, d1u8);
d28u8 = vext_u8(d7u8, d8u8, 1);
d29u8 = vext_u8(d10u8, d11u8, 1);
d30u8 = vext_u8(d13u8, d14u8, 1);
q9u16 = vmlsl_u8(q9u16, d28u8, d1u8);
q11u16 = vmlsl_u8(q11u16, d29u8, d1u8);
q13u16 = vmlsl_u8(q13u16, d30u8, d1u8);
d28u8 = vext_u8(d6u8, d7u8, 4);
d29u8 = vext_u8(d9u8, d10u8, 4);
d30u8 = vext_u8(d12u8, d13u8, 4);
q8u16 = vmlsl_u8(q8u16, d28u8, d4u8);
q10u16 = vmlsl_u8(q10u16, d29u8, d4u8);
q12u16 = vmlsl_u8(q12u16, d30u8, d4u8);
d28u8 = vext_u8(d7u8, d8u8, 4);
d29u8 = vext_u8(d10u8, d11u8, 4);
d30u8 = vext_u8(d13u8, d14u8, 4);
q9u16 = vmlsl_u8(q9u16, d28u8, d4u8);
q11u16 = vmlsl_u8(q11u16, d29u8, d4u8);
q13u16 = vmlsl_u8(q13u16, d30u8, d4u8);
d28u8 = vext_u8(d6u8, d7u8, 5);
d29u8 = vext_u8(d9u8, d10u8, 5);
d30u8 = vext_u8(d12u8, d13u8, 5);
q8u16 = vmlal_u8(q8u16, d28u8, d5u8);
q10u16 = vmlal_u8(q10u16, d29u8, d5u8);
q12u16 = vmlal_u8(q12u16, d30u8, d5u8);
d28u8 = vext_u8(d7u8, d8u8, 5);
d29u8 = vext_u8(d10u8, d11u8, 5);
d30u8 = vext_u8(d13u8, d14u8, 5);
q9u16 = vmlal_u8(q9u16, d28u8, d5u8);
q11u16 = vmlal_u8(q11u16, d29u8, d5u8);
q13u16 = vmlal_u8(q13u16, d30u8, d5u8);
d28u8 = vext_u8(d6u8, d7u8, 2);
d29u8 = vext_u8(d9u8, d10u8, 2);
d30u8 = vext_u8(d12u8, d13u8, 2);
q8u16 = vmlal_u8(q8u16, d28u8, d2u8);
q10u16 = vmlal_u8(q10u16, d29u8, d2u8);
q12u16 = vmlal_u8(q12u16, d30u8, d2u8);
d28u8 = vext_u8(d7u8, d8u8, 2);
d29u8 = vext_u8(d10u8, d11u8, 2);
d30u8 = vext_u8(d13u8, d14u8, 2);
q9u16 = vmlal_u8(q9u16, d28u8, d2u8);
q11u16 = vmlal_u8(q11u16, d29u8, d2u8);
q13u16 = vmlal_u8(q13u16, d30u8, d2u8);
d28u8 = vext_u8(d6u8, d7u8, 3);
d29u8 = vext_u8(d9u8, d10u8, 3);
d30u8 = vext_u8(d12u8, d13u8, 3);
d15u8 = vext_u8(d7u8, d8u8, 3);
d31u8 = vext_u8(d10u8, d11u8, 3);
d6u8 = vext_u8(d13u8, d14u8, 3);
q4u16 = vmull_u8(d28u8, d3u8);
q5u16 = vmull_u8(d29u8, d3u8);
q6u16 = vmull_u8(d30u8, d3u8);
q4s16 = vreinterpretq_s16_u16(q4u16);
q5s16 = vreinterpretq_s16_u16(q5u16);
q6s16 = vreinterpretq_s16_u16(q6u16);
q8s16 = vreinterpretq_s16_u16(q8u16);
q10s16 = vreinterpretq_s16_u16(q10u16);
q12s16 = vreinterpretq_s16_u16(q12u16);
q8s16 = vqaddq_s16(q8s16, q4s16);
q10s16 = vqaddq_s16(q10s16, q5s16);
q12s16 = vqaddq_s16(q12s16, q6s16);
q6u16 = vmull_u8(d15u8, d3u8);
q7u16 = vmull_u8(d31u8, d3u8);
q3u16 = vmull_u8(d6u8, d3u8);
q3s16 = vreinterpretq_s16_u16(q3u16);
q6s16 = vreinterpretq_s16_u16(q6u16);
q7s16 = vreinterpretq_s16_u16(q7u16);
q9s16 = vreinterpretq_s16_u16(q9u16);
q11s16 = vreinterpretq_s16_u16(q11u16);
q13s16 = vreinterpretq_s16_u16(q13u16);
q9s16 = vqaddq_s16(q9s16, q6s16);
q11s16 = vqaddq_s16(q11s16, q7s16);
q13s16 = vqaddq_s16(q13s16, q3s16);
d6u8 = vqrshrun_n_s16(q8s16, 7);
d7u8 = vqrshrun_n_s16(q9s16, 7);
d8u8 = vqrshrun_n_s16(q10s16, 7);
d9u8 = vqrshrun_n_s16(q11s16, 7);
d10u8 = vqrshrun_n_s16(q12s16, 7);
d11u8 = vqrshrun_n_s16(q13s16, 7);
vst1_u8(tmpp, d6u8);
tmpp += 8;
vst1_u8(tmpp, d7u8);
tmpp += 8;
vst1_u8(tmpp, d8u8);
tmpp += 8;
vst1_u8(tmpp, d9u8);
tmpp += 8;
vst1_u8(tmpp, d10u8);
tmpp += 8;
vst1_u8(tmpp, d11u8);
tmpp += 8;
}
// Second pass: 16x16
dtmps8 = vld1_s8(vp8_sub_pel_filters[yoffset]);
d0s8 = vdup_lane_s8(dtmps8, 0);
d1s8 = vdup_lane_s8(dtmps8, 1);
d2s8 = vdup_lane_s8(dtmps8, 2);
d3s8 = vdup_lane_s8(dtmps8, 3);
d4s8 = vdup_lane_s8(dtmps8, 4);
d5s8 = vdup_lane_s8(dtmps8, 5);
d0u8 = vreinterpret_u8_s8(vabs_s8(d0s8));
d1u8 = vreinterpret_u8_s8(vabs_s8(d1s8));
d2u8 = vreinterpret_u8_s8(vabs_s8(d2s8));
d3u8 = vreinterpret_u8_s8(vabs_s8(d3s8));
d4u8 = vreinterpret_u8_s8(vabs_s8(d4s8));
d5u8 = vreinterpret_u8_s8(vabs_s8(d5s8));
for (i = 0; i < 2; i++) {
dst = dst_ptr + 8 * i;
tmpp = tmp + 8 * i;
d18u8 = vld1_u8(tmpp);
tmpp += 16;
d19u8 = vld1_u8(tmpp);
tmpp += 16;
d20u8 = vld1_u8(tmpp);
tmpp += 16;
d21u8 = vld1_u8(tmpp);
tmpp += 16;
d22u8 = vld1_u8(tmpp);
tmpp += 16;
for (j = 0; j < 4; j++) {
d23u8 = vld1_u8(tmpp);
tmpp += 16;
d24u8 = vld1_u8(tmpp);
tmpp += 16;
d25u8 = vld1_u8(tmpp);
tmpp += 16;
d26u8 = vld1_u8(tmpp);
tmpp += 16;
q3u16 = vmull_u8(d18u8, d0u8);
q4u16 = vmull_u8(d19u8, d0u8);
q5u16 = vmull_u8(d20u8, d0u8);
q6u16 = vmull_u8(d21u8, d0u8);
q3u16 = vmlsl_u8(q3u16, d19u8, d1u8);
q4u16 = vmlsl_u8(q4u16, d20u8, d1u8);
q5u16 = vmlsl_u8(q5u16, d21u8, d1u8);
q6u16 = vmlsl_u8(q6u16, d22u8, d1u8);
q3u16 = vmlsl_u8(q3u16, d22u8, d4u8);
q4u16 = vmlsl_u8(q4u16, d23u8, d4u8);
q5u16 = vmlsl_u8(q5u16, d24u8, d4u8);
q6u16 = vmlsl_u8(q6u16, d25u8, d4u8);
q3u16 = vmlal_u8(q3u16, d20u8, d2u8);
q4u16 = vmlal_u8(q4u16, d21u8, d2u8);
q5u16 = vmlal_u8(q5u16, d22u8, d2u8);
q6u16 = vmlal_u8(q6u16, d23u8, d2u8);
q3u16 = vmlal_u8(q3u16, d23u8, d5u8);
q4u16 = vmlal_u8(q4u16, d24u8, d5u8);
q5u16 = vmlal_u8(q5u16, d25u8, d5u8);
q6u16 = vmlal_u8(q6u16, d26u8, d5u8);
q7u16 = vmull_u8(d21u8, d3u8);
q8u16 = vmull_u8(d22u8, d3u8);
q9u16 = vmull_u8(d23u8, d3u8);
q10u16 = vmull_u8(d24u8, d3u8);
q3s16 = vreinterpretq_s16_u16(q3u16);
q4s16 = vreinterpretq_s16_u16(q4u16);
q5s16 = vreinterpretq_s16_u16(q5u16);
q6s16 = vreinterpretq_s16_u16(q6u16);
q7s16 = vreinterpretq_s16_u16(q7u16);
q8s16 = vreinterpretq_s16_u16(q8u16);
q9s16 = vreinterpretq_s16_u16(q9u16);
q10s16 = vreinterpretq_s16_u16(q10u16);
q7s16 = vqaddq_s16(q7s16, q3s16);
q8s16 = vqaddq_s16(q8s16, q4s16);
q9s16 = vqaddq_s16(q9s16, q5s16);
q10s16 = vqaddq_s16(q10s16, q6s16);
d6u8 = vqrshrun_n_s16(q7s16, 7);
d7u8 = vqrshrun_n_s16(q8s16, 7);
d8u8 = vqrshrun_n_s16(q9s16, 7);
d9u8 = vqrshrun_n_s16(q10s16, 7);
d18u8 = d22u8;
d19u8 = d23u8;
d20u8 = d24u8;
d21u8 = d25u8;
d22u8 = d26u8;
vst1_u8(dst, d6u8);
dst += dst_pitch;
vst1_u8(dst, d7u8);
dst += dst_pitch;
vst1_u8(dst, d8u8);
dst += dst_pitch;
vst1_u8(dst, d9u8);
dst += dst_pitch;
}
}
return;
}
void vp8_sixtap_predict8x8_neon(
unsigned char *src_ptr,
int src_pixels_per_line,
int xoffset,
int yoffset,
unsigned char *dst_ptr,
int dst_pitch) {
unsigned char *src, *tmpp;
unsigned char tmp[64];
int i;
uint8x8_t d0u8, d1u8, d2u8, d3u8, d4u8, d5u8, d6u8, d7u8, d8u8, d9u8;
uint8x8_t d18u8, d19u8, d20u8, d21u8, d22u8, d23u8, d24u8, d25u8;
uint8x8_t d26u8, d27u8, d28u8, d29u8, d30u8, d31u8;
int8x8_t dtmps8, d0s8, d1s8, d2s8, d3s8, d4s8, d5s8;
uint16x8_t q3u16, q4u16, q5u16, q6u16, q7u16;
uint16x8_t q8u16, q9u16, q10u16, q11u16, q12u16;
int16x8_t q3s16, q4s16, q5s16, q6s16, q7s16;
int16x8_t q8s16, q9s16, q10s16, q11s16, q12s16;
uint8x16_t q3u8, q4u8, q5u8, q6u8, q7u8, q9u8, q10u8, q11u8, q12u8;
if (xoffset == 0) { // secondpass_filter8x8_only
// load second_pass filter
dtmps8 = vld1_s8(vp8_sub_pel_filters[yoffset]);
d0s8 = vdup_lane_s8(dtmps8, 0);
d1s8 = vdup_lane_s8(dtmps8, 1);
d2s8 = vdup_lane_s8(dtmps8, 2);
d3s8 = vdup_lane_s8(dtmps8, 3);
d4s8 = vdup_lane_s8(dtmps8, 4);
d5s8 = vdup_lane_s8(dtmps8, 5);
d0u8 = vreinterpret_u8_s8(vabs_s8(d0s8));
d1u8 = vreinterpret_u8_s8(vabs_s8(d1s8));
d2u8 = vreinterpret_u8_s8(vabs_s8(d2s8));
d3u8 = vreinterpret_u8_s8(vabs_s8(d3s8));
d4u8 = vreinterpret_u8_s8(vabs_s8(d4s8));
d5u8 = vreinterpret_u8_s8(vabs_s8(d5s8));
// load src data
src = src_ptr - src_pixels_per_line * 2;
d18u8 = vld1_u8(src);
src += src_pixels_per_line;
d19u8 = vld1_u8(src);
src += src_pixels_per_line;
d20u8 = vld1_u8(src);
src += src_pixels_per_line;
d21u8 = vld1_u8(src);
src += src_pixels_per_line;
d22u8 = vld1_u8(src);
src += src_pixels_per_line;
d23u8 = vld1_u8(src);
src += src_pixels_per_line;
d24u8 = vld1_u8(src);
src += src_pixels_per_line;
d25u8 = vld1_u8(src);
src += src_pixels_per_line;
d26u8 = vld1_u8(src);
src += src_pixels_per_line;
d27u8 = vld1_u8(src);
src += src_pixels_per_line;
d28u8 = vld1_u8(src);
src += src_pixels_per_line;
d29u8 = vld1_u8(src);
src += src_pixels_per_line;
d30u8 = vld1_u8(src);
for (i = 2; i > 0; i--) {
q3u16 = vmull_u8(d18u8, d0u8);
q4u16 = vmull_u8(d19u8, d0u8);
q5u16 = vmull_u8(d20u8, d0u8);
q6u16 = vmull_u8(d21u8, d0u8);
q3u16 = vmlsl_u8(q3u16, d19u8, d1u8);
q4u16 = vmlsl_u8(q4u16, d20u8, d1u8);
q5u16 = vmlsl_u8(q5u16, d21u8, d1u8);
q6u16 = vmlsl_u8(q6u16, d22u8, d1u8);
q3u16 = vmlsl_u8(q3u16, d22u8, d4u8);
q4u16 = vmlsl_u8(q4u16, d23u8, d4u8);
q5u16 = vmlsl_u8(q5u16, d24u8, d4u8);
q6u16 = vmlsl_u8(q6u16, d25u8, d4u8);
q3u16 = vmlal_u8(q3u16, d20u8, d2u8);
q4u16 = vmlal_u8(q4u16, d21u8, d2u8);
q5u16 = vmlal_u8(q5u16, d22u8, d2u8);
q6u16 = vmlal_u8(q6u16, d23u8, d2u8);
q3u16 = vmlal_u8(q3u16, d23u8, d5u8);
q4u16 = vmlal_u8(q4u16, d24u8, d5u8);
q5u16 = vmlal_u8(q5u16, d25u8, d5u8);
q6u16 = vmlal_u8(q6u16, d26u8, d5u8);
q7u16 = vmull_u8(d21u8, d3u8);
q8u16 = vmull_u8(d22u8, d3u8);
q9u16 = vmull_u8(d23u8, d3u8);
q10u16 = vmull_u8(d24u8, d3u8);
q3s16 = vreinterpretq_s16_u16(q3u16);
q4s16 = vreinterpretq_s16_u16(q4u16);
q5s16 = vreinterpretq_s16_u16(q5u16);
q6s16 = vreinterpretq_s16_u16(q6u16);
q7s16 = vreinterpretq_s16_u16(q7u16);
q8s16 = vreinterpretq_s16_u16(q8u16);
q9s16 = vreinterpretq_s16_u16(q9u16);
q10s16 = vreinterpretq_s16_u16(q10u16);
q7s16 = vqaddq_s16(q7s16, q3s16);
q8s16 = vqaddq_s16(q8s16, q4s16);
q9s16 = vqaddq_s16(q9s16, q5s16);
q10s16 = vqaddq_s16(q10s16, q6s16);
d6u8 = vqrshrun_n_s16(q7s16, 7);
d7u8 = vqrshrun_n_s16(q8s16, 7);
d8u8 = vqrshrun_n_s16(q9s16, 7);
d9u8 = vqrshrun_n_s16(q10s16, 7);
d18u8 = d22u8;
d19u8 = d23u8;
d20u8 = d24u8;
d21u8 = d25u8;
d22u8 = d26u8;
d23u8 = d27u8;
d24u8 = d28u8;
d25u8 = d29u8;
d26u8 = d30u8;
vst1_u8(dst_ptr, d6u8);
dst_ptr += dst_pitch;
vst1_u8(dst_ptr, d7u8);
dst_ptr += dst_pitch;
vst1_u8(dst_ptr, d8u8);
dst_ptr += dst_pitch;
vst1_u8(dst_ptr, d9u8);
dst_ptr += dst_pitch;
}
return;
}
// load first_pass filter
dtmps8 = vld1_s8(vp8_sub_pel_filters[xoffset]);
d0s8 = vdup_lane_s8(dtmps8, 0);
d1s8 = vdup_lane_s8(dtmps8, 1);
d2s8 = vdup_lane_s8(dtmps8, 2);
d3s8 = vdup_lane_s8(dtmps8, 3);
d4s8 = vdup_lane_s8(dtmps8, 4);
d5s8 = vdup_lane_s8(dtmps8, 5);
d0u8 = vreinterpret_u8_s8(vabs_s8(d0s8));
d1u8 = vreinterpret_u8_s8(vabs_s8(d1s8));
d2u8 = vreinterpret_u8_s8(vabs_s8(d2s8));
d3u8 = vreinterpret_u8_s8(vabs_s8(d3s8));
d4u8 = vreinterpret_u8_s8(vabs_s8(d4s8));
d5u8 = vreinterpret_u8_s8(vabs_s8(d5s8));
// First pass: output_height lines x output_width columns (9x4)
if (yoffset == 0) // firstpass_filter4x4_only
src = src_ptr - 2;
else
src = src_ptr - 2 - (src_pixels_per_line * 2);
tmpp = tmp;
for (i = 2; i > 0; i--) {
q3u8 = vld1q_u8(src);
src += src_pixels_per_line;
q4u8 = vld1q_u8(src);
src += src_pixels_per_line;
q5u8 = vld1q_u8(src);
src += src_pixels_per_line;
q6u8 = vld1q_u8(src);
src += src_pixels_per_line;
__builtin_prefetch(src);
__builtin_prefetch(src + src_pixels_per_line);
__builtin_prefetch(src + src_pixels_per_line * 2);
q7u16 = vmull_u8(vget_low_u8(q3u8), d0u8);
q8u16 = vmull_u8(vget_low_u8(q4u8), d0u8);
q9u16 = vmull_u8(vget_low_u8(q5u8), d0u8);
q10u16 = vmull_u8(vget_low_u8(q6u8), d0u8);
d28u8 = vext_u8(vget_low_u8(q3u8), vget_high_u8(q3u8), 1);
d29u8 = vext_u8(vget_low_u8(q4u8), vget_high_u8(q4u8), 1);
d30u8 = vext_u8(vget_low_u8(q5u8), vget_high_u8(q5u8), 1);
d31u8 = vext_u8(vget_low_u8(q6u8), vget_high_u8(q6u8), 1);
q7u16 = vmlsl_u8(q7u16, d28u8, d1u8);
q8u16 = vmlsl_u8(q8u16, d29u8, d1u8);
q9u16 = vmlsl_u8(q9u16, d30u8, d1u8);
q10u16 = vmlsl_u8(q10u16, d31u8, d1u8);
d28u8 = vext_u8(vget_low_u8(q3u8), vget_high_u8(q3u8), 4);
d29u8 = vext_u8(vget_low_u8(q4u8), vget_high_u8(q4u8), 4);
d30u8 = vext_u8(vget_low_u8(q5u8), vget_high_u8(q5u8), 4);
d31u8 = vext_u8(vget_low_u8(q6u8), vget_high_u8(q6u8), 4);
q7u16 = vmlsl_u8(q7u16, d28u8, d4u8);
q8u16 = vmlsl_u8(q8u16, d29u8, d4u8);
q9u16 = vmlsl_u8(q9u16, d30u8, d4u8);
q10u16 = vmlsl_u8(q10u16, d31u8, d4u8);
d28u8 = vext_u8(vget_low_u8(q3u8), vget_high_u8(q3u8), 2);
d29u8 = vext_u8(vget_low_u8(q4u8), vget_high_u8(q4u8), 2);
d30u8 = vext_u8(vget_low_u8(q5u8), vget_high_u8(q5u8), 2);
d31u8 = vext_u8(vget_low_u8(q6u8), vget_high_u8(q6u8), 2);
q7u16 = vmlal_u8(q7u16, d28u8, d2u8);
q8u16 = vmlal_u8(q8u16, d29u8, d2u8);
q9u16 = vmlal_u8(q9u16, d30u8, d2u8);
q10u16 = vmlal_u8(q10u16, d31u8, d2u8);
d28u8 = vext_u8(vget_low_u8(q3u8), vget_high_u8(q3u8), 5);
d29u8 = vext_u8(vget_low_u8(q4u8), vget_high_u8(q4u8), 5);
d30u8 = vext_u8(vget_low_u8(q5u8), vget_high_u8(q5u8), 5);
d31u8 = vext_u8(vget_low_u8(q6u8), vget_high_u8(q6u8), 5);
q7u16 = vmlal_u8(q7u16, d28u8, d5u8);
q8u16 = vmlal_u8(q8u16, d29u8, d5u8);
q9u16 = vmlal_u8(q9u16, d30u8, d5u8);
q10u16 = vmlal_u8(q10u16, d31u8, d5u8);
d28u8 = vext_u8(vget_low_u8(q3u8), vget_high_u8(q3u8), 3);
d29u8 = vext_u8(vget_low_u8(q4u8), vget_high_u8(q4u8), 3);
d30u8 = vext_u8(vget_low_u8(q5u8), vget_high_u8(q5u8), 3);
d31u8 = vext_u8(vget_low_u8(q6u8), vget_high_u8(q6u8), 3);
q3u16 = vmull_u8(d28u8, d3u8);
q4u16 = vmull_u8(d29u8, d3u8);
q5u16 = vmull_u8(d30u8, d3u8);
q6u16 = vmull_u8(d31u8, d3u8);
q3s16 = vreinterpretq_s16_u16(q3u16);
q4s16 = vreinterpretq_s16_u16(q4u16);
q5s16 = vreinterpretq_s16_u16(q5u16);
q6s16 = vreinterpretq_s16_u16(q6u16);
q7s16 = vreinterpretq_s16_u16(q7u16);
q8s16 = vreinterpretq_s16_u16(q8u16);
q9s16 = vreinterpretq_s16_u16(q9u16);
q10s16 = vreinterpretq_s16_u16(q10u16);
q7s16 = vqaddq_s16(q7s16, q3s16);
q8s16 = vqaddq_s16(q8s16, q4s16);
q9s16 = vqaddq_s16(q9s16, q5s16);
q10s16 = vqaddq_s16(q10s16, q6s16);
d22u8 = vqrshrun_n_s16(q7s16, 7);
d23u8 = vqrshrun_n_s16(q8s16, 7);
d24u8 = vqrshrun_n_s16(q9s16, 7);
d25u8 = vqrshrun_n_s16(q10s16, 7);
if (yoffset == 0) { // firstpass_filter8x4_only
vst1_u8(dst_ptr, d22u8);
dst_ptr += dst_pitch;
vst1_u8(dst_ptr, d23u8);
dst_ptr += dst_pitch;
vst1_u8(dst_ptr, d24u8);
dst_ptr += dst_pitch;
vst1_u8(dst_ptr, d25u8);
dst_ptr += dst_pitch;
} else {
vst1_u8(tmpp, d22u8);
tmpp += 8;
vst1_u8(tmpp, d23u8);
tmpp += 8;
vst1_u8(tmpp, d24u8);
tmpp += 8;
vst1_u8(tmpp, d25u8);
tmpp += 8;
}
}
if (yoffset == 0)
return;
// First Pass on rest 5-line data
q3u8 = vld1q_u8(src);
src += src_pixels_per_line;
q4u8 = vld1q_u8(src);
src += src_pixels_per_line;
q5u8 = vld1q_u8(src);
src += src_pixels_per_line;
q6u8 = vld1q_u8(src);
src += src_pixels_per_line;
q7u8 = vld1q_u8(src);
q8u16 = vmull_u8(vget_low_u8(q3u8), d0u8);
q9u16 = vmull_u8(vget_low_u8(q4u8), d0u8);
q10u16 = vmull_u8(vget_low_u8(q5u8), d0u8);
q11u16 = vmull_u8(vget_low_u8(q6u8), d0u8);
q12u16 = vmull_u8(vget_low_u8(q7u8), d0u8);
d27u8 = vext_u8(vget_low_u8(q3u8), vget_high_u8(q3u8), 1);
d28u8 = vext_u8(vget_low_u8(q4u8), vget_high_u8(q4u8), 1);
d29u8 = vext_u8(vget_low_u8(q5u8), vget_high_u8(q5u8), 1);
d30u8 = vext_u8(vget_low_u8(q6u8), vget_high_u8(q6u8), 1);
d31u8 = vext_u8(vget_low_u8(q7u8), vget_high_u8(q7u8), 1);
q8u16 = vmlsl_u8(q8u16, d27u8, d1u8);
q9u16 = vmlsl_u8(q9u16, d28u8, d1u8);
q10u16 = vmlsl_u8(q10u16, d29u8, d1u8);
q11u16 = vmlsl_u8(q11u16, d30u8, d1u8);
q12u16 = vmlsl_u8(q12u16, d31u8, d1u8);
d27u8 = vext_u8(vget_low_u8(q3u8), vget_high_u8(q3u8), 4);
d28u8 = vext_u8(vget_low_u8(q4u8), vget_high_u8(q4u8), 4);
d29u8 = vext_u8(vget_low_u8(q5u8), vget_high_u8(q5u8), 4);
d30u8 = vext_u8(vget_low_u8(q6u8), vget_high_u8(q6u8), 4);
d31u8 = vext_u8(vget_low_u8(q7u8), vget_high_u8(q7u8), 4);
q8u16 = vmlsl_u8(q8u16, d27u8, d4u8);
q9u16 = vmlsl_u8(q9u16, d28u8, d4u8);
q10u16 = vmlsl_u8(q10u16, d29u8, d4u8);
q11u16 = vmlsl_u8(q11u16, d30u8, d4u8);
q12u16 = vmlsl_u8(q12u16, d31u8, d4u8);
d27u8 = vext_u8(vget_low_u8(q3u8), vget_high_u8(q3u8), 2);
d28u8 = vext_u8(vget_low_u8(q4u8), vget_high_u8(q4u8), 2);
d29u8 = vext_u8(vget_low_u8(q5u8), vget_high_u8(q5u8), 2);
d30u8 = vext_u8(vget_low_u8(q6u8), vget_high_u8(q6u8), 2);
d31u8 = vext_u8(vget_low_u8(q7u8), vget_high_u8(q7u8), 2);
q8u16 = vmlal_u8(q8u16, d27u8, d2u8);
q9u16 = vmlal_u8(q9u16, d28u8, d2u8);
q10u16 = vmlal_u8(q10u16, d29u8, d2u8);
q11u16 = vmlal_u8(q11u16, d30u8, d2u8);
q12u16 = vmlal_u8(q12u16, d31u8, d2u8);
d27u8 = vext_u8(vget_low_u8(q3u8), vget_high_u8(q3u8), 5);
d28u8 = vext_u8(vget_low_u8(q4u8), vget_high_u8(q4u8), 5);
d29u8 = vext_u8(vget_low_u8(q5u8), vget_high_u8(q5u8), 5);
d30u8 = vext_u8(vget_low_u8(q6u8), vget_high_u8(q6u8), 5);
d31u8 = vext_u8(vget_low_u8(q7u8), vget_high_u8(q7u8), 5);
q8u16 = vmlal_u8(q8u16, d27u8, d5u8);
q9u16 = vmlal_u8(q9u16, d28u8, d5u8);
q10u16 = vmlal_u8(q10u16, d29u8, d5u8);
q11u16 = vmlal_u8(q11u16, d30u8, d5u8);
q12u16 = vmlal_u8(q12u16, d31u8, d5u8);
d27u8 = vext_u8(vget_low_u8(q3u8), vget_high_u8(q3u8), 3);
d28u8 = vext_u8(vget_low_u8(q4u8), vget_high_u8(q4u8), 3);
d29u8 = vext_u8(vget_low_u8(q5u8), vget_high_u8(q5u8), 3);
d30u8 = vext_u8(vget_low_u8(q6u8), vget_high_u8(q6u8), 3);
d31u8 = vext_u8(vget_low_u8(q7u8), vget_high_u8(q7u8), 3);
q3u16 = vmull_u8(d27u8, d3u8);
q4u16 = vmull_u8(d28u8, d3u8);
q5u16 = vmull_u8(d29u8, d3u8);
q6u16 = vmull_u8(d30u8, d3u8);
q7u16 = vmull_u8(d31u8, d3u8);
q3s16 = vreinterpretq_s16_u16(q3u16);
q4s16 = vreinterpretq_s16_u16(q4u16);
q5s16 = vreinterpretq_s16_u16(q5u16);
q6s16 = vreinterpretq_s16_u16(q6u16);
q7s16 = vreinterpretq_s16_u16(q7u16);
q8s16 = vreinterpretq_s16_u16(q8u16);
q9s16 = vreinterpretq_s16_u16(q9u16);
q10s16 = vreinterpretq_s16_u16(q10u16);
q11s16 = vreinterpretq_s16_u16(q11u16);
q12s16 = vreinterpretq_s16_u16(q12u16);
q8s16 = vqaddq_s16(q8s16, q3s16);
q9s16 = vqaddq_s16(q9s16, q4s16);
q10s16 = vqaddq_s16(q10s16, q5s16);
q11s16 = vqaddq_s16(q11s16, q6s16);
q12s16 = vqaddq_s16(q12s16, q7s16);
d26u8 = vqrshrun_n_s16(q8s16, 7);
d27u8 = vqrshrun_n_s16(q9s16, 7);
d28u8 = vqrshrun_n_s16(q10s16, 7);
d29u8 = vqrshrun_n_s16(q11s16, 7);
d30u8 = vqrshrun_n_s16(q12s16, 7);
// Second pass: 8x8
dtmps8 = vld1_s8(vp8_sub_pel_filters[yoffset]);
d0s8 = vdup_lane_s8(dtmps8, 0);
d1s8 = vdup_lane_s8(dtmps8, 1);
d2s8 = vdup_lane_s8(dtmps8, 2);
d3s8 = vdup_lane_s8(dtmps8, 3);
d4s8 = vdup_lane_s8(dtmps8, 4);
d5s8 = vdup_lane_s8(dtmps8, 5);
d0u8 = vreinterpret_u8_s8(vabs_s8(d0s8));
d1u8 = vreinterpret_u8_s8(vabs_s8(d1s8));
d2u8 = vreinterpret_u8_s8(vabs_s8(d2s8));
d3u8 = vreinterpret_u8_s8(vabs_s8(d3s8));
d4u8 = vreinterpret_u8_s8(vabs_s8(d4s8));
d5u8 = vreinterpret_u8_s8(vabs_s8(d5s8));
tmpp = tmp;
q9u8 = vld1q_u8(tmpp);
tmpp += 16;
q10u8 = vld1q_u8(tmpp);
tmpp += 16;
q11u8 = vld1q_u8(tmpp);
tmpp += 16;
q12u8 = vld1q_u8(tmpp);
d18u8 = vget_low_u8(q9u8);
d19u8 = vget_high_u8(q9u8);
d20u8 = vget_low_u8(q10u8);
d21u8 = vget_high_u8(q10u8);
d22u8 = vget_low_u8(q11u8);
d23u8 = vget_high_u8(q11u8);
d24u8 = vget_low_u8(q12u8);
d25u8 = vget_high_u8(q12u8);
for (i = 2; i > 0; i--) {
q3u16 = vmull_u8(d18u8, d0u8);
q4u16 = vmull_u8(d19u8, d0u8);
q5u16 = vmull_u8(d20u8, d0u8);
q6u16 = vmull_u8(d21u8, d0u8);
q3u16 = vmlsl_u8(q3u16, d19u8, d1u8);
q4u16 = vmlsl_u8(q4u16, d20u8, d1u8);
q5u16 = vmlsl_u8(q5u16, d21u8, d1u8);
q6u16 = vmlsl_u8(q6u16, d22u8, d1u8);
q3u16 = vmlsl_u8(q3u16, d22u8, d4u8);
q4u16 = vmlsl_u8(q4u16, d23u8, d4u8);
q5u16 = vmlsl_u8(q5u16, d24u8, d4u8);
q6u16 = vmlsl_u8(q6u16, d25u8, d4u8);
q3u16 = vmlal_u8(q3u16, d20u8, d2u8);
q4u16 = vmlal_u8(q4u16, d21u8, d2u8);
q5u16 = vmlal_u8(q5u16, d22u8, d2u8);
q6u16 = vmlal_u8(q6u16, d23u8, d2u8);
q3u16 = vmlal_u8(q3u16, d23u8, d5u8);
q4u16 = vmlal_u8(q4u16, d24u8, d5u8);
q5u16 = vmlal_u8(q5u16, d25u8, d5u8);
q6u16 = vmlal_u8(q6u16, d26u8, d5u8);
q7u16 = vmull_u8(d21u8, d3u8);
q8u16 = vmull_u8(d22u8, d3u8);
q9u16 = vmull_u8(d23u8, d3u8);
q10u16 = vmull_u8(d24u8, d3u8);
q3s16 = vreinterpretq_s16_u16(q3u16);
q4s16 = vreinterpretq_s16_u16(q4u16);
q5s16 = vreinterpretq_s16_u16(q5u16);
q6s16 = vreinterpretq_s16_u16(q6u16);
q7s16 = vreinterpretq_s16_u16(q7u16);
q8s16 = vreinterpretq_s16_u16(q8u16);
q9s16 = vreinterpretq_s16_u16(q9u16);
q10s16 = vreinterpretq_s16_u16(q10u16);
q7s16 = vqaddq_s16(q7s16, q3s16);
q8s16 = vqaddq_s16(q8s16, q4s16);
q9s16 = vqaddq_s16(q9s16, q5s16);
q10s16 = vqaddq_s16(q10s16, q6s16);
d6u8 = vqrshrun_n_s16(q7s16, 7);
d7u8 = vqrshrun_n_s16(q8s16, 7);
d8u8 = vqrshrun_n_s16(q9s16, 7);
d9u8 = vqrshrun_n_s16(q10s16, 7);
d18u8 = d22u8;
d19u8 = d23u8;
d20u8 = d24u8;
d21u8 = d25u8;
d22u8 = d26u8;
d23u8 = d27u8;
d24u8 = d28u8;
d25u8 = d29u8;
d26u8 = d30u8;
vst1_u8(dst_ptr, d6u8);
dst_ptr += dst_pitch;
vst1_u8(dst_ptr, d7u8);
dst_ptr += dst_pitch;
vst1_u8(dst_ptr, d8u8);
dst_ptr += dst_pitch;
vst1_u8(dst_ptr, d9u8);
dst_ptr += dst_pitch;
}
return;
}
