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 ConvertARGBToUV_SSE41(const uint32_t* argb,
uint8_t* u, uint8_t* v,
int src_width, int do_store) {
const int max_width = src_width & ~31;
int i;
for (i = 0; i < max_width; i += 32, u += 16, v += 16) {
__m128i rgb[6], U0, V0, U1, V1;
RGB32PackedToPlanar_SSE41(&argb[i], rgb);
HorizontalAddPack_SSE41(&rgb[0], &rgb[1], &rgb[0]);
HorizontalAddPack_SSE41(&rgb[2], &rgb[3], &rgb[2]);
HorizontalAddPack_SSE41(&rgb[4], &rgb[5], &rgb[4]);
ConvertRGBToUV_SSE41(&rgb[0], &rgb[2], &rgb[4], &U0, &V0);
RGB32PackedToPlanar_SSE41(&argb[i + 16], rgb);
HorizontalAddPack_SSE41(&rgb[0], &rgb[1], &rgb[0]);
HorizontalAddPack_SSE41(&rgb[2], &rgb[3], &rgb[2]);
HorizontalAddPack_SSE41(&rgb[4], &rgb[5], &rgb[4]);
ConvertRGBToUV_SSE41(&rgb[0], &rgb[2], &rgb[4], &U1, &V1);
U0 = _mm_packus_epi16(U0, U1);
V0 = _mm_packus_epi16(V0, V1);
if (!do_store) {
const __m128i prev_u = LOAD_16(u);
const __m128i prev_v = LOAD_16(v);
U0 = _mm_avg_epu8(U0, prev_u);
V0 = _mm_avg_epu8(V0, prev_v);
}
STORE_16(U0, u);
STORE_16(V0, v);
}
if (i < src_width) { // left-over
WebPConvertARGBToUV_C(argb + i, u, v, src_width - i, do_store);
}
}
