105   const int16x8_t tmp3 = vcombine_s16(vshrn_n_s32(tmp2_lo, 16),  \
122     int16x8_t U, V;  in ConvertRGBA32ToUV_NEON()
142     int16x8_t U_tmp, V_tmp;  in ConvertARGBToUV_NEON()
186   const int16x8_t zero = vdupq_n_s16(0);  in SharpYUVUpdateY_NEON()
187   const int16x8_t max = vdupq_n_s16(MAX_Y);  in SharpYUVUpdateY_NEON()
192     const int16x8_t A = vreinterpretq_s16_u16(vld1q_u16(ref + i));  in SharpYUVUpdateY_NEON()
193     const int16x8_t B = vreinterpretq_s16_u16(vld1q_u16(src + i));  in SharpYUVUpdateY_NEON()
194     const int16x8_t C = vreinterpretq_s16_u16(vld1q_u16(dst + i));  in SharpYUVUpdateY_NEON()
195     const int16x8_t D = vsubq_s16(A, B);       // diff_y  in SharpYUVUpdateY_NEON()
196     const int16x8_t F = vaddq_s16(C, D);       // new_y  in SharpYUVUpdateY_NEON()
199     const int16x8_t I = vabsq_s16(D);          // abs(diff_y)  in SharpYUVUpdateY_NEON()
217     const int16x8_t A = vld1q_s16(ref + i);  in SharpYUVUpdateRGB_NEON()
218     const int16x8_t B = vld1q_s16(src + i);  in SharpYUVUpdateRGB_NEON()
219     const int16x8_t C = vld1q_s16(dst + i);  in SharpYUVUpdateRGB_NEON()
220     const int16x8_t D = vsubq_s16(A, B);   // diff_uv  in SharpYUVUpdateRGB_NEON()
221     const int16x8_t E = vaddq_s16(C, D);   // new_uv  in SharpYUVUpdateRGB_NEON()
233   const int16x8_t max = vdupq_n_s16(MAX_Y);  in SharpYUVFilterRow_NEON()
234   const int16x8_t zero = vdupq_n_s16(0);  in SharpYUVFilterRow_NEON()
236     const int16x8_t a0 = vld1q_s16(A + i + 0);  in SharpYUVFilterRow_NEON()
237     const int16x8_t a1 = vld1q_s16(A + i + 1);  in SharpYUVFilterRow_NEON()
238     const int16x8_t b0 = vld1q_s16(B + i + 0);  in SharpYUVFilterRow_NEON()
239     const int16x8_t b1 = vld1q_s16(B + i + 1);  in SharpYUVFilterRow_NEON()
240     const int16x8_t a0b1 = vaddq_s16(a0, b1);  in SharpYUVFilterRow_NEON()
241     const int16x8_t a1b0 = vaddq_s16(a1, b0);  in SharpYUVFilterRow_NEON()
242     const int16x8_t a0a1b0b1 = vaddq_s16(a0b1, a1b0);  // A0+A1+B0+B1  in SharpYUVFilterRow_NEON()
243     const int16x8_t a0b1_2 = vaddq_s16(a0b1, a0b1);    // 2*(A0+B1)  in SharpYUVFilterRow_NEON()
244     const int16x8_t a1b0_2 = vaddq_s16(a1b0, a1b0);    // 2*(A1+B0)  in SharpYUVFilterRow_NEON()
245     const int16x8_t c0 = vshrq_n_s16(vaddq_s16(a0b1_2, a0a1b0b1), 3);  in SharpYUVFilterRow_NEON()
246     const int16x8_t c1 = vshrq_n_s16(vaddq_s16(a1b0_2, a0a1b0b1), 3);  in SharpYUVFilterRow_NEON()
247     const int16x8_t d0 = vaddq_s16(c1, a0);  in SharpYUVFilterRow_NEON()
248     const int16x8_t d1 = vaddq_s16(c0, a1);  in SharpYUVFilterRow_NEON()
249     const int16x8_t e0 = vrshrq_n_s16(d0, 1);  in SharpYUVFilterRow_NEON()
250     const int16x8_t e1 = vrshrq_n_s16(d1, 1);  in SharpYUVFilterRow_NEON()
252     const int16x8_t g0 = vreinterpretq_s16_u16(vld1q_u16(best_y + 2 * i + 0));  in SharpYUVFilterRow_NEON()
253     const int16x8_t g1 = vreinterpretq_s16_u16(vld1q_u16(best_y + 2 * i + 8));  in SharpYUVFilterRow_NEON()
254     const int16x8_t h0 = vaddq_s16(g0, f.val[0]);  in SharpYUVFilterRow_NEON()
255     const int16x8_t h1 = vaddq_s16(g1, f.val[1]);  in SharpYUVFilterRow_NEON()
256     const int16x8_t i0 = vmaxq_s16(vminq_s16(h0, max), zero);  in SharpYUVFilterRow_NEON()
257     const int16x8_t i1 = vmaxq_s16(vminq_s16(h1, max), zero);  in SharpYUVFilterRow_NEON()