306     const int32x4_t kCst937 = vdupq_n_s32(937);  in FTransform()
307     const int32x4_t kCst1812 = vdupq_n_s32(1812);  in FTransform()
315     const int32x4_t a3_2217 = vmull_n_s16(vget_low_s16(a3a2), 2217);  in FTransform()
316     const int32x4_t a2_2217 = vmull_n_s16(vget_high_s16(a3a2), 2217);  in FTransform()
317     const int32x4_t a2_p_a3 = vmlal_n_s16(a2_2217, vget_low_s16(a3a2), 5352);  in FTransform()
318     const int32x4_t a3_m_a2 = vmlsl_n_s16(a3_2217, vget_high_s16(a3a2), 5352);  in FTransform()
325     const int32x4_t kCst12000 = vdupq_n_s32(12000 + (1 << 16));  in FTransform()
326     const int32x4_t kCst51000 = vdupq_n_s32(51000);  in FTransform()
332     const int32x4_t a3_2217 = vmull_n_s16(vget_low_s16(a3a2), 2217);  in FTransform()
333     const int32x4_t a2_2217 = vmull_n_s16(vget_high_s16(a3a2), 2217);  in FTransform()
334     const int32x4_t a2_p_a3 = vmlal_n_s16(a2_2217, vget_low_s16(a3a2), 5352);  in FTransform()
335     const int32x4_t a3_m_a2 = vmlsl_n_s16(a3_2217, vget_high_s16(a3a2), 5352);  in FTransform()
509     const int32x4_t a0 = vaddl_s16(in.val[0], in.val[2]);  in FTransformWHT()
510     const int32x4_t a1 = vaddl_s16(in.val[1], in.val[3]);  in FTransformWHT()
511     const int32x4_t a2 = vsubl_s16(in.val[1], in.val[3]);  in FTransformWHT()
512     const int32x4_t a3 = vsubl_s16(in.val[0], in.val[2]);  in FTransformWHT()
524     const int32x4_t a0 = vaddq_s32(tmp1.val[0], tmp1.val[2]);  in FTransformWHT()
525     const int32x4_t a1 = vaddq_s32(tmp1.val[1], tmp1.val[3]);  in FTransformWHT()
526     const int32x4_t a2 = vsubq_s32(tmp1.val[1], tmp1.val[3]);  in FTransformWHT()
527     const int32x4_t a3 = vsubq_s32(tmp1.val[0], tmp1.val[2]);  in FTransformWHT()
528     const int32x4_t b0 = vhaddq_s32(a0, a1);  // (a0 + a1) >> 1  in FTransformWHT()
529     const int32x4_t b1 = vhaddq_s32(a3, a2);  // (a3 + a2) >> 1  in FTransformWHT()
530     const int32x4_t b2 = vhsubq_s32(a3, a2);  // (a3 - a2) >> 1  in FTransformWHT()
531     const int32x4_t b3 = vhsubq_s32(a0, a1);  // (a0 - a1) >> 1  in FTransformWHT()
648   int32x4_t q_sum0 = vmull_s16(d4_w.val[0], vget_low_s16(q4_in.val[0]));  in DistoSum()
649   int32x4_t q_sum1 = vmull_s16(d4_w.val[1], vget_low_s16(q4_in.val[1]));  in DistoSum()
650   int32x4_t q_sum2 = vmull_s16(d4_w.val[2], vget_low_s16(q4_in.val[2]));  in DistoSum()
651   int32x4_t q_sum3 = vmull_s16(d4_w.val[3], vget_low_s16(q4_in.val[3]));  in DistoSum()