38 int16x8_t q0s16, q2s16, q3s16, q8s16, q10s16, q11s16, q13s16;
39 int16x8_t q14s16, q15s16, qzs16;
40 int16x4_t d0s16, d2s16, d3s16, dzs16;
41 uint16x8_t q1u16, q9u16;
44 dzs16 = vdup_n_s16(0);
45 qzs16 = vdupq_n_s16(0);
47 q15s16 = vdupq_n_s16(qscale << 1);
48 q14s16 = vdupq_n_s16(qadd);
49 q13s16 = vnegq_s16(q14s16);
52 for (; nCoeffs > 8; nCoeffs -= 16, block += 16) {
53 q0s16 = vld1q_s16(block);
54 q3s16 = vreinterpretq_s16_u16(vcltq_s16(q0s16, qzs16));
55 q8s16 = vld1q_s16(block + 8);
56 q1u16 = vceqq_s16(q0s16, qzs16);
57 q2s16 = vmulq_s16(q0s16, q15s16);
58 q11s16 = vreinterpretq_s16_u16(vcltq_s16(q8s16, qzs16));
59 q10s16 = vmulq_s16(q8s16, q15s16);
60 q3s16 = vbslq_s16(vreinterpretq_u16_s16(q3s16), q13s16, q14s16);
61 q11s16 = vbslq_s16(vreinterpretq_u16_s16(q11s16), q13s16, q14s16);
62 q2s16 = vaddq_s16(q2s16, q3s16);
63 q9u16 = vceqq_s16(q8s16, qzs16);
64 q10s16 = vaddq_s16(q10s16, q11s16);
65 q0s16 = vbslq_s16(q1u16, q0s16, q2s16);
66 q8s16 = vbslq_s16(q9u16, q8s16, q10s16);
67 vst1q_s16(block, q0s16);
68 vst1q_s16(block + 8, q8s16);
74 d0s16 = vld1_s16(block);
75 d3s16 = vreinterpret_s16_u16(vclt_s16(d0s16, dzs16));
76 d1u16 = vceq_s16(d0s16, dzs16);
77 d2s16 = vmul_s16(d0s16, vget_high_s16(q15s16));
78 d3s16 = vbsl_s16(vreinterpret_u16_s16(d3s16),
79 vget_high_s16(q13s16), vget_high_s16(q14s16));
80 d2s16 = vadd_s16(d2s16, d3s16);
81 d0s16 = vbsl_s16(d1u16, d0s16, d2s16);
82 vst1_s16(block, d0s16);
89 int qadd = (qscale - 1) | 1;
105 qadd = (qscale - 1) | 1;