/****************************************************************************** * * Copyright (C) 2022 The Android Open Source Project * * Licensed under the Apache License, Version 2.0 (the "License"); * you may not use this file except in compliance with the License. * You may obtain a copy of the License at: * * http://www.apache.org/licenses/LICENSE-2.0 * * Unless required by applicable law or agreed to in writing, software * distributed under the License is distributed on an "AS IS" BASIS, * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. * See the License for the specific language governing permissions and * limitations under the License. * ***************************************************************************** * Originally developed and contributed by Ittiam Systems Pvt. Ltd, Bangalore */ /** ******************************************************************************* * @file * isvcd_iquant_itrans_residual_neonintr.c * * @brief * Contains definition of functions for h264 inverse quantization inverse * transformation and resd comp * * @author * Kishore * * @par List of Functions: * - isvcd_iquant_itrans_residual_4x4_neonintr() * - isvcd_iquant_itrans_residual_8x8_neonintr() * - isvcd_iquant_itrans_residual_4x4_dc_neonintr() * - isvcd_iquant_itrans_residual_8x8_dc_neonintr() * - isvcd_iquant_itrans_residual_chroma_4x4_neonintr() * - isvcd_iquant_itrans_residual_chroma_4x4_dc_neonintr() * * @remarks * ******************************************************************************* */ /*****************************************************************************/ /* File Includes */ /*****************************************************************************/ #include #include /* User include files */ #include "ih264_typedefs.h" #include "ih264_defs.h" #include "ih264_trans_macros.h" #include "ih264_macros.h" #include "ih264_platform_macros.h" #include "ih264_trans_data.h" #include "ih264_size_defs.h" #include "ih264_structs.h" #include "isvcd_iquant_itrans_residual.h" /*****************************************************************************/ /* */ /* Function Name : isvcd_iquant_itrans_residual_4x4_neonintr */ /* */ /* Description : this function computes the resd output from the */ /* IQ+IT */ /* */ /* Inputs : */ /* Globals : none */ /* Processing : */ /* */ /* Outputs : i4_nnz */ /* Returns : none */ /* */ /* Issues : none */ /* */ /* Revision History: */ /* */ /* DD MM YYYY Author(s) Changes (Describe the changes made) */ /* 25 11 2021 Kishore creation */ /* */ /*****************************************************************************/ WORD32 isvcd_iquant_itrans_residual_4x4_neonintr(WORD16 *pi2_src, WORD16 *pi2_pred, WORD16 *pi2_out, WORD32 pred_strd, WORD32 out_strd, const UWORD16 *pu2_iscal_mat, const UWORD16 *pu2_weigh_mat, UWORD32 u4_qp_div_6, WORD16 *pi2_tmp, WORD32 iq_start_idx, WORD16 *pi2_dc_ld_addr) { int16x4x4_t src_16x4x2; int16x4x4_t iscal_16x4x2; int16x4x4_t weigh_16x4x2; WORD16 rnd_factor = (u4_qp_div_6 < 4) ? 1 << (3 - u4_qp_div_6) : 0; int32x4_t rnd_fact = vdupq_n_s32(rnd_factor); int16x4_t pred0, pred1, pred2, pred3; int16x4_t q0_16x4, q1_16x4, q2_16x4, q3_16x4; int32x4_t q0_32x4, q1_32x4, q2_32x4, q3_32x4; int16x4_t rq1_16x4, rq3_16x4; int16x4_t x0_16x4, x1_16x4, x2_16x4, x3_16x4; int16x4_t xx0_16x4, xx1_16x4, xx2_16x4, xx3_16x4; int16x4_t xx0_0_16x4, xx0_1_16x4, xx2_0_16x4, xx2_1_16x4; int32x2_t x0_32x2, x1_32x2, x2_32x2, x3_32x2; int16x4_t weigh0_16x4, weigh1_16x4, weigh2_16x4, weigh3_16x4, dup_min, dup_max; int32x4_t qp_div_6_32x4 = vdupq_n_s32(u4_qp_div_6); int16x8_t pred01_in, pred23_in; WORD32 i4_nnz; int16x4x2_t xx0_16x4_2, xx2_16x4_2; int32x2x2_t x0_32x2_2, x1_32x2_2; int16x8_t dup_val_1, dup_val_2, dup_abs; UNUSED(pi2_tmp); src_16x4x2 = vld4_s16(pi2_src); iscal_16x4x2 = vld4_s16((const int16_t *) pu2_iscal_mat); weigh_16x4x2 = vld4_s16((const int16_t *) pu2_weigh_mat); weigh0_16x4 = vmul_s16(weigh_16x4x2.val[0], iscal_16x4x2.val[0]); weigh1_16x4 = vmul_s16(weigh_16x4x2.val[1], iscal_16x4x2.val[1]); weigh2_16x4 = vmul_s16(weigh_16x4x2.val[2], iscal_16x4x2.val[2]); weigh3_16x4 = vmul_s16(weigh_16x4x2.val[3], iscal_16x4x2.val[3]); dup_min = vdup_n_s16(RSD_MIN); dup_max = vdup_n_s16(RSD_MAX); q0_32x4 = vmull_s16(weigh0_16x4, src_16x4x2.val[0]); q1_32x4 = vmull_s16(weigh1_16x4, src_16x4x2.val[1]); q2_32x4 = vmull_s16(weigh2_16x4, src_16x4x2.val[2]); q3_32x4 = vmull_s16(weigh3_16x4, src_16x4x2.val[3]); q0_32x4 = vaddq_s32(q0_32x4, rnd_fact); q1_32x4 = vaddq_s32(q1_32x4, rnd_fact); q2_32x4 = vaddq_s32(q2_32x4, rnd_fact); q3_32x4 = vaddq_s32(q3_32x4, rnd_fact); q0_32x4 = vshlq_s32(q0_32x4, qp_div_6_32x4); q1_32x4 = vshlq_s32(q1_32x4, qp_div_6_32x4); q2_32x4 = vshlq_s32(q2_32x4, qp_div_6_32x4); q3_32x4 = vshlq_s32(q3_32x4, qp_div_6_32x4); q0_16x4 = vqshrn_n_s32(q0_32x4, 4); q1_16x4 = vqshrn_n_s32(q1_32x4, 4); q2_16x4 = vqshrn_n_s32(q2_32x4, 4); q3_16x4 = vqshrn_n_s32(q3_32x4, 4); if(iq_start_idx == 1) { q0_16x4 = vset_lane_s16(pi2_dc_ld_addr[0], q0_16x4, 0); } rq1_16x4 = vshr_n_s16(q1_16x4, 1); // q1 >>1 rq3_16x4 = vshr_n_s16(q3_16x4, 1); // q3 >>1 x0_16x4 = vadd_s16(q0_16x4, q2_16x4); // x0 = q0 + q2 x1_16x4 = vsub_s16(q0_16x4, q2_16x4); // x1 = q0 - q2 x2_16x4 = vsub_s16(rq1_16x4, q3_16x4); // x2 = q1>>1 - q3 x3_16x4 = vadd_s16(q1_16x4, rq3_16x4); // x2 = q1 + q3>>1 xx0_16x4 = vadd_s16(x0_16x4, x3_16x4); // x0+x3 xx1_16x4 = vadd_s16(x1_16x4, x2_16x4); // x1+x2 xx2_16x4 = vsub_s16(x1_16x4, x2_16x4); // x1-x2 xx3_16x4 = vsub_s16(x0_16x4, x3_16x4); // x0-x3 xx0_16x4_2 = vtrn_s16(xx0_16x4, xx1_16x4); xx0_0_16x4 = xx0_16x4_2.val[0]; xx0_1_16x4 = xx0_16x4_2.val[1]; xx2_16x4_2 = vtrn_s16(xx2_16x4, xx3_16x4); xx2_0_16x4 = xx2_16x4_2.val[0]; xx2_1_16x4 = xx2_16x4_2.val[1]; x0_32x2_2 = vtrn_s32(vreinterpret_s32_s16(xx0_0_16x4), vreinterpret_s32_s16(xx2_0_16x4)); x1_32x2_2 = vtrn_s32(vreinterpret_s32_s16(xx0_1_16x4), vreinterpret_s32_s16(xx2_1_16x4)); x0_32x2 = x0_32x2_2.val[0]; x1_32x2 = x1_32x2_2.val[0]; x2_32x2 = x0_32x2_2.val[1]; x3_32x2 = x1_32x2_2.val[1]; x0_16x4 = vreinterpret_s16_s32(x0_32x2); x1_16x4 = vreinterpret_s16_s32(x1_32x2); x2_16x4 = vreinterpret_s16_s32(x2_32x2); x3_16x4 = vreinterpret_s16_s32(x3_32x2); /* vertical inverse transform */ rq1_16x4 = vshr_n_s16(x1_16x4, 1); // q1 >> 1 rq3_16x4 = vshr_n_s16(x3_16x4, 1); // q3 >> 1 xx0_16x4 = vadd_s16(x0_16x4, x2_16x4); // x0 = q0 + q2 xx1_16x4 = vsub_s16(x0_16x4, x2_16x4); // x1 = q0 - q2 xx2_16x4 = vsub_s16(rq1_16x4, x3_16x4); // x2 = q1>>1 - q3 xx3_16x4 = vadd_s16(x1_16x4, rq3_16x4); // x3 = q1 + q3>>1 x0_16x4 = vadd_s16(xx0_16x4, xx3_16x4); // imacro = x0 + x3 x1_16x4 = vadd_s16(xx1_16x4, xx2_16x4); // imacro = x1 + x2 x2_16x4 = vsub_s16(xx1_16x4, xx2_16x4); // imacro = x1 - x2 x3_16x4 = vsub_s16(xx0_16x4, xx3_16x4); // imacro = x0 - x3 x0_16x4 = vrshr_n_s16(x0_16x4, 6); x1_16x4 = vrshr_n_s16(x1_16x4, 6); x2_16x4 = vrshr_n_s16(x2_16x4, 6); x3_16x4 = vrshr_n_s16(x3_16x4, 6); pred0 = vld1_s16((int16_t *) pi2_pred); pred1 = vld1_s16((int16_t *) pi2_pred + pred_strd); pred2 = vld1_s16((int16_t *) pi2_pred + (pred_strd * 2)); pred3 = vld1_s16((int16_t *) pi2_pred + (pred_strd * 3)); x0_16x4 = vadd_s16(pred0, x0_16x4); x1_16x4 = vadd_s16(pred1, x1_16x4); x2_16x4 = vadd_s16(pred2, x2_16x4); x3_16x4 = vadd_s16(pred3, x3_16x4); x0_16x4 = vmin_s16(x0_16x4, dup_max); x0_16x4 = vmax_s16(x0_16x4, dup_min); x1_16x4 = vmin_s16(x1_16x4, dup_max); x1_16x4 = vmax_s16(x1_16x4, dup_min); x2_16x4 = vmin_s16(x2_16x4, dup_max); x2_16x4 = vmax_s16(x2_16x4, dup_min); x3_16x4 = vmin_s16(x3_16x4, dup_max); x3_16x4 = vmax_s16(x3_16x4, dup_min); pred01_in = vcombine_s16(x0_16x4, x1_16x4); pred23_in = vcombine_s16(x2_16x4, x3_16x4); dup_val_1 = vabsq_s16(pred01_in); dup_val_2 = vabsq_s16(pred23_in); dup_abs = vqaddq_s16(dup_val_1, dup_val_2); i4_nnz = dup_abs[0] || dup_abs[1] || dup_abs[2] || dup_abs[3] || dup_abs[4] || dup_abs[5] || dup_abs[6] || dup_abs[7]; vst1_s16(pi2_out, x0_16x4); vst1_s16(pi2_out + out_strd, x1_16x4); vst1_s16(pi2_out + (out_strd << 1), x2_16x4); vst1_s16(pi2_out + ((out_strd << 1) + out_strd), x3_16x4); return i4_nnz; } /*****************************************************************************/ /* */ /* Function Name : isvcd_iquant_itrans_residual_4x4_dc_neonintr */ /* */ /* Description : this function computes the resd output from the */ /* IQ+IT */ /* */ /* Inputs : */ /* Globals : none */ /* Processing : */ /* */ /* Outputs : i4_nnz */ /* Returns : none */ /* */ /* Issues : none */ /* */ /* Revision History: */ /* */ /* DD MM YYYY Author(s) Changes (Describe the changes made) */ /* 25 11 2021 Kishore creation */ /* */ /*****************************************************************************/ WORD32 isvcd_iquant_itrans_residual_4x4_dc_neonintr(WORD16 *pi2_src, WORD16 *pi2_pred, WORD16 *pi2_out, WORD32 pred_strd, WORD32 out_strd, const UWORD16 *pu2_iscal_mat, const UWORD16 *pu2_weigh_mat, UWORD32 u4_qp_div_6, WORD16 *pi2_tmp, WORD32 iq_start_idx, WORD16 *pi2_dc_ld_addr) { WORD32 i4_iq_out_temp; int16x4_t temp_0; int16x4_t pred0_in, pred1_in, pred2_in, pred3_in, dup_min, dup_max; int16x8_t pred01_in, pred23_in; WORD32 i4_nnz; int16x8_t dup_val_1, dup_val_2, dup_abs; WORD16 rnd_fact = (u4_qp_div_6 < 4) ? 1 << (3 - u4_qp_div_6) : 0; UNUSED(pi2_tmp); if(iq_start_idx == 0) { i4_iq_out_temp = pi2_src[0]; INV_QUANT(i4_iq_out_temp, pu2_iscal_mat[0], pu2_weigh_mat[0], u4_qp_div_6, rnd_fact, 4); } else { i4_iq_out_temp = pi2_dc_ld_addr[0]; } /* inv transform and residual comp */ pred0_in = vld1_s16((int16_t *) pi2_pred); pi2_pred = pi2_pred + pred_strd; pred1_in = vld1_s16((int16_t *) pi2_pred); pi2_pred = pi2_pred + pred_strd; pred2_in = vld1_s16((int16_t *) pi2_pred); pi2_pred = pi2_pred + pred_strd; pred3_in = vld1_s16((int16_t *) pi2_pred); temp_0 = vdup_n_s16((i4_iq_out_temp + 32) >> 6); dup_min = vdup_n_s16(RSD_MIN); dup_max = vdup_n_s16(RSD_MAX); pred0_in = vadd_s16(pred0_in, temp_0); pred1_in = vadd_s16(pred1_in, temp_0); pred2_in = vadd_s16(pred2_in, temp_0); pred3_in = vadd_s16(pred3_in, temp_0); pred0_in = vmin_s16(pred0_in, dup_max); pred0_in = vmax_s16(pred0_in, dup_min); pred1_in = vmin_s16(pred1_in, dup_max); pred1_in = vmax_s16(pred1_in, dup_min); pred2_in = vmin_s16(pred2_in, dup_max); pred2_in = vmax_s16(pred2_in, dup_min); pred3_in = vmin_s16(pred3_in, dup_max); pred3_in = vmax_s16(pred3_in, dup_min); pred01_in = vcombine_s16(pred0_in, pred1_in); pred23_in = vcombine_s16(pred2_in, pred3_in); dup_val_1 = vabsq_s16(pred01_in); dup_val_2 = vabsq_s16(pred23_in); dup_abs = vqaddq_s16(dup_val_1, dup_val_2); i4_nnz = dup_abs[0] || dup_abs[1] || dup_abs[2] || dup_abs[3] || dup_abs[4] || dup_abs[5] || dup_abs[6] || dup_abs[7]; vst1_s16((int16_t *) (pi2_out), pred0_in); vst1_s16((int16_t *) (pi2_out + out_strd), pred1_in); vst1_s16((int16_t *) (pi2_out + (out_strd * 2)), pred2_in); vst1_s16((int16_t *) (pi2_out + (out_strd * 3)), pred3_in); return i4_nnz; } /*****************************************************************************/ /* */ /* Function Name : isvcd_iquant_itrans_residual_chroma_4x4_neonintr */ /* */ /* Description : this function computes the resd output from the */ /* IQ+IT */ /* */ /* Inputs : */ /* Globals : none */ /* Processing : */ /* */ /* Outputs : i4_nnz */ /* Returns : none */ /* */ /* Issues : none */ /* */ /* Revision History: */ /* */ /* DD MM YYYY Author(s) Changes (Describe the changes made) */ /* 25 11 2021 Kishore creation */ /* */ /*****************************************************************************/ WORD32 isvcd_iquant_itrans_residual_chroma_4x4_neonintr( WORD16 *pi2_src, WORD16 *pi2_pred, WORD16 *pi2_out, WORD32 pred_strd, WORD32 out_strd, const UWORD16 *pu2_iscal_mat, const UWORD16 *pu2_weigh_mat, UWORD32 u4_qp_div_6, WORD16 *pi2_tmp, WORD16 *pi2_dc_src) { int16x4x4_t src_16x4x2; int16x4x4_t iscal_16x4x2; int16x4x4_t weigh_16x4x2; int16x8_t pred0, pred1, pred2, pred3; int16x4_t q0_16x4, q1_16x4, q2_16x4, q3_16x4; int32x4_t q0_32x4, q1_32x4, q2_32x4, q3_32x4; int16x4_t rq1_16x4, rq3_16x4; int16x4_t x0_16x4, x1_16x4, x2_16x4, x3_16x4; int16x4_t xx0_16x4, xx1_16x4, xx2_16x4, xx3_16x4; int16x4_t xx0_0_16x4, xx0_1_16x4, xx2_0_16x4, xx2_1_16x4; int32x2_t x0_32x2, x1_32x2, x2_32x2, x3_32x2; int16x4_t weigh0_16x4, weigh1_16x4, weigh2_16x4, weigh3_16x4; int32x4_t qp_div_6_32x4 = vdupq_n_s32(u4_qp_div_6); int16x4_t zero_16x4 = vdup_n_s16(0); int16x4_t x_16x4_low, x_16x4_high; int16x8_t x0_16x8, x1_16x8, x2_16x8, x3_16x8; int16x8_t dup_min, dup_max; WORD32 i4_nnz; int16x4x2_t xx0_16x4_2, xx2_16x4_2, x_16x4x2_t; int32x2x2_t x0_32x2_2, x1_32x2_2; int16x8_t dup_val_1, dup_val_2, dup_val_3, dup_val_4, dup_val_5, dup_abs; int16x8_t i4_out_horz_16x8_r0, i4_out_horz_16x8_r1, i4_out_horz_16x8_r2, i4_out_horz_16x8_r3; uint16x8_t chroma_mask_16x8 = vreinterpretq_u16_u32(vdupq_n_u32(0x0000ffff)); int16x8_t chroma_mask_16x8_2 = vreinterpretq_s16_s32(vdupq_n_s32(0x0000ffff)); WORD16 rnd_factor = (u4_qp_div_6 < 4) ? 1 << (3 - u4_qp_div_6) : 0; int32x4_t rnd_fact = vdupq_n_s32(rnd_factor); UNUSED(pi2_tmp); dup_min = vdupq_n_s16(RSD_MIN); dup_max = vdupq_n_s16(RSD_MAX); src_16x4x2 = vld4_s16(pi2_src); iscal_16x4x2 = vld4_s16((const int16_t *) pu2_iscal_mat); weigh_16x4x2 = vld4_s16((const int16_t *) pu2_weigh_mat); weigh0_16x4 = vmul_s16(weigh_16x4x2.val[0], iscal_16x4x2.val[0]); weigh1_16x4 = vmul_s16(weigh_16x4x2.val[1], iscal_16x4x2.val[1]); weigh2_16x4 = vmul_s16(weigh_16x4x2.val[2], iscal_16x4x2.val[2]); weigh3_16x4 = vmul_s16(weigh_16x4x2.val[3], iscal_16x4x2.val[3]); q0_32x4 = vmull_s16(weigh0_16x4, src_16x4x2.val[0]); q1_32x4 = vmull_s16(weigh1_16x4, src_16x4x2.val[1]); q2_32x4 = vmull_s16(weigh2_16x4, src_16x4x2.val[2]); q3_32x4 = vmull_s16(weigh3_16x4, src_16x4x2.val[3]); q0_32x4 = vaddq_s32(q0_32x4, rnd_fact); q1_32x4 = vaddq_s32(q1_32x4, rnd_fact); q2_32x4 = vaddq_s32(q2_32x4, rnd_fact); q3_32x4 = vaddq_s32(q3_32x4, rnd_fact); q0_32x4 = vshlq_s32(q0_32x4, qp_div_6_32x4); q1_32x4 = vshlq_s32(q1_32x4, qp_div_6_32x4); q2_32x4 = vshlq_s32(q2_32x4, qp_div_6_32x4); q3_32x4 = vshlq_s32(q3_32x4, qp_div_6_32x4); q0_16x4 = vqshrn_n_s32(q0_32x4, 4); q1_16x4 = vqshrn_n_s32(q1_32x4, 4); q2_16x4 = vqshrn_n_s32(q2_32x4, 4); q3_16x4 = vqshrn_n_s32(q3_32x4, 4); rq1_16x4 = vshr_n_s16(q1_16x4, 1); // q1 >>1 rq3_16x4 = vshr_n_s16(q3_16x4, 1); // q3 >>1 q0_16x4 = vset_lane_s16(pi2_dc_src[0], q0_16x4, 0); x0_16x4 = vadd_s16(q0_16x4, q2_16x4); // x0 = q0 + q2 x1_16x4 = vsub_s16(q0_16x4, q2_16x4); // x1 = q0 - q2 x2_16x4 = vsub_s16(rq1_16x4, q3_16x4); // x2 = q1>>1 - q3 x3_16x4 = vadd_s16(q1_16x4, rq3_16x4); // x2 = q1 + q3>>1 xx0_16x4 = vadd_s16(x0_16x4, x3_16x4); // x0+x3 xx1_16x4 = vadd_s16(x1_16x4, x2_16x4); // x1+x2 xx2_16x4 = vsub_s16(x1_16x4, x2_16x4); // x1-x2 xx3_16x4 = vsub_s16(x0_16x4, x3_16x4); // x0-x3 xx0_16x4_2 = vtrn_s16(xx0_16x4, xx1_16x4); xx0_0_16x4 = xx0_16x4_2.val[0]; xx0_1_16x4 = xx0_16x4_2.val[1]; xx2_16x4_2 = vtrn_s16(xx2_16x4, xx3_16x4); xx2_0_16x4 = xx2_16x4_2.val[0]; xx2_1_16x4 = xx2_16x4_2.val[1]; x0_32x2_2 = vtrn_s32(vreinterpret_s32_s16(xx0_0_16x4), vreinterpret_s32_s16(xx2_0_16x4)); x1_32x2_2 = vtrn_s32(vreinterpret_s32_s16(xx0_1_16x4), vreinterpret_s32_s16(xx2_1_16x4)); x0_32x2 = x0_32x2_2.val[0]; x1_32x2 = x1_32x2_2.val[0]; x2_32x2 = x0_32x2_2.val[1]; x3_32x2 = x1_32x2_2.val[1]; x0_16x4 = vreinterpret_s16_s32(x0_32x2); x1_16x4 = vreinterpret_s16_s32(x1_32x2); x2_16x4 = vreinterpret_s16_s32(x2_32x2); x3_16x4 = vreinterpret_s16_s32(x3_32x2); /* vertical inverse transform */ rq1_16x4 = vshr_n_s16(x1_16x4, 1); // q1 >> 1 rq3_16x4 = vshr_n_s16(x3_16x4, 1); // q3 >> 1 xx0_16x4 = vadd_s16(x0_16x4, x2_16x4); // x0 = q0 + q2 xx1_16x4 = vsub_s16(x0_16x4, x2_16x4); // x1 = q0 - q2 xx2_16x4 = vsub_s16(rq1_16x4, x3_16x4); // x2 = q1>>1 - q3 xx3_16x4 = vadd_s16(x1_16x4, rq3_16x4); // x3 = q1 + q3>>1 x0_16x4 = vadd_s16(xx0_16x4, xx3_16x4); // imacro = x0 + x3 x1_16x4 = vadd_s16(xx1_16x4, xx2_16x4); // imacro = x1 + x2 x2_16x4 = vsub_s16(xx1_16x4, xx2_16x4); // imacro = x1 - x2 x3_16x4 = vsub_s16(xx0_16x4, xx3_16x4); // imacro = x0 - x3 x0_16x4 = vrshr_n_s16(x0_16x4, 6); x1_16x4 = vrshr_n_s16(x1_16x4, 6); x2_16x4 = vrshr_n_s16(x2_16x4, 6); x3_16x4 = vrshr_n_s16(x3_16x4, 6); x_16x4x2_t = vzip_s16(x0_16x4, zero_16x4); x_16x4_low = x_16x4x2_t.val[0]; x_16x4_high = x_16x4x2_t.val[1]; x0_16x8 = vcombine_s16(x_16x4_low, x_16x4_high); x_16x4x2_t = vzip_s16(x1_16x4, zero_16x4); x_16x4_low = x_16x4x2_t.val[0]; x_16x4_high = x_16x4x2_t.val[1]; x1_16x8 = vcombine_s16(x_16x4_low, x_16x4_high); x_16x4x2_t = vzip_s16(x2_16x4, zero_16x4); x_16x4_low = x_16x4x2_t.val[0]; x_16x4_high = x_16x4x2_t.val[1]; x2_16x8 = vcombine_s16(x_16x4_low, x_16x4_high); x_16x4x2_t = vzip_s16(x3_16x4, zero_16x4); x_16x4_low = x_16x4x2_t.val[0]; x_16x4_high = x_16x4x2_t.val[1]; x3_16x8 = vcombine_s16(x_16x4_low, x_16x4_high); pred0 = vld1q_s16((int16_t *) pi2_pred); pred1 = vld1q_s16((int16_t *) pi2_pred + pred_strd); pred2 = vld1q_s16((int16_t *) pi2_pred + (pred_strd * 2)); pred3 = vld1q_s16((int16_t *) pi2_pred + (pred_strd * 3)); x0_16x8 = vaddq_s16(pred0, x0_16x8); x1_16x8 = vaddq_s16(pred1, x1_16x8); x2_16x8 = vaddq_s16(pred2, x2_16x8); x3_16x8 = vaddq_s16(pred3, x3_16x8); x0_16x8 = vminq_s16(x0_16x8, dup_max); x0_16x8 = vmaxq_s16(x0_16x8, dup_min); x1_16x8 = vminq_s16(x1_16x8, dup_max); x1_16x8 = vmaxq_s16(x1_16x8, dup_min); x2_16x8 = vminq_s16(x2_16x8, dup_max); x2_16x8 = vmaxq_s16(x2_16x8, dup_min); x3_16x8 = vminq_s16(x3_16x8, dup_max); x3_16x8 = vmaxq_s16(x3_16x8, dup_min); x0_16x8 = vandq_s16(x0_16x8, chroma_mask_16x8_2); x1_16x8 = vandq_s16(x1_16x8, chroma_mask_16x8_2); x2_16x8 = vandq_s16(x2_16x8, chroma_mask_16x8_2); x3_16x8 = vandq_s16(x3_16x8, chroma_mask_16x8_2); dup_val_1 = vabsq_s16(x0_16x8); dup_val_2 = vabsq_s16(x1_16x8); dup_val_3 = vabsq_s16(x2_16x8); dup_val_4 = vabsq_s16(x3_16x8); dup_val_5 = vqaddq_s16(dup_val_1, dup_val_2); dup_val_1 = vqaddq_s16(dup_val_3, dup_val_4); dup_abs = vqaddq_s16(dup_val_1, dup_val_5); i4_nnz = dup_abs[0] || dup_abs[1] || dup_abs[2] || dup_abs[3] || dup_abs[4] || dup_abs[5] || dup_abs[6] || dup_abs[7]; i4_out_horz_16x8_r0 = vld1q_s16(pi2_out); i4_out_horz_16x8_r1 = vld1q_s16(pi2_out + out_strd); i4_out_horz_16x8_r2 = vld1q_s16(pi2_out + out_strd * 2); i4_out_horz_16x8_r3 = vld1q_s16(pi2_out + out_strd * 3); i4_out_horz_16x8_r0 = vbslq_s16(chroma_mask_16x8, x0_16x8, i4_out_horz_16x8_r0); i4_out_horz_16x8_r1 = vbslq_s16(chroma_mask_16x8, x1_16x8, i4_out_horz_16x8_r1); i4_out_horz_16x8_r2 = vbslq_s16(chroma_mask_16x8, x2_16x8, i4_out_horz_16x8_r2); i4_out_horz_16x8_r3 = vbslq_s16(chroma_mask_16x8, x3_16x8, i4_out_horz_16x8_r3); vst1q_s16((int16_t *) (pi2_out), i4_out_horz_16x8_r0); vst1q_s16((int16_t *) (pi2_out + out_strd), i4_out_horz_16x8_r1); vst1q_s16((int16_t *) (pi2_out + out_strd * 2), i4_out_horz_16x8_r2); vst1q_s16((int16_t *) (pi2_out + out_strd * 3), i4_out_horz_16x8_r3); return i4_nnz; } /*****************************************************************************/ /* */ /* Function Name : isvcd_iquant_itrans_residual_chroma_4x4_dc_neonintr */ /* */ /* Description : this function computes the resd output from the */ /* IQ+IT */ /* */ /* Inputs : */ /* Globals : none */ /* Processing : */ /* */ /* Outputs : i4_nnz */ /* Returns : none */ /* */ /* Issues : none */ /* */ /* Revision History: */ /* */ /* DD MM YYYY Author(s) Changes (Describe the changes made) */ /* 25 11 2021 Kishore creation */ /* */ /*****************************************************************************/ WORD32 isvcd_iquant_itrans_residual_chroma_4x4_dc_neonintr( WORD16 *pi2_src, WORD16 *pi2_pred, WORD16 *pi2_out, WORD32 pred_strd, WORD32 out_strd, const UWORD16 *pu2_iscal_mat, const UWORD16 *pu2_weigh_mat, UWORD32 u4_qp_div_6, WORD16 *pi2_tmp, WORD16 *pi2_dc_src) { int16x8_t pred0, pred1, pred2, pred3; int16x8_t temp_0, dup_min, dup_max; WORD32 i4_iq_out_temp; int16x8_t dup_val_1, dup_val_2, dup_val_3, dup_val_4, dup_val_5, dup_abs; int16x8_t i4_out_horz_16x8_r0, i4_out_horz_16x8_r1, i4_out_horz_16x8_r2, i4_out_horz_16x8_r3; uint16x8_t chroma_mask_16x8 = vreinterpretq_u16_u32(vdupq_n_u32(0x0000ffff)); int16x8_t chroma_mask_16x8_2 = vreinterpretq_s16_s32(vdupq_n_s32(0x0000ffff)); WORD32 i4_nnz; UNUSED(pi2_src); UNUSED(pu2_iscal_mat); UNUSED(pu2_weigh_mat); UNUSED(u4_qp_div_6); UNUSED(pi2_tmp); UNUSED(pi2_dc_src); i4_iq_out_temp = pi2_dc_src[0]; temp_0 = vdupq_n_s16((i4_iq_out_temp + 32) >> 6); dup_min = vdupq_n_s16(RSD_MIN); dup_max = vdupq_n_s16(RSD_MAX); pred0 = vld1q_s16((int16_t *) pi2_pred); pred1 = vld1q_s16((int16_t *) pi2_pred + pred_strd); pred2 = vld1q_s16((int16_t *) pi2_pred + (pred_strd * 2)); pred3 = vld1q_s16((int16_t *) pi2_pred + (pred_strd * 3)); pred0 = vaddq_s16(pred0, temp_0); pred1 = vaddq_s16(pred1, temp_0); pred2 = vaddq_s16(pred2, temp_0); pred3 = vaddq_s16(pred3, temp_0); pred0 = vminq_s16(pred0, dup_max); pred0 = vmaxq_s16(pred0, dup_min); pred1 = vminq_s16(pred1, dup_max); pred1 = vmaxq_s16(pred1, dup_min); pred2 = vminq_s16(pred2, dup_max); pred2 = vmaxq_s16(pred2, dup_min); pred3 = vminq_s16(pred3, dup_max); pred3 = vmaxq_s16(pred3, dup_min); pred0 = vandq_s16(pred0, chroma_mask_16x8_2); pred1 = vandq_s16(pred1, chroma_mask_16x8_2); pred2 = vandq_s16(pred2, chroma_mask_16x8_2); pred3 = vandq_s16(pred3, chroma_mask_16x8_2); dup_val_1 = vabsq_s16(pred0); dup_val_2 = vabsq_s16(pred1); dup_val_3 = vabsq_s16(pred2); dup_val_4 = vabsq_s16(pred2); dup_val_5 = vqaddq_s16(dup_val_1, dup_val_2); dup_val_1 = vqaddq_s16(dup_val_3, dup_val_4); dup_abs = vqaddq_s16(dup_val_1, dup_val_5); i4_nnz = dup_abs[0] || dup_abs[1] || dup_abs[2] || dup_abs[3] || dup_abs[4] || dup_abs[5] || dup_abs[6] || dup_abs[7]; i4_out_horz_16x8_r0 = vld1q_s16(pi2_out); i4_out_horz_16x8_r1 = vld1q_s16(pi2_out + out_strd); i4_out_horz_16x8_r2 = vld1q_s16(pi2_out + (out_strd * 2)); i4_out_horz_16x8_r3 = vld1q_s16(pi2_out + (out_strd * 3)); i4_out_horz_16x8_r0 = vbslq_s16(chroma_mask_16x8, pred0, i4_out_horz_16x8_r0); i4_out_horz_16x8_r1 = vbslq_s16(chroma_mask_16x8, pred1, i4_out_horz_16x8_r1); i4_out_horz_16x8_r2 = vbslq_s16(chroma_mask_16x8, pred2, i4_out_horz_16x8_r2); i4_out_horz_16x8_r3 = vbslq_s16(chroma_mask_16x8, pred3, i4_out_horz_16x8_r3); vst1q_s16((int16_t *) (pi2_out), i4_out_horz_16x8_r0); vst1q_s16((int16_t *) (pi2_out + out_strd), i4_out_horz_16x8_r1); vst1q_s16((int16_t *) (pi2_out + (out_strd * 2)), i4_out_horz_16x8_r2); vst1q_s16((int16_t *) (pi2_out + (out_strd * 3)), i4_out_horz_16x8_r3); return i4_nnz; } /*****************************************************************************/ /* */ /* Function Name : isvcd_iquant_itrans_residual_8x8_neonintr */ /* */ /* Description : this function computes the resd output from the */ /* IQ+IT */ /* */ /* Inputs : */ /* Globals : none */ /* Processing : */ /* */ /* Outputs : i4_nnz */ /* Returns : none */ /* */ /* Issues : none */ /* */ /* Revision History: */ /* */ /* DD MM YYYY Author(s) Changes (Describe the changes made) */ /* 25 11 2021 Kishore creation */ /* */ /*****************************************************************************/ WORD32 isvcd_iquant_itrans_residual_8x8_neonintr(WORD16 *pi2_src, WORD16 *pi2_pred, WORD16 *pi2_out, WORD32 pred_strd, WORD32 out_strd, const UWORD16 *pu2_iscale_mat, const UWORD16 *pu2_weigh_mat, UWORD32 qp_div, WORD16 *pi2_tmp, WORD32 iq_start_idx, WORD16 *pi2_dc_ld_addr) { int16x8_t iscal_16x8_0, iscal_16x8_1, iscal_16x8_2, iscal_16x8_3, iscal_16x8_4, iscal_16x8_5, iscal_16x8_6, iscal_16x8_7; int16x8_t weigh_16x8_0, weigh_16x8_1, weigh_16x8_2, weigh_16x8_3, weigh_16x8_4, weigh_16x8_5, weigh_16x8_6, weigh_16x8_7; int16x8_t src_16x8_0, src_16x8_1, src_16x8_2, src_16x8_3, src_16x8_4, src_16x8_5, src_16x8_6, src_16x8_7; int16x8_t coeff_mul_16x8_0, coeff_mul_16x8_1, coeff_mul_16x8_2, coeff_mul_16x8_3, coeff_mul_16x8_4, coeff_mul_16x8_5, coeff_mul_16x8_6, coeff_mul_16x8_7; int32x4_t quant_res_32x4_l_0, quant_res_32x4_l_1, quant_res_32x4_l_2, quant_res_32x4_l_3, quant_res_32x4_l_4, quant_res_32x4_l_5, quant_res_32x4_l_6, quant_res_32x4_l_7; int32x4_t quant_res_32x4_h_0, quant_res_32x4_h_1, quant_res_32x4_h_2, quant_res_32x4_h_3, quant_res_32x4_h_4, quant_res_32x4_h_5, quant_res_32x4_h_6, quant_res_32x4_h_7; int16x4_t quant_res_16x4_l_0, quant_res_16x4_l_1, quant_res_16x4_l_2, quant_res_16x4_l_3, quant_res_16x4_l_4, quant_res_16x4_l_5, quant_res_16x4_l_6, quant_res_16x4_l_7; int16x4_t quant_res_16x4_h_0, quant_res_16x4_h_1, quant_res_16x4_h_2, quant_res_16x4_h_3, quant_res_16x4_h_4, quant_res_16x4_h_5, quant_res_16x4_h_6, quant_res_16x4_h_7; int16x8_t quant_res_16x8_0, quant_res_16x8_1, quant_res_16x8_2, quant_res_16x8_3, quant_res_16x8_4, quant_res_16x8_5, quant_res_16x8_6, quant_res_16x8_7; int16x8_t trans_16x8_0, trans_16x8_1, trans_16x8_2, trans_16x8_3, trans_16x8_4, trans_16x8_5, trans_16x8_6, trans_16x8_7; int32x4_t trans_32x4_0, trans_32x4_1, trans_32x4_2, trans_32x4_3, trans_32x4_4, trans_32x4_5, trans_32x4_6, trans_32x4_7; int64x2_t trans_64x2_0, trans_64x2_1, trans_64x2_2, trans_64x2_3, trans_64x2_4, trans_64x2_5, trans_64x2_6, trans_64x2_7; int16x4_t trans_16x4_1_l, trans_16x4_3_l, trans_16x4_5_l, trans_16x4_7_l; int16x8_t rs_trans_16x8_1, rs_trans_16x8_2, rs_trans_16x8_3, rs_trans_16x8_5, rs_trans_16x8_6, rs_trans_16x8_7; int32x4_t sub_3_5_l, sub_3_5_h; int32x4_t add_3_5_l, add_3_5_h; int32x4_t sub_1_7_l, sub_1_7_h; int32x4_t add_1_7_l, add_1_7_h; int32x4_t sub_357_l, sub_357_h; int32x4_t add_351_l, add_351_h; int32x4_t add_175_l, add_175_h; int32x4_t sub_173_l, sub_173_h; int32x4_t y1_32x4_l, y1_32x4_h; int32x4_t y3_32x4_l, y3_32x4_h; int32x4_t y5_32x4_l, y5_32x4_h; int32x4_t y7_32x4_l, y7_32x4_h; int16x4_t y1_16x4_l, y3_16x4_l, y5_16x4_l, y7_16x4_l; int16x8_t y0_16x8, y1_16x8, y2_16x8, y3_16x8, y4_16x8, y5_16x8, y6_16x8, y7_16x8; int16x8_t rs_y1_16x8, rs_y3_16x8, rs_y5_16x8, rs_y7_16x8; int16x8_t z0_16x8, z1_16x8, z2_16x8, z3_16x8, z4_16x8, z5_16x8, z6_16x8, z7_16x8; int16x8_t pred0, pred1, pred2, pred3, pred4, pred5, pred6, pred7; int64x2_t pred0_in_64x2, pred1_in_64x2, pred2_in_64x2, pred3_in_64x2, pred4_in_64x2, pred5_in_64x2, pred6_in_64x2, pred7_in_64x2; int32x4_t qp_div_6_32x4 = vdupq_n_s32(qp_div); int16x8_t pred_b0_r01_in; int16x8_t pred_b0_r23_in; int16x8_t pred_b1_r01_in; int16x8_t pred_b1_r23_in; int16x8_t pred_b2_r45_in; int16x8_t pred_b2_r67_in; int16x8_t pred_b3_r45_in; int16x8_t pred_b3_r67_in; int16x8_t dup_min, dup_max; int16x8x2_t trans_16x8_0_1, trans_16x8_2_3, trans_16x8_4_5, trans_16x8_6_7; int32x4x2_t trans_32x4_0_2, trans_32x4_1_3, trans_32x4_4_6, trans_32x4_5_7; int16x4_t y1_16x4_h, y3_16x4_h, y5_16x4_h, y7_16x4_h; WORD32 nnz, nnz_b0, nnz_b1, nnz_b2, nnz_b3, i; int16x8_t dup_val_1, dup_val_2, dup_abs; UNUSED(pi2_tmp); UNUSED(iq_start_idx); UNUSED(pi2_dc_ld_addr); dup_min = vdupq_n_s16(RSD_MIN); dup_max = vdupq_n_s16(RSD_MAX); iscal_16x8_0 = vld1q_s16((const int16_t *) pu2_iscale_mat); iscal_16x8_1 = vld1q_s16((const int16_t *) (pu2_iscale_mat + 8)); iscal_16x8_2 = vld1q_s16((const int16_t *) (pu2_iscale_mat + 16)); iscal_16x8_3 = vld1q_s16((const int16_t *) (pu2_iscale_mat + 24)); iscal_16x8_4 = vld1q_s16((const int16_t *) (pu2_iscale_mat + 32)); iscal_16x8_5 = vld1q_s16((const int16_t *) (pu2_iscale_mat + 40)); iscal_16x8_6 = vld1q_s16((const int16_t *) (pu2_iscale_mat + 48)); iscal_16x8_7 = vld1q_s16((const int16_t *) (pu2_iscale_mat + 56)); weigh_16x8_0 = vld1q_s16((const int16_t *) pu2_weigh_mat); weigh_16x8_1 = vld1q_s16((const int16_t *) (pu2_weigh_mat + 8)); weigh_16x8_2 = vld1q_s16((const int16_t *) (pu2_weigh_mat + 16)); weigh_16x8_3 = vld1q_s16((const int16_t *) (pu2_weigh_mat + 24)); weigh_16x8_4 = vld1q_s16((const int16_t *) (pu2_weigh_mat + 32)); weigh_16x8_5 = vld1q_s16((const int16_t *) (pu2_weigh_mat + 40)); weigh_16x8_6 = vld1q_s16((const int16_t *) (pu2_weigh_mat + 48)); weigh_16x8_7 = vld1q_s16((const int16_t *) (pu2_weigh_mat + 56)); src_16x8_0 = vld1q_s16((const int16_t *) pi2_src); // a0 a1 a2 a3 a4 a5 a6 a7 src_16x8_1 = vld1q_s16((const int16_t *) (pi2_src + 8)); // b0 b1 b2 b3 b4 b5 b6 b7 src_16x8_2 = vld1q_s16((const int16_t *) (pi2_src + 16)); src_16x8_3 = vld1q_s16((const int16_t *) (pi2_src + 24)); src_16x8_4 = vld1q_s16((const int16_t *) (pi2_src + 32)); src_16x8_5 = vld1q_s16((const int16_t *) (pi2_src + 40)); src_16x8_6 = vld1q_s16((const int16_t *) (pi2_src + 48)); src_16x8_7 = vld1q_s16((const int16_t *) (pi2_src + 56)); coeff_mul_16x8_0 = vmulq_s16(iscal_16x8_0, weigh_16x8_0); coeff_mul_16x8_1 = vmulq_s16(iscal_16x8_1, weigh_16x8_1); coeff_mul_16x8_2 = vmulq_s16(iscal_16x8_2, weigh_16x8_2); coeff_mul_16x8_3 = vmulq_s16(iscal_16x8_3, weigh_16x8_3); coeff_mul_16x8_4 = vmulq_s16(iscal_16x8_4, weigh_16x8_4); coeff_mul_16x8_5 = vmulq_s16(iscal_16x8_5, weigh_16x8_5); coeff_mul_16x8_6 = vmulq_s16(iscal_16x8_6, weigh_16x8_6); coeff_mul_16x8_7 = vmulq_s16(iscal_16x8_7, weigh_16x8_7); quant_res_32x4_l_0 = vmull_s16(vget_low_s16(coeff_mul_16x8_0), vget_low_s16(src_16x8_0)); quant_res_32x4_l_1 = vmull_s16(vget_low_s16(coeff_mul_16x8_1), vget_low_s16(src_16x8_1)); quant_res_32x4_l_2 = vmull_s16(vget_low_s16(coeff_mul_16x8_2), vget_low_s16(src_16x8_2)); quant_res_32x4_l_3 = vmull_s16(vget_low_s16(coeff_mul_16x8_3), vget_low_s16(src_16x8_3)); quant_res_32x4_l_4 = vmull_s16(vget_low_s16(coeff_mul_16x8_4), vget_low_s16(src_16x8_4)); quant_res_32x4_l_5 = vmull_s16(vget_low_s16(coeff_mul_16x8_5), vget_low_s16(src_16x8_5)); quant_res_32x4_l_6 = vmull_s16(vget_low_s16(coeff_mul_16x8_6), vget_low_s16(src_16x8_6)); quant_res_32x4_l_7 = vmull_s16(vget_low_s16(coeff_mul_16x8_7), vget_low_s16(src_16x8_7)); quant_res_32x4_h_0 = vmull_s16(vget_high_s16(coeff_mul_16x8_0), vget_high_s16(src_16x8_0)); quant_res_32x4_h_1 = vmull_s16(vget_high_s16(coeff_mul_16x8_1), vget_high_s16(src_16x8_1)); quant_res_32x4_h_2 = vmull_s16(vget_high_s16(coeff_mul_16x8_2), vget_high_s16(src_16x8_2)); quant_res_32x4_h_3 = vmull_s16(vget_high_s16(coeff_mul_16x8_3), vget_high_s16(src_16x8_3)); quant_res_32x4_h_4 = vmull_s16(vget_high_s16(coeff_mul_16x8_4), vget_high_s16(src_16x8_4)); quant_res_32x4_h_5 = vmull_s16(vget_high_s16(coeff_mul_16x8_5), vget_high_s16(src_16x8_5)); quant_res_32x4_h_6 = vmull_s16(vget_high_s16(coeff_mul_16x8_6), vget_high_s16(src_16x8_6)); quant_res_32x4_h_7 = vmull_s16(vget_high_s16(coeff_mul_16x8_7), vget_high_s16(src_16x8_7)); quant_res_32x4_l_0 = vshlq_s32(quant_res_32x4_l_0, qp_div_6_32x4); quant_res_32x4_l_1 = vshlq_s32(quant_res_32x4_l_1, qp_div_6_32x4); quant_res_32x4_l_2 = vshlq_s32(quant_res_32x4_l_2, qp_div_6_32x4); quant_res_32x4_l_3 = vshlq_s32(quant_res_32x4_l_3, qp_div_6_32x4); quant_res_32x4_l_4 = vshlq_s32(quant_res_32x4_l_4, qp_div_6_32x4); quant_res_32x4_l_5 = vshlq_s32(quant_res_32x4_l_5, qp_div_6_32x4); quant_res_32x4_l_6 = vshlq_s32(quant_res_32x4_l_6, qp_div_6_32x4); quant_res_32x4_l_7 = vshlq_s32(quant_res_32x4_l_7, qp_div_6_32x4); quant_res_32x4_h_0 = vshlq_s32(quant_res_32x4_h_0, qp_div_6_32x4); quant_res_32x4_h_1 = vshlq_s32(quant_res_32x4_h_1, qp_div_6_32x4); quant_res_32x4_h_2 = vshlq_s32(quant_res_32x4_h_2, qp_div_6_32x4); quant_res_32x4_h_3 = vshlq_s32(quant_res_32x4_h_3, qp_div_6_32x4); quant_res_32x4_h_4 = vshlq_s32(quant_res_32x4_h_4, qp_div_6_32x4); quant_res_32x4_h_5 = vshlq_s32(quant_res_32x4_h_5, qp_div_6_32x4); quant_res_32x4_h_6 = vshlq_s32(quant_res_32x4_h_6, qp_div_6_32x4); quant_res_32x4_h_7 = vshlq_s32(quant_res_32x4_h_7, qp_div_6_32x4); quant_res_16x4_l_0 = vqrshrn_n_s32(quant_res_32x4_l_0, 6); quant_res_16x4_l_1 = vqrshrn_n_s32(quant_res_32x4_l_1, 6); quant_res_16x4_l_2 = vqrshrn_n_s32(quant_res_32x4_l_2, 6); quant_res_16x4_l_3 = vqrshrn_n_s32(quant_res_32x4_l_3, 6); quant_res_16x4_l_4 = vqrshrn_n_s32(quant_res_32x4_l_4, 6); quant_res_16x4_l_5 = vqrshrn_n_s32(quant_res_32x4_l_5, 6); quant_res_16x4_l_6 = vqrshrn_n_s32(quant_res_32x4_l_6, 6); quant_res_16x4_l_7 = vqrshrn_n_s32(quant_res_32x4_l_7, 6); quant_res_16x4_h_0 = vqrshrn_n_s32(quant_res_32x4_h_0, 6); quant_res_16x4_h_1 = vqrshrn_n_s32(quant_res_32x4_h_1, 6); quant_res_16x4_h_2 = vqrshrn_n_s32(quant_res_32x4_h_2, 6); quant_res_16x4_h_3 = vqrshrn_n_s32(quant_res_32x4_h_3, 6); quant_res_16x4_h_4 = vqrshrn_n_s32(quant_res_32x4_h_4, 6); quant_res_16x4_h_5 = vqrshrn_n_s32(quant_res_32x4_h_5, 6); quant_res_16x4_h_6 = vqrshrn_n_s32(quant_res_32x4_h_6, 6); quant_res_16x4_h_7 = vqrshrn_n_s32(quant_res_32x4_h_7, 6); quant_res_16x8_0 = vcombine_s16(quant_res_16x4_l_0, quant_res_16x4_h_0); quant_res_16x8_1 = vcombine_s16(quant_res_16x4_l_1, quant_res_16x4_h_1); quant_res_16x8_2 = vcombine_s16(quant_res_16x4_l_2, quant_res_16x4_h_2); quant_res_16x8_3 = vcombine_s16(quant_res_16x4_l_3, quant_res_16x4_h_3); quant_res_16x8_4 = vcombine_s16(quant_res_16x4_l_4, quant_res_16x4_h_4); quant_res_16x8_5 = vcombine_s16(quant_res_16x4_l_5, quant_res_16x4_h_5); quant_res_16x8_6 = vcombine_s16(quant_res_16x4_l_6, quant_res_16x4_h_6); quant_res_16x8_7 = vcombine_s16(quant_res_16x4_l_7, quant_res_16x4_h_7); for(i = 0; i < 2; i++) { trans_16x8_0_1 = vtrnq_s16(quant_res_16x8_0, quant_res_16x8_1); trans_16x8_0 = trans_16x8_0_1.val[0]; trans_16x8_1 = trans_16x8_0_1.val[1]; trans_16x8_2_3 = vtrnq_s16(quant_res_16x8_2, quant_res_16x8_3); trans_16x8_2 = trans_16x8_2_3.val[0]; trans_16x8_3 = trans_16x8_2_3.val[1]; trans_16x8_4_5 = vtrnq_s16(quant_res_16x8_4, quant_res_16x8_5); trans_16x8_4 = trans_16x8_4_5.val[0]; trans_16x8_5 = trans_16x8_4_5.val[1]; trans_16x8_6_7 = vtrnq_s16(quant_res_16x8_6, quant_res_16x8_7); trans_16x8_6 = trans_16x8_6_7.val[0]; trans_16x8_7 = trans_16x8_6_7.val[1]; trans_32x4_0_2 = vtrnq_s32(vreinterpretq_s32_s16(trans_16x8_0), vreinterpretq_s32_s16(trans_16x8_2)); trans_32x4_0 = trans_32x4_0_2.val[0]; trans_32x4_2 = trans_32x4_0_2.val[1]; trans_32x4_1_3 = vtrnq_s32(vreinterpretq_s32_s16(trans_16x8_1), vreinterpretq_s32_s16(trans_16x8_3)); trans_32x4_1 = trans_32x4_1_3.val[0]; trans_32x4_3 = trans_32x4_1_3.val[1]; trans_32x4_4_6 = vtrnq_s32(vreinterpretq_s32_s16(trans_16x8_4), vreinterpretq_s32_s16(trans_16x8_6)); trans_32x4_4 = trans_32x4_4_6.val[0]; trans_32x4_6 = trans_32x4_4_6.val[1]; trans_32x4_5_7 = vtrnq_s32(vreinterpretq_s32_s16(trans_16x8_5), vreinterpretq_s32_s16(trans_16x8_7)); trans_32x4_5 = trans_32x4_5_7.val[0]; trans_32x4_7 = trans_32x4_5_7.val[1]; trans_64x2_0 = vcombine_s64(vreinterpret_s64_s32(vget_low_s32(trans_32x4_0)), vreinterpret_s64_s32(vget_low_s32(trans_32x4_4))); trans_64x2_4 = vcombine_s64(vreinterpret_s64_s32(vget_high_s32(trans_32x4_0)), vreinterpret_s64_s32(vget_high_s32(trans_32x4_4))); trans_64x2_1 = vcombine_s64(vreinterpret_s64_s32(vget_low_s32(trans_32x4_1)), vreinterpret_s64_s32(vget_low_s32(trans_32x4_5))); trans_64x2_5 = vcombine_s64(vreinterpret_s64_s32(vget_high_s32(trans_32x4_1)), vreinterpret_s64_s32(vget_high_s32(trans_32x4_5))); trans_64x2_2 = vcombine_s64(vreinterpret_s64_s32(vget_low_s32(trans_32x4_2)), vreinterpret_s64_s32(vget_low_s32(trans_32x4_6))); trans_64x2_6 = vcombine_s64(vreinterpret_s64_s32(vget_high_s32(trans_32x4_2)), vreinterpret_s64_s32(vget_high_s32(trans_32x4_6))); trans_64x2_3 = vcombine_s64(vreinterpret_s64_s32(vget_low_s32(trans_32x4_3)), vreinterpret_s64_s32(vget_low_s32(trans_32x4_7))); trans_64x2_7 = vcombine_s64(vreinterpret_s64_s32(vget_high_s32(trans_32x4_3)), vreinterpret_s64_s32(vget_high_s32(trans_32x4_7))); trans_16x8_0 = vreinterpretq_s16_s64(trans_64x2_0); trans_16x8_1 = vreinterpretq_s16_s64(trans_64x2_1); trans_16x8_2 = vreinterpretq_s16_s64(trans_64x2_2); trans_16x8_3 = vreinterpretq_s16_s64(trans_64x2_3); trans_16x8_4 = vreinterpretq_s16_s64(trans_64x2_4); trans_16x8_5 = vreinterpretq_s16_s64(trans_64x2_5); trans_16x8_6 = vreinterpretq_s16_s64(trans_64x2_6); trans_16x8_7 = vreinterpretq_s16_s64(trans_64x2_7); rs_trans_16x8_1 = vshrq_n_s16(trans_16x8_1, 1); //(pi2_tmp_ptr[1] >> 1) rs_trans_16x8_2 = vshrq_n_s16(trans_16x8_2, 1); //(pi2_tmp_ptr[2] >> 1) rs_trans_16x8_3 = vshrq_n_s16(trans_16x8_3, 1); //(pi2_tmp_ptr[3] >> 1) rs_trans_16x8_5 = vshrq_n_s16(trans_16x8_5, 1); //(pi2_tmp_ptr[5] >> 1) rs_trans_16x8_6 = vshrq_n_s16(trans_16x8_6, 1); //(pi2_tmp_ptr[6] >> 1) rs_trans_16x8_7 = vshrq_n_s16(trans_16x8_7, 1); //(pi2_tmp_ptr[7] >> 1) y0_16x8 = vaddq_s16(trans_16x8_0, trans_16x8_4); // i_y0 = (pi2_tmp_ptr[0] + pi2_tmp_ptr[4] ); y2_16x8 = vsubq_s16(trans_16x8_0, trans_16x8_4); // i_y2 = (pi2_tmp_ptr[0] - pi2_tmp_ptr[4] ); y4_16x8 = vsubq_s16(rs_trans_16x8_2, trans_16x8_6); // i_y4 = ((pi2_tmp_ptr[2] >> 1) - pi2_tmp_ptr[6] ); y6_16x8 = vaddq_s16(trans_16x8_2, rs_trans_16x8_6); // i_y6 = (pi2_tmp_ptr[2] + (pi2_tmp_ptr[6] >> 1)); trans_16x4_3_l = vget_low_s16(trans_16x8_3); trans_16x4_5_l = vget_low_s16(trans_16x8_5); //-w3 + w5 sub_3_5_l = vsubl_s16(vget_low_s16(trans_16x8_5), vget_low_s16(trans_16x8_3)); sub_3_5_h = vsubl_s16(vget_high_s16(trans_16x8_5), vget_high_s16(trans_16x8_3)); // w3 + w5 add_3_5_l = vaddl_s16(trans_16x4_3_l, trans_16x4_5_l); add_3_5_h = vaddl_s16(vget_high_s16(trans_16x8_3), vget_high_s16(trans_16x8_5)); trans_16x4_1_l = vget_low_s16(trans_16x8_1); trans_16x4_7_l = vget_low_s16(trans_16x8_7); //-w1 + w7 sub_1_7_l = vsubl_s16(trans_16x4_7_l, trans_16x4_1_l); sub_1_7_h = vsubl_s16(vget_high_s16(trans_16x8_7), vget_high_s16(trans_16x8_1)); // w1 + w7 add_1_7_l = vaddl_s16(trans_16x4_1_l, trans_16x4_7_l); add_1_7_h = vaddl_s16(vget_high_s16(trans_16x8_1), vget_high_s16(trans_16x8_7)); //-w3 + w5 - w7 sub_357_l = vsubw_s16(sub_3_5_l, trans_16x4_7_l); sub_357_h = vsubw_s16(sub_3_5_h, vget_high_s16(trans_16x8_7)); // w3 + w5 + w1 add_351_l = vaddw_s16(add_3_5_l, trans_16x4_1_l); add_351_h = vaddw_s16(add_3_5_h, vget_high_s16(trans_16x8_1)); //-w1 + w7 + w5 add_175_l = vaddw_s16(sub_1_7_l, trans_16x4_5_l); add_175_h = vaddw_s16(sub_1_7_h, vget_high_s16(trans_16x8_5)); // w1 + w7 - w3 sub_173_l = vsubw_s16(add_1_7_l, trans_16x4_3_l); sub_173_h = vsubw_s16(add_1_7_h, vget_high_s16(trans_16x8_3)); //-w3 + w5 - w7 - (w7 >> 1) y1_32x4_l = vsubw_s16(sub_357_l, vget_low_s16(rs_trans_16x8_7)); y1_32x4_h = vsubw_s16(sub_357_h, vget_high_s16(rs_trans_16x8_7)); // w1 + w7 - w3 - (w3 >> 1) y3_32x4_l = vsubw_s16(sub_173_l, vget_low_s16(rs_trans_16x8_3)); y3_32x4_h = vsubw_s16(sub_173_h, vget_high_s16(rs_trans_16x8_3)); //-w1 + w7 + w5 + (w5 >> 1) y5_32x4_l = vaddw_s16(add_175_l, vget_low_s16(rs_trans_16x8_5)); y5_32x4_h = vaddw_s16(add_175_h, vget_high_s16(rs_trans_16x8_5)); // w3 + w5 + w1 + (w1 >> 1) y7_32x4_l = vaddw_s16(add_351_l, vget_low_s16(rs_trans_16x8_1)); y7_32x4_h = vaddw_s16(add_351_h, vget_high_s16(rs_trans_16x8_1)); y1_16x4_l = vmovn_s32(y1_32x4_l); y1_16x4_h = vmovn_s32(y1_32x4_h); y1_16x8 = vcombine_s16(y1_16x4_l, y1_16x4_h); y3_16x4_l = vmovn_s32(y3_32x4_l); y3_16x4_h = vmovn_s32(y3_32x4_h); y3_16x8 = vcombine_s16(y3_16x4_l, y3_16x4_h); y5_16x4_l = vmovn_s32(y5_32x4_l); y5_16x4_h = vmovn_s32(y5_32x4_h); y5_16x8 = vcombine_s16(y5_16x4_l, y5_16x4_h); y7_16x4_l = vmovn_s32(y7_32x4_l); y7_16x4_h = vmovn_s32(y7_32x4_h); y7_16x8 = vcombine_s16(y7_16x4_l, y7_16x4_h); rs_y1_16x8 = vshrq_n_s16(y1_16x8, 2); rs_y3_16x8 = vshrq_n_s16(y3_16x8, 2); rs_y5_16x8 = vshrq_n_s16(y5_16x8, 2); rs_y7_16x8 = vshrq_n_s16(y7_16x8, 2); z0_16x8 = vaddq_s16(y0_16x8, y6_16x8); // z0 = y0 + y6 z1_16x8 = vaddq_s16(y1_16x8, rs_y7_16x8); // z1 = y1 + (y7 >> 2) z2_16x8 = vaddq_s16(y2_16x8, y4_16x8); // z2 = y2 + y4 z3_16x8 = vaddq_s16(y3_16x8, rs_y5_16x8); // z3 = y3 + (y5 >> 2) z4_16x8 = vsubq_s16(y2_16x8, y4_16x8); // z4 = y2 - y4 z5_16x8 = vsubq_s16(rs_y3_16x8, y5_16x8); // z5 = (y3 >> 2) - y5 z6_16x8 = vsubq_s16(y0_16x8, y6_16x8); // z6 = y0 - y6 z7_16x8 = vsubq_s16(y7_16x8, rs_y1_16x8); // z7 = y7 - (y1 >> 2) quant_res_16x8_0 = vaddq_s16(z0_16x8, z7_16x8); // x0 = z0 + z7 quant_res_16x8_1 = vaddq_s16(z2_16x8, z5_16x8); // x1 = z2 + z5 quant_res_16x8_2 = vaddq_s16(z4_16x8, z3_16x8); // x2 = z4 + z3 quant_res_16x8_3 = vaddq_s16(z6_16x8, z1_16x8); // x3 = z6 + z1 quant_res_16x8_4 = vsubq_s16(z6_16x8, z1_16x8); // x4 = z6 - z1 quant_res_16x8_5 = vsubq_s16(z4_16x8, z3_16x8); // x5 = z4 - z3 quant_res_16x8_6 = vsubq_s16(z2_16x8, z5_16x8); // x6 = z2 - z5 quant_res_16x8_7 = vsubq_s16(z0_16x8, z7_16x8); // x7 = z0 - z7 } quant_res_16x8_0 = vrshrq_n_s16(quant_res_16x8_0, 6); quant_res_16x8_1 = vrshrq_n_s16(quant_res_16x8_1, 6); quant_res_16x8_2 = vrshrq_n_s16(quant_res_16x8_2, 6); quant_res_16x8_3 = vrshrq_n_s16(quant_res_16x8_3, 6); quant_res_16x8_4 = vrshrq_n_s16(quant_res_16x8_4, 6); quant_res_16x8_5 = vrshrq_n_s16(quant_res_16x8_5, 6); quant_res_16x8_6 = vrshrq_n_s16(quant_res_16x8_6, 6); quant_res_16x8_7 = vrshrq_n_s16(quant_res_16x8_7, 6); pred0 = vld1q_s16((int16_t *) pi2_pred); pred1 = vld1q_s16((int16_t *) pi2_pred + pred_strd); pred2 = vld1q_s16((int16_t *) pi2_pred + (pred_strd * 2)); pred3 = vld1q_s16((int16_t *) pi2_pred + (pred_strd * 3)); pred4 = vld1q_s16((int16_t *) pi2_pred + (pred_strd * 4)); pred5 = vld1q_s16((int16_t *) pi2_pred + (pred_strd * 5)); pred6 = vld1q_s16((int16_t *) pi2_pred + (pred_strd * 6)); pred7 = vld1q_s16((int16_t *) pi2_pred + (pred_strd * 7)); quant_res_16x8_0 = vaddq_s16(pred0, quant_res_16x8_0); quant_res_16x8_1 = vaddq_s16(pred1, quant_res_16x8_1); quant_res_16x8_2 = vaddq_s16(pred2, quant_res_16x8_2); quant_res_16x8_3 = vaddq_s16(pred3, quant_res_16x8_3); quant_res_16x8_4 = vaddq_s16(pred4, quant_res_16x8_4); quant_res_16x8_5 = vaddq_s16(pred5, quant_res_16x8_5); quant_res_16x8_6 = vaddq_s16(pred6, quant_res_16x8_6); quant_res_16x8_7 = vaddq_s16(pred7, quant_res_16x8_7); quant_res_16x8_0 = vminq_s16(quant_res_16x8_0, dup_max); quant_res_16x8_0 = vmaxq_s16(quant_res_16x8_0, dup_min); quant_res_16x8_1 = vminq_s16(quant_res_16x8_1, dup_max); quant_res_16x8_1 = vmaxq_s16(quant_res_16x8_1, dup_min); quant_res_16x8_2 = vminq_s16(quant_res_16x8_2, dup_max); quant_res_16x8_2 = vmaxq_s16(quant_res_16x8_2, dup_min); quant_res_16x8_3 = vminq_s16(quant_res_16x8_3, dup_max); quant_res_16x8_3 = vmaxq_s16(quant_res_16x8_3, dup_min); quant_res_16x8_4 = vminq_s16(quant_res_16x8_4, dup_max); quant_res_16x8_4 = vmaxq_s16(quant_res_16x8_4, dup_min); quant_res_16x8_5 = vminq_s16(quant_res_16x8_5, dup_max); quant_res_16x8_5 = vmaxq_s16(quant_res_16x8_5, dup_min); quant_res_16x8_6 = vminq_s16(quant_res_16x8_6, dup_max); quant_res_16x8_6 = vmaxq_s16(quant_res_16x8_6, dup_min); quant_res_16x8_7 = vminq_s16(quant_res_16x8_7, dup_max); quant_res_16x8_7 = vmaxq_s16(quant_res_16x8_7, dup_min); pred0_in_64x2 = vreinterpretq_s64_s16(quant_res_16x8_0); pred1_in_64x2 = vreinterpretq_s64_s16(quant_res_16x8_1); pred2_in_64x2 = vreinterpretq_s64_s16(quant_res_16x8_2); pred3_in_64x2 = vreinterpretq_s64_s16(quant_res_16x8_3); pred4_in_64x2 = vreinterpretq_s64_s16(quant_res_16x8_4); pred5_in_64x2 = vreinterpretq_s64_s16(quant_res_16x8_5); pred6_in_64x2 = vreinterpretq_s64_s16(quant_res_16x8_6); pred7_in_64x2 = vreinterpretq_s64_s16(quant_res_16x8_7); pred_b0_r01_in = vreinterpretq_s16_s64( vcombine_s64(vget_low_s64(pred0_in_64x2), vget_low_s64(pred1_in_64x2))); pred_b0_r23_in = vreinterpretq_s16_s64( vcombine_s64(vget_low_s64(pred2_in_64x2), vget_low_s64(pred3_in_64x2))); pred_b1_r01_in = vreinterpretq_s16_s64( vcombine_s64(vget_high_s64(pred0_in_64x2), vget_high_s64(pred1_in_64x2))); pred_b1_r23_in = vreinterpretq_s16_s64( vcombine_s64(vget_high_s64(pred2_in_64x2), vget_high_s64(pred3_in_64x2))); pred_b2_r45_in = vreinterpretq_s16_s64( vcombine_s64(vget_low_s64(pred4_in_64x2), vget_low_s64(pred5_in_64x2))); pred_b2_r67_in = vreinterpretq_s16_s64( vcombine_s64(vget_low_s64(pred6_in_64x2), vget_low_s64(pred7_in_64x2))); pred_b3_r45_in = vreinterpretq_s16_s64( vcombine_s64(vget_high_s64(pred4_in_64x2), vget_high_s64(pred5_in_64x2))); pred_b3_r67_in = vreinterpretq_s16_s64( vcombine_s64(vget_high_s64(pred6_in_64x2), vget_high_s64(pred7_in_64x2))); dup_val_1 = vabsq_s16(pred_b0_r01_in); dup_val_2 = vabsq_s16(pred_b0_r23_in); dup_abs = vqaddq_s16(dup_val_1, dup_val_2); nnz_b0 = dup_abs[0] || dup_abs[1] || dup_abs[2] || dup_abs[3] || dup_abs[4] || dup_abs[5] || dup_abs[6] || dup_abs[7]; dup_val_1 = vabsq_s16(pred_b1_r01_in); dup_val_2 = vabsq_s16(pred_b1_r23_in); dup_abs = vqaddq_s16(dup_val_1, dup_val_2); nnz_b1 = dup_abs[0] || dup_abs[1] || dup_abs[2] || dup_abs[3] || dup_abs[4] || dup_abs[5] || dup_abs[6] || dup_abs[7]; dup_val_1 = vabsq_s16(pred_b2_r45_in); dup_val_2 = vabsq_s16(pred_b2_r67_in); dup_abs = vqaddq_s16(dup_val_1, dup_val_2); nnz_b2 = dup_abs[0] || dup_abs[1] || dup_abs[2] || dup_abs[3] || dup_abs[4] || dup_abs[5] || dup_abs[6] || dup_abs[7]; dup_val_1 = vabsq_s16(pred_b3_r45_in); dup_val_2 = vabsq_s16(pred_b3_r67_in); dup_abs = vqaddq_s16(dup_val_1, dup_val_2); nnz_b3 = dup_abs[0] || dup_abs[1] || dup_abs[2] || dup_abs[3] || dup_abs[4] || dup_abs[5] || dup_abs[6] || dup_abs[7]; nnz = (nnz_b0 | (nnz_b1 << 1) | (nnz_b2 << 4) | (nnz_b3 << 5)); vst1q_s16(pi2_out, quant_res_16x8_0); vst1q_s16(pi2_out + out_strd, quant_res_16x8_1); vst1q_s16(pi2_out + out_strd * 2, quant_res_16x8_2); vst1q_s16(pi2_out + out_strd * 3, quant_res_16x8_3); vst1q_s16(pi2_out + out_strd * 4, quant_res_16x8_4); vst1q_s16(pi2_out + out_strd * 5, quant_res_16x8_5); vst1q_s16(pi2_out + out_strd * 6, quant_res_16x8_6); vst1q_s16(pi2_out + out_strd * 7, quant_res_16x8_7); return nnz; } /*****************************************************************************/ /* */ /* Function Name : isvcd_iquant_itrans_residual_8x8_dc_neonintr */ /* */ /* Description : this function computes the resd dc output from the */ /* IQ+IT */ /* */ /* Inputs : */ /* Globals : none */ /* Processing : */ /* */ /* Outputs : i4_nnz */ /* Returns : none */ /* */ /* Issues : none */ /* */ /* Revision History: */ /* */ /* DD MM YYYY Author(s) Changes (Describe the changes made) */ /* 25 11 2021 Kishore creation */ /* */ /*****************************************************************************/ WORD32 isvcd_iquant_itrans_residual_8x8_dc_neonintr(WORD16 *pi2_src, WORD16 *pi2_pred, WORD16 *pi2_out, WORD32 pred_strd, WORD32 out_strd, const UWORD16 *pu2_iscale_mat, const UWORD16 *pu2_weigh_mat, UWORD32 qp_div, WORD16 *pi2_tmp, WORD32 iq_start_idx, WORD16 *pi2_dc_ld_addr) { WORD32 i4_iq_out_temp; int16x8_t temp_0, dup_min, dup_max; int16x8_t pred0_in, pred1_in, pred2_in, pred3_in; int16x8_t pred4_in, pred5_in, pred6_in, pred7_in; WORD32 rnd_fact = (qp_div < 6) ? (1 << (5 - qp_div)) : 0; int64x2_t pred0_in_64x2, pred1_in_64x2, pred2_in_64x2, pred3_in_64x2, pred4_in_64x2, pred5_in_64x2, pred6_in_64x2, pred7_in_64x2; int16x8_t pred_b0_r01_in; int16x8_t pred_b0_r23_in; int16x8_t pred_b1_r01_in; int16x8_t pred_b1_r23_in; int16x8_t pred_b2_r45_in; int16x8_t pred_b2_r67_in; int16x8_t pred_b3_r45_in; int16x8_t pred_b3_r67_in; int16x8_t dup_val_1, dup_val_2, dup_abs; WORD32 nnz, nnz_b0, nnz_b1, nnz_b2, nnz_b3; UNUSED(pi2_tmp); UNUSED(iq_start_idx); UNUSED(pi2_dc_ld_addr); i4_iq_out_temp = pi2_src[0]; INV_QUANT(i4_iq_out_temp, pu2_iscale_mat[0], pu2_weigh_mat[0], qp_div, rnd_fact, 6); /* inv transform and residual comp */ pred0_in = vld1q_s16((int16_t *) pi2_pred); pi2_pred = pi2_pred + pred_strd; pred1_in = vld1q_s16((int16_t *) pi2_pred); pi2_pred = pi2_pred + pred_strd; pred2_in = vld1q_s16((int16_t *) pi2_pred); pi2_pred = pi2_pred + pred_strd; pred3_in = vld1q_s16((int16_t *) pi2_pred); pi2_pred = pi2_pred + pred_strd; pred4_in = vld1q_s16((int16_t *) pi2_pred); pi2_pred = pi2_pred + pred_strd; pred5_in = vld1q_s16((int16_t *) pi2_pred); pi2_pred = pi2_pred + pred_strd; pred6_in = vld1q_s16((int16_t *) pi2_pred); pi2_pred = pi2_pred + pred_strd; pred7_in = vld1q_s16((int16_t *) pi2_pred); temp_0 = vdupq_n_s16((i4_iq_out_temp + 32) >> 6); dup_min = vdupq_n_s16(RSD_MIN); dup_max = vdupq_n_s16(RSD_MAX); pred0_in = vaddq_s16(pred0_in, temp_0); pred1_in = vaddq_s16(pred1_in, temp_0); pred2_in = vaddq_s16(pred2_in, temp_0); pred3_in = vaddq_s16(pred3_in, temp_0); pred4_in = vaddq_s16(pred4_in, temp_0); pred5_in = vaddq_s16(pred5_in, temp_0); pred6_in = vaddq_s16(pred6_in, temp_0); pred7_in = vaddq_s16(pred7_in, temp_0); pred0_in = vminq_s16(pred0_in, dup_max); pred0_in = vmaxq_s16(pred0_in, dup_min); pred1_in = vminq_s16(pred1_in, dup_max); pred1_in = vmaxq_s16(pred1_in, dup_min); pred2_in = vminq_s16(pred2_in, dup_max); pred2_in = vmaxq_s16(pred2_in, dup_min); pred3_in = vminq_s16(pred3_in, dup_max); pred3_in = vmaxq_s16(pred3_in, dup_min); pred4_in = vminq_s16(pred4_in, dup_max); pred4_in = vmaxq_s16(pred4_in, dup_min); pred5_in = vminq_s16(pred5_in, dup_max); pred5_in = vmaxq_s16(pred5_in, dup_min); pred6_in = vminq_s16(pred6_in, dup_max); pred6_in = vmaxq_s16(pred6_in, dup_min); pred7_in = vminq_s16(pred7_in, dup_max); pred7_in = vmaxq_s16(pred7_in, dup_min); pred0_in_64x2 = vreinterpretq_s64_s16(pred0_in); pred1_in_64x2 = vreinterpretq_s64_s16(pred1_in); pred2_in_64x2 = vreinterpretq_s64_s16(pred2_in); pred3_in_64x2 = vreinterpretq_s64_s16(pred3_in); pred4_in_64x2 = vreinterpretq_s64_s16(pred4_in); pred5_in_64x2 = vreinterpretq_s64_s16(pred5_in); pred6_in_64x2 = vreinterpretq_s64_s16(pred6_in); pred7_in_64x2 = vreinterpretq_s64_s16(pred7_in); pred_b0_r01_in = vreinterpretq_s16_s64( vcombine_s64(vget_low_s64(pred0_in_64x2), vget_low_s64(pred1_in_64x2))); pred_b0_r23_in = vreinterpretq_s16_s64( vcombine_s64(vget_low_s64(pred2_in_64x2), vget_low_s64(pred3_in_64x2))); pred_b1_r01_in = vreinterpretq_s16_s64( vcombine_s64(vget_high_s64(pred0_in_64x2), vget_high_s64(pred1_in_64x2))); pred_b1_r23_in = vreinterpretq_s16_s64( vcombine_s64(vget_high_s64(pred2_in_64x2), vget_high_s64(pred3_in_64x2))); pred_b2_r45_in = vreinterpretq_s16_s64( vcombine_s64(vget_low_s64(pred4_in_64x2), vget_low_s64(pred5_in_64x2))); pred_b2_r67_in = vreinterpretq_s16_s64( vcombine_s64(vget_low_s64(pred6_in_64x2), vget_low_s64(pred7_in_64x2))); pred_b3_r45_in = vreinterpretq_s16_s64( vcombine_s64(vget_high_s64(pred4_in_64x2), vget_high_s64(pred5_in_64x2))); pred_b3_r67_in = vreinterpretq_s16_s64( vcombine_s64(vget_high_s64(pred6_in_64x2), vget_high_s64(pred7_in_64x2))); dup_val_1 = vabsq_s16(pred_b0_r01_in); dup_val_2 = vabsq_s16(pred_b0_r23_in); dup_abs = vqaddq_s16(dup_val_1, dup_val_2); nnz_b0 = dup_abs[0] || dup_abs[1] || dup_abs[2] || dup_abs[3] || dup_abs[4] || dup_abs[5] || dup_abs[6] || dup_abs[7]; dup_val_1 = vabsq_s16(pred_b1_r01_in); dup_val_2 = vabsq_s16(pred_b1_r23_in); dup_abs = vqaddq_s16(dup_val_1, dup_val_2); nnz_b1 = dup_abs[0] || dup_abs[1] || dup_abs[2] || dup_abs[3] || dup_abs[4] || dup_abs[5] || dup_abs[6] || dup_abs[7]; dup_val_1 = vabsq_s16(pred_b2_r45_in); dup_val_2 = vabsq_s16(pred_b2_r67_in); dup_abs = vqaddq_s16(dup_val_1, dup_val_2); nnz_b2 = dup_abs[0] || dup_abs[1] || dup_abs[2] || dup_abs[3] || dup_abs[4] || dup_abs[5] || dup_abs[6] || dup_abs[7]; dup_val_1 = vabsq_s16(pred_b3_r45_in); dup_val_2 = vabsq_s16(pred_b3_r67_in); dup_abs = vqaddq_s16(dup_val_1, dup_val_2); nnz_b3 = dup_abs[0] || dup_abs[1] || dup_abs[2] || dup_abs[3] || dup_abs[4] || dup_abs[5] || dup_abs[6] || dup_abs[7]; nnz = (nnz_b0 | (nnz_b1 << 1) | (nnz_b2 << 4) | (nnz_b3 << 5)); vst1q_s16((int16_t *) (pi2_out), pred0_in); vst1q_s16((int16_t *) (pi2_out + out_strd), pred1_in); vst1q_s16((int16_t *) (pi2_out + (out_strd * 2)), pred2_in); vst1q_s16((int16_t *) (pi2_out + (out_strd * 3)), pred3_in); vst1q_s16((int16_t *) (pi2_out + (out_strd * 4)), pred4_in); vst1q_s16((int16_t *) (pi2_out + (out_strd * 5)), pred5_in); vst1q_s16((int16_t *) (pi2_out + (out_strd * 6)), pred6_in); vst1q_s16((int16_t *) (pi2_out + (out_strd * 7)), pred7_in); return nnz; }