May 2015 - ffmpeg-devel - lists.ffmpeg.org

[PATCH 01/16] vp9: profile 1 header decoding.
by Ronald S. Bultje 01 May '15

01 May '15

--- libavcodec/vp9.c | 88 ++++++++++++++++++++++++++++++++++++++++---------------- 1 file changed, 64 insertions(+), 24 deletions(-) diff --git a/libavcodec/vp9.c b/libavcodec/vp9.c index 1310798..3487f99 100644 --- a/libavcodec/vp9.c +++ b/libavcodec/vp9.c @@ -112,8 +112,7 @@ typedef struct VP9Context { uint8_t invisible; uint8_t use_last_frame_mvs; uint8_t errorres; - uint8_t colorspace; - uint8_t fullrange; + uint8_t ss_h, ss_v; uint8_t intraonly; uint8_t resetctx; uint8_t refreshrefmask; @@ -463,11 +462,56 @@ static int update_prob(VP56RangeCoder *c, int p) 255 - inv_recenter_nonneg(inv_map_table[d], 255 - p); } +static enum AVPixelFormat read_colorspace_details(AVCodecContext *ctx) +{ + static const enum AVColorSpace colorspaces[8] = { + AVCOL_SPC_UNSPECIFIED, AVCOL_SPC_BT470BG, AVCOL_SPC_BT709, AVCOL_SPC_SMPTE170M, + AVCOL_SPC_SMPTE240M, AVCOL_SPC_BT2020_NCL, AVCOL_SPC_RESERVED, AVCOL_SPC_RGB, + }; + VP9Context *s = ctx->priv_data; + enum AVPixelFormat res; + + ctx->colorspace = colorspaces[get_bits(&s->gb, 3)]; + if (ctx->colorspace == AVCOL_SPC_RGB) { // RGB = profile 1 + if (s->profile == 1) { + s->ss_h = s->ss_v = 1; + res = AV_PIX_FMT_GBRP; + ctx->color_range = AVCOL_RANGE_JPEG; + } else { + av_log(ctx, AV_LOG_ERROR, "RGB not supported in profile 0\n"); + return AVERROR_INVALIDDATA; + } + } else { + static const enum AVPixelFormat pix_fmt_for_ss[2 /* v */][2 /* h */] = { + { AV_PIX_FMT_YUV444P, AV_PIX_FMT_YUV422P }, + { AV_PIX_FMT_YUV440P, AV_PIX_FMT_YUV420P }, + }; + ctx->color_range = get_bits1(&s->gb) ? AVCOL_RANGE_JPEG : AVCOL_RANGE_MPEG; + if (s->profile == 1) { + s->ss_h = get_bits1(&s->gb); + s->ss_v = get_bits1(&s->gb); + if ((res = pix_fmt_for_ss[s->ss_v][s->ss_h]) == AV_PIX_FMT_YUV420P) { + av_log(ctx, AV_LOG_ERROR, "YUV 4:2:0 not supported in profile 1\n"); + return AVERROR_INVALIDDATA; + } else if (get_bits1(&s->gb)) { + av_log(ctx, AV_LOG_ERROR, "Profile 1 color details reserved bit set\n"); + return AVERROR_INVALIDDATA; + } + } else { + s->ss_h = s->ss_v = 1; + res = AV_PIX_FMT_YUV420P; + } + } + + return res; +} + static int decode_frame_header(AVCodecContext *ctx, const uint8_t *data, int size, int *ref) { VP9Context *s = ctx->priv_data; int c, i, j, k, l, m, n, w, h, max, size2, res, sharp; + enum AVPixelFormat fmt = ctx->pix_fmt; int last_invisible; const uint8_t *data2; @@ -481,8 +525,9 @@ static int decode_frame_header(AVCodecContext *ctx, return AVERROR_INVALIDDATA; } s->profile = get_bits1(&s->gb); - if (get_bits1(&s->gb)) { // reserved bit - av_log(ctx, AV_LOG_ERROR, "Reserved bit should be zero\n"); + s->profile |= get_bits1(&s->gb) << 1; + if (s->profile > 1) { + av_log(ctx, AV_LOG_ERROR, "Profile %d is not yet supported\n", s->profile); return AVERROR_INVALIDDATA; } if (get_bits1(&s->gb)) { @@ -500,12 +545,8 @@ static int decode_frame_header(AVCodecContext *ctx, av_log(ctx, AV_LOG_ERROR, "Invalid sync code\n"); return AVERROR_INVALIDDATA; } - s->colorspace = get_bits(&s->gb, 3); - if (s->colorspace == 7) { // RGB = profile 1 - av_log(ctx, AV_LOG_ERROR, "RGB not supported in profile 0\n"); - return AVERROR_INVALIDDATA; - } - s->fullrange = get_bits1(&s->gb); + if ((fmt = read_colorspace_details(ctx)) < 0) + return fmt; // for profile 1, here follows the subsampling bits s->refreshrefmask = 0xff; w = get_bits(&s->gb, 16) + 1; @@ -520,6 +561,15 @@ static int decode_frame_header(AVCodecContext *ctx, av_log(ctx, AV_LOG_ERROR, "Invalid sync code\n"); return AVERROR_INVALIDDATA; } + if (s->profile == 1) { + if ((fmt = read_colorspace_details(ctx)) < 0) + return fmt; + } else { + s->ss_h = s->ss_v = 1; + fmt = AV_PIX_FMT_YUV420P; + ctx->colorspace = AVCOL_SPC_BT470BG; + ctx->color_range = AVCOL_RANGE_JPEG; + } s->refreshrefmask = get_bits(&s->gb, 8); w = get_bits(&s->gb, 16) + 1; h = get_bits(&s->gb, 16) + 1; @@ -3808,18 +3858,6 @@ static int vp9_decode_frame(AVCodecContext *ctx, void *frame, return res; } - if (s->fullrange) - ctx->color_range = AVCOL_RANGE_JPEG; - else - ctx->color_range = AVCOL_RANGE_MPEG; - - switch (s->colorspace) { - case 1: ctx->colorspace = AVCOL_SPC_BT470BG; break; - case 2: ctx->colorspace = AVCOL_SPC_BT709; break; - case 3: ctx->colorspace = AVCOL_SPC_SMPTE170M; break; - case 4: ctx->colorspace = AVCOL_SPC_SMPTE240M; break; - } - // main tile decode loop memset(s->above_partition_ctx, 0, s->cols); memset(s->above_skip_ctx, 0, s->cols); @@ -3829,8 +3867,8 @@ static int vp9_decode_frame(AVCodecContext *ctx, void *frame, memset(s->above_mode_ctx, NEARESTMV, s->cols); } memset(s->above_y_nnz_ctx, 0, s->sb_cols * 16); - memset(s->above_uv_nnz_ctx[0], 0, s->sb_cols * 8); - memset(s->above_uv_nnz_ctx[1], 0, s->sb_cols * 8); + memset(s->above_uv_nnz_ctx[0], 0, s->sb_cols * 16 >> s->ss_h); + memset(s->above_uv_nnz_ctx[1], 0, s->sb_cols * 16 >> s->ss_h); memset(s->above_segpred_ctx, 0, s->cols); s->pass = s->frames[CUR_FRAME].uses_2pass = ctx->active_thread_type == FF_THREAD_FRAME && s->refreshctx && !s->parallelmode; @@ -4087,6 +4125,8 @@ static int vp9_decode_update_thread_context(AVCodecContext *dst, const AVCodecCo s->invisible = ssrc->invisible; s->keyframe = ssrc->keyframe; + s->ss_v = ssrc->ss_v; + s->ss_h = ssrc->ss_h; s->segmentation.enabled = ssrc->segmentation.enabled; s->segmentation.update_map = ssrc->segmentation.update_map; memcpy(&s->prob_ctx, &ssrc->prob_ctx, sizeof(s->prob_ctx)); -- 2.1.2

2 19

Re: [FFmpeg-devel] Aggiornamenti importanti
by Stefano Sabatini 01 May '15

01 May '15

On date Sunday 2015-04-26 16:19:50 +0200, Giangiuseppe Pili encoded: > Cari tutti, > Vi aggiorno su un paio di cose: [...] > d. E' un vecchio ritornello per alcuni di noi, ma io ci terrei a > inserire la possibilità di fornire delle donazioni. Il sistema > donativo deve essere trasparente e non richiede oneri da parte > nostra, se non il fatto di continuare a lavorare ed esistere. Anche > in questo caso si tratta di lanciare il messaggio che noi facciamo > tutto gratis, ma che questo "gratis" ha un valore. Sicché diamo la > possibilità di aiutarci anche fornendo un aiuto economico. Non > abbiamo ancora la logistica per dare un contraccambio, anche se con > Wolfi e Riccardo abbiamo già in programma di far stampare delle > magliette. Ma per il momento queste andranno a noi del gruppo. Però > bisogna sempre pensare in prospettiva. Perciò io vi invito a darmi > delle vostre opinioni in materia. Aggiungo che personalmente ritengo > che il sistema donativo vada assolutamente inserito. Non credo > avremo delle entrate milionarie, ma è più il fatto di mostrare che > dietro Scuola Filosofica c'è un grande lavoro collettivo che vorrei > far passare. Se c'è qualcuno con qualche dote nella grafica si può produrre del materiale di "merchandise" producendo magliette con scritte e grafica attinenti. Questo implica ovviamente un certo lavoro, poi c'è il fatto che non è chiaro come ridistribuire il "ricavato" sempre che esista. In generale si dovrebbe adottare il principio che il "guadagno" deve essere proporzionato all'entità dei contributi. > e. In merito a (c&d) aggiungo che è opinione mia, di Wolfi e di > Riccardo quella di riuscire a fondare una associazione > culturale. Sia per gestire i fondi in modo totalmente trasparente, > sia per poterci promuovere in modo più efficiente. Fondare > un'associazione richiede tempo. Ma è, a mio parere, il passo > successivo a tutto questo lavoro per poter proseguire. Inoltre, con > essa abbiamo una visibilità maggiorata e la possibilità di > richiedere fondi a enti esterni in modo non individuale (il che fa > spesso la differenza). Quest'ultimo punto lo dico solo per vostra > cognizione. Si tratta, cioè, di un intento a lungo periodo ma che > non siamo ancora in grado di portare avanti nell'immediato. Anche > se, vi dirò, che una bozza parziale era già stata fatta. Però la > stesura del testo ufficiale, che poi va ufficializzato, richiede un > suo tempo ed è una cosa estremamente delicata. Per cui ci si > ritornerà a tempo debito. Per il momento, dunque, si tratta > solamente di una cosa che volevo dirvi per mantenervi informati. > Vorrei una vostra opinione su (c) e su (d) e se avete qualcosa in > particolare da proporre. Un caro saluto a tutti e un augurio per un > sereno fine settimana. Giangi Penso che sia una bella idea. Al momento partecipo a diverse associazioni a vario titolo, posso informarmi per dare un'idea delle procedure e dei costi. Nel caso dell'associazione culturale del mio gruppo di Aikido abbiamo dovuto stendere lo statuto, registrarlo e nominare il direttivo. In quel caso esistono delle limitazioni sui profitti e dei vincoli sul bilancio (le spese devono coprire tutte le entrate nel bilancio annuale). Ciao

3 3

[PATCH] Adobe HTTP Dynamic Streaming (HDS) demuxer improvements
by Gorilla Maguila 01 May '15

01 May '15

After contacting the original author of the the patch posted in: http://lists.ffmpeg.org/pipermail/ffmpeg-devel/2014-February/154612.html I attach a new patch with improvements: - Better codec detection in amfmetata - Better live streams handling (Get current segment, calculate next segment..etc) - Parse tabs in raw xml (0x09) Tested streams: http://livehds.rasset.ie/hds-live/_definst_/rte1/rte1_288p.f4m http://livehds.rasset.ie/hds-live/_definst_/rte2/rte2_288p.f4m http://livehds.rasset.ie/hds-live/_definst_/newsnow/newsnow_288p.f4m TODO: - Parse child manifest, ex: http://www.rte.ie/manifests/rte1.f4m http://www.rte.ie/manifests/newsnow.f4m - Implement fragment caching.

5 13

h264 : data partitioning is not implemented
by しらせ　けんじ 01 May '15

01 May '15

Hello. I have a troube when I dump live stream image from IP camera. The command line I use is below. ffmpeg -i "http://10.12.27.5/livestream.cgi?user=xxx&pwd=xxx&streamid=0&audio=0&filena…" -vcodec copy \ -movflags faststart+empty_moov \ -f segment \ -segment_atclocktime 1 \ -segment_time 60 \ -reset_timestamps 1 \ -strftime 1 \ %y%m%d%H%M.mp4 The segment time is 60 seconds. But the duration of the file is only about 25 seconeds!! I dumped live stream by curl to investigate. In this investigation, I got this message, "data partitining is not implemented. ･･･ If you want to help, upload a sample of this file". So, I uploaded the file to the ftp server. (filename : night.dump) I don't know "data partitioning is not implemented" is the reason of this duration missmatch problem. I hope this file is useful and my trouble will go away. Regard.

3 2

[PATCH] Fix segfault with x11grab when switching windows.
by Matthew Smiglarski 01 May '15

01 May '15

The segfault occurred when running ffmpeg with x11grab and specifying a resolution size greater than the screen, alongside an offset: ./ffmpeg -f x11grab -r 30 -s 1920x1147 -i :0.0+0,53 output.mkv Signed-off-by: Matt Smiglarski <matthew.smiglarski(a)gmail.com> --- libavdevice/xcbgrab.c | 3 +++ 1 file changed, 3 insertions(+) diff --git a/libavdevice/xcbgrab.c b/libavdevice/xcbgrab.c index 777bd25..79324dd 100644 --- a/libavdevice/xcbgrab.c +++ b/libavdevice/xcbgrab.c @@ -309,6 +309,9 @@ static void xcbgrab_draw_mouse(AVFormatContext *s, AVPacket *pkt, xcb_xfixes_get_cursor_image_reply_t *ci; int cx, cy, x, y, w, h, c_off, i_off; + if (!image) + return; + cc = xcb_xfixes_get_cursor_image(gr->conn); ci = xcb_xfixes_get_cursor_image_reply(gr->conn, cc, NULL); if (!ci) -- 2.1.4

2 2

[PATCH] avcodec/mips: MSA (MIPS-SIMD-Arch) optimizations for H264 lpf and weight/biweight functions
by shivraj.patil＠imgtec.com 01 May '15

01 May '15

From: Shivraj Patil <shivraj.patil(a)imgtec.com> Signed-off-by: Shivraj Patil <shivraj.patil(a)imgtec.com> --- libavcodec/h264dsp.c | 1 + libavcodec/h264dsp.h | 2 + libavcodec/mips/Makefile | 2 + libavcodec/mips/h264dsp_init_mips.c | 74 + libavcodec/mips/h264dsp_mips.h | 71 + libavcodec/mips/h264dsp_msa.c | 3037 +++++++++++++++++++++++++++++++++++ libavutil/mips/generic_macros_msa.h | 518 ++++++ 7 files changed, 3705 insertions(+) create mode 100644 libavcodec/mips/h264dsp_init_mips.c create mode 100644 libavcodec/mips/h264dsp_mips.h create mode 100644 libavcodec/mips/h264dsp_msa.c diff --git a/libavcodec/h264dsp.c b/libavcodec/h264dsp.c index ddcd6e6..bfb6f1e 100644 --- a/libavcodec/h264dsp.c +++ b/libavcodec/h264dsp.c @@ -157,4 +157,5 @@ av_cold void ff_h264dsp_init(H264DSPContext *c, const int bit_depth, if (ARCH_ARM) ff_h264dsp_init_arm(c, bit_depth, chroma_format_idc); if (ARCH_PPC) ff_h264dsp_init_ppc(c, bit_depth, chroma_format_idc); if (ARCH_X86) ff_h264dsp_init_x86(c, bit_depth, chroma_format_idc); + if (ARCH_MIPS) ff_h264dsp_init_mips(c, bit_depth, chroma_format_idc); } diff --git a/libavcodec/h264dsp.h b/libavcodec/h264dsp.h index 85ea40b..c232193 100644 --- a/libavcodec/h264dsp.h +++ b/libavcodec/h264dsp.h @@ -126,5 +126,7 @@ void ff_h264dsp_init_ppc(H264DSPContext *c, const int bit_depth, const int chroma_format_idc); void ff_h264dsp_init_x86(H264DSPContext *c, const int bit_depth, const int chroma_format_idc); +void ff_h264dsp_init_mips(H264DSPContext *c, const int bit_depth, + const int chroma_format_idc); #endif /* AVCODEC_H264DSP_H */ diff --git a/libavcodec/mips/Makefile b/libavcodec/mips/Makefile index b41d8c7..eaedd7f 100644 --- a/libavcodec/mips/Makefile +++ b/libavcodec/mips/Makefile @@ -19,4 +19,6 @@ OBJS-$(CONFIG_AAC_DECODER) += mips/aacdec_mips.o \ MIPSDSPR1-OBJS-$(CONFIG_AAC_ENCODER) += mips/aaccoder_mips.o MIPSFPU-OBJS-$(CONFIG_AAC_ENCODER) += mips/iirfilter_mips.o OBJS-$(CONFIG_HEVC_DECODER) += mips/hevcdsp_init_mips.o +OBJS-$(CONFIG_H264DSP) += mips/h264dsp_init_mips.o MSA-OBJS-$(CONFIG_HEVC_DECODER) += mips/hevcdsp_msa.o +MSA-OBJS-$(CONFIG_H264DSP) += mips/h264dsp_msa.o diff --git a/libavcodec/mips/h264dsp_init_mips.c b/libavcodec/mips/h264dsp_init_mips.c new file mode 100644 index 0000000..8d3d760 --- /dev/null +++ b/libavcodec/mips/h264dsp_init_mips.c @@ -0,0 +1,74 @@ +/* + * Copyright (c) 2015 Parag Salasakar (Parag.Salasakar(a)imgtec.com) + * + * This file is part of FFmpeg. + * + * FFmpeg is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * FFmpeg is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with FFmpeg; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +#include "h264dsp_mips.h" + +#if HAVE_MSA +static av_cold void h264dsp_init_msa(H264DSPContext *c, + const int bit_depth, + const int chroma_format_idc) +{ + if (8 == bit_depth) { + c->h264_v_loop_filter_luma = ff_h264_v_lpf_luma_inter_msa; + c->h264_h_loop_filter_luma = ff_h264_h_lpf_luma_inter_msa; + c->h264_h_loop_filter_luma_mbaff = + ff_h264_h_loop_filter_luma_mbaff_msa; + c->h264_v_loop_filter_luma_intra = ff_h264_v_lpf_luma_intra_msa; + c->h264_h_loop_filter_luma_intra = ff_h264_h_lpf_luma_intra_msa; + c->h264_h_loop_filter_luma_mbaff_intra = + ff_h264_h_loop_filter_luma_mbaff_intra_msa; + c->h264_v_loop_filter_chroma = ff_h264_v_lpf_chroma_inter_msa; + + if (chroma_format_idc <= 1) + c->h264_h_loop_filter_chroma = ff_h264_h_lpf_chroma_inter_msa; + else + c->h264_h_loop_filter_chroma = + ff_h264_h_loop_filter_chroma422_msa; + + if (chroma_format_idc > 1) + c->h264_h_loop_filter_chroma_mbaff = + ff_h264_h_loop_filter_chroma422_mbaff_msa; + + c->h264_v_loop_filter_chroma_intra = + ff_h264_v_lpf_chroma_intra_msa; + + if (chroma_format_idc <= 1) + c->h264_h_loop_filter_chroma_intra = + ff_h264_h_lpf_chroma_intra_msa; + + /* Weighted MC */ + c->weight_h264_pixels_tab[0] = ff_weight_h264_pixels16_8_msa; + c->weight_h264_pixels_tab[1] = ff_weight_h264_pixels8_8_msa; + c->weight_h264_pixels_tab[2] = ff_weight_h264_pixels4_8_msa; + + c->biweight_h264_pixels_tab[0] = ff_biweight_h264_pixels16_8_msa; + c->biweight_h264_pixels_tab[1] = ff_biweight_h264_pixels8_8_msa; + c->biweight_h264_pixels_tab[2] = ff_biweight_h264_pixels4_8_msa; + } // if (8 == bit_depth) +} +#endif // #if HAVE_MSA + +av_cold void ff_h264dsp_init_mips(H264DSPContext *c, const int bit_depth, + const int chroma_format_idc) +{ +#if HAVE_MSA + h264dsp_init_msa(c, bit_depth, chroma_format_idc); +#endif // #if HAVE_MSA +} diff --git a/libavcodec/mips/h264dsp_mips.h b/libavcodec/mips/h264dsp_mips.h new file mode 100644 index 0000000..df9b0b2 --- /dev/null +++ b/libavcodec/mips/h264dsp_mips.h @@ -0,0 +1,71 @@ +/* + * Copyright (c) 2015 Parag Salasakar (Parag.Salasakar(a)imgtec.com) + * + * This file is part of FFmpeg. + * + * FFmpeg is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * FFmpeg is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with FFmpeg; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +#ifndef H264_DSP_MIPS_H +#define H264_DSP_MIPS_H + +#include "libavcodec/h264.h" + +void ff_h264_h_lpf_luma_inter_msa(uint8_t *src, int stride, + int alpha, int beta, int8_t *tc0); +void ff_h264_v_lpf_luma_inter_msa(uint8_t *src, int stride, + int alpha, int beta, int8_t *tc0); +void ff_h264_h_lpf_chroma_inter_msa(uint8_t *src, int stride, + int alpha, int beta, int8_t *tc0); +void ff_h264_v_lpf_chroma_inter_msa(uint8_t *src, int stride, + int alpha, int beta, int8_t *tc0); +void ff_h264_h_loop_filter_chroma422_msa(uint8_t *src, int32_t stride, + int32_t alpha, int32_t beta, + int8_t *tc0); +void ff_h264_h_loop_filter_chroma422_mbaff_msa(uint8_t *src, int32_t stride, + int32_t alpha, int32_t beta, + int8_t *tc0); +void ff_h264_h_loop_filter_luma_mbaff_msa(uint8_t *src, int32_t stride, + int32_t alpha, int32_t beta, + int8_t *tc0); + +void ff_h264_h_lpf_luma_intra_msa(uint8_t *src, int stride, + int alpha, int beta); +void ff_h264_v_lpf_luma_intra_msa(uint8_t *src, int stride, + int alpha, int beta); +void ff_h264_h_lpf_chroma_intra_msa(uint8_t *src, int stride, + int alpha, int beta); +void ff_h264_v_lpf_chroma_intra_msa(uint8_t *src, int stride, + int alpha, int beta); +void ff_h264_h_loop_filter_luma_mbaff_intra_msa(uint8_t *src, int stride, + int alpha, int beta); + +void ff_biweight_h264_pixels16_8_msa(uint8_t *dst, uint8_t *src, + int stride, int height, int log2_denom, + int weightd, int weights, int offset); +void ff_biweight_h264_pixels8_8_msa(uint8_t *dst, uint8_t *src, + int stride, int height, int log2_denom, + int weightd, int weights, int offset); +void ff_biweight_h264_pixels4_8_msa(uint8_t *dst, uint8_t *src, + int stride, int height, int log2_denom, + int weightd, int weights, int offset); +void ff_weight_h264_pixels16_8_msa(uint8_t *src, int stride, int height, + int log2_denom, int weight, int offset); +void ff_weight_h264_pixels8_8_msa(uint8_t *src, int stride, int height, + int log2_denom, int weight, int offset); +void ff_weight_h264_pixels4_8_msa(uint8_t *src, int stride, int height, + int log2_denom, int weight, int offset); + +#endif // #ifndef H264_DSP_MIPS_H diff --git a/libavcodec/mips/h264dsp_msa.c b/libavcodec/mips/h264dsp_msa.c new file mode 100644 index 0000000..8aa8aeb --- /dev/null +++ b/libavcodec/mips/h264dsp_msa.c @@ -0,0 +1,3037 @@ +/* + * Copyright (c) 2015 Parag Salasakar (Parag.Salasakar(a)imgtec.com) + * + * This file is part of FFmpeg. + * + * FFmpeg is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * FFmpeg is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with FFmpeg; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +#include "libavutil/mips/generic_macros_msa.h" +#include "h264dsp_mips.h" + +static void avc_wgt_4x2_msa(uint8_t *data, + int32_t stride, + int32_t log2_denom, + int32_t src_weight, + int32_t offset_in) +{ + uint32_t data0, data1; + v16u8 zero = { 0 }; + v16u8 vec0, vec1; + v4i32 src0, src1; + v16i8 res0, res1; + v8i16 temp0, temp1; + v8i16 wgt, denom, offset; + + offset_in <<= (log2_denom); + + if (log2_denom) { + offset_in += (1 << (log2_denom - 1)); + } + + wgt = __msa_fill_h(src_weight); + offset = __msa_fill_h(offset_in); + denom = __msa_fill_h(log2_denom); + + data0 = LOAD_WORD(data); + data1 = LOAD_WORD(data + stride); + + src0 = __msa_fill_w(data0); + src1 = __msa_fill_w(data1); + + ILVR_B_2VECS_UB(src0, src1, zero, zero, vec0, vec1); + + temp0 = wgt * (v8i16) vec0; + temp1 = wgt * (v8i16) vec1; + + temp0 = __msa_adds_s_h(temp0, offset); + temp1 = __msa_adds_s_h(temp1, offset); + + temp0 = __msa_maxi_s_h(temp0, 0); + temp1 = __msa_maxi_s_h(temp1, 0); + + temp0 = __msa_srl_h(temp0, denom); + temp1 = __msa_srl_h(temp1, denom); + + temp0 = (v8i16) __msa_sat_u_h((v8u16) temp0, 7); + temp1 = (v8i16) __msa_sat_u_h((v8u16) temp1, 7); + + res0 = __msa_pckev_b((v16i8) temp0, (v16i8) temp0); + res1 = __msa_pckev_b((v16i8) temp1, (v16i8) temp1); + + data0 = __msa_copy_u_w((v4i32) res0, 0); + data1 = __msa_copy_u_w((v4i32) res1, 0); + + STORE_WORD(data, data0); + data += stride; + STORE_WORD(data, data1); +} + +static void avc_wgt_4x4multiple_msa(uint8_t *data, + int32_t stride, + int32_t height, + int32_t log2_denom, + int32_t src_weight, + int32_t offset_in) +{ + uint8_t cnt; + uint32_t data0, data1, data2, data3; + v16u8 zero = { 0 }; + v4i32 src0, src1, src2, src3; + v8u16 temp0, temp1, temp2, temp3; + v8i16 wgt, denom, offset; + + offset_in <<= (log2_denom); + + if (log2_denom) { + offset_in += (1 << (log2_denom - 1)); + } + + wgt = __msa_fill_h(src_weight); + offset = __msa_fill_h(offset_in); + denom = __msa_fill_h(log2_denom); + + for (cnt = height / 4; cnt--;) { + LOAD_4WORDS_WITH_STRIDE(data, stride, data0, data1, data2, data3); + + src0 = __msa_fill_w(data0); + src1 = __msa_fill_w(data1); + src2 = __msa_fill_w(data2); + src3 = __msa_fill_w(data3); + + ILVR_B_4VECS_UH(src0, src1, src2, src3, zero, zero, zero, zero, + temp0, temp1, temp2, temp3); + + temp0 *= wgt; + temp1 *= wgt; + temp2 *= wgt; + temp3 *= wgt; + + ADDS_S_H_4VECS_UH(temp0, offset, temp1, offset, + temp2, offset, temp3, offset, + temp0, temp1, temp2, temp3); + + MAXI_S_H_4VECS_UH(temp0, temp1, temp2, temp3, 0); + + SRL_H_4VECS_UH(temp0, temp1, temp2, temp3, + temp0, temp1, temp2, temp3, denom); + + SAT_U_H_4VECS_UH(temp0, temp1, temp2, temp3, 7); + + PCKEV_B_STORE_4_BYTES_4(temp0, temp1, temp2, temp3, data, stride); + data += (4 * stride); + } +} + +static void avc_wgt_4width_msa(uint8_t *data, + int32_t stride, + int32_t height, + int32_t log2_denom, + int32_t src_weight, + int32_t offset_in) +{ + if (2 == height) { + avc_wgt_4x2_msa(data, stride, log2_denom, src_weight, offset_in); + } else { + avc_wgt_4x4multiple_msa(data, stride, height, log2_denom, + src_weight, offset_in); + } +} + +static void avc_wgt_8width_msa(uint8_t *data, + int32_t stride, + int32_t height, + int32_t log2_denom, + int32_t src_weight, + int32_t offset_in) +{ + uint8_t cnt; + v16u8 zero = { 0 }; + v16u8 src0, src1, src2, src3; + v8u16 src0_r, src1_r, src2_r, src3_r; + v8u16 temp0, temp1, temp2, temp3; + v8u16 wgt; + v8i16 denom, offset; + + offset_in <<= (log2_denom); + + if (log2_denom) { + offset_in += (1 << (log2_denom - 1)); + } + + wgt = (v8u16) __msa_fill_h(src_weight); + offset = __msa_fill_h(offset_in); + denom = __msa_fill_h(log2_denom); + + for (cnt = height / 4; cnt--;) { + LOAD_4VECS_UB(data, stride, src0, src1, src2, src3); + + ILVR_B_4VECS_UH(src0, src1, src2, src3, zero, zero, zero, zero, + src0_r, src1_r, src2_r, src3_r); + + temp0 = wgt * src0_r; + temp1 = wgt * src1_r; + temp2 = wgt * src2_r; + temp3 = wgt * src3_r; + + ADDS_S_H_4VECS_UH(temp0, offset, temp1, offset, + temp2, offset, temp3, offset, + temp0, temp1, temp2, temp3); + + MAXI_S_H_4VECS_UH(temp0, temp1, temp2, temp3, 0); + + SRL_H_4VECS_UH(temp0, temp1, temp2, temp3, + temp0, temp1, temp2, temp3, denom); + + SAT_U_H_4VECS_UH(temp0, temp1, temp2, temp3, 7); + + PCKEV_B_STORE_8_BYTES_4(temp0, temp1, temp2, temp3, data, stride); + data += (4 * stride); + } +} + +static void avc_wgt_16width_msa(uint8_t *data, + int32_t stride, + int32_t height, + int32_t log2_denom, + int32_t src_weight, + int32_t offset_in) +{ + uint8_t cnt; + v16u8 zero = { 0 }; + v16u8 src0, src1, src2, src3; + v16u8 dst0, dst1, dst2, dst3; + v8u16 src0_l, src1_l, src2_l, src3_l; + v8u16 src0_r, src1_r, src2_r, src3_r; + v8u16 temp0, temp1, temp2, temp3, temp4, temp5, temp6, temp7; + v8u16 wgt; + v8i16 denom, offset; + + offset_in <<= (log2_denom); + + if (log2_denom) { + offset_in += (1 << (log2_denom - 1)); + } + + wgt = (v8u16) __msa_fill_h(src_weight); + offset = __msa_fill_h(offset_in); + denom = __msa_fill_h(log2_denom); + + for (cnt = height / 4; cnt--;) { + LOAD_4VECS_UB(data, stride, src0, src1, src2, src3); + + ILV_B_LRLR_UH(src0, zero, src1, zero, src0_l, src0_r, src1_l, src1_r); + ILV_B_LRLR_UH(src2, zero, src3, zero, src2_l, src2_r, src3_l, src3_r); + + temp0 = wgt * src0_r; + temp1 = wgt * src0_l; + temp2 = wgt * src1_r; + temp3 = wgt * src1_l; + temp4 = wgt * src2_r; + temp5 = wgt * src2_l; + temp6 = wgt * src3_r; + temp7 = wgt * src3_l; + + ADDS_S_H_4VECS_UH(temp0, offset, temp1, offset, + temp2, offset, temp3, offset, + temp0, temp1, temp2, temp3); + + ADDS_S_H_4VECS_UH(temp4, offset, temp5, offset, + temp6, offset, temp7, offset, + temp4, temp5, temp6, temp7); + + MAXI_S_H_4VECS_UH(temp0, temp1, temp2, temp3, 0); + MAXI_S_H_4VECS_UH(temp4, temp5, temp6, temp7, 0); + + SRL_H_4VECS_UH(temp0, temp1, temp2, temp3, + temp0, temp1, temp2, temp3, denom); + + SRL_H_4VECS_UH(temp4, temp5, temp6, temp7, + temp4, temp5, temp6, temp7, denom); + + SAT_U_H_4VECS_UH(temp0, temp1, temp2, temp3, 7); + SAT_U_H_4VECS_UH(temp4, temp5, temp6, temp7, 7); + + PCKEV_B_4VECS_UB(temp1, temp3, temp5, temp7, temp0, temp2, temp4, temp6, + dst0, dst1, dst2, dst3); + + STORE_4VECS_UB(data, stride, dst0, dst1, dst2, dst3); + data += 4 * stride; + } +} + +static void avc_biwgt_4x2_msa(uint8_t *src, + int32_t src_stride, + uint8_t *dst, + int32_t dst_stride, + int32_t log2_denom, + int32_t src_weight, + int32_t dst_weight, + int32_t offset_in) +{ + uint32_t load0, load1, out0, out1; + v16i8 src_wgt, dst_wgt, wgt; + v16i8 src0, src1, dst0, dst1; + v8i16 temp0, temp1, denom, offset, add_val; + int32_t val = 128 * (src_weight + dst_weight); + + offset_in = ((offset_in + 1) | 1) << log2_denom; + + src_wgt = __msa_fill_b(src_weight); + dst_wgt = __msa_fill_b(dst_weight); + offset = __msa_fill_h(offset_in); + denom = __msa_fill_h(log2_denom + 1); + add_val = __msa_fill_h(val); + offset += add_val; + + wgt = __msa_ilvev_b(dst_wgt, src_wgt); + + load0 = LOAD_WORD(src); + src += src_stride; + load1 = LOAD_WORD(src); + + src0 = (v16i8) __msa_fill_w(load0); + src1 = (v16i8) __msa_fill_w(load1); + + load0 = LOAD_WORD(dst); + load1 = LOAD_WORD(dst + dst_stride); + + dst0 = (v16i8) __msa_fill_w(load0); + dst1 = (v16i8) __msa_fill_w(load1); + + XORI_B_4VECS_SB(src0, src1, dst0, dst1, + src0, src1, dst0, dst1, 128); + + ILVR_B_2VECS_SH(src0, src1, dst0, dst1, temp0, temp1); + + temp0 = __msa_dpadd_s_h(offset, wgt, (v16i8) temp0); + temp1 = __msa_dpadd_s_h(offset, wgt, (v16i8) temp1); + + temp0 >>= denom; + temp1 >>= denom; + + temp0 = CLIP_UNSIGNED_CHAR_H(temp0); + temp1 = CLIP_UNSIGNED_CHAR_H(temp1); + + dst0 = __msa_pckev_b((v16i8) temp0, (v16i8) temp0); + dst1 = __msa_pckev_b((v16i8) temp1, (v16i8) temp1); + + out0 = __msa_copy_u_w((v4i32) dst0, 0); + out1 = __msa_copy_u_w((v4i32) dst1, 0); + + STORE_WORD(dst, out0); + dst += dst_stride; + STORE_WORD(dst, out1); +} + +static void avc_biwgt_4x4multiple_msa(uint8_t *src, + int32_t src_stride, + uint8_t *dst, + int32_t dst_stride, + int32_t height, + int32_t log2_denom, + int32_t src_weight, + int32_t dst_weight, + int32_t offset_in) +{ + uint8_t cnt; + uint32_t load0, load1, load2, load3; + v16i8 src_wgt, dst_wgt, wgt; + v16i8 src0, src1, src2, src3; + v16i8 dst0, dst1, dst2, dst3; + v8i16 temp0, temp1, temp2, temp3; + v8i16 denom, offset, add_val; + int32_t val = 128 * (src_weight + dst_weight); + + offset_in = ((offset_in + 1) | 1) << log2_denom; + + src_wgt = __msa_fill_b(src_weight); + dst_wgt = __msa_fill_b(dst_weight); + offset = __msa_fill_h(offset_in); + denom = __msa_fill_h(log2_denom + 1); + add_val = __msa_fill_h(val); + offset += add_val; + + wgt = __msa_ilvev_b(dst_wgt, src_wgt); + + for (cnt = height / 4; cnt--;) { + LOAD_4WORDS_WITH_STRIDE(src, src_stride, load0, load1, load2, load3); + src += (4 * src_stride); + + src0 = (v16i8) __msa_fill_w(load0); + src1 = (v16i8) __msa_fill_w(load1); + src2 = (v16i8) __msa_fill_w(load2); + src3 = (v16i8) __msa_fill_w(load3); + + LOAD_4WORDS_WITH_STRIDE(dst, dst_stride, load0, load1, load2, load3); + + dst0 = (v16i8) __msa_fill_w(load0); + dst1 = (v16i8) __msa_fill_w(load1); + dst2 = (v16i8) __msa_fill_w(load2); + dst3 = (v16i8) __msa_fill_w(load3); + + XORI_B_4VECS_SB(src0, src1, src2, src3, + src0, src1, src2, src3, 128); + + XORI_B_4VECS_SB(dst0, dst1, dst2, dst3, + dst0, dst1, dst2, dst3, 128); + + ILVR_B_4VECS_SH(src0, src1, src2, src3, dst0, dst1, dst2, dst3, + temp0, temp1, temp2, temp3); + + temp0 = __msa_dpadd_s_h(offset, wgt, (v16i8) temp0); + temp1 = __msa_dpadd_s_h(offset, wgt, (v16i8) temp1); + temp2 = __msa_dpadd_s_h(offset, wgt, (v16i8) temp2); + temp3 = __msa_dpadd_s_h(offset, wgt, (v16i8) temp3); + + SRA_4VECS(temp0, temp1, temp2, temp3, + temp0, temp1, temp2, temp3, denom); + + temp0 = CLIP_UNSIGNED_CHAR_H(temp0); + temp1 = CLIP_UNSIGNED_CHAR_H(temp1); + temp2 = CLIP_UNSIGNED_CHAR_H(temp2); + temp3 = CLIP_UNSIGNED_CHAR_H(temp3); + + PCKEV_B_STORE_4_BYTES_4(temp0, temp1, temp2, temp3, dst, dst_stride); + dst += (4 * dst_stride); + } +} + +static void avc_biwgt_4width_msa(uint8_t *src, + int32_t src_stride, + uint8_t *dst, + int32_t dst_stride, + int32_t height, + int32_t log2_denom, + int32_t src_weight, + int32_t dst_weight, + int32_t offset_in) +{ + if (2 == height) { + avc_biwgt_4x2_msa(src, src_stride, dst, dst_stride, + log2_denom, src_weight, dst_weight, + offset_in); + } else { + avc_biwgt_4x4multiple_msa(src, src_stride, dst, dst_stride, + height, log2_denom, src_weight, + dst_weight, offset_in); + } +} + +static void avc_biwgt_8width_msa(uint8_t *src, int32_t src_stride, + uint8_t *dst, int32_t dst_stride, + int32_t height, int32_t log2_denom, + int32_t src_weight, int32_t dst_weight, + int32_t offset_in) +{ + uint8_t cnt; + v16i8 src_wgt, dst_wgt, wgt; + v16i8 src0, src1, src2, src3; + v16i8 dst0, dst1, dst2, dst3; + v8i16 temp0, temp1, temp2, temp3; + v8i16 denom, offset, add_val; + int32_t val = 128 * (src_weight + dst_weight); + + offset_in = ((offset_in + 1) | 1) << log2_denom; + + src_wgt = __msa_fill_b(src_weight); + dst_wgt = __msa_fill_b(dst_weight); + offset = __msa_fill_h(offset_in); + denom = __msa_fill_h(log2_denom + 1); + add_val = __msa_fill_h(val); + offset += add_val; + + wgt = __msa_ilvev_b(dst_wgt, src_wgt); + + for (cnt = height / 4; cnt--;) { + LOAD_4VECS_SB(src, src_stride, src0, src1, src2, src3); + src += (4 * src_stride); + + LOAD_4VECS_SB(dst, dst_stride, dst0, dst1, dst2, dst3); + + XORI_B_4VECS_SB(src0, src1, src2, src3, + src0, src1, src2, src3, 128); + + XORI_B_4VECS_SB(dst0, dst1, dst2, dst3, + dst0, dst1, dst2, dst3, 128); + + ILVR_B_4VECS_SH(src0, src1, src2, src3, dst0, dst1, dst2, dst3, + temp0, temp1, temp2, temp3); + + temp0 = __msa_dpadd_s_h(offset, wgt, (v16i8) temp0); + temp1 = __msa_dpadd_s_h(offset, wgt, (v16i8) temp1); + temp2 = __msa_dpadd_s_h(offset, wgt, (v16i8) temp2); + temp3 = __msa_dpadd_s_h(offset, wgt, (v16i8) temp3); + + SRA_4VECS(temp0, temp1, temp2, temp3, + temp0, temp1, temp2, temp3, denom); + + temp0 = CLIP_UNSIGNED_CHAR_H(temp0); + temp1 = CLIP_UNSIGNED_CHAR_H(temp1); + temp2 = CLIP_UNSIGNED_CHAR_H(temp2); + temp3 = CLIP_UNSIGNED_CHAR_H(temp3); + + PCKEV_B_STORE_8_BYTES_4(temp0, temp1, temp2, temp3, dst, dst_stride); + dst += 4 * dst_stride; + } +} + +static void avc_biwgt_16width_msa(uint8_t *src, int32_t src_stride, + uint8_t *dst, int32_t dst_stride, + int32_t height, int32_t log2_denom, + int32_t src_weight, int32_t dst_weight, + int32_t offset_in) +{ + uint8_t cnt; + v16i8 src_wgt, dst_wgt, wgt; + v16i8 src0, src1, src2, src3; + v16i8 dst0, dst1, dst2, dst3; + v8i16 temp0, temp1, temp2, temp3, temp4, temp5, temp6, temp7; + v8i16 denom, offset, add_val; + int32_t val = 128 * (src_weight + dst_weight); + + offset_in = ((offset_in + 1) | 1) << log2_denom; + + src_wgt = __msa_fill_b(src_weight); + dst_wgt = __msa_fill_b(dst_weight); + offset = __msa_fill_h(offset_in); + denom = __msa_fill_h(log2_denom + 1); + add_val = __msa_fill_h(val); + offset += add_val; + + wgt = __msa_ilvev_b(dst_wgt, src_wgt); + + for (cnt = height / 4; cnt--;) { + LOAD_4VECS_SB(src, src_stride, src0, src1, src2, src3); + src += (4 * src_stride); + + LOAD_4VECS_SB(dst, dst_stride, dst0, dst1, dst2, dst3); + + XORI_B_4VECS_SB(src0, src1, src2, src3, + src0, src1, src2, src3, 128); + + XORI_B_4VECS_SB(dst0, dst1, dst2, dst3, + dst0, dst1, dst2, dst3, 128); + + ILV_B_LRLR_SH(src0, dst0, src1, dst1, temp1, temp0, temp3, temp2); + ILV_B_LRLR_SH(src2, dst2, src3, dst3, temp5, temp4, temp7, temp6); + + temp0 = __msa_dpadd_s_h(offset, wgt, (v16i8) temp0); + temp1 = __msa_dpadd_s_h(offset, wgt, (v16i8) temp1); + temp2 = __msa_dpadd_s_h(offset, wgt, (v16i8) temp2); + temp3 = __msa_dpadd_s_h(offset, wgt, (v16i8) temp3); + temp4 = __msa_dpadd_s_h(offset, wgt, (v16i8) temp4); + temp5 = __msa_dpadd_s_h(offset, wgt, (v16i8) temp5); + temp6 = __msa_dpadd_s_h(offset, wgt, (v16i8) temp6); + temp7 = __msa_dpadd_s_h(offset, wgt, (v16i8) temp7); + + SRA_4VECS(temp0, temp1, temp2, temp3, + temp0, temp1, temp2, temp3, denom); + SRA_4VECS(temp4, temp5, temp6, temp7, + temp4, temp5, temp6, temp7, denom); + + temp0 = CLIP_UNSIGNED_CHAR_H(temp0); + temp1 = CLIP_UNSIGNED_CHAR_H(temp1); + temp2 = CLIP_UNSIGNED_CHAR_H(temp2); + temp3 = CLIP_UNSIGNED_CHAR_H(temp3); + temp4 = CLIP_UNSIGNED_CHAR_H(temp4); + temp5 = CLIP_UNSIGNED_CHAR_H(temp5); + temp6 = CLIP_UNSIGNED_CHAR_H(temp6); + temp7 = CLIP_UNSIGNED_CHAR_H(temp7); + + PCKEV_B_4VECS_SB(temp1, temp3, temp5, temp7, temp0, temp2, temp4, temp6, + dst0, dst1, dst2, dst3); + + STORE_4VECS_SB(dst, dst_stride, dst0, dst1, dst2, dst3); + dst += 4 * dst_stride; + } +} + +#define AVC_LOOP_FILTER_P0P1P2_OR_Q0Q1Q2(p3_or_q3_org_in, p0_or_q0_org_in, \ + q3_or_p3_org_in, p1_or_q1_org_in, \ + p2_or_q2_org_in, q1_or_p1_org_in, \ + p0_or_q0_out, \ + p1_or_q1_out, p2_or_q2_out) \ +{ \ + v8i16 threshold; \ + v8i16 const3 = __msa_ldi_h(3); \ + \ + threshold = (p0_or_q0_org_in) + (q3_or_p3_org_in); \ + threshold += (p1_or_q1_org_in); \ + \ + (p0_or_q0_out) = threshold << 1; \ + (p0_or_q0_out) += (p2_or_q2_org_in); \ + (p0_or_q0_out) += (q1_or_p1_org_in); \ + (p0_or_q0_out) = __msa_srari_h((p0_or_q0_out), 3); \ + \ + (p1_or_q1_out) = (p2_or_q2_org_in) + threshold; \ + (p1_or_q1_out) = __msa_srari_h((p1_or_q1_out), 2); \ + \ + (p2_or_q2_out) = (p2_or_q2_org_in) * const3; \ + (p2_or_q2_out) += (p3_or_q3_org_in); \ + (p2_or_q2_out) += (p3_or_q3_org_in); \ + (p2_or_q2_out) += threshold; \ + (p2_or_q2_out) = __msa_srari_h((p2_or_q2_out), 3); \ +} + +/* data[-u32_img_width] = (uint8_t)((2 * p1 + p0 + q1 + 2) >> 2); */ +#define AVC_LOOP_FILTER_P0_OR_Q0(p0_or_q0_org_in, q1_or_p1_org_in, \ + p1_or_q1_org_in, p0_or_q0_out) \ +{ \ + (p0_or_q0_out) = (p0_or_q0_org_in) + (q1_or_p1_org_in); \ + (p0_or_q0_out) += (p1_or_q1_org_in); \ + (p0_or_q0_out) += (p1_or_q1_org_in); \ + (p0_or_q0_out) = __msa_srari_h((p0_or_q0_out), 2); \ +} + +#define AVC_LOOP_FILTER_P1_OR_Q1(p0_or_q0_org_in, q0_or_p0_org_in, \ + p1_or_q1_org_in, p2_or_q2_org_in, \ + negate_tc_in, tc_in, p1_or_q1_out) \ +{ \ + v8i16 clip3, temp; \ + \ + clip3 = (v8i16) __msa_aver_u_h((v8u16) (p0_or_q0_org_in), \ + (v8u16) (q0_or_p0_org_in)); \ + temp = (p1_or_q1_org_in) << 1; \ + clip3 = clip3 - temp; \ + clip3 = __msa_ave_s_h((p2_or_q2_org_in), clip3); \ + clip3 = CLIP_MIN_TO_MAX_H(clip3, negate_tc_in, tc_in); \ + (p1_or_q1_out) = (p1_or_q1_org_in) + clip3; \ +} + +#define AVC_LOOP_FILTER_P0Q0(q0_or_p0_org_in, p0_or_q0_org_in, \ + p1_or_q1_org_in, q1_or_p1_org_in, \ + negate_threshold_in, threshold_in, \ + p0_or_q0_out, q0_or_p0_out) \ +{ \ + v8i16 q0_sub_p0, p1_sub_q1, delta; \ + \ + q0_sub_p0 = (q0_or_p0_org_in) - (p0_or_q0_org_in); \ + p1_sub_q1 = (p1_or_q1_org_in) - (q1_or_p1_org_in); \ + q0_sub_p0 <<= 2; \ + p1_sub_q1 += 4; \ + delta = q0_sub_p0 + p1_sub_q1; \ + delta >>= 3; \ + \ + delta = CLIP_MIN_TO_MAX_H(delta, negate_threshold_in, \ + threshold_in); \ + \ + (p0_or_q0_out) = (p0_or_q0_org_in) + delta; \ + p0_or_q0_out = CLIP_UNSIGNED_CHAR_H(p0_or_q0_out); \ + \ + (q0_or_p0_out) = (q0_or_p0_org_in) - delta; \ + q0_or_p0_out = CLIP_UNSIGNED_CHAR_H(q0_or_p0_out); \ +} + +#define AVC_LPF_H_CHROMA_422(src, stride, tc_val, alpha, beta, res) \ +{ \ + uint32_t load0, load1, load2, load3; \ + v16u8 src0 = { 0 }; \ + v16u8 src1 = { 0 }; \ + v16u8 src2 = { 0 }; \ + v16u8 src3 = { 0 }; \ + v16u8 p0_asub_q0, p1_asub_p0, q1_asub_q0; \ + v16u8 is_less_than, is_less_than_alpha, is_less_than_beta; \ + v8i16 tc, q0_sub_p0, p1_sub_q1, delta; \ + v8i16 res0_r, res1_r; \ + v16i8 zeros = { 0 }; \ + v16u8 res0, res1; \ + \ + LOAD_4WORDS_WITH_STRIDE((src - 2), stride, \ + load0, load1, load2, load3); \ + \ + src0 = (v16u8) __msa_insert_w((v4i32) src0, 0, load0); \ + src1 = (v16u8) __msa_insert_w((v4i32) src1, 0, load1); \ + src2 = (v16u8) __msa_insert_w((v4i32) src2, 0, load2); \ + src3 = (v16u8) __msa_insert_w((v4i32) src3, 0, load3); \ + \ + TRANSPOSE4x4_B_UB(src0, src1, src2, src3, \ + src0, src1, src2, src3); \ + \ + p0_asub_q0 = __msa_asub_u_b(src2, src1); \ + p1_asub_p0 = __msa_asub_u_b(src1, src0); \ + q1_asub_q0 = __msa_asub_u_b(src2, src3); \ + \ + tc = __msa_fill_h(tc_val); \ + \ + is_less_than_alpha = (p0_asub_q0 < alpha); \ + is_less_than_beta = (p1_asub_p0 < beta); \ + is_less_than = is_less_than_alpha & is_less_than_beta; \ + is_less_than_beta = (q1_asub_q0 < beta); \ + is_less_than = is_less_than_beta & is_less_than; \ + \ + q0_sub_p0 = (v8i16) __msa_ilvr_b((v16i8) src2, (v16i8) src1); \ + p1_sub_q1 = (v8i16) __msa_ilvr_b((v16i8) src0, (v16i8) src3); \ + \ + q0_sub_p0 = __msa_hsub_u_h((v16u8) q0_sub_p0, (v16u8) q0_sub_p0); \ + p1_sub_q1 = __msa_hsub_u_h((v16u8) p1_sub_q1, (v16u8) p1_sub_q1); \ + \ + q0_sub_p0 <<= 2; \ + delta = q0_sub_p0 + p1_sub_q1; \ + delta = __msa_srari_h(delta, 3); \ + \ + delta = CLIP_MIN_TO_MAX_H(delta, -tc, tc); \ + \ + res0_r = (v8i16) __msa_ilvr_b(zeros, (v16i8) src1); \ + res1_r = (v8i16) __msa_ilvr_b(zeros, (v16i8) src2); \ + \ + res0_r += delta; \ + res1_r -= delta; \ + \ + res0_r = CLIP_UNSIGNED_CHAR_H(res0_r); \ + res1_r = CLIP_UNSIGNED_CHAR_H(res1_r); \ + \ + res0 = (v16u8) __msa_pckev_b((v16i8) res0_r, (v16i8) res0_r); \ + res1 = (v16u8) __msa_pckev_b((v16i8) res1_r, (v16i8) res1_r); \ + \ + res0 = __msa_bmnz_v(src1, res0, is_less_than); \ + res1 = __msa_bmnz_v(src2, res1, is_less_than); \ + \ + res = (v16u8) __msa_ilvr_b((v16i8) res1, (v16i8) res0); \ +} + +#define TRANSPOSE2x4_B_UB(in0, in1, \ + out0, out1, out2, out3) \ +{ \ + v16i8 zero_m = { 0 }; \ + \ + out0 = (v16u8) __msa_ilvr_b((v16i8) in1, (v16i8) in0); \ + \ + out1 = (v16u8) __msa_sldi_b(zero_m, (v16i8) out0, 2); \ + out2 = (v16u8) __msa_sldi_b(zero_m, (v16i8) out1, 2); \ + out3 = (v16u8) __msa_sldi_b(zero_m, (v16i8) out2, 2); \ +} + +#define AVC_LPF_H_2BYTE_CHROMA_422(src, stride, \ + tc_val, alpha, beta, res) \ +{ \ + uint32_t load0, load1; \ + v16u8 src0 = { 0 }; \ + v16u8 src1 = { 0 }; \ + v16u8 src2 = { 0 }; \ + v16u8 src3 = { 0 }; \ + v16u8 p0_asub_q0, p1_asub_p0, q1_asub_q0; \ + v16u8 is_less_than, is_less_than_alpha, is_less_than_beta; \ + v8i16 tc, q0_sub_p0, p1_sub_q1, delta; \ + v8i16 res0_r, res1_r; \ + v16i8 zeros = { 0 }; \ + v16u8 res0, res1; \ + \ + load0 = LOAD_WORD(src - 2); \ + load1 = LOAD_WORD(src - 2 + stride); \ + \ + src0 = (v16u8) __msa_insert_w((v4i32) src0, 0, load0); \ + src1 = (v16u8) __msa_insert_w((v4i32) src1, 0, load1); \ + \ + TRANSPOSE2x4_B_UB(src0, src1, \ + src0, src1, src2, src3); \ + \ + p0_asub_q0 = __msa_asub_u_b(src2, src1); \ + p1_asub_p0 = __msa_asub_u_b(src1, src0); \ + q1_asub_q0 = __msa_asub_u_b(src2, src3); \ + \ + tc = __msa_fill_h(tc_val); \ + \ + is_less_than_alpha = (p0_asub_q0 < alpha); \ + is_less_than_beta = (p1_asub_p0 < beta); \ + is_less_than = is_less_than_alpha & is_less_than_beta; \ + is_less_than_beta = (q1_asub_q0 < beta); \ + is_less_than = is_less_than_beta & is_less_than; \ + \ + q0_sub_p0 = (v8i16) __msa_ilvr_b((v16i8) src2, (v16i8) src1); \ + p1_sub_q1 = (v8i16) __msa_ilvr_b((v16i8) src0, (v16i8) src3); \ + \ + q0_sub_p0 = __msa_hsub_u_h((v16u8) q0_sub_p0, (v16u8) q0_sub_p0); \ + p1_sub_q1 = __msa_hsub_u_h((v16u8) p1_sub_q1, (v16u8) p1_sub_q1); \ + \ + q0_sub_p0 <<= 2; \ + delta = q0_sub_p0 + p1_sub_q1; \ + delta = __msa_srari_h(delta, 3); \ + \ + delta = CLIP_MIN_TO_MAX_H(delta, -tc, tc); \ + \ + res0_r = (v8i16) __msa_ilvr_b(zeros, (v16i8) src1); \ + res1_r = (v8i16) __msa_ilvr_b(zeros, (v16i8) src2); \ + \ + res0_r += delta; \ + res1_r -= delta; \ + \ + res0_r = CLIP_UNSIGNED_CHAR_H(res0_r); \ + res1_r = CLIP_UNSIGNED_CHAR_H(res1_r); \ + \ + res0 = (v16u8) __msa_pckev_b((v16i8) res0_r, (v16i8) res0_r); \ + res1 = (v16u8) __msa_pckev_b((v16i8) res1_r, (v16i8) res1_r); \ + \ + res0 = __msa_bmnz_v(src1, res0, is_less_than); \ + res1 = __msa_bmnz_v(src2, res1, is_less_than); \ + \ + res = (v16u8) __msa_ilvr_b((v16i8) res1, (v16i8) res0); \ +} + +static void avc_loopfilter_luma_intra_edge_hor_msa(uint8_t *data, + uint8_t alpha_in, + uint8_t beta_in, + uint32_t img_width) +{ + v16u8 p2_asub_p0, q2_asub_q0, p0_asub_q0; + v16u8 alpha, beta; + v16u8 is_less_than, is_less_than_beta, negate_is_less_than_beta; + v16u8 p2, p1, p0, q0, q1, q2; + v16u8 p3_org, p2_org, p1_org, p0_org, q0_org, q1_org, q2_org, q3_org; + v8i16 p1_org_r, p0_org_r, q0_org_r, q1_org_r; + v8i16 p1_org_l, p0_org_l, q0_org_l, q1_org_l; + v8i16 p2_r = { 0 }; + v8i16 p1_r = { 0 }; + v8i16 p0_r = { 0 }; + v8i16 q0_r = { 0 }; + v8i16 q1_r = { 0 }; + v8i16 q2_r = { 0 }; + v8i16 p2_l = { 0 }; + v8i16 p1_l = { 0 }; + v8i16 p0_l = { 0 }; + v8i16 q0_l = { 0 }; + v8i16 q1_l = { 0 }; + v8i16 q2_l = { 0 }; + v16u8 tmp_flag; + v16i8 zero = { 0 }; + + alpha = (v16u8) __msa_fill_b(alpha_in); + beta = (v16u8) __msa_fill_b(beta_in); + + p1_org = LOAD_UB(data - (img_width << 1)); + p0_org = LOAD_UB(data - img_width); + q0_org = LOAD_UB(data); + q1_org = LOAD_UB(data + img_width); + + { + v16u8 p1_asub_p0, q1_asub_q0, is_less_than_alpha; + + p0_asub_q0 = __msa_asub_u_b(p0_org, q0_org); + p1_asub_p0 = __msa_asub_u_b(p1_org, p0_org); + q1_asub_q0 = __msa_asub_u_b(q1_org, q0_org); + + is_less_than_alpha = (p0_asub_q0 < alpha); + is_less_than_beta = (p1_asub_p0 < beta); + is_less_than = is_less_than_beta & is_less_than_alpha; + is_less_than_beta = (q1_asub_q0 < beta); + is_less_than = is_less_than_beta & is_less_than; + } + + if (!__msa_test_bz_v(is_less_than)) { + q2_org = LOAD_UB(data + (2 * img_width)); + p3_org = LOAD_UB(data - (img_width << 2)); + p2_org = LOAD_UB(data - (3 * img_width)); + + p1_org_r = (v8i16) __msa_ilvr_b(zero, (v16i8) p1_org); + p0_org_r = (v8i16) __msa_ilvr_b(zero, (v16i8) p0_org); + q0_org_r = (v8i16) __msa_ilvr_b(zero, (v16i8) q0_org); + + p1_org_l = (v8i16) __msa_ilvl_b(zero, (v16i8) p1_org); + p0_org_l = (v8i16) __msa_ilvl_b(zero, (v16i8) p0_org); + q0_org_l = (v8i16) __msa_ilvl_b(zero, (v16i8) q0_org); + + tmp_flag = alpha >> 2; + tmp_flag = tmp_flag + 2; + tmp_flag = (p0_asub_q0 < tmp_flag); + + p2_asub_p0 = __msa_asub_u_b(p2_org, p0_org); + is_less_than_beta = (p2_asub_p0 < beta); + is_less_than_beta = is_less_than_beta & tmp_flag; + + negate_is_less_than_beta = __msa_xori_b(is_less_than_beta, 0xff); + is_less_than_beta = is_less_than_beta & is_less_than; + negate_is_less_than_beta = negate_is_less_than_beta & is_less_than; + + { + v16i8 is_less_than_beta_l, is_less_than_beta_r; + + q1_org_r = (v8i16) __msa_ilvr_b(zero, (v16i8) q1_org); + + is_less_than_beta_r = __msa_sldi_b((v16i8) is_less_than_beta, zero, + 8); + if (!__msa_test_bz_v((v16u8) is_less_than_beta_r)) { + v8i16 p3_org_r; + + p3_org_r = (v8i16) __msa_ilvr_b(zero, (v16i8) p3_org); + p2_r = (v8i16) __msa_ilvr_b(zero, (v16i8) p2_org); + + AVC_LOOP_FILTER_P0P1P2_OR_Q0Q1Q2(p3_org_r, p0_org_r, + q0_org_r, p1_org_r, + p2_r, q1_org_r, + p0_r, p1_r, p2_r); + } + + q1_org_l = (v8i16) __msa_ilvl_b(zero, (v16i8) q1_org); + + is_less_than_beta_l = __msa_sldi_b(zero, (v16i8) is_less_than_beta, + 8); + + if (!__msa_test_bz_v((v16u8) is_less_than_beta_l)) { + v8i16 p3_org_l; + + p3_org_l = (v8i16) __msa_ilvl_b(zero, (v16i8) p3_org); + p2_l = (v8i16) __msa_ilvl_b(zero, (v16i8) p2_org); + + AVC_LOOP_FILTER_P0P1P2_OR_Q0Q1Q2(p3_org_l, p0_org_l, + q0_org_l, p1_org_l, + p2_l, q1_org_l, + p0_l, p1_l, p2_l); + } + } + + /* combine and store */ + if (!__msa_test_bz_v(is_less_than_beta)) { + p0 = (v16u8) __msa_pckev_b((v16i8) p0_l, (v16i8) p0_r); + p1 = (v16u8) __msa_pckev_b((v16i8) p1_l, (v16i8) p1_r); + p2 = (v16u8) __msa_pckev_b((v16i8) p2_l, (v16i8) p2_r); + + p0_org = __msa_bmnz_v(p0_org, p0, is_less_than_beta); + p1_org = __msa_bmnz_v(p1_org, p1, is_less_than_beta); + p2_org = __msa_bmnz_v(p2_org, p2, is_less_than_beta); + + STORE_UB(p1_org, data - (2 * img_width)); + STORE_UB(p2_org, data - (3 * img_width)); + } + + { + v16i8 negate_is_less_than_beta_r, negate_is_less_than_beta_l; + + negate_is_less_than_beta_r = + __msa_sldi_b((v16i8) negate_is_less_than_beta, zero, 8); + if (!__msa_test_bz_v((v16u8) negate_is_less_than_beta_r)) { + AVC_LOOP_FILTER_P0_OR_Q0(p0_org_r, q1_org_r, p1_org_r, p0_r); + } + + negate_is_less_than_beta_l = + __msa_sldi_b(zero, (v16i8) negate_is_less_than_beta, 8); + if (!__msa_test_bz_v((v16u8) negate_is_less_than_beta_l)) { + AVC_LOOP_FILTER_P0_OR_Q0(p0_org_l, q1_org_l, p1_org_l, p0_l); + } + } + + /* combine */ + if (!__msa_test_bz_v(negate_is_less_than_beta)) { + p0 = (v16u8) __msa_pckev_b((v16i8) p0_l, (v16i8) p0_r); + p0_org = __msa_bmnz_v(p0_org, p0, negate_is_less_than_beta); + } + + STORE_UB(p0_org, data - img_width); + + /* if (tmpFlag && (unsigned)ABS(q2-q0) < thresholds->beta_in) */ + + q3_org = LOAD_UB(data + (3 * img_width)); + + q2_asub_q0 = __msa_asub_u_b(q2_org, q0_org); + is_less_than_beta = (q2_asub_q0 < beta); + is_less_than_beta = is_less_than_beta & tmp_flag; + negate_is_less_than_beta = __msa_xori_b(is_less_than_beta, 0xff); + is_less_than_beta = is_less_than_beta & is_less_than; + negate_is_less_than_beta = negate_is_less_than_beta & is_less_than; + + { + v16i8 is_less_than_beta_l, is_less_than_beta_r; + + is_less_than_beta_r = __msa_sldi_b((v16i8) is_less_than_beta, zero, + 8); + if (!__msa_test_bz_v((v16u8) is_less_than_beta_r)) { + v8i16 q3_org_r; + + q3_org_r = (v8i16) __msa_ilvr_b(zero, (v16i8) q3_org); + q2_r = (v8i16) __msa_ilvr_b(zero, (v16i8) q2_org); + + AVC_LOOP_FILTER_P0P1P2_OR_Q0Q1Q2(q3_org_r, q0_org_r, + p0_org_r, q1_org_r, + q2_r, p1_org_r, + q0_r, q1_r, q2_r); + } + + is_less_than_beta_l = __msa_sldi_b(zero, (v16i8) is_less_than_beta, + 8); + if (!__msa_test_bz_v((v16u8) is_less_than_beta_l)) { + v8i16 q3_org_l; + + q3_org_l = (v8i16) __msa_ilvl_b(zero, (v16i8) q3_org); + q2_l = (v8i16) __msa_ilvl_b(zero, (v16i8) q2_org); + + AVC_LOOP_FILTER_P0P1P2_OR_Q0Q1Q2(q3_org_l, q0_org_l, + p0_org_l, q1_org_l, + q2_l, p1_org_l, + q0_l, q1_l, q2_l); + } + } + + /* combine and store */ + if (!__msa_test_bz_v(is_less_than_beta)) { + q0 = (v16u8) __msa_pckev_b((v16i8) q0_l, (v16i8) q0_r); + q1 = (v16u8) __msa_pckev_b((v16i8) q1_l, (v16i8) q1_r); + q2 = (v16u8) __msa_pckev_b((v16i8) q2_l, (v16i8) q2_r); + + q0_org = __msa_bmnz_v(q0_org, q0, is_less_than_beta); + q1_org = __msa_bmnz_v(q1_org, q1, is_less_than_beta); + q2_org = __msa_bmnz_v(q2_org, q2, is_less_than_beta); + + STORE_UB(q1_org, data + img_width); + STORE_UB(q2_org, data + 2 * img_width); + } + + { + v16i8 negate_is_less_than_beta_r, negate_is_less_than_beta_l; + + negate_is_less_than_beta_r = + __msa_sldi_b((v16i8) negate_is_less_than_beta, zero, 8); + if (!__msa_test_bz_v((v16u8) negate_is_less_than_beta_r)) { + AVC_LOOP_FILTER_P0_OR_Q0(q0_org_r, p1_org_r, q1_org_r, q0_r); + } + + negate_is_less_than_beta_l = + __msa_sldi_b(zero, (v16i8) negate_is_less_than_beta, 8); + if (!__msa_test_bz_v((v16u8) negate_is_less_than_beta_l)) { + AVC_LOOP_FILTER_P0_OR_Q0(q0_org_l, p1_org_l, q1_org_l, q0_l); + } + } + + /* combine */ + if (!__msa_test_bz_v(negate_is_less_than_beta)) { + q0 = (v16u8) __msa_pckev_b((v16i8) q0_l, (v16i8) q0_r); + q0_org = __msa_bmnz_v(q0_org, q0, negate_is_less_than_beta); + } + + STORE_UB(q0_org, data); + } +} + +static void avc_loopfilter_luma_intra_edge_ver_msa(uint8_t *data, + uint8_t alpha_in, + uint8_t beta_in, + uint32_t img_width) +{ + uint8_t *src; + v16u8 alpha, beta, p0_asub_q0; + v16u8 is_less_than_alpha, is_less_than; + v16u8 is_less_than_beta, negate_is_less_than_beta; + v8i16 p2_r = { 0 }; + v8i16 p1_r = { 0 }; + v8i16 p0_r = { 0 }; + v8i16 q0_r = { 0 }; + v8i16 q1_r = { 0 }; + v8i16 q2_r = { 0 }; + v8i16 p2_l = { 0 }; + v8i16 p1_l = { 0 }; + v8i16 p0_l = { 0 }; + v8i16 q0_l = { 0 }; + v8i16 q1_l = { 0 }; + v8i16 q2_l = { 0 }; + v16u8 p3_org, p2_org, p1_org, p0_org, q0_org, q1_org, q2_org, q3_org; + v8i16 p1_org_r, p0_org_r, q0_org_r, q1_org_r; + v8i16 p1_org_l, p0_org_l, q0_org_l, q1_org_l; + v16i8 zero = { 0 }; + v16u8 tmp_flag; + + src = data - 4; + + { + v16u8 row0, row1, row2, row3, row4, row5, row6, row7; + v16u8 row8, row9, row10, row11, row12, row13, row14, row15; + + LOAD_8VECS_UB(src, img_width, + row0, row1, row2, row3, row4, row5, row6, row7); + LOAD_8VECS_UB(src + (8 * img_width), img_width, + row8, row9, row10, row11, row12, row13, row14, row15); + + TRANSPOSE16x8_B_UB(row0, row1, row2, row3, row4, row5, row6, row7, + row8, row9, row10, row11, row12, row13, row14, row15, + p3_org, p2_org, p1_org, p0_org, + q0_org, q1_org, q2_org, q3_org); + } + + p1_org_r = (v8i16) __msa_ilvr_b(zero, (v16i8) p1_org); + p0_org_r = (v8i16) __msa_ilvr_b(zero, (v16i8) p0_org); + q0_org_r = (v8i16) __msa_ilvr_b(zero, (v16i8) q0_org); + q1_org_r = (v8i16) __msa_ilvr_b(zero, (v16i8) q1_org); + + p1_org_l = (v8i16) __msa_ilvl_b(zero, (v16i8) p1_org); + p0_org_l = (v8i16) __msa_ilvl_b(zero, (v16i8) p0_org); + q0_org_l = (v8i16) __msa_ilvl_b(zero, (v16i8) q0_org); + q1_org_l = (v8i16) __msa_ilvl_b(zero, (v16i8) q1_org); + + /* if ( ((unsigned)ABS(p0-q0) < thresholds->alpha_in) && + ((unsigned)ABS(p1-p0) < thresholds->beta_in) && + ((unsigned)ABS(q1-q0) < thresholds->beta_in) ) */ + { + v16u8 p1_asub_p0, q1_asub_q0; + + p0_asub_q0 = __msa_asub_u_b(p0_org, q0_org); + p1_asub_p0 = __msa_asub_u_b(p1_org, p0_org); + q1_asub_q0 = __msa_asub_u_b(q1_org, q0_org); + + alpha = (v16u8) __msa_fill_b(alpha_in); + beta = (v16u8) __msa_fill_b(beta_in); + + is_less_than_alpha = (p0_asub_q0 < alpha); + is_less_than_beta = (p1_asub_p0 < beta); + is_less_than = is_less_than_beta & is_less_than_alpha; + is_less_than_beta = (q1_asub_q0 < beta); + is_less_than = is_less_than_beta & is_less_than; + } + + if (!__msa_test_bz_v(is_less_than)) { + tmp_flag = alpha >> 2; + tmp_flag = tmp_flag + 2; + tmp_flag = (p0_asub_q0 < tmp_flag); + + { + v16u8 p2_asub_p0; + + p2_asub_p0 = __msa_asub_u_b(p2_org, p0_org); + is_less_than_beta = (p2_asub_p0 < beta); + } + + is_less_than_beta = tmp_flag & is_less_than_beta; + + negate_is_less_than_beta = __msa_xori_b(is_less_than_beta, 0xff); + is_less_than_beta = is_less_than_beta & is_less_than; + negate_is_less_than_beta = negate_is_less_than_beta & is_less_than; + + /* right */ + { + v16i8 is_less_than_beta_r; + + is_less_than_beta_r = __msa_sldi_b((v16i8) is_less_than_beta, zero, + 8); + if (!__msa_test_bz_v((v16u8) is_less_than_beta_r)) { + v8i16 p3_org_r; + + p3_org_r = (v8i16) __msa_ilvr_b(zero, (v16i8) p3_org); + p2_r = (v8i16) __msa_ilvr_b(zero, (v16i8) p2_org); + + AVC_LOOP_FILTER_P0P1P2_OR_Q0Q1Q2(p3_org_r, p0_org_r, + q0_org_r, p1_org_r, + p2_r, q1_org_r, + p0_r, p1_r, p2_r); + } + } + /* left */ + { + v16i8 is_less_than_beta_l; + + is_less_than_beta_l = __msa_sldi_b(zero, (v16i8) is_less_than_beta, + 8); + if (!__msa_test_bz_v((v16u8) is_less_than_beta_l)) { + v8i16 p3_org_l; + + p3_org_l = (v8i16) __msa_ilvl_b(zero, (v16i8) p3_org); + p2_l = (v8i16) __msa_ilvl_b(zero, (v16i8) p2_org); + + AVC_LOOP_FILTER_P0P1P2_OR_Q0Q1Q2(p3_org_l, p0_org_l, + q0_org_l, p1_org_l, + p2_l, q1_org_l, + p0_l, p1_l, p2_l); + } + } + + /* combine and store */ + if (!__msa_test_bz_v(is_less_than_beta)) { + v16u8 p0, p2, p1; + + p0 = (v16u8) __msa_pckev_b((v16i8) p0_l, (v16i8) p0_r); + p1 = (v16u8) __msa_pckev_b((v16i8) p1_l, (v16i8) p1_r); + p2 = (v16u8) __msa_pckev_b((v16i8) p2_l, (v16i8) p2_r); + + p0_org = __msa_bmnz_v(p0_org, p0, is_less_than_beta); + p1_org = __msa_bmnz_v(p1_org, p1, is_less_than_beta); + p2_org = __msa_bmnz_v(p2_org, p2, is_less_than_beta); + } + + /* right */ + { + v16i8 negate_is_less_than_beta_r; + + negate_is_less_than_beta_r = + __msa_sldi_b((v16i8) negate_is_less_than_beta, zero, 8); + + if (!__msa_test_bz_v((v16u8) negate_is_less_than_beta_r)) { + AVC_LOOP_FILTER_P0_OR_Q0(p0_org_r, q1_org_r, p1_org_r, p0_r); + } + } + + /* left */ + { + v16i8 negate_is_less_than_beta_l; + + negate_is_less_than_beta_l = + __msa_sldi_b(zero, (v16i8) negate_is_less_than_beta, 8); + if (!__msa_test_bz_v((v16u8) negate_is_less_than_beta_l)) { + AVC_LOOP_FILTER_P0_OR_Q0(p0_org_l, q1_org_l, p1_org_l, p0_l); + } + } + + if (!__msa_test_bz_v(negate_is_less_than_beta)) { + v16u8 p0; + + p0 = (v16u8) __msa_pckev_b((v16i8) p0_l, (v16i8) p0_r); + p0_org = __msa_bmnz_v(p0_org, p0, negate_is_less_than_beta); + } + + { + v16u8 q2_asub_q0; + + q2_asub_q0 = __msa_asub_u_b(q2_org, q0_org); + is_less_than_beta = (q2_asub_q0 < beta); + } + + is_less_than_beta = is_less_than_beta & tmp_flag; + negate_is_less_than_beta = __msa_xori_b(is_less_than_beta, 0xff); + + is_less_than_beta = is_less_than_beta & is_less_than; + negate_is_less_than_beta = negate_is_less_than_beta & is_less_than; + + /* right */ + { + v16i8 is_less_than_beta_r; + + is_less_than_beta_r = + __msa_sldi_b((v16i8) is_less_than_beta, zero, 8); + if (!__msa_test_bz_v((v16u8) is_less_than_beta_r)) { + v8i16 q3_org_r; + + q3_org_r = (v8i16) __msa_ilvr_b(zero, (v16i8) q3_org); + q2_r = (v8i16) __msa_ilvr_b(zero, (v16i8) q2_org); + + AVC_LOOP_FILTER_P0P1P2_OR_Q0Q1Q2(q3_org_r, q0_org_r, + p0_org_r, q1_org_r, + q2_r, p1_org_r, + q0_r, q1_r, q2_r); + } + } + + /* left */ + { + v16i8 is_less_than_beta_l; + + is_less_than_beta_l = + __msa_sldi_b(zero, (v16i8) is_less_than_beta, 8); + if (!__msa_test_bz_v((v16u8) is_less_than_beta_l)) { + v8i16 q3_org_l; + + q3_org_l = (v8i16) __msa_ilvl_b(zero, (v16i8) q3_org); + q2_l = (v8i16) __msa_ilvl_b(zero, (v16i8) q2_org); + + AVC_LOOP_FILTER_P0P1P2_OR_Q0Q1Q2(q3_org_l, q0_org_l, + p0_org_l, q1_org_l, + q2_l, p1_org_l, + q0_l, q1_l, q2_l); + } + } + + /* combine and store */ + if (!__msa_test_bz_v(is_less_than_beta)) { + v16u8 q0, q1, q2; + + q0 = (v16u8) __msa_pckev_b((v16i8) q0_l, (v16i8) q0_r); + q1 = (v16u8) __msa_pckev_b((v16i8) q1_l, (v16i8) q1_r); + q2 = (v16u8) __msa_pckev_b((v16i8) q2_l, (v16i8) q2_r); + + q0_org = __msa_bmnz_v(q0_org, q0, is_less_than_beta); + q1_org = __msa_bmnz_v(q1_org, q1, is_less_than_beta); + q2_org = __msa_bmnz_v(q2_org, q2, is_less_than_beta); + } + + /* right */ + { + v16i8 negate_is_less_than_beta_r; + + negate_is_less_than_beta_r = + __msa_sldi_b((v16i8) negate_is_less_than_beta, zero, 8); + if (!__msa_test_bz_v((v16u8) negate_is_less_than_beta_r)) { + AVC_LOOP_FILTER_P0_OR_Q0(q0_org_r, p1_org_r, q1_org_r, q0_r); + } + } + /* left */ + { + v16i8 negate_is_less_than_beta_l; + + negate_is_less_than_beta_l = + __msa_sldi_b(zero, (v16i8) negate_is_less_than_beta, 8); + if (!__msa_test_bz_v((v16u8) negate_is_less_than_beta_l)) { + AVC_LOOP_FILTER_P0_OR_Q0(q0_org_l, p1_org_l, q1_org_l, q0_l); + } + } + + if (!__msa_test_bz_v(negate_is_less_than_beta)) { + v16u8 q0; + + q0 = (v16u8) __msa_pckev_b((v16i8) q0_l, (v16i8) q0_r); + q0_org = __msa_bmnz_v(q0_org, q0, negate_is_less_than_beta); + } + } + + { + v16u8 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7; + uint32_t out0, out2; + uint16_t out1, out3; + + tmp0 = (v16u8) __msa_ilvr_b((v16i8) p1_org, (v16i8) p2_org); + tmp1 = (v16u8) __msa_ilvr_b((v16i8) q0_org, (v16i8) p0_org); + tmp2 = (v16u8) __msa_ilvr_b((v16i8) q2_org, (v16i8) q1_org); + tmp3 = (v16u8) __msa_ilvr_h((v8i16) tmp1, (v8i16) tmp0); + tmp4 = (v16u8) __msa_ilvl_h((v8i16) tmp1, (v8i16) tmp0); + + tmp0 = (v16u8) __msa_ilvl_b((v16i8) p1_org, (v16i8) p2_org); + tmp1 = (v16u8) __msa_ilvl_b((v16i8) q0_org, (v16i8) p0_org); + tmp5 = (v16u8) __msa_ilvl_b((v16i8) q2_org, (v16i8) q1_org); + tmp6 = (v16u8) __msa_ilvr_h((v8i16) tmp1, (v8i16) tmp0); + tmp7 = (v16u8) __msa_ilvl_h((v8i16) tmp1, (v8i16) tmp0); + + src = data - 3; + + out0 = __msa_copy_u_w((v4i32) tmp3, 0); + out1 = __msa_copy_u_h((v8i16) tmp2, 0); + out2 = __msa_copy_u_w((v4i32) tmp3, 1); + out3 = __msa_copy_u_h((v8i16) tmp2, 1); + + STORE_WORD(src, out0); + STORE_HWORD((src + 4), out1); + src += img_width; + STORE_WORD(src, out2); + STORE_HWORD((src + 4), out3); + + out0 = __msa_copy_u_w((v4i32) tmp3, 2); + out1 = __msa_copy_u_h((v8i16) tmp2, 2); + out2 = __msa_copy_u_w((v4i32) tmp3, 3); + out3 = __msa_copy_u_h((v8i16) tmp2, 3); + + src += img_width; + STORE_WORD(src, out0); + STORE_HWORD((src + 4), out1); + src += img_width; + STORE_WORD(src, out2); + STORE_HWORD((src + 4), out3); + + out0 = __msa_copy_u_w((v4i32) tmp4, 0); + out1 = __msa_copy_u_h((v8i16) tmp2, 4); + out2 = __msa_copy_u_w((v4i32) tmp4, 1); + out3 = __msa_copy_u_h((v8i16) tmp2, 5); + + src += img_width; + STORE_WORD(src, out0); + STORE_HWORD((src + 4), out1); + src += img_width; + STORE_WORD(src, out2); + STORE_HWORD((src + 4), out3); + + out0 = __msa_copy_u_w((v4i32) tmp4, 2); + out1 = __msa_copy_u_h((v8i16) tmp2, 6); + out2 = __msa_copy_u_w((v4i32) tmp4, 3); + out3 = __msa_copy_u_h((v8i16) tmp2, 7); + + src += img_width; + STORE_WORD(src, out0); + STORE_HWORD((src + 4), out1); + src += img_width; + STORE_WORD(src, out2); + STORE_HWORD((src + 4), out3); + + out0 = __msa_copy_u_w((v4i32) tmp6, 0); + out1 = __msa_copy_u_h((v8i16) tmp5, 0); + out2 = __msa_copy_u_w((v4i32) tmp6, 1); + out3 = __msa_copy_u_h((v8i16) tmp5, 1); + + src += img_width; + STORE_WORD(src, out0); + STORE_HWORD((src + 4), out1); + src += img_width; + STORE_WORD(src, out2); + STORE_HWORD((src + 4), out3); + + out0 = __msa_copy_u_w((v4i32) tmp6, 2); + out1 = __msa_copy_u_h((v8i16) tmp5, 2); + out2 = __msa_copy_u_w((v4i32) tmp6, 3); + out3 = __msa_copy_u_h((v8i16) tmp5, 3); + + src += img_width; + STORE_WORD(src, out0); + STORE_HWORD((src + 4), out1); + src += img_width; + STORE_WORD(src, out2); + STORE_HWORD((src + 4), out3); + + out0 = __msa_copy_u_w((v4i32) tmp7, 0); + out1 = __msa_copy_u_h((v8i16) tmp5, 4); + out2 = __msa_copy_u_w((v4i32) tmp7, 1); + out3 = __msa_copy_u_h((v8i16) tmp5, 5); + + src += img_width; + STORE_WORD(src, out0); + STORE_HWORD((src + 4), out1); + src += img_width; + STORE_WORD(src, out2); + STORE_HWORD((src + 4), out3); + + out0 = __msa_copy_u_w((v4i32) tmp7, 2); + out1 = __msa_copy_u_h((v8i16) tmp5, 6); + out2 = __msa_copy_u_w((v4i32) tmp7, 3); + out3 = __msa_copy_u_h((v8i16) tmp5, 7); + + src += img_width; + STORE_WORD(src, out0); + STORE_HWORD((src + 4), out1); + src += img_width; + STORE_WORD(src, out2); + STORE_HWORD((src + 4), out3); + } +} + +static void avc_h_loop_filter_luma_mbaff_intra_msa(uint8_t *src, + int32_t stride, + int32_t alpha_in, + int32_t beta_in) +{ + uint64_t load0, load1; + uint32_t out0, out2; + uint16_t out1, out3; + v8u16 src0_r, src1_r, src2_r, src3_r, src4_r, src5_r, src6_r, src7_r; + v8u16 dst0_r, dst1_r, dst4_r, dst5_r; + v8u16 dst2_x_r, dst2_y_r, dst3_x_r, dst3_y_r; + v16u8 dst0, dst1, dst4, dst5, dst2_x, dst2_y, dst3_x, dst3_y; + v16u8 tmp0, tmp1, tmp2, tmp3; + v16u8 alpha, beta; + v16u8 p0_asub_q0, p1_asub_p0, q1_asub_q0, p2_asub_p0, q2_asub_q0; + v16u8 is_less_than, is_less_than_alpha, is_less_than_beta; + v16u8 is_less_than_beta1, is_less_than_beta2; + v16u8 src0 = { 0 }; + v16u8 src1 = { 0 }; + v16u8 src2 = { 0 }; + v16u8 src3 = { 0 }; + v16u8 src4 = { 0 }; + v16u8 src5 = { 0 }; + v16u8 src6 = { 0 }; + v16u8 src7 = { 0 }; + v16i8 zeros = { 0 }; + + load0 = LOAD_DWORD(src - 4); + load1 = LOAD_DWORD(src + stride - 4); + src0 = (v16u8) __msa_insert_d((v2i64) src0, 0, load0); + src1 = (v16u8) __msa_insert_d((v2i64) src1, 0, load1); + + load0 = LOAD_DWORD(src + (2 * stride) - 4); + load1 = LOAD_DWORD(src + (3 * stride) - 4); + src2 = (v16u8) __msa_insert_d((v2i64) src2, 0, load0); + src3 = (v16u8) __msa_insert_d((v2i64) src3, 0, load1); + + load0 = LOAD_DWORD(src + (4 * stride) - 4); + load1 = LOAD_DWORD(src + (5 * stride) - 4); + src4 = (v16u8) __msa_insert_d((v2i64) src4, 0, load0); + src5 = (v16u8) __msa_insert_d((v2i64) src5, 0, load1); + + load0 = LOAD_DWORD(src + (6 * stride) - 4); + load1 = LOAD_DWORD(src + (7 * stride) - 4); + src6 = (v16u8) __msa_insert_d((v2i64) src6, 0, load0); + src7 = (v16u8) __msa_insert_d((v2i64) src7, 0, load1); + + src0 = (v16u8) __msa_ilvr_b((v16i8) src1, (v16i8) src0); + src1 = (v16u8) __msa_ilvr_b((v16i8) src3, (v16i8) src2); + src2 = (v16u8) __msa_ilvr_b((v16i8) src5, (v16i8) src4); + src3 = (v16u8) __msa_ilvr_b((v16i8) src7, (v16i8) src6); + tmp0 = (v16u8) __msa_ilvr_h((v8i16) src1, (v8i16) src0); + tmp1 = (v16u8) __msa_ilvl_h((v8i16) src1, (v8i16) src0); + tmp2 = (v16u8) __msa_ilvr_h((v8i16) src3, (v8i16) src2); + tmp3 = (v16u8) __msa_ilvl_h((v8i16) src3, (v8i16) src2); + src6 = (v16u8) __msa_ilvr_w((v4i32) tmp2, (v4i32) tmp0); + src0 = (v16u8) __msa_sldi_b(zeros, (v16i8) src6, 8); + src1 = (v16u8) __msa_ilvl_w((v4i32) tmp2, (v4i32) tmp0); + src2 = (v16u8) __msa_sldi_b(zeros, (v16i8) src1, 8); + src3 = (v16u8) __msa_ilvr_w((v4i32) tmp3, (v4i32) tmp1); + src4 = (v16u8) __msa_sldi_b(zeros, (v16i8) src3, 8); + src5 = (v16u8) __msa_ilvl_w((v4i32) tmp3, (v4i32) tmp1); + src7 = (v16u8) __msa_sldi_b(zeros, (v16i8) src5, 8); + + p0_asub_q0 = __msa_asub_u_b(src2, src3); + p1_asub_p0 = __msa_asub_u_b(src1, src2); + q1_asub_q0 = __msa_asub_u_b(src4, src3); + + alpha = (v16u8) __msa_fill_b(alpha_in); + beta = (v16u8) __msa_fill_b(beta_in); + + is_less_than_alpha = (p0_asub_q0 < alpha); + is_less_than_beta = (p1_asub_p0 < beta); + is_less_than = is_less_than_alpha & is_less_than_beta; + is_less_than_beta = (q1_asub_q0 < beta); + is_less_than = is_less_than & is_less_than_beta; + + alpha >>= 2; + alpha += 2; + + is_less_than_alpha = (p0_asub_q0 < alpha); + + p2_asub_p0 = __msa_asub_u_b(src0, src2); + is_less_than_beta1 = (p2_asub_p0 < beta); + q2_asub_q0 = __msa_asub_u_b(src5, src3); + is_less_than_beta2 = (q2_asub_q0 < beta); + + src0_r = (v8u16) __msa_ilvr_b(zeros, (v16i8) src0); + src1_r = (v8u16) __msa_ilvr_b(zeros, (v16i8) src1); + src2_r = (v8u16) __msa_ilvr_b(zeros, (v16i8) src2); + src3_r = (v8u16) __msa_ilvr_b(zeros, (v16i8) src3); + src4_r = (v8u16) __msa_ilvr_b(zeros, (v16i8) src4); + src5_r = (v8u16) __msa_ilvr_b(zeros, (v16i8) src5); + src6_r = (v8u16) __msa_ilvr_b(zeros, (v16i8) src6); + src7_r = (v8u16) __msa_ilvr_b(zeros, (v16i8) src7); + + dst2_x_r = src1_r + src2_r + src3_r; + dst2_x_r = src0_r + (2 * (dst2_x_r)) + src4_r; + dst2_x_r = (v8u16) __msa_srari_h((v8i16) dst2_x_r, 3); + + dst1_r = src0_r + src1_r + src2_r + src3_r; + dst1_r = (v8u16) __msa_srari_h((v8i16) dst1_r, 2); + + dst0_r = (2 * src6_r) + (3 * src0_r); + dst0_r += src1_r + src2_r + src3_r; + dst0_r = (v8u16) __msa_srari_h((v8i16) dst0_r, 3); + + dst2_y_r = (2 * src1_r) + src2_r + src4_r; + dst2_y_r = (v8u16) __msa_srari_h((v8i16) dst2_y_r, 2); + + dst2_x = (v16u8) __msa_pckev_b((v16i8) dst2_x_r, (v16i8) dst2_x_r); + dst2_y = (v16u8) __msa_pckev_b((v16i8) dst2_y_r, (v16i8) dst2_y_r); + dst2_x = __msa_bmnz_v(dst2_y, dst2_x, is_less_than_beta1); + + dst3_x_r = src2_r + src3_r + src4_r; + dst3_x_r = src1_r + (2 * dst3_x_r) + src5_r; + dst3_x_r = (v8u16) __msa_srari_h((v8i16) dst3_x_r, 3); + + dst4_r = src2_r + src3_r + src4_r + src5_r; + dst4_r = (v8u16) __msa_srari_h((v8i16) dst4_r, 2); + + dst5_r = (2 * src7_r) + (3 * src5_r); + dst5_r += src4_r + src3_r + src2_r; + dst5_r = (v8u16) __msa_srari_h((v8i16) dst5_r, 3); + + dst3_y_r = (2 * src4_r) + src3_r + src1_r; + dst3_y_r = (v8u16) __msa_srari_h((v8i16) dst3_y_r, 2); + dst3_x = (v16u8) __msa_pckev_b((v16i8) dst3_x_r, (v16i8) dst3_x_r); + dst3_y = (v16u8) __msa_pckev_b((v16i8) dst3_y_r, (v16i8) dst3_y_r); + dst3_x = __msa_bmnz_v(dst3_y, dst3_x, is_less_than_beta2); + + dst2_y_r = (2 * src1_r) + src2_r + src4_r; + dst2_y_r = (v8u16) __msa_srari_h((v8i16) dst2_y_r, 2); + + dst3_y_r = (2 * src4_r) + src3_r + src1_r; + dst3_y_r = (v8u16) __msa_srari_h((v8i16) dst3_y_r, 2); + + dst2_y = (v16u8) __msa_pckev_b((v16i8) dst2_y_r, (v16i8) dst2_y_r); + dst3_y = (v16u8) __msa_pckev_b((v16i8) dst3_y_r, (v16i8) dst3_y_r); + dst2_x = __msa_bmnz_v(dst2_y, dst2_x, is_less_than_alpha); + dst3_x = __msa_bmnz_v(dst3_y, dst3_x, is_less_than_alpha); + dst2_x = __msa_bmnz_v(src2, dst2_x, is_less_than); + dst3_x = __msa_bmnz_v(src3, dst3_x, is_less_than); + + is_less_than = is_less_than_alpha & is_less_than; + dst1 = (v16u8) __msa_pckev_b((v16i8) dst1_r, (v16i8) dst1_r); + is_less_than_beta1 = is_less_than_beta1 & is_less_than; + dst1 = __msa_bmnz_v(src1, dst1, is_less_than_beta1); + + dst0 = (v16u8) __msa_pckev_b((v16i8) dst0_r, (v16i8) dst0_r); + dst0 = __msa_bmnz_v(src0, dst0, is_less_than_beta1); + + dst4 = (v16u8) __msa_pckev_b((v16i8) dst4_r, (v16i8) dst4_r); + is_less_than_beta2 = is_less_than_beta2 & is_less_than; + dst4 = __msa_bmnz_v(src4, dst4, is_less_than_beta2); + + dst5 = (v16u8) __msa_pckev_b((v16i8) dst5_r, (v16i8) dst5_r); + dst5 = __msa_bmnz_v(src5, dst5, is_less_than_beta2); + + dst0 = (v16u8) __msa_ilvr_b((v16i8) dst1, (v16i8) dst0); + dst1 = (v16u8) __msa_ilvr_b((v16i8) dst3_x, (v16i8) dst2_x); + dst2_x = (v16u8) __msa_ilvr_b((v16i8) dst5, (v16i8) dst4); + tmp0 = (v16u8) __msa_ilvr_h((v8i16) dst1, (v8i16) dst0); + tmp1 = (v16u8) __msa_ilvl_h((v8i16) dst1, (v8i16) dst0); + tmp2 = (v16u8) __msa_ilvr_h((v8i16) zeros, (v8i16) dst2_x); + tmp3 = (v16u8) __msa_ilvl_h((v8i16) zeros, (v8i16) dst2_x); + dst0 = (v16u8) __msa_ilvr_w((v4i32) tmp2, (v4i32) tmp0); + dst1 = (v16u8) __msa_sldi_b(zeros, (v16i8) dst0, 8); + dst2_x = (v16u8) __msa_ilvl_w((v4i32) tmp2, (v4i32) tmp0); + dst3_x = (v16u8) __msa_sldi_b(zeros, (v16i8) dst2_x, 8); + dst4 = (v16u8) __msa_ilvr_w((v4i32) tmp3, (v4i32) tmp1); + dst5 = (v16u8) __msa_sldi_b(zeros, (v16i8) dst4, 8); + dst2_y = (v16u8) __msa_ilvl_w((v4i32) tmp3, (v4i32) tmp1); + dst3_y = (v16u8) __msa_sldi_b(zeros, (v16i8) dst2_y, 8); + + out0 = __msa_copy_u_w((v4i32) dst0, 0); + out1 = __msa_copy_u_h((v8i16) dst0, 2); + out2 = __msa_copy_u_w((v4i32) dst1, 0); + out3 = __msa_copy_u_h((v8i16) dst1, 2); + + STORE_WORD((src - 3), out0); + STORE_HWORD((src + 1), out1); + src += stride; + STORE_WORD((src - 3), out2); + STORE_HWORD((src + 1), out3); + src += stride; + + out0 = __msa_copy_u_w((v4i32) dst2_x, 0); + out1 = __msa_copy_u_h((v8i16) dst2_x, 2); + out2 = __msa_copy_u_w((v4i32) dst3_x, 0); + out3 = __msa_copy_u_h((v8i16) dst3_x, 2); + + STORE_WORD((src - 3), out0); + STORE_HWORD((src + 1), out1); + src += stride; + STORE_WORD((src - 3), out2); + STORE_HWORD((src + 1), out3); + src += stride; + + out0 = __msa_copy_u_w((v4i32) dst4, 0); + out1 = __msa_copy_u_h((v8i16) dst4, 2); + out2 = __msa_copy_u_w((v4i32) dst5, 0); + out3 = __msa_copy_u_h((v8i16) dst5, 2); + + STORE_WORD((src - 3), out0); + STORE_HWORD((src + 1), out1); + src += stride; + STORE_WORD((src - 3), out2); + STORE_HWORD((src + 1), out3); + src += stride; + + out0 = __msa_copy_u_w((v4i32) dst2_y, 0); + out1 = __msa_copy_u_h((v8i16) dst2_y, 2); + out2 = __msa_copy_u_w((v4i32) dst3_y, 0); + out3 = __msa_copy_u_h((v8i16) dst3_y, 2); + + STORE_WORD((src - 3), out0); + STORE_HWORD((src + 1), out1); + src += stride; + STORE_WORD((src - 3), out2); + STORE_HWORD((src + 1), out3); +} + +static void avc_loopfilter_cb_or_cr_intra_edge_hor_msa(uint8_t *data_cb_or_cr, + uint8_t alpha_in, + uint8_t beta_in, + uint32_t img_width) +{ + v16u8 alpha, beta; + v16u8 is_less_than; + v8i16 p0_or_q0, q0_or_p0; + v16u8 p1_or_q1_org, p0_or_q0_org, q0_or_p0_org, q1_or_p1_org; + v16i8 zero = { 0 }; + v16u8 p0_asub_q0, p1_asub_p0, q1_asub_q0; + v16u8 is_less_than_alpha, is_less_than_beta; + v8i16 p1_org_r, p0_org_r, q0_org_r, q1_org_r; + + alpha = (v16u8) __msa_fill_b(alpha_in); + beta = (v16u8) __msa_fill_b(beta_in); + + p1_or_q1_org = LOAD_UB(data_cb_or_cr - (img_width << 1)); + p0_or_q0_org = LOAD_UB(data_cb_or_cr - img_width); + q0_or_p0_org = LOAD_UB(data_cb_or_cr); + q1_or_p1_org = LOAD_UB(data_cb_or_cr + img_width); + + p0_asub_q0 = __msa_asub_u_b(p0_or_q0_org, q0_or_p0_org); + p1_asub_p0 = __msa_asub_u_b(p1_or_q1_org, p0_or_q0_org); + q1_asub_q0 = __msa_asub_u_b(q1_or_p1_org, q0_or_p0_org); + + is_less_than_alpha = (p0_asub_q0 < alpha); + is_less_than_beta = (p1_asub_p0 < beta); + is_less_than = is_less_than_beta & is_less_than_alpha; + is_less_than_beta = (q1_asub_q0 < beta); + is_less_than = is_less_than_beta & is_less_than; + + is_less_than = (v16u8) __msa_ilvr_d((v2i64) zero, (v2i64) is_less_than); + + if (!__msa_test_bz_v(is_less_than)) { + p1_org_r = (v8i16) __msa_ilvr_b(zero, (v16i8) p1_or_q1_org); + p0_org_r = (v8i16) __msa_ilvr_b(zero, (v16i8) p0_or_q0_org); + q0_org_r = (v8i16) __msa_ilvr_b(zero, (v16i8) q0_or_p0_org); + q1_org_r = (v8i16) __msa_ilvr_b(zero, (v16i8) q1_or_p1_org); + + AVC_LOOP_FILTER_P0_OR_Q0(p0_org_r, q1_org_r, p1_org_r, p0_or_q0); + AVC_LOOP_FILTER_P0_OR_Q0(q0_org_r, p1_org_r, q1_org_r, q0_or_p0); + + p0_or_q0 = (v8i16) __msa_pckev_b(zero, (v16i8) p0_or_q0); + q0_or_p0 = (v8i16) __msa_pckev_b(zero, (v16i8) q0_or_p0); + + p0_or_q0_org = + __msa_bmnz_v(p0_or_q0_org, (v16u8) p0_or_q0, is_less_than); + q0_or_p0_org = + __msa_bmnz_v(q0_or_p0_org, (v16u8) q0_or_p0, is_less_than); + + STORE_UB(q0_or_p0_org, data_cb_or_cr); + STORE_UB(p0_or_q0_org, data_cb_or_cr - img_width); + } +} + +static void avc_loopfilter_cb_or_cr_intra_edge_ver_msa(uint8_t *data_cb_or_cr, + uint8_t alpha_in, + uint8_t beta_in, + uint32_t img_width) +{ + uint16_t out0, out1, out2, out3; + v16u8 tmp1, alpha, beta, is_less_than; + v8i16 p0_or_q0, q0_or_p0; + v16u8 p1_or_q1_org, p0_or_q0_org, q0_or_p0_org, q1_or_p1_org; + v16i8 zero = { 0 }; + v16u8 p0_asub_q0, p1_asub_p0, q1_asub_q0; + v16u8 is_less_than_alpha, is_less_than_beta; + v8i16 p1_org_r, p0_org_r, q0_org_r, q1_org_r; + + { + v16u8 row0, row1, row2, row3, row4, row5, row6, row7; + + LOAD_8VECS_UB((data_cb_or_cr - 2), img_width, + row0, row1, row2, row3, row4, row5, row6, row7); + + TRANSPOSE8x4_B_UB(row0, row1, row2, row3, row4, row5, row6, row7, + p1_or_q1_org, p0_or_q0_org, + q0_or_p0_org, q1_or_p1_org); + } + + alpha = (v16u8) __msa_fill_b(alpha_in); + beta = (v16u8) __msa_fill_b(beta_in); + + p0_asub_q0 = __msa_asub_u_b(p0_or_q0_org, q0_or_p0_org); + p1_asub_p0 = __msa_asub_u_b(p1_or_q1_org, p0_or_q0_org); + q1_asub_q0 = __msa_asub_u_b(q1_or_p1_org, q0_or_p0_org); + + is_less_than_alpha = (p0_asub_q0 < alpha); + is_less_than_beta = (p1_asub_p0 < beta); + is_less_than = is_less_than_beta & is_less_than_alpha; + is_less_than_beta = (q1_asub_q0 < beta); + is_less_than = is_less_than_beta & is_less_than; + + is_less_than = (v16u8) __msa_ilvr_d((v2i64) zero, (v2i64) is_less_than); + + if (!__msa_test_bz_v(is_less_than)) { + p1_org_r = (v8i16) __msa_ilvr_b(zero, (v16i8) p1_or_q1_org); + p0_org_r = (v8i16) __msa_ilvr_b(zero, (v16i8) p0_or_q0_org); + q0_org_r = (v8i16) __msa_ilvr_b(zero, (v16i8) q0_or_p0_org); + q1_org_r = (v8i16) __msa_ilvr_b(zero, (v16i8) q1_or_p1_org); + + AVC_LOOP_FILTER_P0_OR_Q0(p0_org_r, q1_org_r, p1_org_r, p0_or_q0); + AVC_LOOP_FILTER_P0_OR_Q0(q0_org_r, p1_org_r, q1_org_r, q0_or_p0); + + /* convert 16 bit output into 8 bit output */ + p0_or_q0 = (v8i16) __msa_pckev_b(zero, (v16i8) p0_or_q0); + q0_or_p0 = (v8i16) __msa_pckev_b(zero, (v16i8) q0_or_p0); + + p0_or_q0_org = + __msa_bmnz_v(p0_or_q0_org, (v16u8) p0_or_q0, is_less_than); + q0_or_p0_org = + __msa_bmnz_v(q0_or_p0_org, (v16u8) q0_or_p0, is_less_than); + + tmp1 = (v16u8) __msa_ilvr_b((v16i8) q0_or_p0_org, (v16i8) p0_or_q0_org); + + data_cb_or_cr -= 1; + + out0 = __msa_copy_u_h((v8i16) tmp1, 0); + out1 = __msa_copy_u_h((v8i16) tmp1, 1); + out2 = __msa_copy_u_h((v8i16) tmp1, 2); + out3 = __msa_copy_u_h((v8i16) tmp1, 3); + + STORE_HWORD(data_cb_or_cr, out0); + data_cb_or_cr += img_width; + STORE_HWORD(data_cb_or_cr, out1); + data_cb_or_cr += img_width; + STORE_HWORD(data_cb_or_cr, out2); + data_cb_or_cr += img_width; + STORE_HWORD(data_cb_or_cr, out3); + data_cb_or_cr += img_width; + + out0 = __msa_copy_u_h((v8i16) tmp1, 4); + out1 = __msa_copy_u_h((v8i16) tmp1, 5); + out2 = __msa_copy_u_h((v8i16) tmp1, 6); + out3 = __msa_copy_u_h((v8i16) tmp1, 7); + + STORE_HWORD(data_cb_or_cr, out0); + data_cb_or_cr += img_width; + STORE_HWORD(data_cb_or_cr, out1); + data_cb_or_cr += img_width; + STORE_HWORD(data_cb_or_cr, out2); + data_cb_or_cr += img_width; + STORE_HWORD(data_cb_or_cr, out3); + } +} + +static void avc_loopfilter_luma_inter_edge_ver_msa(uint8_t *data, + uint8_t bs0, uint8_t bs1, + uint8_t bs2, uint8_t bs3, + uint8_t tc0, uint8_t tc1, + uint8_t tc2, uint8_t tc3, + uint8_t alpha_in, + uint8_t beta_in, + uint32_t img_width) +{ + uint8_t *src; + v16u8 beta, bs = { 0 }; + v16i8 tmp_vec; + v16u8 tc = { 0 }; + v16u8 is_less_than, is_less_than_beta; + v16u8 p1, p0, q0, q1; + v8i16 p0_r, q0_r, p1_r = { 0 }; + v8i16 q1_r = { 0 }; + v8i16 p0_l, q0_l, p1_l = { 0 }; + v8i16 q1_l = { 0 }; + v16u8 p3_org, p2_org, p1_org, p0_org, q0_org, q1_org, q2_org, q3_org; + v8i16 p2_org_r, p1_org_r, p0_org_r, q0_org_r, q1_org_r, q2_org_r; + v8i16 p2_org_l, p1_org_l, p0_org_l, q0_org_l, q1_org_l, q2_org_l; + v8i16 tc_r, tc_l; + v16i8 zero = { 0 }; + v16u8 is_bs_greater_than0; + + tmp_vec = __msa_fill_b(bs0); + bs = (v16u8) __msa_insve_w((v4i32) bs, 0, (v4i32) tmp_vec); + tmp_vec = __msa_fill_b(bs1); + bs = (v16u8) __msa_insve_w((v4i32) bs, 1, (v4i32) tmp_vec); + tmp_vec = __msa_fill_b(bs2); + bs = (v16u8) __msa_insve_w((v4i32) bs, 2, (v4i32) tmp_vec); + tmp_vec = __msa_fill_b(bs3); + bs = (v16u8) __msa_insve_w((v4i32) bs, 3, (v4i32) tmp_vec); + + if (!__msa_test_bz_v(bs)) { + tmp_vec = __msa_fill_b(tc0); + tc = (v16u8) __msa_insve_w((v4i32) tc, 0, (v4i32) tmp_vec); + tmp_vec = __msa_fill_b(tc1); + tc = (v16u8) __msa_insve_w((v4i32) tc, 1, (v4i32) tmp_vec); + tmp_vec = __msa_fill_b(tc2); + tc = (v16u8) __msa_insve_w((v4i32) tc, 2, (v4i32) tmp_vec); + tmp_vec = __msa_fill_b(tc3); + tc = (v16u8) __msa_insve_w((v4i32) tc, 3, (v4i32) tmp_vec); + + is_bs_greater_than0 = (zero < bs); + + { + v16u8 row0, row1, row2, row3, row4, row5, row6, row7; + v16u8 row8, row9, row10, row11, row12, row13, row14, row15; + + src = data; + src -= 4; + + LOAD_8VECS_UB(src, img_width, + row0, row1, row2, row3, row4, row5, row6, row7); + src += (8 * img_width); + LOAD_8VECS_UB(src, img_width, + row8, row9, row10, row11, row12, row13, row14, row15); + + TRANSPOSE16x8_B_UB(row0, row1, row2, row3, row4, row5, row6, row7, + row8, row9, row10, row11, + row12, row13, row14, row15, + p3_org, p2_org, p1_org, p0_org, + q0_org, q1_org, q2_org, q3_org); + } + + { + v16u8 p0_asub_q0, p1_asub_p0, q1_asub_q0, alpha; + v16u8 is_less_than_alpha; + + p0_asub_q0 = __msa_asub_u_b(p0_org, q0_org); + p1_asub_p0 = __msa_asub_u_b(p1_org, p0_org); + q1_asub_q0 = __msa_asub_u_b(q1_org, q0_org); + + alpha = (v16u8) __msa_fill_b(alpha_in); + beta = (v16u8) __msa_fill_b(beta_in); + + is_less_than_alpha = (p0_asub_q0 < alpha); + is_less_than_beta = (p1_asub_p0 < beta); + is_less_than = is_less_than_beta & is_less_than_alpha; + is_less_than_beta = (q1_asub_q0 < beta); + is_less_than = is_less_than_beta & is_less_than; + is_less_than = is_less_than & is_bs_greater_than0; + } + + if (!__msa_test_bz_v(is_less_than)) { + v16i8 negate_tc, sign_negate_tc; + v8i16 negate_tc_r, i16_negatetc_l; + + negate_tc = zero - (v16i8) tc; + sign_negate_tc = __msa_clti_s_b(negate_tc, 0); + + negate_tc_r = (v8i16) __msa_ilvr_b(sign_negate_tc, negate_tc); + i16_negatetc_l = (v8i16) __msa_ilvl_b(sign_negate_tc, negate_tc); + + tc_r = (v8i16) __msa_ilvr_b(zero, (v16i8) tc); + tc_l = (v8i16) __msa_ilvl_b(zero, (v16i8) tc); + + p1_org_r = (v8i16) __msa_ilvr_b(zero, (v16i8) p1_org); + p0_org_r = (v8i16) __msa_ilvr_b(zero, (v16i8) p0_org); + q0_org_r = (v8i16) __msa_ilvr_b(zero, (v16i8) q0_org); + + p1_org_l = (v8i16) __msa_ilvl_b(zero, (v16i8) p1_org); + p0_org_l = (v8i16) __msa_ilvl_b(zero, (v16i8) p0_org); + q0_org_l = (v8i16) __msa_ilvl_b(zero, (v16i8) q0_org); + + { + v16u8 p2_asub_p0; + v8u16 is_less_than_beta_r, is_less_than_beta_l; + + p2_asub_p0 = __msa_asub_u_b(p2_org, p0_org); + is_less_than_beta = (p2_asub_p0 < beta); + is_less_than_beta = is_less_than_beta & is_less_than; + + is_less_than_beta_r = + (v8u16) __msa_sldi_b((v16i8) is_less_than_beta, zero, 8); + if (!__msa_test_bz_v((v16u8) is_less_than_beta_r)) { + p2_org_r = (v8i16) __msa_ilvr_b(zero, (v16i8) p2_org); + + AVC_LOOP_FILTER_P1_OR_Q1(p0_org_r, q0_org_r, + p1_org_r, p2_org_r, + negate_tc_r, tc_r, p1_r); + } + + is_less_than_beta_l = + (v8u16) __msa_sldi_b(zero, (v16i8) is_less_than_beta, 8); + if (!__msa_test_bz_v((v16u8) is_less_than_beta_l)) { + p2_org_l = (v8i16) __msa_ilvl_b(zero, (v16i8) p2_org); + + AVC_LOOP_FILTER_P1_OR_Q1(p0_org_l, q0_org_l, + p1_org_l, p2_org_l, + i16_negatetc_l, tc_l, p1_l); + } + } + + if (!__msa_test_bz_v(is_less_than_beta)) { + p1 = (v16u8) __msa_pckev_b((v16i8) p1_l, (v16i8) p1_r); + p1_org = __msa_bmnz_v(p1_org, p1, is_less_than_beta); + + is_less_than_beta = __msa_andi_b(is_less_than_beta, 1); + tc = tc + is_less_than_beta; + } + + { + v16u8 u8_q2asub_q0; + v8u16 is_less_than_beta_l, is_less_than_beta_r; + + u8_q2asub_q0 = __msa_asub_u_b(q2_org, q0_org); + is_less_than_beta = (u8_q2asub_q0 < beta); + is_less_than_beta = is_less_than_beta & is_less_than; + + q1_org_r = (v8i16) __msa_ilvr_b(zero, (v16i8) q1_org); + + is_less_than_beta_r = + (v8u16) __msa_sldi_b((v16i8) is_less_than_beta, zero, 8); + if (!__msa_test_bz_v((v16u8) is_less_than_beta_r)) { + q2_org_r = (v8i16) __msa_ilvr_b(zero, (v16i8) q2_org); + + AVC_LOOP_FILTER_P1_OR_Q1(p0_org_r, q0_org_r, + q1_org_r, q2_org_r, + negate_tc_r, tc_r, q1_r); + } + + q1_org_l = (v8i16) __msa_ilvl_b(zero, (v16i8) q1_org); + + is_less_than_beta_l = + (v8u16) __msa_sldi_b(zero, (v16i8) is_less_than_beta, 8); + if (!__msa_test_bz_v((v16u8) is_less_than_beta_l)) { + q2_org_l = (v8i16) __msa_ilvl_b(zero, (v16i8) q2_org); + + AVC_LOOP_FILTER_P1_OR_Q1(p0_org_l, q0_org_l, + q1_org_l, q2_org_l, + i16_negatetc_l, tc_l, q1_l); + } + } + + if (!__msa_test_bz_v(is_less_than_beta)) { + q1 = (v16u8) __msa_pckev_b((v16i8) q1_l, (v16i8) q1_r); + q1_org = __msa_bmnz_v(q1_org, q1, is_less_than_beta); + + is_less_than_beta = __msa_andi_b(is_less_than_beta, 1); + tc = tc + is_less_than_beta; + } + + { + v8i16 threshold_r, negate_thresh_r; + v8i16 threshold_l, negate_thresh_l; + v16i8 negate_thresh, sign_negate_thresh; + + negate_thresh = zero - (v16i8) tc; + sign_negate_thresh = __msa_clti_s_b(negate_thresh, 0); + + threshold_r = (v8i16) __msa_ilvr_b(zero, (v16i8) tc); + negate_thresh_r = (v8i16) __msa_ilvr_b(sign_negate_thresh, + negate_thresh); + + AVC_LOOP_FILTER_P0Q0(q0_org_r, p0_org_r, p1_org_r, q1_org_r, + negate_thresh_r, threshold_r, p0_r, q0_r); + + threshold_l = (v8i16) __msa_ilvl_b(zero, (v16i8) tc); + negate_thresh_l = (v8i16) __msa_ilvl_b(sign_negate_thresh, + negate_thresh); + + AVC_LOOP_FILTER_P0Q0(q0_org_l, p0_org_l, p1_org_l, q1_org_l, + negate_thresh_l, threshold_l, p0_l, q0_l); + } + + p0 = (v16u8) __msa_pckev_b((v16i8) p0_l, (v16i8) p0_r); + q0 = (v16u8) __msa_pckev_b((v16i8) q0_l, (v16i8) q0_r); + + p0_org = __msa_bmnz_v(p0_org, p0, is_less_than); + q0_org = __msa_bmnz_v(q0_org, q0, is_less_than); + } + + { + v16u8 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7; + uint32_t out0, out2; + uint16_t out1, out3; + + src = data - 3; + + tmp0 = (v16u8) __msa_ilvr_b((v16i8) p1_org, (v16i8) p2_org); + tmp1 = (v16u8) __msa_ilvr_b((v16i8) q0_org, (v16i8) p0_org); + tmp2 = (v16u8) __msa_ilvr_b((v16i8) q2_org, (v16i8) q1_org); + tmp3 = (v16u8) __msa_ilvr_h((v8i16) tmp1, (v8i16) tmp0); + tmp4 = (v16u8) __msa_ilvl_h((v8i16) tmp1, (v8i16) tmp0); + + tmp0 = (v16u8) __msa_ilvl_b((v16i8) p1_org, (v16i8) p2_org); + tmp1 = (v16u8) __msa_ilvl_b((v16i8) q0_org, (v16i8) p0_org); + tmp5 = (v16u8) __msa_ilvl_b((v16i8) q2_org, (v16i8) q1_org); + tmp6 = (v16u8) __msa_ilvr_h((v8i16) tmp1, (v8i16) tmp0); + tmp7 = (v16u8) __msa_ilvl_h((v8i16) tmp1, (v8i16) tmp0); + + out0 = __msa_copy_u_w((v4i32) tmp3, 0); + out1 = __msa_copy_u_h((v8i16) tmp2, 0); + out2 = __msa_copy_u_w((v4i32) tmp3, 1); + out3 = __msa_copy_u_h((v8i16) tmp2, 1); + + STORE_WORD(src, out0); + STORE_HWORD((src + 4), out1); + src += img_width; + STORE_WORD(src, out2); + STORE_HWORD((src + 4), out3); + + out0 = __msa_copy_u_w((v4i32) tmp3, 2); + out1 = __msa_copy_u_h((v8i16) tmp2, 2); + out2 = __msa_copy_u_w((v4i32) tmp3, 3); + out3 = __msa_copy_u_h((v8i16) tmp2, 3); + + src += img_width; + STORE_WORD(src, out0); + STORE_HWORD((src + 4), out1); + src += img_width; + STORE_WORD(src, out2); + STORE_HWORD((src + 4), out3); + + out0 = __msa_copy_u_w((v4i32) tmp4, 0); + out1 = __msa_copy_u_h((v8i16) tmp2, 4); + out2 = __msa_copy_u_w((v4i32) tmp4, 1); + out3 = __msa_copy_u_h((v8i16) tmp2, 5); + + src += img_width; + STORE_WORD(src, out0); + STORE_HWORD((src + 4), out1); + src += img_width; + STORE_WORD(src, out2); + STORE_HWORD((src + 4), out3); + + out0 = __msa_copy_u_w((v4i32) tmp4, 2); + out1 = __msa_copy_u_h((v8i16) tmp2, 6); + out2 = __msa_copy_u_w((v4i32) tmp4, 3); + out3 = __msa_copy_u_h((v8i16) tmp2, 7); + + src += img_width; + STORE_WORD(src, out0); + STORE_HWORD((src + 4), out1); + src += img_width; + STORE_WORD(src, out2); + STORE_HWORD((src + 4), out3); + + out0 = __msa_copy_u_w((v4i32) tmp6, 0); + out1 = __msa_copy_u_h((v8i16) tmp5, 0); + out2 = __msa_copy_u_w((v4i32) tmp6, 1); + out3 = __msa_copy_u_h((v8i16) tmp5, 1); + + src += img_width; + STORE_WORD(src, out0); + STORE_HWORD((src + 4), out1); + src += img_width; + STORE_WORD(src, out2); + STORE_HWORD((src + 4), out3); + + out0 = __msa_copy_u_w((v4i32) tmp6, 2); + out1 = __msa_copy_u_h((v8i16) tmp5, 2); + out2 = __msa_copy_u_w((v4i32) tmp6, 3); + out3 = __msa_copy_u_h((v8i16) tmp5, 3); + + src += img_width; + STORE_WORD(src, out0); + STORE_HWORD((src + 4), out1); + src += img_width; + STORE_WORD(src, out2); + STORE_HWORD((src + 4), out3); + + out0 = __msa_copy_u_w((v4i32) tmp7, 0); + out1 = __msa_copy_u_h((v8i16) tmp5, 4); + out2 = __msa_copy_u_w((v4i32) tmp7, 1); + out3 = __msa_copy_u_h((v8i16) tmp5, 5); + + src += img_width; + STORE_WORD(src, out0); + STORE_HWORD((src + 4), out1); + src += img_width; + STORE_WORD(src, out2); + STORE_HWORD((src + 4), out3); + + out0 = __msa_copy_u_w((v4i32) tmp7, 2); + out1 = __msa_copy_u_h((v8i16) tmp5, 6); + out2 = __msa_copy_u_w((v4i32) tmp7, 3); + out3 = __msa_copy_u_h((v8i16) tmp5, 7); + + src += img_width; + STORE_WORD(src, out0); + STORE_HWORD((src + 4), out1); + src += img_width; + STORE_WORD(src, out2); + STORE_HWORD((src + 4), out3); + } + } +} + +static void avc_loopfilter_luma_inter_edge_hor_msa(uint8_t *data, + uint8_t bs0, uint8_t bs1, + uint8_t bs2, uint8_t bs3, + uint8_t tc0, uint8_t tc1, + uint8_t tc2, uint8_t tc3, + uint8_t alpha_in, + uint8_t beta_in, + uint32_t image_width) +{ + v16u8 p2_asub_p0, u8_q2asub_q0; + v16u8 alpha, beta, is_less_than, is_less_than_beta; + v16u8 p1, p0, q0, q1; + v8i16 p1_r = { 0 }; + v8i16 p0_r, q0_r, q1_r = { 0 }; + v8i16 p1_l = { 0 }; + v8i16 p0_l, q0_l, q1_l = { 0 }; + v16u8 p2_org, p1_org, p0_org, q0_org, q1_org, q2_org; + v8i16 p2_org_r, p1_org_r, p0_org_r, q0_org_r, q1_org_r, q2_org_r; + v8i16 p2_org_l, p1_org_l, p0_org_l, q0_org_l, q1_org_l, q2_org_l; + v16i8 zero = { 0 }; + v16i8 tmp_vec; + v16u8 bs = { 0 }; + v16u8 tc = { 0 }; + + tmp_vec = __msa_fill_b(bs0); + bs = (v16u8) __msa_insve_w((v4i32) bs, 0, (v4i32) tmp_vec); + tmp_vec = __msa_fill_b(bs1); + bs = (v16u8) __msa_insve_w((v4i32) bs, 1, (v4i32) tmp_vec); + tmp_vec = __msa_fill_b(bs2); + bs = (v16u8) __msa_insve_w((v4i32) bs, 2, (v4i32) tmp_vec); + tmp_vec = __msa_fill_b(bs3); + bs = (v16u8) __msa_insve_w((v4i32) bs, 3, (v4i32) tmp_vec); + + if (!__msa_test_bz_v(bs)) { + tmp_vec = __msa_fill_b(tc0); + tc = (v16u8) __msa_insve_w((v4i32) tc, 0, (v4i32) tmp_vec); + tmp_vec = __msa_fill_b(tc1); + tc = (v16u8) __msa_insve_w((v4i32) tc, 1, (v4i32) tmp_vec); + tmp_vec = __msa_fill_b(tc2); + tc = (v16u8) __msa_insve_w((v4i32) tc, 2, (v4i32) tmp_vec); + tmp_vec = __msa_fill_b(tc3); + tc = (v16u8) __msa_insve_w((v4i32) tc, 3, (v4i32) tmp_vec); + + alpha = (v16u8) __msa_fill_b(alpha_in); + beta = (v16u8) __msa_fill_b(beta_in); + + p2_org = LOAD_UB(data - (3 * image_width)); + p1_org = LOAD_UB(data - (image_width << 1)); + p0_org = LOAD_UB(data - image_width); + q0_org = LOAD_UB(data); + q1_org = LOAD_UB(data + image_width); + + { + v16u8 p0_asub_q0, p1_asub_p0, q1_asub_q0; + v16u8 is_less_than_alpha, is_bs_greater_than0; + + is_bs_greater_than0 = (zero < bs); + p0_asub_q0 = __msa_asub_u_b(p0_org, q0_org); + p1_asub_p0 = __msa_asub_u_b(p1_org, p0_org); + q1_asub_q0 = __msa_asub_u_b(q1_org, q0_org); + + is_less_than_alpha = (p0_asub_q0 < alpha); + is_less_than_beta = (p1_asub_p0 < beta); + is_less_than = is_less_than_beta & is_less_than_alpha; + is_less_than_beta = (q1_asub_q0 < beta); + is_less_than = is_less_than_beta & is_less_than; + is_less_than = is_less_than & is_bs_greater_than0; + } + + if (!__msa_test_bz_v(is_less_than)) { + v16i8 sign_negate_tc, negate_tc; + v8i16 negate_tc_r, i16_negatetc_l, tc_l, tc_r; + + q2_org = LOAD_UB(data + (2 * image_width)); + + negate_tc = zero - (v16i8) tc; + sign_negate_tc = __msa_clti_s_b(negate_tc, 0); + + negate_tc_r = (v8i16) __msa_ilvr_b(sign_negate_tc, negate_tc); + i16_negatetc_l = (v8i16) __msa_ilvl_b(sign_negate_tc, negate_tc); + + tc_r = (v8i16) __msa_ilvr_b(zero, (v16i8) tc); + tc_l = (v8i16) __msa_ilvl_b(zero, (v16i8) tc); + + p1_org_r = (v8i16) __msa_ilvr_b(zero, (v16i8) p1_org); + p0_org_r = (v8i16) __msa_ilvr_b(zero, (v16i8) p0_org); + q0_org_r = (v8i16) __msa_ilvr_b(zero, (v16i8) q0_org); + + p1_org_l = (v8i16) __msa_ilvl_b(zero, (v16i8) p1_org); + p0_org_l = (v8i16) __msa_ilvl_b(zero, (v16i8) p0_org); + q0_org_l = (v8i16) __msa_ilvl_b(zero, (v16i8) q0_org); + + p2_asub_p0 = __msa_asub_u_b(p2_org, p0_org); + is_less_than_beta = (p2_asub_p0 < beta); + is_less_than_beta = is_less_than_beta & is_less_than; + + { + v8u16 is_less_than_beta_r, is_less_than_beta_l; + + is_less_than_beta_r = + (v8u16) __msa_sldi_b((v16i8) is_less_than_beta, zero, 8); + + if (!__msa_test_bz_v((v16u8) is_less_than_beta_r)) { + p2_org_r = (v8i16) __msa_ilvr_b(zero, (v16i8) p2_org); + + AVC_LOOP_FILTER_P1_OR_Q1(p0_org_r, q0_org_r, + p1_org_r, p2_org_r, + negate_tc_r, tc_r, p1_r); + } + + is_less_than_beta_l = + (v8u16) __msa_sldi_b(zero, (v16i8) is_less_than_beta, 8); + if (!__msa_test_bz_v((v16u8) is_less_than_beta_l)) { + p2_org_l = (v8i16) __msa_ilvl_b(zero, (v16i8) p2_org); + + AVC_LOOP_FILTER_P1_OR_Q1(p0_org_l, q0_org_l, + p1_org_l, p2_org_l, + i16_negatetc_l, tc_l, p1_l); + } + } + + if (!__msa_test_bz_v(is_less_than_beta)) { + p1 = (v16u8) __msa_pckev_b((v16i8) p1_l, (v16i8) p1_r); + p1_org = __msa_bmnz_v(p1_org, p1, is_less_than_beta); + STORE_UB(p1_org, data - (2 * image_width)); + + is_less_than_beta = __msa_andi_b(is_less_than_beta, 1); + tc = (v16u8) ((v16i8) tc + (v16i8) is_less_than_beta); + } + + u8_q2asub_q0 = __msa_asub_u_b(q2_org, q0_org); + is_less_than_beta = (u8_q2asub_q0 < beta); + is_less_than_beta = is_less_than_beta & is_less_than; + + { + v8u16 is_less_than_beta_r, is_less_than_beta_l; + + is_less_than_beta_r = + (v8u16) __msa_sldi_b((v16i8) is_less_than_beta, zero, 8); + + q1_org_r = (v8i16) __msa_ilvr_b(zero, (v16i8) q1_org); + if (!__msa_test_bz_v((v16u8) is_less_than_beta_r)) { + q2_org_r = (v8i16) __msa_ilvr_b(zero, (v16i8) q2_org); + + AVC_LOOP_FILTER_P1_OR_Q1(p0_org_r, q0_org_r, + q1_org_r, q2_org_r, + negate_tc_r, tc_r, q1_r); + } + + is_less_than_beta_l = + (v8u16) __msa_sldi_b(zero, (v16i8) is_less_than_beta, 8); + + q1_org_l = (v8i16) __msa_ilvl_b(zero, (v16i8) q1_org); + if (!__msa_test_bz_v((v16u8) is_less_than_beta_l)) { + q2_org_l = (v8i16) __msa_ilvl_b(zero, (v16i8) q2_org); + + AVC_LOOP_FILTER_P1_OR_Q1(p0_org_l, q0_org_l, + q1_org_l, q2_org_l, + i16_negatetc_l, tc_l, q1_l); + } + } + + if (!__msa_test_bz_v(is_less_than_beta)) { + q1 = (v16u8) __msa_pckev_b((v16i8) q1_l, (v16i8) q1_r); + q1_org = __msa_bmnz_v(q1_org, q1, is_less_than_beta); + STORE_UB(q1_org, data + image_width); + + is_less_than_beta = __msa_andi_b(is_less_than_beta, 1); + tc = (v16u8) ((v16i8) tc + (v16i8) is_less_than_beta); + } + + { + v16i8 negate_thresh, sign_negate_thresh; + v8i16 threshold_r, threshold_l; + v8i16 negate_thresh_l, negate_thresh_r; + + negate_thresh = (v16i8) zero - (v16i8) tc; + sign_negate_thresh = __msa_clti_s_b(negate_thresh, 0); + + threshold_r = (v8i16) __msa_ilvr_b(zero, (v16i8) tc); + negate_thresh_r = (v8i16) __msa_ilvr_b(sign_negate_thresh, + negate_thresh); + + AVC_LOOP_FILTER_P0Q0(q0_org_r, p0_org_r, p1_org_r, q1_org_r, + negate_thresh_r, threshold_r, p0_r, q0_r); + + threshold_l = (v8i16) __msa_ilvl_b(zero, (v16i8) tc); + negate_thresh_l = (v8i16) __msa_ilvl_b(sign_negate_thresh, + negate_thresh); + + AVC_LOOP_FILTER_P0Q0(q0_org_l, p0_org_l, p1_org_l, q1_org_l, + negate_thresh_l, threshold_l, p0_l, q0_l); + } + + p0 = (v16u8) __msa_pckev_b((v16i8) p0_l, (v16i8) p0_r); + q0 = (v16u8) __msa_pckev_b((v16i8) q0_l, (v16i8) q0_r); + + p0_org = __msa_bmnz_v(p0_org, p0, is_less_than); + q0_org = __msa_bmnz_v(q0_org, q0, is_less_than); + + STORE_UB(p0_org, (data - image_width)); + STORE_UB(q0_org, data); + } + } +} + +static void avc_h_loop_filter_luma_mbaff_msa(uint8_t *in, int32_t stride, + int32_t alpha_in, int32_t beta_in, + int8_t *tc0) +{ + uint8_t *data = in; + uint32_t out0, out1, out2, out3; + uint64_t load; + uint32_t tc_val; + v16u8 alpha, beta; + v16u8 inp0 = { 0 }; + v16u8 inp1 = { 0 }; + v16u8 inp2 = { 0 }; + v16u8 inp3 = { 0 }; + v16u8 inp4 = { 0 }; + v16u8 inp5 = { 0 }; + v16u8 inp6 = { 0 }; + v16u8 inp7 = { 0 }; + v16u8 src0, src1, src2, src3, src4, src5, src6, src7; + v16u8 p0_asub_q0, p1_asub_p0, q1_asub_q0, p2_asub_p0, q2_asub_q0; + v16u8 is_less_than, is_less_than_alpha, is_less_than_beta; + v16u8 is_less_than_beta1, is_less_than_beta2; + v8i16 tc, tc_orig_r, tc_plus1; + v16u8 is_tc_orig1, is_tc_orig2, tc_orig = { 0 }; + v8i16 p0_ilvr_q0, p0_add_q0, q0_sub_p0, p1_sub_q1; + v8u16 src2_r, src3_r; + v8i16 p2_r, p1_r, q2_r, q1_r; + v16u8 p2, q2, p0, q0; + v16u8 dst0, dst1; + v16i8 zeros = { 0 }; + + alpha = (v16u8) __msa_fill_b(alpha_in); + beta = (v16u8) __msa_fill_b(beta_in); + + if (tc0[0] < 0) { + data += (2 * stride); + } else { + load = LOAD_DWORD(data - 3); + inp0 = (v16u8) __msa_insert_d((v2i64) inp0, 0, load); + load = LOAD_DWORD(data - 3 + stride); + inp1 = (v16u8) __msa_insert_d((v2i64) inp1, 0, load); + data += (2 * stride); + } + + if (tc0[1] < 0) { + data += (2 * stride); + } else { + load = LOAD_DWORD(data - 3); + inp2 = (v16u8) __msa_insert_d((v2i64) inp2, 0, load); + load = LOAD_DWORD(data - 3 + stride); + inp3 = (v16u8) __msa_insert_d((v2i64) inp3, 0, load); + data += (2 * stride); + } + + if (tc0[2] < 0) { + data += (2 * stride); + } else { + load = LOAD_DWORD(data - 3); + inp4 = (v16u8) __msa_insert_d((v2i64) inp4, 0, load); + load = LOAD_DWORD(data - 3 + stride); + inp5 = (v16u8) __msa_insert_d((v2i64) inp5, 0, load); + data += (2 * stride); + } + + if (tc0[3] < 0) { + data += (2 * stride); + } else { + load = LOAD_DWORD(data - 3); + inp6 = (v16u8) __msa_insert_d((v2i64) inp6, 0, load); + load = LOAD_DWORD(data - 3 + stride); + inp7 = (v16u8) __msa_insert_d((v2i64) inp7, 0, load); + data += (2 * stride); + } + + src0 = (v16u8) __msa_ilvr_b((v16i8) inp1, (v16i8) inp0); + src1 = (v16u8) __msa_ilvr_b((v16i8) inp3, (v16i8) inp2); + src2 = (v16u8) __msa_ilvr_b((v16i8) inp5, (v16i8) inp4); + src3 = (v16u8) __msa_ilvr_b((v16i8) inp7, (v16i8) inp6); + src4 = (v16u8) __msa_ilvr_h((v8i16) src1, (v8i16) src0); + src5 = (v16u8) __msa_ilvl_h((v8i16) src1, (v8i16) src0); + src6 = (v16u8) __msa_ilvr_h((v8i16) src3, (v8i16) src2); + src7 = (v16u8) __msa_ilvl_h((v8i16) src3, (v8i16) src2); + src0 = (v16u8) __msa_ilvr_w((v4i32) src6, (v4i32) src4); + src1 = (v16u8) __msa_sldi_b(zeros, (v16i8) src0, 8); + src2 = (v16u8) __msa_ilvl_w((v4i32) src6, (v4i32) src4); + src3 = (v16u8) __msa_sldi_b(zeros, (v16i8) src2, 8); + src4 = (v16u8) __msa_ilvr_w((v4i32) src7, (v4i32) src5); + src5 = (v16u8) __msa_sldi_b(zeros, (v16i8) src4, 8); + + p0_asub_q0 = __msa_asub_u_b(src2, src3); + p1_asub_p0 = __msa_asub_u_b(src1, src2); + q1_asub_q0 = __msa_asub_u_b(src4, src3); + p2_asub_p0 = __msa_asub_u_b(src0, src2); + q2_asub_q0 = __msa_asub_u_b(src5, src3); + + is_less_than_alpha = (p0_asub_q0 < alpha); + is_less_than_beta = (p1_asub_p0 < beta); + is_less_than = is_less_than_alpha & is_less_than_beta; + is_less_than_beta = (q1_asub_q0 < beta); + is_less_than = is_less_than_beta & is_less_than; + + is_less_than_beta1 = (p2_asub_p0 < beta); + is_less_than_beta2 = (q2_asub_q0 < beta); + + p0_ilvr_q0 = (v8i16) __msa_ilvr_b((v16i8) src3, (v16i8) src2); + p0_add_q0 = (v8i16) __msa_hadd_u_h((v16u8) p0_ilvr_q0, (v16u8) p0_ilvr_q0); + p0_add_q0 = __msa_srari_h(p0_add_q0, 1); + + p2_r = (v8i16) __msa_ilvr_b(zeros, (v16i8) src0); + p1_r = (v8i16) __msa_ilvr_b(zeros, (v16i8) src1); + p2_r += p0_add_q0; + p2_r >>= 1; + p2_r -= p1_r; + q2_r = (v8i16) __msa_ilvr_b(zeros, (v16i8) src5); + q1_r = (v8i16) __msa_ilvr_b(zeros, (v16i8) src4); + q2_r += p0_add_q0; + q2_r >>= 1; + q2_r -= q1_r; + + tc_val = LOAD_WORD(tc0); + tc_orig = (v16u8) __msa_insert_w((v4i32) tc_orig, 0, tc_val); + tc_orig = (v16u8) __msa_ilvr_b((v16i8) tc_orig, (v16i8) tc_orig); + is_tc_orig1 = tc_orig; + is_tc_orig2 = tc_orig; + tc_orig_r = (v8i16) __msa_ilvr_b(zeros, (v16i8) tc_orig); + tc = tc_orig_r; + + p2_r = CLIP_MIN_TO_MAX_H(p2_r, -tc_orig_r, tc_orig_r); + q2_r = CLIP_MIN_TO_MAX_H(q2_r, -tc_orig_r, tc_orig_r); + + p2_r += p1_r; + q2_r += q1_r; + + p2 = (v16u8) __msa_pckev_b((v16i8) p2_r, (v16i8) p2_r); + q2 = (v16u8) __msa_pckev_b((v16i8) q2_r, (v16i8) q2_r); + + is_tc_orig1 = (zeros < is_tc_orig1); + is_tc_orig2 = is_tc_orig1; + is_tc_orig1 = is_less_than_beta1 & is_tc_orig1; + is_tc_orig2 = is_less_than_beta2 & is_tc_orig2; + is_tc_orig1 = is_less_than & is_tc_orig1; + is_tc_orig2 = is_less_than & is_tc_orig2; + + p2 = __msa_bmnz_v(src1, p2, is_tc_orig1); + q2 = __msa_bmnz_v(src4, q2, is_tc_orig2); + + q0_sub_p0 = __msa_hsub_u_h((v16u8) p0_ilvr_q0, (v16u8) p0_ilvr_q0); + q0_sub_p0 <<= 2; + p1_sub_q1 = p1_r - q1_r; + q0_sub_p0 += p1_sub_q1; + q0_sub_p0 = __msa_srari_h(q0_sub_p0, 3); + + tc_plus1 = tc + 1; + is_less_than_beta1 = (v16u8) __msa_ilvr_b((v16i8) is_less_than_beta1, + (v16i8) is_less_than_beta1); + tc = (v8i16) __msa_bmnz_v((v16u8) tc, (v16u8) tc_plus1, is_less_than_beta1); + tc_plus1 = tc + 1; + is_less_than_beta2 = (v16u8) __msa_ilvr_b((v16i8) is_less_than_beta2, + (v16i8) is_less_than_beta2); + tc = (v8i16) __msa_bmnz_v((v16u8) tc, (v16u8) tc_plus1, is_less_than_beta2); + + q0_sub_p0 = CLIP_MIN_TO_MAX_H(q0_sub_p0, -tc, tc); + + src2_r = (v8u16) __msa_ilvr_b(zeros, (v16i8) src2); + src3_r = (v8u16) __msa_ilvr_b(zeros, (v16i8) src3); + src2_r += q0_sub_p0; + src3_r -= q0_sub_p0; + + src2_r = (v8u16) CLIP_UNSIGNED_CHAR_H(src2_r); + src3_r = (v8u16) CLIP_UNSIGNED_CHAR_H(src3_r); + + p0 = (v16u8) __msa_pckev_b((v16i8) src2_r, (v16i8) src2_r); + q0 = (v16u8) __msa_pckev_b((v16i8) src3_r, (v16i8) src3_r); + + p0 = __msa_bmnz_v(src2, p0, is_less_than); + q0 = __msa_bmnz_v(src3, q0, is_less_than); + + p2 = (v16u8) __msa_ilvr_b((v16i8) p0, (v16i8) p2); + q2 = (v16u8) __msa_ilvr_b((v16i8) q2, (v16i8) q0); + dst0 = (v16u8) __msa_ilvr_h((v8i16) q2, (v8i16) p2); + dst1 = (v16u8) __msa_ilvl_h((v8i16) q2, (v8i16) p2); + + data = in; + + out0 = __msa_copy_u_w((v4i32) dst0, 0); + out1 = __msa_copy_u_w((v4i32) dst0, 1); + out2 = __msa_copy_u_w((v4i32) dst0, 2); + out3 = __msa_copy_u_w((v4i32) dst0, 3); + + if (tc0[0] < 0) { + data += (2 * stride); + } else { + STORE_WORD((data - 2), out0); + data += stride; + STORE_WORD((data - 2), out1); + data += stride; + } + + if (tc0[1] < 0) { + data += (2 * stride); + } else { + STORE_WORD((data - 2), out2); + data += stride; + STORE_WORD((data - 2), out3); + data += stride; + } + + out0 = __msa_copy_u_w((v4i32) dst1, 0); + out1 = __msa_copy_u_w((v4i32) dst1, 1); + out2 = __msa_copy_u_w((v4i32) dst1, 2); + out3 = __msa_copy_u_w((v4i32) dst1, 3); + + if (tc0[2] < 0) { + data += (2 * stride); + } else { + STORE_WORD((data - 2), out0); + data += stride; + STORE_WORD((data - 2), out1); + data += stride; + } + + if (tc0[3] >= 0) { + STORE_WORD((data - 2), out2); + data += stride; + STORE_WORD((data - 2), out3); + } +} + +static void avc_loopfilter_cb_or_cr_inter_edge_hor_msa(uint8_t *data, + uint8_t bs0, uint8_t bs1, + uint8_t bs2, uint8_t bs3, + uint8_t tc0, uint8_t tc1, + uint8_t tc2, uint8_t tc3, + uint8_t alpha_in, + uint8_t beta_in, + uint32_t img_width) +{ + v16u8 alpha, beta, bs = { 0 }; + v16i8 tmp_vec; + v16u8 tc = { 0 }; + v16u8 p0, q0, p0_asub_q0, p1_asub_p0, q1_asub_q0; + v16u8 is_less_than; + v16u8 is_less_than_beta, is_less_than_alpha, is_bs_greater_than0; + v8i16 p0_r, q0_r; + v16u8 p1_org, p0_org, q0_org, q1_org; + v8i16 p1_org_r, p0_org_r, q0_org_r, q1_org_r; + v16i8 negate_tc, sign_negate_tc; + v8i16 tc_r, negate_tc_r; + v16i8 zero = { 0 }; + + tmp_vec = __msa_fill_b(bs0); + bs = (v16u8) __msa_insve_h((v8i16) bs, 0, (v8i16) tmp_vec); + tmp_vec = __msa_fill_b(bs1); + bs = (v16u8) __msa_insve_h((v8i16) bs, 1, (v8i16) tmp_vec); + tmp_vec = __msa_fill_b(bs2); + bs = (v16u8) __msa_insve_h((v8i16) bs, 2, (v8i16) tmp_vec); + tmp_vec = __msa_fill_b(bs3); + bs = (v16u8) __msa_insve_h((v8i16) bs, 3, (v8i16) tmp_vec); + + if (!__msa_test_bz_v(bs)) { + tmp_vec = __msa_fill_b(tc0); + tc = (v16u8) __msa_insve_h((v8i16) tc, 0, (v8i16) tmp_vec); + tmp_vec = __msa_fill_b(tc1); + tc = (v16u8) __msa_insve_h((v8i16) tc, 1, (v8i16) tmp_vec); + tmp_vec = __msa_fill_b(tc2); + tc = (v16u8) __msa_insve_h((v8i16) tc, 2, (v8i16) tmp_vec); + tmp_vec = __msa_fill_b(tc3); + tc = (v16u8) __msa_insve_h((v8i16) tc, 3, (v8i16) tmp_vec); + + is_bs_greater_than0 = (zero < bs); + + alpha = (v16u8) __msa_fill_b(alpha_in); + beta = (v16u8) __msa_fill_b(beta_in); + + p1_org = LOAD_UB(data - (img_width << 1)); + p0_org = LOAD_UB(data - img_width); + q0_org = LOAD_UB(data); + q1_org = LOAD_UB(data + img_width); + + p0_asub_q0 = __msa_asub_u_b(p0_org, q0_org); + p1_asub_p0 = __msa_asub_u_b(p1_org, p0_org); + q1_asub_q0 = __msa_asub_u_b(q1_org, q0_org); + + is_less_than_alpha = (p0_asub_q0 < alpha); + is_less_than_beta = (p1_asub_p0 < beta); + is_less_than = is_less_than_beta & is_less_than_alpha; + is_less_than_beta = (q1_asub_q0 < beta); + is_less_than = is_less_than_beta & is_less_than; + is_less_than = is_less_than & is_bs_greater_than0; + + is_less_than = (v16u8) __msa_ilvr_d((v2i64) zero, (v2i64) is_less_than); + + if (!__msa_test_bz_v(is_less_than)) { + negate_tc = zero - (v16i8) tc; + sign_negate_tc = __msa_clti_s_b(negate_tc, 0); + + negate_tc_r = (v8i16) __msa_ilvr_b(sign_negate_tc, negate_tc); + + tc_r = (v8i16) __msa_ilvr_b(zero, (v16i8) tc); + + p1_org_r = (v8i16) __msa_ilvr_b(zero, (v16i8) p1_org); + p0_org_r = (v8i16) __msa_ilvr_b(zero, (v16i8) p0_org); + q0_org_r = (v8i16) __msa_ilvr_b(zero, (v16i8) q0_org); + q1_org_r = (v8i16) __msa_ilvr_b(zero, (v16i8) q1_org); + + AVC_LOOP_FILTER_P0Q0(q0_org_r, p0_org_r, p1_org_r, q1_org_r, + negate_tc_r, tc_r, p0_r, q0_r); + + p0 = (v16u8) __msa_pckev_b(zero, (v16i8) p0_r); + q0 = (v16u8) __msa_pckev_b(zero, (v16i8) q0_r); + + p0_org = __msa_bmnz_v(p0_org, p0, is_less_than); + q0_org = __msa_bmnz_v(q0_org, q0, is_less_than); + + STORE_UB(q0_org, data); + STORE_UB(p0_org, (data - img_width)); + } + } +} + +static void avc_loopfilter_cb_or_cr_inter_edge_ver_msa(uint8_t *data, + uint8_t bs0, uint8_t bs1, + uint8_t bs2, uint8_t bs3, + uint8_t tc0, uint8_t tc1, + uint8_t tc2, uint8_t tc3, + uint8_t alpha_in, + uint8_t beta_in, + uint32_t img_width) +{ + uint8_t *src; + uint16_t out0, out1, out2, out3; + v16u8 alpha, beta; + v16u8 p0_asub_q0, p1_asub_p0, q1_asub_q0; + v16u8 is_less_than, is_less_than_beta, is_less_than_alpha; + v16u8 p0, q0; + v8i16 p0_r = { 0 }; + v8i16 q0_r = { 0 }; + v16u8 p1_org, p0_org, q0_org, q1_org; + v8i16 p1_org_r, p0_org_r, q0_org_r, q1_org_r; + v16u8 is_bs_greater_than0; + v8i16 tc_r, negate_tc_r, tmp1; + v16i8 negate_tc, sign_negate_tc, tmp_vec; + v16i8 zero = { 0 }; + v16u8 row0, row1, row2, row3, row4, row5, row6, row7; + v16u8 bs = { 0 }, tc = { 0 }; + + tmp_vec = __msa_fill_b(bs0); + bs = (v16u8) __msa_insve_h((v8i16) bs, 0, (v8i16) tmp_vec); + tmp_vec = __msa_fill_b(bs1); + bs = (v16u8) __msa_insve_h((v8i16) bs, 1, (v8i16) tmp_vec); + tmp_vec = __msa_fill_b(bs2); + bs = (v16u8) __msa_insve_h((v8i16) bs, 2, (v8i16) tmp_vec); + tmp_vec = __msa_fill_b(bs3); + bs = (v16u8) __msa_insve_h((v8i16) bs, 3, (v8i16) tmp_vec); + + if (!__msa_test_bz_v(bs)) { + tmp_vec = __msa_fill_b(tc0); + tc = (v16u8) __msa_insve_h((v8i16) tc, 0, (v8i16) tmp_vec); + tmp_vec = __msa_fill_b(tc1); + tc = (v16u8) __msa_insve_h((v8i16) tc, 1, (v8i16) tmp_vec); + tmp_vec = __msa_fill_b(tc2); + tc = (v16u8) __msa_insve_h((v8i16) tc, 2, (v8i16) tmp_vec); + tmp_vec = __msa_fill_b(tc3); + tc = (v16u8) __msa_insve_h((v8i16) tc, 3, (v8i16) tmp_vec); + + is_bs_greater_than0 = (zero < bs); + + LOAD_8VECS_UB((data - 2), img_width, + row0, row1, row2, row3, row4, row5, row6, row7); + + TRANSPOSE8x4_B_UB(row0, row1, row2, row3, + row4, row5, row6, row7, + p1_org, p0_org, q0_org, q1_org); + + p0_asub_q0 = __msa_asub_u_b(p0_org, q0_org); + p1_asub_p0 = __msa_asub_u_b(p1_org, p0_org); + q1_asub_q0 = __msa_asub_u_b(q1_org, q0_org); + + alpha = (v16u8) __msa_fill_b(alpha_in); + beta = (v16u8) __msa_fill_b(beta_in); + + is_less_than_alpha = (p0_asub_q0 < alpha); + is_less_than_beta = (p1_asub_p0 < beta); + is_less_than = is_less_than_beta & is_less_than_alpha; + is_less_than_beta = (q1_asub_q0 < beta); + is_less_than = is_less_than_beta & is_less_than; + is_less_than = is_bs_greater_than0 & is_less_than; + + is_less_than = (v16u8) __msa_ilvr_d((v2i64) zero, (v2i64) is_less_than); + + if (!__msa_test_bz_v(is_less_than)) { + p1_org_r = (v8i16) __msa_ilvr_b(zero, (v16i8) p1_org); + p0_org_r = (v8i16) __msa_ilvr_b(zero, (v16i8) p0_org); + q0_org_r = (v8i16) __msa_ilvr_b(zero, (v16i8) q0_org); + q1_org_r = (v8i16) __msa_ilvr_b(zero, (v16i8) q1_org); + + negate_tc = zero - (v16i8) tc; + sign_negate_tc = __msa_clti_s_b(negate_tc, 0); + + negate_tc_r = (v8i16) __msa_ilvr_b(sign_negate_tc, negate_tc); + + tc_r = (v8i16) __msa_ilvr_b(zero, (v16i8) tc); + + AVC_LOOP_FILTER_P0Q0(q0_org_r, p0_org_r, p1_org_r, q1_org_r, + negate_tc_r, tc_r, p0_r, q0_r); + + p0 = (v16u8) __msa_pckev_b(zero, (v16i8) p0_r); + q0 = (v16u8) __msa_pckev_b(zero, (v16i8) q0_r); + + p0_org = __msa_bmnz_v(p0_org, p0, is_less_than); + q0_org = __msa_bmnz_v(q0_org, q0, is_less_than); + + tmp1 = (v8i16) __msa_ilvr_b((v16i8) q0_org, (v16i8) p0_org); + + src = data - 1; + + out0 = __msa_copy_u_h(tmp1, 0); + out1 = __msa_copy_u_h(tmp1, 1); + out2 = __msa_copy_u_h(tmp1, 2); + out3 = __msa_copy_u_h(tmp1, 3); + + STORE_HWORD(src, out0); + src += img_width; + STORE_HWORD(src, out1); + src += img_width; + STORE_HWORD(src, out2); + src += img_width; + STORE_HWORD(src, out3); + + out0 = __msa_copy_u_h(tmp1, 4); + out1 = __msa_copy_u_h(tmp1, 5); + out2 = __msa_copy_u_h(tmp1, 6); + out3 = __msa_copy_u_h(tmp1, 7); + + src += img_width; + STORE_HWORD(src, out0); + src += img_width; + STORE_HWORD(src, out1); + src += img_width; + STORE_HWORD(src, out2); + src += img_width; + STORE_HWORD(src, out3); + } + } +} + +static void avc_h_loop_filter_chroma422_msa(uint8_t *src, + int32_t stride, + int32_t alpha_in, + int32_t beta_in, + int8_t *tc0) +{ + int32_t col, tc_val; + int16_t out0, out1, out2, out3; + v16u8 alpha, beta, res; + + alpha = (v16u8) __msa_fill_b(alpha_in); + beta = (v16u8) __msa_fill_b(beta_in); + + for (col = 0; col < 4; col++) { + tc_val = (tc0[col] - 1) + 1; + + if (tc_val <= 0) { + src += (4 * stride); + continue; + } + + AVC_LPF_H_CHROMA_422(src, stride, tc_val, alpha, beta, res); + + out0 = __msa_copy_s_h((v8i16) res, 0); + out1 = __msa_copy_s_h((v8i16) res, 1); + out2 = __msa_copy_s_h((v8i16) res, 2); + out3 = __msa_copy_s_h((v8i16) res, 3); + + STORE_HWORD((src - 1), out0); + src += stride; + STORE_HWORD((src - 1), out1); + src += stride; + STORE_HWORD((src - 1), out2); + src += stride; + STORE_HWORD((src - 1), out3); + src += stride; + } +} + +static void avc_h_loop_filter_chroma422_mbaff_msa(uint8_t *src, + int32_t stride, + int32_t alpha_in, + int32_t beta_in, + int8_t *tc0) +{ + int32_t col, tc_val; + int16_t out0, out1; + v16u8 alpha, beta, res; + + alpha = (v16u8) __msa_fill_b(alpha_in); + beta = (v16u8) __msa_fill_b(beta_in); + + for (col = 0; col < 4; col++) { + tc_val = (tc0[col] - 1) + 1; + + if (tc_val <= 0) { + src += 2 * stride; + continue; + } + + AVC_LPF_H_2BYTE_CHROMA_422(src, stride, tc_val, alpha, beta, res); + + out0 = __msa_copy_s_h((v8i16) res, 0); + out1 = __msa_copy_s_h((v8i16) res, 1); + + STORE_HWORD((src - 1), out0); + src += stride; + STORE_HWORD((src - 1), out1); + src += stride; + } +} + +void ff_h264_h_lpf_luma_inter_msa(uint8_t *data, int img_width, + int alpha, int beta, int8_t *tc) +{ + uint8_t bs0 = 1; + uint8_t bs1 = 1; + uint8_t bs2 = 1; + uint8_t bs3 = 1; + + if (tc[0] < 0) + bs0 = 0; + if (tc[1] < 0) + bs1 = 0; + if (tc[2] < 0) + bs2 = 0; + if (tc[3] < 0) + bs3 = 0; + + avc_loopfilter_luma_inter_edge_ver_msa(data, + bs0, bs1, bs2, bs3, + tc[0], tc[1], tc[2], tc[3], + alpha, beta, img_width); +} + +void ff_h264_v_lpf_luma_inter_msa(uint8_t *data, int img_width, + int alpha, int beta, int8_t *tc) +{ + + uint8_t bs0 = 1; + uint8_t bs1 = 1; + uint8_t bs2 = 1; + uint8_t bs3 = 1; + + if (tc[0] < 0) + bs0 = 0; + if (tc[1] < 0) + bs1 = 0; + if (tc[2] < 0) + bs2 = 0; + if (tc[3] < 0) + bs3 = 0; + + avc_loopfilter_luma_inter_edge_hor_msa(data, + bs0, bs1, bs2, bs3, + tc[0], tc[1], tc[2], tc[3], + alpha, beta, img_width); +} + +void ff_h264_h_lpf_chroma_inter_msa(uint8_t *data, int img_width, + int alpha, int beta, int8_t *tc) +{ + uint8_t bs0 = 1; + uint8_t bs1 = 1; + uint8_t bs2 = 1; + uint8_t bs3 = 1; + + if (tc[0] < 0) + bs0 = 0; + if (tc[1] < 0) + bs1 = 0; + if (tc[2] < 0) + bs2 = 0; + if (tc[3] < 0) + bs3 = 0; + + avc_loopfilter_cb_or_cr_inter_edge_ver_msa(data, + bs0, bs1, bs2, bs3, + tc[0], tc[1], tc[2], tc[3], + alpha, beta, + img_width); +} + +void ff_h264_v_lpf_chroma_inter_msa(uint8_t *data, int img_width, + int alpha, int beta, int8_t *tc) +{ + uint8_t bs0 = 1; + uint8_t bs1 = 1; + uint8_t bs2 = 1; + uint8_t bs3 = 1; + + if (tc[0] < 0) + bs0 = 0; + if (tc[1] < 0) + bs1 = 0; + if (tc[2] < 0) + bs2 = 0; + if (tc[3] < 0) + bs3 = 0; + + avc_loopfilter_cb_or_cr_inter_edge_hor_msa(data, + bs0, bs1, bs2, bs3, + tc[0], tc[1], tc[2], tc[3], + alpha, beta, + img_width); +} + +void ff_h264_h_lpf_luma_intra_msa(uint8_t *data, int img_width, + int alpha, int beta) +{ + avc_loopfilter_luma_intra_edge_ver_msa(data, (uint8_t) alpha, + (uint8_t) beta, + (unsigned int) img_width); +} + +void ff_h264_v_lpf_luma_intra_msa(uint8_t *data, int img_width, + int alpha, int beta) +{ + avc_loopfilter_luma_intra_edge_hor_msa(data, (uint8_t) alpha, + (uint8_t) beta, + (unsigned int) img_width); +} + +void ff_h264_h_lpf_chroma_intra_msa(uint8_t *data, int img_width, + int alpha, int beta) +{ + avc_loopfilter_cb_or_cr_intra_edge_ver_msa(data, (uint8_t) alpha, + (uint8_t) beta, + (unsigned int) img_width); +} + +void ff_h264_v_lpf_chroma_intra_msa(uint8_t *data, int img_width, + int alpha, int beta) +{ + avc_loopfilter_cb_or_cr_intra_edge_hor_msa(data, (uint8_t) alpha, + (uint8_t) beta, + (unsigned int) img_width); +} + +void ff_h264_h_loop_filter_chroma422_msa(uint8_t *src, + int32_t ystride, + int32_t alpha, int32_t beta, + int8_t *tc0) +{ + avc_h_loop_filter_chroma422_msa(src, ystride, alpha, beta, tc0); +} + +void ff_h264_h_loop_filter_chroma422_mbaff_msa(uint8_t *src, + int32_t ystride, + int32_t alpha, + int32_t beta, + int8_t *tc0) +{ + avc_h_loop_filter_chroma422_mbaff_msa(src, ystride, alpha, beta, tc0); +} + +void ff_h264_h_loop_filter_luma_mbaff_msa(uint8_t *src, + int32_t ystride, + int32_t alpha, + int32_t beta, + int8_t *tc0) +{ + avc_h_loop_filter_luma_mbaff_msa(src, ystride, alpha, beta, tc0); +} + +void ff_h264_h_loop_filter_luma_mbaff_intra_msa(uint8_t *src, + int32_t ystride, + int32_t alpha, + int32_t beta) +{ + avc_h_loop_filter_luma_mbaff_intra_msa(src, ystride, alpha, beta); +} + +void ff_weight_h264_pixels16_8_msa(uint8_t *src, int stride, + int height, int log2_denom, + int weight_src, int offset) +{ + avc_wgt_16width_msa(src, stride, + height, log2_denom, weight_src, offset); +} + +void ff_weight_h264_pixels8_8_msa(uint8_t *src, int stride, + int height, int log2_denom, + int weight_src, int offset) +{ + avc_wgt_8width_msa(src, stride, + height, log2_denom, weight_src, offset); +} + +void ff_weight_h264_pixels4_8_msa(uint8_t *src, int stride, + int height, int log2_denom, + int weight_src, int offset) +{ + avc_wgt_4width_msa(src, stride, + height, log2_denom, weight_src, offset); +} + +void ff_biweight_h264_pixels16_8_msa(uint8_t *dst, uint8_t *src, + int stride, int height, + int log2_denom, int weight_dst, + int weight_src, int offset) +{ + avc_biwgt_16width_msa(src, stride, + dst, stride, + height, log2_denom, + weight_src, weight_dst, offset); +} + +void ff_biweight_h264_pixels8_8_msa(uint8_t *dst, uint8_t *src, + int stride, int height, + int log2_denom, int weight_dst, + int weight_src, int offset) +{ + avc_biwgt_8width_msa(src, stride, + dst, stride, + height, log2_denom, + weight_src, weight_dst, offset); +} + +void ff_biweight_h264_pixels4_8_msa(uint8_t *dst, uint8_t *src, + int stride, int height, + int log2_denom, int weight_dst, + int weight_src, int offset) +{ + avc_biwgt_4width_msa(src, stride, + dst, stride, + height, log2_denom, + weight_src, weight_dst, offset); +} diff --git a/libavutil/mips/generic_macros_msa.h b/libavutil/mips/generic_macros_msa.h index d2fd87c..c84baef 100644 --- a/libavutil/mips/generic_macros_msa.h +++ b/libavutil/mips/generic_macros_msa.h @@ -45,12 +45,95 @@ out_m; \ } ) +#define STORE_UB(vec, pdest) \ +{ \ + *((v16u8 *) (pdest)) = (vec); \ +} + +#define STORE_SB(vec, pdest) \ +{ \ + *((v16i8 *) (pdest)) = (vec); \ +} + #define STORE_SH(vec, pdest) \ { \ *((v8i16 *) (pdest)) = (vec); \ } #if (__mips_isa_rev >= 6) + #define LOAD_WORD(psrc) \ + ( { \ + uint8_t *src_m = (uint8_t *) (psrc); \ + uint32_t val_m; \ + \ + __asm__ __volatile__ ( \ + "lw %[val_m], %[src_m] \n\t" \ + \ + : [val_m] "=r" (val_m) \ + : [src_m] "m" (*src_m) \ + ); \ + \ + val_m; \ + } ) + + #if (__mips == 64) + #define LOAD_DWORD(psrc) \ + ( { \ + uint8_t *src_m = (uint8_t *) (psrc); \ + uint64_t val_m = 0; \ + \ + __asm__ __volatile__ ( \ + "ld %[val_m], %[src_m] \n\t" \ + \ + : [val_m] "=r" (val_m) \ + : [src_m] "m" (*src_m) \ + ); \ + \ + val_m; \ + } ) + #else + #define LOAD_DWORD(psrc) \ + ( { \ + uint8_t *src1_m = (uint8_t *) (psrc); \ + uint8_t *src2_m = ((uint8_t *) (psrc)) + 4; \ + uint32_t val0_m, val1_m; \ + uint64_t genval_m = 0; \ + \ + __asm__ __volatile__ ( \ + "lw %[val0_m], %[src1_m] \n\t" \ + \ + : [val0_m] "=r" (val0_m) \ + : [src1_m] "m" (*src1_m) \ + ); \ + \ + __asm__ __volatile__ ( \ + "lw %[val1_m], %[src2_m] \n\t" \ + \ + : [val1_m] "=r" (val1_m) \ + : [src2_m] "m" (*src2_m) \ + ); \ + \ + genval_m = (uint64_t) (val1_m); \ + genval_m = (uint64_t) ((genval_m << 32) & 0xFFFFFFFF00000000); \ + genval_m = (uint64_t) (genval_m | (uint64_t) val0_m); \ + \ + genval_m; \ + } ) + #endif + + #define STORE_WORD(pdst, val) \ + { \ + uint8_t *dst_ptr_m = (uint8_t *) (pdst); \ + uint32_t val_m = (val); \ + \ + __asm__ __volatile__ ( \ + "sw %[val_m], %[dst_ptr_m] \n\t" \ + \ + : [dst_ptr_m] "=m" (*dst_ptr_m) \ + : [val_m] "r" (val_m) \ + ); \ + } + #define STORE_DWORD(pdst, val) \ { \ uint8_t *dst_ptr_m = (uint8_t *) (pdst); \ @@ -63,7 +146,93 @@ : [val_m] "r" (val_m) \ ); \ } + + #define STORE_HWORD(pdst, val) \ + { \ + uint8_t *dst_ptr_m = (uint8_t *) (pdst); \ + uint16_t val_m = (val); \ + \ + __asm__ __volatile__ ( \ + "sh %[val_m], %[dst_ptr_m] \n\t" \ + \ + : [dst_ptr_m] "=m" (*dst_ptr_m) \ + : [val_m] "r" (val_m) \ + ); \ + } + #else + #define LOAD_WORD(psrc) \ + ( { \ + uint8_t *src_m = (uint8_t *) (psrc); \ + uint32_t val_m; \ + \ + __asm__ __volatile__ ( \ + "ulw %[val_m], %[src_m] \n\t" \ + \ + : [val_m] "=r" (val_m) \ + : [src_m] "m" (*src_m) \ + ); \ + \ + val_m; \ + } ) + + #if (__mips == 64) + #define LOAD_DWORD(psrc) \ + ( { \ + uint8_t *src_m = (uint8_t *) (psrc); \ + uint64_t val_m = 0; \ + \ + __asm__ __volatile__ ( \ + "uld %[val_m], %[src_m] \n\t" \ + \ + : [val_m] "=r" (val_m) \ + : [src_m] "m" (*src_m) \ + ); \ + \ + val_m; \ + } ) + #else + #define LOAD_DWORD(psrc) \ + ( { \ + uint8_t *src1_m = (uint8_t *) (psrc); \ + uint8_t *src2_m = ((uint8_t *) (psrc)) + 4; \ + uint32_t val0_m, val1_m; \ + uint64_t genval_m = 0; \ + \ + __asm__ __volatile__ ( \ + "ulw %[val0_m], %[src1_m] \n\t" \ + \ + : [val0_m] "=r" (val0_m) \ + : [src1_m] "m" (*src1_m) \ + ); \ + \ + __asm__ __volatile__ ( \ + "ulw %[val1_m], %[src2_m] \n\t" \ + \ + : [val1_m] "=r" (val1_m) \ + : [src2_m] "m" (*src2_m) \ + ); \ + \ + genval_m = (uint64_t) (val1_m); \ + genval_m = (uint64_t) ((genval_m << 32) & 0xFFFFFFFF00000000); \ + genval_m = (uint64_t) (genval_m | (uint64_t) val0_m); \ + \ + genval_m; \ + } ) + #endif + #define STORE_WORD(pdst, val) \ + { \ + uint8_t *dst_ptr_m = (uint8_t *) (pdst); \ + uint32_t val_m = (val); \ + \ + __asm__ __volatile__ ( \ + "usw %[val_m], %[dst_ptr_m] \n\t" \ + \ + : [dst_ptr_m] "=m" (*dst_ptr_m) \ + : [val_m] "r" (val_m) \ + ); \ + } + #define STORE_DWORD(pdst, val) \ { \ uint8_t *dst1_m = (uint8_t *) (pdst); \ @@ -81,8 +250,40 @@ : [val0_m] "r" (val0_m), [val1_m] "r" (val1_m) \ ); \ } + + #define STORE_HWORD(pdst, val) \ + { \ + uint8_t *dst_ptr_m = (uint8_t *) (pdst); \ + uint16_t val_m = (val); \ + \ + __asm__ __volatile__ ( \ + "ush %[val_m], %[dst_ptr_m] \n\t" \ + \ + : [dst_ptr_m] "=m" (*dst_ptr_m) \ + : [val_m] "r" (val_m) \ + ); \ + } + #endif +#define LOAD_4WORDS_WITH_STRIDE(psrc, src_stride, \ + src0, src1, src2, src3) \ +{ \ + src0 = LOAD_WORD(psrc + 0 * src_stride); \ + src1 = LOAD_WORD(psrc + 1 * src_stride); \ + src2 = LOAD_WORD(psrc + 2 * src_stride); \ + src3 = LOAD_WORD(psrc + 3 * src_stride); \ +} + +#define LOAD_4VECS_UB(psrc, stride, \ + val0, val1, val2, val3) \ +{ \ + val0 = LOAD_UB(psrc + 0 * stride); \ + val1 = LOAD_UB(psrc + 1 * stride); \ + val2 = LOAD_UB(psrc + 2 * stride); \ + val3 = LOAD_UB(psrc + 3 * stride); \ +} + #define LOAD_4VECS_SB(psrc, stride, \ val0, val1, val2, val3) \ { \ @@ -105,6 +306,16 @@ val6 = LOAD_SB((psrc) + 6 * (stride)); \ } +#define LOAD_8VECS_UB(psrc, stride, \ + out0, out1, out2, out3, \ + out4, out5, out6, out7) \ +{ \ + LOAD_4VECS_UB((psrc), (stride), \ + (out0), (out1), (out2), (out3)); \ + LOAD_4VECS_UB((psrc + 4 * stride), (stride), \ + (out4), (out5), (out6), (out7)); \ +} + #define LOAD_8VECS_SB(psrc, stride, \ out0, out1, out2, out3, \ out4, out5, out6, out7) \ @@ -115,6 +326,162 @@ (out4), (out5), (out6), (out7)); \ } +#define STORE_4VECS_UB(dst_out, pitch, \ + in0, in1, in2, in3) \ +{ \ + STORE_UB((in0), (dst_out)); \ + STORE_UB((in1), ((dst_out) + (pitch))); \ + STORE_UB((in2), ((dst_out) + 2 * (pitch))); \ + STORE_UB((in3), ((dst_out) + 3 * (pitch))); \ +} + +#define STORE_4VECS_SB(dst_out, pitch, \ + in0, in1, in2, in3) \ +{ \ + STORE_SB((in0), (dst_out)); \ + STORE_SB((in1), ((dst_out) + (pitch))); \ + STORE_SB((in2), ((dst_out) + 2 * (pitch))); \ + STORE_SB((in3), ((dst_out) + 3 * (pitch))); \ +} + +/* out = (((in) < (min)) ? (min) : (((in) > (max)) ? (max) : (in))) */ +#define CLIP_MIN_TO_MAX_H(in, min, max) \ +( { \ + v8i16 out_m; \ + \ + out_m = __msa_max_s_h((v8i16) (min), (v8i16) (in)); \ + out_m = __msa_min_s_h((v8i16) (max), (v8i16) out_m); \ + out_m; \ +} ) + +#define CLIP_UNSIGNED_CHAR_H(in) \ +( { \ + v8i16 max_m = __msa_ldi_h(255); \ + v8i16 out_m; \ + \ + out_m = __msa_maxi_s_h((v8i16) (in), 0); \ + out_m = __msa_min_s_h((v8i16) max_m, (v8i16) out_m); \ + out_m; \ +} ) + +/* Byte Transpose */ +#define TRANSPOSE4x4_B_UB(in0, in1, in2, in3, \ + out0, out1, out2, out3) \ +{ \ + v16i8 zero_m = { 0 }; \ + v16i8 s0_m, s1_m, s2_m, s3_m; \ + \ + s0_m = (v16i8) __msa_ilvr_d((v2i64) (in1), (v2i64) (in0)); \ + s1_m = (v16i8) __msa_ilvr_d((v2i64) (in3), (v2i64) (in2)); \ + s2_m = __msa_ilvr_b(s1_m, s0_m); \ + s3_m = __msa_ilvl_b(s1_m, s0_m); \ + \ + out0 = (v16u8) __msa_ilvr_b(s3_m, s2_m); \ + out1 = (v16u8) __msa_sldi_b(zero_m, (v16i8) out0, 4); \ + out2 = (v16u8) __msa_sldi_b(zero_m, (v16i8) out1, 4); \ + out3 = (v16u8) __msa_sldi_b(zero_m, (v16i8) out2, 4); \ +} + +/* transpose 8x4 to 4x8 */ +#define TRANSPOSE8x4_B_UB(in0, in1, in2, in3, \ + in4, in5, in6, in7, \ + out0, out1, out2, out3) \ +{ \ + v16i8 tmp0_m, tmp1_m, tmp2_m, tmp3_m; \ + \ + tmp0_m = (v16i8) __msa_ilvev_w((v4i32) (in4), (v4i32) (in0)); \ + tmp1_m = (v16i8) __msa_ilvev_w((v4i32) (in5), (v4i32) (in1)); \ + tmp2_m = __msa_ilvr_b(tmp1_m, tmp0_m); \ + tmp0_m = (v16i8) __msa_ilvev_w((v4i32) (in6), (v4i32) (in2)); \ + tmp1_m = (v16i8) __msa_ilvev_w((v4i32) (in7), (v4i32) (in3)); \ + \ + tmp3_m = __msa_ilvr_b(tmp1_m, tmp0_m); \ + tmp0_m = (v16i8) __msa_ilvr_h((v8i16) tmp3_m, (v8i16) tmp2_m); \ + tmp1_m = (v16i8) __msa_ilvl_h((v8i16) tmp3_m, (v8i16) tmp2_m); \ + \ + out0 = (v16u8) __msa_ilvr_w((v4i32) tmp1_m, (v4i32) tmp0_m); \ + out2 = (v16u8) __msa_ilvl_w((v4i32) tmp1_m, (v4i32) tmp0_m); \ + out1 = (v16u8) __msa_ilvl_d((v2i64) out2, (v2i64) out0); \ + out3 = (v16u8) __msa_ilvl_d((v2i64) out0, (v2i64) out2); \ +} + +/* transpose 16x8 matrix into 8x16 */ +#define TRANSPOSE16x8_B_UB(in0, in1, in2, in3, \ + in4, in5, in6, in7, \ + in8, in9, in10, in11, \ + in12, in13, in14, in15, \ + out0, out1, out2, out3, \ + out4, out5, out6, out7) \ +{ \ + v16u8 tmp0_m, tmp1_m, tmp2_m, tmp3_m; \ + v16u8 tmp4_m, tmp5_m, tmp6_m, tmp7_m; \ + \ + (out7) = (v16u8) __msa_ilvev_d((v2i64) (in8), (v2i64) (in0)); \ + (out6) = (v16u8) __msa_ilvev_d((v2i64) (in9), (v2i64) (in1)); \ + (out5) = (v16u8) __msa_ilvev_d((v2i64) (in10), (v2i64) (in2)); \ + (out4) = (v16u8) __msa_ilvev_d((v2i64) (in11), (v2i64) (in3)); \ + (out3) = (v16u8) __msa_ilvev_d((v2i64) (in12), (v2i64) (in4)); \ + (out2) = (v16u8) __msa_ilvev_d((v2i64) (in13), (v2i64) (in5)); \ + (out1) = (v16u8) __msa_ilvev_d((v2i64) (in14), (v2i64) (in6)); \ + (out0) = (v16u8) __msa_ilvev_d((v2i64) (in15), (v2i64) (in7)); \ + \ + tmp0_m = (v16u8) __msa_ilvev_b((v16i8) (out6), (v16i8) (out7)); \ + tmp4_m = (v16u8) __msa_ilvod_b((v16i8) (out6), (v16i8) (out7)); \ + tmp1_m = (v16u8) __msa_ilvev_b((v16i8) (out4), (v16i8) (out5)); \ + tmp5_m = (v16u8) __msa_ilvod_b((v16i8) (out4), (v16i8) (out5)); \ + (out5) = (v16u8) __msa_ilvev_b((v16i8) (out2), (v16i8) (out3)); \ + tmp6_m = (v16u8) __msa_ilvod_b((v16i8) (out2), (v16i8) (out3)); \ + (out7) = (v16u8) __msa_ilvev_b((v16i8) (out0), (v16i8) (out1)); \ + tmp7_m = (v16u8) __msa_ilvod_b((v16i8) (out0), (v16i8) (out1)); \ + \ + tmp2_m = (v16u8) __msa_ilvev_h((v8i16) tmp1_m, (v8i16) tmp0_m); \ + tmp3_m = (v16u8) __msa_ilvev_h((v8i16) (out7), (v8i16) (out5)); \ + (out0) = (v16u8) __msa_ilvev_w((v4i32) tmp3_m, (v4i32) tmp2_m); \ + (out4) = (v16u8) __msa_ilvod_w((v4i32) tmp3_m, (v4i32) tmp2_m); \ + \ + tmp2_m = (v16u8) __msa_ilvod_h((v8i16) tmp1_m, (v8i16) tmp0_m); \ + tmp3_m = (v16u8) __msa_ilvod_h((v8i16) (out7), (v8i16) (out5)); \ + (out2) = (v16u8) __msa_ilvev_w((v4i32) tmp3_m, (v4i32) tmp2_m); \ + (out6) = (v16u8) __msa_ilvod_w((v4i32) tmp3_m, (v4i32) tmp2_m); \ + \ + tmp2_m = (v16u8) __msa_ilvev_h((v8i16) tmp5_m, (v8i16) tmp4_m); \ + tmp3_m = (v16u8) __msa_ilvev_h((v8i16) tmp7_m, (v8i16) tmp6_m); \ + (out1) = (v16u8) __msa_ilvev_w((v4i32) tmp3_m, (v4i32) tmp2_m); \ + (out5) = (v16u8) __msa_ilvod_w((v4i32) tmp3_m, (v4i32) tmp2_m); \ + \ + tmp2_m = (v16u8) __msa_ilvod_h((v8i16) tmp5_m, (v8i16) tmp4_m); \ + tmp2_m = (v16u8) __msa_ilvod_h((v8i16) tmp5_m, (v8i16) tmp4_m); \ + tmp3_m = (v16u8) __msa_ilvod_h((v8i16) tmp7_m, (v8i16) tmp6_m); \ + tmp3_m = (v16u8) __msa_ilvod_h((v8i16) tmp7_m, (v8i16) tmp6_m); \ + (out3) = (v16u8) __msa_ilvev_w((v4i32) tmp3_m, (v4i32) tmp2_m); \ + (out7) = (v16u8) __msa_ilvod_w((v4i32) tmp3_m, (v4i32) tmp2_m); \ +} + +#define ILV_B_LRLR_UH(in0, in1, in2, in3, \ + out0, out1, out2, out3) \ +{ \ + out0 = (v8u16) __msa_ilvl_b((v16i8) (in1), (v16i8) (in0)); \ + out1 = (v8u16) __msa_ilvr_b((v16i8) (in1), (v16i8) (in0)); \ + out2 = (v8u16) __msa_ilvl_b((v16i8) (in3), (v16i8) (in2)); \ + out3 = (v8u16) __msa_ilvr_b((v16i8) (in3), (v16i8) (in2)); \ +} + +#define ILV_B_LRLR_SH(in0, in1, in2, in3, \ + out0, out1, out2, out3) \ +{ \ + out0 = (v8i16) __msa_ilvl_b((v16i8) (in1), (v16i8) (in0)); \ + out1 = (v8i16) __msa_ilvr_b((v16i8) (in1), (v16i8) (in0)); \ + out2 = (v8i16) __msa_ilvl_b((v16i8) (in3), (v16i8) (in2)); \ + out3 = (v8i16) __msa_ilvr_b((v16i8) (in3), (v16i8) (in2)); \ +} + +#define ILVR_B_2VECS_UB(in0_r, in1_r, in0_l, in1_l, \ + out0, out1) \ +{ \ + out0 = (v16u8) __msa_ilvr_b((v16i8) (in0_l), (v16i8) (in0_r)); \ + out1 = (v16u8) __msa_ilvr_b((v16i8) (in1_l), (v16i8) (in1_r)); \ +} + #define ILVR_B_2VECS_SB(in0_r, in1_r, in0_l, in1_l, \ out0, out1) \ { \ @@ -164,6 +531,40 @@ out6, out7); \ } +#define ILVR_B_2VECS_UH(in0_r, in1_r, in0_l, in1_l, \ + out0, out1) \ +{ \ + out0 = (v8u16) __msa_ilvr_b((v16i8) (in0_l), (v16i8) (in0_r)); \ + out1 = (v8u16) __msa_ilvr_b((v16i8) (in1_l), (v16i8) (in1_r)); \ +} + +#define ILVR_B_2VECS_SH(in0_r, in1_r, in0_l, in1_l, \ + out0, out1) \ +{ \ + out0 = (v8i16) __msa_ilvr_b((v16i8) (in0_l), (v16i8) (in0_r)); \ + out1 = (v8i16) __msa_ilvr_b((v16i8) (in1_l), (v16i8) (in1_r)); \ +} + +#define ILVR_B_4VECS_UH(in0_r, in1_r, in2_r, in3_r, \ + in0_l, in1_l, in2_l, in3_l, \ + out0, out1, out2, out3) \ +{ \ + ILVR_B_2VECS_UH(in0_r, in1_r, in0_l, in1_l, \ + out0, out1); \ + ILVR_B_2VECS_UH(in2_r, in3_r, in2_l, in3_l, \ + out2, out3); \ +} + +#define ILVR_B_4VECS_SH(in0_r, in1_r, in2_r, in3_r, \ + in0_l, in1_l, in2_l, in3_l, \ + out0, out1, out2, out3) \ +{ \ + ILVR_B_2VECS_SH(in0_r, in1_r, in0_l, in1_l, \ + out0, out1); \ + ILVR_B_2VECS_SH(in2_r, in3_r, in2_l, in3_l, \ + out2, out3); \ +} + #define ILVL_B_2VECS_SB(in0_r, in1_r, in0_l, in1_l, \ out0, out1) \ { \ @@ -223,6 +624,44 @@ out3, in3_l, in3_r); \ } +#define MAXI_S_H_4VECS_UH(vec0, vec1, vec2, vec3, \ + max_value) \ +{ \ + vec0 = (v8u16) __msa_maxi_s_h((v8i16) (vec0), (max_value)); \ + vec1 = (v8u16) __msa_maxi_s_h((v8i16) (vec1), (max_value)); \ + vec2 = (v8u16) __msa_maxi_s_h((v8i16) (vec2), (max_value)); \ + vec3 = (v8u16) __msa_maxi_s_h((v8i16) (vec3), (max_value)); \ +} + +#define SAT_U_H_4VECS_UH(vec0, vec1, vec2, vec3, \ + sat_value) \ +{ \ + vec0 = __msa_sat_u_h((v8u16) (vec0), (sat_value)); \ + vec1 = __msa_sat_u_h((v8u16) (vec1), (sat_value)); \ + vec2 = __msa_sat_u_h((v8u16) (vec2), (sat_value)); \ + vec3 = __msa_sat_u_h((v8u16) (vec3), (sat_value)); \ +} + +#define PCKEV_B_4VECS_UB(in0_l, in1_l, in2_l, in3_l, \ + in0_r, in1_r, in2_r, in3_r, \ + out0, out1, out2, out3) \ +{ \ + out0 = (v16u8) __msa_pckev_b((v16i8) (in0_l), (v16i8) (in0_r)); \ + out1 = (v16u8) __msa_pckev_b((v16i8) (in1_l), (v16i8) (in1_r)); \ + out2 = (v16u8) __msa_pckev_b((v16i8) (in2_l), (v16i8) (in2_r)); \ + out3 = (v16u8) __msa_pckev_b((v16i8) (in3_l), (v16i8) (in3_r)); \ +} + +#define PCKEV_B_4VECS_SB(in0_l, in1_l, in2_l, in3_l, \ + in0_r, in1_r, in2_r, in3_r, \ + out0, out1, out2, out3) \ +{ \ + out0 = __msa_pckev_b((v16i8) (in0_l), (v16i8) (in0_r)); \ + out1 = __msa_pckev_b((v16i8) (in1_l), (v16i8) (in1_r)); \ + out2 = __msa_pckev_b((v16i8) (in2_l), (v16i8) (in2_r)); \ + out3 = __msa_pckev_b((v16i8) (in3_l), (v16i8) (in3_r)); \ +} + #define XORI_B_2VECS_SB(val0, val1, \ out0, out1, xor_val) \ { \ @@ -282,4 +721,83 @@ out4, out5, out6, out7, xor_val); \ } +#define ADDS_S_H_4VECS_UH(in0, in1, in2, in3, \ + in4, in5, in6, in7, \ + out0, out1, out2, out3) \ +{ \ + out0 = (v8u16) __msa_adds_s_h((v8i16) (in0), (v8i16) (in1)); \ + out1 = (v8u16) __msa_adds_s_h((v8i16) (in2), (v8i16) (in3)); \ + out2 = (v8u16) __msa_adds_s_h((v8i16) (in4), (v8i16) (in5)); \ + out3 = (v8u16) __msa_adds_s_h((v8i16) (in6), (v8i16) (in7)); \ +} + +#define SRA_4VECS(in0, in1, in2, in3, \ + out0, out1, out2, out3, \ + shift_right_vec) \ +{ \ + out0 = (in0) >> (shift_right_vec); \ + out1 = (in1) >> (shift_right_vec); \ + out2 = (in2) >> (shift_right_vec); \ + out3 = (in3) >> (shift_right_vec); \ +} + +#define SRL_H_4VECS_UH(in0, in1, in2, in3, \ + out0, out1, out2, out3, \ + shift_right_vec) \ +{ \ + out0 = (v8u16) __msa_srl_h((v8i16) (in0), (v8i16) (shift_right_vec)); \ + out1 = (v8u16) __msa_srl_h((v8i16) (in1), (v8i16) (shift_right_vec)); \ + out2 = (v8u16) __msa_srl_h((v8i16) (in2), (v8i16) (shift_right_vec)); \ + out3 = (v8u16) __msa_srl_h((v8i16) (in3), (v8i16) (shift_right_vec)); \ +} + +#define PCKEV_B_STORE_4_BYTES_4(in1, in2, in3, in4, \ + pdst, stride) \ +{ \ + uint32_t out0_m, out1_m, out2_m, out3_m; \ + v16i8 tmp0_m, tmp1_m; \ + uint8_t *dst_m = (uint8_t *) (pdst); \ + \ + tmp0_m = __msa_pckev_b((v16i8) (in2), (v16i8) (in1)); \ + tmp1_m = __msa_pckev_b((v16i8) (in4), (v16i8) (in3)); \ + \ + out0_m = __msa_copy_u_w((v4i32) tmp0_m, 0); \ + out1_m = __msa_copy_u_w((v4i32) tmp0_m, 2); \ + out2_m = __msa_copy_u_w((v4i32) tmp1_m, 0); \ + out3_m = __msa_copy_u_w((v4i32) tmp1_m, 2); \ + \ + STORE_WORD(dst_m, out0_m); \ + dst_m += stride; \ + STORE_WORD(dst_m, out1_m); \ + dst_m += stride; \ + STORE_WORD(dst_m, out2_m); \ + dst_m += stride; \ + STORE_WORD(dst_m, out3_m); \ +} + +#define PCKEV_B_STORE_8_BYTES_4(in1, in2, in3, in4, \ + pdst, stride) \ +{ \ + uint64_t out0_m, out1_m, out2_m, out3_m; \ + v16i8 tmp0_m, tmp1_m; \ + uint8_t *dst_m = (uint8_t *) (pdst); \ + \ + tmp0_m = __msa_pckev_b((v16i8) (in2), (v16i8) (in1)); \ + tmp1_m = __msa_pckev_b((v16i8) (in4), (v16i8) (in3)); \ + \ + out0_m = __msa_copy_u_d((v2i64) tmp0_m, 0); \ + out1_m = __msa_copy_u_d((v2i64) tmp0_m, 1); \ + out2_m = __msa_copy_u_d((v2i64) tmp1_m, 0); \ + out3_m = __msa_copy_u_d((v2i64) tmp1_m, 1); \ + \ + STORE_DWORD(dst_m, out0_m); \ + dst_m += stride; \ + STORE_DWORD(dst_m, out1_m); \ + dst_m += stride; \ + STORE_DWORD(dst_m, out2_m); \ + dst_m += stride; \ + STORE_DWORD(dst_m, out3_m); \ +} + + #endif /* AVUTIL_MIPS_GENERIC_MACROS_MSA_H */ -- 2.3.2

7 10

[PATCH 1/2] avformat/mpegts: Factorize version checking code out
by Michael Niedermayer 01 May '15

01 May '15

Signed-off-by: Michael Niedermayer <michaelni(a)gmx.at> --- libavformat/mpegts.c | 22 ++++++++++++++-------- 1 file changed, 14 insertions(+), 8 deletions(-) diff --git a/libavformat/mpegts.c b/libavformat/mpegts.c index 0e5c2ba..d707cc3 100644 --- a/libavformat/mpegts.c +++ b/libavformat/mpegts.c @@ -580,6 +580,16 @@ typedef struct SectionHeader { uint8_t last_sec_num; } SectionHeader; +static int skip_identical(const SectionHeader *h, MpegTSSectionFilter *tssf) +{ + if (h->version == tssf->last_ver) + return 1; + + tssf->last_ver = h->version; + + return 0; +} + static inline int get8(const uint8_t **pp, const uint8_t *p_end) { const uint8_t *p; @@ -1469,9 +1479,8 @@ static void m4sl_cb(MpegTSFilter *filter, const uint8_t *section, return; if (h.tid != M4OD_TID) return; - if (h.version == tssf->last_ver) + if (skip_identical(&h, tssf)) return; - tssf->last_ver = h.version; mp4_read_od(s, p, (unsigned) (p_end - p), mp4_descr, &mp4_descr_count, MAX_MP4_DESCR_COUNT); @@ -1816,9 +1825,8 @@ static void pmt_cb(MpegTSFilter *filter, const uint8_t *section, int section_len p = section; if (parse_section_header(h, &p, p_end) < 0) return; - if (h->version == tssf->last_ver) + if (skip_identical(h, tssf)) return; - tssf->last_ver = h->version; av_log(ts->stream, AV_LOG_TRACE, "sid=0x%x sec_num=%d/%d version=%d\n", h->id, h->sec_num, h->last_sec_num, h->version); @@ -1986,9 +1994,8 @@ static void pat_cb(MpegTSFilter *filter, const uint8_t *section, int section_len if (ts->skip_changes) return; - if (h->version == tssf->last_ver) + if (skip_identical(h, tssf)) return; - tssf->last_ver = h->version; ts->stream->ts_id = h->id; clear_programs(ts); @@ -2061,9 +2068,8 @@ static void sdt_cb(MpegTSFilter *filter, const uint8_t *section, int section_len return; if (ts->skip_changes) return; - if (h->version == tssf->last_ver) + if (skip_identical(h, tssf)) return; - tssf->last_ver = h->version; onid = get16(&p, p_end); if (onid < 0) -- 1.7.9.5

3 5