From 9606e4b6e68de8c5d9ffe0c829716f0d5dbf1215 Mon Sep 17 00:00:00 2001
From: Paul B Mahol <onemda@gmail.com>
Date: Wed, 28 Aug 2019 17:28:16 +0200
Subject: avcodec/dsddec: add slice threading support

---
 libavcodec/dsd.c    | 12 ++++++++----
 libavcodec/dsddec.c | 47 ++++++++++++++++++++++++++++++++---------------
 2 files changed, 40 insertions(+), 19 deletions(-)

diff --git a/libavcodec/dsd.c b/libavcodec/dsd.c
index 9104f38476..8a1fc21b0d 100644
--- a/libavcodec/dsd.c
+++ b/libavcodec/dsd.c
@@ -56,23 +56,26 @@ void ff_dsd2pcm_translate(DSDContext* s, size_t samples, int lsbf,
                           const unsigned char *src, ptrdiff_t src_stride,
                           float *dst, ptrdiff_t dst_stride)
 {
+    unsigned char buf[FIFOSIZE];
     unsigned pos, i;
     unsigned char* p;
     double sum;
 
     pos = s->pos;
 
+    memcpy(buf, s->buf, sizeof(buf));
+
     while (samples-- > 0) {
-        s->buf[pos] = lsbf ? ff_reverse[*src] : *src;
+        buf[pos] = lsbf ? ff_reverse[*src] : *src;
         src += src_stride;
 
-        p = s->buf + ((pos - CTABLES) & FIFOMASK);
+        p = buf + ((pos - CTABLES) & FIFOMASK);
         *p = ff_reverse[*p];
 
         sum = 0.0;
         for (i = 0; i < CTABLES; i++) {
-            unsigned char a = s->buf[(pos                   - i) & FIFOMASK];
-            unsigned char b = s->buf[(pos - (CTABLES*2 - 1) + i) & FIFOMASK];
+            unsigned char a = buf[(pos                   - i) & FIFOMASK];
+            unsigned char b = buf[(pos - (CTABLES*2 - 1) + i) & FIFOMASK];
             sum += ctables[i][a] + ctables[i][b];
         }
 
@@ -83,4 +86,5 @@ void ff_dsd2pcm_translate(DSDContext* s, size_t samples, int lsbf,
     }
 
     s->pos = pos;
+    memcpy(s->buf, buf, sizeof(buf));
 }
diff --git a/libavcodec/dsddec.c b/libavcodec/dsddec.c
index 2c5c357acc..7c3ae15768 100644
--- a/libavcodec/dsddec.c
+++ b/libavcodec/dsddec.c
@@ -61,17 +61,20 @@ static av_cold int decode_init(AVCodecContext *avctx)
     return 0;
 }
 
-static int decode_frame(AVCodecContext *avctx, void *data,
-                        int *got_frame_ptr, AVPacket *avpkt)
+typedef struct ThreadData {
+    AVFrame *frame;
+    AVPacket *avpkt;
+} ThreadData;
+
+static int dsd_channel(AVCodecContext *avctx, void *tdata, int j, int threadnr)
 {
-    DSDContext * s = avctx->priv_data;
-    AVFrame *frame = data;
-    int ret, i;
     int lsbf = avctx->codec_id == AV_CODEC_ID_DSD_LSBF || avctx->codec_id == AV_CODEC_ID_DSD_LSBF_PLANAR;
-    int src_next;
-    int src_stride;
-
-    frame->nb_samples = avpkt->size / avctx->channels;
+    DSDContext *s = avctx->priv_data;
+    ThreadData *td = tdata;
+    AVFrame *frame = td->frame;
+    AVPacket *avpkt = td->avpkt;
+    int src_next, src_stride;
+    float *dst = ((float **)frame->extended_data)[j];
 
     if (avctx->codec_id == AV_CODEC_ID_DSD_LSBF_PLANAR || avctx->codec_id == AV_CODEC_ID_DSD_MSBF_PLANAR) {
         src_next   = frame->nb_samples;
@@ -81,15 +84,28 @@ static int decode_frame(AVCodecContext *avctx, void *data,
         src_stride = avctx->channels;
     }
 
+    ff_dsd2pcm_translate(&s[j], frame->nb_samples, lsbf,
+                         avpkt->data + j * src_next, src_stride,
+                         dst, 1);
+
+    return 0;
+}
+
+static int decode_frame(AVCodecContext *avctx, void *data,
+                        int *got_frame_ptr, AVPacket *avpkt)
+{
+    ThreadData td;
+    AVFrame *frame = data;
+    int ret;
+
+    frame->nb_samples = avpkt->size / avctx->channels;
+
     if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
         return ret;
 
-    for (i = 0; i < avctx->channels; i++) {
-        float * dst = ((float **)frame->extended_data)[i];
-        ff_dsd2pcm_translate(&s[i], frame->nb_samples, lsbf,
-            avpkt->data + i * src_next, src_stride,
-            dst, 1);
-    }
+    td.frame = frame;
+    td.avpkt = avpkt;
+    avctx->execute2(avctx, dsd_channel, &td, NULL, avctx->channels);
 
     *got_frame_ptr = 1;
     return frame->nb_samples * avctx->channels;
@@ -103,6 +119,7 @@ AVCodec ff_##name_##_decoder = { \
     .id           = AV_CODEC_ID_##id_, \
     .init         = decode_init, \
     .decode       = decode_frame, \
+    .capabilities = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_SLICE_THREADS, \
     .sample_fmts  = (const enum AVSampleFormat[]){ AV_SAMPLE_FMT_FLTP, \
                                                    AV_SAMPLE_FMT_NONE }, \
 };
-- 
cgit v1.2.3