avfilter/vf_spp: support 10bit per sample

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

2 files changed, 79 insertions, 23 deletions

diff --git a/libavfilter/vf_spp.c b/libavfilter/vf_spp.c
index 43fce691c6..48d23a1da2 100644
--- a/libavfilter/vf_spp.c
+++ b/libavfilter/vf_spp.c
@@ -186,6 +186,36 @@ static void store_slice_c(uint8_t *dst, const uint16_t *src,
     }
 }
 
+static void store_slice16_c(uint16_t *dst, const uint16_t *src,
+                            int dst_linesize, int src_linesize,
+                            int width, int height, int log2_scale,
+                            const uint8_t dither[8][8])
+{
+    int y, x;
+
+#define STORE16(pos) do {                                                   \
+    temp = ((src[x + y*src_linesize + pos] << log2_scale) + (d[pos]>>1)) >> 5;   \
+    if (temp & 0x400)                                                       \
+        temp = ~(temp >> 31);                                               \
+    dst[x + y*dst_linesize + pos] = temp;                                   \
+} while (0)
+
+    for (y = 0; y < height; y++) {
+        const uint8_t *d = dither[y];
+        for (x = 0; x < width; x += 8) {
+            int temp;
+            STORE16(0);
+            STORE16(1);
+            STORE16(2);
+            STORE16(3);
+            STORE16(4);
+            STORE16(5);
+            STORE16(6);
+            STORE16(7);
+        }
+    }
+}
+
 static inline void add_block(uint16_t *dst, int linesize, const int16_t block[64])
 {
     int y;
@@ -212,7 +242,7 @@ static inline int norm_qscale(int qscale, int type)
 
 static void filter(SPPContext *p, uint8_t *dst, uint8_t *src,
                    int dst_linesize, int src_linesize, int width, int height,
-                   const uint8_t *qp_table, int qp_stride, int is_luma)
+                   const uint8_t *qp_table, int qp_stride, int is_luma, int sample_bytes)
 {
     int x, y, i;
     const int count = 1 << p->log2_count;
@@ -220,18 +250,26 @@ static void filter(SPPContext *p, uint8_t *dst, uint8_t *src,
     DECLARE_ALIGNED(16, uint64_t, block_align)[32];
     int16_t *block  = (int16_t *)block_align;
     int16_t *block2 = (int16_t *)(block_align + 16);
+    uint16_t *psrc16 = (uint16_t*)p->src;
 
     for (y = 0; y < height; y++) {
         int index = 8 + 8*linesize + y*linesize;
-        memcpy(p->src + index, src + y*src_linesize, width);
-        for (x = 0; x < 8; x++) {
-            p->src[index         - x - 1] = p->src[index +         x    ];
-            p->src[index + width + x    ] = p->src[index + width - x - 1];
+        memcpy(p->src + index*sample_bytes, src + y*src_linesize, width*sample_bytes);
+        if (sample_bytes == 1) {
+            for (x = 0; x < 8; x++) {
+                p->src[index         - x - 1] = p->src[index +         x    ];
+                p->src[index + width + x    ] = p->src[index + width - x - 1];
+            }
+        } else {
+            for (x = 0; x < 8; x++) {
+                psrc16[index         - x - 1] = psrc16[index +         x    ];
+                psrc16[index + width + x    ] = psrc16[index + width - x - 1];
+            }
         }
     }
     for (y = 0; y < 8; y++) {
-        memcpy(p->src + (       7-y)*linesize, p->src + (       y+8)*linesize, linesize);
-        memcpy(p->src + (height+8+y)*linesize, p->src + (height-y+7)*linesize, linesize);
+        memcpy(p->src + (       7-y)*linesize * sample_bytes, p->src + (       y+8)*linesize * sample_bytes, linesize * sample_bytes);
+        memcpy(p->src + (height+8+y)*linesize * sample_bytes, p->src + (height-y+7)*linesize * sample_bytes, linesize * sample_bytes);
     }
 
     for (y = 0; y < height + 8; y += 8) {
@@ -250,18 +288,26 @@ static void filter(SPPContext *p, uint8_t *dst, uint8_t *src,
                 const int x1 = x + offset[i + count - 1][0];
                 const int y1 = y + offset[i + count - 1][1];
                 const int index = x1 + y1*linesize;
-                p->dct->get_pixels(block, p->src + index, linesize);
+                p->dct->get_pixels(block, p->src + sample_bytes*index, sample_bytes*linesize);
                 p->dct->fdct(block);
                 p->requantize(block2, block, qp, p->dct->idct_permutation);
                 p->dct->idct(block2);
                 add_block(p->temp + index, linesize, block2);
             }
         }
-        if (y)
-            p->store_slice(dst + (y - 8) * dst_linesize, p->temp + 8 + y*linesize,
-                           dst_linesize, linesize, width,
-                           FFMIN(8, height + 8 - y), MAX_LEVEL - p->log2_count,
-                           ldither);
+        if (y) {
+            if (sample_bytes == 1) {
+                p->store_slice(dst + (y - 8) * dst_linesize, p->temp + 8 + y*linesize,
+                               dst_linesize, linesize, width,
+                               FFMIN(8, height + 8 - y), MAX_LEVEL - p->log2_count,
+                               ldither);
+            } else {
+                store_slice16_c(dst + (y - 8) * dst_linesize, p->temp + 8 + y*linesize,
+                                dst_linesize/2, linesize, width,
+                                FFMIN(8, height + 8 - y), MAX_LEVEL - p->log2_count,
+                                ldither);
+            }
+        }
     }
 }
 
@@ -273,6 +319,8 @@ static int query_formats(AVFilterContext *ctx)
         AV_PIX_FMT_YUV410P,  AV_PIX_FMT_YUV440P,
         AV_PIX_FMT_YUVJ444P, AV_PIX_FMT_YUVJ422P,
         AV_PIX_FMT_YUVJ420P, AV_PIX_FMT_YUVJ440P,
+        AV_PIX_FMT_YUV444P10,  AV_PIX_FMT_YUV422P10,
+        AV_PIX_FMT_YUV420P10,
         AV_PIX_FMT_NONE
     };
     ff_set_common_formats(ctx, ff_make_format_list(pix_fmts));
@@ -284,12 +332,19 @@ static int config_input(AVFilterLink *inlink)
     SPPContext *spp = inlink->dst->priv;
     const int h = FFALIGN(inlink->h + 16, 16);
     const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
+    const int bps = desc->comp[0].depth_minus1 + 1;
+
+    av_opt_set_int(spp->dct, "bits_per_sample", bps, 0);
+    avcodec_dct_init(spp->dct);
+
+    if (ARCH_X86)
+        ff_spp_init_x86(spp);
 
     spp->hsub = desc->log2_chroma_w;
     spp->vsub = desc->log2_chroma_h;
     spp->temp_linesize = FFALIGN(inlink->w + 16, 16);
     spp->temp = av_malloc_array(spp->temp_linesize, h * sizeof(*spp->temp));
-    spp->src  = av_malloc_array(spp->temp_linesize, h * sizeof(*spp->src));
+    spp->src  = av_malloc_array(spp->temp_linesize, h * sizeof(*spp->src) * 2);
 
     if (!spp->temp || !spp->src)
         return AVERROR(ENOMEM);
@@ -304,6 +359,8 @@ static int filter_frame(AVFilterLink *inlink, AVFrame *in)
     AVFrame *out = in;
     int qp_stride = 0;
     const int8_t *qp_table = NULL;
+    const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
+    const int sample_bytes = desc->comp[0].depth_minus1 < 8 ? 1 : 2;
 
     /* if we are not in a constant user quantizer mode and we don't want to use
      * the quantizers from the B-frames (B-frames often have a higher QP), we
@@ -363,9 +420,9 @@ static int filter_frame(AVFilterLink *inlink, AVFrame *in)
                 out->height = in->height;
             }
 
-            filter(spp, out->data[0], in->data[0], out->linesize[0], in->linesize[0], inlink->w, inlink->h, qp_table, qp_stride, 1);
-            filter(spp, out->data[1], in->data[1], out->linesize[1], in->linesize[1], cw,        ch,        qp_table, qp_stride, 0);
-            filter(spp, out->data[2], in->data[2], out->linesize[2], in->linesize[2], cw,        ch,        qp_table, qp_stride, 0);
+            filter(spp, out->data[0], in->data[0], out->linesize[0], in->linesize[0], inlink->w, inlink->h, qp_table, qp_stride, 1, sample_bytes);
+            filter(spp, out->data[1], in->data[1], out->linesize[1], in->linesize[1], cw,        ch,        qp_table, qp_stride, 0, sample_bytes);
+            filter(spp, out->data[2], in->data[2], out->linesize[2], in->linesize[2], cw,        ch,        qp_table, qp_stride, 0, sample_bytes);
             emms_c();
         }
     }
@@ -415,14 +472,11 @@ static av_cold int init_dict(AVFilterContext *ctx, AVDictionary **opts)
         av_dict_free(opts);
     }
 
-    avcodec_dct_init(spp->dct);
     spp->store_slice = store_slice_c;
     switch (spp->mode) {
     case MODE_HARD: spp->requantize = hardthresh_c; break;
     case MODE_SOFT: spp->requantize = softthresh_c; break;
     }
-    if (ARCH_X86)
-        ff_spp_init_x86(spp);
     return 0;
 }
 
diff --git a/libavfilter/x86/vf_spp.c b/libavfilter/x86/vf_spp.c
index 862190b928..016f526a6d 100644
--- a/libavfilter/x86/vf_spp.c
+++ b/libavfilter/x86/vf_spp.c
@@ -224,9 +224,11 @@ av_cold void ff_spp_init_x86(SPPContext *s)
 
     if (cpu_flags & AV_CPU_FLAG_MMX) {
         s->store_slice = store_slice_mmx;
-        switch (s->mode) {
-        case 0: s->requantize = hardthresh_mmx; break;
-        case 1: s->requantize = softthresh_mmx; break;
+        if (av_get_int(s->dct, "bits_per_sample", NULL) <= 8) {
+            switch (s->mode) {
+            case 0: s->requantize = hardthresh_mmx; break;
+            case 1: s->requantize = softthresh_mmx; break;
+            }
         }
     }
 #endif