swscale: 9,10 bits pixel format output support

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

libswscale/swscale.c

@@ -401,23 +401,76 @@ static av_always_inline void yuv2yuvX16inC_template(const int16_t *lumFilter, co
     }
 }
 
+static av_always_inline void yuv2yuvXNinC_template(const int16_t *lumFilter, const int16_t **lumSrc, int lumFilterSize,
+                                                   const int16_t *chrFilter, const int16_t **chrSrc, int chrFilterSize,
+                                                   const int16_t **alpSrc, uint16_t *dest, uint16_t *uDest, uint16_t *vDest, uint16_t *aDest,
+                                                   int dstW, int chrDstW, int big_endian, int depth)
+{
+    //FIXME Optimize (just quickly written not optimized..)
+    int i;
+
+    for (i = 0; i < dstW; i++) {
+        int val = 1 << (26-depth);
+        int j;
+
+        for (j = 0; j < lumFilterSize; j++)
+            val += lumSrc[j][i] * lumFilter[j];
+
+        if (big_endian) {
+            AV_WB16(&dest[i], av_clip(val >> (27-depth), 0, (1<<depth)-1));
+        } else {
+            AV_WL16(&dest[i], av_clip(val >> (27-depth), 0, (1<<depth)-1));
+        }
+    }
+
+    if (uDest) {
+        for (i = 0; i < chrDstW; i++) {
+            int u = 1 << (26-depth);
+            int v = 1 << (26-depth);
+            int j;
+
+            for (j = 0; j < chrFilterSize; j++) {
+                u += chrSrc[j][i       ] * chrFilter[j];
+                v += chrSrc[j][i + VOFW] * chrFilter[j];
+            }
+
+            if (big_endian) {
+                AV_WB16(&uDest[i], av_clip(u >> (27-depth), 0, (1<<depth)-1));
+                AV_WB16(&vDest[i], av_clip(v >> (27-depth), 0, (1<<depth)-1));
+            } else {
+                AV_WL16(&uDest[i], av_clip(u >> (27-depth), 0, (1<<depth)-1));
+                AV_WL16(&vDest[i], av_clip(v >> (27-depth), 0, (1<<depth)-1));
+            }
+        }
+    }
+}
+
 static inline void yuv2yuvX16inC(const int16_t *lumFilter, const int16_t **lumSrc, int lumFilterSize,
                                  const int16_t *chrFilter, const int16_t **chrSrc, int chrFilterSize,
                                  const int16_t **alpSrc, uint16_t *dest, uint16_t *uDest, uint16_t *vDest, uint16_t *aDest, int dstW, int chrDstW,
                                  enum PixelFormat dstFormat)
 {
-    if (isBE(dstFormat)) {
-        yuv2yuvX16inC_template(lumFilter, lumSrc, lumFilterSize,
-                               chrFilter, chrSrc, chrFilterSize,
-                               alpSrc,
-                               dest, uDest, vDest, aDest,
-                               dstW, chrDstW, 1);
+    if (isNBPS(dstFormat)) {
+        const int depth = av_pix_fmt_descriptors[dstFormat].comp[0].depth_minus1+1;
+        yuv2yuvXNinC_template(lumFilter, lumSrc, lumFilterSize,
+                              chrFilter, chrSrc, chrFilterSize,
+                              alpSrc,
+                              dest, uDest, vDest, aDest,
+                              dstW, chrDstW, isBE(dstFormat), depth);
     } else {
-        yuv2yuvX16inC_template(lumFilter, lumSrc, lumFilterSize,
-                               chrFilter, chrSrc, chrFilterSize,
-                               alpSrc,
-                               dest, uDest, vDest, aDest,
-                               dstW, chrDstW, 0);
+        if (isBE(dstFormat)) {
+            yuv2yuvX16inC_template(lumFilter, lumSrc, lumFilterSize,
+                                   chrFilter, chrSrc, chrFilterSize,
+                                   alpSrc,
+                                   dest, uDest, vDest, aDest,
+                                   dstW, chrDstW, 1);
+        } else {
+            yuv2yuvX16inC_template(lumFilter, lumSrc, lumFilterSize,
+                                   chrFilter, chrSrc, chrFilterSize,
+                                   alpSrc,
+                                   dest, uDest, vDest, aDest,
+                                   dstW, chrDstW, 0);
+        }
     }
 }
 
@@ -1823,34 +1876,62 @@ static int planarCopyWrapper(SwsContext *c, const uint8_t* src[], int srcStride[
                 length*=2;
             fillPlane(dst[plane], dstStride[plane], length, height, y, (plane==3) ? 255 : 128);
         } else {
-            if(isNBPS(c->srcFormat)) {
-                const int depth = av_pix_fmt_descriptors[c->srcFormat].comp[plane].depth_minus1+1;
+            if(isNBPS(c->srcFormat) || isNBPS(c->dstFormat)) {
+                const int src_depth = av_pix_fmt_descriptors[c->srcFormat].comp[plane].depth_minus1+1;
+                const int dst_depth = av_pix_fmt_descriptors[c->dstFormat].comp[plane].depth_minus1+1;
                 uint16_t *srcPtr2 = (uint16_t*)srcPtr;
+                uint16_t *dstPtr2 = (uint16_t*)dstPtr;
 
-                if (is16BPS(c->dstFormat)) {
-                    uint16_t *dstPtr2 = (uint16_t*)dstPtr;
+                if (dst_depth == 8) {
+                    for (i = 0; i < height; i++) {
+                        uint8_t *dither= dithers[src_depth-9][i&7];
+                        for (j = 0; j < length-7; j+=8){
+                            dstPtr[j+0] = (srcPtr2[j+0] + dither[0])>>(src_depth-8);
+                            dstPtr[j+1] = (srcPtr2[j+1] + dither[1])>>(src_depth-8);
+                            dstPtr[j+2] = (srcPtr2[j+2] + dither[2])>>(src_depth-8);
+                            dstPtr[j+3] = (srcPtr2[j+3] + dither[3])>>(src_depth-8);
+                            dstPtr[j+4] = (srcPtr2[j+4] + dither[4])>>(src_depth-8);
+                            dstPtr[j+5] = (srcPtr2[j+5] + dither[5])>>(src_depth-8);
+                            dstPtr[j+6] = (srcPtr2[j+6] + dither[6])>>(src_depth-8);
+                            dstPtr[j+7] = (srcPtr2[j+7] + dither[7])>>(src_depth-8);
+                        }
+                        for (; j < length; j++)
+                            dstPtr[j] = (srcPtr2[j] + dither[j&7])>>(src_depth-8);
+                        dstPtr  += dstStride[plane];
+                        srcPtr2 += srcStride[plane]/2;
+                    }
+                } else if (src_depth == 8) {
                     for (i = 0; i < height; i++) {
                         for (j = 0; j < length; j++)
-                            dstPtr2[j] = (srcPtr2[j]<<(16-depth)) | (srcPtr2[j]>>(2*depth-16));
+                            dstPtr2[j] = (srcPtr[j]<<(dst_depth-8)) |
+                                (srcPtr[j]>>(2*8-dst_depth));
+                        dstPtr2 += dstStride[plane]/2;
+                        srcPtr  += srcStride[plane];
+                    }
+                } else if (src_depth < dst_depth) {
+                    for (i = 0; i < height; i++) {
+                        for (j = 0; j < length; j++)
+                            dstPtr2[j] = (srcPtr2[j]<<(dst_depth-src_depth)) |
+                                (srcPtr2[j]>>(2*src_depth-dst_depth));
                         dstPtr2 += dstStride[plane]/2;
                         srcPtr2 += srcStride[plane]/2;
                     }
                 } else {
                     for (i = 0; i < height; i++) {
-                        uint8_t *dither= dithers[depth-9][i&7];
+                        uint8_t *dither= dithers[src_depth-9][i&7];
                         for (j = 0; j < length-7; j+=8){
-                            dstPtr[j+0] = (srcPtr2[j+0] + dither[0])>>(depth-8);
-                            dstPtr[j+1] = (srcPtr2[j+1] + dither[1])>>(depth-8);
-                            dstPtr[j+2] = (srcPtr2[j+2] + dither[2])>>(depth-8);
-                            dstPtr[j+3] = (srcPtr2[j+3] + dither[3])>>(depth-8);
-                            dstPtr[j+4] = (srcPtr2[j+4] + dither[4])>>(depth-8);
-                            dstPtr[j+5] = (srcPtr2[j+5] + dither[5])>>(depth-8);
-                            dstPtr[j+6] = (srcPtr2[j+6] + dither[6])>>(depth-8);
-                            dstPtr[j+7] = (srcPtr2[j+7] + dither[7])>>(depth-8);
+                            dstPtr2[j+0] = (srcPtr2[j+0] + dither[0])>>(src_depth-dst_depth);
+                            dstPtr2[j+1] = (srcPtr2[j+1] + dither[1])>>(src_depth-dst_depth);
+                            dstPtr2[j+2] = (srcPtr2[j+2] + dither[2])>>(src_depth-dst_depth);
+                            dstPtr2[j+3] = (srcPtr2[j+3] + dither[3])>>(src_depth-dst_depth);
+                            dstPtr2[j+4] = (srcPtr2[j+4] + dither[4])>>(src_depth-dst_depth);
+                            dstPtr2[j+5] = (srcPtr2[j+5] + dither[5])>>(src_depth-dst_depth);
+                            dstPtr2[j+6] = (srcPtr2[j+6] + dither[6])>>(src_depth-dst_depth);
+                            dstPtr2[j+7] = (srcPtr2[j+7] + dither[7])>>(src_depth-dst_depth);
                         }
                         for (; j < length; j++)
-                            dstPtr[j] = (srcPtr2[j] + dither[j&7])>>(depth-8);
-                        dstPtr  += dstStride[plane];
+                            dstPtr2[j] = (srcPtr2[j] + dither[j&7])>>(src_depth-dst_depth);
+                        dstPtr2 += dstStride[plane];
                         srcPtr2 += srcStride[plane]/2;
                     }
                 }

libswscale/swscale_internal.h

@@ -359,6 +359,8 @@ const char *sws_format_name(enum PixelFormat format);
         || (x)==PIX_FMT_YUV420P9BE  \
         || (x)==PIX_FMT_YUV420P10LE \
         || (x)==PIX_FMT_YUV420P10BE \
+        || (x)==PIX_FMT_YUV422P10LE \
+        || (x)==PIX_FMT_YUV422P10BE \
     )
 #define isBE(x) ((x)&1)
 #define isPlanar8YUV(x) (           \
@@ -377,11 +379,13 @@ const char *sws_format_name(enum PixelFormat format);
         || (x)==PIX_FMT_YUV420P9LE  \
         || (x)==PIX_FMT_YUV420P10LE \
         || (x)==PIX_FMT_YUV420P16LE \
+        || (x)==PIX_FMT_YUV422P10LE \
         || (x)==PIX_FMT_YUV422P16LE \
         || (x)==PIX_FMT_YUV444P16LE \
         || (x)==PIX_FMT_YUV420P9BE  \
         || (x)==PIX_FMT_YUV420P10BE \
         || (x)==PIX_FMT_YUV420P16BE \
+        || (x)==PIX_FMT_YUV422P10BE \
         || (x)==PIX_FMT_YUV422P16BE \
         || (x)==PIX_FMT_YUV444P16BE \
     )

libswscale/swscale_template.c

@@ -2817,7 +2817,7 @@ static int RENAME(swScale)(SwsContext *c, const uint8_t* src[], int srcStride[],
             } else if (isPlanarYUV(dstFormat) || dstFormat==PIX_FMT_GRAY8) { //YV12 like
                 const int chrSkipMask= (1<<c->chrDstVSubSample)-1;
                 if ((dstY&chrSkipMask) || isGray(dstFormat)) uDest=vDest= NULL; //FIXME split functions in lumi / chromi
-                if (is16BPS(dstFormat)) {
+                if (is16BPS(dstFormat) || isNBPS(dstFormat)) {
                     yuv2yuvX16inC(
                                   vLumFilter+dstY*vLumFilterSize   , lumSrcPtr, vLumFilterSize,
                                   vChrFilter+chrDstY*vChrFilterSize, chrSrcPtr, vChrFilterSize,
@@ -2894,7 +2894,7 @@ static int RENAME(swScale)(SwsContext *c, const uint8_t* src[], int srcStride[],
             } else if (isPlanarYUV(dstFormat) || dstFormat==PIX_FMT_GRAY8) { //YV12
                 const int chrSkipMask= (1<<c->chrDstVSubSample)-1;
                 if ((dstY&chrSkipMask) || isGray(dstFormat)) uDest=vDest= NULL; //FIXME split functions in lumi / chromi
-                if (is16BPS(dstFormat)) {
+                if (is16BPS(dstFormat) || isNBPS(dstFormat)) {
                     yuv2yuvX16inC(
                                   vLumFilter+dstY*vLumFilterSize   , lumSrcPtr, vLumFilterSize,
                                   vChrFilter+chrDstY*vChrFilterSize, chrSrcPtr, vChrFilterSize,
@@ -2979,6 +2979,7 @@ static void RENAME(sws_init_swScale)(SwsContext *c)
         case PIX_FMT_BGR4_BYTE:
         case PIX_FMT_RGB4_BYTE: c->chrToYV12 = palToUV; break;
         case PIX_FMT_YUV420P9 : c->chrToYV12 = (void*)RENAME(yuv9ToUV ); break;
+        case PIX_FMT_YUV422P10:
         case PIX_FMT_YUV420P10: c->chrToYV12 = (void*)RENAME(yuv10ToUV); break;
         case PIX_FMT_YUV420P16BE:
         case PIX_FMT_YUV422P16BE:
@@ -3027,6 +3028,7 @@ static void RENAME(sws_init_swScale)(SwsContext *c)
     c->alpToYV12 = NULL;
     switch (srcFormat) {
     case PIX_FMT_YUV420P9 : c->lumToYV12 = (void*)RENAME(yuv9ToY ); break;
+    case PIX_FMT_YUV422P10:
     case PIX_FMT_YUV420P10: c->lumToYV12 = (void*)RENAME(yuv10ToY); break;
     case PIX_FMT_YUYV422  :
     case PIX_FMT_YUV420P16BE:

libswscale/utils.c

@@ -115,6 +115,7 @@ const char *swscale_license(void)
         || (x)==PIX_FMT_YUV444P16BE   \
         || (x)==PIX_FMT_YUV420P9    \
         || (x)==PIX_FMT_YUV420P10   \
+        || (x)==PIX_FMT_YUV422P10   \
     )
 
 int sws_isSupportedInput(enum PixelFormat pix_fmt)
@@ -142,6 +143,9 @@ int sws_isSupportedInput(enum PixelFormat pix_fmt)
         || (x)==PIX_FMT_GRAY8       \
         || (x)==PIX_FMT_YUV410P     \
         || (x)==PIX_FMT_YUV440P     \
+        || (x)==PIX_FMT_YUV420P9    \
+        || (x)==PIX_FMT_YUV420P10   \
+        || (x)==PIX_FMT_YUV422P10   \
         || (x)==PIX_FMT_YUV420P16LE   \
         || (x)==PIX_FMT_YUV422P16LE   \
         || (x)==PIX_FMT_YUV444P16LE   \

tests/ref/lavfi/pixdesc_le

@@ -27,9 +27,12 @@ uyvy422             adcf64516a19fce44df77082bdb16291
 yuv410p             2d9225153c83ee1132397d619d94d1b3
 yuv411p             8b298af3e43348ca1b11eb8a3252ac6c
 yuv420p             eba2f135a08829387e2f698ff72a2939
+yuv420p10le         4228ee628c6deec123a13b9784516cc7
 yuv420p16be         16c009a235cd52b74791a895423152a3
 yuv420p16le         2d59c4f1d0314a5a957a7cfc4b6fabcc
+yuv420p9le          16543fda8f87d94a6cf857d2e8d4461a
 yuv422p             c9bba4529821d796a6ab09f6a5fd355a
+yuv422p10le         d0607c260a45c973e6639f4e449730ad
 yuv422p16be         5499502e1c29534a158a1fe60e889f60
 yuv422p16le         e3d61fde6978591596bc36b914386623
 yuv440p             5a064afe2b453bb52cdb3f176b1aa1cf

tests/ref/lavfi/pixfmts_copy_le

@@ -27,9 +27,12 @@ uyvy422             adcf64516a19fce44df77082bdb16291
 yuv410p             2d9225153c83ee1132397d619d94d1b3
 yuv411p             8b298af3e43348ca1b11eb8a3252ac6c
 yuv420p             eba2f135a08829387e2f698ff72a2939
+yuv420p10le         4228ee628c6deec123a13b9784516cc7
 yuv420p16be         16c009a235cd52b74791a895423152a3
 yuv420p16le         2d59c4f1d0314a5a957a7cfc4b6fabcc
+yuv420p9le          16543fda8f87d94a6cf857d2e8d4461a
 yuv422p             c9bba4529821d796a6ab09f6a5fd355a
+yuv422p10le         d0607c260a45c973e6639f4e449730ad
 yuv422p16be         5499502e1c29534a158a1fe60e889f60
 yuv422p16le         e3d61fde6978591596bc36b914386623
 yuv440p             5a064afe2b453bb52cdb3f176b1aa1cf

tests/ref/lavfi/pixfmts_null_le

@@ -27,9 +27,12 @@ uyvy422             adcf64516a19fce44df77082bdb16291
 yuv410p             2d9225153c83ee1132397d619d94d1b3
 yuv411p             8b298af3e43348ca1b11eb8a3252ac6c
 yuv420p             eba2f135a08829387e2f698ff72a2939
+yuv420p10le         4228ee628c6deec123a13b9784516cc7
 yuv420p16be         16c009a235cd52b74791a895423152a3
 yuv420p16le         2d59c4f1d0314a5a957a7cfc4b6fabcc
+yuv420p9le          16543fda8f87d94a6cf857d2e8d4461a
 yuv422p             c9bba4529821d796a6ab09f6a5fd355a
+yuv422p10le         d0607c260a45c973e6639f4e449730ad
 yuv422p16be         5499502e1c29534a158a1fe60e889f60
 yuv422p16le         e3d61fde6978591596bc36b914386623
 yuv440p             5a064afe2b453bb52cdb3f176b1aa1cf

tests/ref/lavfi/pixfmts_scale_le

@@ -27,9 +27,12 @@ uyvy422             314bd486277111a95d9369b944fa0400
 yuv410p             7df8f6d69b56a8dcb6c7ee908e5018b5
 yuv411p             1143e7c5cc28fe0922b051b17733bc4c
 yuv420p             fdad2d8df8985e3d17e73c71f713cb14
+yuv420p10le         d510ddbabefd03ef39ec943fcb51b709
 yuv420p16be         29a0265764530070f5cd3251cc01f66a
 yuv420p16le         6f3a265b084a78baec229238d9f7945f
+yuv420p9le          c136dce5913a722eee44ab72cff664b2
 yuv422p             918e37701ee7377d16a8a6c119c56a40
+yuv422p10le         aeb0ef08a883f43429ca9d886d8fc095
 yuv422p16be         ef3e865fc1d0c68977c735323c50af6e
 yuv422p16le         428a9b96214c09cb5a983ce36d6961ff
 yuv440p             461503fdb9b90451020aa3b25ddf041c

tests/ref/lavfi/pixfmts_vflip_le

@@ -27,9 +27,12 @@ uyvy422             ffbd36720c77398d9a0d03ce2625928f
 yuv410p             7bfb39d7afb49d6a6173e6b23ae321eb
 yuv411p             4a90048cc3a65fac150e53289700efe1
 yuv420p             2e6d6062e8cad37fb3ab2c433b55f382
+yuv420p10le         4b5249208602b941332945c926f80ae9
 yuv420p16be         539076782902664a8acf381bf4f713e8
 yuv420p16le         0f609e588e5a258644ef85170d70e030
+yuv420p9le          7e606310d3f5ff12badf911e8f333471
 yuv422p             d7f5cb44d9b0210d66d6a8762640ab34
+yuv422p10le         11b57f2bd9661024153f3973b9090cdb
 yuv422p16be         9bd8f8c961822b586fa4cf992be54acc
 yuv422p16le         9c4a1239605c7952b736ac3130163f14
 yuv440p             876385e96165acf51271b20e5d85a416