swscale: use 15-bit intermediates for 9/10-bit scaling.

8 files changed, 190 insertions, 88 deletions

diff --git a/libswscale/swscale.c b/libswscale/swscale.c
index dd9f4a108f..6fc65a8b2a 100644
--- a/libswscale/swscale.c
+++ b/libswscale/swscale.c
@@ -211,17 +211,9 @@ yuv2yuvX16_c_template(const int16_t *lumFilter, const int32_t **lumSrc,
 
 #define output_pixel(pos, val) \
     if (big_endian) { \
-        if (output_bits == 16) { \
-            AV_WB16(pos, av_clip_uint16(val >> shift)); \
-        } else { \
-            AV_WB16(pos, av_clip_uintp2(val >> shift, output_bits)); \
-        } \
+        AV_WB16(pos, av_clip_uint16(val >> shift)); \
     } else { \
-        if (output_bits == 16) { \
-            AV_WL16(pos, av_clip_uint16(val >> shift)); \
-        } else { \
-            AV_WL16(pos, av_clip_uintp2(val >> shift, output_bits)); \
-        } \
+        AV_WL16(pos, av_clip_uint16(val >> shift)); \
     }
     for (i = 0; i < dstW; i++) {
         int val = 1 << (30-output_bits - 1);
@@ -263,7 +255,67 @@ yuv2yuvX16_c_template(const int16_t *lumFilter, const int32_t **lumSrc,
 #undef output_pixel
 }
 
-#define yuv2NBPS(bits, BE_LE, is_be) \
+static av_always_inline void
+yuv2yuvX10_c_template(const int16_t *lumFilter, const int16_t **lumSrc,
+                      int lumFilterSize, const int16_t *chrFilter,
+                      const int16_t **chrUSrc, const int16_t **chrVSrc,
+                      int chrFilterSize, const int16_t **alpSrc,
+                      uint16_t *dest[4], int dstW, int chrDstW,
+                      int big_endian, int output_bits)
+{
+    //FIXME Optimize (just quickly written not optimized..)
+    int i;
+    uint16_t *yDest = dest[0], *uDest = dest[1], *vDest = dest[2],
+    *aDest = CONFIG_SWSCALE_ALPHA ? dest[3] : NULL;
+    int shift = 15 + 16 - output_bits - 1;
+
+#define output_pixel(pos, val) \
+    if (big_endian) { \
+        AV_WB16(pos, av_clip_uintp2(val >> shift, output_bits)); \
+    } else { \
+        AV_WL16(pos, av_clip_uintp2(val >> shift, output_bits)); \
+    }
+    for (i = 0; i < dstW; i++) {
+        int val = 1 << (30-output_bits - 1);
+        int j;
+
+        for (j = 0; j < lumFilterSize; j++)
+            val += (lumSrc[j][i] * lumFilter[j]) >> 1;
+
+        output_pixel(&yDest[i], val);
+    }
+
+    if (uDest) {
+        for (i = 0; i < chrDstW; i++) {
+            int u = 1 << (30-output_bits - 1);
+            int v = 1 << (30-output_bits - 1);
+            int j;
+
+            for (j = 0; j < chrFilterSize; j++) {
+                u += (chrUSrc[j][i] * chrFilter[j]) >> 1;
+                v += (chrVSrc[j][i] * chrFilter[j]) >> 1;
+            }
+
+            output_pixel(&uDest[i], u);
+            output_pixel(&vDest[i], v);
+        }
+    }
+
+    if (CONFIG_SWSCALE_ALPHA && aDest) {
+        for (i = 0; i < dstW; i++) {
+            int val = 1 << (30-output_bits - 1);
+            int j;
+
+            for (j = 0; j < lumFilterSize; j++)
+                val += (alpSrc[j][i] * lumFilter[j]) >> 1;
+
+            output_pixel(&aDest[i], val);
+        }
+    }
+#undef output_pixel
+}
+
+#define yuv2NBPS(bits, BE_LE, is_be, yuv2yuvX_template_fn, typeX_t) \
 static void yuv2yuvX ## bits ## BE_LE ## _c(SwsContext *c, const int16_t *lumFilter, \
                               const int16_t **_lumSrc, int lumFilterSize, \
                               const int16_t *chrFilter, const int16_t **_chrUSrc, \
@@ -271,21 +323,21 @@ static void yuv2yuvX ## bits ## BE_LE ## _c(SwsContext *c, const int16_t *lumFil
                               int chrFilterSize, const int16_t **_alpSrc, \
                               uint8_t *_dest[4], int dstW, int chrDstW) \
 { \
-    const int32_t **lumSrc  = (const int32_t **) _lumSrc, \
-                  **chrUSrc = (const int32_t **) _chrUSrc, \
-                  **chrVSrc = (const int32_t **) _chrVSrc, \
-                  **alpSrc  = (const int32_t **) _alpSrc; \
-    yuv2yuvX16_c_template(lumFilter, lumSrc, lumFilterSize, \
-                          chrFilter, chrUSrc, chrVSrc, chrFilterSize, \
-                          alpSrc, (uint16_t **) _dest, \
-                          dstW, chrDstW, is_be, bits); \
-}
-yuv2NBPS( 9, BE, 1);
-yuv2NBPS( 9, LE, 0);
-yuv2NBPS(10, BE, 1);
-yuv2NBPS(10, LE, 0);
-yuv2NBPS(16, BE, 1);
-yuv2NBPS(16, LE, 0);
+    const typeX_t **lumSrc  = (const typeX_t **) _lumSrc, \
+                  **chrUSrc = (const typeX_t **) _chrUSrc, \
+                  **chrVSrc = (const typeX_t **) _chrVSrc, \
+                  **alpSrc  = (const typeX_t **) _alpSrc; \
+    yuv2yuvX_template_fn(lumFilter, lumSrc, lumFilterSize, \
+                         chrFilter, chrUSrc, chrVSrc, chrFilterSize, \
+                         alpSrc, (uint16_t **) _dest, \
+                         dstW, chrDstW, is_be, bits); \
+}
+yuv2NBPS( 9, BE, 1, yuv2yuvX10_c_template, int16_t);
+yuv2NBPS( 9, LE, 0, yuv2yuvX10_c_template, int16_t);
+yuv2NBPS(10, BE, 1, yuv2yuvX10_c_template, int16_t);
+yuv2NBPS(10, LE, 0, yuv2yuvX10_c_template, int16_t);
+yuv2NBPS(16, BE, 1, yuv2yuvX16_c_template, int32_t);
+yuv2NBPS(16, LE, 0, yuv2yuvX16_c_template, int32_t);
 
 static void yuv2yuvX_c(SwsContext *c, const int16_t *lumFilter,
                        const int16_t **lumSrc, int lumFilterSize,
@@ -1880,6 +1932,27 @@ static void hScale16_c(SwsContext *c, int16_t *_dst, int dstW, const uint8_t *_s
     }
 }
 
+static void hScale10_c(SwsContext *c, int16_t *dst, int dstW, const uint8_t *_src,
+                       const int16_t *filter,
+                       const int16_t *filterPos, int filterSize)
+{
+    int i;
+    const uint16_t *src = (const uint16_t *) _src;
+    int sh = av_pix_fmt_descriptors[c->srcFormat].comp[0].depth_minus1;
+
+    for (i = 0; i < dstW; i++) {
+        int j;
+        int srcPos = filterPos[i];
+        int val = 0;
+
+        for (j = 0; j < filterSize; j++) {
+            val += src[srcPos + j] * filter[filterSize * i + j];
+        }
+        // filter=14 bit, input=16 bit, output=30 bit, >> 11 makes 19 bit
+        dst[i] = FFMIN(val >> sh, (1 << 15) - 1);
+    }
+}
+
 // bilinear / bicubic scaling
 static void hScale_c(SwsContext *c, int16_t *dst, int dstW, const uint8_t *src,
                      const int16_t *filter, const int16_t *filterPos,
@@ -2025,7 +2098,7 @@ static av_always_inline void hyscale(SwsContext *c, int16_t *dst, int dstWidth,
     if (convertRange)
         convertRange(dst, dstWidth);
 
-    if (av_pix_fmt_descriptors[c->dstFormat].comp[0].depth_minus1 < 8 && c->scalingBpp == 16) {
+    if (av_pix_fmt_descriptors[c->dstFormat].comp[0].depth_minus1 < 10 && c->scalingBpp == 16) {
         c->scale19To15Fw(dst, (int32_t *) dst, dstWidth);
     }
 }
@@ -2052,7 +2125,7 @@ static av_always_inline void hcscale(SwsContext *c, int16_t *dst1, int16_t *dst2
                                      uint8_t *formatConvBuffer, uint32_t *pal)
 {
     if (c->chrToYV12) {
-        uint8_t *buf2 = formatConvBuffer + FFALIGN(srcW * c->scalingBpp >> 3, 16);
+        uint8_t *buf2 = formatConvBuffer + FFALIGN(srcW * FFALIGN(c->scalingBpp, 8) >> 3, 16);
         c->chrToYV12(formatConvBuffer, buf2, src1, src2, srcW, pal);
         src1= formatConvBuffer;
         src2= buf2;
@@ -2076,7 +2149,7 @@ static av_always_inline void hcscale(SwsContext *c, int16_t *dst1, int16_t *dst2
     if (c->chrConvertRange)
         c->chrConvertRange(dst1, dst2, dstWidth);
 
-    if (av_pix_fmt_descriptors[c->dstFormat].comp[0].depth_minus1 < 8 && c->scalingBpp == 16) {
+    if (av_pix_fmt_descriptors[c->dstFormat].comp[0].depth_minus1 < 10 && c->scalingBpp == 16) {
         c->scale19To15Fw(dst1, (int32_t *) dst1, dstWidth);
         c->scale19To15Fw(dst2, (int32_t *) dst2, dstWidth);
     }
@@ -2735,27 +2808,27 @@ static av_cold void sws_init_swScale_c(SwsContext *c)
     }
 
     if (c->scalingBpp == 8) {
-    c->hScale       = hScale_c;
-    if (c->flags & SWS_FAST_BILINEAR) {
-        c->hyscale_fast = hyscale_fast_c;
-        c->hcscale_fast = hcscale_fast_c;
-    }
-
-    if (c->srcRange != c->dstRange && !isAnyRGB(c->dstFormat)) {
-        if (c->srcRange) {
-            c->lumConvertRange = lumRangeFromJpeg_c;
-            c->chrConvertRange = chrRangeFromJpeg_c;
-        } else {
-            c->lumConvertRange = lumRangeToJpeg_c;
-            c->chrConvertRange = chrRangeToJpeg_c;
+        c->hScale       = hScale_c;
+        if (c->flags & SWS_FAST_BILINEAR) {
+            c->hyscale_fast = hyscale_fast_c;
+            c->hcscale_fast = hcscale_fast_c;
         }
-    }
     } else {
-        c->hScale = hScale16_c;
+        c->hScale = c->scalingBpp == 16 ? hScale16_c : hScale10_c;
         c->scale19To15Fw = scale19To15Fw_c;
         c->scale8To16Rv  = scale8To16Rv_c;
+    }
 
-        if (c->srcRange != c->dstRange && !isAnyRGB(c->dstFormat)) {
+    if (c->srcRange != c->dstRange && !isAnyRGB(c->dstFormat)) {
+        if (c->scalingBpp <= 10) {
+            if (c->srcRange) {
+                c->lumConvertRange = lumRangeFromJpeg_c;
+                c->chrConvertRange = chrRangeFromJpeg_c;
+            } else {
+                c->lumConvertRange = lumRangeToJpeg_c;
+                c->chrConvertRange = chrRangeToJpeg_c;
+            }
+        } else {
             if (c->srcRange) {
                 c->lumConvertRange = lumRangeFromJpeg16_c;
                 c->chrConvertRange = chrRangeFromJpeg16_c;
diff --git a/libswscale/swscale_internal.h b/libswscale/swscale_internal.h
index 9492303301..60787aed91 100644
--- a/libswscale/swscale_internal.h
+++ b/libswscale/swscale_internal.h
@@ -64,11 +64,16 @@ typedef int (*SwsFunc)(struct SwsContext *context, const uint8_t* src[],
  * without any additional vertical scaling (or point-scaling).
  *
  * @param c       SWS scaling context
- * @param lumSrc  scaled luma (Y) source data, 15bit for 8bit output
- * @param chrUSrc scaled chroma (U) source data, 15bit for 8bit output
- * @param chrVSrc scaled chroma (V) source data, 15bit for 8bit output
- * @param alpSrc  scaled alpha (A) source data, 15bit for 8bit output
- * @param dest    pointer to the 4 output planes (Y/U/V/A)
+ * @param lumSrc  scaled luma (Y) source data, 15bit for 8-10bit output,
+ *                19-bit for 16bit output (in int32_t)
+ * @param chrUSrc scaled chroma (U) source data, 15bit for 8-10bit output,
+ *                19-bit for 16bit output (in int32_t)
+ * @param chrVSrc scaled chroma (V) source data, 15bit for 8-10bit output,
+ *                19-bit for 16bit output (in int32_t)
+ * @param alpSrc  scaled alpha (A) source data, 15bit for 8-10bit output,
+ *                19-bit for 16bit output (in int32_t)
+ * @param dest    pointer to the 4 output planes (Y/U/V/A). For >8bit
+ *                output, this is in uint16_t
  * @param dstW    width of dest[0], dest[3], lumSrc and alpSrc in pixels
  * @param chrDstW width of dest[1], dest[2], chrUSrc and chrVSrc
  */
@@ -82,14 +87,19 @@ typedef void (*yuv2planar1_fn) (struct SwsContext *c,
  *
  * @param c             SWS scaling context
  * @param lumFilter     vertical luma/alpha scaling coefficients, 12bit [0,4096]
- * @param lumSrc        scaled luma (Y) source data, 15bit for 8bit output
+ * @param lumSrc        scaled luma (Y) source data, 15bit for 8-10bit output,
+ *                      19-bit for 16bit output (in int32_t)
  * @param lumFilterSize number of vertical luma/alpha input lines to scale
  * @param chrFilter     vertical chroma scaling coefficients, 12bit [0,4096]
- * @param chrUSrc       scaled chroma (U) source data, 15bit for 8bit output
- * @param chrVSrc       scaled chroma (V) source data, 15bit for 8bit output
+ * @param chrUSrc       scaled chroma (U) source data, 15bit for 8-10bit output,
+ *                      19-bit for 16bit output (in int32_t)
+ * @param chrVSrc       scaled chroma (V) source data, 15bit for 8-10bit output,
+ *                      19-bit for 16bit output (in int32_t)
  * @param chrFilterSize number of vertical chroma input lines to scale
- * @param alpSrc        scaled alpha (A) source data, 15bit for 8bit output
- * @param dest          pointer to the 4 output planes (Y/U/V/A)
+ * @param alpSrc        scaled alpha (A) source data, 15bit for 8-10bit output,
+ *                      19-bit for 16bit output (in int32_t)
+ * @param dest          pointer to the 4 output planes (Y/U/V/A). For >8bit
+ *                      output, this is in uint16_t
  * @param dstW          width of dest[0], dest[3], lumSrc and alpSrc in pixels
  * @param chrDstW       width of dest[1], dest[2], chrUSrc and chrVSrc
  */
@@ -105,11 +115,16 @@ typedef void (*yuv2planarX_fn) (struct SwsContext *c, const int16_t *lumFilter,
  * that this function may do chroma scaling, see the "uvalpha" argument.
  *
  * @param c       SWS scaling context
- * @param lumSrc  scaled luma (Y) source data, 15bit for 8bit output
- * @param chrUSrc scaled chroma (U) source data, 15bit for 8bit output
- * @param chrVSrc scaled chroma (V) source data, 15bit for 8bit output
- * @param alpSrc  scaled alpha (A) source data, 15bit for 8bit output
- * @param dest    pointer to the output plane
+ * @param lumSrc  scaled luma (Y) source data, 15bit for 8-10bit output,
+ *                19-bit for 16bit output (in int32_t)
+ * @param chrUSrc scaled chroma (U) source data, 15bit for 8-10bit output,
+ *                19-bit for 16bit output (in int32_t)
+ * @param chrVSrc scaled chroma (V) source data, 15bit for 8-10bit output,
+ *                19-bit for 16bit output (in int32_t)
+ * @param alpSrc  scaled alpha (A) source data, 15bit for 8-10bit output,
+ *                19-bit for 16bit output (in int32_t)
+ * @param dest    pointer to the output plane. For 16bit output, this is
+ *                uint16_t
  * @param dstW    width of lumSrc and alpSrc in pixels, number of pixels
  *                to write into dest[]
  * @param uvalpha chroma scaling coefficient for the second line of chroma
@@ -132,11 +147,16 @@ typedef void (*yuv2packed1_fn) (struct SwsContext *c,  const int16_t *lumSrc,
  * output by doing bilinear scaling between two input lines.
  *
  * @param c       SWS scaling context
- * @param lumSrc  scaled luma (Y) source data, 15bit for 8bit output
- * @param chrUSrc scaled chroma (U) source data, 15bit for 8bit output
- * @param chrVSrc scaled chroma (V) source data, 15bit for 8bit output
- * @param alpSrc  scaled alpha (A) source data, 15bit for 8bit output
- * @param dest    pointer to the output plane
+ * @param lumSrc  scaled luma (Y) source data, 15bit for 8-10bit output,
+ *                19-bit for 16bit output (in int32_t)
+ * @param chrUSrc scaled chroma (U) source data, 15bit for 8-10bit output,
+ *                19-bit for 16bit output (in int32_t)
+ * @param chrVSrc scaled chroma (V) source data, 15bit for 8-10bit output,
+ *                19-bit for 16bit output (in int32_t)
+ * @param alpSrc  scaled alpha (A) source data, 15bit for 8-10bit output,
+ *                19-bit for 16bit output (in int32_t)
+ * @param dest    pointer to the output plane. For 16bit output, this is
+ *                uint16_t
  * @param dstW    width of lumSrc and alpSrc in pixels, number of pixels
  *                to write into dest[]
  * @param yalpha  luma/alpha scaling coefficients for the second input line.
@@ -160,14 +180,19 @@ typedef void (*yuv2packed2_fn) (struct SwsContext *c,  const int16_t *lumSrc[2],
  *
  * @param c             SWS scaling context
  * @param lumFilter     vertical luma/alpha scaling coefficients, 12bit [0,4096]
- * @param lumSrc        scaled luma (Y) source data, 15bit for 8bit output
+ * @param lumSrc        scaled luma (Y) source data, 15bit for 8-10bit output,
+ *                      19-bit for 16bit output (in int32_t)
  * @param lumFilterSize number of vertical luma/alpha input lines to scale
  * @param chrFilter     vertical chroma scaling coefficients, 12bit [0,4096]
- * @param chrUSrc       scaled chroma (U) source data, 15bit for 8bit output
- * @param chrVSrc       scaled chroma (V) source data, 15bit for 8bit output
+ * @param chrUSrc       scaled chroma (U) source data, 15bit for 8-10bit output,
+ *                      19-bit for 16bit output (in int32_t)
+ * @param chrVSrc       scaled chroma (V) source data, 15bit for 8-10bit output,
+ *                      19-bit for 16bit output (in int32_t)
  * @param chrFilterSize number of vertical chroma input lines to scale
- * @param alpSrc        scaled alpha (A) source data, 15bit for 8bit output
- * @param dest          pointer to the output plane
+ * @param alpSrc        scaled alpha (A) source data, 15bit for 8-10bit output,
+ *                      19-bit for 16bit output (in int32_t)
+ * @param dest          pointer to the output plane. For 16bit output, this is
+ *                      uint16_t
  * @param dstW          width of lumSrc and alpSrc in pixels, number of pixels
  *                      to write into dest[]
  * @param y             vertical line number for this output. This does not need
diff --git a/libswscale/utils.c b/libswscale/utils.c
index c6abb6b446..18d1227c83 100644
--- a/libswscale/utils.c
+++ b/libswscale/utils.c
@@ -896,11 +896,15 @@ int sws_init_context(SwsContext *c, SwsFilter *srcFilter, SwsFilter *dstFilter)
         }
     }
 
-    c->scalingBpp = FFMAX(av_pix_fmt_descriptors[srcFormat].comp[0].depth_minus1,
-                          av_pix_fmt_descriptors[dstFormat].comp[0].depth_minus1) >= 8 ? 16 : 8;
+    c->scalingBpp = 1 + FFMAX(av_pix_fmt_descriptors[srcFormat].comp[0].depth_minus1,
+                             av_pix_fmt_descriptors[dstFormat].comp[0].depth_minus1);
+    if (c->scalingBpp <= 8)
+        c->scalingBpp = 8;
     if (c->scalingBpp == 16)
         dst_stride <<= 1;
-    FF_ALLOC_OR_GOTO(c, c->formatConvBuffer, FFALIGN(srcW, 16) * 2 * c->scalingBpp >> 3, fail);
+    FF_ALLOC_OR_GOTO(c, c->formatConvBuffer,
+                     FFALIGN(srcW, 16) * 2 * FFALIGN(c->scalingBpp, 8) >> 3,
+                     fail);
     if (HAVE_MMX2 && cpu_flags & AV_CPU_FLAG_MMX2 && c->scalingBpp == 8) {
         c->canMMX2BeUsed= (dstW >=srcW && (dstW&31)==0 && (srcW&15)==0) ? 1 : 0;
         if (!c->canMMX2BeUsed && dstW >=srcW && (srcW&15)==0 && (flags&SWS_FAST_BILINEAR)) {
@@ -1055,7 +1059,7 @@ int sws_init_context(SwsContext *c, SwsFilter *srcFilter, SwsFilter *dstFilter)
         c->lumPixBuf[i] = c->lumPixBuf[i+c->vLumBufSize];
     }
     // 64 / c->scalingBpp is the same as 16 / sizeof(scaling_intermediate)
-    c->uv_off_px   = dst_stride_px + 64 / c->scalingBpp;
+    c->uv_off_px   = dst_stride_px + 64 / (c->scalingBpp &~ 7);
     c->uv_off_byte = dst_stride + 16;
     for (i=0; i<c->vChrBufSize; i++) {
         FF_ALLOC_OR_GOTO(c, c->chrUPixBuf[i+c->vChrBufSize], dst_stride*2+32, fail);
diff --git a/tests/ref/lavfi/pixdesc b/tests/ref/lavfi/pixdesc
index 3730988777..0da94f861e 100644
--- a/tests/ref/lavfi/pixdesc
+++ b/tests/ref/lavfi/pixdesc
@@ -38,8 +38,8 @@ yuv420p16le         2d59c4f1d0314a5a957a7cfc4b6fabcc
 yuv420p9be          ce880fa07830e5297c22acf6e20555ce
 yuv420p9le          16543fda8f87d94a6cf857d2e8d4461a
 yuv422p             c9bba4529821d796a6ab09f6a5fd355a
-yuv422p10be         107c6e31a3d4d598bca1d8426aaa54f5
-yuv422p10le         3f478be644add24b6cc77e718a6e2afa
+yuv422p10be         a4a83d0811280eff7405d94a7de21596
+yuv422p10le         23717b6c73a59912c605f27877ae2fb6
 yuv422p16be         dc9886f2fccf87cc54b27e071a2c251e
 yuv422p16le         f181c8d8436f1233ba566d9bc88005ec
 yuv440p             5a064afe2b453bb52cdb3f176b1aa1cf
diff --git a/tests/ref/lavfi/pixfmts_copy b/tests/ref/lavfi/pixfmts_copy
index 3730988777..0da94f861e 100644
--- a/tests/ref/lavfi/pixfmts_copy
+++ b/tests/ref/lavfi/pixfmts_copy
@@ -38,8 +38,8 @@ yuv420p16le         2d59c4f1d0314a5a957a7cfc4b6fabcc
 yuv420p9be          ce880fa07830e5297c22acf6e20555ce
 yuv420p9le          16543fda8f87d94a6cf857d2e8d4461a
 yuv422p             c9bba4529821d796a6ab09f6a5fd355a
-yuv422p10be         107c6e31a3d4d598bca1d8426aaa54f5
-yuv422p10le         3f478be644add24b6cc77e718a6e2afa
+yuv422p10be         a4a83d0811280eff7405d94a7de21596
+yuv422p10le         23717b6c73a59912c605f27877ae2fb6
 yuv422p16be         dc9886f2fccf87cc54b27e071a2c251e
 yuv422p16le         f181c8d8436f1233ba566d9bc88005ec
 yuv440p             5a064afe2b453bb52cdb3f176b1aa1cf
diff --git a/tests/ref/lavfi/pixfmts_null b/tests/ref/lavfi/pixfmts_null
index 3730988777..0da94f861e 100644
--- a/tests/ref/lavfi/pixfmts_null
+++ b/tests/ref/lavfi/pixfmts_null
@@ -38,8 +38,8 @@ yuv420p16le         2d59c4f1d0314a5a957a7cfc4b6fabcc
 yuv420p9be          ce880fa07830e5297c22acf6e20555ce
 yuv420p9le          16543fda8f87d94a6cf857d2e8d4461a
 yuv422p             c9bba4529821d796a6ab09f6a5fd355a
-yuv422p10be         107c6e31a3d4d598bca1d8426aaa54f5
-yuv422p10le         3f478be644add24b6cc77e718a6e2afa
+yuv422p10be         a4a83d0811280eff7405d94a7de21596
+yuv422p10le         23717b6c73a59912c605f27877ae2fb6
 yuv422p16be         dc9886f2fccf87cc54b27e071a2c251e
 yuv422p16le         f181c8d8436f1233ba566d9bc88005ec
 yuv440p             5a064afe2b453bb52cdb3f176b1aa1cf
diff --git a/tests/ref/lavfi/pixfmts_scale b/tests/ref/lavfi/pixfmts_scale
index 392b9ce601..d68cdd41d0 100644
--- a/tests/ref/lavfi/pixfmts_scale
+++ b/tests/ref/lavfi/pixfmts_scale
@@ -31,15 +31,15 @@ uyvy422             314bd486277111a95d9369b944fa0400
 yuv410p             7df8f6d69b56a8dcb6c7ee908e5018b5
 yuv411p             1143e7c5cc28fe0922b051b17733bc4c
 yuv420p             fdad2d8df8985e3d17e73c71f713cb14
-yuv420p10be         af5429f27b9f95bf955e795921c65cdc
-yuv420p10le         d0b47e6a8a44e6b5ca0fe4349a4e393b
+yuv420p10be         04663d400e44692fe8a622a067f838da
+yuv420p10le         6171850f66df7a727b4bed1c87ef9188
 yuv420p16be         9688e33e03b8c8275ab2fb1df0f06bee
 yuv420p16le         cba8b390ad5e7b8678e419b8ce79c008
-yuv420p9be          a073b2d93b2a7dce2069ba252bc43175
-yuv420p9le          b67233c3c7d93763d07d88f697c145e1
+yuv420p9be          ab163dfef03c4d563aca99b24276b9fd
+yuv420p9le          cd56c5a76ce74e504dd59d25a5e4389c
 yuv422p             918e37701ee7377d16a8a6c119c56a40
-yuv422p10be         533fd21e7943c20a1026b19069b3b867
-yuv422p10le         59b20a4a8609f5da2dc54c78aea11e6c
+yuv422p10be         e8e80fed7121f3afac994f2afac42cd2
+yuv422p10le         370866666f4889ee0928345b16d68fb4
 yuv422p16be         2cf502d7d386db1f1b3b946679d897b1
 yuv422p16le         3002a4e47520731dcee5929aff49eb74
 yuv440p             461503fdb9b90451020aa3b25ddf041c
diff --git a/tests/ref/lavfi/pixfmts_vflip b/tests/ref/lavfi/pixfmts_vflip
index 2b62518a28..a2b11690ab 100644
--- a/tests/ref/lavfi/pixfmts_vflip
+++ b/tests/ref/lavfi/pixfmts_vflip
@@ -38,8 +38,8 @@ yuv420p16le         0f609e588e5a258644ef85170d70e030
 yuv420p9be          be40ec975fb2873891643cbbbddbc3b0
 yuv420p9le          7e606310d3f5ff12badf911e8f333471
 yuv422p             d7f5cb44d9b0210d66d6a8762640ab34
-yuv422p10be         a28b051168af49435c04af5f58dce47b
-yuv422p10le         35936ffff30df2697f47b9b8d2cb7dea
+yuv422p10be         51d9ef13fe43ea9549b3792bfd449bf7
+yuv422p10le         4b286a243ee0e715b2961bc1b469e629
 yuv422p16be         51d9aa4e78d121c226d919ce97976fe4
 yuv422p16le         12965c54bda8932ca72da194419a9908
 yuv440p             876385e96165acf51271b20e5d85a416