swscale: dont loose precission on RGB/BGR48 input, that is dont drop half the bits.

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

3 files changed, 63 insertions, 107 deletions

diff --git a/libswscale/swscale.c b/libswscale/swscale.c
index 84926635c3..00941c9dfc 100644
--- a/libswscale/swscale.c
+++ b/libswscale/swscale.c
@@ -1045,93 +1045,55 @@ static void fillPlane(uint8_t* plane, int stride, int width, int height, int y,
     }
 }
 
-static inline void rgb48ToY(int16_t *dst, const uint8_t *src, long width,
-                            uint32_t *unused)
-{
-    int i;
-    for (i = 0; i < width; i++) {
-        int r = src[i*6+0];
-        int g = src[i*6+2];
-        int b = src[i*6+4];
-
-        dst[i] = (RY*r + GY*g + BY*b + (32<<(RGB2YUV_SHIFT-1)) + (1<<(RGB2YUV_SHIFT-7))) >> (RGB2YUV_SHIFT-6);
-    }
-}
-
-static inline void rgb48ToUV(int16_t *dstU, int16_t *dstV,
-                             const uint8_t *src1, const uint8_t *src2,
-                             long width, uint32_t *unused)
-{
-    int i;
-    assert(src1==src2);
-    for (i = 0; i < width; i++) {
-        int r = src1[6*i + 0];
-        int g = src1[6*i + 2];
-        int b = src1[6*i + 4];
-
-        dstU[i] = (RU*r + GU*g + BU*b + (256<<(RGB2YUV_SHIFT-1)) + (1<<(RGB2YUV_SHIFT-7))) >> (RGB2YUV_SHIFT-6);
-        dstV[i] = (RV*r + GV*g + BV*b + (256<<(RGB2YUV_SHIFT-1)) + (1<<(RGB2YUV_SHIFT-7))) >> (RGB2YUV_SHIFT-6);
-    }
-}
-
-static inline void rgb48ToUV_half(int16_t *dstU, int16_t *dstV,
-                                  const uint8_t *src1, const uint8_t *src2,
-                                  long width, uint32_t *unused)
-{
-    int i;
-    assert(src1==src2);
-    for (i = 0; i < width; i++) {
-        int r= src1[12*i + 0] + src1[12*i + 6];
-        int g= src1[12*i + 2] + src1[12*i + 8];
-        int b= src1[12*i + 4] + src1[12*i + 10];
-
-        dstU[i]= (RU*r + GU*g + BU*b + (256<<(RGB2YUV_SHIFT)) + (1<<(RGB2YUV_SHIFT-6))) >> (RGB2YUV_SHIFT-5);
-        dstV[i]= (RV*r + GV*g + BV*b + (256<<(RGB2YUV_SHIFT)) + (1<<(RGB2YUV_SHIFT-6))) >> (RGB2YUV_SHIFT-5);
-    }
-}
-
-static inline void bgr48ToY(int16_t *dst, const uint8_t *src, long width,
-                            uint32_t *unused)
-{
-    int i;
-    for (i = 0; i < width; i++) {
-        int b = src[i*6+0];
-        int g = src[i*6+2];
-        int r = src[i*6+4];
-
-        dst[i] = (RY*r + GY*g + BY*b + (32<<(RGB2YUV_SHIFT-1)) + (1<<(RGB2YUV_SHIFT-7))) >> (RGB2YUV_SHIFT-6);
-    }
-}
-
-static inline void bgr48ToUV(int16_t *dstU, int16_t *dstV,
-                             const uint8_t *src1, const uint8_t *src2,
-                             long width, uint32_t *unused)
-{
-    int i;
-    for (i = 0; i < width; i++) {
-        int b = src1[6*i + 0];
-        int g = src1[6*i + 2];
-        int r = src1[6*i + 4];
-
-        dstU[i] = (RU*r + GU*g + BU*b + (256<<(RGB2YUV_SHIFT-1)) + (1<<(RGB2YUV_SHIFT-7))) >> (RGB2YUV_SHIFT-6);
-        dstV[i] = (RV*r + GV*g + BV*b + (256<<(RGB2YUV_SHIFT-1)) + (1<<(RGB2YUV_SHIFT-7))) >> (RGB2YUV_SHIFT-6);
-    }
+#define RGB48(name, R, B, READ)\
+static inline void name ## ToY(int16_t *dst, const uint16_t *src, long width, uint32_t *unused)\
+{\
+    int i;\
+    for (i = 0; i < width; i++) {\
+        int r = READ(&src[i*3+R]);\
+        int g = READ(&src[i*3+1]);\
+        int b = READ(&src[i*3+B]);\
+\
+        dst[i] = (RY*r + GY*g + BY*b + (32<<(RGB2YUV_SHIFT-1+8)) + (1<<(RGB2YUV_SHIFT-7+8))) >> (RGB2YUV_SHIFT-6+8);\
+    }\
+}\
+\
+static inline void name ## ToUV(int16_t *dstU, int16_t *dstV,\
+                             const uint16_t *src1, const uint16_t *src2,\
+                             long width, uint32_t *unused)\
+{\
+    int i;\
+    assert(src1==src2);\
+    for (i = 0; i < width; i++) {\
+        int r = READ(&src1[3*i + R]);\
+        int g = READ(&src1[3*i + 1]);\
+        int b = READ(&src1[3*i + B]);\
+\
+        dstU[i] = (RU*r + GU*g + BU*b + (256<<(RGB2YUV_SHIFT-1+8)) + (1<<(RGB2YUV_SHIFT-7+8))) >> (RGB2YUV_SHIFT-6+8);\
+        dstV[i] = (RV*r + GV*g + BV*b + (256<<(RGB2YUV_SHIFT-1+8)) + (1<<(RGB2YUV_SHIFT-7+8))) >> (RGB2YUV_SHIFT-6+8);\
+    }\
+}\
+\
+static inline void name ## ToUV_half(int16_t *dstU, int16_t *dstV,\
+                                  const uint16_t *src1, const uint16_t *src2,\
+                                  long width, uint32_t *unused)\
+{\
+    int i;\
+    assert(src1==src2);\
+    for (i = 0; i < width; i++) {\
+        int r= READ(&src1[6*i + R]) + READ(&src1[6*i + 3+R]);\
+        int g= READ(&src1[6*i + 1]) + READ(&src1[6*i + 4]);\
+        int b= READ(&src1[6*i + B]) + READ(&src1[6*i + 3+B]);\
+\
+        dstU[i]= (RU*r + GU*g + BU*b + (256U<<(RGB2YUV_SHIFT+8)) + (1<<(RGB2YUV_SHIFT-6+8))) >> (RGB2YUV_SHIFT-5+8);\
+        dstV[i]= (RV*r + GV*g + BV*b + (256U<<(RGB2YUV_SHIFT+8)) + (1<<(RGB2YUV_SHIFT-6+8))) >> (RGB2YUV_SHIFT-5+8);\
+    }\
 }
 
-static inline void bgr48ToUV_half(int16_t *dstU, int16_t *dstV,
-                                  const uint8_t *src1, const uint8_t *src2,
-                                  long width, uint32_t *unused)
-{
-    int i;
-    for (i = 0; i < width; i++) {
-        int b= src1[12*i + 0] + src1[12*i + 6];
-        int g= src1[12*i + 2] + src1[12*i + 8];
-        int r= src1[12*i + 4] + src1[12*i + 10];
-
-        dstU[i]= (RU*r + GU*g + BU*b + (256<<(RGB2YUV_SHIFT)) + (1<<(RGB2YUV_SHIFT-6))) >> (RGB2YUV_SHIFT-5);
-        dstV[i]= (RV*r + GV*g + BV*b + (256<<(RGB2YUV_SHIFT)) + (1<<(RGB2YUV_SHIFT-6))) >> (RGB2YUV_SHIFT-5);
-    }
-}
+RGB48(rgb48LE, 0, 2, AV_RL16)
+RGB48(rgb48BE, 0, 2, AV_RB16)
+RGB48(bgr48LE, 2, 0, AV_RL16)
+RGB48(bgr48BE, 2, 0, AV_RB16)
 
 #define BGR2Y(type, name, shr, shg, shb, maskr, maskg, maskb, RY, GY, BY, S)\
 static inline void name(int16_t *dst, const uint8_t *src, long width, uint32_t *unused)\
diff --git a/libswscale/swscale_template.c b/libswscale/swscale_template.c
index 8bf38b5946..5a84ceb871 100644
--- a/libswscale/swscale_template.c
+++ b/libswscale/swscale_template.c
@@ -886,10 +886,10 @@ static void sws_init_swScale_c(SwsContext *c)
     }
     if (c->chrSrcHSubSample) {
         switch(srcFormat) {
-        case PIX_FMT_RGB48BE:
-        case PIX_FMT_RGB48LE: c->chrToYV12 = rgb48ToUV_half; break;
-        case PIX_FMT_BGR48BE:
-        case PIX_FMT_BGR48LE: c->chrToYV12 = bgr48ToUV_half; break;
+        case PIX_FMT_RGB48BE: c->chrToYV12 = rgb48BEToUV_half; break;
+        case PIX_FMT_RGB48LE: c->chrToYV12 = rgb48LEToUV_half; break;
+        case PIX_FMT_BGR48BE: c->chrToYV12 = bgr48BEToUV_half; break;
+        case PIX_FMT_BGR48LE: c->chrToYV12 = bgr48LEToUV_half; break;
         case PIX_FMT_RGB32  : c->chrToYV12 = bgr32ToUV_half;  break;
         case PIX_FMT_RGB32_1: c->chrToYV12 = bgr321ToUV_half; break;
         case PIX_FMT_BGR24  : c->chrToYV12 = bgr24ToUV_half_c; break;
@@ -903,10 +903,10 @@ static void sws_init_swScale_c(SwsContext *c)
         }
     } else {
         switch(srcFormat) {
-        case PIX_FMT_RGB48BE:
-        case PIX_FMT_RGB48LE: c->chrToYV12 = rgb48ToUV; break;
-        case PIX_FMT_BGR48BE:
-        case PIX_FMT_BGR48LE: c->chrToYV12 = bgr48ToUV; break;
+        case PIX_FMT_RGB48BE: c->chrToYV12 = rgb48BEToUV; break;
+        case PIX_FMT_RGB48LE: c->chrToYV12 = rgb48LEToUV; break;
+        case PIX_FMT_BGR48BE: c->chrToYV12 = bgr48BEToUV; break;
+        case PIX_FMT_BGR48LE: c->chrToYV12 = bgr48LEToUV; break;
         case PIX_FMT_RGB32  : c->chrToYV12 = bgr32ToUV;  break;
         case PIX_FMT_RGB32_1: c->chrToYV12 = bgr321ToUV; break;
         case PIX_FMT_BGR24  : c->chrToYV12 = bgr24ToUV_c; break;
@@ -945,10 +945,10 @@ static void sws_init_swScale_c(SwsContext *c)
     case PIX_FMT_RGB32_1: c->lumToYV12 = bgr321ToY; break;
     case PIX_FMT_BGR32  : c->lumToYV12 = rgb32ToY;  break;
     case PIX_FMT_BGR32_1: c->lumToYV12 = rgb321ToY; break;
-    case PIX_FMT_RGB48BE:
-    case PIX_FMT_RGB48LE: c->lumToYV12 = rgb48ToY; break;
-    case PIX_FMT_BGR48BE:
-    case PIX_FMT_BGR48LE: c->lumToYV12 = bgr48ToY; break;
+    case PIX_FMT_RGB48BE: c->lumToYV12 = rgb48BEToY; break;
+    case PIX_FMT_RGB48LE: c->lumToYV12 = rgb48LEToY; break;
+    case PIX_FMT_BGR48BE: c->lumToYV12 = bgr48BEToY; break;
+    case PIX_FMT_BGR48LE: c->lumToYV12 = bgr48LEToY; break;
     }
     if (c->alpPixBuf) {
         switch (srcFormat) {
@@ -972,12 +972,6 @@ static void sws_init_swScale_c(SwsContext *c)
     case PIX_FMT_BGR32  :
         c->alpSrcOffset = 3;
         break;
-    case PIX_FMT_RGB48LE:
-    case PIX_FMT_BGR48LE:
-        c->lumSrcOffset = 1;
-        c->chrSrcOffset = 1;
-        c->alpSrcOffset = 1;
-        break;
     }
 
     if (c->srcRange != c->dstRange && !isAnyRGB(c->dstFormat)) {
diff --git a/tests/ref/lavfi/pixfmts_scale_le b/tests/ref/lavfi/pixfmts_scale_le
index a66138a8a3..6e9ab9ae49 100644
--- a/tests/ref/lavfi/pixfmts_scale_le
+++ b/tests/ref/lavfi/pixfmts_scale_le
@@ -1,8 +1,8 @@
 abgr                cff82561a074874027ac1cc896fd2730
 argb                756dd1eaa5baca2238ce23dbdc452684
 bgr24               e44192347a45586c6c157e3059610cd1
-bgr48be             62e6043fbe9734e63ad679999ca8011c
-bgr48le             61237dad4fa5f3e9109db85f53cd25d9
+bgr48be             6d01b6ccd2ccf18c12985bcb2fde2218
+bgr48le             4caa6914091ad03b8f67c02d6b050bc0
 bgr4_byte           ee1d35a7baf8e9016891929a2f565c0b
 bgr555le            41e3e0961478dc634bf68a7bbd670cc9
 bgr565le            614897eaeb422bd9a972f8ee51909be5
@@ -16,8 +16,8 @@ monow               fd5d417ab7728acddffc06870661df61
 nv12                4676d59db43d657dc12841f6bc3ab452
 nv21                69c699510ff1fb777b118ebee1002f14
 rgb24               13ff53ebeab74dc05492836f1cfbd2c1
-rgb48be             7f6b1f8139c6a64eadf9dfa867ac20e8
-rgb48le             2756d8710c152cbc367656de4d0f1b76
+rgb48be             f82e99f13d5ede2a53cf3bf7178ca350
+rgb48le             3a09d89e4b27ea1a98f762e662e306a7
 rgb4_byte           d81ffd3add95842a618eec81024f0b5c
 rgb555le            bd698d86c03170c4a16607c0fd1f750f
 rgb565le            bfa0c639d80c3c03fd0c9e5f34296a5e