sws/x86: update RENAME(rgb24toyv12)() to using the user provided rgb2yuv table

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

3 files changed, 24 insertions, 11 deletions

diff --git a/libswscale/swscale_internal.h b/libswscale/swscale_internal.h
index 3bdc92eee9..37d6900396 100644
--- a/libswscale/swscale_internal.h
+++ b/libswscale/swscale_internal.h
@@ -358,7 +358,7 @@ typedef struct SwsContext {
     uint8_t *table_gU[256 + 2*YUVRGB_TABLE_HEADROOM];
     int table_gV[256 + 2*YUVRGB_TABLE_HEADROOM];
     uint8_t *table_bU[256 + 2*YUVRGB_TABLE_HEADROOM];
-    int32_t input_rgb2yuv_table[16+32*4]; // This table can contain both C and SIMD formatted values, teh C vales are always at the XY_IDX points
+    int32_t input_rgb2yuv_table[16+40*4]; // This table can contain both C and SIMD formatted values, teh C vales are always at the XY_IDX points
 #define RY_IDX 0
 #define GY_IDX 1
 #define BY_IDX 2
diff --git a/libswscale/utils.c b/libswscale/utils.c
index 4025e7284f..ee522c95fc 100644
--- a/libswscale/utils.c
+++ b/libswscale/utils.c
@@ -834,7 +834,18 @@ static void fill_rgb2yuv_table(SwsContext *c, const int table[4], int dstRange)
     RV_IDX, BV_IDX, RV_IDX, BV_IDX, RV_IDX, BV_IDX, RV_IDX, BV_IDX,
     BV_IDX, RV_IDX, BV_IDX, RV_IDX, BV_IDX, RV_IDX, BV_IDX, RV_IDX,
     GV_IDX, -1    , GV_IDX, -1    , GV_IDX, -1    , GV_IDX, -1    ,
-    -1    , GV_IDX, -1    , GV_IDX, -1    , GV_IDX, -1    , GV_IDX,
+    -1    , GV_IDX, -1    , GV_IDX, -1    , GV_IDX, -1    , GV_IDX, //23
+    -1    , -1    , -1    , -1    , -1    , -1    , -1    , -1    , //24
+    -1    , -1    , -1    , -1    , -1    , -1    , -1    , -1    , //25
+    -1    , -1    , -1    , -1    , -1    , -1    , -1    , -1    , //26
+    -1    , -1    , -1    , -1    , -1    , -1    , -1    , -1    , //27
+    -1    , -1    , -1    , -1    , -1    , -1    , -1    , -1    , //28
+    -1    , -1    , -1    , -1    , -1    , -1    , -1    , -1    , //29
+    -1    , -1    , -1    , -1    , -1    , -1    , -1    , -1    , //30
+    -1    , -1    , -1    , -1    , -1    , -1    , -1    , -1    , //31
+    BY_IDX, GY_IDX, RY_IDX, -1    , -1    , -1    , -1    , -1    , //32
+    BU_IDX, GU_IDX, RU_IDX, -1    , -1    , -1    , -1    , -1    , //33
+    BV_IDX, GV_IDX, RV_IDX, -1    , -1    , -1    , -1    , -1    , //34
     };
 
     dstRange = 0; //FIXME range = 1 is handled elsewhere
@@ -874,7 +885,6 @@ static void fill_rgb2yuv_table(SwsContext *c, const int table[4], int dstRange)
         c->input_rgb2yuv_table[RV_IDX] =  ((int)(0.500 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
         c->input_rgb2yuv_table[RU_IDX] = (-(int)(0.169 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5));
     }
-
     for(i=0; i<FF_ARRAY_ELEMS(map); i++)
         AV_WL16(p + 16*4 + 2*i, map[i] >= 0 ? c->input_rgb2yuv_table[map[i]] : 0);
 }
diff --git a/libswscale/x86/rgb2rgb_template.c b/libswscale/x86/rgb2rgb_template.c
index 68e2821b56..8998fea2d6 100644
--- a/libswscale/x86/rgb2rgb_template.c
+++ b/libswscale/x86/rgb2rgb_template.c
@@ -1615,6 +1615,9 @@ static inline void RENAME(rgb24toyv12)(const uint8_t *src, uint8_t *ydst, uint8_
                                        int lumStride, int chromStride, int srcStride,
                                        int32_t *rgb2yuv)
 {
+#define BGR2Y_IDX "16*4+16*32"
+#define BGR2U_IDX "16*4+16*33"
+#define BGR2V_IDX "16*4+16*34"
     int y;
     const x86_reg chromWidth= width>>1;
     for (y=0; y<height-2; y+=2) {
@@ -1622,7 +1625,7 @@ static inline void RENAME(rgb24toyv12)(const uint8_t *src, uint8_t *ydst, uint8_
         for (i=0; i<2; i++) {
             __asm__ volatile(
                 "mov                        %2, %%"REG_a"   \n\t"
-                "movq  "MANGLE(ff_bgr2YCoeff)", %%mm6       \n\t"
+                "movq          "BGR2Y_IDX"(%3), %%mm6       \n\t"
                 "movq       "MANGLE(ff_w1111)", %%mm5       \n\t"
                 "pxor                    %%mm7, %%mm7       \n\t"
                 "lea (%%"REG_a", %%"REG_a", 2), %%"REG_d"   \n\t"
@@ -1682,7 +1685,7 @@ static inline void RENAME(rgb24toyv12)(const uint8_t *src, uint8_t *ydst, uint8_
                 MOVNTQ"                  %%mm0, (%1, %%"REG_a") \n\t"
                 "add                        $8,      %%"REG_a"  \n\t"
                 " js                        1b                  \n\t"
-                : : "r" (src+width*3), "r" (ydst+width), "g" ((x86_reg)-width)
+                : : "r" (src+width*3), "r" (ydst+width), "g" ((x86_reg)-width), "r"(rgb2yuv)
                 : "%"REG_a, "%"REG_d
             );
             ydst += lumStride;
@@ -1692,7 +1695,7 @@ static inline void RENAME(rgb24toyv12)(const uint8_t *src, uint8_t *ydst, uint8_
         __asm__ volatile(
             "mov                        %4, %%"REG_a"   \n\t"
             "movq       "MANGLE(ff_w1111)", %%mm5       \n\t"
-            "movq  "MANGLE(ff_bgr2UCoeff)", %%mm6       \n\t"
+            "movq          "BGR2U_IDX"(%5), %%mm6       \n\t"
             "pxor                    %%mm7, %%mm7       \n\t"
             "lea (%%"REG_a", %%"REG_a", 2), %%"REG_d"   \n\t"
             "add                 %%"REG_d", %%"REG_d"   \n\t"
@@ -1741,8 +1744,8 @@ static inline void RENAME(rgb24toyv12)(const uint8_t *src, uint8_t *ydst, uint8_
             "psrlw                      $2, %%mm0       \n\t"
             "psrlw                      $2, %%mm2       \n\t"
 #endif
-            "movq  "MANGLE(ff_bgr2VCoeff)", %%mm1       \n\t"
-            "movq  "MANGLE(ff_bgr2VCoeff)", %%mm3       \n\t"
+            "movq          "BGR2V_IDX"(%5), %%mm1       \n\t"
+            "movq          "BGR2V_IDX"(%5), %%mm3       \n\t"
 
             "pmaddwd                 %%mm0, %%mm1       \n\t"
             "pmaddwd                 %%mm2, %%mm3       \n\t"
@@ -1801,8 +1804,8 @@ static inline void RENAME(rgb24toyv12)(const uint8_t *src, uint8_t *ydst, uint8_
             "psrlw                      $2, %%mm4       \n\t"
             "psrlw                      $2, %%mm2       \n\t"
 #endif
-            "movq  "MANGLE(ff_bgr2VCoeff)", %%mm1       \n\t"
-            "movq  "MANGLE(ff_bgr2VCoeff)", %%mm3       \n\t"
+            "movq          "BGR2V_IDX"(%5), %%mm1       \n\t"
+            "movq          "BGR2V_IDX"(%5), %%mm3       \n\t"
 
             "pmaddwd                 %%mm4, %%mm1       \n\t"
             "pmaddwd                 %%mm2, %%mm3       \n\t"
@@ -1830,7 +1833,7 @@ static inline void RENAME(rgb24toyv12)(const uint8_t *src, uint8_t *ydst, uint8_
             "movd                    %%mm0, (%3, %%"REG_a") \n\t"
             "add                        $4, %%"REG_a"       \n\t"
             " js                        1b                  \n\t"
-            : : "r" (src+chromWidth*6), "r" (src+srcStride+chromWidth*6), "r" (udst+chromWidth), "r" (vdst+chromWidth), "g" (-chromWidth)
+            : : "r" (src+chromWidth*6), "r" (src+srcStride+chromWidth*6), "r" (udst+chromWidth), "r" (vdst+chromWidth), "g" (-chromWidth), "r"(rgb2yuv)
             : "%"REG_a, "%"REG_d
         );