sws/x86: update rgb/bgr->yuv code to use provided coefficient table

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

3 files changed, 50 insertions, 17 deletions

diff --git a/libswscale/swscale_internal.h b/libswscale/swscale_internal.h
index 4de94a6247..7208893a21 100644
--- a/libswscale/swscale_internal.h
+++ b/libswscale/swscale_internal.h
@@ -360,7 +360,7 @@ 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]; // 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+32*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 98ba283eb3..ce2604b225 100644
--- a/libswscale/utils.c
+++ b/libswscale/utils.c
@@ -808,6 +808,22 @@ static void fill_rgb2yuv_table(SwsContext *c, const int table[4], int dstRange)
     int64_t vg = -table[3];
     int64_t ONE = 65536;
     int64_t cy = ONE;
+    uint8_t *p = (uint8_t*)c->input_rgb2yuv_table;
+    int i;
+    static const int8_t map[] = {
+    BY_IDX, GY_IDX, -1    , BY_IDX, BY_IDX, GY_IDX, -1    , BY_IDX,
+    RY_IDX, -1    , GY_IDX, RY_IDX, RY_IDX, -1    , GY_IDX, RY_IDX,
+    RY_IDX, GY_IDX, -1    , RY_IDX, RY_IDX, GY_IDX, -1    , RY_IDX,
+    BY_IDX, -1    , GY_IDX, BY_IDX, BY_IDX, -1    , GY_IDX, BY_IDX,
+    BU_IDX, GU_IDX, -1    , BU_IDX, BU_IDX, GU_IDX, -1    , BU_IDX,
+    RU_IDX, -1    , GU_IDX, RU_IDX, RU_IDX, -1    , GU_IDX, RU_IDX,
+    RU_IDX, GU_IDX, -1    , RU_IDX, RU_IDX, GU_IDX, -1    , RU_IDX,
+    BU_IDX, -1    , GU_IDX, BU_IDX, BU_IDX, -1    , GU_IDX, BU_IDX,
+    BV_IDX, GV_IDX, -1    , BV_IDX, BV_IDX, GV_IDX, -1    , BV_IDX,
+    RV_IDX, -1    , GV_IDX, RV_IDX, RV_IDX, -1    , GV_IDX, RV_IDX,
+    RV_IDX, GV_IDX, -1    , RV_IDX, RV_IDX, GV_IDX, -1    , RV_IDX,
+    BV_IDX, -1    , GV_IDX, BV_IDX, BV_IDX, -1    , GV_IDX, BV_IDX,
+    };
 
     dstRange = 0; //FIXME range = 1 is handled elsewhere
 
@@ -846,6 +862,9 @@ 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);
 }
 
 int sws_setColorspaceDetails(struct SwsContext *c, const int inv_table[4],
diff --git a/libswscale/x86/input.asm b/libswscale/x86/input.asm
index 9d5a871255..995b9033f7 100644
--- a/libswscale/x86/input.asm
+++ b/libswscale/x86/input.asm
@@ -37,18 +37,32 @@ SECTION_RODATA
 
 rgb_Yrnd:        times 4 dd 0x80100        ;  16.5 << 15
 rgb_UVrnd:       times 4 dd 0x400100       ; 128.5 << 15
-bgr_Ycoeff_12x4: times 2 dw BY, GY, 0, BY
-bgr_Ycoeff_3x56: times 2 dw RY, 0, GY, RY
-rgb_Ycoeff_12x4: times 2 dw RY, GY, 0, RY
-rgb_Ycoeff_3x56: times 2 dw BY, 0, GY, BY
-bgr_Ucoeff_12x4: times 2 dw BU, GU, 0, BU
-bgr_Ucoeff_3x56: times 2 dw RU, 0, GU, RU
-rgb_Ucoeff_12x4: times 2 dw RU, GU, 0, RU
-rgb_Ucoeff_3x56: times 2 dw BU, 0, GU, BU
-bgr_Vcoeff_12x4: times 2 dw BV, GV, 0, BV
-bgr_Vcoeff_3x56: times 2 dw RV, 0, GV, RV
-rgb_Vcoeff_12x4: times 2 dw RV, GV, 0, RV
-rgb_Vcoeff_3x56: times 2 dw BV, 0, GV, BV
+%define bgr_Ycoeff_12x4 16*4 + 16* 0 + tableq
+%define bgr_Ycoeff_3x56 16*4 + 16* 1 + tableq
+%define rgb_Ycoeff_12x4 16*4 + 16* 2 + tableq
+%define rgb_Ycoeff_3x56 16*4 + 16* 3 + tableq
+%define bgr_Ucoeff_12x4 16*4 + 16* 4 + tableq
+%define bgr_Ucoeff_3x56 16*4 + 16* 5 + tableq
+%define rgb_Ucoeff_12x4 16*4 + 16* 6 + tableq
+%define rgb_Ucoeff_3x56 16*4 + 16* 7 + tableq
+%define bgr_Vcoeff_12x4 16*4 + 16* 8 + tableq
+%define bgr_Vcoeff_3x56 16*4 + 16* 9 + tableq
+%define rgb_Vcoeff_12x4 16*4 + 16*10 + tableq
+%define rgb_Vcoeff_3x56 16*4 + 16*11 + tableq
+
+
+; bgr_Ycoeff_12x4: times 2 dw BY, GY, 0, BY
+; bgr_Ycoeff_3x56: times 2 dw RY, 0, GY, RY
+; rgb_Ycoeff_12x4: times 2 dw RY, GY, 0, RY
+; rgb_Ycoeff_3x56: times 2 dw BY, 0, GY, BY
+; bgr_Ucoeff_12x4: times 2 dw BU, GU, 0, BU
+; bgr_Ucoeff_3x56: times 2 dw RU, 0, GU, RU
+; rgb_Ucoeff_12x4: times 2 dw RU, GU, 0, RU
+; rgb_Ucoeff_3x56: times 2 dw BU, 0, GU, BU
+; bgr_Vcoeff_12x4: times 2 dw BV, GV, 0, BV
+; bgr_Vcoeff_3x56: times 2 dw RV, 0, GV, RV
+; rgb_Vcoeff_12x4: times 2 dw RV, GV, 0, RV
+; rgb_Vcoeff_3x56: times 2 dw BV, 0, GV, BV
 
 rgba_Ycoeff_rb:  times 4 dw RY, BY
 rgba_Ycoeff_br:  times 4 dw BY, RY
@@ -82,7 +96,7 @@ SECTION .text
 ; %1 = nr. of XMM registers
 ; %2 = rgb or bgr
 %macro RGB24_TO_Y_FN 2-3
-cglobal %2 %+ 24ToY, 6, 6, %1, dst, src, u1, u2, w, u3
+cglobal %2 %+ 24ToY, 6, 6, %1, dst, src, u1, u2, w, table
 %if mmsize == 8
     mova           m5, [%2_Ycoeff_12x4]
     mova           m6, [%2_Ycoeff_3x56]
@@ -171,7 +185,7 @@ cglobal %2 %+ 24ToY, 6, 6, %1, dst, src, u1, u2, w, u3
 ; %1 = nr. of XMM registers
 ; %2 = rgb or bgr
 %macro RGB24_TO_UV_FN 2-3
-cglobal %2 %+ 24ToUV, 7, 7, %1, dstU, dstV, u1, src, u2, w, u3
+cglobal %2 %+ 24ToUV, 7, 7, %1, dstU, dstV, u1, src, u2, w, table
 %if ARCH_X86_64
     mova           m8, [%2_Ucoeff_12x4]
     mova           m9, [%2_Ucoeff_3x56]
@@ -311,7 +325,7 @@ RGB24_FUNCS 11, 13
 ; %1 = nr. of XMM registers
 ; %2-5 = rgba, bgra, argb or abgr (in individual characters)
 %macro RGB32_TO_Y_FN 5-6
-cglobal %2%3%4%5 %+ ToY, 6, 6, %1, dst, src, u1, u2, w, u3
+cglobal %2%3%4%5 %+ ToY, 6, 6, %1, dst, src, u1, u2, w, table
     mova           m5, [rgba_Ycoeff_%2%4]
     mova           m6, [rgba_Ycoeff_%3%5]
 %if %0 == 6
@@ -354,7 +368,7 @@ cglobal %2%3%4%5 %+ ToY, 6, 6, %1, dst, src, u1, u2, w, u3
 ; %1 = nr. of XMM registers
 ; %2-5 = rgba, bgra, argb or abgr (in individual characters)
 %macro RGB32_TO_UV_FN 5-6
-cglobal %2%3%4%5 %+ ToUV, 7, 7, %1, dstU, dstV, u1, src, u2, w, u3
+cglobal %2%3%4%5 %+ ToUV, 7, 7, %1, dstU, dstV, u1, src, u2, w, table
 %if ARCH_X86_64
     mova           m8, [rgba_Ucoeff_%2%4]
     mova           m9, [rgba_Ucoeff_%3%5]