common code moved to a #define

added bgr24 output support


git-svn-id: svn://svn.mplayerhq.hu/mplayer/trunk@2735 b3059339-0415-0410-9bf9-f77b7e298cf2

1 files changed, 227 insertions, 187 deletions

diff --git a/postproc/yuv2rgb_template.c b/postproc/yuv2rgb_template.c
index 851182b974..833e6a95f3 100644
--- a/postproc/yuv2rgb_template.c
+++ b/postproc/yuv2rgb_template.c
@@ -69,6 +69,81 @@ uint64_t mmx_blueshift = 0x03;
 #define EMMS "emms;"
 #endif
 
+#define YUV2RGB \
+		     /* Do the multiply part of the conversion for even and odd pixels,
+			register usage:
+			mm0 -> Cblue, mm1 -> Cred, mm2 -> Cgreen even pixels,
+			mm3 -> Cblue, mm4 -> Cred, mm5 -> Cgreen odd pixels,
+			mm6 -> Y even, mm7 -> Y odd */\
+		     /* convert the chroma part */\
+		     "punpcklbw %%mm4, %%mm0;" /* scatter 4 Cb 00 u3 00 u2 00 u1 00 u0 */ \
+		     "punpcklbw %%mm4, %%mm1;" /* scatter 4 Cr 00 v3 00 v2 00 v1 00 v0 */ \
+\
+		     "psubsw mmx_80w, %%mm0;" /* Cb -= 128 */ \
+		     "psubsw mmx_80w, %%mm1;" /* Cr -= 128 */ \
+\
+		     "psllw $3, %%mm0;" /* Promote precision */ \
+		     "psllw $3, %%mm1;" /* Promote precision */ \
+\
+		     "movq %%mm0, %%mm2;" /* Copy 4 Cb 00 u3 00 u2 00 u1 00 u0 */ \
+		     "movq %%mm1, %%mm3;" /* Copy 4 Cr 00 v3 00 v2 00 v1 00 v0 */ \
+\
+		     "pmulhw mmx_U_green, %%mm2;" /* Mul Cb with green coeff -> Cb green */ \
+		     "pmulhw mmx_V_green, %%mm3;" /* Mul Cr with green coeff -> Cr green */ \
+\
+		     "pmulhw mmx_U_blue, %%mm0;" /* Mul Cb -> Cblue 00 b3 00 b2 00 b1 00 b0 */\
+		     "pmulhw mmx_V_red, %%mm1;" /* Mul Cr -> Cred 00 r3 00 r2 00 r1 00 r0 */\
+\
+		     "paddsw %%mm3, %%mm2;" /* Cb green + Cr green -> Cgreen */\
+\
+		     /* convert the luma part */\
+		     "psubusb mmx_10w, %%mm6;" /* Y -= 16 */\
+\
+		     "movq %%mm6, %%mm7;" /* Copy 8 Y Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0 */\
+		     "pand mmx_00ffw, %%mm6;" /* get Y even 00 Y6 00 Y4 00 Y2 00 Y0 */\
+\
+		     "psrlw $8, %%mm7;" /* get Y odd 00 Y7 00 Y5 00 Y3 00 Y1 */\
+\
+		     "psllw $3, %%mm6;" /* Promote precision */\
+		     "psllw $3, %%mm7;" /* Promote precision */\
+\
+		     "pmulhw mmx_Y_coeff, %%mm6;" /* Mul 4 Y even 00 y6 00 y4 00 y2 00 y0 */\
+		     "pmulhw mmx_Y_coeff, %%mm7;" /* Mul 4 Y odd 00 y7 00 y5 00 y3 00 y1 */\
+\
+		     /* Do the addition part of the conversion for even and odd pixels,
+			register usage:
+			mm0 -> Cblue, mm1 -> Cred, mm2 -> Cgreen even pixels,
+			mm3 -> Cblue, mm4 -> Cred, mm5 -> Cgreen odd pixels,
+			mm6 -> Y even, mm7 -> Y odd */\
+		     "movq %%mm0, %%mm3;" /* Copy Cblue */\
+		     "movq %%mm1, %%mm4;" /* Copy Cred */\
+		     "movq %%mm2, %%mm5;" /* Copy Cgreen */\
+\
+		     "paddsw %%mm6, %%mm0;" /* Y even + Cblue 00 B6 00 B4 00 B2 00 B0 */\
+		     "paddsw %%mm7, %%mm3;" /* Y odd + Cblue 00 B7 00 B5 00 B3 00 B1 */\
+\
+		     "paddsw %%mm6, %%mm1;" /* Y even + Cred 00 R6 00 R4 00 R2 00 R0 */\
+		     "paddsw %%mm7, %%mm4;" /* Y odd + Cred 00 R7 00 R5 00 R3 00 R1 */\
+\
+		     "paddsw %%mm6, %%mm2;" /* Y even + Cgreen 00 G6 00 G4 00 G2 00 G0 */\
+		     "paddsw %%mm7, %%mm5;" /* Y odd + Cgreen 00 G7 00 G5 00 G3 00 G1 */\
+\
+		     /* Limit RGB even to 0..255 */\
+		     "packuswb %%mm0, %%mm0;" /* B6 B4 B2 B0  B6 B4 B2 B0 */\
+		     "packuswb %%mm1, %%mm1;" /* R6 R4 R2 R0  R6 R4 R2 R0 */\
+		     "packuswb %%mm2, %%mm2;" /* G6 G4 G2 G0  G6 G4 G2 G0 */\
+\
+		     /* Limit RGB odd to 0..255 */\
+		     "packuswb %%mm3, %%mm3;" /* B7 B5 B3 B1  B7 B5 B3 B1 */\
+		     "packuswb %%mm4, %%mm4;" /* R7 R5 R3 R1  R7 R5 R3 R1 */\
+		     "packuswb %%mm5, %%mm5;" /* G7 G5 G3 G1  G7 G5 G3 G1 */\
+\
+		     /* Interleave RGB even and odd */\
+		     "punpcklbw %%mm3, %%mm0;" /* B7 B6 B5 B4 B3 B2 B1 B0 */\
+		     "punpcklbw %%mm4, %%mm1;" /* R7 R6 R5 R4 R3 R2 R1 R0 */\
+		     "punpcklbw %%mm5, %%mm2;" /* G7 G6 G5 G4 G3 G2 G1 G0 */\
+
+
 static void yuv420_rgb16_mmx (uint8_t * image, uint8_t * py,
 			      uint8_t * pu, uint8_t * pv,
 			      int h_size, int v_size,
@@ -87,9 +162,9 @@ static void yuv420_rgb16_mmx (uint8_t * image, uint8_t * py,
 
 	/* load data for start of next scan line */
 	__asm__ __volatile__ (
-		 "movd (%1), %%mm0;" /* Load 4 Cb 00 00 00 00 u3 u2 u1 u0 */ 
-		 "movd (%2), %%mm1;" /* Load 4 Cr 00 00 00 00 v3 v2 v1 v0 */ 
-		 "movq (%0), %%mm6;" /* Load 8  Y Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0 */ 
+		 "movd (%1), %%mm0;" /* Load 4 Cb 00 00 00 00 u3 u2 u1 u0 */
+		 "movd (%2), %%mm1;" /* Load 4 Cr 00 00 00 00 v3 v2 v1 v0 */
+		 "movq (%0), %%mm6;" /* Load 8  Y Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0 */
 
 		 : : "r" (_py), "r" (_pu), "r" (_pv));
 
@@ -98,117 +173,154 @@ static void yuv420_rgb16_mmx (uint8_t * image, uint8_t * py,
 	       pixels in each iteration */
 
 	    __asm__ __volatile__ (
-		     /* Do the multiply part of the conversion for even and odd pixels,
-			register usage:
-			mm0 -> Cblue, mm1 -> Cred, mm2 -> Cgreen even pixels,
-			mm3 -> Cblue, mm4 -> Cred, mm5 -> Cgreen odd pixels,
-			mm6 -> Y even, mm7 -> Y odd */
-		     /* convert the chroma part */
-		     "punpcklbw %%mm4, %%mm0;" /* scatter 4 Cb 00 u3 00 u2 00 u1 00 u0 */ 
-		     "punpcklbw %%mm4, %%mm1;" /* scatter 4 Cr 00 v3 00 v2 00 v1 00 v0 */ 
-
-		     "psubsw mmx_80w, %%mm0;" /* Cb -= 128 */ 
-		     "psubsw mmx_80w, %%mm1;" /* Cr -= 128 */ 
-
-		     "psllw $3, %%mm0;" /* Promote precision */ 
-		     "psllw $3, %%mm1;" /* Promote precision */ 
-
-		     "movq %%mm0, %%mm2;" /* Copy 4 Cb 00 u3 00 u2 00 u1 00 u0 */ 
-		     "movq %%mm1, %%mm3;" /* Copy 4 Cr 00 v3 00 v2 00 v1 00 v0 */ 
-
-		     "pmulhw mmx_U_green, %%mm2;" /* Mul Cb with green coeff -> Cb green */ 
-		     "pmulhw mmx_V_green, %%mm3;" /* Mul Cr with green coeff -> Cr green */ 
+YUV2RGB
 
-		     "pmulhw mmx_U_blue, %%mm0;" /* Mul Cb -> Cblue 00 b3 00 b2 00 b1 00 b0 */ 
-		     "pmulhw mmx_V_red, %%mm1;" /* Mul Cr -> Cred 00 r3 00 r2 00 r1 00 r0 */ 
+		     /* mask unneeded bits off */
+		     "pand mmx_redmask, %%mm0;" /* b7b6b5b4 b3_0_0_0 b7b6b5b4 b3_0_0_0 */
+		     "pand mmx_grnmask, %%mm2;" /* g7g6g5g4 g3g2_0_0 g7g6g5g4 g3g2_0_0 */
+		     "pand mmx_redmask, %%mm1;" /* r7r6r5r4 r3_0_0_0 r7r6r5r4 r3_0_0_0 */
 
-		     "paddsw %%mm3, %%mm2;" /* Cb green + Cr green -> Cgreen */ 
+		     "psrlw mmx_blueshift,%%mm0;" /* 0_0_0_b7 b6b5b4b3 0_0_0_b7 b6b5b4b3 */
+		     "pxor %%mm4, %%mm4;" /* zero mm4 */
 
-		     /* convert the luma part */
-		     "psubusb mmx_10w, %%mm6;" /* Y -= 16 */ 
+		     "movq %%mm0, %%mm5;" /* Copy B7-B0 */
+		     "movq %%mm2, %%mm7;" /* Copy G7-G0 */
 
-		     "movq %%mm6, %%mm7;" /* Copy 8 Y Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0 */ 
-		     "pand mmx_00ffw, %%mm6;" /* get Y even 00 Y6 00 Y4 00 Y2 00 Y0 */ 
+		     /* convert rgb24 plane to rgb16 pack for pixel 0-3 */
+		     "punpcklbw %%mm4, %%mm2;" /* 0_0_0_0 0_0_0_0 g7g6g5g4 g3g2_0_0 */
+		     "punpcklbw %%mm1, %%mm0;" /* r7r6r5r4 r3_0_0_0 0_0_0_b7 b6b5b4b3 */
 
-		     "psrlw $8, %%mm7;" /* get Y odd 00 Y7 00 Y5 00 Y3 00 Y1 */ 
+		     "psllw mmx_blueshift,%%mm2;" /* 0_0_0_0 0_g7g6g5 g4g3g2_0 0_0_0_0 */
+		     "por %%mm2, %%mm0;" /* r7r6r5r4 r3g7g6g5 g4g3g2b7 b6b5b4b3 */
 
-		     "psllw $3, %%mm6;" /* Promote precision */ 
-		     "psllw $3, %%mm7;" /* Promote precision */ 
+		     "movq 8 (%0), %%mm6;" /* Load 8 Y Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0 */
+		     MOVNTQ " %%mm0, (%3);" /* store pixel 0-3 */
 
-		     "pmulhw mmx_Y_coeff, %%mm6;" /* Mul 4 Y even 00 y6 00 y4 00 y2 00 y0 */ 
-		     "pmulhw mmx_Y_coeff, %%mm7;" /* Mul 4 Y odd 00 y7 00 y5 00 y3 00 y1 */ 
+		     /* convert rgb24 plane to rgb16 pack for pixel 0-3 */
+		     "punpckhbw %%mm4, %%mm7;" /* 0_0_0_0 0_0_0_0 g7g6g5g4 g3g2_0_0 */
+		     "punpckhbw %%mm1, %%mm5;" /* r7r6r5r4 r3_0_0_0 0_0_0_b7 b6b5b4b3 */
 
-		     /* Do the addition part of the conversion for even and odd pixels,
-			register usage:
-			mm0 -> Cblue, mm1 -> Cred, mm2 -> Cgreen even pixels,
-			mm3 -> Cblue, mm4 -> Cred, mm5 -> Cgreen odd pixels,
-			mm6 -> Y even, mm7 -> Y odd */
-		     "movq %%mm0, %%mm3;" /* Copy Cblue */ 
-		     "movq %%mm1, %%mm4;" /* Copy Cred */ 
-		     "movq %%mm2, %%mm5;" /* Copy Cgreen */ 
+		     "psllw mmx_blueshift,%%mm7;" /* 0_0_0_0 0_g7g6g5 g4g3g2_0 0_0_0_0 */
+		     "movd 4 (%1), %%mm0;" /* Load 4 Cb 00 00 00 00 u3 u2 u1 u0 */
 
-		     "paddsw %%mm6, %%mm0;" /* Y even + Cblue 00 B6 00 B4 00 B2 00 B0 */ 
-		     "paddsw %%mm7, %%mm3;" /* Y odd + Cblue 00 B7 00 B5 00 B3 00 B1 */ 
+		     "por %%mm7, %%mm5;" /* r7r6r5r4 r3g7g6g5 g4g3g2b7 b6b5b4b3 */
+		     "movd 4 (%2), %%mm1;" /* Load 4 Cr 00 00 00 00 v3 v2 v1 v0 */
 
-		     "paddsw %%mm6, %%mm1;" /* Y even + Cred 00 R6 00 R4 00 R2 00 R0 */ 
-		     "paddsw %%mm7, %%mm4;" /* Y odd + Cred 00 R7 00 R5 00 R3 00 R1 */ 
+		     MOVNTQ " %%mm5, 8 (%3);" /* store pixel 4-7 */
+		     : : "r" (_py), "r" (_pu), "r" (_pv), "r" (_image));
 
-		     "paddsw %%mm6, %%mm2;" /* Y even + Cgreen 00 G6 00 G4 00 G2 00 G0 */ 
-		     "paddsw %%mm7, %%mm5;" /* Y odd + Cgreen 00 G7 00 G5 00 G3 00 G1 */ 
+	    _py += 8;
+	    _pu += 4;
+	    _pv += 4;
+	    _image += 16;
+	}
 
-		     /* Limit RGB even to 0..255 */
-		     "packuswb %%mm0, %%mm0;" /* B6 B4 B2 B0  B6 B4 B2 B0 */ 
-		     "packuswb %%mm1, %%mm1;" /* R6 R4 R2 R0  R6 R4 R2 R0 */ 
-		     "packuswb %%mm2, %%mm2;" /* G6 G4 G2 G0  G6 G4 G2 G0 */ 
+	if (!even) {
+	    pu += uv_stride;
+	    pv += uv_stride;
+	}
 
-		     /* Limit RGB odd to 0..255 */
-		     "packuswb %%mm3, %%mm3;" /* B7 B5 B3 B1  B7 B5 B3 B1 */ 
-		     "packuswb %%mm4, %%mm4;" /* R7 R5 R3 R1  R7 R5 R3 R1 */ 
-		     "packuswb %%mm5, %%mm5;" /* G7 G5 G3 G1  G7 G5 G3 G1 */ 
+	py += y_stride;
+	image += rgb_stride;
 
-		     /* Interleave RGB even and odd */
-		     "punpcklbw %%mm3, %%mm0;" /* B7 B6 B5 B4 B3 B2 B1 B0 */ 
-		     "punpcklbw %%mm4, %%mm1;" /* R7 R6 R5 R4 R3 R2 R1 R0 */ 
-		     "punpcklbw %%mm5, %%mm2;" /* G7 G6 G5 G4 G3 G2 G1 G0 */ 
+	even = (!even);
+    }
 
-		     /* mask unneeded bits off */
-		     "pand mmx_redmask, %%mm0;" /* b7b6b5b4 b3_0_0_0 b7b6b5b4 b3_0_0_0 */ 
-		     "pand mmx_grnmask, %%mm2;" /* g7g6g5g4 g3g2_0_0 g7g6g5g4 g3g2_0_0 */ 
-		     "pand mmx_redmask, %%mm1;" /* r7r6r5r4 r3_0_0_0 r7r6r5r4 r3_0_0_0 */ 
+    __asm__ __volatile__ (EMMS);
+}
 
-		     "psrlw mmx_blueshift,%%mm0;" /* 0_0_0_b7 b6b5b4b3 0_0_0_b7 b6b5b4b3 */ 
-		     "pxor %%mm4, %%mm4;" /* zero mm4 */ 
+static void yuv420_rgb24_mmx (uint8_t * image, uint8_t * py,
+			      uint8_t * pu, uint8_t * pv,
+			      int h_size, int v_size,
+			      int rgb_stride, int y_stride, int uv_stride)
+{
+    int even = 1;
+    int x, y;
 
-		     "movq %%mm0, %%mm5;" /* Copy B7-B0 */ 
-		     "movq %%mm2, %%mm7;" /* Copy G7-G0 */ 
+    __asm__ __volatile__ ("pxor %mm4, %mm4;" /* zero mm4 */ );
 
-		     /* convert rgb24 plane to rgb16 pack for pixel 0-3 */
-		     "punpcklbw %%mm4, %%mm2;" /* 0_0_0_0 0_0_0_0 g7g6g5g4 g3g2_0_0 */ 
-		     "punpcklbw %%mm1, %%mm0;" /* r7r6r5r4 r3_0_0_0 0_0_0_b7 b6b5b4b3 */ 
+    for (y = v_size; --y >= 0; ) {
+	uint8_t *_image = image;
+	uint8_t *_py = py;
+	uint8_t *_pu = pu;
+	uint8_t *_pv = pv;
 
-		     "psllw mmx_blueshift,%%mm2;" /* 0_0_0_0 0_g7g6g5 g4g3g2_0 0_0_0_0 */ 
-		     "por %%mm2, %%mm0;" /* r7r6r5r4 r3g7g6g5 g4g3g2b7 b6b5b4b3 */ 
+	/* load data for start of next scan line */
+	__asm__ __volatile__ (
+		 "movd (%1), %%mm0;" /* Load 4 Cb 00 00 00 00 u3 u2 u1 u0 */
+		 "movd (%2), %%mm1;" /* Load 4 Cr 00 00 00 00 v3 v2 v1 v0 */
+		 "movq (%0), %%mm6;" /* Load 8  Y Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0 */
 
-		     "movq 8 (%0), %%mm6;" /* Load 8 Y Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0 */ 
-		     MOVNTQ " %%mm0, (%3);" /* store pixel 0-3 */ 
+		 : : "r" (_py), "r" (_pu), "r" (_pv));
 
-		     /* convert rgb24 plane to rgb16 pack for pixel 0-3 */
-		     "punpckhbw %%mm4, %%mm7;" /* 0_0_0_0 0_0_0_0 g7g6g5g4 g3g2_0_0 */ 
-		     "punpckhbw %%mm1, %%mm5;" /* r7r6r5r4 r3_0_0_0 0_0_0_b7 b6b5b4b3 */ 
+	for (x = h_size >> 3; --x >= 0; ) {
+	    /* this mmx assembly code deals with SINGLE scan line at a time, it convert 8
+	       pixels in each iteration */
 
-		     "psllw mmx_blueshift,%%mm7;" /* 0_0_0_0 0_g7g6g5 g4g3g2_0 0_0_0_0 */ 
-		     "movd 4 (%1), %%mm0;" /* Load 4 Cb 00 00 00 00 u3 u2 u1 u0 */ 
+	    __asm__ __volatile__ (
+YUV2RGB
+
+	/* mm0=B, %%mm2=G, %%mm1=R */
+			"pxor %%mm4, %%mm4		\n\t"
+			"movq %%mm0, %%mm5		\n\t" /* B */
+			"movq %%mm1, %%mm6		\n\t" /* R */
+			"punpcklbw %%mm2, %%mm0		\n\t" /* GBGBGBGB 0 */
+			"punpcklbw %%mm4, %%mm1		\n\t" /* 0R0R0R0R 0 */
+			"punpckhbw %%mm2, %%mm5		\n\t" /* GBGBGBGB 2 */
+			"punpckhbw %%mm4, %%mm6		\n\t" /* 0R0R0R0R 2 */
+			"movq %%mm0, %%mm7		\n\t" /* GBGBGBGB 0 */
+			"movq %%mm5, %%mm3		\n\t" /* GBGBGBGB 2 */
+			"punpcklwd %%mm1, %%mm7		\n\t" /* 0RGB0RGB 0 */
+			"punpckhwd %%mm1, %%mm0		\n\t" /* 0RGB0RGB 1 */
+			"punpcklwd %%mm6, %%mm5		\n\t" /* 0RGB0RGB 2 */
+			"punpckhwd %%mm6, %%mm3		\n\t" /* 0RGB0RGB 3 */
+
+			"movq %%mm7, %%mm2		\n\t" /* 0RGB0RGB 0 */
+			"movq %%mm0, %%mm6		\n\t" /* 0RGB0RGB 1 */
+			"movq %%mm5, %%mm1		\n\t" /* 0RGB0RGB 2 */
+			"movq %%mm3, %%mm4		\n\t" /* 0RGB0RGB 3 */
+
+			"psllq $40, %%mm7		\n\t" /* RGB00000 0 */
+			"psllq $40, %%mm0		\n\t" /* RGB00000 1 */
+			"psllq $40, %%mm5		\n\t" /* RGB00000 2 */
+			"psllq $40, %%mm3		\n\t" /* RGB00000 3 */
+
+			"punpckhdq %%mm2, %%mm7		\n\t" /* 0RGBRGB0 0 */
+			"punpckhdq %%mm6, %%mm0		\n\t" /* 0RGBRGB0 1 */
+			"punpckhdq %%mm1, %%mm5		\n\t" /* 0RGBRGB0 2 */
+			"punpckhdq %%mm4, %%mm3		\n\t" /* 0RGBRGB0 3 */
+
+			"psrlq $8, %%mm7		\n\t" /* 00RGBRGB 0 */
+			"movq %%mm0, %%mm6		\n\t" /* 0RGBRGB0 1 */
+			"psllq $40, %%mm0		\n\t" /* GB000000 1 */
+			"por %%mm0, %%mm7		\n\t" /* GBRGBRGB 0 */
+			MOVNTQ" %%mm7, (%3)		\n\t"
+
+			"movd 4 (%1), %%mm0;" /* Load 4 Cb 00 00 00 00 u3 u2 u1 u0 */
+
+			"psrlq $24, %%mm6		\n\t" /* 0000RGBR 1 */
+			"movq %%mm5, %%mm1		\n\t" /* 0RGBRGB0 2 */
+			"psllq $24, %%mm5		\n\t" /* BRGB0000 2 */
+			"por %%mm5, %%mm6		\n\t" /* BRGBRGBR 1 */
+			MOVNTQ" %%mm6, 8(%3)		\n\t"
+
+			"movq 8 (%0), %%mm6;" /* Load 8 Y Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0 */
+
+			"psrlq $40, %%mm1		\n\t" /* 000000RG 2 */
+			"psllq $8, %%mm3		\n\t" /* RGBRGB00 3 */
+			"por %%mm3, %%mm1		\n\t" /* RGBRGBRG 2 */
+			MOVNTQ" %%mm1, 16(%3)		\n\t"
+
+			"movd 4 (%2), %%mm1;" /* Load 4 Cr 00 00 00 00 v3 v2 v1 v0 */
+			"pxor %%mm4, %%mm4		\n\t"
 
-		     "por %%mm7, %%mm5;" /* r7r6r5r4 r3g7g6g5 g4g3g2b7 b6b5b4b3 */ 
-		     "movd 4 (%2), %%mm1;" /* Load 4 Cr 00 00 00 00 v3 v2 v1 v0 */ 
 
-		     MOVNTQ " %%mm5, 8 (%3);" /* store pixel 4-7 */ 
 		     : : "r" (_py), "r" (_pu), "r" (_pv), "r" (_image));
 
 	    _py += 8;
 	    _pu += 4;
 	    _pv += 4;
-	    _image += 16;
+	    _image += 24;
 	}
 
 	if (!even) {
@@ -225,6 +337,7 @@ static void yuv420_rgb16_mmx (uint8_t * image, uint8_t * py,
     __asm__ __volatile__ (EMMS);
 }
 
+
 static void yuv420_argb32_mmx (uint8_t * image, uint8_t * py,
 			       uint8_t * pu, uint8_t * pv,
 			       int h_size, int v_size,
@@ -254,122 +367,48 @@ static void yuv420_argb32_mmx (uint8_t * image, uint8_t * py,
 	    /* this mmx assembly code deals with SINGLE scan line at a time, it convert 8
 	       pixels in each iteration */
 	    __asm__ __volatile__ (
-		     /* Do the multiply part of the conversion for even and odd pixels,
-			register usage:
-			mm0 -> Cblue, mm1 -> Cred, mm2 -> Cgreen even pixels,
-			mm3 -> Cblue, mm4 -> Cred, mm5 -> Cgreen odd pixels,
-			mm6 -> Y even, mm7 -> Y odd */
-
-		     /* convert the chroma part */
-		     "punpcklbw %%mm4, %%mm0;" /* scatter 4 Cb 00 u3 00 u2 00 u1 00 u0 */ 
-		     "punpcklbw %%mm4, %%mm1;" /* scatter 4 Cr 00 v3 00 v2 00 v1 00 v0 */ 
-
-		     "psubsw mmx_80w, %%mm0;" /* Cb -= 128 */ 
-		     "psubsw mmx_80w, %%mm1;" /* Cr -= 128 */ 
-
-		     "psllw $3, %%mm0;" /* Promote precision */ 
-		     "psllw $3, %%mm1;" /* Promote precision */ 
-
-		     "movq %%mm0, %%mm2;" /* Copy 4 Cb 00 u3 00 u2 00 u1 00 u0 */ 
-		     "movq %%mm1, %%mm3;" /* Copy 4 Cr 00 v3 00 v2 00 v1 00 v0 */ 
-
-		     "pmulhw mmx_U_green, %%mm2;" /* Mul Cb with green coeff -> Cb green */ 
-		     "pmulhw mmx_V_green, %%mm3;" /* Mul Cr with green coeff -> Cr green */ 
-
-		     "pmulhw mmx_U_blue, %%mm0;" /* Mul Cb -> Cblue 00 b3 00 b2 00 b1 00 b0 */ 
-		     "pmulhw mmx_V_red, %%mm1;" /* Mul Cr -> Cred 00 r3 00 r2 00 r1 00 r0 */ 
-
-		     "paddsw %%mm3, %%mm2;" /* Cb green + Cr green -> Cgreen */ 
-
-		     /* convert the luma part */
-		     "psubusb mmx_10w, %%mm6;" /* Y -= 16 */ 
-
-		     "movq %%mm6, %%mm7;" /* Copy 8 Y Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0 */ 
-		     "pand mmx_00ffw, %%mm6;" /* get Y even 00 Y6 00 Y4 00 Y2 00 Y0 */ 
-
-		     "psrlw $8, %%mm7;" /* get Y odd 00 Y7 00 Y5 00 Y3 00 Y1 */ 
-
-		     "psllw $3, %%mm6;" /* Promote precision */ 
-		     "psllw $3, %%mm7;" /* Promote precision */ 
-
-		     "pmulhw mmx_Y_coeff, %%mm6;" /* Mul 4 Y even 00 y6 00 y4 00 y2 00 y0 */ 
-		     "pmulhw mmx_Y_coeff, %%mm7;" /* Mul 4 Y odd 00 y7 00 y5 00 y3 00 y1 */ 
-
-		     /* Do the addition part of the conversion for even and odd pixels,
-			register usage:
-			mm0 -> Cblue, mm1 -> Cred, mm2 -> Cgreen even pixels,
-			mm3 -> Cblue, mm4 -> Cred, mm5 -> Cgreen odd pixels,
-			mm6 -> Y even, mm7 -> Y odd */
-
-		     "movq %%mm0, %%mm3;" /* Copy Cblue */ 
-		     "movq %%mm1, %%mm4;" /* Copy Cred */ 
-		     "movq %%mm2, %%mm5;" /* Copy Cgreen */ 
-
-		     "paddsw %%mm6, %%mm0;" /* Y even + Cblue 00 B6 00 B4 00 B2 00 B0 */ 
-		     "paddsw %%mm7, %%mm3;" /* Y odd + Cblue 00 B7 00 B5 00 B3 00 B1 */ 
-
-		     "paddsw %%mm6, %%mm1;" /* Y even + Cred 00 R6 00 R4 00 R2 00 R0 */ 
-		     "paddsw %%mm7, %%mm4;" /* Y odd + Cred 00 R7 00 R5 00 R3 00 R1 */ 
-
-		     "paddsw %%mm6, %%mm2;" /* Y even + Cgreen 00 G6 00 G4 00 G2 00 G0 */ 
-		     "paddsw %%mm7, %%mm5;" /* Y odd + Cgreen 00 G7 00 G5 00 G3 00 G1 */ 
-
-		     /* Limit RGB even to 0..255 */
-		     "packuswb %%mm0, %%mm0;" /* B6 B4 B2 B0 B6 B4 B2 B0 */ 
-		     "packuswb %%mm1, %%mm1;" /* R6 R4 R2 R0 R6 R4 R2 R0 */ 
-		     "packuswb %%mm2, %%mm2;" /* G6 G4 G2 G0 G6 G4 G2 G0 */ 
-
-		     /* Limit RGB odd to 0..255 */
-		     "packuswb %%mm3, %%mm3;" /* B7 B5 B3 B1 B7 B5 B3 B1 */ 
-		     "packuswb %%mm4, %%mm4;" /* R7 R5 R3 R1 R7 R5 R3 R1 */ 
-		     "packuswb %%mm5, %%mm5;" /* G7 G5 G3 G1 G7 G5 G3 G1 */ 
-
-		     /* Interleave RGB even and odd */
-		     "punpcklbw %%mm3, %%mm0;" /* B7 B6 B5 B4 B3 B2 B1 B0 */ 
-		     "punpcklbw %%mm4, %%mm1;" /* R7 R6 R5 R4 R3 R2 R1 R0 */ 
-		     "punpcklbw %%mm5, %%mm2;" /* G7 G6 G5 G4 G3 G2 G1 G0 */ 
-
-		     /* convert RGB plane to RGB packed format, 
+YUV2RGB
+		     /* convert RGB plane to RGB packed format,
 			mm0 -> B, mm1 -> R, mm2 -> G, mm3 -> 0,
 			mm4 -> GB, mm5 -> AR pixel 4-7,
 			mm6 -> GB, mm7 -> AR pixel 0-3 */
-		     "pxor %%mm3, %%mm3;" /* zero mm3 */ 
+		     "pxor %%mm3, %%mm3;" /* zero mm3 */
+
+		     "movq %%mm0, %%mm6;" /* B7 B6 B5 B4 B3 B2 B1 B0 */
+		     "movq %%mm1, %%mm7;" /* R7 R6 R5 R4 R3 R2 R1 R0 */
 
-		     "movq %%mm0, %%mm6;" /* B7 B6 B5 B4 B3 B2 B1 B0 */ 
-		     "movq %%mm1, %%mm7;" /* R7 R6 R5 R4 R3 R2 R1 R0 */ 
+		     "movq %%mm0, %%mm4;" /* B7 B6 B5 B4 B3 B2 B1 B0 */
+		     "movq %%mm1, %%mm5;" /* R7 R6 R5 R4 R3 R2 R1 R0 */
 
-		     "movq %%mm0, %%mm4;" /* B7 B6 B5 B4 B3 B2 B1 B0 */ 
-		     "movq %%mm1, %%mm5;" /* R7 R6 R5 R4 R3 R2 R1 R0 */ 
+		     "punpcklbw %%mm2, %%mm6;" /* G3 B3 G2 B2 G1 B1 G0 B0 */
+		     "punpcklbw %%mm3, %%mm7;" /* 00 R3 00 R2 00 R1 00 R0 */
 
-		     "punpcklbw %%mm2, %%mm6;" /* G3 B3 G2 B2 G1 B1 G0 B0 */ 
-		     "punpcklbw %%mm3, %%mm7;" /* 00 R3 00 R2 00 R1 00 R0 */ 
+		     "punpcklwd %%mm7, %%mm6;" /* 00 R1 B1 G1 00 R0 B0 G0 */
+		     MOVNTQ " %%mm6, (%3);" /* Store ARGB1 ARGB0 */
 
-		     "punpcklwd %%mm7, %%mm6;" /* 00 R1 B1 G1 00 R0 B0 G0 */ 
-		     MOVNTQ " %%mm6, (%3);" /* Store ARGB1 ARGB0 */ 
+		     "movq %%mm0, %%mm6;" /* B7 B6 B5 B4 B3 B2 B1 B0 */
+		     "punpcklbw %%mm2, %%mm6;" /* G3 B3 G2 B2 G1 B1 G0 B0 */
 
-		     "movq %%mm0, %%mm6;" /* B7 B6 B5 B4 B3 B2 B1 B0 */ 
-		     "punpcklbw %%mm2, %%mm6;" /* G3 B3 G2 B2 G1 B1 G0 B0 */ 
+		     "punpckhwd %%mm7, %%mm6;" /* 00 R3 G3 B3 00 R2 B3 G2 */
+		     MOVNTQ " %%mm6, 8 (%3);" /* Store ARGB3 ARGB2 */
 
-		     "punpckhwd %%mm7, %%mm6;" /* 00 R3 G3 B3 00 R2 B3 G2 */ 
-		     MOVNTQ " %%mm6, 8 (%3);" /* Store ARGB3 ARGB2 */ 
+		     "punpckhbw %%mm2, %%mm4;" /* G7 B7 G6 B6 G5 B5 G4 B4 */
+		     "punpckhbw %%mm3, %%mm5;" /* 00 R7 00 R6 00 R5 00 R4 */
 
-		     "punpckhbw %%mm2, %%mm4;" /* G7 B7 G6 B6 G5 B5 G4 B4 */ 
-		     "punpckhbw %%mm3, %%mm5;" /* 00 R7 00 R6 00 R5 00 R4 */ 
-				
-		     "punpcklwd %%mm5, %%mm4;" /* 00 R5 B5 G5 00 R4 B4 G4 */ 
-		     MOVNTQ " %%mm4, 16 (%3);" /* Store ARGB5 ARGB4 */ 
+		     "punpcklwd %%mm5, %%mm4;" /* 00 R5 B5 G5 00 R4 B4 G4 */
+		     MOVNTQ " %%mm4, 16 (%3);" /* Store ARGB5 ARGB4 */
 
-		     "movq %%mm0, %%mm4;" /* B7 B6 B5 B4 B3 B2 B1 B0 */ 
-		     "punpckhbw %%mm2, %%mm4;" /* G7 B7 G6 B6 G5 B5 G4 B4 */ 
+		     "movq %%mm0, %%mm4;" /* B7 B6 B5 B4 B3 B2 B1 B0 */
+		     "punpckhbw %%mm2, %%mm4;" /* G7 B7 G6 B6 G5 B5 G4 B4 */
 
-		     "punpckhwd %%mm5, %%mm4;" /* 00 R7 G7 B7 00 R6 B6 G6 */ 
-		     MOVNTQ " %%mm4, 24 (%3);" /* Store ARGB7 ARGB6 */ 
+		     "punpckhwd %%mm5, %%mm4;" /* 00 R7 G7 B7 00 R6 B6 G6 */
+		     MOVNTQ " %%mm4, 24 (%3);" /* Store ARGB7 ARGB6 */
 
-		     "movd 4 (%1), %%mm0;" /* Load 4 Cb 00 00 00 00 u3 u2 u1 u0 */ 
-		     "movd 4 (%2), %%mm1;" /* Load 4 Cr 00 00 00 00 v3 v2 v1 v0 */ 
+		     "movd 4 (%1), %%mm0;" /* Load 4 Cb 00 00 00 00 u3 u2 u1 u0 */
+		     "movd 4 (%2), %%mm1;" /* Load 4 Cr 00 00 00 00 v3 v2 v1 v0 */
 
-		     "pxor %%mm4, %%mm4;" /* zero mm4 */ 
-		     "movq 8 (%0), %%mm6;" /* Load 8 Y Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0 */ 
+		     "pxor %%mm4, %%mm4;" /* zero mm4 */
+		     "movq 8 (%0), %%mm6;" /* Load 8 Y Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0 */
 
 		     : : "r" (_py), "r" (_pu), "r" (_pv), "r" (_image));
 
@@ -397,6 +436,7 @@ yuv2rgb_fun yuv2rgb_init_mmx (int bpp, int mode)
 {
 //    if (bpp == 15 || bpp == 16) {
     if (bpp == 16 && mode == MODE_RGB) return yuv420_rgb16_mmx;
+    if (bpp == 24 && mode == MODE_RGB) return yuv420_rgb24_mmx;
     if (bpp == 32 && mode == MODE_RGB) return yuv420_argb32_mmx;
     return NULL; // Fallback to C.
 }