rixed@happyleptic.org
 发送至 loongson-dev
	
A patch for siliconmotion 1.7.3 is available in the bug trackers
at freedesktop.org, but a better one will be available monday.
Still, it lacks the most interresting part : the MMX pack function,
which should not be implemented there but probably either directly
in Xv or in pixman (and make Xv use pixman).

So for now the simpliest is to fix these manually.
For the impatient I attach a patch against SM 1.7.3
With it, 16bpp and AccelMethod = "XAA" works quite well.

Just an ugly hack really.
diff -ur orig/src/smi_video.c mod/src/smi_video.c
--- orig/src/smi_video.c	2010-02-05 13:11:18.000000000 +0100
+++ mod/src/smi_video.c	2010-02-05 13:20:36.000000000 +0100
@@ -276,6 +276,7 @@
     XVIMAGE_YUY2,
     XVIMAGE_YV12,
     XVIMAGE_I420,
+    XVIMAGE_UYVY,
     {
 	FOURCC_RV15,			/* id				*/
 	XvRGB,				/* type				*/
@@ -1462,6 +1463,117 @@
     LEAVE();
 }
 
+static void myXVCopyYUV12ToPacked(const unsigned char *srcy, const unsigned char *srcv, const unsigned char *srcu,
+		unsigned char *dst, int srcPitchy, int srcPitchuv, int dstPitch, int h, int w)
+{
+	int i, j;
+	unsigned char const *y, *u, *v;
+	int dstinc, yinc, uinc, vinc;
+
+	y = srcy;
+	u = srcu;
+	v = srcv;
+
+	dstinc = dstPitch - 2*w;
+	yinc = srcPitchy - w;
+	uinc = srcPitchuv - w/2;
+	vinc = srcPitchuv - w/2;
+
+	for (i = 0; i < h; i++) {
+		asm (
+//			".set arch=loongson2f\n\t"
+			".set noreorder\n\t"
+			"move $8, %8 \n\t"
+			"1: \n\t"
+			"beqz $8, 2f \n\t"
+			"xor $f0, $f0, $f0 \n\t"
+			"ldc1 $f4, (%0) \n\t"
+			"punpcklbh $f2, $f4, $f0 \n\t"
+			"punpckhbh $f4, $f4, $f0 \n\t"
+			"ldc1 $f16, 8(%0) \n\t"
+			"punpcklbh $f14, $f16, $f0 \n\t"
+			"punpckhbh $f16, $f16, $f0 \n\t"
+			
+			"lwc1 $f8, (%1) \n\t"
+			"lwc1 $f12, (%2) \n\t"
+			"punpcklbh $f8, $f8, $f12 \n\t"
+			"punpcklbh $f6, $f0, $f8 \n\t"
+			"punpckhbh $f8, $f0, $f8 \n\t"
+			"lwc1 $f18, 4(%1) \n\t"
+			"lwc1 $f12, 4(%2) \n\t"
+			"punpcklbh $f18, $f18, $f12 \n\t"
+			"punpcklbh $f10, $f0, $f18 \n\t"
+			"punpckhbh $f12, $f0, $f18 \n\t"
+
+			"or $f2, $f2, $f6 \n\t"
+			"or $f4, $f4, $f8 \n\t"
+			"or $f14, $f14, $f10 \n\t"
+			"or $f16, $f16, $f12 \n\t"
+
+			"sdc1 $f2, (%3) \n\t"
+			"sdc1 $f4, 8(%3) \n\t"
+			"add %0, 16 \n\t"
+			"add %1, 8 \n\t"
+			"add %2, 8 \n\t"
+			"sdc1 $f14, 0x10(%3) \n\t"
+			"sdc1 $f16, 0x18(%3) \n\t"
+			"add $8, -1 \n\t"
+			"b 1b \n\t"
+			"add %3, 32 \n\t"
+			"2: \n\t"
+			".set reorder\n\t"
+			: "=r" (y), "=r" (u), "=r" (v), "=r" (dst)
+			: "0" (y), "1" (u), "2" (v), "3" (dst), "r" (w>>4)
+			: "memory","$8"
+		);
+
+		asm (
+//			".set arch=loongson2f\n\t"
+			".set noreorder\n\t"
+			"move $8, %8 \n\t"
+			"1: \n\t"
+			"beqz $8, 2f \n\t"
+			"xor $f0, $f0, $f0 \n\t"
+			"ldc1 $f4, (%0) \n\t"
+			"punpcklbh $f2, $f4, $f0 \n\t"
+			"punpckhbh $f4, $f4, $f0 \n\t"
+			
+			"lwc1 $f8, (%1) \n\t"
+			"lwc1 $f12, (%2) \n\t"
+			"punpcklbh $f8, $f8, $f12 \n\t"
+			"punpcklbh $f6, $f0, $f8 \n\t"
+			"punpckhbh $f8, $f0, $f8 \n\t"
+
+			"or $f2, $f2, $f6 \n\t"
+			"or $f4, $f4, $f8 \n\t"
+
+			"sdc1 $f2, (%3) \n\t"
+			"sdc1 $f4, 8(%3) \n\t"
+			"add %0, 8 \n\t"
+			"add %1, 4 \n\t"
+			"add %2, 4 \n\t"
+			"add $8, -1 \n\t"
+			"b 1b \n\t"
+			"add %3, 16 \n\t"
+			"2:\n\t"
+			".set reorder\n\t"
+			: "=r" (y), "=r" (u), "=r" (v), "=r" (dst)
+			: "0" (y), "1" (u), "2" (v), "3" (dst), "r" ((w&0xf)/8)
+			: "memory","$8"
+		);
+
+		for (j = (w&7)/2; j; j--) {
+			*dst++ = *y++;
+			*dst++ = *u++;
+			*dst++ = *y++;
+			*dst++ = *v++;
+		}
+		y += yinc;
+		u = (i%2) ? (u + uinc): (u - w/2);
+		v = (i%2) ? (v + vinc): (v - w/2);
+		dst += dstinc;
+	}
+}
 
 static int
 SMI_PutImage(
@@ -1593,7 +1705,7 @@
 		offset3 = tmp;
 	    }
 	    nLines = ((((y2 + 0xffff) >> 16) + 1) & ~1) - top;
-	    xf86XVCopyYUV12ToPacked(buf + (top * srcPitch) + (left >> 1), 
+	    myXVCopyYUV12ToPacked(buf + (top * srcPitch) + (left >> 1), 
 				    buf + offset2, buf + offset3, dstStart,
 				    srcPitch, srcPitch2, dstPitch, nLines,
 				    nPixels);
@@ -1793,6 +1905,7 @@
     WRITE_VPR(pSmi, 0x1C, offset >> 3);
     WRITE_VPR(pSmi, 0x20, (pitch >> 3) | ((pitch >> 3) << 16));
     WRITE_VPR(pSmi, 0x24, (hstretch & 0xff00) | ((vstretch & 0xff00) >> 8));
+    WRITE_VPR(pSmi, 0x68, ((hstretch & 0xff) << 8) | (vstretch & 0xff));
     if (pSmi->Chipset == SMI_LYNXEMplus) {	/* This one can store additional precision */
 	WRITE_VPR(pSmi, 0x68, ((hstretch & 0xff) << 8) | (vstretch & 0xff));
     }