[directfb-dev] Re: patch stretchblitting YUV in software

[Claudio \"KLaN\" Ciccani]

Ville Syrj�l� wrote:

On Thu, Jun 10, 2004 at 02:56:12PM +0200, Claudio KLaN Ciccani wrote:

If you don't have objections, I'm going to apply this patch (see below) to DirectFB/src/gfx/generic/generic.c.
It enables stretchblitting YUV surfaces and requires only small changes.

See my comments below.

<snip>

@@ -4623,20 +4623,24 @@

    int f;
    int i = 0;
+     int h;

    D_ASSERT( gfxs != NULL );

    CHECK_PIPELINE();

- gfxs->length = drect->w;
gfxs->SperD = (srect->w << 16) / drect->w;
+ gfxs->length = drect->w;
+ if (gfxs->src_format == DSPF_YUY2 || gfxs->src_format == DSPF_UYVY)
+ gfxs->length /= 2;

This will probably look quite bad. But I suppose it's better than nothing...

Yes, the image will look striped, but I don't know other ways to stretch
yuv420 with acceptable results.

    f = (srect->h << 16) / drect->h;
+     h = drect->h;

    Aop_xy( gfxs, gfxs->dst_org, drect->x, drect->y, gfxs->dst_pitch );
    Bop_xy( gfxs, gfxs->src_org, srect->x, srect->y, gfxs->src_pitch );

-     while (drect->h--) {
+     while (h--) {
         RUN_PIPELINE();

Aop_next( gfxs, gfxs->dst_pitch );
@@ -4648,6 +4652,39 @@
Bop_next( gfxs, gfxs->src_pitch );
}
}
+
+ /* scale other planes */
+ if (gfxs->src_format == DSPF_YV12 || gfxs->src_format == DSPF_I420) {
+ void *splane1 = gfxs->src_org + (gfxs->src_pitch * gfxs->src_height)
+ + ((gfxs->src_pitch * srect->y) / 4) + (srect->x / 2);
+ void *splane2 = splane1 + ((gfxs->src_pitch * gfxs->src_height) / 4);
+ void *dplane1 = gfxs->dst_org + (gfxs->dst_pitch * gfxs->dst_height)
+ + ((gfxs->dst_pitch * drect->y) / 4) + (drect->x / 2);
+ void *dplane2 = dplane1 + ((gfxs->dst_pitch * gfxs->dst_height) / 4);
+ int spitch = gfxs->src_pitch / 2;
+ int dpitch = gfxs->dst_pitch / 2;
+
+ gfxs->length = drect->w / 2;
+ h = drect->h / 2;
+ i = 0;
+
+ while (h--) {
+ gfxs->Bop = splane1 + (i >> 16) * spitch;
+ gfxs->Aop = dplane1;

You should use ?op_xy() and ?op_next() to do this stuff.

I can use ?op_xy but not ?op_next because I'm processing two
planes at once to save time.

+
+ RUN_PIPELINE();
+
+ gfxs->Bop = splane2 + (i >> 16) * spitch;
+ gfxs->Aop = dplane2;
+
+ RUN_PIPELINE();
+
+ dplane1 += dpitch;
+ dplane2 += dpitch;

I think blitting the planes one at a time would be better.

It would be only slower, according to me.

+
+ i += f;
+ }
+ }
}

Regards,
Claudio