On 02/28/2011 09:42 AM, Christian König wrote:
Am Montag, den 28.02.2011, 08:32 -0800 schrieb Brian Paul:
....
diff --git a/src/mesa/state_tracker/st_draw.c b/src/mesa/state_tracker/st_draw.c
....
min_index = MIN2(min_index, prims[i].start);
max_index = MAX2(max_index, prims[i].start + prims[i].count - 1);
+ max_index = MAX2(max_index, prims[i].num_instances);
....
I don't understand that last bit. If we're doing regular,
non-instanced drawing, prims[i].num_instances will be 1. So max_index
will always be one. That doesn't seem right.
The maximum check is additional to the existing calculation, not a
replacement. So max_index only gets increased if the number of instances
to draw is higher than the number of vertices.
My initial comment was wrong- I was thinking "min" instead of "max".
Ugh. Still, this line seems weird. The code which allocates the
vertex output buffer should check for instanced drawing and adjust the
size accordingly. max_index here refers to the incoming vertex data
and probably shouldn't be manipulated just to allocate the right size
of output buffer.
But indeed that's just a workaround to make piglits
draw-instanced-divisor test happy. max_index seems to be used to
calculate the size of the buffer needed to hold the vertex data, but in
the case of instanced drawing this isn't correct.
It wouldn't surprise me a bit if there's issues like this in a few
places. Wherever we try to compute a max index or figure out the the
size of a destination buffer from the vertex info we need to consider
how/if the instance count and divisor factors into it.
@Marek:
Since you are the author of the code maybe you could help here out a
bit. Where is max_index used to calculate the buffer size? I couldn't
find a call to the buffer creation function. The size calculation
somewhere around this call needs to be adopted to the case when we draw
more instances than vertices or indexes.
Christian.
-Brian
_______________________________________________
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/mesa-dev
