the explanation i gave was not good at all .......
i need to process a lightmap (like a post process but in cpu).
for each texel, i need to recover the model position that "generated" it (the triangle position that this texel will be "pasted" ).
so i did:
Rasterized all the triangles and discovered each UV, world position and normal for all rasterized "pixel", then i add it in a associative array (uv -> world position/Normal)
then when processing th lightmap (image processing ..) i just used this map to get the properties i needed !!!
is this a reasonable way ? (for me it looked like a hack)
The second part
Just extending the UV to the tessellated triangles fixed resolved. Thanks