<tomeu>
thanks, he is trying with annother version of tflite
pcercuei has joined #etnaviv
<tomeu>
austriancoder: I have a kernel that when executed by etnaviv is giving me a result different to that of galcore, even when all inputs are the same
<tomeu>
it is a quantize operation, so it is doing floating point operations
<tomeu>
the difference between individual values is small, looks like some rounding error, but my command stream sets the same rounding method as the blob
<tomeu>
wonder if it could be related to the flop reset mentioned in the kernel sources
<szemzoa>
with tensorflow 2.13 the preparedelegate() problem is gone