For the quantized_bits quantizer, this seems to result in a duplication of the inputs, for example:

void myproject(
    result_t quantizer_input[1],
    result_t quantizer_input[1]
) { ...

This doesn't compile due to redefinition of the variable. Haven't dug deep enough to actually find the cause of the problem.

For the quantized_bits quantizer, this seems to result in a duplication of the inputs, for example:

void myproject(
    result_t quantizer_input[1],
    result_t quantizer_input[1]
) { ...

This doesn't compile due to redefinition of the variable. Haven't dug deep enough to actually find the cause of the problem.

This test converts an orphaned quantizer. The fuse quantizer flow merged it with the input layer by overridding the precision, and there is 0 layer left in the model and hence inp=out, the duplicated name.

Made fuse only happen is bit-exact is enabled, though in this case I think removing this test would also make sense.

Now I see that in the hgq_layer tests all test_syn_players tests with io_steam fail with

NotImplementedError: Heterogenous quantization for activations is only supported with IOType=io_parallel

Can you have a look at that as well?

Trigger cond was reversed

Last remaining test failure is already fixed by #1377. So this is good to go now.