To calculate the output width, the core combines data width, coefficient width, and bit growth (due to the natural gain of the filter).
This gives the full precision output width.
If the input width and output width are the same, some rounding mode will be applied to the output, and additional LSBs will be trimmed.
Normally the value will be attenuated, depending on how many bits are rounded.
To investigate further, you can add another FIR Compiler with full precision output and compare the output values.