Potential xs3_filter_fir_s32 buffer overrun issue

[keithm-xmos]

It has been observed when the number of filter taps is 1024 that a buffer may be overrun. This was observed in an application where other threads were running and an overrun corrupted some data causing an exception. It is not known with 100% certainty that the xs3_filter_fir_s32 function is the culprit. Here is the evidence:

• We do not observe any memory corruption when calling the C reference version of xs3_filter_fir_s32
• The input data and/or the filter coeffs can be all zeros. The corruption is not a function of the input values.
• If we reduce the the number of taps to 896 we do not observe the corruption. Increase to 900+ and we do observe the corruption.

(mention @QuinnWang)

[astewart-xmos]

Extensive debugging has gone into this between myself and @keithm-xmos. There are a couple pieces of evidence that point to xs3_filter_fir_s32(), but no smoking gun has been found, and I don't see any way in which xs3_filter_fir_s32() could be clobbering anyone else's memory.

Unless new evidence is found pointing to xs3_filter_fir_s32(), I am going to assume it is not the culprit.

[keithm-xmos]

The source of the buffer overrun has been found and lib_xcore_math is no longer a suspect in the case.

Those interested can follow this issue: https://github.com/xmos/xscope_fileio/issues/52