Albert, Sean and I tracked this down to samples_needed going negative in EngineBufferScaleLinear. The logic to subtract the remainder was subtracting from samples_needed such that it went negative. In r2407, two asserts in EBSL that verify that samples_needed is greater than 0 were disabled, so the problem propagated to the RAMAN once EBSL tried to read with a negative samples_needed.
Albert, Sean and I tracked this down to samples_needed going negative in EngineBufferSca leLinear. The logic to subtract the remainder was subtracting from samples_needed such that it went negative. In r2407, two asserts in EBSL that verify that samples_needed is greater than 0 were disabled, so the problem propagated to the RAMAN once EBSL tried to read with a negative samples_needed.