maybe you knew about this already? anyways I don't think RTBF_STI_GPU_BLOCK gets set correctly. meson.build was setting RTBF_STI_GPU_BLOC, so RTBF_STI_GPU_BLOCK got its default value of 16 even though RTBF_STI_LENGTH was correctly set to 21. so I think we're dropping samples since we only sum 16 of the 21...before I changed this the run_beamformer() time was 6 ms, now it's 12 ms.