Skip to content

Fix DivSqrt lanes synchronization

Eclipse Webmaster requested to merge fix/divsqrt_synchronization into develop

Created by: lucabertaccini

This PR fixes the synchronization scheme for DivSqrt lanes. In the case of CVFPU instances supporting SIMD operations, such a scheme made the lanes wait for the others to complete their calculations even when the operation was scalar (i.e. EnableVectors = 1'b1 and vectorial_op_i = 1'b0), thus never completing the scalar computation.

With this PR, the synchronization scheme is modified, deciding whether to synchronize with other lanes based on whether the current operation is vectorial (this guarantees correct execution also in the case of interleaved scalar/SIMD DivSqrt executions).

This PR targets another minor modification: some additional unused signals in specific configurations of fpnew_opgroup_multifmt_slice are tied to '0.

Merge request reports

Loading