The MX/transpose compute ops are validated by dedicated testbenches in test/backend/ (directed corners, random stall campaigns, throughput gates, negative fences), all on idma_backend_rw_axi only — the generic test/tpl/tb_idma_backend.sv.tpl job flow does not know about compute. Extending the job schema with a compute op/format field and hooking the DPI-C goldens (test/golden/) into the generated TB's checker would give every topology byte-exact compute coverage through the existing harness, with no new TB files.
This is the stated path (see the #170 known-scope notes) to validate compute on non-AXI ports and lift the ComputeMxSrcProtocol/ComputeMxDstProtocol fences (OBI-write is the natural first candidate — its write ready is already per-beat). Once landed, the directed byte-exact cases can migrate into jobs and the dedicated TBs slim down to the checks the job model cannot express (utilization measurement, stall-shim/interlock campaigns, assert-firing negative tests). Best done now that the compute programming model is frozen by #170.
The MX/transpose compute ops are validated by dedicated testbenches in
test/backend/(directed corners, random stall campaigns, throughput gates, negative fences), all onidma_backend_rw_axionly — the generictest/tpl/tb_idma_backend.sv.tpljob flow does not know about compute. Extending the job schema with a compute op/format field and hooking the DPI-C goldens (test/golden/) into the generated TB's checker would give every topology byte-exact compute coverage through the existing harness, with no new TB files.This is the stated path (see the #170 known-scope notes) to validate compute on non-AXI ports and lift the
ComputeMxSrcProtocol/ComputeMxDstProtocolfences (OBI-write is the natural first candidate — its write ready is already per-beat). Once landed, the directed byte-exact cases can migrate into jobs and the dedicated TBs slim down to the checks the job model cannot express (utilization measurement, stall-shim/interlock campaigns, assert-firing negative tests). Best done now that the compute programming model is frozen by #170.