Skip to content

Superseded Cortex-M7 FP smoke setup attempts

These are development attempts from Round 530, not the final platform result. The runner's per-target working artifacts were replaced by later retries before the failed attempts were copied, so this note reconstructs their outcome from the runner output; it is not a substitute for raw simulation evidence. The final measured result is under docs/coverage/renode-onboarding/metrics/stm32h743/.

  1. At 22:36 UTC, the firmware build used thumbv7em-none-eabi, which does not enable the FP register set. The assembler rejected vldmia, vmsr, vstmia, and vmrs with instruction requires: fp registers. No simulation ran.
  2. At 22:37 UTC, the build used thumbv7em-none-eabihf, but the smoke YAML still referenced the old soft-float output path. The simulator booted the prior image and reached its 4,000,000-instruction limit; the new FP UART assertion failed because that image had no FP test. The metric reported failure_stage: run_smoke and captured len=7.

The mapping and smoke input now both use the hard-float target. The next one-target run passed the FP frame, FPCAR, S0-S15/FPSCR restoration, and existing Cortex-M7 smoke assertions. These setup failures are not counted as platform measurements in the generated scoreboard.