Fix-Point opened a new pull request, #4040:
URL: https://github.com/apache/nuttx-apps/pull/4040

   ## Summary
   
   This PR shrinks the flash footprint of the `ostest` and `dd` test
   utilities to keep the 128 KiB G431 boards linking alongside
   https://github.com/apache/nuttx/pull/21491.
   
   This is entirely about flash size, not behaviour. The math32 changes in
   apache/nuttx#21491 replace constant-divisor 64-bit divisions with inline
   multiply-shift magic, which is a performance win but grows the image by
   ~256 bytes on 32-bit targets. `b-g431b-esc1` and `nucleo-g431kb` already
   have only 127/55 bytes of flash free on master, so #21491 pushes them
   over the edge (the `Linux (arm-stm32l-u-g)` CI job reports `region
   'flash' overflowed by 132 bytes` and `by 188 bytes`). To make room
   without touching the boards' configuration, this PR frees flash in the
   test utilities:
   
   1. **Narrow the wdog delay math to 32 bits** (`testing/ostest/wdog.c`):
      every delay in the test stays below 2^32 ns, so keeping the division
      and modulo 32-bit avoids pulling the 64-bit division library into the
      image.
   
   2. **Compute the dd copy throughput with integer arithmetic**
      (`system/dd/dd_main.c`): the throughput was computed in `double`,
      which pulls the whole soft-float cluster (`__aeabi_dadd` and friends,
      ~2 KiB) into the image. Integer arithmetic --
      `(total * USEC_PER_SEC / 1024) / elapsed`, with the division by 1024
      a shift -- keeps one 64-bit division against the division library the
      image already contains and drops the soft-float cluster.
   
   Measured on `nucleo-g431kb:nsh`, the image shrinks by **2052 bytes**
   (128536 -> 126484), leaving headroom over the 188-byte shortfall; the
   same two commits clear `b-g431b-esc1:nsh` (132 bytes).
   
   The 64-bit division library is still linked after this PR (the unpatched
   `gmtime`/`printf` sites keep it); removing it entirely is a separate,
   larger change and is deliberately out of scope here.
   
   ## Impact
   
   - User impact: none beyond the test utilities. `dd` prints its KB/s
     line with integer rounding (off by at most ~1 KB/s versus the old
     double math) and `ostest`'s wdog timings are unchanged (the delays fit
     in 32 bits).
   - Build: no `CONFIG_*` change, no build-system change, no new header.
   - Hardware: targets the small-flash STM32G4 boards; no functional change
     on any other target.
   - Documentation: none needed.
   - Security: no security impact.
   - Compatibility: `dd`'s output format is unchanged.
   
   ## Testing
   
   - Host: Ubuntu 22.04 x86-64, `arm-none-eabi-gcc` 10.3.1.
   - Board/Configuration: `b-g431b-esc1:nsh` (Make), `nucleo-g431kb:nsh`
     (CMake).
   - Build: clean on both, no new warnings.
   - Size (`nucleo-g431kb:nsh`, text+data): 128536 -> 126484, **-2052 B**.
   
   ```
   $ arm-none-eabi-size build/nuttx   # after
     125548    936   8784  135268  21064  build/nuttx
   ```
   
   - Run: `qemu-armv7a` boots; `ostest` exits 0 (wdog, spinlock and POSIX
     timer cases pass); `dd` prints a KB/s line consistent with the integer
     formula. No real G431 hardware available -- the run is QEMU only.
   - `tools/checkpatch.sh`: passes on both commits.
   


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]

Reply via email to