xiaoxiang781216 opened a new pull request, #3712:
URL: https://github.com/apache/nuttx-apps/pull/3712
## Summary
`apps/testing/libc/arch_libc` only exercised `strcpy()`, so the architecture
optimized implementations of every other string and memory routine were never
covered by the test suite. This series turns it into a full correctness and
speed harness for the whole string/memory family, which is the prerequisite
for
landing architecture optimized assembly with confidence.
Four commits:
1. **`Add tests for all string/memory functions.`**
Adds a correctness test (sweeping buffer alignment and transfer size) and
a
speed test (average cycles via `perf_gettime()`) for `memcpy`, `memmove`,
`memset`, `memcmp`, `memchr`, `strlen`, `strcmp`, `strchr`, `strncmp`,
`strnlen`, `strncpy`, `stpcpy`, `strcat` and `strrchr`. Each test is
selected by its own `CONFIG_TESTING_ARCH_LIBC_<FUNC>` option (default y)
so
a target can drop the ones it does not need.
2. **`Add strchrnul test and sweep size boundaries.`**
Adds `strchrnul` coverage, and sweeps alignment 0..7 together with the
boundary sizes {0, 1, 7, 8, 9, 15, 16, 17, 31, 32, 33, 63, 64, 65, 127,
128,
129, 255, 256, 257} in the scan function tests and in `memmove`. Those
sizes sit on the 8/16-byte chunk edges and on the sub-word tails, so
vectorized (NEON/MVE) and word-at-a-time implementations are stressed
exactly at their alignment and size boundaries. `memmove` is additionally
exercised across four overlap layouts: forward, backward, contained and
adjacent.
3. **`Fix out-of-bounds write in memmove test.`**
The adjacent overlap layout introduced by (2) started at a fixed
`g_buf1 + align + 64`, so the destination tail ran past `g_buf1` for the
largest swept sizes (AddressSanitizer reported a global-buffer-overflow).
Start the layout at `g_buf1 + align` instead.
4. **`Cover unaligned src/dst copy paths.`**
`strcpy`/`strncpy`/`stpcpy` applied the same offset to source and
destination, so the two pointers always shared the same word congruence
and
the byte prologue plus shift-merge path of optimized copy routines was
never
reached. Vary both offsets independently over 0..7. Also drops the
`ARCH_TOOLCHAIN_GNU` dependency from `TESTING_ARCH_LIBC`: the test only
uses
standard C string functions and `perf_gettime()`, with no GNU specific
construct, so it builds with non-GNU toolchains such as TASKING as well.
Commit 3 is a fix for commit 2 rather than a squash because the two commits
have different authors; each commit still builds and runs standalone
(see Testing).
## Impact
* Test code only. Nothing is built unless `CONFIG_TESTING_ARCH_LIBC`
(default n) is selected, so no existing board configuration or defconfig
changes, and no size impact on any shipped build.
* New Kconfig options
`TESTING_ARCH_LIBC_{MEMCHR,MEMCMP,MEMCPY,MEMMOVE,MEMSET,
STRCHR,STRCMP,STRCPY,STRLEN,STRNCMP,STRNLEN,STRNCPY,STPCPY,STRCAT,STRRCHR,
STRCHRNUL}`, all default y inside `TESTING_ARCH_LIBC`.
* Dropping the `ARCH_TOOLCHAIN_GNU` dependency only widens the set of
toolchains that may select the test; no existing configuration changes
behaviour.
* No user API/ABI, hardware, security or documentation impact.
## Testing
Host: Ubuntu 24.04 x86_64, gcc 13.3.0
Target: `sim:nsh` with `CONFIG_TESTING_ARCH_LIBC=y` (all 16 function options
enabled) and `CONFIG_TESTING_ARCH_LIBC_VERBOSE=y`
Every commit of the series was built and run standalone; no build warnings,
and
all enabled functions report `PASSED`:
| commit | functions PASSED | result |
| --- | --- | --- |
| 1 `Add tests for all string/memory functions.` | 15 | `arch_libc_test
Passed` |
| 2 `Add strchrnul test and sweep size boundaries.` | 16 | `arch_libc_test
Passed` |
| 3 `Fix out-of-bounds write in memmove test.` | 16 | `arch_libc_test
Passed` |
| 4 `Cover unaligned src/dst copy paths.` | 16 | `arch_libc_test Passed` |
A build with only `CONFIG_TESTING_ARCH_LIBC_STRCPY=y` (every other function
test disabled) was also checked to make sure the reduced configurations stay
warning free.
**Before** (`apache/master`, only `strcpy` is covered):
```
NuttShell (NSH) NuttX-10.4.0
nsh> arch_libctest
arch_libc_test_strcpy Test Passed
strcpy total(run 25 times) cpu cycles 1749
strcpy average cpu cycles 69
nsh> exit
```
**After** (this series):
```
NuttShell (NSH) NuttX-10.4.0
nsh> arch_libctest
Testing memcpy...
memcpy: PASSED
memcpy(128) avg cycles: 96
Testing memmove...
memmove: PASSED
memmove(128) avg cycles: 4
Testing memset...
memset: PASSED
memset(128) avg cycles: 53
Testing memcmp...
memcmp: PASSED
memcmp(128) avg cycles: 121
Testing memchr...
memchr: PASSED
memchr(128) avg cycles: 12
Testing strlen...
strlen: PASSED
strlen(128) avg cycles: 50
Testing strcmp...
strcmp: PASSED
strcmp(128) avg cycles: 133
Testing strcpy...
strcpy: PASSED
strcpy(128) avg cycles: 234
Testing strchr...
strchr: PASSED
strchr(128) avg cycles: 54
Testing strncmp...
strncmp: PASSED
strncmp(128) avg cycles: 469
Testing strnlen...
strnlen: PASSED
strnlen(128) avg cycles: 281
Testing strncpy...
strncpy: PASSED
strncpy(128) avg cycles: 142
Testing stpcpy...
stpcpy: PASSED
stpcpy(128) avg cycles: 68
Testing strcat...
strcat: PASSED
strcat(64) avg cycles: 37
Testing strrchr...
strrchr: PASSED
strrchr(128) avg cycles: 65
Testing strchrnul...
strchrnul: PASSED
strchrnul(128) avg cycles: 18
arch_libc_test Passed
nsh> exit
```
`tools/checkpatch.sh -c -u -m -g apache/master..HEAD` reports
`All checks pass.` for the series.
--
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]
