doc: NEWS: adjust 'wc -l' aarch64 benchmark after recent commit

After commit e0190a9d1 (wc: improve aarch64 Neon optimization for
'wc -l', 2026-03-09), on a Ampere eMAG machine:

    $ yes | head -n 10000000000 > input
    $ (time ./src/wc -l input)
    10000000000 input

    real	0m3.447s
    user	0m1.533s
    sys	0m1.913s
    $ (export GLIBC_TUNABLES='glibc.cpu.hwcaps=-ASIMD,-AVX2,-AVX512F'; \
       time ./src/wc -l input)
    10000000000 input

    real	0m15.758s
    user	0m14.039s
    sys	0m1.720s

* NEWS: Mention the improved benchmark.

diff --git a/NEWS b/NEWS
index a42670f8c6ebc1e0efa2c966a769eade34c0479e..97ed7e721b08b376deea412e602e6a670a21d942 100644 (file)
--- a/NEWS
+++ b/NEWS
@@ -51,8 +51,8 @@ GNU coreutils NEWS                                    -*- outline -*-
   'shuf -i' now operates up to two times faster on systems with unlocked stdio
   functions.
 
-  'wc -l' now operates up to three times faster on hosts that support Neon
-  instructions.
+  'wc -l' now operates up to four and a half times faster on hosts that support
+  Neon instructions.
 
   'yes' now uses zero-copy I/O on Linux to significantly increase throughput.
   E.g., increases from 12GiB/s to 175GiB/s were seen on some systems.