[PATCH 0/2] string: vectorize the strspn single-accept-character case

These two patches replace "[PATCH 16/20] alpha: add vectorized single-char
fast path for strspn" from the alpha series, which added a strspn to
sysdeps/alpha rather than improving the generic one.  Adhemerval was right
that this belongs in the generic implementation, and the result is both
faster and useful to every target that lacks an assembly strspn.

Patch 1 adds find_ne_all (), which searches for byte inequality alone.  The
existing find_zero_ne_all () also searches for a zero byte in X1, which the
strspn caller does not need: the accept character is not NUL there, so a NUL
byte already differs from it.  Dropping the zero test is worth 13-16% on an
EV68CB and 1.3-1.8x on x86_64 over using find_zero_ne_all () in patch 2, so
it is not just tidiness.  Each target that provides its own string-fza.h
needs its own version, since the mask conventions differ and a plain bitwise
inverse of find_eq_all () is only correct where the mask is a full byte.

Patch 2 rewrites the single-accept-character case as a word-at-a-time scan.
Speedup over the byte-at-a-time loop it replaces:

  length      8    32   128   512    2K    8K   32K
  EV68CB    1.4x  3.3x  4.5x  5.7x  6.4x  6.6x  6.6x
  i7-1370P  0.8x  1.6x  4.3x  3.6x  2.7x  2.8x  2.9x

The x86_64 numbers are for the generic C string-fza.h, which x86_64 itself
only reaches where SSE4.2 is unavailable; they stand in for the targets that
use the generic implementation for real.  Below 16 bytes it loses under half
a nanosecond per call, which I am happy to trade but can gate on a length
check if anyone objects.

Checked against strspn () for lengths 0 to 300 at every byte alignment on
x86_64, s390x, powerpc64, arm and alpha, covering the generic, cmpb, uqsub8
and cmpbge string-fza.h implementations.  The riscv orc.b one was built but
not run.  string/test-strspn, string/bug-strspn1 and string/test-strcspn
pass on x86_64.

Matt Turner (2):
  string: add find_ne_all to string-fza.h
  string: vectorize strspn single-accept-char case

 string/strspn.c                  | 27 ++++++++++++++++++++++++---
 sysdeps/alpha/string-fza.h       |  6 ++++++
 sysdeps/arm/armv6t2/string-fza.h |  8 ++++++++
 sysdeps/generic/string-fza.h     | 10 ++++++++++
 sysdeps/powerpc/string-fza.h     |  8 ++++++++
 sysdeps/riscv/string-fza.h       |  7 +++++++
 6 files changed, 63 insertions(+), 3 deletions(-)

-- 
2.54.0