[PATCH] x86_64: Prefer EVEX512 code-path on AMD Zen5 CPUs

[Sajan Karumanchi]

This patch introduces a performance optimization for AMD Zen5 processors
by preferring 512-bit EVEX variants of string and memory functions when
AVX-512 is available.

Background
----------
Current Glibc's IFUNC dispatch mechanism selects 256-bit EVEX
variants for several string and memory functions on CPUs with AVX-512
support, regardless of their 512-bit execution capabilities. This
leaves significant performance potential untapped on Zen5, which
benefits substantially from using the full 512-bit EVEX512 variants.

Implementation
--------------
The patch introduces a new synthetic CPU feature flag 'Prefer_EVEX512'
that is automatically enabled on Zen5 CPUs when AVX-512F, AVX-512VL,
and AVX-512BW are supported. This flag influences IFUNC dispatch to
select EVEX512 implementations over EVEX (256-bit) variants.

Performance Results
-------------------
Benchmarks conducted on AMD Zen5 hardware demonstrate significant
performance improvements across all affected functions:

Function    Baseline   Patched    Avg         Avg        Avg      Max
            Variant    Variant    Baseline    Patched    Change   Improve
                                  (ns)        (ns)       %        %
------------+----------+----------+-----------+----------+--------+--------
STRCHR      evex       evex512    16.408      12.293     25.08%   37.69%
STRLEN      evex       evex512    16.862      11.436     32.18%   56.74%
STRNLEN     evex       evex512    18.493      11.762     36.40%   64.40%
STRRCHR     evex       evex512    15.154      10.874     28.24%   44.38%
STRCHRNUL   evex       evex512    16.464      12.605     23.44%   45.56%
MEMCHR      evex       evex512    9.984       8.268      17.19%   39.99%

Key findings:
- Average performance improvements range from 17% to 36%
- Maximum improvements reach up to 64% in some test cases

Testing
-------
- All glibc string and memory tests pass on Zen5 hardware
- IFUNC selection verified to correctly choose evex512 variants
- Backward compatibility maintained for non-Zen5 AMD processors
- No impact on Intel or other x86_64 processors

Affected Functions
------------------
- strchr
- strlen
- strnlen
- strrchr
- strchrnul
- memchr

The patch adds a new ifunc-evex512.h header for common IFUNC selection
logic and updates the affected functions to use it.

Sajan Karumanchi (1):
  x86_64: Prefer EVEX512 code-path on AMD Zen5 CPUs

 sysdeps/x86/cpu-features.c                        | 11 ++++
 sysdeps/x86/include/cpu-features-preferred_feature_index_1.def |  1 +
 sysdeps/x86_64/multiarch/ifunc-evex512.h          | 60 +++++++++++++++++++
 sysdeps/x86_64/multiarch/ifunc-impl-list.c        | 28 +++++++--
 sysdeps/x86_64/multiarch/memchr.c                 | 48 ++++++++++++++-
 sysdeps/x86_64/multiarch/strchr.c                 |  8 ++-
 sysdeps/x86_64/multiarch/strchrnul.c              |  4 +-
 sysdeps/x86_64/multiarch/strlen.c                 |  2 +-
 sysdeps/x86_64/multiarch/strnlen.c                |  2 +-
 sysdeps/x86_64/multiarch/strrchr.c                |  2 +-
 10 files changed, 151 insertions(+), 15 deletions(-)
 create mode 100644 sysdeps/x86_64/multiarch/ifunc-evex512.h