bench(string): add bench_string.c vs glibc/musl

System-headers-only harness timing the five L1 string functions
(strlen, strcmp, memcpy, memmove, memset) at 8- and 64-byte inputs.
Calls go through volatile function pointers to defeat builtin
substitution and LICM hoisting; correctness self-asserts route through
the same pointers. Prints FUNC@SIZE ns/call lines.

bench_string uses empty per-target CPPFLAGS (no -I include) so it links
against whichever libc --with-libc= selects; the bench target runs it
always and bench_vlibc only under BENCH_LINK_VLIBC. Both programs
relink when config.status changes, so a reconfigure cannot silently
reuse a binary from the previous libc.

Measured vs glibc 2.44 (median of 3 runs): mem* 1.0-3.3x,
strlen/strcmp 3.2-3.9x — word-at-a-time vs AVX2, all sane.

Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-openagent)

Co-authored-by: Sisyphus <[email protected]>

benchmarks/Makefile.am

@@ -5,23 +5,49 @@ AM_CFLAGS = @VLIBC_CFLAGS@
 
 # Built only on demand (via `make bench`), so `make all` does not need the
 # library to be built first.
-EXTRA_PROGRAMS = bench_vlibc
+EXTRA_PROGRAMS = bench_vlibc bench_string
 bench_vlibc_SOURCES = bench_vlibc.c
+bench_string_SOURCES = bench_string.c
+
+# bench_string is a system-headers-only translation unit (it times whichever
+# libc it is linked against), so it must NOT inherit AM_CPPFLAGS: -I include
+# would pull vlibc's self-contained <stddef.h>/<string.h> into the same TU as
+# the system <stdio.h>/<time.h> and double-define size_t/NULL/offsetof.
+# bench_vlibc keeps the inherited -I include.
+bench_string_CPPFLAGS =
 
 # By default benchmarks link against vlibc itself. --with-libc=glibc links the
 # same harness against the system glibc; --with-libc=musl builds it with
 # musl-gcc for the musl reference.
 if BENCH_LINK_VLIBC
 bench_vlibc_LDADD = ../libvlibc.la
+bench_string_LDADD = ../libvlibc.la
 else
 bench_vlibc_LDADD =
+bench_string_LDADD =
 endif
 
 if BENCH_LINK_MUSL
 CC = $(MUSL_CC)
 endif
 
-bench: bench_vlibc
+# Relink whenever the configuration changes (autogen.sh reruns configure and
+# rewrites config.status). --with-libc only affects the link (LDADD), and the
+# objects are compile-time identical, so make would otherwise reuse a stale
+# binary from the previous configuration and the wrong libc would be measured.
+EXTRA_bench_string_DEPENDENCIES = $(top_builddir)/config.status
+EXTRA_bench_vlibc_DEPENDENCIES = $(top_builddir)/config.status
+
+# bench_vlibc.c calls vlibc_version(), which exists only in vlibc — under
+# --with-libc=glibc/musl it cannot link, so it is built and run only in the
+# vlibc build. bench_string links either way and runs always.
+if BENCH_LINK_VLIBC
+bench: bench_vlibc bench_string
 	./bench_vlibc
+	./bench_string
+else
+bench: bench_string
+	./bench_string
+endif
 
 .PHONY: bench