pystrhex: Use configure.ac detection for SIMD support

Replace the hardcoded arch/compiler preprocessor checks with a
configure.ac probe that tests __builtin_shufflevector with 128-bit
vectors directly. This follows the existing HAVE_BUILTIN_ATOMIC
pattern and lets configure determine what the compiler actually
supports rather than maintaining a list of arch/compiler combos.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Author: gpshead

Python/pystrhex.c

@@ -41,15 +41,7 @@ _Py_hexlify_scalar(const unsigned char *src, Py_UCS1 *dst, Py_ssize_t len)
 
    Speeding up the 16-64 byte cases fits nicely with md5 through sha512.
 */
-#if (defined(__x86_64__) || defined(__aarch64__) || \
-     (defined(__arm__) && defined(__ARM_NEON))) && \
-    (defined(__clang__) || (defined(__GNUC__) && __GNUC__ >= 12))
-#  define PY_HEXLIFY_CAN_COMPILE_SIMD 1
-#else
-#  define PY_HEXLIFY_CAN_COMPILE_SIMD 0
-#endif
-
-#if PY_HEXLIFY_CAN_COMPILE_SIMD
+#ifdef HAVE_BUILTIN_SHUFFLEVECTOR_AND_IT_IS_WORTH_USING
 
 /* 128-bit vector of 16 unsigned bytes */
 typedef unsigned char v16u8 __attribute__((vector_size(16)));
@@ -122,7 +114,7 @@ _Py_hexlify_simd(const unsigned char *src, Py_UCS1 *dst, Py_ssize_t len)
     _Py_hexlify_scalar(src + i, dst, len - i);
 }
 
-#endif /* PY_HEXLIFY_CAN_COMPILE_SIMD */
+#endif /* HAVE_BUILTIN_SHUFFLEVECTOR_AND_IT_IS_WORTH_USING */
 
 static PyObject *_Py_strhex_impl(const char* argbuf, const Py_ssize_t arglen,
                                  PyObject* sep, int bytes_per_sep_group,
@@ -202,7 +194,7 @@ static PyObject *_Py_strhex_impl(const char* argbuf, const Py_ssize_t arglen,
     unsigned char c;
 
     if (bytes_per_sep_group == 0) {
-#if PY_HEXLIFY_CAN_COMPILE_SIMD
+#ifdef HAVE_BUILTIN_SHUFFLEVECTOR_AND_IT_IS_WORTH_USING
         if (arglen >= 16) {
             // little vector units go brrrr...
             _Py_hexlify_simd((const unsigned char *)argbuf, retbuf, arglen);

configure

@@ -5014,6 +5014,40 @@ AS_VAR_IF([ac_cv_builtin_atomic], [yes], [
     AC_DEFINE(HAVE_BUILTIN_ATOMIC, 1, [Has builtin __atomic_load_n() and __atomic_store_n() functions])
 ])
 
+# Check for __builtin_shufflevector with 128-bit vector support on an
+# architecture where it compiles to worthwhile native SIMD instructions.
+# Used for SIMD-accelerated bytes.hex() in Python/pystrhex.c.
+AC_CACHE_CHECK([for __builtin_shufflevector], [ac_cv_builtin_shufflevector], [
+AC_LINK_IFELSE([
+  AC_LANG_PROGRAM([[
+    /* __builtin_shufflevector is available on many platforms, but 128-bit
+       vector code is only worthwhile on architectures with native SIMD:
+       x86-64 (SSE2, always available), ARM64 (NEON, always available),
+       or ARM32 when NEON is enabled via compiler flags (e.g. -march=native
+       on RPi3+).  On ARM32 without NEON (e.g. armv6 builds), the compiler
+       has the builtin but generates slow scalar code instead. */
+    #if !defined(__x86_64__) && !defined(__aarch64__) && \
+        !(defined(__arm__) && defined(__ARM_NEON))
+    #  error "128-bit vector SIMD not worthwhile on this architecture"
+    #endif
+    typedef unsigned char v16u8 __attribute__((vector_size(16)));
+  ]], [[
+    v16u8 a = {0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15};
+    v16u8 b = {16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31};
+    v16u8 c = __builtin_shufflevector(a, b,
+        0, 16, 1, 17, 2, 18, 3, 19, 4, 20, 5, 21, 6, 22, 7, 23);
+    (void)c;
+    return 0;
+  ]])
+],[ac_cv_builtin_shufflevector=yes],[ac_cv_builtin_shufflevector=no])
+])
+
+AS_VAR_IF([ac_cv_builtin_shufflevector], [yes], [
+    AC_DEFINE([HAVE_BUILTIN_SHUFFLEVECTOR_AND_IT_IS_WORTH_USING], [1],
+              [Define if compiler supports __builtin_shufflevector with
+               128-bit vectors on an architecture with native SIMD])
+])
+
 # --with-mimalloc
 AC_MSG_CHECKING([for --with-mimalloc])
 AC_ARG_WITH([mimalloc],

configure.ac

@@ -126,6 +126,10 @@
 /* Has builtin __atomic_load_n() and __atomic_store_n() functions */
 #undef HAVE_BUILTIN_ATOMIC
 
+/* Define if compiler supports __builtin_shufflevector with 128-bit vectors on
+   an architecture with native SIMD */
+#undef HAVE_BUILTIN_SHUFFLEVECTOR_AND_IT_IS_WORTH_USING
+
 /* Define to 1 if you have the <bzlib.h> header file. */
 #undef HAVE_BZLIB_H