Clarify Armv9 Support

[jserv]

To consolidate Armv9 support, I conducted a review of current architecture detection logic with a specific focus on Armv9-A. The objective was to determine if existing preprocessor conditions require modification to support the newer standard.

1. Armv9-A (AArch64) is already fully supported by the current logic; no changes are required.
2. Armv8-A (AArch32) detection is currently too restrictive. A minor adjustment is proposed to ensure forward compatibility with updated ACLE specifications (Armv8.1+).

The findings are as follows:

• Execution State: Armv9-A primarily targets the AArch64 execution state. Modern Armv9 implementations (e.g., Cortex-A715, X3, X4) dropped AArch32 support entirely, while others treat it as optional/legacy.
• Compiler Behavior: When targeting Armv9-A, standard toolchains do not define __arm__. Instead, they define __aarch64__ or __arm64__, mirroring Armv8-A behavior.

Consequently, the existing detection logic in SSE2NEON remains valid for Armv9-A:

#if defined(__aarch64__) || defined(__arm64__)
    /* Correctly captures Armv8-A and Armv9-A in 64-bit mode */
#endif

While the AArch64 logic is robust, a limitation was identified in the Legacy AArch32 detection logic.

According to the Arm C Language Extensions (ACLE), the __ARM_ARCH macro typically encodes the version as an integer. However, for minor versions beyond 8.0, the value scales (e.g., 8.1 becomes 801).

Architecture	__ARM_ARCH Value	Current Status in SSE2NEON
Armv8.0-A	8	✅ Detected
Armv8.1-A	801	❌ Ignored
Armv8.2-A	802	❌ Ignored
Armv9.x	9xx	N/A (Uses AArch64 path)

The current check is overly strict:

/* Current Logic */
__ARM_ARCH == 8

This condition implicitly excludes Armv8.1 through Armv8.6 AArch32 targets. While this does not impact Armv9 (which uses the AArch64 path), it is a compatibility gap for updated 32-bit toolchains.

To improve ACLE compliance and ensure forward compatibility for 32-bit builds, I propose relaxing the equality check to a range check. Proposed Change:

Replace:

__ARM_ARCH == 8

With:

__ARM_ARCH >= 8

The following diff illustrates the minimal changes required to implement this fix.

diff --git a/sse2neon.h b/sse2neon.h
index 1234567..abcdef0 100644
--- a/sse2neon.h
+++ b/sse2neon.h
@@ -45,7 +45,7 @@
 #elif SSE2NEON_ARCH_AARCH64
     /* AArch64 NEON path */
 
-#elif __ARM_ARCH == 8
+#elif __ARM_ARCH >= 8
 #  if !defined(__ARM_NEON) && !defined(__ARM_NEON__)
 #    error "You must enable NEON instructions (e.g. -mfpu=neon-fp-armv8) to use SSE2NEON."
 #  endif
@@ -220,7 +220,7 @@
 #include <math.h>
 
 /* Include arm_acle.h only for Armv8-A AArch32 */
-#if (!SSE2NEON_ARCH_AARCH64) && (__ARM_ARCH == 8)
+#if (!SSE2NEON_ARCH_AARCH64) && (__ARM_ARCH >= 8)
 #  if defined __has_include && __has_include(<arm_acle.h>)
 #    include <arm_acle.h>
 #  endif