diff --git a/.github/workflows/test.yml b/.github/workflows/test.yml
index cde4e62a..f8496d93 100644
--- a/.github/workflows/test.yml
+++ b/.github/workflows/test.yml
@@ -62,6 +62,8 @@ jobs:
             platform: "arm64"
           - runner: "ubuntu-22.04-arm"
             platform: "aarch64"
+          - runner: "windows-11-arm"
+            platform: "ARM64"
     steps:
       - name: Checkout
         uses: actions/checkout@v4
@@ -141,7 +143,7 @@ jobs:
     steps:
       - name: Checkout
         uses: actions/checkout@v4
-      - uses: uraimo/run-on-arch-action@v2
+      - uses: uraimo/run-on-arch-action@v3
         with:
           arch: ${{matrix.arch}}
           distro: alpine_latest
@@ -162,7 +164,7 @@ jobs:
     steps:
       - name: Checkout
         uses: actions/checkout@v4
-      - uses: uraimo/run-on-arch-action@v2
+      - uses: uraimo/run-on-arch-action@v3
         with:
           arch: ${{matrix.arch}}
           distro: alpine_latest
diff --git a/cmake/test-arch.c b/cmake/test-arch.c
index b438405e..a62c1443 100644
--- a/cmake/test-arch.c
+++ b/cmake/test-arch.c
@@ -12,7 +12,8 @@
 ////////////////////////////////////////////////////////////////////////////////
 
 // ARM 64-Bit
-#if defined(__aarch64__)
+#if defined(__aarch64__) \
+    || defined(_M_ARM64)
 #error ##arch=arm64##
 
 // ARM 32-Bit
diff --git a/lib/arch/neon64/codec.c b/lib/arch/neon64/codec.c
index 6b664b40..7ac1a8a7 100644
--- a/lib/arch/neon64/codec.c
+++ b/lib/arch/neon64/codec.c
@@ -8,13 +8,7 @@
 #include "config.h"
 #include "../../env.h"
 
-#ifdef __aarch64__
-#  if (defined(__ARM_NEON__) || defined(__ARM_NEON)) && HAVE_NEON64
-#    define BASE64_USE_NEON64
-#  endif
-#endif
-
-#ifdef BASE64_USE_NEON64
+#if HAVE_NEON64
 #include <arm_neon.h>
 
 // Only enable inline assembly on supported compilers.
@@ -66,7 +60,7 @@ load_64byte_table (const uint8_t *p)
 # include "enc_loop.c"
 #endif
 
-#endif	// BASE64_USE_NEON64
+#endif	// HAVE_NEON64
 
 // Stride size is so large on these NEON 64-bit functions
 // (48 bytes encode, 64 bytes decode) that we inline the
@@ -75,7 +69,7 @@ load_64byte_table (const uint8_t *p)
 void
 base64_stream_encode_neon64 BASE64_ENC_PARAMS
 {
-#ifdef BASE64_USE_NEON64
+#if HAVE_NEON64
 	#include "../generic/enc_head.c"
 	enc_loop_neon64(&s, &slen, &o, &olen);
 	enc_loop_generic_64(&s, &slen, &o, &olen);
@@ -88,7 +82,7 @@ base64_stream_encode_neon64 BASE64_ENC_PARAMS
 int
 base64_stream_decode_neon64 BASE64_DEC_PARAMS
 {
-#ifdef BASE64_USE_NEON64
+#if HAVE_NEON64
 	#include "../generic/dec_head.c"
 	dec_loop_neon64(&s, &slen, &o, &olen);
 	dec_loop_generic_32(&s, &slen, &o, &olen);
diff --git a/lib/arch/neon64/dec_loop.c b/lib/arch/neon64/dec_loop.c
index 48232f20..428e0651 100644
--- a/lib/arch/neon64/dec_loop.c
+++ b/lib/arch/neon64/dec_loop.c
@@ -100,10 +100,10 @@ dec_loop_neon64 (const uint8_t **s, size_t *slen, uint8_t **o, size_t *olen)
 
 		// Check for invalid input, any value larger than 63:
 		const uint8x16_t classified
-			= vcgtq_u8(str.val[0], vdupq_n_u8(63))
-			| vcgtq_u8(str.val[1], vdupq_n_u8(63))
-			| vcgtq_u8(str.val[2], vdupq_n_u8(63))
-			| vcgtq_u8(str.val[3], vdupq_n_u8(63));
+			= vorrq_u8(
+				vorrq_u8(vcgtq_u8(str.val[0], vdupq_n_u8(63)), vcgtq_u8(str.val[1], vdupq_n_u8(63))),
+				vorrq_u8(vcgtq_u8(str.val[2], vdupq_n_u8(63)), vcgtq_u8(str.val[3], vdupq_n_u8(63)))
+			);
 
 		// Check that all bits are zero:
 		if (vmaxvq_u8(classified) != 0U) {
@@ -111,9 +111,9 @@ dec_loop_neon64 (const uint8_t **s, size_t *slen, uint8_t **o, size_t *olen)
 		}
 
 		// Compress four bytes into three:
-		dec.val[0] = vshlq_n_u8(str.val[0], 2) | vshrq_n_u8(str.val[1], 4);
-		dec.val[1] = vshlq_n_u8(str.val[1], 4) | vshrq_n_u8(str.val[2], 2);
-		dec.val[2] = vshlq_n_u8(str.val[2], 6) | str.val[3];
+		dec.val[0] = vorrq_u8(vshlq_n_u8(str.val[0], 2), vshrq_n_u8(str.val[1], 4));
+		dec.val[1] = vorrq_u8(vshlq_n_u8(str.val[1], 4), vshrq_n_u8(str.val[2], 2));
+		dec.val[2] = vorrq_u8(vshlq_n_u8(str.val[2], 6), str.val[3]);
 
 		// Interleave and store decoded result:
 		vst3q_u8((uint8_t *) *o, dec);
diff --git a/lib/codec_choose.c b/lib/codec_choose.c
index a98b9472..8b5938dd 100644
--- a/lib/codec_choose.c
+++ b/lib/codec_choose.c
@@ -163,13 +163,13 @@ codec_choose_forced (struct codec *codec, int flags)
 static bool
 codec_choose_arm (struct codec *codec)
 {
-#if (defined(__ARM_NEON__) || defined(__ARM_NEON)) && ((defined(__aarch64__) && HAVE_NEON64) || HAVE_NEON32)
+#if HAVE_NEON64 || ((defined(__ARM_NEON__) || defined(__ARM_NEON)) && HAVE_NEON32)
 
 	// Unfortunately there is no portable way to check for NEON
 	// support at runtime from userland in the same way that x86
 	// has cpuid, so just stick to the compile-time configuration:
 
-	#if defined(__aarch64__) && HAVE_NEON64
+	#if HAVE_NEON64
 	codec->enc = base64_stream_encode_neon64;
 	codec->dec = base64_stream_decode_neon64;
 	#else