From b93a91f4f0effc1252aed3738603067d3b4e783c Mon Sep 17 00:00:00 2001
From: Mathias Berchtold <mberchtold@gmail.com>
Date: Thu, 3 Sep 2026 05:04:18 +0200
Subject: [PATCH 1/2] chacha20poly1305: uses native unaligned 32-bit
 little-endian loads/stores

uses native unaligned 32-bit little-endian loads and stores on WebAssembly.
- scalar and SIMD ChaCha20 improved by 0.82% and 0.23%, respectively;
- scalar and SIMD-target Poly1305 improved by 35.32% and 34.44%; and
- scalar and production SQLite page encryption improved by 3.91% and 13.21%.
---
 src/chacha20poly1305.c | 39 ++++++++++++++++++++++++++++-----------
 1 file changed, 28 insertions(+), 11 deletions(-)

diff --git a/src/chacha20poly1305.c b/src/chacha20poly1305.c
index 86157dc0..30bd3600 100644
--- a/src/chacha20poly1305.c
+++ b/src/chacha20poly1305.c
@@ -13,12 +13,33 @@
 #define ROL32(x, c) (((x) << (c)) | ((x) >> (32-(c))))
 #define ROR32(x, c) (((x) >> (c)) | ((x) << (32-(c))))
 
-#define LOAD32_LE(p)            \
-  ( ((uint32_t)((p)[0]) <<  0)  \
-  | ((uint32_t)((p)[1]) <<  8)  \
-  | ((uint32_t)((p)[2]) << 16)  \
-  | ((uint32_t)((p)[3]) << 24)  \
-  )
+static inline uint32_t load32_le_(const void* p)
+{
+#if defined(__wasm__) || (defined(__BYTE_ORDER__) && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__)
+  uint32_t v;
+  memcpy(&v, p, 4);
+  return v;
+#else
+  const uint8_t* b = (const uint8_t*)p;
+  return (uint32_t)b[0] | (uint32_t)b[1] << 8
+       | (uint32_t)b[2] << 16 | (uint32_t)b[3] << 24;
+#endif
+}
+
+static inline void store32_le_(void* p, uint32_t v)
+{
+#if defined(__wasm__) || (defined(__BYTE_ORDER__) && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__)
+  memcpy(p, &v, 4);
+#else
+  uint8_t* b = (uint8_t*)p;
+  b[0] = (uint8_t)(v);
+  b[1] = (uint8_t)(v >> 8);
+  b[2] = (uint8_t)(v >> 16);
+  b[3] = (uint8_t)(v >> 24);
+#endif
+}
+
+#define LOAD32_LE(p) load32_le_(p)
 #define LOAD32_BE(p)            \
   ( ((uint32_t)((p)[3]) <<  0)  \
   | ((uint32_t)((p)[2]) <<  8)  \
@@ -26,11 +47,7 @@
   | ((uint32_t)((p)[0]) << 24)  \
   )
 
-#define STORE32_LE(p, v)        \
-  (p)[0] = ((v) >>  0) & 0xFF;  \
-  (p)[1] = ((v) >>  8) & 0xFF;  \
-  (p)[2] = ((v) >> 16) & 0xFF;  \
-  (p)[3] = ((v) >> 24) & 0xFF;
+#define STORE32_LE(p, v) store32_le_((p), (v))
 #define STORE32_BE(p, v)        \
   (p)[3] = ((v) >>  0) & 0xFF;  \
   (p)[2] = ((v) >>  8) & 0xFF;  \

From 10f6fb4014403e7536b3095c51fb66b7530438a7 Mon Sep 17 00:00:00 2001
From: Mathias Berchtold <mberchtold@gmail.com>
Date: Thu, 3 Sep 2026 16:51:34 +0200
Subject: [PATCH 2/2] Added check for_MSC_VER

---
 src/chacha20poly1305.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

diff --git a/src/chacha20poly1305.c b/src/chacha20poly1305.c
index 30bd3600..369e7a7f 100644
--- a/src/chacha20poly1305.c
+++ b/src/chacha20poly1305.c
@@ -15,7 +15,8 @@
 
 static inline uint32_t load32_le_(const void* p)
 {
-#if defined(__wasm__) || (defined(__BYTE_ORDER__) && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__)
+#if defined(__wasm__) || defined(_MSC_VER) \
+    || (defined(__BYTE_ORDER__) && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__)
   uint32_t v;
   memcpy(&v, p, 4);
   return v;
@@ -28,7 +29,8 @@ static inline uint32_t load32_le_(const void* p)
 
 static inline void store32_le_(void* p, uint32_t v)
 {
-#if defined(__wasm__) || (defined(__BYTE_ORDER__) && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__)
+#if defined(__wasm__) || defined(_MSC_VER) \
+    || (defined(__BYTE_ORDER__) && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__)
   memcpy(p, &v, 4);
 #else
   uint8_t* b = (uint8_t*)p;