Avoid overflow.

sqlite3/libc/libc.wat

@@ -1325,11 +1325,11 @@
    )
    (i32.const 0)
    (i32.le_u
-    (local.get $0)
-    (i32.add
-     (local.get $1)
+    (i32.sub
+     (local.get $0)
      (local.get $3)
     )
+    (local.get $1)
    )
   )
  )

sqlite3/libc/string.h

@@ -113,7 +113,7 @@ void *memchr(const void *v, int c, size_t n) {
         // That's a match, unless it is beyond the end of the object.
         // Recall that we decremented n, so less-than-or-equal-to is correct.
         size_t ctz = __builtin_ctz(mask);
-        return ctz <= n + align ? (char *)w + ctz : NULL;
+        return ctz - align <= n ? (char *)w + ctz : NULL;
       }
     }
     // Decrement n; if it overflows we're done.
@@ -166,6 +166,8 @@ size_t strlen(const char *s) {
       // At least one bit will be set, unless we cleared them.
       // Knowing this helps the compiler.
       __builtin_assume(mask || align);
+      // If the mask is zero because of alignment,
+      // it's as if we didn't find anything.
       if (mask) {
         // Find the offset of the first one bit (little-endian).
         return (char *)w - s + __builtin_ctz(mask);
@@ -280,6 +282,8 @@ static char *__strchrnul(const char *s, int c) {
       // At least one bit will be set, unless we cleared them.
       // Knowing this helps the compiler.
       __builtin_assume(mask || align);
+      // If the mask is zero because of alignment,
+      // it's as if we didn't find anything.
       if (mask) {
         // Find the offset of the first one bit (little-endian).
         return (char *)w + __builtin_ctz(mask);

sqlite3/libc/strings.h

@@ -57,6 +57,7 @@ int bcmp(const void *v1, const void *v2, size_t n) {
 
 #endif  // __OPTIMIZE_SIZE__
 
+__attribute__((always_inline))
 static v128_t __tolower8x16(v128_t v) {
   __i8x16 i = v;
   i = i + wasm_i8x16_splat(INT8_MAX - ('Z'));