crates/asm: fix aarch64 stack alignment and add bcmp/memcmp

aarch64 logical instructions encode register 31 as xzr, not sp, so
`and sp, sp, #-16` is an invalid operand. Use a temp register (x9)
to align the stack instead.

Also add bcmp and memcmp implementations needed for -static linking
without libc (riscv64 codegen emits calls to bcmp for byte comparisons).

Author: amaanq

crates/asm/src/lib.rs

@@ -55,6 +55,33 @@ pub unsafe extern "C" fn memset(s: *mut u8, c: i32, n: usize) -> *mut u8 {
   s
 }
 
+/// Compares two byte sequences.
+///
+/// # Safety
+///
+/// `s1` and `s2` must be valid pointers to memory of at least `n` bytes.
+#[unsafe(no_mangle)]
+pub unsafe extern "C" fn bcmp(s1: *const u8, s2: *const u8, n: usize) -> i32 {
+  for i in 0..n {
+    let a = unsafe { *s1.add(i) };
+    let b = unsafe { *s2.add(i) };
+    if a != b {
+      return i32::from(a) - i32::from(b);
+    }
+  }
+  0
+}
+
+/// Compares two byte sequences.
+///
+/// # Safety
+///
+/// `s1` and `s2` must be valid pointers to memory of at least `n` bytes.
+#[unsafe(no_mangle)]
+pub unsafe extern "C" fn memcmp(s1: *const u8, s2: *const u8, n: usize) -> i32 {
+  unsafe { bcmp(s1, s2, n) }
+}
+
 /// Calculates the length of a null-terminated string.
 ///
 /// # Safety
@@ -198,13 +225,16 @@ mod entry {
       // Move stack pointer to first argument register
       "mov x0, sp",
       // Align stack to 16-byte boundary (AArch64 ABI requirement)
-      "and sp, sp, -16",
+      // Note: `and sp, sp, #imm` is invalid because logical instructions
+      // encode register 31 as xzr, not sp. Use a temp register instead.
+      "mov x9, sp",
+      "and x9, x9, #-16",
+      "mov sp, x9",
       // Call into Rust code
       "bl {entry_rust}",
-      // Move return code to syscall argument
-      "mov x0, x0",
+      // Return code is already in x0 from entry_rust
       // Exit syscall
-      "mov x8, 93",  // SYS_exit
+      "mov x8, #93",  // SYS_exit
       "svc #0",
       entry_rust = sym super::entry_rust,
     );