clean up runtime

src/runtime/baseline/entry.s

@@ -2,9 +2,10 @@
 .extern main
 .section .text
 
-.globl _start
+.global _start
 _start:
     mov rdi, rsp
+    # main([]^string): i64
     call main
     mov rdi, rax
     mov rax, 60

src/runtime/baseline/nub_memcpy.s

@@ -1,24 +1,20 @@
 .intel_syntax noprefix
 .section .text
 
-# Memory copy function
-# Arguments: rdi = destination, rsi = source, rdx = count
-# Returns: rdi (original destination pointer)
-.globl nub_memcpy
+# func nub_memcpy(destination: ^u8, source: ^u8, count: u64): ^u8
+.global nub_memcpy
 nub_memcpy:
-    push rdi                # Save original destination for return value
-    mov rcx, rdx            # Load count into counter register
-    # Handle zero count case
+    push rdi
+    mov rcx, rdx
     test rcx, rcx
-    jz .done
-    # Simple byte-by-byte copy (no overlap handling)
-.loop:
-    mov al, BYTE PTR [rsi]  # Load byte from source
-    mov BYTE PTR [rdi], al  # Store byte to destination
-    inc rsi                 # Move to next source byte
-    inc rdi                 # Move to next destination byte
-    dec rcx                 # Decrement counter
-    jnz .loop               # Continue if counter not zero
-.done:
-    pop rax                 # Return original destination pointer
+    jz memcpy_done
+memcpy_loop:
+    mov al, BYTE PTR [rsi]
+    mov BYTE PTR [rdi], al
+    inc rsi
+    inc rdi
+    dec rcx
+    jnz memcpy_loop
+memcpy_done:
+    pop rax
     ret

src/runtime/baseline/nub_memset.s

@@ -1,22 +1,19 @@
 .intel_syntax noprefix
 .section .text
 
-# Memory set function
-# Arguments: rdi = destination pointer, rsi = value (byte), rdx = count
-# Returns: rdi (original destination pointer)
-.globl nub_memset
+# func nub_memset(destination: ^u8, value: i8, count: u64): ^u8
+.global nub_memset
 nub_memset:
-    push rdi                # Save original destination for return value
-    mov rcx, rdx            # Load count into counter register
-    mov al, sil             # Move byte value to al (lower 8 bits of rsi)
-    # Handle zero count case
+    push rdi
+    mov rcx, rdx
+    mov al, sil
     test rcx, rcx
-    jz .done
-.loop:
-    mov BYTE PTR [rdi], al  # Store byte at current position
-    inc rdi                 # Move to next byte
-    dec rcx                 # Decrement counter
-    jnz .loop               # Continue if counter not zero
-.done:
-    pop rax                 # Return original destination pointer
+    jz memset_done
+memset_loop:
+    mov BYTE PTR [rdi], al
+    inc rdi
+    dec rcx
+    jnz memset_loop
+memset_done:
+    pop rax
     ret

src/runtime/baseline/nub_panic.s

@@ -1,10 +1,8 @@
 .intel_syntax noprefix
 .section .text
 
-# Panic function with message
-# Arguments: rdi = message (*char), rsi = message length (long)
-# Remarks: exits the program
-.globl nub_panic
+# func nub_panic(message: ^u8, message_length: u64): void
+.global nub_panic
 nub_panic:
     mov rdx, rsi
     mov rsi, rdi

src/runtime/baseline/nub_strcmp.s

@@ -1,24 +1,22 @@
 .intel_syntax noprefix
 .section .text
 
-# String comparison function null-terminated strings
-# Arguments: rdi = lhs (*char), rsi = rhs (*char)
-# Returns: 1 if equal, else 0
-.globl nub_strcmp
+# func nub_strcmp(lhs: ^u8, rhs: ^u8): bool
+.global nub_strcmp
 nub_strcmp:
     xor rdx, rdx
-.loop:
+strcmp_loop:
     mov al, BYTE PTR [rsi + rdx]
     mov bl, BYTE PTR [rdi + rdx]
     inc rdx
     cmp al, bl
-    jne .not_equal
+    jne strcmp_not_equal
     cmp al, 0
-    je .equal
-    jmp .loop
-.not_equal:
+    je strcmp_equal
+    jmp strcmp_loop
+strcmp_not_equal:
     mov rax, 0
     ret
-.equal:
+strcmp_equal:
     mov rax, 1
     ret

src/runtime/core/syscall.s

@@ -1,5 +1,5 @@
 .intel_syntax noprefix
-.globl core_syscall
+.global core_syscall
 .section .text
 
 core_syscall: