Reduced operations to get -1 in register

array.asm

@@ -11,7 +11,7 @@ pub find_index:
     push r12 ; We will store the predicate pointer here
     push r11 ; We will store the current pointer here
     push r10 ; We will store the stride here
-    push r9 ; We will store the bytes left here
+    push r9 ; We will store the final address here
     push r8 ; We will store the mask here
     push r1 ; We need the array pointer to calculate the item index
 
@@ -19,14 +19,14 @@ pub find_index:
     mov r11, r1
     mov r10, r3
     mov r9, r2
-    lsl r9, r9, r3
+    lsl r9, r9, r3 ; We calculate bytes left
+    add r9, r9, r1 ; We add the start address to get the final address
 
     push r13 ; We save up the return address because we will provide our own to the predicate
     counter r13
     add r13, r13, 52 ; Point to just after the predicate call - we can set this up now so we don't waste loop cycles
 
-    mov r8, 0
+    nand r8, zr, zr ; We create a mask of 0xFFFFFFFF
-    not r8, r8 ; We create a mask of 0xFFFFFFFF
     mov r6, 4
     sub r6, r6, r3 ; We create a "negative stride", e.g. 4 -> 0, 2 -> 2, 1 -> 3
     lsl r6, r6, 3
@@ -45,16 +45,14 @@ pub find_index:
         jneq find_index_found_item ; If we found the item, we jump out
         ; If we didn't, move to next item
         add r11, r11, r10 ; We add the stride to the pointer
-        sub r9, r9, r10 ; We decrease bytes left
+        cmp r11, r9 ; We compare with the final address
-        cmp r9, zr
         jeq find_index_not_found ; If we reached the end we're done
         jmp find_index_loop
     
     find_index_not_found:
         add sp, sp, 4 ; The predicate context is not useful
         pop r13 ; We get our return address
-        mov r1, 0
+        nand r1, zr, zr ; We put -1 in r1
-        not r1, r1 ; We put -1 in r1
         add sp, sp, 4 ; The old array pointer are not useful
         jmp find_index_postamble