Adjust rounding, loop exits

Still slight differences between positive and negative Y halves
Not sure what's wrong

mandel.s

@@ -359,48 +359,38 @@ positive_result:
 
 .macro round16 arg
     ; Round top 16 bits of 32-bit fixed-point number in-place
-    .local zero
-    .local one
     .local positive
     .local negative
-    .local neg2
+    .local increment
     .local next
 
-    ; no round            - 5 cycles
-    ; round pos, no carry - 17
-    ; round pos, carry    - 22
-    ; round neg, no carry - 23
-    ; round neg, carry    - 28
-    ; average = 5 / 2 + (17 + 22 + 23 + 28) / 8
-    ;         = 5 / 2 + 90 / 8
-    ;         = 2.5 + 11.25 = 13.75 cycles average on evenly distributed input
+    ; positive input:
+    ;   0 rounds down, no change - 10 cyc
+    ;   1 rounds up, add one - 17-22 cyc
+    ; negative input:
+    ;   0 rounds up, add one - 19-24 cyc
+    ;   1 rounds down, no change - 10 cyc
 
-    lda arg + 1  ; 3 cyc
-    bpl zero     ; 2 cyc
-
-one:
     ; check sign bit
     lda arg + 3  ; 3 cyc
     bpl positive ; 2 cyc
 
 negative:
-    lda arg + 2  ; 3 cyc
-    beq neg2     ; 2 cyc
-
-    dec arg + 2  ; 5 cyc
-    jmp next     ; 3 cyc
-
-neg2:
-    dec arg + 2  ; 5 cyc
-    dec arg + 3  ; 5 cyc
-    jmp next     ; 3 cyc
+    ; check 16th bit down
+    lda arg + 1   ; 3 cyc
+    bmi next      ; 2 cyc
+    bpl increment ; 2 cyc
 
 positive:
+    ; check 16th bit down
+    lda arg + 1   ; 3 cyc
+    bpl next      ; 2 cyc
+
+increment:
     inc arg + 2  ; 5 cyc
     bne next     ; 2 cyc
     inc arg + 3  ; 5 cyc
 
-zero:
 next:
 
 .endmacro
@@ -435,17 +425,34 @@ loop:
 keep_going:
 
     .macro quick_exit arg, max
-        .local keep_going
-        .local keep_going2
+        .local positive
+        .local negative
+        .local nope_out
+        .local first_equal
+        .local all_done
+
+        ; check sign bit
         lda arg + 1
+        bmi negative
+
+    positive:
         cmp #((max) << 4)
-        bmi keep_going
+        bmi all_done ; 'less than'
         rts
-    keep_going:
+
+    negative:
         cmp #(256 - ((max) << 4))
-        bpl keep_going2
+        beq first_equal ; 'equal' on first byte
+        bpl all_done    ; 'greater than'
+
+    nope_out:
         rts
-    keep_going2:
+    
+    first_equal:
+        lda arg
+        beq nope_out  ; 2nd byte 0 shows it's really 'equal'
+
+    all_done:
     .endmacro
 
     ; 4.12: (-8 .. +7.9)