unclobber arg in sqr16

this'll make it easier to use direct targets bypassing function args

costs 2 cycles on negatives in this version for now
but it should save us later

mandel-core.s

@@ -533,6 +533,20 @@ input_max:
     neg 4, arg
 .endmacro
 
+.macro copy_neg bytes, dest, arg
+    sec ; 2 cyc
+    .repeat bytes, byte ; 8 * byte cycles
+        lda #00         ; 2 cyc
+        sbc arg + byte  ; 3 cyc
+        sta dest + byte  ; 3 cyc
+    .endrepeat
+.endmacro
+
+; 18 cycles
+.macro copy_neg16 dest, arg
+    copy_neg 2, dest, arg
+.endmacro
+
 ; 11-27 + 18 * shift cycles
 ; 65-81 cycles for shift=3
 .macro shift_round_16 arg, shift
@@ -790,16 +804,11 @@ inner_loop:
 ; min ext-mem: 50 * 4 + 22 * 2 + 10 + 6 = 260 cyc
 ; max ext-mem: 55 * 4 + 27 * 2 + 50 + 6 = 330 cyc
 .macro imul16_impl xe
-    .local arg1
+    .scope
-    .local arg2
+        arg1 = FR0    ; 16-bit arg
-    .local result
+        arg2 = FR1    ; 16-bit arg
-    .local inter
+        result = FR2  ; 32-bit result (output)
-    .local arg1_pos
+        inter = temp2 ; 16-bit temporary (clobbered)
-    .local arg2_pos
-    arg1 = FR0   ; 16-bit arg (clobbered)
-    arg2 = FR1   ; 16-bit arg (clobbered)
-    result = FR2 ; 32-bit result
-    inter = temp2
 
         ; h1l1 * h2l2
         ; (h1*256 + l1) * (h2*256 + l2)
@@ -828,32 +837,25 @@ inner_loop:
         lda arg1 + 1                       ; 3 cyc
         bpl arg1_pos                       ; 2 cyc
         sub16 result + 2, result + 2, arg2 ; 20 cyc
-arg1_pos:
+    arg1_pos:
+
         lda arg2 + 1                       ; 3 cyc
         bpl arg2_pos                       ; 2 cyc
         sub16 result + 2, result + 2, arg1 ; 20 cyc
-arg2_pos:
+    arg2_pos:
 
         rts ; 6 cyc
+    .endscope
 .endmacro
 
-; min base ram: 5 + 19 * 2 + 81 + 22 * 2 + 6 = 174
+
-; max base ram: 23 + 19 * 2 + 92 + 27 * 2 + 6 = 213
+; min base: 19 * 2 + 81 + 22 * 2 = 163
-; min ext ram: 5 + 19 * 2 + 50 + 22 * 2 + 6 = 143
+; max base: 19 * 2 + 92 + 27 * 2 = 184
-; max ext ram: 23 + 19 * 2 + 55 + 27 * 2 + 6 = 176
+; min ext: 19 * 2 + 50 + 22 * 2 = 132
-.macro sqr16_impl xe
+; max ext: 19 * 2 + 55 + 27 * 2 = 147
+.macro sqr16_impl_inner result, arg, xe
     .scope
-        arg = FR0    ; 16-bit arg (clobbered)
+    inter = temp2
-        result = FR2 ; 32-bit result
-        ;inter = temp2
-        inter = FR1
-
-        ; 5-23 cycles
-        lda arg + 1  ; 3 cyc
-        bpl arg_pos  ; 2 cyc
-        neg16 arg    ; 18 cyc
-    arg_pos:
-
         ; hl * hl
         ; (h*256 + l) * (h*256 + l)
         ; h*256*(h*256 + l) + l*(h*256 + l)
@@ -871,7 +873,32 @@ arg2_pos:
         add_carry result + 3                 ; 2-7 cyc
         add16 result + 1, result + 1, inter  ; 20 cyc
         add_carry result + 3                 ; 2-7 cyc
+    .endscope
+.endmacro
 
+
+; min base ram: 5 + 163 + 6 = 174
+; max base ram: 7 + 184 + 18 + 6 = 215
+; min ext ram: 5 + 132 + 6 = 143
+; max ext ram: 7 + 147 + 18 + 6 = 178
+.macro sqr16_impl xe
+    .scope
+        arg = FR0    ; 16-bit arg
+        negated = FR1
+        result = FR2 ; 32-bit result
+
+        ; 5-7 cycles
+        lda arg + 1                ; 3 cyc
+        bpl arg_pos                ; 2 cyc
+        jmp arg_neg                ; 3 cyc
+
+    arg_pos:
+        sqr16_impl_inner result, arg, xe      ; 163-184 / 132-147 cyc
+        rts                                   ; 6 cyc
+
+    arg_neg:
+        copy_neg16 negated, arg               ; 18 cyc
+        sqr16_impl_inner result, negated, xe  ; 163-184 / 132-147 cyc
         rts                                   ; 6 cyc
     .endscope
 .endmacro