From 0797e69106dccbbe6cead5640fc189d5860750a7 Mon Sep 17 00:00:00 2001
From: hugova <hugova@kth.se>
Date: Thu, 6 Mar 2025 01:50:37 +0100
Subject: [PATCH] Add all optimisations done on line_down to line_down_inv. DIx
 a problem that the pixel was drawin a bit to late in the loop on line_down.

---
 wip-hugo/macros/16aritmatic.s          | 14 -------
 wip-hugo/routines/line/line_down.s     | 16 ++++----
 wip-hugo/routines/line/line_down_inv.s | 53 +++++++++++++++++---------
 wip-hugo/source.s                      |  2 +-
 4 files changed, 44 insertions(+), 41 deletions(-)

diff --git a/wip-hugo/macros/16aritmatic.s b/wip-hugo/macros/16aritmatic.s
index c9010e0..46eb0be 100755
--- a/wip-hugo/macros/16aritmatic.s
+++ b/wip-hugo/macros/16aritmatic.s
@@ -10,12 +10,6 @@
 .endif
 ;; If b_low != A
 .if .match(.mid (0, 1, {b_low}), A )
-.elseif .match(.mid (0, 1, {b_low}), X)
-LDX
-kssk
-.match(.mid (0, 1, {b_low}), Y)
-LDY
-kkdkd
 .else
     LDA b_low
 .endif
@@ -23,15 +17,7 @@ kkdkd
 	ADC a_low
 	STA a_low
 	LDA b_hi
-.if .match(.mid (0, 1, {b_low}), X )
-	LDX
-	kdkdkd
-.elseif .match(.mid (0, 1, {b_low}), Y )
-	LDY
-	dkkdk
-.else
     LDA b_hi
-.endif
 	ADC a_hi
 	STA a_hi
 .endmacro
diff --git a/wip-hugo/routines/line/line_down.s b/wip-hugo/routines/line/line_down.s
index 3fc4aaa..2319b3d 100644
--- a/wip-hugo/routines/line/line_down.s
+++ b/wip-hugo/routines/line/line_down.s
@@ -48,21 +48,20 @@
     Add_16 >D, <D, #$ff, #$01, !
     Sub_16 >D, <D, dx, #$00
 
-
     LDY #$00
     jsr pixel_draw ;;only used first pixel. after this relative position is abused
 for_x:
     ;; Lets increment btp_mem_pos with +8
     ;; Read more in pixel_draw to understand this!
-increment_pixel_x:
-    CLC
-    LDA byte_to_paint
-    ROR byte_to_paint
     LDX #$00
+    LDA byte_to_paint
     ORA (>btp_mem_pos, X)
     STA (>btp_mem_pos, X)
+increment_pixel_x:
+    CLC
+    ROR byte_to_paint
     BCS move_8px_left
-    jmp increment_pixel_x_end
+    JMP increment_pixel_x_end
 move_8px_left:
     ;; add +8 to btp_mem_pos. Find more of why in pixel_draw
     Add_16 >btp_mem_pos, <btp_mem_pos, #$08, #$00
@@ -79,17 +78,18 @@ increment_pixel_x_end:
     Lag_16 >D, <D, #$00, #$02, case_2
 case_1:; C =1 so we can use !
     Sub_16 >D, <D, >V, <V, ! ; D = D - V
-    increment_y_pos:
+increment_y_pos:
     Add_16 >btp_mem_pos, <btp_mem_pos, #$01, #$00
     INY
     CPY #$08 ;
     BEQ move_8px_down
     jmp for_x
-move_8px_down:
+move_8px_down: ; Z=1 --> C=1
     ;; +320-8 bytes
     LDY #$00
     Add_16 >btp_mem_pos, <btp_mem_pos,#$38 , #$01
     jmp for_x
+increment_y_pos_end:
 case_2: ;; C =0 because LAG_16 so we can use !
     Add_16 >D, <D, >dy_2, <dy_2, ! ;D = D + 2*dy
     JMP for_x
diff --git a/wip-hugo/routines/line/line_down_inv.s b/wip-hugo/routines/line/line_down_inv.s
index 9e5831f..517882f 100644
--- a/wip-hugo/routines/line/line_down_inv.s
+++ b/wip-hugo/routines/line/line_down_inv.s
@@ -8,16 +8,6 @@
     ;;NOTE THAT X_pos <= X_end, Y_pos <= Y_end. Min 45deg!
     .include "line.inc"; Defines memory positions, ex X_pos
 
-    ;example values ~~~~~ SHOULD BE PRECOMPILED
-    ;LDA #$00
-    ;STA X_pos
-    ;STA Y_pos
-    ;LDA #$40
-    ;STA X_end
-    ;LDA #$a0
-    ;STA Y_end
-    ;;~~~~~~~~~~
-
     ;;We need to clear this memory
     LDA #$00
     STA <V
@@ -52,23 +42,50 @@
     Add_16 >D, <D, #$ff, #$01
     Sub_16 >D, <D, dy, #$00
 
+    LDY #$00
+    jsr pixel_draw ;;only used first pixel. after this relative position is abused
+    CLC ; this makes it that C =0 always on for_x
 for_y:
-    jsr pixel_draw
-    ;;Increment Y until Y_pos = Y_end and X_pos = X_end
-    INC Y_pos
-    LDY Y_pos
-    CPY Y_end
+LDX #$00
+LDA byte_to_paint
+ORA (>btp_mem_pos, X)
+STA (>btp_mem_pos, X)
+increment_y_pos:
+    Add_16 >btp_mem_pos, <btp_mem_pos, #$01, #$00, ! ; I know that C ==0 on for_y
+    INY
+    CPY #$08 ;
+    BEQ move_8px_down
+    jmp increment_y_pos_end
+move_8px_down:
+    ;; +320-8 bytes
+    LDY #$00
+    Add_16 >btp_mem_pos, <btp_mem_pos,#$38 , #$01
+increment_y_pos_end:
+    INC Y_pos;; legacy
+    LDX Y_pos
+    CPX Y_end
     BEQ end
 
     ;;If D <  %00000010 00000000:  case_2
     ;;else case 1.
     Lag_16 >D, <D, #$00, #$02, case_2
 case_1:
-    INC X_pos
     Sub_16 >D, <D, >V, <V; D = D - V
+increment_pixel_x:
+    CLC
+    LDA byte_to_paint
+    ROR byte_to_paint
+    BCS move_8px_left
     JMP for_y
-case_2:
-    Add_16 >D, <D, >dx_2, <dx_2;D = D + 2*dx
+move_8px_left:
+    ;; add +8 to btp_mem_pos. Find more of why in pixel_draw
+    Add_16 >btp_mem_pos, <btp_mem_pos, #$08, #$00
+    ;; reset byte_to_paint
+    LDA #%10000000
+    STA byte_to_paint
+    JMP for_y
+case_2:  ;; C =0 because LAG_16 so we can use !
+    Add_16 >D, <D, >dx_2, <dx_2, ! ;D = D + 2*dx
     JMP for_y
 end:
     RTS
diff --git a/wip-hugo/source.s b/wip-hugo/source.s
index 47b370b..7392a30 100755
--- a/wip-hugo/source.s
+++ b/wip-hugo/source.s
@@ -2,7 +2,7 @@
 .include "macros/timer.s"
 
 .include "STARTUP.s"
-.include "routines/line/line_test_time.s"
+.include "routines/line/line_test.s"
 
 ;;This is used by .s files to terminate nicely without reading includes at the end!
 exit: