WIP savings of half a cycle per imul8_xe

Uses X to cache arg1, which is always used, instead of
arg2, which is only used on odds.

Should save half a cycle per imul8_xe, untested

Makefile

@@ -2,11 +2,8 @@
 
 all : mandel.xex
 
-mandel.xex : mandel.o mandel-core.o tables.o atari-xex.cfg
+mandel.xex : mandel.o tables.o atari-asm-xex.cfg
-	ld65 -C ./atari-xex.cfg --mapfile mandel.map -o $@ mandel.o mandel-core.o tables.o atari.lib
+	ld65 -C ./atari-asm-xex.cfg --mapfile mandel.map -o $@ mandel.o tables.o
-
-mandel.s : mandel.c mandel.h
-	cc65 -o $@ mandel.c
 
 %.o : %.s
 	ca65 -o $@ $<
@@ -16,7 +13,8 @@ tables.s : tables.js
 
 clean :
 	rm -f tables.s
-	rm -f mandel.s
 	rm -f *.o
 	rm -f *.xex
 	rm -f mandel.map
+
+

atari-xex.cfg

@@ -1,69 +0,0 @@
-# Sample linker configuration for C programs using the Atari binary file support.
-# Use with: cl65 -tatari -Catari-xex.cfg prog.c -o prog.xex
-FEATURES {
-    STARTADDRESS: default = $8000;
-}
-SYMBOLS {
-    __SYSTEM_CHECK__:    type = import;  # force inclusion of "system check" load chunk
-    __STACKSIZE__:       type = weak, value = $0800; # 2k stack
-    __STARTADDRESS__:    type = export, value = %S;
-    __RESERVED_MEMORY__: type = weak, value = $0000;
-    __SYSCHKHDR__:       type = export, value = 0; # Disable system check header
-    __SYSCHKTRL__:       type = export, value = 0; # Disable system check trailer
-    __TABLESEG_START__:    type = weak, value = $2E00 + $0300;
-    __TABLESEG_SIZE__:     type = weak, value = 6 * $100;
-    __BANKSY_START__:  type = weak, value = $4000;
-    __BANKSY_SIZE__:   type = weak, value = $4000;
-    __FRAMEBUFFER_START__: type = weak, value = $A000;
-}
-MEMORY {
-# Note -- $80 and $81 (LOMEM) appear to be reserved in ZP.
-    ZP:         file = "", define = yes, start = $0082, size = $007E;
-# "system check" load chunk
-    SYSCHKCHNK: file = %O,               start = $2E00, size = $0300;
-# Note $a000-$bfff is against the BASIC cartridge, may require booting with OPTION.
-    TABLES:     file = %O, define = yes, start = __TABLESEG_START__, size = __TABLESEG_SIZE__;
-# We reserve $4000-7fff for the bank-switch window.
-# In theory we could keep data and code here that we only use on 48k/64k systems.
-    BANKSWITCH: file = "", define = yes, start = __BANKSY_START__, size = __BANKSY_SIZE__;
-# "main program" load chunk
-    MAIN:       file = %O, define = yes, start = %S, size = __FRAMEBUFFER_START__ - __STACKSIZE__ - __RESERVED_MEMORY__ - %S;
-}
-FILES {
-    %O: format = atari;
-}
-FORMATS {
-    atari: runad = start,
-           initad = SYSCHKCHNK: __SYSTEM_CHECK__;
-}
-SEGMENTS {
-    ZEROPAGE:  load = ZP,         type = zp;
-    EXTZP:     load = ZP,         type = zp,                optional = yes;
-    SYSCHK:    load = SYSCHKCHNK, type = rw,  define = yes, optional = yes;
-    TABLES:    load = TABLES,     type = ro,  optional = yes, align = 256;
-    BANKSWICH: load = BANKSWITCH, type = ro,  optional = yes;
-    STARTUP:   load = MAIN,       type = ro,  define = yes;
-    LOWBSS:    load = MAIN,       type = rw,                optional = yes;  # not zero initialized
-    LOWCODE:   load = MAIN,       type = ro,  define = yes, optional = yes;
-    ONCE:      load = MAIN,       type = ro,                optional = yes;
-    CODE:      load = MAIN,       type = ro,  define = yes;
-    RODATA:    load = MAIN,       type = ro;
-    DATA:      load = MAIN,       type = rw;
-    INIT:      load = MAIN,       type = rw,                optional = yes;
-    BSS:       load = MAIN,       type = bss, define = yes;
-}
-FEATURES {
-    CONDES: type    = constructor,
-            label   = __CONSTRUCTOR_TABLE__,
-            count   = __CONSTRUCTOR_COUNT__,
-            segment = ONCE;
-    CONDES: type    = destructor,
-            label   = __DESTRUCTOR_TABLE__,
-            count   = __DESTRUCTOR_COUNT__,
-            segment = RODATA;
-    CONDES: type    = interruptor,
-            label   = __INTERRUPTOR_TABLE__,
-            count   = __INTERRUPTOR_COUNT__,
-            segment = RODATA,
-            import  = __CALLIRQ__;
-}

mandel.c

@@ -1,15 +0,0 @@
-/**
- * The UI and I/O wrapper for the Mandelbrot runner, in C.
- *
- * For the moment *all* logic is in mandel-core.s, I'm just
- * trying to get this to run within a cc65 environment.
- * Eventually just the inner loop fun will live in there.
- */
-
-#include <stdlib.h>
-#include <stdio.h>
-#include "mandel.h"
-
-void main(void) {
-    mandel_start();
-}

mandel.h

@@ -1,4 +0,0 @@
-#include <inttypes.h>
-
-// From mandel-core.s:
-extern void mandel_start(void);

mandel.s

@@ -1,44 +1,44 @@
-.zeropage
+; Our zero-page vars
+ox              = $80 ; fixed6.26: center point x
+oy              = $84 ; fixed6.26: center point y
+cx              = $88 ; fixed6.26: c_x
+cy              = $8c ; fixed6.26: c_y
 
-ox:              .res 4 ; fixed6.26: center point x
+zx              = $90 ; fixed6.26: z_x
-oy:              .res 4 ; fixed6.26: center point y
+zy              = $94 ; fixed6.26: z_y
-cx:              .res 4 ; fixed6.26: c_x
+zx_2            = $98 ; fixed6.26: z_x^2
-cy:              .res 4 ; fixed6.26: c_y
+zy_2            = $9c ; fixed6.26: z_y^2
 
-zx:              .res 4 ; fixed6.26: z_x
+zx_zy           = $a0 ; fixed6.26: z_x * z_y
-zy:              .res 4 ; fixed6.26: z_y
+dist            = $a4 ; fixed6.26: z_x^2 + z_y^2
-zx_2:            .res 4 ; fixed6.26: z_x^2
+sx              = $a8 ; i16: screen pixel x
-zy_2:            .res 4 ; fixed6.26: z_y^2
+sy              = $aa ; i16: screen pixel y
+z_buffer_active = $ac ; boolean: 1 if we triggered the lake, 0 if not
+z_buffer_start  = $ad ; u8: index into z_buffer
+z_buffer_end    = $ae ; u8: index into z_buffer
+iter            = $af ; u8: iteration count
 
-zx_zy:           .res 4 ; fixed6.26: z_x * z_y
+ptr             = $b0 ; u16
-dist:            .res 4 ; fixed6.26: z_x^2 + z_y^2
+pixel_ptr       = $b2 ; u16
+zoom            = $b4 ; u8: zoom shift level
+fill_level      = $b5 ; u8
+pixel_color     = $b6 ; u8
+pixel_mask      = $b7 ; u8
+pixel_shift     = $b8 ; u8
+pixel_offset    = $b9 ; u8
+palette_offset  = $ba ; u8
+chroma_offset   = $bb ; u8
+palette_ticks   = $bc ; u8
+chroma_ticks    = $bd ; u8
+count_frames    = $be ; u8
+; free space $bf
 
-z_buffer_active: .res 1 ; boolean: 1 if we triggered the lake, 0 if not
+count_iters     = $c0 ; u16
-z_buffer_start:  .res 1 ; u8: index into z_buffer
+text_col        = $c2 ; u8
-z_buffer_end:    .res 1 ; u8: index into z_buffer
+text_row        = $c3 ; u8
-iter:            .res 1 ; u8: iteration count
+; free space c4-cb
-ptr:             .res 2 ; u16
+temp            = $cc ; u16
-temp:            .res 2 ; u16
+temp2           = $ce ; u16
-temp2:           .res 2 ; u16
-
-.data
-; can move to .data
-sx:              .res 2 ; i16: screen pixel x
-sy:              .res 2 ; i16: screen pixel y
-zoom:            .res 1 ; u8: zoom shift level
-fill_level:      .res 1 ; u8
-pixel_color:     .res 1 ; u8
-pixel_mask:      .res 1 ; u8
-pixel_shift:     .res 1 ; u8
-pixel_offset:    .res 1 ; u8
-palette_offset:  .res 1 ; u8
-chroma_offset:   .res 1 ; u8
-palette_ticks:   .res 1 ; u8
-chroma_ticks:    .res 1 ; u8
-count_frames:    .res 1 ; u8
-count_iters:     .res 2 ; u16
-text_col:        .res 1 ; u8
-text_row:        .res 1 ; u8
 
 palette_delay = 23
 chroma_delay = 137
@@ -131,8 +131,6 @@ KEY_E     = 42
 KEY_X     = 22
 KEY_Y     = 43
 
-.data
-
 .struct float48
     exponent .byte
     mantissa .byte 5
@@ -144,6 +142,7 @@ KEY_Y     = 43
 .import sqr_lobyte
 .import sqr_hibyte
 
+.data
 
 strings:
 str_self:
@@ -362,7 +361,7 @@ z_buffer:
         .word 0
     .endrepeat
 
-.export _mandel_start
+.export start
 
 ;max_fill_level = 6
 max_fill_level = 3
@@ -589,7 +588,7 @@ bank_switch_table:
 .macro imul8 dest, arg1, arg2, xe
     .if xe
         ; using 64KB lookup table
-        ; 51-70 cycles
+        ; 50-70 cycles
         ; clobbers x, y, dest, ptr
         .scope
             output = dest
@@ -601,13 +600,13 @@ bank_switch_table:
 
             ; bottom 14 bits except the LSB are the per-bank table index
             ; add $4000 for the bank pointer
-            txa          ; 2 cyc
             and #$3f     ; 2 cyc
             ora #$40     ; 2 cyc
             sta ptr + 1  ; 3 cyc
 
             ; copy the entry into output
             lda arg1     ; 3 cyc
+            tax          ; 2 cyc
             and #$fe     ; 2 cyc
             tay          ; 2 cyc
             lda (ptr),y  ; 5 cyc
@@ -624,13 +623,13 @@ bank_switch_table:
             ;;sta PORTB    ; 4 cyc - disabled
 
             ; check that 1 bit we skipped to fit into space
-            lda arg1     ; 3 cyc
+            txa          ; 2 cyc
             and #1       ; 2 cyc
             beq done     ; 2 cyc
 
             ; add arg2 one last time for the skipped bit
             clc          ; 2 cyc
-            txa          ; 2 cyc
+            lda arg1     ; 3 cyc
             adc output   ; 3 cyc
             sta output   ; 3 cyc
             lda #0       ; 2 cyc
@@ -1254,21 +1253,21 @@ enough:
 negative:
     ; temp1 = top half
     lda #.lobyte(framebuffer_top + stride * half_height)
-    sta ptr
+    sta pixel_ptr
     lda #.hibyte(framebuffer_top + stride * half_height)
-    sta ptr + 1
+    sta pixel_ptr + 1
     jmp point
 
 positive:
 
     lda #.lobyte(framebuffer_bottom)
-    sta ptr
+    sta pixel_ptr
     lda #.hibyte(framebuffer_bottom)
-    sta ptr + 1
+    sta pixel_ptr + 1
 
 point:
 
-    ; ptr += sy * stride
+    ; pixel_ptr += sy * stride
     ;    temp * 40
     ; =  temp * 32  +  temp * 8
     ; = (temp << 5) + (temp << 3)
@@ -1276,10 +1275,10 @@ point:
     shl16 temp
     shl16 temp
     shl16 temp
-    add16 ptr, ptr, temp
+    add16 pixel_ptr, pixel_ptr, temp
     shl16 temp
     shl16 temp
-    add16 ptr, ptr, temp
+    add16 pixel_ptr, pixel_ptr, temp
 
     ; Ok so temp1 points to the start of the line, which is 40 bytes.
     ; Get the byte and bit offsets
@@ -1319,20 +1318,20 @@ shift_done:
 
 draw_pixel:
     ; read, mask, or, write
-    lda (ptr),y
+    lda (pixel_ptr),y
     and pixel_mask
     ora pixel_color
-    sta (ptr),y
+    sta (pixel_ptr),y
 
     dex
     beq done
     clc
     lda #40
-    adc ptr
+    adc pixel_ptr
-    sta ptr
+    sta pixel_ptr
     lda #0
-    adc ptr + 1
+    adc pixel_ptr + 1
-    sta ptr + 1
+    sta pixel_ptr + 1
     jmp draw_pixel
 
 done:
@@ -1746,7 +1745,7 @@ zero_byte_loop:
     rts
 .endproc
 
-.proc _mandel_start
+.proc start
 
     jsr imul8xe_init