xmas2018: scrolling optimizing

This commit is contained in:
Vince Weaver 2018-12-20 22:03:31 -05:00
parent fd1377c10c
commit cd6d5e9f4c

View File

@ -196,17 +196,18 @@ OFFSCREEN = $05
; Timing
; scroll_hgr_left: 8
; 140* scroll_hgr_loop: 10 setup
; 64*left_one_loop 6+3651
; 64*left_one_loop 6+3225
; 23 (increments)
; 29 increment counts
; 10 check and loop
; 6 return
; total time = 14 + 140*(10+29+10+64*(41+23+(3657)))
; total time = 14 + 140*(10+29+10+64*(41+23+(3225)))
; 67,431,293 cycles = roughly 67s -- original
; 64,564,093 cycles = roughly 64s -- optimize inner loop a bit
; 33,347,034 cycles = roughly 33s -- don't shift hidden page
; 30,569,434 cycles = roughly 30s -- unroll 4 times
; 29,476,314 cycles = roughly 29s -- add back INH for +1 address
scroll_hgr_left:
lda #$0 ; 2
@ -282,11 +283,16 @@ scroll_done:
;===========================================
;
; 86 init
; 40*
; hgr_scroll_line_loop: 25
; high bit 20
; prepare bits: 16
; output new byte: 21
; 10* (unrolled)
; 3* hgr_scroll_line_loop: 16
; high bit 20
; prepare bits: 16
; output new byte: 23
; 1* hgr_scroll_line_loop: 21
; high bit 20
; prepare bits: 16
; output new byte: 23
; 1*
; increment and loop: 7
; 5 return
;
@ -294,7 +300,8 @@ scroll_done:
; (91*40)+7=3647 -- remove branch in highbit code
; (89*40)+7=3567 -- convert 5 asl to 4 ror
; (89*40)+91=3651 -- re-write with col40 pre-calculated
; (79*3 + 84*1 + 7)*10+91 = 3341
; (75*3 + 80*1 + 7)*10+105= 3225
hgr_scroll_line:
setup_column_40:
@ -383,8 +390,204 @@ done_count:
; best case(0)= 19
; worse case(3)= 56
; repeated 40 times
ldx OUTL ; 3
inx ; 2
stx INL ; 3
ldx OUTH ; 3
stx INH ; 3
;===========
; 14
; repeated 10 times
hgr_scroll_line_loop:
;============= Unroll 0
lda (OUTL),Y ; get pixel block of interest ; 5
sta CURRENT ; 3
lda (INL),Y ; get subsequent pixel block ; 5
sta NEXT ; 3
;===================
; 20
; if in bit 2 or 6 of horiz scroll, shift the color bit over
; makes some color flicker, is there a better way?
high_bit0:
lda COUNTL ; 3
and #$3 ; 2
cmp #2 ; 2
bne keep_high_bit0 ; 3
move_high_bit0:
; -1
lda NEXT ; 3
jmp done_high_bit0 ; 3
keep_high_bit0:
lda CURRENT ; 3
done_high_bit0:
and #$80 ; 2
sta HIGH ; 3
;===================
; 2or6: 20
; else: 18
prepare_bits0:
; get right byte, bottom 2 bits, shifted left to be in 6+5
lda NEXT ; 3
; this method 2 cycles faster than asl x 5
ror ; 2
ror ; 2
ror ; 2
ror ; 2
and #$60 ; 2
sta NEXT ; 3
;==========
; 16
output_new0:
; get current, mask off bottom 2 bits (no longer needed)
; then OR in the saved high (color) bit as well as NEXT bits
lda CURRENT ; 3
lsr ; 2
lsr ; 2
and #$1f ; 2
ora HIGH ; 3
ora NEXT ; 3
sta (OUTL),Y ; 6
iny ; 2
;===========
; 23
;============= Unroll 1
lda (OUTL),Y ; get pixel block of interest ; 5
sta CURRENT ; 3
lda (INL),Y ; get subsequent pixel block ; 5
sta NEXT ; 3
;===================
; 20
; if in bit 2 or 6 of horiz scroll, shift the color bit over
; makes some color flicker, is there a better way?
high_bit1:
lda COUNTL ; 3
and #$3 ; 2
cmp #2 ; 2
bne keep_high_bit1 ; 3
move_high_bit1:
; -1
lda NEXT ; 3
jmp done_high_bit1 ; 3
keep_high_bit1:
lda CURRENT ; 3
done_high_bit1:
and #$80 ; 2
sta HIGH ; 3
;===================
; 2or6: 20
; else: 18
prepare_bits1:
; get right byte, bottom 2 bits, shifted left to be in 6+5
lda NEXT ; 3
; this method 2 cycles faster than asl x 5
ror ; 2
ror ; 2
ror ; 2
ror ; 2
and #$60 ; 2
sta NEXT ; 3
;==========
; 16
output_new1:
; get current, mask off bottom 2 bits (no longer needed)
; then OR in the saved high (color) bit as well as NEXT bits
lda CURRENT ; 3
lsr ; 2
lsr ; 2
and #$1f ; 2
ora HIGH ; 3
ora NEXT ; 3
sta (OUTL),Y ; 6
iny ; 2
;===========
; 23
;============= Unroll 2
lda (OUTL),Y ; get pixel block of interest ; 5
sta CURRENT ; 3
lda (INL),Y ; get subsequent pixel block ; 5
sta NEXT ; 3
;===================
; 16
; if in bit 2 or 6 of horiz scroll, shift the color bit over
; makes some color flicker, is there a better way?
high_bit2:
lda COUNTL ; 3
and #$3 ; 2
cmp #2 ; 2
bne keep_high_bit2 ; 3
move_high_bit2:
; -1
lda NEXT ; 3
jmp done_high_bit2 ; 3
keep_high_bit2:
lda CURRENT ; 3
done_high_bit2:
and #$80 ; 2
sta HIGH ; 3
;===================
; 2or6: 20
; else: 18
prepare_bits2:
; get right byte, bottom 2 bits, shifted left to be in 6+5
lda NEXT ; 3
; this method 2 cycles faster than asl x 5
ror ; 2
ror ; 2
ror ; 2
ror ; 2
and #$60 ; 2
sta NEXT ; 3
;==========
; 16
output_new2:
; get current, mask off bottom 2 bits (no longer needed)
; then OR in the saved high (color) bit as well as NEXT bits
lda CURRENT ; 3
lsr ; 2
lsr ; 2
and #$1f ; 2
ora HIGH ; 3
ora NEXT ; 3
sta (OUTL),Y ; 6
iny ; 2
;===========
; 23
;============= Unroll 3
lda (OUTL),Y ; get pixel block of interest ; 5
sta CURRENT ; 3
@ -395,13 +598,11 @@ thirtynine:
lda OFFSCREEN ; 3
jmp done_thirtynine ; 3
not_thirtynine:
iny ; 2
lda (OUTL),Y ; get subsequent pixel block ; 5
dey ; 2
lda (INL),Y ; get subsequent pixel block ; 5
done_thirtynine:
sta NEXT ; 3
;===================
; usually: 25
; usually: 21
; rarely: 18
; if in bit 2 or 6 of horiz scroll, shift the color bit over
@ -449,14 +650,18 @@ output_new:
ora HIGH ; 3
ora NEXT ; 3
sta (OUTL),Y ; 6
;===========
; 21
iny ; 2
;===========
; 23
cpy #40 ; 2
bne hgr_scroll_line_loop ; 3
beq hgr_scroll_return ; 3
; -1
jmp hgr_scroll_line_loop ; 3
;===========
; 7
hgr_scroll_return:
; -1
rts ; 6