dranke/upkr
1
0
Fork 0
mirror of https://github.com/exoticorn/upkr.git synced 2026-01-20 19:46:42 +01:00
Code Issues Projects Releases Wiki Activity

z80_unpacker: optimisation: -3B and ~-10T in decode_bit = 174B

Browse Source 
unpack zx48.rom is now ~22.6s (from 23.0s)
(performance version is now 199 bytes, zx48.rom unpack 19.4s -> 19.0s)
This commit is contained in:
Peter Helcmanovsky (Ped)
2022-09-18 22:54:10 +02:00
parent b13fa05413
commit d4bce4bf7c
2 changed files with 15 additions and 16 deletions
Download Patch File Download Diff File Expand all files Collapse all files

BIN
z80_unpacker/example/example.sna
View File

Binary file not shown.

31
z80_unpacker/unpack.asm
View File

@@ -250,27 +250,26 @@ decode_bit:
ENDIF ENDIF
add hl,bc ; HL = state_scale * (upkr_state >> 8) + (upkr_state & 255) add hl,bc ; HL = state_scale * (upkr_state >> 8) + (upkr_state & 255)
pop af pop af ; restore prob and CF=bit
ld d,-16 ; D = -prob_offset (-16 0xF0 when bit = 0)
jr nc,.bit_is_0_2 jr nc,.bit_is_0_2
ld d,b ; D = -prob_offset (0 when bit = 1) (also does fix following ADD) dec d ; DE = -prob (also D = bit ? $FF : $00)
dec h add hl,de ; HL += -prob
add hl,de ; HL += -prob (HL += (256 - prob) - 256) ; ^ this always preserves CF=1, because (state>>8) >= 128, state_scale: 7..250, prob: 7..250,
.bit_is_0_2: ; HL = state_offset + state_scale * (upkr_state >> 8) + (upkr_state & 255) ; new upkr_state ; so 7*128 > 250 and thus edge case `ADD hl=(7*128+0),de=(-250)` => CF=1
.bit_is_0_2:
; *** adjust probs[context_index] ; *** adjust probs[context_index]
ld e,a ; D:E = -prob_offset:prob, A = prob ld e,a ; preserve prob
and $F8 rra ; + (bit<<4) ; part of -prob_offset, needs another -16
and $FC ; clear/keep correct bits to get desired (prob>>4) + extras, CF=0
rra rra
rra rra
rra rra ; A = (bit<<4) + (prob>>4), CF=(prob & 8)
rra adc a,-16 ; A = (bit<<4) - 16 + ((prob + 8)>>4) ; -prob_offset = (bit<<4) - 16
adc a,d ; A = -prob_offset + ((prob + 8) >> 4) sub e ; A = (bit<<4) - 16 + ((prob + 8)>>4) - prob ; = ((prob + 8)>>4) - prob_offset - prob
neg neg ; A = prob_offset + prob - ((prob + 8)>>4)
add a,e ; A = prob_offset + prob - ((prob + 8) >> 4)
pop bc pop bc
ld (bc),a ; update probs[context_index] ld (bc),a ; probs[context_index] = prob_offset + prob - ((prob + 8) >> 4);
add a,d ; bit=0: A = 23..249, D = 240 -> CF=1 || bit=1: D=0 -> CF=0 add a,d ; restore CF = bit (D = bit ? $FF : $00 && A > 0)
ccf ; resulting CF = bit restored
pop de pop de
ret ret