From d4bce4bf7cb1095a797ca5853c0c01fcba4f603f Mon Sep 17 00:00:00 2001
From: "Peter Helcmanovsky (Ped)" <ped@7gods.org>
Date: Sun, 18 Sep 2022 22:54:10 +0200
Subject: [PATCH] z80_unpacker: optimisation: -3B and ~-10T in decode_bit =
 174B

unpack zx48.rom is now ~22.6s (from 23.0s)
(performance version is now 199 bytes, zx48.rom unpack 19.4s -> 19.0s)
---
 z80_unpacker/example/example.sna | Bin 49179 -> 49179 bytes
 z80_unpacker/unpack.asm          |  31 +++++++++++++++----------------
 2 files changed, 15 insertions(+), 16 deletions(-)

diff --git a/z80_unpacker/example/example.sna b/z80_unpacker/example/example.sna
index 78df3d8a2cf9da82d8237271b0d006240e53706b..515f5d68690562b02a92e501cbcea636569d53df 100644
GIT binary patch
delta 175
zcmbQ;z&yKwd4u>u#tV}r7KZU{eJEYE`NZ4c^oM*3XSYtSS!ge8z$7XeFaPY1yuAFm
z50l@z9As*md~u<Hj{GAF0iLt+k4_63{H(kwBz-eL;$;QPt}F7740w3XCOv`}B>D(w
zP|{@GMM_Mnk0vi%C}n`fyo)CV|JO@!f=!b;Ee)bAKG~lJGA(|d+^G0Ve51i5mdPdi
GmjeKk{!YRG

delta 167
zcmbQ;z&yKwd4u>u#><l>7KZWdcqm=9`NZ4c^oM*3XLn4lS!gdN_Q8NTKvgpS*$;Uj
z=z8nYe2}The)74620BWQEChJYDm^+aXz;W0rjYc_0Ew3sEW55KJu=|oIh*kaVwS`s
upjjD{wH7HcX*`;|aG{h2b|WvI6#QQ=!3j1{>a;YF-rT$Bq0Hox{mTKZu}H!I

diff --git a/z80_unpacker/unpack.asm b/z80_unpacker/unpack.asm
index b7e5120..b1bcbce 100644
--- a/z80_unpacker/unpack.asm
+++ b/z80_unpacker/unpack.asm
@@ -250,27 +250,26 @@ decode_bit:
   ENDIF
 
     add     hl,bc                   ; HL = state_scale * (upkr_state >> 8) + (upkr_state & 255)
-    pop     af
-    ld      d,-16                   ; D = -prob_offset (-16 0xF0 when bit = 0)
+    pop     af                      ; restore prob and CF=bit
     jr      nc,.bit_is_0_2
-    ld      d,b                     ; D = -prob_offset (0 when bit = 1) (also does fix following ADD)
-    dec     h
-    add     hl,de                   ; HL += -prob (HL += (256 - prob) - 256)
-.bit_is_0_2:                        ; HL = state_offset + state_scale * (upkr_state >> 8) + (upkr_state & 255) ; new upkr_state
+    dec     d                       ; DE = -prob (also D = bit ? $FF : $00)
+    add     hl,de                   ; HL += -prob
+    ; ^ this always preserves CF=1, because (state>>8) >= 128, state_scale: 7..250, prob: 7..250,
+    ; 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]
-    ld      e,a                     ; D:E = -prob_offset:prob, A = prob
-    and     $F8
+    ld      e,a                     ; preserve prob
+    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
-    adc     a,d                     ; A = -prob_offset + ((prob + 8) >> 4)
-    neg
-    add     a,e                     ; A = prob_offset + prob - ((prob + 8) >> 4)
+    rra                             ; A = (bit<<4) + (prob>>4), CF=(prob & 8)
+    adc     a,-16                   ; A = (bit<<4) - 16 + ((prob + 8)>>4) ; -prob_offset = (bit<<4) - 16
+    sub     e                       ; A = (bit<<4) - 16 + ((prob + 8)>>4) - prob ; = ((prob + 8)>>4) - prob_offset - prob
+    neg                             ; A = prob_offset + prob - ((prob + 8)>>4)
     pop     bc
-    ld      (bc),a                  ; update probs[context_index]
-    add     a,d                     ; bit=0: A = 23..249, D = 240 -> CF=1 || bit=1: D=0 -> CF=0
-    ccf                             ; resulting CF = bit restored
+    ld      (bc),a                  ; probs[context_index] = prob_offset + prob - ((prob + 8) >> 4);
+    add     a,d                     ; restore CF = bit (D = bit ? $FF : $00 && A > 0)
     pop     de
     ret