Soul Blazer input lag

[Gideon Zhi]

Hi guys, I had a few minutes during lunch today and I figured I'd look into why Soul Blazer has some very noticeable input lag. It's especially noticeable if you try to use the sword strafe while also walking (L/R) and infects just about all aspects of the game's control. Brief examination of the game's code yielded this as being unique to the sword strafe on L:

Code Select Expand


$00/9D7C: 20 9D 9F     JSR $9F9D [$00:9F9D]      A:0002 X:0800 Y:0800 D:0000 DB:01 S:1FF1 P:envmxdIzc HC:138 VC:253
$00/9D7F: 02 80        COP #$80                  A:0000 X:0800 Y:0A00 D:0000 DB:01 S:1FF1 P:envmxdIZc HC:876 VC:253
$00/9D82: 02 82        COP #$82                  A:9D82 X:0800 Y:0800 D:0000 DB:01 S:1FF1 P:envmxdIZc HC:696 VC:254


A quick glance at 65816info.txt yields COP as "Coprocessor Empowerement." Loads a hardware vector from the cart header. Wait, what? Soul Blazer doesn't have any extra hardware - no DSP or SA-1 or anything like that. What the hell is going on here?

[Revenant]

The COP opcode just triggers a software interrupt exactly like BRK does, and they're both treated as two-byte instructions. The second byte can be (and sometimes is, though not commonly on the NES/SNES) used as an actual operand; in this case, the interrupt handler uses it as an index into a jump table:

COP #$80 in Soul Blazer does this (I omitted a few instructions at the beginning):

Code Select Expand

$00/D627 A3 02       LDA $02,s  [$00:1FEF]   A:0000 X:0800 Y:0800 D:0000 DB:01 S:1FED P:envmxdIZc HC:0674 VC:000 FC:38 I:00
$00/D629 3A          DEC A                   A:9D71 X:0800 Y:0800 D:0000 DB:01 S:1FED P:eNvmxdIzc HC:0728 VC:000 FC:38 I:00
$00/D62A 85 38       STA $38    [$00:0038]   A:9D70 X:0800 Y:0800 D:0000 DB:01 S:1FED P:eNvmxdIzc HC:0758 VC:000 FC:38 I:00
$00/D62C A7 38       LDA [$38]  [$00:9D70]   A:9D70 X:0800 Y:0800 D:0000 DB:01 S:1FED P:eNvmxdIzc HC:0830 VC:000 FC:38 I:00
$00/D62E E6 38       INC $38    [$00:0038]   A:2980 X:0800 Y:0800 D:0000 DB:01 S:1FED P:envmxdIzc HC:0926 VC:000 FC:38 I:00
$00/D630 29 FF 00    AND #$00FF              A:2980 X:0800 Y:0800 D:0000 DB:01 S:1FED P:eNvmxdIzc HC:0996 VC:000 FC:38 I:00
$00/D633 0A          ASL A                   A:0080 X:0800 Y:0800 D:0000 DB:01 S:1FED P:envmxdIzc HC:1036 VC:000 FC:38 I:00
$00/D634 AA          TAX                     A:0100 X:0800 Y:0800 D:0000 DB:01 S:1FED P:envmxdIzc HC:1066 VC:000 FC:38 I:00
$00/D635 7C 38 D6    JMP ($D638,x)[$00:DE7A] A:0100 X:0100 Y:0800 D:0000 DB:01 S:1FED P:envmxdIzc HC:1096 VC:000 FC:38 I:00

So it's basically just a "syscall" sort of deal (like the x86 "int" instruction), but for the SNES

[Gideon Zhi]

Alright, that makes sense. Slightly boggled by the whole "coprocessor empowerement" thing; I'd never seen COP before today and only had a few minutes in which to digest the code. I'll dig a bit further into this, but the input lag is something I'd really like to see fixed (if the possibility of an easy fix exists, anyway.)

[Vehek]

Terranigma does that too, where events are assembly mixed with event commands called by COP. If Soul Blazer's doing the same thing, then you're looking at event code.