Editing Talk:PS2 Emulation

Jump to navigation Jump to search
Warning: You are not logged in. Your IP address will be publicly visible if you make any edits. If you log in or create an account, your edits will be attributed to your username, along with other benefits.

The edit can be undone. Please check the comparison below to verify that this is what you want to do, and then publish the changes below to finish undoing the edit.

Latest revision Your text
Line 203: Line 203:
|}{{clear}}
|}{{clear}}


====Function 0x0D====
===ps2_netemu command 0x4===
What is the purpose of this function? Could it be used as a potential fix for the various DMA issues (similar to the EE timing hack in PCSX2)? I may be wrong, but I think the majority of netemu's emulation problems are caused by DMA issues, which are unfortunately hard to fix.
Patch SPE 3 program (eedma) by searching for ila r4, xxxxx, starting at 0x178A0 and replacing them with (0x42000004 | ((value << 7) & 0x1FFFF80)<br>
* This hack allow to run all spe cores while doing nothing on ppe side. Is like giving spe more time (100 msec). This can be used to fix some timing issues here or there. But if you know game offset that you want to use it, you probably can already fix it in different way. Also this probably won't affect "emulation cycles", so is not like pcsx2 EE timing hack. About pcsx2 EE timing hack.. This is really stupid hack if you ask me. Hack make all events listed [[https://github.com/PCSX2/pcsx2/blob/5bdec2f532e94065655032eb6cf7f7715c075e3b/pcsx2/R5900.h#L403 here]] to take 8 cycles. No matter that really it was 1, or 2000 cycles, it will make all of that 8 cycles. Idea of that hack is not bad itself, but it is terrible implementation that make a lot of things random. I think that ps2 emu on ps4 do this much better, as you can select only one event, and set cycles for it. While on pcsx2 if you want DMAC_FROM_IPU to take 8 cycles, you also make ALL OTHER events to take 8 cycles. I don't know how lucky that hack is to not break other stuff. --[[User:Kozarovv|Kozarovv]] ([[User talk:Kozarovv|talk]]) 07:47, 12 March 2022 (UTC)
0x42000004 is ila r4 opcode. Due to opcode encoding example result of that patch with value 0x08 will be 0x42000404 (ila r4, 0x08).
There is little bit more than that, but main purpose is just to patch SPE program behavior.
* What are the valid values? The official config from The Suffering uses a 0x8 value, yet the flashing does still happen. Increasing it to 0x20 seems to fix the flashing.--[[User:Agrippa|Agrippa]] ([[User talk:Agrippa|talk]]) 14:42, 22 February 2022 (UTC)
** 0x00 - 0x3FFFF. Well you can use higher values, but it will be truncated by mask to something below 0x40000 anyway. Default is 0x12345 if i understand correctly. --[[User:Kozarovv|Kozarovv]] ([[User talk:Kozarovv|talk]]) 16:29, 22 February 2022 (UTC)


====Function 0x12====
===ps2_netemu command 0x5===
The 0x4FC500 EE memory offset is not hardcoded, isn't it? Since this hook fixes other Traveller's Tales games, like Wrath of Cortex, the memory offset should be taken from the opcode itself (if I remember correctly, it is "load word from global pointer").--[[User:Agrippa|Agrippa]] ([[User talk:Agrippa|talk]]) 11:48, 2 October 2022 (UTC)
The external config format used by ps2_netemu.self doesnt supports command ID 0x5. But this same command with a different ID (0x4) was used by this game configs embedded inside ps2_gxemu.self: Grand Theft Auto: Liberty City Stories (SLES-54135, SLES-54136, SLUS-21423), Grand Theft Auto: Vice City Stories (SLES-54622, SLES-54623, SLUS-21590), Hunter: The Reckoning Wayward (SLES-51823), Onimusha: Dawn of Dreams (SLPM-66275), Shinseiki Evangelion: Ayanami Ikusei Keikaku with Asuka Hokan Keikaku (SLPM-65340), Tekken Tag Tournament (SLUS-20001)<br>
*Fun(?)fact, it is hardcoded.
The reason why command ID 0x5 is not supported in the config files is because the related emulator code is enabled permanently in ps2_netemu.self, so in the practise is like if all config files have this command enabled<br>
oris r0, r0, 0x204F
The command patches EEDMA SPE program (at eedma init, 0x1F77C in latest netemu) to set different handling for DIRECT/DIRECTHL VIF1 commands. Weird solution, wasn't better to just change pointers in spe program instead of patching that on init?
ori r0, r0, 0xC500.
But i think i know what happen. hook land here (on finding nemo):
lw          $s0, dword_4FC500
bnez        $s0, loc_187568
Notice that config beside 0x4FC500, set $s0 to zero. And here is another fun fact. At least on Nemo this is only place that read from that address. Making patch to 4FC500 meaningless, and not really needed. Assuming that other game doesn't use that exact address, this config is just additionally corrupting some random thing at 0x4FC500. I think that is probably the case also for other Nemo games. To be honest that config can be replaced to just nop that branch there and should do exactly the same thing. ¯\_(ツ)_/¯ --[[User:Kozarovv|Kozarovv]] ([[User talk:Kozarovv|talk]]) 12:23, 2 October 2022 (UTC)
** Hmm, or not exactly the same thing as it also check that COP0 and DMAC stuff. Anyway, i think that patch to EE address do nothing there.
*** NET config for Finding Nemo (SLES-51755) seems to have a 0x42 replacement for 0x01 hook, including the COP0 and D_STAT checks you have mentioned.--[[User:Agrippa|Agrippa]] ([[User talk:Agrippa|talk]]) 13:33, 2 October 2022 (UTC)


====Function 0x35====
===ps2_netemu command 0x0B===
''Ninkyouden: Toseinin Ichidaiki'' (SLPM-66274) is using command 0x1 with function 0x35 at 0x13ED94...it seems like its executed at the VSync function of the game. This hook is also not in gxemu judging from the table above. No immediate change with or without the command that I can tell. Any idea on what this is doing?
There is a lot of misunderstanding about that command.
* Function check some bits in GIF, but i don't know what exactly (offset 0x300/0x310 in fe SPU). When bits are 0 command end. When bits are not 0, PPU side of PS3 sleeps 256 cycles, and then check again fe SPU. That loop won't ends unless bits are 0. Just small tip about hooks that are close to VSync. This is not always related to VSync per se. Of course can be, but don't really need to. I mean, if you want to run hook in predictable place, and with similar time steps. VSync is where you hook game code then. --[[User:Kozarovv|Kozarovv]] ([[User talk:Kozarovv|talk]]) 05:31, 22 August 2022 (UTC)
** Yes, I first thought it would be VSync related, but I figured it could be something executed at every VSync like you said. Thanks for checking in on this hook. I am thinking of testing games like Tenchu: Wrath of Heaven and Street Fighter EX3 with this hook to see if anything changes (at least Tenchu's issue is directly related to GIF, SFEX3 not too sure). --[[User:Mrjaredbeta|Mrjaredbeta]] ([[User talk:Mrjaredbeta|talk]]) 00:54, 23 August 2022 (UTC)


===ps2_netemu command 0x5===
Offset seems to be dependent on read mode, is not about what media we use. This is dependent how game read data, more precisely how game read that one sector we want to patch.  
The external config format used by ps2_netemu.self doesnt supports command ID 0x5. But this same command with a different ID (0x4) was used by this game configs embedded inside ps2_gxemu.self: Grand Theft Auto: Liberty City Stories (SLES-54135, SLES-54136, SLUS-21423), Grand Theft Auto: Vice City Stories (SLES-54622, SLES-54623, SLUS-21590), Hunter: The Reckoning Wayward (SLES-51823), Onimusha: Dawn of Dreams (SLPM-66275), Shinseiki Evangelion: Ayanami Ikusei Keikaku with Asuka Hokan Keikaku (SLPM-65340), Tekken Tag Tournament (SLUS-20001). The reason why this command is not supported in the external config files is because the related emulator code is enabled permanently in ps2_netemu.self for all games


This command patches EEDMA SPE program during emulator init (0x1F77C in latest netemu) to set different handling for DIRECT/DIRECTHL VIF1 commands. Weird solution, wasn't better to just change pointers in spe program instead of patching that on init?
PCSX2 "CDVD reads" logs can help here:
* Is there any way to patch the emulator to prevent that command to apply? I wonder if it is affecting the behaviour of some games.--[[User:Agrippa|Agrippa]] ([[User talk:Agrippa|talk]]) 20:00, 11 March 2022 (UTC)
** Yeah, you can patch it on decrypted emu.
Find
91 48 00 00 90 09 00 1C 90 09 00 10 90 09 00 14 90 09 00 18 90 0b 00 1C 90 0b 00 10 90 0b 00 14 90 0b 00 18
Replace
91 48 00 00 60 00 00 00 60 00 00 00 60 00 00 00 60 00 00 00 60 00 00 00 60 00 00 00 60 00 00 00 60 00 00 00


And find
'''CDRead requested block size (CD disc):'''
E9 42 8C 20 3C 00 00 01 60 00 72 80 E9 2A 00 00
*2048 = Offset + 0x18 (skip 12 sync bytes, 4 of header, and 8 of subheader)
replace
*2328 = Offset + 0x18 (skip 12 sync bytes, 4 of header, and 8 of subheader)
4E 80 00 20 3C 00 00 01 60 00 72 80 E9 2A 00 00
*2340 = Offset + 0x0C (skip only 12 bytes of sync data)


Now you need to encrypt it with scetool, using original emu as a template.  
'''DVDRead requested block size (DVD Disc):'''
scetool --template orig_ps2_netemu.self --sce-type=SELF --compress-data=TRUE --encrypt ps2_netemu.elf ps2_netemu.self
*2064 = Offset match, but only until the 349th sector. Otherwise is offset - 0x0C because that read mode see data as ID DATA (4) + ID DATA EDC (2) + Reserved bytes (6) + 2048 data + EDC (4). Why there is some weirdness that about first sectors, no idea. Maybe it is something common for DVD discs that i'm not aware off.  
Remember to delete netemu from flash, then copy new one. Overwriting can fail as there is not enough space on dev_flash. --[[User:Kozarovv|Kozarovv]] ([[User talk:Kozarovv|talk]]) 07:35, 12 March 2022 (UTC)


===ps2_netemu command 0x0B===
"Offset + XX" for CD assume that you use Isobuster RAW mode. "Offset - XX" for DVD assume that you use Isobuster NON RAW mode (ISO can't store all data, so is missing ID/Resv bytes too.<br>
Keep in mind there is a bug in pcsx2 where fastboot "force" 2048 CD read on DVD disc for executable. That one will match 2064 read for us.


SRS:Street Racing Syndicate - sector to patch: 3021/1392496 (1.03) or 1402736 (2.00), offset correction: no, PCSX2 log:
* You are very right. I was not aware about different read modes you can specify in the sceCdRead command. That makes sense and that explains that Freekstyle issue. Regarding the whole offset misunderstanding, I know it could be confusing sometimes when you open the mounted file system through the HxD for example (only data bytes are seen). It is important to load the image file in the hex editor directly (or use the "Load image file" in HxD), or check the RAW box in the Isobuster's sector viewer.<br> When it comes to the DVD discs, I know the offset correction is somehow related to the DVD RAW 2064 bytes per sector mode. But I am not sure if it is not applied until the 349th sector precisely - it is what I noticed by looking into the Psychonauts and Street Racing Syndicate configs. The latter has got the patch data applied to the 349th sector without the 0xC correction at all. It is the farthest example I have found.--[[User:Agrippa|Agrippa]] ([[User talk:Agrippa|talk]]) 14:47, 20 February 2022 (UTC)
CDRead: Reading Sector 0003021 (008 Blocks of Size 2048) at Speed=4x(CAV) Spindle=83
* The next 0x0B sector DVD patch in ascending order is in the Ace Combat Zero: The Belkan War config (402nd sector). It does use the +0xC correction. So it is somewhere between 349th and 402nd sector the correction starts to be applied.--[[User:Agrippa|Agrippa]] ([[User talk:Agrippa|talk]]) 14:00, 22 February 2022 (UTC)
WRC: Rally Evolved - sector to patch: 3235/1792256 (1.01), offset correction: no, PCSX2 log:
CDRead: Reading Sector 0003235 (008 Blocks of Size 2048) at Speed=4x(CAV) Spindle=83
Notice the CDRead command instead of DvdRead.
* Not sure this is the case here. But PCSX2 have bug in fastboot, where it always fallback to CDRead for few first reads. To confirm you can launch game with fastboot disabled, pcsx2 now should read those sectors with proper mode. At the second hand.. Maybe that's what happen on PS3? Probably no, as far as i know netemu bios perform something similar to fullboot. But maybe, sony can break unbreakable, and this actually seems to be reasonable explanation. --[[User:Kozarovv|Kozarovv]] ([[User talk:Kozarovv|talk]]) 05:18, 22 August 2022 (UTC)
** That is how it does look like on the newest PCSX2 build at the moment of posting it. It does use a CDRead for few reads after loading an ELF file, even in the full boot mode. --[[User:Agrippa|Agrippa]] ([[User talk:Agrippa|talk]]) 17:27, 22 August 2022 (UTC)
Emulator does panic itself, when two separate 0x0B commands are used.--[[User:Agrippa|Agrippa]] ([[User talk:Agrippa|talk]]) 20:25, 15 November 2022 (UTC)


===ps2_netemu command 0x0C===
===ps2_netemu command 0x0C===
Line 262: Line 243:
*** Well that's "unfortunate" because Shadowman2 would be perfect test case here. I noticed that Shadowma2 have hardcoded bios settings "CDVD_READ_DELAY", so maybe is handled there. --[[User:Kozarovv|Kozarovv]] ([[User talk:Kozarovv|talk]]) 19:11, 4 March 2022 (UTC)
*** Well that's "unfortunate" because Shadowman2 would be perfect test case here. I noticed that Shadowma2 have hardcoded bios settings "CDVD_READ_DELAY", so maybe is handled there. --[[User:Kozarovv|Kozarovv]] ([[User talk:Kozarovv|talk]]) 19:11, 4 March 2022 (UTC)


It seems like this command can further improve FMVs that still have stuttering issues with command 0x21 enabled. The Fear's opening FMVs are heavily improved with command 0x21 set at 1, but there is still some video slowdown and audio stuttering and popping remaining. With the addition of command 0x0C set to 0x1 and 0x400, the slowdown and stutter is completely fixed. I have only tested 0x400 as a value so far.
===ps2_netemu command 0x0F===
Apparently command is combined 0x26, and 0x27. Address range is added to both 0x26, and 0x27 list.
This is probably because someone realized later that you not always need accuracy on both FPU, and COP2 to make things work, and that game speed suffer from it. So command was split into 2 separate commands, leaving combined 0x0F for backward compatibility.


The ingame FMVs with the graphic overlay are still stuttering heavily, though, and I am still unsure why. It seems like the shorter FMVs run fine, and the longer they are they more they have slowdown/stutter. This only applies to the "ingame" FMVs and not the opening ones.
0x0F use 2 list counters. From 0x26, and from 0x27. This make usage limit variable. With overall limit 31 for 0x27, and 31 for 0x26. When we have 20 0x0F commands, and 12 0x27, or 0x26. Emulator will panic as one of counters will be above allowed (31) number.


I did some work on it, and it seems to modify read speed. Higher values should give faster read speed. Eventually higher values will give more frequent event checks, but read speed modifier seems to be more probable now when we know default values. To get faster read speed we should use (1, x > 0x400) for CD, and (1, x > 0x1000) for DVD. At this point Shadowman 2 is good test candidate again, just with some really high value. All configs from GX emu looks like they slowing down reads, because values for DVD are below 0x1000, Shadowman2 without config doesn't change anything because this config use default value... Burnout 3 is interesting here, only config that seems to use 0 (yolo read speed) option. --[[User:Kozarovv|Kozarovv]] ([[User talk:Kozarovv|talk]]) 22:24, 3 July 2023 (CEST) <br>
Edit: There are some additional runtimes which check 0x26/0x0F/0x10/0x27 and 0x0E. Those runtimes check that current PC match one from configs, and return true or false. Is unknown what is purpose of that check, and even if it is really used (no xrefs, but can be ppc bctr jump). I'm suspecting that check is what make slowdown when commands are used on unsupported opcodes. This is weird if is really working, but that will explain slowdowns.. Supported opcodes perform own check for current PC, and supported command anyway. So maybe that's some kind of hint for recompiler. No idea, really.
void MECHA_on_state_3(__int64* mecha)
* In other words, the 0x0E command is equal to the 0x0F one, but it does work on a specific offset instead of range, am I right? And neither 0x0E/0x0F/0x26/0x27 should affect the mul/div accuracy, but the add/sub only.--[[User:Agrippa|Agrippa]] ([[User talk:Agrippa|talk]]) 14:26, 6 March 2022 (UTC)
{
  ReadSectorSize = *(mecha + 0x70);
  if ( !ReadSectorSize )
    panic(1u);
   
  do
    current_tb = get_current_timebase()
  while ( !current_tb );
 
  readed_sectors_count = readed_size_in_bytes / ReadSectorSize;
  readed_sectors_count_mul_by_79800000 = 79800000LL * readed_sectors_count
  tb_for_seek_and_read = *(mecha + 0x18);
  skip = current_tb - tb_for_seek_and_read >= readed_sectors_count_mul_by_79800000 / val_from_cmd_0x0C_1)
 
  if ( cmd_0x0C_0 != 1 || val_from_cmd_0x0C_1 == 0 || skip )
  {
    if ( *(mecha + 0xD4) <= BytesToRead && !*(mecha + 0xE4) )
    {
      *(mecha + 0xD4) = readed_size_in_bytes;
      *mecha_state = 4; // Reading done here. Going to next step.
      return;
    }
    panic(1u);
  }
  set_scheduler(
    5uLL,                                                                              // 5 - CDVD event
    tb_for_seek_and_read + readed_sectors_count_mul_by_79800000 / val_from_cmd_0x0C_1,  // When
    &syscall8_200,                                                                      // Event Handler.
    0LL);                                                                              // unk
}


* Netemu does crash when the first param is set to 0x2.--[[User:Agrippa|Agrippa]] ([[User talk:Agrippa|talk]]) 21:03, 7 July 2023 (CEST)
===ps2_netemu command 0x12===
** No idea why, function that check config compare to 2 and panic if greater than. So '''2''' is accepted at least there. This config is used in 4 places. All of them use next part of config only if this one is '''1'''. 3 places check explicitly for '''1''', one place check for '''0'''. Setting '''2''' is weird from code point of view because it's almost like 0, but allow code to do one check in read function, and to be honest this doesn't look like intentional behavior because similar check is performed earlier in previous N command state, and there is handled properly including setting 1F402006 to error code instead of panic. --[[User:Kozarovv|Kozarovv]] ([[User talk:Kozarovv|talk]]) 22:31, 7 July 2023 (CEST)


===ps2_netemu command 0x12===
First 8 bytes of that command are special flags. Not quite sure about bytes 5-8 yet, because at some point they are used to "andc" with first 4 bytes.
Some examples for first 4 bytes:
0x100000  = Different code path for VU0 opcodes that do ADD/SUB with multiply (MSUB, MADDA, etc.).
0x200000  = Run some additional code in VU0 load/store opcodes (ILW, LQI, ISWR, etc.)
0x400000  = Skip emu syscall 3 (3)
0x800000  = Skip emu syscall 3 (4)
0x4000000  = This flag ensure that type 2 config from cmd 0x12 will run. Otherwise it seems to be skipped.
0x8000000  = Run some additional code for VU0 DIV opcode
0x30000000 = Different code path for VU0 MUL opcodes, include opcodes like MSUB for mul part. So 0x30100000 work for mul, and sub part.


====type 1====
====type 1====
Line 338: Line 298:
seg017:00000000001984C0                bgt      cr6, next_value
seg017:00000000001984C0                bgt      cr6, next_value
</pre>
</pre>
<br>
Leaving old info for now as it have good explanation of code itself. Command patch vu0_jit_data memory directly. For example above patch will write 0x188 to 0x1C018994 address. vu0 jit mem hold info about microprograms, like how many cycles will run, where program start/end, where is ebit, etc. For now is unknown which info is patched, but shift by 3 suggest it will be vu0 address related info. --[[User:Kozarovv|Kozarovv]] ([[User talk:Kozarovv|talk]]) 12:23, 14 June 2023 (CEST)


====type 2====
====type 2====
Line 404: Line 362:
         stw      r9, 0x1238(r31) save count>>1
         stw      r9, 0x1238(r31) save count>>1
         std      r11, 0x1240(r31) save ptr to table values start
         std      r11, 0x1240(r31) save ptr to table values start
---big handler, different register settings?---


One correct entry is 2x 32 bit. First 32 bits are EE Opcode in little endian, second 32 bits are the same as first 4 bits of type 3 (interlocking). This type of config work in conjunction with type 2 config, and was required because R&C series use EE memory overlays. So type 3 config can't be used here as EE offset change per game level.
===ps2_netemu command 0x29===
 
Something related with read time, maybe seek time. First value is meant to be lower than second value, but this is not requirement.
Here is alternative config for Rayman 3 (SLUS_206.01), this time using type 4 instead of 3. Just to confirm above is true. Untested, because i have no ability to do that.
Code that use it seems to delay some read/seek operation by multiply of first, or second value depending which sector is currently read (or maybe which part of disc actually). Here is code from one of fuctions that use values from that command, keep in mind that "mecha" is just fancy name for cdvd in that emu.
<pre>
if ((75 * cdvd.CrtSecond + 4500 * cdvd.CrtMinute + cdvd.CrtFrame - 150) >= *(mecha.unk_0x60))
12 00 00 00 0C 00 00 00 00 00 00 04 00 00 00 00
    a = *(cmd_0x29_val_2);
02 00 04 00 00 00 00 00 AE B3 4E 5D 20 02 00 00
  else
46 02 00 00 04 00 04 00 00 30 D3 48 01 00 00 00
    a = *(cmd_0x29_val_1);
00 30 C8 48 01 00 00 00
   
</pre>
  b = 4835703278458516699;        // read https://munroesj52.github.io/vec__int64__ppc_8h.html (search on page for that number).
 
  c = (79800000 * a * b) >> 64;  // 0x4C1A6C0 (79800000) is value that lv1 repo key be.clock return.  
===ps2_netemu command 0x17===
d = c >> 18;                    // This and 2 above are generally used as a division by multiply.
This command is used only in Bully which is only known game by R* Vancouver, there is a chance that is only game which need it. But just in case, what game do is MTC0 $zero, Count <br>
e = get_timebase_reg();
And read that at some point. Since we are in recompiler, count is updated only on events. This can create situation where game retrieve 0 value from count reg. Which according to pcsx2 source is wrong ([[https://github.com/PCSX2/pcsx2/blob/b3590430c97e8df170df0c00ebf466ce848edbf9/pcsx2/COP0.cpp#L56 | referred as TIMR]]). To be honest I'm not 100% sure that is issue here. Only other problem i can imagine is potential divide by 0 in emu itself. Anyway, if game do MTC0 zero, Count (00 48 80 40), or any weird operations with Count. This config should be tested.
if ( e == 0 )
 
{
===ps2_netemu command 0x21===
  do
Well, i don't know where to start.. This command with param 0 or 1 at some point prevent writes to TagLo register. Not with MTC0 tho, but during different opcode that is handled separately. I got this unresolved yet, but it seems that is CACHE opcode. That potentially mean ps2_netemu support some parts of instruction cache. But this need to be tested on real hardware. Good candidate here is WRC4. When command 0x21 with 0 or 1 is added to standard config, game should fail to load at all (unless game getting lucky like described in emu bugs section).
    e = get_timebase_reg();
*I assume the cache handling would be the cause for games panicking/shutting down when 0x21 is set at 0? I cannot remember correctly if WRC panics (Agrippa would know better), but games like Shadow Hearts 2/3 will shut down after loading a save when instructions are being loaded in the EE memory from outside the main executable. My mind is blanking on other games besides those two, but surely a handful of games I have tested have had shutdowns caused by this command.--[[User:Mrjaredbeta|Mrjaredbeta]] ([[User talk:Mrjaredbeta|talk]]) 20:13, 1 October 2022 (UTC)
  while ( e == 0 );
** Yes, since that command disable cache handling only partially. There could be some unexpected issues. Problem is that 0x21 do little bit more, and little bit weird. Not gonna lie, cache is new territory for me since i never worked with any emu that support it. PCSX2 support good chunk of cache, but only in interpreter mode, and usually i just patched games to not rely on cache. This is what i got for now.  
}
*** If my memory serves me right:
**** WRC4 - panic at the loading of stages (0x21-0)
f = e - *(mecha.unk_0x24);
**** VP2 - panic at the switching to the battle mode (0x21-0) or player models fail to load in the battle mode (0x21-1)
if ( f >= d )
**** MGS3 - frame rate improvement, at the expense of awfully longer loading times (0x21-0).--[[User:Agrippa|Agrippa]] ([[User talk:Agrippa|talk]]) 11:40, 2 October 2022 (UTC)
{
**Since it’s worth mentioning, 0x21 set to 0 also fixes the camera angle issue in Radiata Stories after one or two battle sequences.--[[User:Mrjaredbeta|Mrjaredbeta]] ([[User talk:Mrjaredbeta|talk]]) 15:02, 2 October 2022 (UTC)
  MECHA_update_status(mecha);
***I think the issue is not related to the camera at all. It looks similar to the VP2 case I posted earlier - the models fail to load on the post battle screen (shadows are drawn, though). Moreover, various models do disappear for a second in the battle mode too, if I remember correctly. Just a guess, as both games share the same engine.--[[User:Agrippa|Agrippa]] ([[User talk:Agrippa|talk]]) 18:22, 4 October 2022 (UTC)
  result = unlock_sc06(0x8000LL);
I messed little bit with this code today, and 0x21_1, plus 0x03 at the same time seems to be most aggressive combo to remove instruction cache emulation. Not sure about compatibility, but this two configs together skip most cache code in emulator. --[[User:Kozarovv|Kozarovv]] ([[User talk:Kozarovv|talk]]) 10:21, 21 October 2022 (UTC)
  }
*Quick test with this, random game (ATV Offroad Fury 4). It seems like 0x21_0 still gave a better improvement to menu FMV playback/stutter than 0x21_1 and 0x03. I didn't notice any change in performance with 0x21_1 in this scenario. Caching is still enabled with IHIN/IXIN with 0x21_0?--[[User:Mrjaredbeta|Mrjaredbeta]] ([[User talk:Mrjaredbeta|talk]]) 00:15, 22 October 2022 (UTC)
**Whoops, sorry i meant 0x21_0 and 0x03. 0x21_1 actually create a lot of additional recompiled code. 0x21_0 is so fast because it skips lookup for cached addresses also outside of general cache opcodes (that how it looks like, i need to do little bit more work here). IHIN/IXIN (hit/index invalidation) can be totally disabled only with 0x03. 0x21 (0/1) still generate a lot of code for IHIN/IXIN when 0x03 is not used, even more that without 0x21. --[[User:Kozarovv|Kozarovv]] ([[User talk:Kozarovv|talk]]) 06:53, 22 October 2022 (UTC)
***Okay, that makes sense. It is difficult to tell if adding 0x03 with 0x21_0 made any difference with this game as FMV is fixed with just 0x21_0, but it may be beneficial to test games that still have FMV playback issues with 0x21_0 enabled (slight desync still with Shadow Hearts).--[[User:Mrjaredbeta|Mrjaredbeta]] ([[User talk:Mrjaredbeta|talk]]) 15:05, 22 October 2022 (UTC)
<br><br>'''Emotion Engine Cache support'''<br>
Apparently, emulator support some parts of cache. More precisely instruction cache. Implementation seems to be simple and not 100% accurate, and there is no data cache support (obviously - speed). Still impressive there is something at all in recompiler.
 
'''Supported opcodes:'''
{| border="1" cellspacing="0" cellpadding="5" border="#999" class="wikitable" style="border:1px solid #999; border-collapse: collapse;"
|-
! Opcode || Info
|-
|r5900 CACHE BXLBT || Unsupported, store 0 on COP0 TagLo and TagHi
|-
|r5900 CACHE DXLDT ||  Unsupported, store 0 on COP0 TagLo
|-
|r5900 CACHE DXLTG || Unsupported, store 0 on COP0 TagLo
|-
|r5900 CACHE IXLDT ||  Unsupported, store 0 on COP0 TagLo
|-
|r5900 CACHE IXLTG ||  Supported, cmd 0x21 (0,1) skip function and store 0 on COP0 TagLo
|-
|r5900 CACHE DXWBIN || Not supported at all, empty function.
|-
|r5900 CACHE DXIN ||  Not supported at all, empty function.
|-
|r5900 CACHE DHWBIN || Not supported at all, empty function.
|-
|r5900 CACHE DHIN  ||  Not supported at all, empty function.
|-
|r5900 CACHE DHWOIN || Not supported at all, empty function.
|-
|r5900 CACHE IHIN  ||  Full support, cmd 0x21 (0,1) change path of that opcode, 0x03 practically skip whole op.
|-
|r5900 CACHE IXIN  ||  Full support, cmd 0x21 (0,1) change path of that opcode, 0x03 practically skip whole op.
|-
|r5900 CACHE misc  ||  Do something, but function is little bit complicated to tell for now that it is something that make sense.
|}
--[[User:Kozarovv|Kozarovv]] ([[User talk:Kozarovv|talk]]) 21:03, 1 October 2022 (UTC)
 
Example of recompiled code with enabled cache emulation: https://pastebin.com/sDwc1MLX
* Every operation with r15 and following r7/r8 loads, compares, and cr operations will be skipped by emitter with 0x21 --> 0
* r15 is loaded with current address (page?), to get full value shift it left by 6 eg.  0x808710 << 6 = 0x2021C400 (shift left by 6 = multiplying by 0x40)
* For small part of code posted on pastebin, cache check is performed 33 times * 8 opcodes = 264 opcodes from 801 are cache checks! A lot of work, a lot of code mostly for nothing.
* Conclusion. Use 0x21 --> 0 whenever you can. iCache emulation is pointless for 99% of games.
* Small oftopic. Every opcode that write to r13 increase emulated cycles. Every opcode writing to r14 modify current r5900 pc. Every branch to 0x3800 mean that event test will be performed.
* Problematic games (have issues with 0x21 enabled): Shadow Hearts 3, Jak & Daxter, VP2, Hot Shots series, Kingdom Hearts. Any game that load additional code overlays and do that custom way, for example Shadow Hearts 3 load code from bin files hidden in cpx files. All those games doesn't really need iCache emulation to work, but emulator seems to have issue with code invalidation. Only small amount of games used this method of loading, so it still should be fine for 99% of games.
--[[User:Kozarovv|Kozarovv]] ([[User talk:Kozarovv|talk]]) 07:32, 22 May 2023 (CEST)
* Radiata does not crash. English patch for FFXII IZJS does crash though (0x21 fixes stuttering in turbo mode).--[[User:Agrippa|Agrippa]] ([[User talk:Agrippa|talk]]) 20:33, 22 May 2023 (CEST)
** Whoops, my bad. Fixed. :P --[[User:Kozarovv|Kozarovv]] ([[User talk:Kozarovv|talk]]) 09:44, 23 May 2023 (CEST)
*** Actually, the icache invalidation issues could explain why the infamous Maori VP2 overlay anti-protection patch is not working on the netemu.--[[User:Agrippa|Agrippa]] ([[User talk:Agrippa|talk]]) 22:14, 15 June 2023 (CEST)
**** His patch is basically self modifying code, so its possible. You can try cmd 0x01 with 0x0E subcommand hook at 0x01feb1c0. This invalidate whole mem every time offset is hit, ensuring that new code that is written from patcher is not in conflict with cache. Also remember that 0x42 is applied only on syscall 7, so if game modify 0x1feb1c0 - 0x1feb2e0 region, it will wipe patcher. And games often memzero whole data memory right after start. This issue don't exist on pcsx2 because it will reapply it on next vsync, while on netemu only 0x09/0x0A patches are reapplied. --[[User:Kozarovv|Kozarovv]] ([[User talk:Kozarovv|talk]]) 15:37, 16 June 2023 (CEST)
 
===ps2_netemu command 0x22===
Weird command. Sets something 1 (CDVD/MECHA), but seems to never use it.
 
===ps2_netemu command 0x28===
This command is heavily stripped version of command from gxemu. On netemu only option that seems to change anything is 0. From what i gathered in emu, when set to 0 SEEK ncmd will to be ignored and instant command ack is set. When 0 is used cdvd.SeekToSector is not updated from NCMD params. For emulated vm this mean next read command will perform seek to sector from position where last read end. Not sure if i'm understanding that correctly, it seems to be very hacky solution.
 
After some more work, its either skip seek, or make seek instant. Still a hack, but may give some loading boosts here and there (and break a lot of stuff too).
 
===ps2_netemu command 0x2A===
Any idea what this command is doing? It seems to be directly related to IPU. I thought it only fixed black screen issues on certain games, but it also fixes Mystery Mayhem’s glitching/freezing/stuttering FMVs.
* This config skip check of another value, which finally make one of functions to never run. This affect "SYS" thread, and value that is checked seems to be related to EE/COP2 recompilers (pure guess by addresses that write there...). Nothing related directly to IPU, that's for sure. But this is part of emu that i have basically untouched, so i can't tell what its really skipped here. Function that is skipped trigger other functions which read cached mips code. So maybe something related to recompiler flushing, or constant propagation. Hard to tell at this point. For sure this is something i want to look more into, but not yet. --[[User:Kozarovv|Kozarovv]] ([[User talk:Kozarovv|talk]]) 05:56, 8 July 2022 (UTC)
** Gotcha. So, I guess it is indirectly affecting the IPU? Or at least something EE/COP2 touches after IPU stuff. The current games that this command fixes are: All Star Baseball 2004 (GX config), Scooby-Doo! Mystery Mayhem (fixes repeating audio and FMV slowdown), Pro Evolution Soccer 6 (PAL v2.00 and NTSC-J only, unneeded in other versions, fixes black screen after opening FMV), and SCORE International Baja 1000 (fixes black screen after career mode FMV). These are all the currently known uses of the command. --[[User:Mrjaredbeta|Mrjaredbeta]] ([[User talk:Mrjaredbeta|talk]]) 02:15, 10 July 2022 (UTC)
 
===ps2_netemu command 0x2E===
Without this command applied there is a black screen and no sound after the PS2LOGO, but the game (Growlanser Generations) is working in the background. Pressing the "PS" button fixes it. After applying this command (0x172 parameter) everything works correctly.
* this (like most "mecha" commands) is messing with timing.
lwz      r0, 0x2C(r31)           # cmd_0x2E
  cmpwi    cr7, r0, 0
beq      cr7, cdvd_error_1349B4  # skip cdvd error if cmd not 0.
nop
   
   
  loc_134988:
  else
  mftb      r9
  {
cmpwi     cr7, r9, 0
     do
beq      cr7, loc_134988        # Loop until timebase is not 0
    e = get_timebase_reg();
clrldi   r0, r0, 32
   while ( e == 0 );
add      r0, r0, r9
  *(mecha.unk_20) = d - f + e;
li        r9, 5
  *mecha.unk_00 = 5;
std      r0, 0x20(r31)         # cmd_0x2E + timebase
  result = unlock_sc06(0x8000LL);
stw      r9, 0(r31)
}
b        end_134844
Value from 0x20(r31) is later used in compare. That result in cdvd error, or in setting which seems schedule event to happen after time from timebase pass. This event is netemu syscall 8 (0x200) which is related to all ps2 cdvd reads. Tl;dr is that value give emulator some more time before cdvd error. Weird thing is that PS button fix it.. --[[User:Kozarovv|Kozarovv]] ([[User talk:Kozarovv|talk]]) 07:05, 7 March 2022 (UTC)


===ps2_gxemu command 0x19===
===ps2_netemu command 0x4D===
Command disable reads/writes to selected IOP HW registers. Disabled regs are:
Ok, i don't get that config. Here is what happen in assembly:
 
0xD820  ilhu     r19, 0x7FFF
* 0x10000000 - 0x10003FFF (DEV9/SPEED/ATA/SMAP)
  0xD824  lqr      r20, Q_cfg_0x4D  ; 0x3F800000 in wild arms
* 0x1F801460 - 0x1F80147F (Unknown)
  0xD82C  iohl      r19, 0xFFFF
 
  0xD834  and      r17, r80, r19    ; r17 = Q & 0x7FFFFFFF mask
Command is auto applied to titles listed here (duplicates are really there in emu too):
  0xD840  ceqi      r15, r17, 0      ; r15 = r17 (shortcut to move 0 or value if exist to r15)
     Hash          ID                Name
  0xD844  lqr      r10, ST_Q
  0x0CD1298155  SLES_540.13  World Snooker Championship 2007
  0xD84C  cwd      r9, 0x30+var_30+8(sp)
  0x12C93199A5  SLES_518.40  NHL Hitz Pro
  0xD850  rotqbyi  r16, r20, 4      ; load mask from config to r16
  0x15C93199AD  SLES_549.13  Pro Evolution Soccer 2008
  0xD858  and       r12, r15, r16    ; tempQ & 0x3F800000 (r15 and with mask from cfg 0x3F800000)
  0x24D92589A5  SLUS_211.86  NBA Ballers - Phenom
  0xD860  or        r5, r80, r12    ; or r80(Q) with r12(Q masked with 0x3F800000)
0x2CD12D8125  SLUS_212.35  MLB 2K6
  0xD868  shufb    r7, r5, r10, r9 ; Prepare correct write for Q (r5 stored to r10 + 8)
0x34C9359935  SLUS_211.38  X-Men Legends II - Rise of Apocalypse
  0xD870  stqr      r7, ST_Q        ; write result as Q value in STQ
0x34C93599E5  SLUS_211.28  Blitz - The League
I removed irrelevant code that setup RGBA for readability, its not affecting Q. So my point is that all that masked Q is finally ored with r80. So with whole untouched Q value. Doen't that make all those operations irrelevant, or i made some mistake here?<br>
0x34C93599E5  SLUS_211.28  Blitz - The League
This config can be quite important because it should help to fix issues like Galerians Ash without dirty static patches. More games affected: [[https://github.com/PCSX2/pcsx2/issues/5137 | List]]
0x449961C9E5  SLES_542.10  NBA 2K7
0x4C9169C1CD  SLES_542.46  FIFA '07
0x4C9169C1D5  SLES_542.45  NHL '07
0x4C9169C1DD  SLES_542.44  FIFA '07
0x4C9169C1E5  SLES_542.43  FIFA '07
0x4C9169C1F5  SLES_542.41  FIFA '07
0x4C9169C1FD  SLES_542.40  FIFA '07
0x4CB14DE12D  SLUS_213.74  Marvel - Ultimate Alliance
0x54A955F915  SLUS_212.74  Outrun 2006 - Coast 2 Coas
0x5CA15DF165  SLUS_213.01  World Series of Poker
0x5CA15DF1FD  SLUS_212.86  WWE SmackDown! vs RAW 2006
0x5CA15DF1FD  SLUS_212.86  WWE SmackDown! vs RAW 2006
0x649965C94D  SLUS_214.63  Tom Clancy's Ghost Recon 2
0x649965C955  SLUS_214.60  NBA Live '07
0x649965C95D  SLUS_214.61  NASCAR '07
0x649965C965  SLUS_214.58  NHL '07
0x649965C96D  SLUS_214.59  NCAA Football '07
0x6BB149E15D  SLES_531.04  Tom Clancy's Rainbow Six - Lockdown
0x6C916DC165  SLUS_214.91  World Series of Poker - Tournament of Champions
0x6C916DC1A5  SLUS_214.83  Tiger Woods PGA Tour '07
0x6C916DC1AD  SLUS_214.82  NFL Street 3
0x6C916DC1B5  SLUS_214.81  NCAA March Madness '07
0x6C916DC1D5  SLUS_214.77  Madden NFL '07 [Hall of Fame Edition]
0x6C916DC1DD  SLUS_214.76  Madden NFL '07
0x748975D9DD  SLUS_213.83  Fight Night - Round 3
0x7C817DD125  SLUS_214.33  FIFA Soccer '07
0x7C817DD165  SLUS_214.25  NHL 2K7
0x7C817DD16D  SLUS_214.24  NBA 2K7
0x7C817DD175  SLUS_214.27  WWE SmackDown! vs RAW 2007
0x7C817DD1CD  SLUS_214.12  World Championship Poker featuring Howard Lederer - All-In
0x84798529BD  SLUS_205.65  Champions of Norrath
0x8559A109AD  uuunnnnnkkk 
0x8579852915  SLUS_215.68  Arena Football - Road to Glory
0x8579852965  SLUS_215.82  MVP '07 - NCAA Baseball
0x8D51A90145  SLES_545.11  UEFA Champions League
0x8D51A901B5  SLES_545.13  UEFA Champions League
0x8D51A901BD  SLES_545.12  UEFA Champions League
0x8D718D21BD  SLUS_216.20  NCAA Football '08
0x9C619D31E5  SLUS_205.41  NBA Ballers
0x9D41B911AD  SLES_544.48  World Series of Poker - Tournament of Champions
0x9D619D31C5  SLUS_215.61  Major League Baseball 2K7
0x9F29357805  SCUS_975.44  NBA '07 featuring The Life Vol.2
0x9F293578E5  SCUS_975.56  MLB '07 - The Show
0xB549B51915  SLUS_216.38  Madden NFL '0
  0xB549B51925  SLUS_216.32  NHL 2K8
  0xB549B5195D  SLUS_216.47  NHL '08
  0xB549B519A5  SLUS_216.48  FIFA Soccer '08
  0xB549B519AD  SLUS_216.49  NBA Live '08
0xBC61793025  SCES_532.85  Ratchet - Gladiator
0xBD41BD1105  SLUS_216.69  NBA 2K8
0xC439C569F5  SLUS_208.20  Tom Clancy's Ghost Recon - Jungle Storm
0xC7716D20D5  SCUS_974.01  Hot Shots Golf FORE!
0xC7716D20D5  SCUS_974.01  Hot Shots Golf FORE!
0xCA11E941F5  SLES_516.97  SSX 3
0xCF7965285D  SCUS_973.53  Ratchet and Clank - Up Your Arsenal
0xCF7965285D  SCUS_973.53  Ratchet and Clank - Up Your Arsenal
  0xD20911582D  SCES_515.93  Hardware Online Arena [Beta, Promo & Full Release]
  0xD7617D308D SCUS_973.28  Gran Turismo 4
  0xE339C1695D  SLES_525.45  Star Wars Battlefront
0xE794CCB06D  PCPX_980.42  Minna no Tennis
0xEA3129608D  SCES_515.78  Network Access Disc [Original, v4.02 & v4.03]
0xEC11ED4115  SLUS_209.73  Champions - Return to Arms
0xEF594508D5  SCUS_975.00  MLB '06 - The Show
0xF409F559AD  SLUS_208.89  MLB Slugfest - Loaded
0xF7415D10E5  SCUS_974.65  Ratchet - Deadlocked
0xF7415D10E5  SCUS_974.65  Ratchet - Deadlocked
As a result config disable HDD/NET capabilities.


==XMB messages related with PS2 Emulation==
==XMB messages related with PS2 Emulation==
Line 689: Line 507:
* Without Factory Service Mode : gives "Incompatible Data" when inserting PS2 disc
* Without Factory Service Mode : gives "Incompatible Data" when inserting PS2 disc


* When enabling [https://web.archive.org/web/*/http://ps3devwiki.com/files/devtools/lv2-v9-pkg/ LV2Patcher] without factory service mode (patch4 set as http://pastie.org/private/jp8zhvuocjz95cfrjm0uzg) : no changes in XMB:Game (still only PS upscaler/smoothing, no PS2 mention at all)
* When enabling [http://www.ps3devwiki.com/files/devtools/lv2-v9-pkg/ LV2Patcher] without factory service mode (patch4 set as http://pastie.org/private/jp8zhvuocjz95cfrjm0uzg) : no changes in XMB:Game (still only PS upscaler/smoothing, no PS2 mention at all)


* When enabling [https://web.archive.org/web/*/http://ps3devwiki.com/files/devtools/lv2-v9-pkg/ LV2Patcher] without factory service mode (patch4 set as http://pastie.org/4355919) : gives XMB:Game PS2 smoothing/upscaling options, it also make an inserted disk to be seen as PS2 format. Still same problem of ¨incompatible title¨ and loss of BT/settings. Also after returning to XMB, it no longer sees the disc as PS2 format but as incompatible data (which suggests the lv2 patch is undone, as lv2 is reloaded when returning from the ps2 lpar)
* When enabling [http://www.ps3devwiki.com/files/devtools/lv2-v9-pkg/ LV2Patcher] without factory service mode (patch4 set as http://pastie.org/4355919) : gives XMB:Game PS2 smoothing/upscaling options, it also make an inserted disk to be seen as PS2 format. Still same problem of ¨incompatible title¨ and loss of BT/settings. Also after returning to XMB, it no longer sees the disc as PS2 format but as incompatible data (which suggests the lv2 patch is undone, as lv2 is reloaded when returning from the ps2 lpar)


* Using [https://web.archive.org/web/*/http://ps3devwiki.com/files/OtherOSplusplus/misc/boot_ps2.pkg boot_ps2.pkg] without factory service mode : no resetting of date/time/displayoutput (still output on mainscreen), but all connection to any bound bluetooth device is lost, even when connected via USB (need PS button reactivation), and after a long while comes up with the message that the title is not compatible and that the ps3 needs to be updated (Basic nag screen that is on BC PS3s when inserting a noncompatible title).  
* Using [http://www.ps3devwiki.com/files/OtherOSplusplus/misc/boot_ps2.pkg boot_ps2.pkg] without factory service mode : no resetting of date/time/displayoutput (still output on mainscreen), but all connection to any bound bluetooth device is lost, even when connected via USB (need PS button reactivation), and after a long while comes up with the message that the title is not compatible and that the ps3 needs to be updated (Basic nag screen that is on BC PS3s when inserting a noncompatible title).  


* With Factory Service Mode enabled (there are no Xmb options to combinetest with [https://web.archive.org/web/*/http://ps3devwiki.com/files/devtools/lv2-v9-pkg/ LV2Patcher] or [https://web.archive.org/web/*/http://ps3devwiki.com/files/OtherOSplusplus/misc/boot_ps2.pkg boot_ps2.pkg]): gives ´PS2 disc´ detected at disc icon, but starting gives: resetting of date/time/displayoutput (effectively disabling my mainscreen), then all connection to any bound bluetooth device is lost, even when connected via USB (needs multiple PS button reactivation), and after a long while comes up with the message that the title is not compatible and that the ps3 needs to be updated (Basic nag screen that is on BC PS3s when inserting a noncompatible title).
* With Factory Service Mode enabled (there are no Xmb options to combinetest with [http://www.ps3devwiki.com/files/devtools/lv2-v9-pkg/ LV2Patcher] or [http://www.ps3devwiki.com/files/OtherOSplusplus/misc/boot_ps2.pkg boot_ps2.pkg]): gives ´PS2 disc´ detected at disc icon, but starting gives: resetting of date/time/displayoutput (effectively disabling my mainscreen), then all connection to any bound bluetooth device is lost, even when connected via USB (needs multiple PS button reactivation), and after a long while comes up with the message that the title is not compatible and that the ps3 needs to be updated (Basic nag screen that is on BC PS3s when inserting a noncompatible title).


In short: [https://web.archive.org/web/*/http://ps3devwiki.com/files/OtherOSplusplus/misc/boot_ps2.pkg boot_ps2.pkg] and Factory Service Mode seem to enable simulare (it tries to boot it) while [https://web.archive.org/web/*/http://ps3devwiki.com/files/OtherOSplusplus/misc/boot_ps2.pkg boot_ps2.pkg] gives you more options e.g. using [https://web.archive.org/web/*/http://ps3devwiki.com/files/devtools/lv2-v9-pkg/ LV2Patcher].
In short: [http://www.ps3devwiki.com/files/OtherOSplusplus/misc/boot_ps2.pkg boot_ps2.pkg] and Factory Service Mode seem to enable simulare (it tries to boot it) while [http://www.ps3devwiki.com/files/OtherOSplusplus/misc/boot_ps2.pkg boot_ps2.pkg] gives you more options e.g. using [http://www.ps3devwiki.com/files/devtools/lv2-v9-pkg/ LV2Patcher].
Perhaps hardswapping out all the dev_flash ps2 emu files for the same software only emulator would circumvent the 'incompatible title' message.
Perhaps hardswapping out all the dev_flash ps2 emu files for the same software only emulator would circumvent the 'incompatible title' message.




==== Second test: FW 2.70/3.15 ====  
==== Second test: FW 2.70/3.15 ====  
Silent Hill : gives disk icon "unsupported data" and error message like "This model of the PS3 system is not compatible with Playstation2 format software" when run via disc icon. Using [https://web.archive.org/web/*/http://ps3devwiki.com/files/OtherOSplusplus/misc/boot_ps2.pkg boot_ps2.pkg] gives title not supported error message like "This title is not currently compatible with the PS3 system".
Silent Hill : gives disk icon "unsupported data" and error message like "This model of the PS3 system is not compatible with Playstation2 format software" when run via disc icon. Using [http://www.ps3devwiki.com/files/OtherOSplusplus/misc/boot_ps2.pkg boot_ps2.pkg] gives title not supported error message like "This title is not currently compatible with the PS3 system".


==== Third test: FW 3.55 OtherOS++22GB (with SS Patches) ====  
==== Third test: FW 3.55 OtherOS++22GB (with SS Patches) ====  
Silent Hill : gives disk icon "unsupported data" and error message like "This model of the PS3 system is not compatible with Playstation2 format software" when run via disc icon. Using [https://web.archive.org/web/*/http://ps3devwiki.com/files/OtherOSplusplus/misc/boot_ps2.pkg boot_ps2.pkg] gives blackscreen lockup, not reacting on PS button, or powerbutton, requiring removing powercord.
Silent Hill : gives disk icon "unsupported data" and error message like "This model of the PS3 system is not compatible with Playstation2 format software" when run via disc icon. Using [http://www.ps3devwiki.com/files/OtherOSplusplus/misc/boot_ps2.pkg boot_ps2.pkg] gives blackscreen lockup, not reacting on PS button, or powerbutton, requiring removing powercord.




Line 904: Line 722:
ID match US release of Crazy Taxi. This id is kinda special, because Swap Magic CD version, and some other Datel products like Action Replay use Crazy Taxi TOC in their retail discs.
ID match US release of Crazy Taxi. This id is kinda special, because Swap Magic CD version, and some other Datel products like Action Replay use Crazy Taxi TOC in their retail discs.
Is known that they literally ripped part of disc (with key/logo, and TOC), and frankesteined it with own products.
Is known that they literally ripped part of disc (with key/logo, and TOC), and frankesteined it with own products.
So mentioned check first compare hash, and if that match, then run function that perform another check at disc sector 267559 (0x41527), so exactly where main executable is.  
So mentioned check first compare hash, and if that match, then run function that perform another check at disc sector 267559 (0x41527), so exactly where main executable is.
I didn't figured out what next, but this is probably anti AR/Datel/SM check. What's weird, there seems to be nothing for TimeSplitters2 which if i recall correctly was used for DVD version of Swap Magic.
I didn't figured out what next, but this is probably anti AR/Datel/SM check. What's weird, there seems to be nothing for TimeSplitters2 which if i recall correctly was used for DVD version of Swap Magic.  
<br>There is also check for 0xE794CCB06D  PCPX_980.42  Minna no Tennis entry, emu also refuse to boot it, and throw the same error as with SwapMagic/SCED discs.


==CDVD Commands==
==CDVD Commands==
Line 1,088: Line 905:
Every "mechacon_auth" command return zeroed result with different size. Only exception here is 0x81 which return 1.
Every "mechacon_auth" command return zeroed result with different size. Only exception here is 0x81 which return 1.
</pre>
</pre>
==r5900 regs memory map==
Registers are updated only when recompiler flush them to memory, if there is pending operation on the same register in next instruction, recompiler have no need to update reg before that. So it flush result there.
===Netemu===
Address in memory |  r5900 register
FFFFFFFFFFFF8980  |  5900_zero_upper64bits
FFFFFFFFFFFF8988  |  5900_zero_lower64bits
FFFFFFFFFFFF8990  |  5900_at_reg_upper64
FFFFFFFFFFFF8998  |  5900_at_reg_lower64
FFFFFFFFFFFF89A0  |  5900_v0_reg_upper64
FFFFFFFFFFFF89A8  |  5900_v0_reg_lower64
FFFFFFFFFFFF89B0  |  5900_v1_reg_upper64
FFFFFFFFFFFF89B8  |  5900_v1_reg_lower64
FFFFFFFFFFFF89C0  |  5900_a0_reg_upper64
FFFFFFFFFFFF89C8  |  5900_a0_reg_lower64
FFFFFFFFFFFF89D0  |  5900_a1_reg_upper64
FFFFFFFFFFFF89D8  |  5900_a1_reg_lower64
FFFFFFFFFFFF89E0  |  5900_a2_reg_upper64
FFFFFFFFFFFF89E8  |  5900_a2_reg_lower64
FFFFFFFFFFFF89F0  |  5900_a3_reg_upper64
FFFFFFFFFFFF89F8  |  5900_a3_reg_lower64
FFFFFFFFFFFF8A00  |  5900_t0_reg_upper64
FFFFFFFFFFFF8A08  |  5900_t0_reg_lower64
FFFFFFFFFFFF8A10  |  5900_t1_reg_upper64
FFFFFFFFFFFF8A18  |  5900_t1_reg_lower64
FFFFFFFFFFFF8A20  |  5900_t2_reg_upper64
FFFFFFFFFFFF8A28  |  5900_t2_reg_lower64
FFFFFFFFFFFF8A30  |  5900_t3_reg_upper64
FFFFFFFFFFFF8A38  |  5900_t3_reg_lower64
FFFFFFFFFFFF8A40  |  5900_t4_reg_upper64
FFFFFFFFFFFF8A48  |  5900_t4_reg_lower64
FFFFFFFFFFFF8A50  |  5900_t5_reg_upper64
FFFFFFFFFFFF8A58  |  5900_t5_reg_lower64
FFFFFFFFFFFF8A60  |  5900_t6_reg_upper64
FFFFFFFFFFFF8A68  |  5900_t6_reg_lower64
FFFFFFFFFFFF8A70  |  5900_t7_reg_upper64
FFFFFFFFFFFF8A78  |  5900_t7_reg_lower64
FFFFFFFFFFFF8A80  |  5900_s0_reg_upper64
FFFFFFFFFFFF8A88  |  5900_s0_reg_lower64
FFFFFFFFFFFF8A90  |  5900_s1_reg_upper64
FFFFFFFFFFFF8A98  |  5900_s1_reg_lower64
FFFFFFFFFFFF8AA0  |  5900_s2_reg_upper64
FFFFFFFFFFFF8AA8  |  5900_s2_reg_lower64
FFFFFFFFFFFF8AB0  |  5900_s3_reg_upper64
FFFFFFFFFFFF8AB8  |  5900_s3_reg_lower64
FFFFFFFFFFFF8AC0  |  5900_s4_reg_upper64
FFFFFFFFFFFF8AC8  |  5900_s4_reg_lower64
FFFFFFFFFFFF8AD0  |  5900_s5_reg_upper64
FFFFFFFFFFFF8AD8  |  5900_s5_reg_lower64
FFFFFFFFFFFF8AE0  |  5900_s6_reg_upper64
FFFFFFFFFFFF8AE8  |  5900_s6_reg_lower64
FFFFFFFFFFFF8AF0  |  5900_s7_reg_upper64
FFFFFFFFFFFF8AF8  |  5900_s7_reg_lower64
FFFFFFFFFFFF8B00  |  5900_t8_reg_upper64
FFFFFFFFFFFF8B08  |  5900_t8_reg_lower64
FFFFFFFFFFFF8B10  |  5900_t9_reg_upper64
FFFFFFFFFFFF8B18  |  5900_t9_reg_lower64
FFFFFFFFFFFF8B20  |  5900_k0_reg_upper64
FFFFFFFFFFFF8B28  |  5900_k0_reg_lower64
FFFFFFFFFFFF8B30  |  5900_k1_reg_upper64
FFFFFFFFFFFF8B38  |  5900_k1_reg_lower64
FFFFFFFFFFFF8B40  |  5900_gp_reg_upper64
FFFFFFFFFFFF8B48  |  5900_gp_reg_lower64
FFFFFFFFFFFF8B50  |  5900_sp_reg_upper64
FFFFFFFFFFFF8B58  |  5900_sp_reg_lower64
FFFFFFFFFFFF8B60  |  5900_fp_reg_upper64
FFFFFFFFFFFF8B68  |  5900_fp_reg_lower64
FFFFFFFFFFFF8B70  |  5900_ra_reg_upper64
FFFFFFFFFFFF8B78  |  5900_ra_reg_lower64
FFFFFFFFFFFF9100  |  5900_pc_reg_lower64
Below FFFFFFFFFFFF8B78 you can find COP0, and finally VFxx, and VIxx regs. Todo..


==EE I/O Handlers list==
==EE I/O Handlers list==
Line 1,678: Line 1,424:
|-
|-
|}
|}
1000F800 to 1000F8B0 seems to be some fake regs for testing purposes. Probably not existing on real PS2. Addresses not described below are handled as a unk rw. Which mean they return 0 on read, and do nothing on write.
1000F800 to 1000F8B0 seems to be some fake regs for testing purposes. Probably not existing on real PS2.  
* 1000F800 do nothing (blr). Have own handler, which suggest it was used for something earlier. [r/w]
* 1000F820 return "DrJock TV Quiz P"
* 1000F810 get current ppc time base using mftb opcode, if tb == 0 then loop until its not. Could be useful, tb is used also in scheduler. [r only] 
* 1000F830 return "hD bags few lynx"
* 1000F820 return "DrJock TV Quiz P" [r only]
* 1000F830 return "hD bags few lynx" [r only]
That make string "DrJock TV Quiz PhD bags few lynx" - This is perfect summary of Sony work. Since correct pangram should use "MrJock". So even here they made mistake.  
That make string "DrJock TV Quiz PhD bags few lynx" - This is perfect summary of Sony work. Since correct pangram should use "MrJock". So even here they made mistake.  
* 1000F860 seems to return halfword of current pad state (0xFFFF if there is no button input currently). [r only]
* 1F00F880 return hardcoded value of 0x4457, which match emu revision i'm working on. Can be just coincidence.
* 1000F880 return hardcoded value of 0x4457, which match emu revision i'm working on. Can be just coincidence. [r only]
* 1000F8B0 return unk value. Seems to be related to fe spe. [r only]


==Random notes about SPE in ps2_netemu==
==Random notes about SPE in ps2_netemu==


===IOP SPE===
===IOP SPE===
This
SPE run not only dma related stuff, but also fully fledged r3000 instruction interpreter (yes interpreter, not recompiler).
<pre>
  opcode      | address
--------------+-------- 
r3000_ADDI      0x317C0
r3000_ADDIU    0x31800
r3000_SLTI      0x31840
r3000_SLTIU    0x31880
r3000_ANDI      0x318C0
r3000_ORI      0x31900
r3000_XORI      0x31940
r3000_LUI      0x31980
r3000_ADD      0x319C0
r3000_ADDU      0x31A00
r3000_SUB      0x31A40
r3000_SUBU      0x31A80
r3000_SLT      0x31AC0
r3000_SLTU      0x31B00
r3000_AND      0x31B40
r3000_OR        0x31B80
r3000_XOR      0x31BC0
r3000_NOR      0x31C00
r3000_SLL      0x31C40
r3000_unk1      0x31C80 Primary opcode field (Bit 26..31) = 0x3E (debug stuff, not existing on real r3000)
r3000_unk2      0x31CC0 Primary opcode field (Bit 26..31) = 0x3F (debug stuff, not existing on real r3000)
r3000_SRL      0x31D00
r3000_SRA      0x31D40
r3000_SLLV      0x31D80
r3000_SRLV      0x31DC0
r3000_SRAV      0x31E00
r3000_MULT      0x31E40
r3000_MULTU    0x31F00
r3000_DIV      0x31F80
r3000_MFHI      0x32080
r3000_MFLO      0x320C0
r3000_MTHI      0x32100
r3000_MTLO      0x32140
r3000_J        0x32180
r3000_JAL      0x32200
r3000_JR        0x32280
r3000_JALR      0x322C0
r3000_BEQ      0x32300
r3000_BNE      0x32340
r3000_BLEZ      0x32380
r3000_BGTZ      0x323C0
r3000_BcondZ    0x32400
r3000_SYSCALL  0x32480
r3000_BREAK    0x324C0
r3000_COP_bad  0x32500
r3000_COP0      0x32540
r3000_bad_op    0x32740
r3000_LB        0x32840
r3000_LH        0x32940
r3000_LW        0x32A40
r3000_LBU      0x32B80
r3000_LHU      0x32C80
r3000_LWL      0x32D80
r3000_LWR      0x32E80
r3000_SB        0x32F80
r3000_SH        0x33080
r3000_SW        0x33180
r3000_SWL      0x33300
r3000_SWR      0x33400
Addresses above ARE NOT the ones in emulator memory, this is local storage address in IOP SPE program!
Opcodes LWCx 0-3, and SWCx 0-3 are not supported at all.
Same goes for general COP 1-3 opcodes, but those cause COPx unusable r3000 exception.
While LWC/SWC are triggering bad opcode exception.
Not sure if LWC0/SWC0 should be allowed.
COP0 is working in PS2 mode, so those probably shouldn't cause exception.
At the second hand i don't know single IOP module that use that instruction.
</pre>
This is unconfirmed by any code reversing for now, but IOP emulator print messages like:  
This is unconfirmed by any code reversing for now, but IOP emulator print messages like:  
  Cache write (IOPADDR/LSADDR/SIZE)
  Cache write (IOPADDR/LSADDR/SIZE)
  Cache read  (IOPADDR/LSADDR/SIZE)
  Cache read  (IOPADDR/LSADDR/SIZE)
  ERROR: Double ICACHE fault
  ERROR: Double ICACHE fault
Which suggest that instruction cache is emulated for IOP. Making this (ps2/gx/net) emu only PS2 emulator that support cache emulation for IOP. For now even most ps1 emulators lack of that feature, and none of known PS2 emulators do that (including Pcsx2/Play!/Dobiestation). With this we can safely assume that also load delay slots are handled correctly here. Unrelated, but is hard to believe that someone implemented icache, but not load delay slots. Which again make it only known emu set that support this afaik.
Which suggest that instruction cache is emulated for IOP. Making this (ps2/gx/net) emu only PS2 emulator that support cache emulation for IOP. For now even most ps1 emulators lack of that feature, and none of known PS2 emulators do that (including Pcsx2/Play!/Dobiestation). With this we can safely assume that also load delay slots are handled correctly here. Unrelated, but is hard to believe that someone implemented icache, but not load delay slots. Which again make it only known emu set that support this afaik.  


===EEDMA on SPE3===
===EEDMA on SPE3===
Line 1,785: Line 1,452:
*8 - SPRfrom dma is handled on PPE only it seems
*8 - SPRfrom dma is handled on PPE only it seems
*9 - SPRto dma is handled on PPE only it seems
*9 - SPRto dma is handled on PPE only it seems
Additionally EEDMA handle VU1 code writes/reads, VU1 code is stored at 0x1A000 of LS in big endian. Only VU1 code, VU1 data is handled by SPE2 (VU1), and any VU0 r/w is handled by PPU only.<br>
Additionally EEDMA handle VU1 code writes/reads. Only VU1 code, VU1 data is handled by SPE2 (VU1), and any VU0 r/w is handled by PPU only.<br>
So is more like "Close to GS" DMA handler.
So is more like "Close to GS" DMA handler.


Line 1,791: Line 1,458:


===VU1 emulation on SPE===
===VU1 emulation on SPE===
VU1 SPE program starts at 0x4000 of LS. Start function is responsible for loading all 128 spe registers with values from memory starting from 0x4230 of LS for r0, ends at 0x4A20 for r127 (r127 is trashed literally in next opcode after that...). This should result in all 0 value regs, excluding r0 which will be set to float(1.0, 0, 0, 0) like ps2 vf00. After that function notify PPE side that its done, and jump to 0x8000, here fun begins. SPE program now clear LS memory from 0x0000 to 0x7FF0, this include VU1 data mem that is located at first 0x4000 bytes. What's more important function that we started at 0x4000 will be deleted. Yup, self modifying code on SPU (good luck for eventual emulator that will try to run ps2_netemu).
<br>
When I disassembled VU1 SPE program, i noticed that real code is really small part of that. Not much to run real VU recompiler/interpreter.
When I disassembled VU1 SPE program, i noticed that real code is really small part of that. Not much to run real VU recompiler/interpreter.
Then i found out something impressive in my opinion. Real deal is that real code delivered to SPE is created on PPE dynamically based on real PS2 VU1 code. Due to similarity of SPE with VU requested in IBM by Sony at design level, there is no VU1 interpreter or recompiler per se. Emulator take VU1 code, dismount it to parts by OP field types, and reassemble into ready SPE code using ready hex templates. I'm not familiar with professional naming of that operation, but its like ahead of time translation of code. So when VU1 code reach SPE is already translated to SPE opcodes. In other terms, SPE responsible for running VU1 is really running VU1 code in some way.
Then i found out something impressive in my opinion. Real deal is that real code delivered to SPE is created on PPE dynamically based on real PS2 VU1 code. Due to similarity of SPE with VU requested in IBM by Sony at design level, there is no VU1 interpreter or recompiler per se. Emulator take VU1 code, dismount it to parts by OP field types, and reassemble into ready SPE code using ready hex templates. I'm not familiar with professional naming of that operation, but its like ahead of time translation of code. So when VU1 code reach SPE is already translated to SPE opcodes. In other terms, SPE responsible for running VU1 is really running VU1 code in some way.


In latest ps2_netemu function responsible for translating VU1 code into SPE ready code is located at 0x13C69C
In latest ps2_netemu function responsible for translating VU1 code into SPE ready code is located at 0x13C69C
<br><br>
SPU registers always follow this layout when VU1 is running:
*r0  - r31 = vf00 - vf31
*r32 - r57 = vi00 - vi25 (vu ctrl regs above vi25 are mapped to vu0 anyway)
*r48      = ACC (accessible from VU0 as vi19)
Additionally ppu can access those regs with addresses 0x3000xyy0.
Where x is 4,5,6, or 7 (mirrors), and yy is reg number in hex.
*0x30004100 = vf16
*0x30004200 = vi00
*0x30004350 = I register (vi21)
etc.
Is worth to note that ppu is not accessing spu regs directly, instead 0x3000x000 is mapped to 0x3F000 of vu1 local store. On real PS2 hardware regs are accessible only when VU1 is stopped so program just dump regs to that location on every vu1 stop (ebit/tbit/force break).
This mapping is added specially for VU0 which can access VU1 registers by use of memory address, when vu0 recompiled program run ppu always load 0x30000000 to r1 to compute address if needed.


===IPU SPE6===
===IPU skip mpeg hack===
====IDEC start code detection====
IDEC perform compare to 0 while starting to search code, but not check that required bits are even available in stream.
Buffer is not refilled when there is more than 0 bits available (function proceed even if there is only 1 bit available). Next compare whole word to 0, which is wrong thing to do. Specially that whole word can be not even available at this point. More over there is no check that at least 8 "0" bits are in stream that way. So start code detection is totally broken. Surprisingly this seems to be handled properly for BDEC..
This probably break Onimusha DoD, and many other random games.
====IPU skip mpeg hack====
There are some leftovers of SKIP MPEG hack in SPE 6 (IPU), i'm not sure that is still available.
There are some leftovers of SKIP MPEG hack in SPE 6 (IPU), i'm not sure that is still available.
Looking at cmd 0x1A there is small chance that is mentioned hack, but i can't confirm yet.


===SPE4 and SPE5===
===SPE4 and SPE5===
Line 1,853: Line 1,499:
== ps2_netemu supported video modes ==
== ps2_netemu supported video modes ==


Both HDTV modes (0x51 1080i and 0x52 720p) crash the emulator. 0x53 576p mode does not, but the image is squished and displaced (looks like the DW is 2560, instead of 1440 and I cannot find any reliable documentation about this mode). Tested by forcing the values in the sceGsResetGraph function.
0x53 video mode (576p) seems to work right. Both HDTV modes (0x51 1080i and 0x52 720p) crash the emulator. Tested by forcing the values in the sceGsResetGraph function.
* None of released games support 576p, so they probably didn't bothered. Afaik that mode is not really used for anything in PS2, like some leftover after testing, or something. I remember that older PS2 don't support it at all, so maybe it was planned as new feature? Also does forcing 480p work? I remember that 480p patching from Managunz gave similar result to what you described for 0x53 mode. But maybe its just broken feature in Managunz itself, i never tested manual patch. --[[User:Kozarovv|Kozarovv]] ([[User talk:Kozarovv|talk]]) 06:15, 29 March 2022 (UTC)
 
** Only the 2.20 BOOT ROM versions and PSX DVR BIOSes support that mode. But in the netemu, I think this mode is not compatible with the real PS2 one. It looks like a regular PAL (0x03) mode to me with a progressive support (SMODE2 INT field set to zero sends correctly a full 576 frame, instead of 288 lines with a PAL 0x03). Judging by the GS DISPLAY registers of OPL's 576p setting, the parameters of this mode should be more similar to the 480p mode than 576i on the real hardware.<br> You are right, it is actually the same case with the forced 480p through sceGsResetGraph. If you force this mode and leave the NTSC output buffer sizes, you get a horizontally magnified image (because the MAGH bit is switched in the register and the requested DW is bigger (2560) than expected (1440)). It could be related to the PCRTC/DVE emulation (maybe conversion of VCK units to pixels, I do not know how it does look like on the PS2). At least the parameters of 480p mode seems to follow the correct specifications, in contrary to the 576p.<br> Anyway, forcing the 480p mode in the netemu is not needed at all (apart from the issue mentioned earlier). Even using the interlaced modes, it does output a progressive frame with the full height front buffer games. With the half height front buffer games, the deinterlacing quality is pretty good in my opinion, as long as the 50/60 fps is maintained.--[[User:Agrippa|Agrippa]] ([[User talk:Agrippa|talk]]) 18:03, 30 March 2022 (UTC)
=== Widescreen support ===
=== Widescreen support ===
If the game does retrieve the widescreen flag from the GetOSDConfigParam through the scescfGetAspect function, the results are as follows:
If the game does retrieve the widescreen flag from the GetOSDConfigParam through the scescfGetAspect function, the results are as follows:
Line 1,864: Line 1,509:


== Emu Patches ==
== Emu Patches ==
===Remove PCRTC Blur for Netemu===
Adds new config 0x4F (no param) which disables PCRTC blur offset which many games use. Patch additionally changes debug menu entry XOR CSR to NO BLUR setting. XOR CSR from file CONFIG is unaffected and still work as expected. I decided to add this entry in menu because it's nice way to compare how good this setting is when game use blur offset. While games which need XOR CSR have very obvious screen corruptions so there is no need for "live" view for it. New config obviously not work on HEN, but CONFIG files which include 0x4F work fine on HEN, command is just skipped, this keeps configs compatible with all current hacks. Big thanks to mrjaredbeta for testing all of this!
Under the hood code compares DISPLAY1 and DISPLAY2 registers and when DX and DY difference is not greater than 6 it mirrors DISPLAY1 to DISPLAY2. Many checks are added to prevent false detection: PMODE need to have enabled 2 circuits, any of DISPLAY registers can't be all 0's, etc. Detection also works for different DW/DH, for games like Soul Calibur 3. Results really vary per game, many games stays unaffected because they modify OFFSET by GIF 0x18/0x19 commands. Good example of game that is nicely improved is Kingdom Hearts 2.
<br><br>Patch is for Evilnat 4.90 ps2_netemu.elf file.
<br>https://www.mediafire.com/file/s83lt9zkwlyhpc2/No_Blur.ppf/file
<br><br>MD5 before patch is applied:
70D22D79A5BB876B8EA2D0FE55D046C7
MD5 after patch is applied:
0B632F05371215AA6BE2C976924737AE
Preview:
https://ibb.co/9sSjmCm (default netemu)
https://ibb.co/G7t11Nq (No Blur enabled)
===Display current PC/RA values for r5900===
Require '''ps2_netemu''' with 3141card patches to display temperatures (every current cfw have it). Mod update values in crazy interval, this is intended. Because mod is used mostly to figure out when game stuck on specific loop. But this can be easily modified to allow less frequent updates. Please note that PC and RA update only when recompiler registers are flushed to memory, luckily for us this happen on every r5900 branch test.
search for (2 times):
3F 40 02 41 63 5A E3 0C 89 3A 00 00 2F 89 00 00
39 29 FF FF 41 9E 00 7C 99 3A 00 00 88 DA 00 01
88 FA 00 02
replace to (2 times):
60 00 00 00 60 00 00 00 60 00 00 00 60 00 00 00
60 00 00 00 60 00 00 00 60 00 00 00 80 C0 91 00
80 E0 8B 7C
search for:
43 45 4C 4C 3A 20 25 30 32 64 20 C2 B0 43 26 23
78 61 3B 52 53 58 3A 20 20 20 25 30 32 64 20 C2
B0 43 00
replace to:
50 43 3A 20 25 30 38 58 26 23 78 61 3B 52 41 3A
20 25 30 38 58 00 00 00 00 00 00 00 00 00 00 00
00 00 00


===Skip demo disc check===
===Skip demo disc check===
Line 1,923: Line 1,531:
   
   
  in 4.75+ emu: 0x8EEA0  beq  cr7, demo_check  ---> nop
  in 4.75+ emu: 0x8EEA0  beq  cr7, demo_check  ---> nop
===Extend config table by 50% in ps2_gxemu===
Patch increase config slots count from 0x314(788) to 0x49E (1182), so exactly 50% more than before. Patch is dedicated for advanced users, and there is no profit of using it if we don't plan to add any new configs.
We are abusing fact that emulator isn't compiled by usual "GameOS" PS3SDK, and pointers are 8 bytes. Also fact that compiler to keep everything aligned is forced to add 4 zeroed bytes after config count.
8 bytes for hash, 8 bytes for pointer, 4 bytes for count, and 4 bytes of align to make data 0x18 sized. Without that PPC64 will throw exception because that data is read in a loop, so next read will be 8 bytes hash from xxxxx4 address.
Sony (LV2/emus/guestOS) compiler isn't aware that emulator need really only 4 bytes for pointer, we have unused 4 bytes. Plus 4 bytes at the end used as alignment.
Now:
hash      00 00 00 FF FF FF FF FF
pointer  00 00 00 00 FF FF FF FF
cmd count FF FF FF FF
align    00 00 00 00 (previously we thought that is terminator)
Compacted:
hash      00 00 00 FF FF FF FF FF
pointer  FF FF FF FF
cmd count FF FF FF FF
We have reduced size from 0x18 to 0x10 for single title without losing any data, free 1576 bytes to use :) .
After that i wrote small dumper that allowed me to edit table in easy way to recreate new table with empty slots.
Hardest part behind us, now just patches to respect new table elements size, and we are done.
4.88.2 Evilnat ps2_gxemu.elf offsets:
0x6F938: 38 00 03 14 --> 38 00 04 9E  (increase loop count)
0x6F950: 39 29 00 18 --> 39 29 00 10  (decrease step incrementer)
0x6F968: 79 59 2E A4
          79 5A 1F 24
          7C 1A C8 50
              -->
          79 59 26 E4
          60 00 00 00
          7B 20 00 20                  (change way of config data offset calculation)
0x6F978: EB EB 00 08 --> 83 EB 00 08  (load config data pointer as zero extended word, instead of doubleword)
0x6F998: 80 0B 00 10 --> 80 0B 00 0C  (change load offset for config count)
0x6F9DC: 7D 3A C8 50 --> 7B 29 00 20  (change way of config data offset calculation for config with 2 or more cmds)
0x6F9F8: 80 09 00 10 --> 80 09 00 0C  (change load offset for config count)
For now we can add some configs to table, and use rest of new zeroed space to store configs itself.
Alternatively we can use whole new space for extended table, and use language strings for configs itself. But that make things "per region" which i don't like. All that without removing or editing any existing config.
Patch need to be applied directly to decrypted elf file, using ppf-o-matic (select all files to see emus in browser), or other tool.<br>
gxemu.elf MD5 before patch
38D891A6A67F9275D1B6D18E04BDCAA7
gxemu.elf MD5 after patch
B452CCB51348127DAF8A931B621E5E39
DL: https://www.mediafire.com/file/kpno5mubyy7q9p0/gx_cfg_ext.ppf/file
===Patch System Manager to allow PS2 emulators Fan Control===
Patch is for LV1 file, should be 4.75+ but it's based on 4.78 LV1 file. '''Don't try to modify your LV1 without hardware flasher!''' This patch enables ability to get fan readings and to set fan speed and policy. This patch doesn't implement fan controls from webman to work in ps2 mode, additional code needs to be patched on PS2 emulators side to make something useful from it.
Search for (4.78 LV1):
E8 03 01 C8 54 00 05 EE  2F A0 00 00 41 9E 00 E4
Replace to (4.78 LV1):
E8 03 01 C8 54 00 05 EE  2F A0 00 00 60 00 00 00
==Inject LIBSD into netemu Bios==
This patch improves compatibility with homebrew. Many homebrews still need hex edits and supplying X modules on disc. But this patch takes care of LIBSD at least. For example, Multiloader 1.41 with midi player mod now works out of the box.
Offsets for unpacked elf file.
0x893470: string TBIN --> LIBSD
0x8DC200: Paste LIBSD file here (Ctrl-B in HxD to overwrite old data).
TBIN file which is replaced is not used in PS2 mode, we can safely patch it. Same can be done for other emulators, just need correcting offsets.
== SPE programs dumper ==
Script for IDA to dump SPE programs from PS2 emulators, don't work with ps2_emu. Script is basing on SPE program names to find correct offsets. Due to that is will miss any SPE program which name is not listed inside script. Tested with ps2_gxemu/ps2_softemu/ps2_netemu.
https://pastebin.com/x8fvY1UL
== Random ps2_netemu notes ==
* Emulator not only patch SPU programs on init, but also patch own PPU code. Which is hard to understand when you can just make changes in source code... eg. 0x1F128 - 0x1F134 in latest emu.
* GUI seems to be tied to GIF/GS emulation. That research was inspired by Dolphin progress report, and it seems to be correct. Fe/be (frontend/backend) spus are involved here. Which explain some UI slowdowns on GIF intensive games.
* Emulator is full of unused functions. Everything that is compiled inline its also there as separate unreachable function.
* DEV9 registers are not implemented. Reads return 0, writes are void. In simple words Net and PS2 HDD emulation is impossible. Fun fact is that emu perform meaningless check for HDD supported titles.
* USB registers are IMPLEMENTED and seems to be fully functional on ppe and spe side. We are missing something else here, not sure what.
===Registers===
It seems that emulator try to keep lower 64 bits of some r5900 registers in specific ppc registers. At least at the time when recompiler is running, also when 0x01 command run.
This helps little bit with understanding some hooks (0x01 commands). Reason is probably optimization, and cached regs are written only when they need to. At least that how it looks like from analyzing recompiled code.
r0  = 0x200000000, used when recompiler load something directly from EE memory.
r13 = Cycles. When recompiled code run, value is frequently compared to value from negmem F000 and event test is triggered depending on compare result.
r14 = r5900_current_pc
r15 = r5900 pc >> 6, used for cache checks. r15 is also used as temp register when fallback to interpreter, etc.
r16 = r5900_v0_reg_lower64
r17 = r5900_v1_reg_lower64
r18 = r5900_a0_reg_lower64
r19 = r5900_a1_reg_lower64
r20 = r5900_a2_reg_lower64
r21 = r5900_a3_reg_lower64
r31 = r5900_ra_reg_lower64
Additionally, Onimusha hooks expect r25 to be r5900 s0, and r28 to be r5900 s3. But i didn't found other parts of recompiler that expect the same. So, for now only values above are confirmed.
128 bit Altivec registers v20 - v23 are used to keep masks which is most likely used to extract/select/insert COP1 regs for next operations. Additionally v0, and v2 are used for other constants.
v0  = 0x0F0E0D0C, 0x0B0A0908, 0x07060504, 0x03020100
v2  = 0x80000000, 0x80000000, 0x80000000, 0x80000000
v20 = 0xFFFFFFFF, 0x00000000, 0x00000000, 0x00000000
v21 = 0x00000000, 0xFFFFFFFF, 0x00000000, 0x00000000
v22 = 0x00000000, 0x00000000, 0xFFFFFFFF, 0x00000000
v22 = 0x00000000, 0x00000000, 0x00000000, 0xFFFFFFFF
128 bit Altivec registers v24 - v31 are used as COP1 (fpu) regs.
v24 = f00, f01, f02, f03
v25 = f04, f05, f06, f07
v26 = f08, f09, f10, f11
v27 = f12, f13, f14, f15
v28 = f16, f17, f18, f19
v29 = f20, f21, f22, f23
v30 = f24, f25, f26, f27
v31 = f28, f29, f30, f31
Register that handle ACC is taken from different pool (same pool as all vfXX regs when in COP2 mode) with param 32 as reg nr (not real reg, probably part of one of vXX regs).
Most likely those regs are flushed to memory when COP2 opcode is running, for sure they are flushed when VU0 microprogram is running.
=== EE Timers Count Read ===
Emulator have bizarre behavior for EE Tx Count read (0x10000000, 10000800, etc). In specific situation (related to pending edge triggered irq) instead of Count value emulator returns Mode value. This doesn't look like programming error and can be some kind of ps2 undocumented behavior implementation.
=== DataStorage vector hook ===
What normally should work as DataStorage exception handler is hacked into very ugly dispatcher for EE related handlers. This code is used for example to read/write IPU registers. At the time when vector is reached:
* Emulator preserves few registers on custom stack at 0x800000. Registers seems to be little random, but they are not. This code is launched from recompiled mips code.
* srr0 is backed up to r3 register (address where exception occurred + 4, rfid opcode jump to address from that reg) and since now it is also used as argument for next steps.
* srr0 is given new value of 0x2EFCC which is custom "dispatcher", link register changes to 0x28F8C8 which is return from that custom piece of... code.
* rfid is hit, let's go to our newly hooked srr0 with 0x2EFCC address.
* Time to use address preserved in r3. This address going thru few checks, it needs to be in 0x10000000- 0x12FFFFFF range (EE JIT Code).
* From this address emulator get single word, that word is used to figure out what mips code wanted to do. This isn't simple offset but some kind of custom identifier.
* When matching identifier is found, task is performed. Some tasks just jump to function and do what is needed, some continue that hackfest and instead are injected into recompiled code as branches to functions that will perform what game want them to do.
* blr is hit, remember that link register is patched earlier to 0x28F8C8
* This function restores previously backed up regs and set link register to value returned by hook. That's all.
=== Free sapce in config parser ===
Since CELL B.E. is big endian machine is only natural to Sony to use little endian for CONFIG files and byte reverse every single word for them. :) What's more important for us, CELL have special opcodes for situation like that one, lwbrx and stwbrx (load/store word byte reversed indexed). But compiler decided that it will be better to do old fashioned 8 opcodes load/swap by shifts and masks and 'or'. This leaves us a lot of space when implementing new configs, making things nice and clean without need to jump outside of function, etc. This code:
seg020:000000000012EECC        lwzx      r0, r9, r31
seg020:000000000012EED0        srwi      r11, r0, 24
seg020:000000000012EED4        rlwinm    r9, r0, 8,8,15
seg020:000000000012EED8        slwi      r3, r0, 24
seg020:000000000012EEDC        rlwinm    r0, r0, 24,16,23
seg020:000000000012EEE0        or        r3, r3, r9
seg020:000000000012EEE4        or        r0, r0, r11
seg020:000000000012EEE8        or        r3, r3, r0
seg020:000000000012EEEC        clrldi    r3, r3, 32
Can be replaced by simple:
seg020:000000000012EECC        lwbrx    r3, r9, r31
seg020:000000000012EED0        clrldi    r3, r3, 32
== RSX workload on the netemu ==
Looks like there is a way to overclock the RSX core by 50 or 100 MHz through LV1 patches. Will the netemu benefit from it?
* Yes, i don't see why not. Assuming that is static patch to elf file, not some cobra style on the fly patch. But don't expect some magic from that. I don't know too much about RSX and not really much about GS. But PS2 emulation is usually limited by CPU power, specially in native resolution. But for example games that need 0x44 cmd, maybe they will work with smoothing now. Maybe some minor slowdowns will be fixed. I still don't know which parts of GS are emulated on RSX, for example softemu used something similar to pcsx2 software render. So there you will get almost nothing from RSX OC. But netemu is different. --[[User:Kozarovv|Kozarovv]] ([[User talk:Kozarovv|talk]]) 04:56, 20 March 2022 (UTC)
** Tested the 600/750 MHz overclock with a few intensive games (SC3, ToCA3, CMR3, VP2, GT4). Assuming the patches are correctly applied (I have no idea at all), there is no performance boost at all.--[[User:Agrippa|Agrippa]] ([[User talk:Agrippa|talk]]) 15:24, 29 May 2022 (UTC)
*** Little bit late, but it seems that GS is almost fully "software" emulated (PPE + BE SPE). It was mistake to not touch ps2_softemu earlier, because that's what helped me to find out that all shaders in netemu are not strictly emulation related. While softemu have Cg shaders for hardware render. --[[User:Kozarovv|Kozarovv]] ([[User talk:Kozarovv|talk]]) 14:48, 29 June 2023 (CEST)
== Netemu load/store with r0 register ==
Emulator access weird addresses which are interpreted as eg. 0xFFFFFFFFFFFF9480, or load/store like std r14, -0x6B80(r0) which translate to std r14, -FFFF9480(r0), and its sign extended by CPU to 64 bit.
Also easier example (without using negative addressing because this is additional emu quirk..). ld r2, 0x3008(r0). This opcode will load double word from 0x3008 address no matter what we currently have in r0, because RA is 0 which is badly interpreted as r0 base.
This is because of PowerPC quirk that i (and apparently IDA in 64 bit mode) wasn't aware. From IBM manual:
ld RT, Disp(RA)
RT Specifies target general-purpose register where result of operation is stored.
Disp Specifies a 16-bit signed number that is a multiple of 4. The assembler divides this number by 4 when generating the instruction.
RA Specifies source general-purpose register for EA calculation.
But according to the same manual:
If GPR RA is not 0, the EA is the sum of the contents of GPR RA and Disp. If GPR RA is 0, then the EA is Disp.
Tl;dr is that if RA is 0 (which disassemblers show as r0), then Disp is real load/store address. This is used many times in emu itself to access negative addresses (0xFFFFFFFFxxxxxxxx), and is used in all netemu cmd 0x01 hooks.
While this is more PPC itself than emu stuff, i feel is important to mention this here.
Now if we remember that emu have mapped "negative address", loads/stores with r0 starting to make sense.
std      r4, 0x3008(r0) # store r4 on 0x3008, no matter what r0 actually is at the moment.
ld        r0, -0x6BF0(r0) # load to r0 from address 0xFFFFFFFFFFFF9410, no matter what r0 actually is at the moment.
ld        r4, 0x3008(r0) # load to r4 from address 0x3008, no matter what r0 actually is at the moment.
== Communication with Graphics Synthesizer in ps2_gxemu ==
Communication from emu level is done with rw to special addresses of what seems to be RSX ports.
Emu includes thin translation layer that intercept rw operations to GS privilaged registers.
Emulator to write or read GS register first need to write register number that we want to access. To do that we use one of two exposed 32 bit ports, separate for reads and writes.
To write GS register first write 64 bit data to write buffer. Separate for lower 32 bit (0xA000304C) and upper (0xA0003048) part of GS write. GS regs are 64 bit while RSX operate on 32 ports.
Finally write translated register number to 0xA0003040. This write starts transfer to GS. To ensure everything went ok emulator performs 32 bit read from 0xA0003000 and check if bit 1 is active.
This operation is performed in loop up to 1000 times until bit 1 is not 0, if that is not the case in mentioned 1000 loops panic is called.
To read GS register first write translated register number to 0xA0003050, then wait for bit 1 of 0xA0003000 to be active. Emulator wait up to 1000 read loops if GS didn't answered in that time emu panic.
Now when bit 1 is not 0, data can be read from 0xA0003058 for upper 32 bits and 0xA000305C for lower 32 bits.
Emulator translate almost all reads to CSR. Only SIGLBLID is readable beside CSR. This is real PS2 GS behavior. Although there is unused runtime that allow read any register. Behavior in that case is unknown.
0x00 = PMODE
0x01 = SMODE1
0x02 = SMODE2
0x03 = SRFSH
0x04 = SYNCH1
0x05 = SYNCH2
0x06 = SYNCV
0x07 = DISPFB1
0x08 = DISPLAY1
0x09 = DISPFB2
0x0A = DISPLAY2
0x0B = EXTBUF
0x0C = EXTDATA
0x0D = EXTWRITE
0x0E = BDCOLOR
0x40 = CSR
0x41 = IMR
0x44 = BUSDIR
0x48 = SIGLBLID
== ps2_gxemu external bios/rom loading. ==
Emulator can load external bios file. File need to be 4 bytes aligned, and 4MB or less size. File name need to be rom.bin (lower case), and probably need to have patched out RDRAM module to success RD check. At least that check is patched in current ps2_gxemu.self rom, maybe is not really needed.
To load bios we need to set emulator status flag bit 2 to true. I'm not sure that is already set, because i don't have memory dump, only elf file. In case is not, we can probably just ignore that check by patch. Emulator read file using lv1_read_remote_file with type 1, and path "rom.bin". So it seems to be relative to emulator self file (dev_flash/ps2emu)? All code is there and is valid, so external bios loading is fully implemented. Only two things required to do that is second bit of status flag set to true, and rom.bin file in correct path.
Note: This code don't exist in ps2_netemu.
== Shin Sangoku Musou config ==
This config does something with sceGsSyncPath by storing 0x4 to INTC_STAT before jumping to it (from one of the main function, so only a specific jump to sceGsSyncPath). I am thinking it fixes this: WhECT9RGZ0k (YouTube link)
What could be happening internally for something like this to be fixed? Interrupt delay to avoid an infinite loop or something? The freeze looks similar to something like the Ecole games (Melty Blood) where the game can move slightly after the freeze, or even break out of it entirely (rare).
* It does write an interrupt request for the start of the VBlank. Looks like the game does sync with the refresh rate and the game logic (at least the specific function you have mentioned) does break itself when the timing is not right.--[[User:Agrippa|Agrippa]] ([[User talk:Agrippa|talk]]) 14:59, 10 August 2022 (UTC)
** Writing to INTC_STAT remove vblank start bit in this case (acknowledge that interrupt happened). But yeah, generally this is timing issue. Probably similar issue to what pcsx2 have: [[https://github.com/PCSX2/pcsx2/issues/5369#issuecomment-1146042222 | here.]] But affect different games due to other differences in emulator. --[[User:Kozarovv|Kozarovv]] ([[User talk:Kozarovv|talk]]) 19:38, 10 August 2022 (UTC)
== Universal DNAS anti-wipe patch ==
Tested with 2.8.0 and 3.0.0 DNAS libraries. EESYNC wipes the memory from 0x00084000 to 0x00100000. Patch is needed for 0x42 command to work with DNAS games when the overlay is installed in this memory region.
<pre>
Original: 18 00 A2 AF 95 00 00 0C 1C 00 A0 AF
Patched:  18 00 A2 AF 01 00 02 24 1C 00 A0 AF</pre>
== Accurate ADD/SUB ==
Emulator need accurate math for PS2 floating points operations because PS2 FPU/VU are special (partially also PS3 SPEs, but that's story for different wiki page).
PS3 VMX unit in PPU have special compatibility mode with SPU, activated by lv1_set_vmx_graphics_mode(1) hvcall. This is supposed to make floating points operations results as close to SPU as possible. This set default rounding mode to round to zero, denormals are treated as zero, and there are no infinities or NaNs. So practically same mode as PS2 use.
Every PS2 EE emulator on PS3 use this mode as otherwise many of that need to be done by software which is inconvenient, and tank performance. There are some suggestions that Sony requested this mode from IBM specifically for PS2 emulators. At this point floats should be handled like on PS2 right? Nope.
PS2 FPU/VU is even more "special" than that, and it seems that Sony missed, or IBM wasn't able to implement that. What is so special? Since PS2 guarantee to round to zero, there is no need for 3 guard bits (more precisely guard/round/sticky bits).
Testing shows that PS2 use only 1 of those bits in way known from other platforms. This makes every FPU/VU ADD/SUB/MUL/DIV operation possibly inaccurate on hardware other than PS2.
There is no way to reproduce that on hardware that is not created with that in mind. No matter it is PS3 VMX, or x86/ARM SSE/AVX. This behavior is not reproductible on currently used hardware without software approach.
But this still doesn't sound that bad, how much developers can rely on bit precise floating point results? 80 from 788 ps2_gxemu configs is using accurate math command, and that's only small chunk of what is really needed. Not to mention direct patches like Persona 4, etc, and tons of custom configs.
<br><br>
Here is function used when accurate ADD/SUB math is requested: https://pastebin.com/sX6vfxin This code run for EVERY SINGLE add/sub separately when opcode is in command range.
<br>
Recompiled code for eg. fpu '''add d,c,b''' looks like this:
vaddfp d,c,b
While recompiled code for accurate fpu '''add d,c,b''' looks like this (simplified):
lis a,0xXXXX
ori a,a, 0xXXXX
stvx b,a, 0x0
add a,0x10
stvx c,a, 0x0
bla https://pastebin.com/sX6vfxin
lis a,0xXXXX
ori a,a, 0xXXXX
lvx d,a, 0x0
Accurate add/sub is really over-complicated, I think is possible to shrink it to prevents slow downs and cache resets when overused. For example instead of creating "constants", we can really use constants from unused memory. On every single accurate add/sub emulator run this code:
0x18726C    li        r9, -1                # --- start of "constants" creation ---
0x187270    clrlwi    r9, r9, 1            # r9 = 0x7FFFFFFF
0x187274    stw      r9, 0x20(r31)
0x187278    li        r9, 0xFF
0x18727C    stw      r9, 0x24(r31)
0x187280    li        r9, 31
0x187284    stw      r9, 0x28(r31)
0x187288    li        r9, 23
0x18728C    stw      r9, 0x2C(r31)
0x187290    li        r9, 0x20 # ' '
0x187294    lvxl      v16, r9, r31          # Load from 0x20(r31)
0x187294                                    # [0x7FFFFFFF, 255, 31, 23]
0x187298    li        r9, 0x97
0x18729C    stw      r9, 0x20(r31)
0x1872A0    clrlslwi  r9, r9, 31,23        # 4B800000 ?
0x1872A4    stw      r9, 0x24(r31)
0x1872A8    li        r9, 25
0x1872AC    stw      r9, 0x28(r31)
0x1872B0    li        r9, -1
0x1872B4    clrrwi    r9, r9, 8            # r9 = 0xFFFFFF00
0x1872B8    stw      r9, 0x2C(r31)
0x1872BC    li        r9, 0x20 # ' '
0x1872C0    lvxl      v15, r9, r31          # Load from 0x20(r31)
0x1872C0                                    # [0x97, 0x4B800000, 25, 0xFFFFFF00]
0x1872C0                                    # End of "constants" creation.
All that to load vector v15 with [0x97, 0x4B800000, 25, 0xFFFFFF00], and v16 with [0x7FFFFFFF, 255, 31, 23]. This is really painful when you need to do that on every accurate add/sub, 22 opcodes, even more cycles. But what if we patch emulator to have those values for example at 0x7000 - 0x701F offset? Then loading constants will be:
0x18726C    li        r9, 0x7000
0x187270    lvxl      v16, r9, r0
0x187274    addi      r9, r9, 0x10
0x187278    lvxl      v16, r9, r0
From 22 opcodes we end up with 4. 18 opcodes less on every accurate add/sub. Additionally we can skip preserving/restoring of LR, CTR, and maybe XER. That give another saving of 12 opcodes. Additional 2 opcodes could be saved by not using r6 as temp for save/restore values. We can reuse r9 instead, so there is no need to preserve/restore old r6. All that shrink add/sub from 216 opcodes to 184. For FPU add/sub even more could be done, but for VU/COP2 its most likely only possible optimisation. Still worth it, should bring noticeable difference in framerate on intensive games that need it.
== Ugly fix for HDD whitelisted games loaded through Cobra ==
The HDD whitelisted games are crashing when loaded through the Cobra (they do work as a 2P package, but the HDD install prompt does not appear of course). I do not know whether the backwards compatible models are affected. The serial ID is compared in the game_ext_plugin module and currently the webMAN MOD does rename the executable inside the ISO to avoid the check. Here is a patch meant to be done on the fly in the payload when mounting the game. Only the SOCOM games have been tested - the HDD install prompt appears and regardless of what you choose, the game will boot successfully.
<pre>
Original: 4B FF 1A 11 38 81 00 A0 38 61 00 AC 4B FF ED 3D
Patched:  60 00 00 00 38 81 00 A0 38 61 00 AC 4B FF ED 3D</pre>
== Unknown VIF issue - bad cmd ==
Emulator seems to do something wrong when bad (unknown) command is sent to VIF. Since all examples are related to VIF1 lets assume that maybe only that part is broken.
* Buffy send command 0x12 - At the time when command is sent VIF1_ERR.ME1 bit is not set. Correct behavior will be stall VIF, seems to be broken because stall take too long.
* Jak TPL accidentally send bad data (vectors floats) as cmd - At the time when command is sent VIF1_ERR.ME1 bit is set. Correct behavior will be ignore bad cmd, and handle it as nop, no stall.
* Formula One 05 bad cmd when entering pitstop - Unknown reason, fixed in pcsx2 at some unknown point. Can be dma issue, but worth to check if netemu is affected in any way (https://github.com/PCSX2/pcsx2/issues/2918)
* General note. PS2 VIF seems to be broken by design. Manual have following warnings:
** Due to a failure in determining DMAtag Mismatch error, setting ME0 bit to 0 may generate an error even though a correct packet has been transmitted.  Set the ME0 bit to 1 when using it.
** In V3_16 format, set the ME1 bit to 1.
** The ER0 bit cannot be used appropriately due to a failure in determining DMAtag Mismatch error. Mask the DMAtag Mismatch error detection feature by setting VIF0_ERR.ME0 to 1.
Code which handle bad cmd:
0x14D00  ila    r4, 0x2000
0x14D04  shufb  r2, r104, r104, r89
0x14D08  ila    sp, sub_147C0
0x14D0C  a      r2, r2, r77
0x14D10  shufb  r2, r103, r102, r2
0x14D14  stqa    r2, VIF1_CODE
0x14D18  lqa    r2, VIF1_ERR
0x14D1C  lqa    r3, VIF1_STAT
0x14D20  andi    r2, r2, 4          ; VIF1_ERR.ME1
0x14D24  ceqi    r2, r2, 0          ; set if bad cmd is unmasked
0x14D28  selb    r3, r3, r2, r4    ; Set VIF_STAT.ER1 if bad cmd bit mask is not active
0x14D2C  andi    r4, r2, 0x10
0x14D30  selb    r2, sp, r39, r2    ; sub_147C0[sp] or (sub_12840 or sub_17700)[r39] depending on ME1 bit // Probably issue starts here.
0x14D34  lnop
0x14D38  or      r87, r87, r4      ; Unk flags, stored at 0x290
0x14D3C  stqa    r3, VIF1_STAT
0x14D40  nop
0x14D44  bi      r2
*Shoutout to Agrippa for finding where the bad cmd in J&D was sent out. Formula One 05 is affected and results in a freeze when entering the pit stop, but I created a patch a while ago to skip the sending of it, so current config fixes the freeze without issue.--[[User:Mrjaredbeta|Mrjaredbeta]] ([[User talk:Mrjaredbeta|talk]]) 03:37, 12 February 2023 (CET)
== Games with EE threading/interrupt issues ==
* '''Harry Potter and the Chamber of Secrets'''
** Seems to randomly hang on the last stage of loading (ra=0x19bf70). WaitSema calls spotted during the hang.
* '''Gran Turismo 4'''
** Hangs during the FMV playback on the idle wait loop. Game does wait for the interrupt that does not happen, probably.
***Game wait in 81FC0 loop or some custom one? Never leave it? --[[User:Kozarovv|Kozarovv]] ([[User talk:Kozarovv|talk]]) 19:01, 11 April 2023 (CEST)
****In the 81FC0 loop. It does leave for a split second indeed, but it is impossible to read the offset. The game is working right in gxemu, I have always wondered if the built-in CDVD emulation may be bugged somehow, introducing new glitches versus gxemu.--[[User:Agrippa|Agrippa]] ([[User talk:Agrippa|talk]]) 19:39, 11 April 2023 (CEST)
Assuming the game programming is right, the underlying issues are data transfer related, probably.
== Games with GS emulation issues ==
Neither performance, nor mipmap and SCANMSK related.
* '''Prince of Persia: Warrior Within'''
** Weapons' trails are rainbow (fixed by Auto-flush in PCSX2).
* '''Lord of the Rings: The Return of the King'''
** Top-left corner issue when the mist/fog is rendered.
* '''Dance Summit 2001: Bust-a-Move'''
** Front buffer is not flushed most of the time. Game seems to apply additional effects there, apart from the downsampling.
* '''Snowblind Engine 2003+ games'''
** Shows the very last "interlaced" frame when switching back to the interlaced mode from progressive one. I thought it could be a VBLANK issue, but that old frame should be long gone by then. The issue seems to be related to the PCRTC. The SMODE2 register is updated in the VBLANK handler. That old frame is shown when the FFMD bit is switched to 1. Looking for better workarounds than delaying the VSYNC or lowering the resolution in 60 fps mode. By the way, the 0x20 command does work with negative values too. Moreover, the max positive value for NTSC is something like 0x106. Anything higher makes the screen freeze on PS2 logo (but the game is working in the background).
== Stuntman/Driv3r research ==
Well, I decided to take another look at this game a few days ago after a weird dream I had. I am glad I did, since I have discovered some new things. This is with 0x0E accuracy adjustments.
Stuntman US
0x100000-0x1c5e68 - no change
0x1c5e6c - improvement in ApplyTyreForces__5WheelR3CarffG4MAv3N24R4MAm4Rf
  vadd.xyzw vf01, vf01, vf02
0x1c7128 - improvement in Update__5WheelR3CarR4MAm4R4MAv3T3
  vadd.xyzw vf01, vf01, vf02
0x1c727c - imrpovement in Update__5WheelR3CarR4MAm4R4MAv3T3
  vadd.xyzw vf01, vf01, vf02
    - both of the above with the others make most accurate path
    - vm insts 0x1c6e78-0x1c6ecc basically negate this additional accuracy?
0x1c769c-0x211f04 - nothing
0x211f08 - improvement in _dyForwardDynamics__FP10DYdynamicsf
  vadd.xyzw vf02, vf02, vf01
0x211f94 - improvement in dyStepRunge1__FP10DYdynamicsf
  vadd.xyzw vf04, vf04, vf01
0x2140b0-0x215118 - no change
0x215118 - obUpdateTransMatrix__FP8OBobject
    - negates additional accuracy of Update__5WheelR3CarR4MAm4R4MAv3T3?
0x215260-0x1ffffff - no change
This does not seem to make any stages completable, but it at least helps the AI car in 1-4 get past the indoor section and not get caught up on the turn/area after. Instead, it now hits the orange crate left of the path it is supposed to go. It could potentially have worse effects on other stages as well... Maybe these offsets/functions in particular can help PCSX2 research.
*Try with EU version, you should be able to complete that mission with accu config. I remember that on pcsx2 with hacked vadd almost full EU game is playable. At some point i also hacked dyStepRunge little bit. dyStepRunge1 is less accurate, dyStepRunge5 is most accurate. Game decide which function should be used. But you can patch it to branch always to dyStepRunge5 etc. --[[User:Kozarovv|Kozarovv]] ([[User talk:Kozarovv|talk]]) 20:03, 4 January 2023 (UTC)
== Patches for SCCS region games ==
Most Chinese region games (SCCS) require the ROM2 modules to be present. However, even if they are required by the game in order to boot, they are not needed at all and are fully playable without ROM2 the vast majority of the time. I have compiled patches for all SCCS games that allow each one to at least boot on PS3 and PCSX2 (older versions, or newer ones without the correct BIOS) by simply skipping over the infinite loops the games get caught in: https://www.mediafire.com/file/skdka574f23b96g/SCCS_patches.zip/file
Mostly untested on PS3, but all have been tested on an older PCSX2 version. PS4 seems to not work with these patches. Even the bootable games without patching needed do not work, so I suspect there is some blacklist preventing them from booting on that emulator. In the end, only two games ended up having text bugs due to the character data not being loaded from ROM2:
{| class="wikitable"
|-
! Serial Number !! Game !! Bootable (without patch) !! ROM2 required
|-
| SCCS-40001 || Ape Escape 2 || No || Yes (missing text)
|-
| SCCS-40002 || Devil May Cry 2 (Disc 1) || No || No
|-
| SCCS-40003 || Devil May Cry 2 (Disc 2) || No || No
|-
| SCCS-40004 || XIGO: Zuihou de Touzi || No || No
|-
| SCCS-40005 || Ico || No || No
|-
| SCCS-40006 || Zhen Sanguo Wushuang 2 || No || No
|-
| SCCS-40007 || Arc the Lad: Seirei no Tasogare || No || No
|-
| SCCS-40009 || Dragon Ball Z 2 || Yes || No
|-
| SCCS-40010 || Super Puzzle Bobble 2 || No || No
|-
| SCCS-40011 || Armored Core 2: Another Age || No || No
|-
| SCCS-40014 || World Soccer Winning Eleven 7 International || No || No
|-
| SCCS-40015 || Viorate no Atelier: Gramnad no Renkinjutsushi 2 || No || Yes (text bugs)
|-
| SCCS-40016 || Ape Escape: Pumped & Primed || Yes || No
|-
| SCCS-40017 || EyeToy: Play || No || EyeToy game
|-
| SCCS-40018 || Saru EyeToy Oosawagi: Wakki Waki Game Tenkomori!! || No || EyeToy game
|-
| SCCS-40019 || Formula One 04 || No || No (English)
|-
| SCCS-40022 || World Soccer Winning Eleven 8: Asia Championship || No || No
|-
| SCCS-60002 || Gran Turismo 4 || No || No
|}
* I believe that real PS2 also need those patches. For OPL, but not only. ROM2 with CH fonts is not existing outside of 3 ps2 models. All of them released only in China, and one of them is still not even dumped (devkit). This ROM2 is even missing in Hong Kong region PS2. Btw. When i implemented ROM2 support in pcsx2, i tested patching game first. Unluckily for netemu, and fortunately for pcsx2. First and only game i tried to patch was Ape Escape 2. :) --[[User:Kozarovv|Kozarovv]] ([[User talk:Kozarovv|talk]]) 12:39, 23 November 2022 (UTC)
** Yup, lucky for me I tested Dynasty Warriors 3 first, then Ape Escape 2 was the second if I recall. I was worried but pleasantly surprised by the rest of the results. I assume ROM2 has data for simplified Chinese characters while Hong Kong/Taiwan uses traditional characters, and that’s probably why there is no ROM2 in the HK/TW region. I am not sure why they even included ROM2 in the first place if barely any games utilize it, but oh well. --[[User:Mrjaredbeta|Mrjaredbeta]] ([[User talk:Mrjaredbeta|talk]]) 18:01, 28 November 2022 (UTC)
*** "PS4 seems to not work with these patches. Even the bootable games without patching needed do not work, so I suspect there is some blacklist preventing them from booting on that emulator.". Just for the record, ps4 don't blacklist them. It's issue with cdvdkey (known for PS3 users as "game hash". ;) ). Emulator fail to respond correctly to one of CDVD register reads. Issue is already patched with my ugly hacks (TC:LA-a-like code swap). Your patches are still required on ps4 after fixing other issues. --[[User:Kozarovv|Kozarovv]] ([[User talk:Kozarovv|talk]]) 17:39, 20 June 2023 (CEST)
== Removing stuff from ps2 emu page ==
I'm thinking about removing some data from PS2 Emulation page. Mostly stuff that is strictly PS2 thing.
*PS2 Memory and Hardware Mapped Registers Layout
*Video Modes
*Config related info
'''Video Modes''' listed there are not even supported by emulators without GS, and likely to fail even on PS3 with GS. This is really info for PS2 wiki in my opinion. '''PS2 Memory and Hardware Mapped Registers Layout''' also fit more in PS2 wiki. This is more like general PS2 dev knowledge than emulation related stuff. Eventually keep them as a links to ps2tek or ps2 devwiki, or something. Let me know if you think this is/isn't good idea. For example PS1 page don't list stuff like this, same goes for PSP page. In case of Config stuff. This is crucial part of this page, but i feel that harm general readability. Due to complicated nature of PS2 config descriptions are getting bigger, and bigger. Honestly this is still missing a lot of info because many times we are limiting ourself to not make descriptions too extensive. All that to not flood page too much. Maybe it's time to move most of that to new dedicated page? We can leave some basic info, like that small table, plus some '''BOLD''' link to "advanced page". This should allow to wikify and move some non-config stuff from talk page. This are only ideas, i expect not everyone will be happy about all of them. Lets talk. :P --[[User:Kozarovv|Kozarovv]] ([[User talk:Kozarovv|talk]]) 08:21, 16 January 2023 (UTC)
* I think these are all good ideas. I occasionally use the PS2 register info for help sometimes, but I could easily use PSI’s page for that (https://psi-rockin.github.io/ps2tek/). I agree that it doesn’t really belong on the page. I agree that emulation page is getting extensive, so maybe it would be good to move the config info to a new page. I am not opposed and are all for it. --[[User:Mrjaredbeta|Mrjaredbeta]] ([[User talk:Mrjaredbeta|talk]]) 15:12, 16 January 2023 (UTC)
== Floating point arithmetic inaccuracies ==
While the software emulators are doing a great job with their accurate paths, there are still issues with some games, unfortunately.
* Stuntman/Driv3r
Completely broken AI, game is unplayable.
* Gran Turismo series
An infamous licence demonstration bug - cars deviate from the racing line. It may affect the AI in the race too. Since the GT's AI is retarded, the subject is difficult to confirm.
* Tokyo Xtreme Racer Zero
It is unconfirmed, but the cars path from PS2 imported replays are played back wrongly in the PCSX2.
* PRIDE FC: Fighting Championships
It seems that y × 1 ≠ y sometimes. Fixed by patching the game code.
== Manipulating r5900 cyclerate ==
Cycles are count in function located at 0x17C9D0 (latest netemu). Emitter with addi to r13 register is what we are looking for. Since pcsx2 use different "unit", lets just call cycles here a... unit.
  _________________________________________________
|  Opcode type  |  Netemu/Gxemu  |    PCSX2  |
|----------------|------------------|-------------|
| Default opcode |    1 unit      |    9 units  |
| Load/Store    |    2 units      |  14 units  |
| Multiply      |    4 units      |  16 units  |
| Divide        |    37 units      |  112 units  |
| COP 0          |    1 unit      |    7 units  |
| COP 1          |  1 unit(some 2)  |    7 units  |
| COP 2          |    1 unit      |    7 units  |
| NOP            |  0 units(!!!)  |  7? units  |
---------------------------------------------------
Additionally pcsx2 use different cycles for many other opcodes that ps3 emus just count as one.
  * MMI      =  14 units
  * MMI Mult =  24 units
  * MMI Div  = 176 units
  * FPU Mul  =  32 units
At the second hand gx/net emu do some weird shenigans with cycles based on... Opcode number, this is still small unknown here. Yeah...
  It turns out that emu is not counting nop cycles. Weird shenigans mentioned above check if opcode number is 1, which belongs to ee NOP in internal emu table.
  When opcode is NOP then emulator is not adding even single cycle. This explains a lot of issues with dma and why EE always seemed too fast for DMA and other components.
  This code takes 2 cycles (units) per loop in emu:
loop_here:
  addiu v0, -1
  nop
  nop
  nop
  nop
  nop
  bnez v0, loop_here
I still need to fully confirm that nop isn't handled in some special way elsewhere but looks like it's not.
Underclocking/Overclocking could be done by modifying opcodes at 0x17CEB0, 0x17CFF8, 0x17D4C8.
What has been tested do far is shifting by 1 or 2 to the left (underclock by multiplying passed cycles). Sadly without any possitive result to framerate, shift by 2 was even slower.
== Emulator status flags ==
We can't really modify this flag from outside on retail emu, but we can always patch function that do init.
===Netemu===
Localised at 0x934158 of emu memory.
1        = init done?
2        = unk thr sys
4        = unk thr sys
8        = unk
10      = run watchpoint opcode (debug halt) 0x11DD80
20      = set mecha 0x245828A to 2, and run sc8 (0x200)
40      = some setting for spe0_class0_int_handler
80      = enable cmd 0xF/0x26/0x27 for whole memory range
200      = set when emulated hardware init is done
400      = enable 0x934160 debug flag (watchpoint on ee and on iop)
800      = skip parsing config file
1000    = do cmd 0x1F with 0x3E8(1000) param
4000    = enable cmd 0x11 for whole VU0 mem (per range setting is inaccessible from config!)
20000    = skip SCED discs error (more, todo)
40000    = do cmd 0x14
80000    = memory card related (0x12A6D8) also run sc_7(2) in thr hdd
200000  = do cmd 0x05
400000  = enable config cmd 0x10 for whole memory range
800000  = do cmd 0x12 with 0x2DFF000 flags
4000000  = mecha related (todo 0x135664)
8000000  = run loop that wait for 0xB52BB8 to be non zero
10000000 = watchpoint (if thread flag is set)
20000000 = timing setting for media validity check
80000000 = show health warning message
Need work
===Gxemu===
Localised at 0x3ED0 of emu memory.
1        = init done?
2        = load external bios.rom
4        = run sub_56784 and sub_56A6C
8        = unk
10      = run watchpoint opcode (debug halt) 0x3F02C
20      = set mecha 0x177AA4F to 2, and run sc8 (0x200) ___todo
40      = some setting for spe0_class0_int_handler
80      = enable gx cmd 0xC/0x22/0x23 for whole memory range
200      = check byte 0x6E6302 and do something if 2. Seems to disable media validity test. Set after EE hw init.
400      = set 0x3D60 to 1
800      = skip parsing internal config list
1000    = do gx cmd 0x1C with param 0x3E8 (1000)
4000    = enable cmd 0x0F for whole VU0 mem (per range setting is inaccessible from config!)
10000    = used as a param for sub_40054, unknown. Not available in netemu
20000    = skip auto setting gx cmd 0x19 for games from hashlist, and later in different sub. Also skip setting SCED error.
40000    = do gx cmd 0x12
80000    = skip running unk syscall
100000  = do gx cmd 0x19 (command unavailable in netemu? Looks different)
200000  = do gx cmd 0x04
400000  = enable gx cmd 0x0E for whole memory range
800000  = run sub sub_F1D78, seems to be much more complicated version of what netemu do here.
1000000  = set 0x3D63 to 1
Need even more work...
== Memory Map ==
=== Netemu ===
Tables at main page taken directly from emulator memory seems to be wrong. I double checked this and values are really there, but ea just don't match to what emulator code do (lpars are most likely correct there). So here it is memory map with addresses that emulator use when want to reach any of mentioned regions.
{| class="wikitable" style="font-size:small;"
|-
! Name !! Start !! End !! Flags !! Notes
|-
| work
|| 0
|| 0x3200000
|| 3
|| separated into work 0x0-0x300000, ro_work 0x300000-0x800000, and rw_work 0x800000-0x32000000.
|-
| menu_heap
|| 0xE000000
|| 0xE400000
|| 3
||
|-
| negmem
|| 0xFFFFFFFFFFFF0000
|| 0
|| 0x10000001
||
|-
| ?
|| 0x30000000
|| 0x30008000
|| 0x30000000
|| vu0
|-
| spu_iop
|| 0x40000000
|| 0x40060000
|| 0xE0000001
||
|-
| spu_spu2
|| 0x40080000
|| 0x400E0000
|| 0xE0000001
||
|-
| spu_vu1
|| 0x40100000
|| 0x40160000
|| 0xE0000001
||
|-
| spu_eedma
|| 0x40180000
|| 0x401E0000
|| 0xE0000001
||
|-
| fe
|| 0x40200000
|| 0x40260000
|| 0xE0000001
||
|-
| be
|| 0x40280000
|| 0x402E0000
|| 0xE0000001
||
|-
| spu_ipu
|| 0x40300000
|| 0x40360000
|| 0xE0000001
||
|-
| be
|| 0x40380000
|| 0x403E0000
|| 0xE0000001
|| not working spu be on spe7
|-
| SPC_SLB
|| 0x50061000
|| 0x50062000
|| 0xF0000000
|| for spu_iop
|-
| 
|| 0x50063000
|| 0x50066000
|| 0xF0000000
||
3000 = SPC_MFC_CNTL
4000 = SPC_PUINT_MB
5000 = SPC_CSR1
|-
| SPC_CSR
|| 0x50062000
|| 0x50063000
|| 0xF0000000
||
|-
| SPC_SLB
|| 0x500E1000
|| 0x500E2000
|| 0xF0000000
|| for spu_spu2
|-
| 
|| 0x500E3000
|| 0x500E6000
|| 0xF0000000
||
3000 = SPC_MFC_CNTL
4000 = SPC_PUINT_MB
5000 = SPC_CSR1
|-
| SPC_CSR
|| 0x500E2000
|| 0x500E3000
|| 0xF0000000
||
|-
| SPC_SLB
|| 0x50161000
|| 0x50162000
|| 0xF0000000
|| for spu_vu1
|-
| 
|| 0x50163000
|| 0x50166000
|| 0xF0000000
||
3000 = SPC_MFC_CNTL
4000 = SPC_PUINT_MB
5000 = SPC_CSR1
|-
| SPC_CSR
|| 0x50162000
|| 0x50163000
|| 0xF0000000
||
|-
| SPC_SLB
|| 0x501E1000
|| 0x501E2000
|| 0xF0000000
|| for spu_eedma
|-
| 
|| 0x501E3000
|| 0x501E6000
|| 0xF0000000
||
3000 = SPC_MFC_CNTL
4000 = SPC_PUINT_MB
5000 = SPC_CSR1
|-
| SPC_CSR
|| 0x501E2000
|| 0x501E3000
|| 0xF0000000
||
|-
| SPC_SLB
|| 0x50261000
|| 0x50262000
|| 0xF0000000
|| for fe
|-
| 
|| 0x50263000
|| 0x50266000
|| 0xF0000000
||
3000 = SPC_MFC_CNTL
4000 = SPC_PUINT_MB
5000 = SPC_CSR1
|-
| SPC_CSR
|| 0x50262000
|| 0x50263000
|| 0xF0000000
||
|-
| SPC_SLB
|| 0x502E1000
|| 0x502E2000
|| 0xF0000000
|| for be
|-
| 
|| 0x502E3000
|| 0x502E6000
|| 0xF0000000
||
3000 = SPC_MFC_CNTL
4000 = SPC_PUINT_MB
5000 = SPC_CSR1
|-
| SPC_CSR
|| 0x502E2000
|| 0x502E3000
|| 0xF0000000
||
|-
| SPC_SLB
|| 0x50361000
|| 0x50362000
|| 0xF0000000
|| for spu_ipu
|-
| 
|| 0x50363000
|| 0x50366000
|| 0xF0000000
||
3000 = SPC_MFC_CNTL
4000 = SPC_PUINT_MB
5000 = SPC_CSR1
|-
| SPC_CSR
|| 0x50362000
|| 0x50363000
|| 0xF0000000
||
|-
| SPC_SLB
|| 0x503E1000
|| 0x503E2000
|| 0xF0000000
|| for unused be on spe7
|-
| 
|| 0x503E3000
|| 0x503E6000
|| 0xF0000000
||
3000 = SPC_MFC_CNTL
4000 = SPC_PUINT_MB
5000 = SPC_CSR1
|-
| SPC_CSR
|| 0x503E2000
|| 0x503E3000
|| 0xF0000000
||
|-
| GSIF
|| 0xA0020000
|| 0xA0028000
|| 0x80000000
|| rsx
|-
| NL
|| 0xA0009000
|| 0xA000A000
|| 0x80000000
|| rsx
|-
| rsx_driver_info
|| 0xA0004000
|| 0xA0007000
|| 0x80000000
|| rsx
|-
| rsx_dma_ctrl
|| 0xA0030000
|| 0xA0050000
|| 0x80000000
|| rsx
|-
| ?
|| 0xA0008000
|| 0xA0009000
|| 0x80000000
|| rsx
|-
| ohci_mmio
|| 0x60000000
|| 0x60001000
|| 0x60000000
||
|-
| ehci_mmio
|| 0x60200000
|| 0x60201000
|| 0x60000000
||
|-
| ohci_mmio
|| 0x60100000
|| 0x60101000
|| 0x60000000
||
|-
| ehci_mmio
|| 0x60300000
|| 0x60301000
|| 0x60000000
||
|-
| base
|| 0x80000000
|| 0x80100000
|| 0x70000003
|| usb
|-
| rawdata
|| 0x80800000
|| 0x80900000
|| 0x70000003
|| usb
|-
| iopdata
|| 0x81000000
|| 0x81100000
|| 0x70000003
|| usb
|-
| base
|| 0x82000000
|| 0x82100000
|| 0x70000003
|| ehci? usb
|-
| ehci_vars
|| 0x82800000
|| 0x82900000
|| 0x70000003
|| usb
|-
| bchs_pmalloc
|| 0x83000000
|| 0x83100000
|| 0x70000003
|| usb
|-
| ?
|| 0x83800000
|| 0x83900000
|| 0x70000003
||
|-
| SB_CDVD
|| 0x84000000
|| 0x84100000
|| 0x70000003
||
|-
| SB_HDD
|| 0x84800000
|| 0x84900000
|| 0x70000003
||
|-
| ?
|| 0x85000000
|| 0x85100000
|| 0x70000003
||
|-
| SB_SIO2
|| 0x85800000
|| 0x85900000
|| 0x70000003
|| only first 0x4000?
|-
| vrc_area
|| 0x90000000
|| 0x91000000
|| 0xC00000003
|| vu1 recompiler area
|-
| ee_cached
|| 0x200000000
|| 0x202000000
|| 0x100000003
||
|-
| ?
|| 0x211004000
|| 0x211008000
|| 0x300000000
|| vu1 memory direct rw
|-
| ee_rom
|| 0x21FC00000
|| 0x220000000
|| 0x300000000
||
|-
| ee_uncached
|| 0x220000000
|| 0x222000000
|| 0x100000003
||
|-
| ee_uncacc
|| 0x230000000
|| 0x232000000
|| 0x100000003
||
|-
| ee_spr_lo
|| 0x260000000
|| 0x260002000
|| 0x700000000
|| weird, that map directly to 0x60000000 ps2 address
|-
| ee_spr_hi
|| 0x270000000
|| 0x270006000
|| 0x800000000
|| ee scratchpad
|-
| ee_kmem0
|| 0x280000000
|| 0x282000000
|| 0x100000003
||
|-
| ?
|| 0x291004000
|| 0x291008000
|| 0x300000000
|| vu1 data rw in kernel mode
|-
| ?
|| 0x29FC00000
|| 0x2A0000000
|| 0x300000000
|| ee_rom in kernel mode
|-
| ?
|| 0x2A0000000
|| 0x2A2000000
|| 0x100000003
|| ee_uncached in kernel mode
|-
| ?
|| 0x2B1004000
|| 0x2B1008000
|| 0x300000000
|| vu1 data in kernel mode. Uncached, i doubt that is even accessible in ps2.
|-
| ?
|| 0x2BFC00000
|| 0x2C0000000
|| 0x300000000
|| ee_rom in uncached kernel mode
|-
| ee_dbg
|| 0x2FFFF8000
|| 0x300000000
|| 0x900000000
||
|-
| iop_mem
|| 0x400000000
|| 0x400200000
|| 0x400000003
||
|-
| iop_rom
|| 0x4BFC00000
|| 0x4C0000000
|| 0x500000001
|| kernel mode
|-
| iop_kmmio0
|| 0x4BF800000
|| 0x4BF810000
|| 0x500000001
|| including iop_spad at first 0x1000
|-
| ?
|| 0x4AFC00000
|| 0x4B0000000
|| 0x500000001
|| iop_rom, uncached?
|-
| iop_kmmio1
|| 0x4AF800000
|| 0x4AF810000
|| 0x500000001
|| including iop_spad at first 0x1000
|-
| spu2_ram
|| 0x600000000
|| 0x600400000
|| 0x600000003
|| starting from 0x200000 spu2_ram2
|-
| ?
|| 0x800000000
|| 0x800010000
|| 0x1000000001
||
|-
| ?
|| 0xC0000000
|| 0xCF800000
|| 0x1600000003
||
|-
| sgs_xdr
|| 0xB4000000
|| 0xB5700000
|| 0x1900000003
||
|-
| ?
|| 0x40000000000
|| 0x40000010000
|| 0x1400000001
||
|-
| ee_jit_code
|| 0x10000000
|| 0x13000000
|| 0xD00000003
||
|-
| vu0_jit_code
|| 0x18000000
|| 0x18400000
|| 0xD00000003
||
|-
| vu0_jit_data
|| 0x1C000000
|| 0x1C400000
|| 0xD00000003
||
|-
| eeram_jit_lut
|| 0x20000000
|| 0x22000000
|| 0xE00000003
||
|-
| eerom_jit_lut
|| 0x2FC00000
|| 0x30000000
|| 0xE00000003
||
|-
| ustack (micro stack)
|| 0xAAA70000000
|| 0xAAA70040000
|| 0x4200000000001
|| Thread KRN0:1 (aka BOOT00)
|-
| ustack (micro stack)
|| 0xAAA70040000
|| 0xAAA70080000
|| 0x4200010000001
|| Thread KRN1:1 (aka BOOT01)
|-
| ustack (micro stack)
|| 0xAAA70080000
|| 0xAAA700C0000
|| 0x42000B0000001
|| VU0 stack (used in EEonBE when VU0 is running)
|-
| ustack (micro stack)
|| 0xAAA700C0000
|| 0xAAA70100000
|| 0x4200030000001
|| Thread SYS
|-
| ustack (micro stack)
|| 0xAAA70100000
|| 0xAAA70140000
|| 0x4200040000001
|| Thread VRC
|-
| ustack (micro stack)
|| 0xAAA70140000
|| 0xAAA70180000
|| 0x4200050000001
|| Thread MECHA
|-
| ustack (micro stack)
|| 0xAAA70180000
|| 0xAAA701C0000
|| 0x4200060000001
|| Thread HDD
|-
| ustack (micro stack)
|| 0xAAA701C0000
|| 0xAAA70200000
|| 0x4200070000001
|| Thread UI
|-
| ustack (micro stack)
|| 0xAAA70200000
|| 0xAAA70280000
|| 0x4200090000001
|| Thread BL2LNK and BL2MAIN
|-
| ustack (micro stack)
|| 0xAAA70280000
|| 0xAAA702C0000
|| 0x42000A0000001
|| Thread USB
|-
| ustack (micro stack)
|| 0xAAA702C0000
|| 0xAAA70300000
|| 0x42000B0000001
|| Thread EEonBE
|-
|}
== Negmem ==
Netemu use memory part called internally negmem. Memory is mapped to 0xFFFFFFFFFFFF0000, but emulator take advantage only of memory from 0xFFFFFFFFFFFF8000 to 0xFFFFFFFFFFFFFFFF. Why? Because that range is loadable with single opcode. How? By "r0" loads/stores (https://www.psdevwiki.com/ps3/Talk:PS2_Emulation#Netemu_load.2Fstore_with_r0_register). With that clever way emulator can access high addresses by setting negative offset for r0 register. This range is used for PS2 registers, DMAC status, next event delta, INTC, and many more. That way emu is able to load value from cached ps2 register in single opcode! Very clever idea!
Negmem memory map.
// Accurate Mul/Div
0xFFFFFFFFFFFF8094 = Storage for accurate mul/div
0xFFFFFFFFFFFF8098 = Storage for accurate mul/div
0xFFFFFFFFFFFF809C = Storage for accurate mul/div
// Cycles
0xFFFFFFFFFFFF8110 = r5900_cycles
// Accurate Add/Sub
0xFFFFFFFFFFFF8138 = Storage for accurate add/sub fs, also for result
0xFFFFFFFFFFFF8140 = Storage for accurate add/sub ft
// Cache checks
0xFFFFFFFFFFFF8180
\
  |- 512 entries. 23 bits for PFN + 2 upper bits for Valid, and LFR flags.
  |- Checks are performed for 2 entries at 1 time.
/
0xFFFFFFFFFFFF897C
// r5900 GPRs
0xFFFFFFFFFFFF8980 = r5900_zero_upper64bits
0xFFFFFFFFFFFF8988 = r5900_zero_lower64bits
0xFFFFFFFFFFFF8990 = r5900_at_reg_upper64
0xFFFFFFFFFFFF8998 = r5900_at_reg_lower64
0xFFFFFFFFFFFF89a0 = r5900_v0_reg_upper64
0xFFFFFFFFFFFF89a8 = r5900_v0_reg_lower64
0xFFFFFFFFFFFF89b0 = r5900_v1_reg_upper64
0xFFFFFFFFFFFF89b8 = r5900_v1_reg_lower64
0xFFFFFFFFFFFF89c0 = r5900_a0_reg_upper64
0xFFFFFFFFFFFF89c8 = r5900_a0_reg_lower64
0xFFFFFFFFFFFF89d0 = r5900_a1_reg_upper64
0xFFFFFFFFFFFF89d8 = r5900_a1_reg_lower64
0xFFFFFFFFFFFF89e0 = r5900_a2_reg_upper64
0xFFFFFFFFFFFF89e8 = r5900_a2_reg_lower64
0xFFFFFFFFFFFF89f0 = r5900_a3_reg_upper64
0xFFFFFFFFFFFF89f8 = r5900_a3_reg_lower64
0xFFFFFFFFFFFF8a00 = r5900_t0_reg_upper64
0xFFFFFFFFFFFF8a08 = r5900_t0_reg_lower64
0xFFFFFFFFFFFF8a10 = r5900_t1_reg_upper64
0xFFFFFFFFFFFF8a18 = r5900_t1_reg_lower64
0xFFFFFFFFFFFF8a20 = r5900_t2_reg_upper64
0xFFFFFFFFFFFF8a28 = r5900_t2_reg_lower64
0xFFFFFFFFFFFF8a30 = r5900_t3_reg_upper64
0xFFFFFFFFFFFF8a38 = r5900_t3_reg_lower64
0xFFFFFFFFFFFF8a40 = r5900_t4_reg_upper64
0xFFFFFFFFFFFF8a48 = r5900_t4_reg_lower64
0xFFFFFFFFFFFF8a50 = r5900_t5_reg_upper64
0xFFFFFFFFFFFF8a58 = r5900_t5_reg_lower64
0xFFFFFFFFFFFF8a60 = r5900_t6_reg_upper64
0xFFFFFFFFFFFF8a68 = r5900_t6_reg_lower64
0xFFFFFFFFFFFF8a70 = r5900_t7_reg_upper64
0xFFFFFFFFFFFF8a78 = r5900_t7_reg_lower64
0xFFFFFFFFFFFF8a80 = r5900_s0_reg_upper64
0xFFFFFFFFFFFF8a88 = r5900_s0_reg_lower64
0xFFFFFFFFFFFF8a90 = r5900_s1_reg_upper64
0xFFFFFFFFFFFF8a98 = r5900_s1_reg_lower64
0xFFFFFFFFFFFF8aa0 = r5900_s2_reg_upper64
0xFFFFFFFFFFFF8aa8 = r5900_s2_reg_lower64
0xFFFFFFFFFFFF8ab0 = r5900_s3_reg_upper64
0xFFFFFFFFFFFF8ab8 = r5900_s3_reg_lower64
0xFFFFFFFFFFFF8ac0 = r5900_s4_reg_upper64
0xFFFFFFFFFFFF8ac8 = r5900_s4_reg_lower64
0xFFFFFFFFFFFF8ad0 = r5900_s5_reg_upper64
0xFFFFFFFFFFFF8ad8 = r5900_s5_reg_lower64
0xFFFFFFFFFFFF8ae0 = r5900_s6_reg_upper64
0xFFFFFFFFFFFF8ae8 = r5900_s6_reg_lower64
0xFFFFFFFFFFFF8af0 = r5900_s7_reg_upper64
0xFFFFFFFFFFFF8af8 = r5900_s7_reg_lower64
0xFFFFFFFFFFFF8b00 = r5900_t8_reg_upper64
0xFFFFFFFFFFFF8b08 = r5900_t8_reg_lower64
0xFFFFFFFFFFFF8b10 = r5900_t9_reg_upper64
0xFFFFFFFFFFFF8b18 = r5900_t9_reg_lower64
0xFFFFFFFFFFFF8b20 = r5900_k0_reg_upper64
0xFFFFFFFFFFFF8b28 = r5900_k0_reg_lower64
0xFFFFFFFFFFFF8b30 = r5900_k1_reg_upper64
0xFFFFFFFFFFFF8b38 = r5900_k1_reg_lower64
0xFFFFFFFFFFFF8b40 = r5900_gp_reg_upper64
0xFFFFFFFFFFFF8b48 = r5900_gp_reg_lower64
0xFFFFFFFFFFFF8b50 = r5900_sp_reg_upper64
0xFFFFFFFFFFFF8b58 = r5900_sp_reg_lower64
0xFFFFFFFFFFFF8b60 = r5900_fp_reg_upper64
0xFFFFFFFFFFFF8b68 = r5900_fp_reg_lower64
0xFFFFFFFFFFFF8b70 = r5900_ra_reg_upper64
0xFFFFFFFFFFFF8b78 = r5900_ra_reg_lower64
// r5900 COP0 Registers
0xFFFFFFFFFFFF8b80 = COP0_Index_REG
0xFFFFFFFFFFFF8b84 = COP0_Random_REG (unused)
0xFFFFFFFFFFFF8b88 = COP0_EntryLo0_REG
0xFFFFFFFFFFFF8b8c = COP0_EntryLo1_REG
0xFFFFFFFFFFFF8b90 = COP0_Context_REG
0xFFFFFFFFFFFF8b94 = COP0_PageMask_REG
0xFFFFFFFFFFFF8b98 = COP0_Wired_REG
0xFFFFFFFFFFFF8b9c = COP0_7_REG
0xFFFFFFFFFFFF8ba0 = COP0_BadVAddr_REG
0xFFFFFFFFFFFF8ba4 = COP0_9_REG
0xFFFFFFFFFFFF8ba8 = COP0_EntryHi_REG
0xFFFFFFFFFFFF8bac = COP0_Compare_REG
0xFFFFFFFFFFFF8bb0 = COP0_Status_REG
0xFFFFFFFFFFFF8bb4 = COP0_Cause_REG
0xFFFFFFFFFFFF8bb8 = COP0_EPC_REG
0xFFFFFFFFFFFF8bbc = COP0_15_REG
0xFFFFFFFFFFFF8bc0 = COP0_PRid_REG
0xFFFFFFFFFFFF8bc4 = COP0_17_REG
0xFFFFFFFFFFFF8bc8 = COP0_18_REG
0xFFFFFFFFFFFF8bcc = COP0_19_REG
0xFFFFFFFFFFFF8bd0 = COP0_20_REG
0xFFFFFFFFFFFF8bd4 = COP0_21_REG
0xFFFFFFFFFFFF8bd8 = COP0_22_REG
0xFFFFFFFFFFFF8bdc = COP0_BadPaddr_REG
0xFFFFFFFFFFFF8be0 = COP0_24_REG
0xFFFFFFFFFFFF8be4 = COP0_Perf_REG
0xFFFFFFFFFFFF8be8 = COP0_26_REG
0xFFFFFFFFFFFF8bec = COP0_27_REG
0xFFFFFFFFFFFF8bf0 = COP0_TagLo_REG
0xFFFFFFFFFFFF8bf4 = COP0_TagHi_REG
0xFFFFFFFFFFFF8bf8 = COP0_ErrorEPC_REG
0xFFFFFFFFFFFF8bfc = COP0_31_REG
// r5900 TLB
0xFFFFFFFFFFFF9110 = r5900_tlb
\
  |- 48 entries
/
0xFFFFFFFFFFFF9400 = r5900_tlb_last
// IPU
0xFFFFFFFFFFFF9770 = IPU_cycles_unk
0xFFFFFFFFFFFF9774 = IPU_FFFF9774
0xFFFFFFFFFFFF9778 = IPU_CTRL
0xFFFFFFFFFFFF9780 = IPU_BP
0xFFFFFFFFFFFF9788 = IPU_TOP
0xFFFFFFFFFFFF9790 = IPU_FFFF9790
// Unk
0xFFFFFFFFFFFF97B0 = 0x20000000000
// Dump regs for debug
0xFFFFFFFFFFFF97C8 = Regs for dump
\
  |- Used in sub_11A640, to dump regs, then notify debugger.
/
0xFFFFFFFFFFFF98E0 = Last reg for dump
// Backup Regs
0xFFFFFFFFFFFF9920
\
|- Another region do backup registers.
|- Used mostly when cmd 0x01 is running, but not only then.
/
0xFFFFFFFFFFFF99A0
// uintc (micro intc)
0xFFFFFFFFFFFFF000 = next event test delta (doubleword)
0xFFFFFFFFFFFFF08x = INTC_STAT      bits
0xFFFFFFFFFFFFF09x = INTC_MASK      bits
0xFFFFFFFFFFFFF0Ax = D_STAT.CIS      bits
0xFFFFFFFFFFFFF0Bx = D_STAT.CIM      bits
0xFFFFFFFFFFFFF0Dx = D_PCR.CPC      bits
0xFFFFFFFFFFFFF0E0 = D_PCR          full
0xFFFFFFFFFFFFF0E4 = INTC_STAT      full
0xFFFFFFFFFFFFF0E8 = INTC_MASK      full
0xFFFFFFFFFFFFF0EC = D_STAT          full
0xFFFFFFFFFFFFF0F0 = D_STAT upper 16 bits
== Shaders available in emulators ==
V for Vertex, F for Fragment, P for Pixel
===Netemu===
FontShader  V
FontShader  F
ImageShader  V
ImageShader  F
GaussShader0 V
GaussShader0 F
GaussShader1 V
GaussShader1 F
ScanShader  V
ScanShader  F
additionally 2 unnamed fragment shaders, 1 vertex shader,
and one "hidden" unaligned vertex shader (at 0x950270).
===Softemu===
GS Base Vertex program              V
GS Base Fragment program            P
Page Translate Out Vertex          V
Page Translate In Vertex            V
Page Translate Out Fragment        F
Page Translate Out Fragment_0      F
Page Translate Out PSMCT16 Fragment F
Page Translate Out PSMCT24 Fragment F
Page Translate Out PSMZ16 Fragment  F
Page Translate In Fragment          F
Page Translate In Fragment_0        F
Page Translate In PSMCT16 Fragment  F
Page Translate In PSMZ24 Fragment  F
Page Translate In PSMZ16 Fragment  F
Block PSMCT4 Translate Frag        F
Block PSMCT8 Translate Frag        F
Block PSMCT16 Translate Frag        F
Block PSMCT24 Translate Frag        F
Block PSMCT32 Translate Frag        F
Block PSMZ24 Translate Frag        F
Texture Block Translate Vertex      V
Notice that GS Base Fragment program is really not fragment shader, but pixel shader.
===Gxemu===
CG_fp_dilate        F
CG_fp_shrink        F
CG_fp_adjweave      F
CG_fp_motionadj    F
CG_fp_undither      F
CG_fp_deinterlace  F  x
CG_fp_weave        F
CG_fp_edgeinterp    F
CG_fp_smooth        F
CG_fp_sharpen      F  x
CG_fp_upscale_smart F  x
CG_fp_upscale      F  x
CG_fp_smart        F  x
CG_fp_mofix        F  x
CG_fp_orientation  F  x
CG_fp_gradient      F  x
CG_vpshader        V  x
CG_interlace        F
CG_fpshader16      F
CG_fpshader        F  x
Plus 2 unnamed shaders, one F, one V.<br>
Shaders with "x" are also available in PS1 emulators, and i sucessfully replaced some of them in ps1 emu on rpcs3 for testing purpose.
===Emu===
Ps2emu have available 14 fragment shaders, and 8 vertex shaders. All of them don't give any hint about name or usage.
== Custom command ideas ==
* '''Disable GS downloads.'''<br>
It is definitely a performance improvement for many titles. In theory, easy to implement (force 0x3 TRXDIR value for every 0x1 write instead). The point is, per-game patches are superior and more robust.<br>
* '''Disable PCRTC blur.'''<br>
PCRTC merge circuits are mostly used for pathetic blurry anti-aliasing. Looks awful on modern TV screens (ToCA 3 is unreadable completely). Blending settings are controlled through the PMODE privileged register. Partially implemented here: [[Talk:PS2_Emulation#Remove_PCRTC_Blur_for_Netemu|Link]]
===Discussion===
GS download config is partially done, need little bit more than TRXDIR patch. But this is something we can patch per-game too. Games do very obvious things to reverse VIF1 FIFO, and also BUSDIR write. From there you can disable whole function that need it. Config will be better of course, but HEN users... Anti Blur is kinda easy to do. Writes to DISPLAY1 can write also to 2 and write to 2 also to 1. So they always match, and config implementation will be rather easy. Most games just offset DX/DY, and never touch those regs again. For games that mess with it, more serious approach is needed. But again HEN users are out of luck if we create command for that. It doesn't help that this can be done by EE patches too. What need to be done is removing offset between DISPLAY1 and DISPLAY2.
*Example for TOCA3 SLUS which use hard coded offset:
004C55F0 00000000
004C55F4 00000000
I'm not saying no, but for now i'm kinda lacking of motivation if not so small HEN user base will be out of luck. But more ideas can help with motivation. :P --[[User:Kozarovv|Kozarovv]] ([[User talk:Kozarovv|talk]]) 22:00, 8 August 2023 (CEST)
== IOP Handling in ps2_emu? (CECHA/CECHB) ==
So, I was reading about how the emulation works here, but I'm not sure how the hardware-based emulation works in the IOP part.
The chart clearly shows that IOP is being 100% emulated via software (Cell) on ps2_gxemu and ps2_netemu, but what about ps2_emu? Is it being emulated using hardware like the EE and GS? Or is it software? I couldn't find any physical chip leading to believe the hardware IOP is there (although the PS2 Bridge has a very similar number model), but I'm aware that some games which have IOP issues on ps2_gxemu and ps2_netemu are not affected in ps2_emu. Also some games that could go online were working fine on ps2_emu, but not so in ps2_gxemu and ps2_netemu.
Is the IOP really fully emulated via software in CECHA/CECHB consoles?
* It's emulated. While other emus use new IOP emulator that is fully running on SPE, ps2_emu use IOP emulator running mostly on PPE core. Emulated IOP ram is mapped to 0x100000000 address of emulator memory, and it's accessible by whole emu. Interpreter is quite simple but handle all needed stuff. IOP hardware registers are mapped in emulator memory and read/write handlers are all PPE functions (that includes DEV9, USB, etc.). IOP Timers run on one of SPE cores, which is interesting solution. SPU2 is running on separate SPE core too. IOP side of SIF communication is done thru "SIF" named SPE program, this program is communicating directly with CXD9208GP hardware. This includes 0x1D0000XX, SIF DMAs, etc. That part is generally not part of IOP per se, I'm just mentioning it for clarity. About compatibility. IOP emus in other PS2 emulators on PS3 were rewritten from scratch, i guess that's why they are less accurate. Plus, fact that with real EE Sony was able to drive most timings inside emu by EE vblk/hblk, which simplifies emulated communication. There is not much to do with accurate IOP when your emulated EE timings are off. --[[User:Kozarovv|Kozarovv]] ([[User talk:Kozarovv|talk]]) 20:19, 24 September 2024 (CEST)
* Thank you very much. all of that is pretty informative and fully answers my questions. The last one I have, does this means then that the software emulation that ps2_emu is doing to the IOP part is more accurate than the one DECKARD does on Slims PS2 right? I tried games like Beyond Good & Evil and it seems to run well without the sounds issues that game has on slims models.
** Yeah, looks like it is. Possibly that wasn't the case back in the release days. PS2 emu has been updated many times since launch in 2006, who knows why... You may want to check this compatibility list: https://en.everybodywiki.com/List_of_PlayStation_2_games_compatible_with_PlayStation_3 It's not that accurate because for example Persona 4 entry is partially bullshit because issue with the bar in the lower left-hand corner just can't exist on real EE. But generally, it will give you some info what doesn't worked back then. Entries like Battle Stadium D.O.N, Naruto Shippuden: Ultimate Ninja 5, Orphen: Scion of Sorcery, Wild Arms 5 or Ibara are really interesting, but I'm not sure if they are correct. Back then people recognized PS3 model by HDD size, You can imagine that it's not most accurate way to do that. :D --[[User:Kozarovv|Kozarovv]] ([[User talk:Kozarovv|talk]]) 08:32, 25 September 2024 (CEST)
***Of course, the HDD size is not the reliable way to identify models, especially when you can just swap the HDD haha. Thanks, you gave me a lot of interesting info, checking the compatibility list I saw a lot of interesting cases, perhaps if one day I'm bored enough I will try the examples of games you gave me, and also try some others. Would be fun if ps2_gxemu could be run on CECHA/B models, I'm aware its not possible, at least not right now (I also tried by swapping them in dev_blind with no success, just a black screen crash, but since it's looking for different hardware on the motherboard this was kind of the expected behavior).
In the end, I guess that games like Ratchet & Clank or Tekken Tag Tournament which runs really slow on ps2_gxemu is because Cell can't keep up with the EE emulation.
Please note that all contributions to PS3 Developer wiki are considered to be released under the GNU Free Documentation License 1.2 (see PS3 Developer wiki:Copyrights for details). If you do not want your writing to be edited mercilessly and redistributed at will, then do not submit it here.
You are also promising us that you wrote this yourself, or copied it from a public domain or similar free resource. Do not submit copyrighted work without permission!

To protect the wiki against automated edit spam, we kindly ask you to solve the following hCaptcha:

Cancel Editing help (opens in new window)

Templates used on this page:

Retrieved from "http://www.psdevwiki.com/ps3/Talk:PS2_Emulation"