Emulator

Forgot to add, you can load the same savestate and you don't have to do anything inbetween. :P

Same for me. After like 100 times my entire RAM is basically taken by Dolphin lol

Yep, that's really like YOLO problem, I'm pretty sure it's gonna get fixed soon :D

The size of a save-state can vary by game. Brawl's save-states are typically around 65MB in size, whereas Worms 3D's are only around 9MB.

65MB O_o, the memory stacks up so fast then :/
I really hope this glitch can be fixed :D
Thanks

I seem EmuCodeBlock protect member is one of the main causes. Those unordered_map is not clear.

Ah, cool you found one of the causes, that'll help a lot!

can confirm that this is a valid issue, not sure if it's release blocking as it has been around since 3.5 at least

I analyze a bit those unordered_map, it's result have 2 problem.

1. When load state, not consider these.(this issue, one of the causes)
   Most games are around several MB.
   (specification implement jit? a tacit understanding since several years?)
2. Regardless load state, it enlarged steadily by specific game.
   for exmaple, The Last Story(SLSJ01).
   (Last Story start, leave anything without play it. And disconnect wiimote guidance. Monitoring dolphin process memory.)
   Sorry my bad English.

By the way, it seems to happens on Linux too. didn't tested on a mac

I'm looking into this issue.

I cannot reproduce this issue on Archlinux x86_64.

I made saves of Smash Bros Brawl using the Fn keys and tried to load them many times while monitoring the memory usage in htop. The loading works fine but I don't see any change in memory consumption.

And I did try both a Debug build and a Release build

seems like to be a Windows exclusive issue(?)

The memory usage increases when loading a savestate and then letting Dolphin run for a second or so. It doesn't happen when loading savestates when Dolphin is paused. This fits with what hk.konpie found - the JIT is only adding entries to the unordered_maps when it's running.

In my #7 comment

2. Regardless load state, it enlarged steadily by specific game.
   for exmaple, The Last Story(SLSJ01).

Have no relationship to this issue8738.
This was due to another.
(The cause was dcbi instruction in LCDisable. LCDisable is executed every 1fps, maybe. As a result, exception handler(JIT_PPC_00000xxx) is generated repeat every 1fps. This seem bug to me. Another for example, Tower of Pandora(SX3J01), ONEPIECE UC EP1(ROUJAF)&EP2(RIUJAF).)

HdkR looked into this today and reproduced it on Ubuntu. Some of the conclusions:

(15:27:18) HdkR: I'm also 100% sure that it isn't an issue with our block cache
(15:33:47) Fog_TAS: hmm
(15:33:52) Fog_TAS: where does that leave us?
(15:35:48) HdkR: Iunno
(15:36:07) HdkR: Figure out what is leaking the memory instead of blaming the block cache?
(15:37:18) Fog_TAS: I'm guessing the unordered_map was the block cache
(15:37:20) Fog_TAS: ?
(15:37:24) JosJuice: Should be somewhere in the JIT code at least, since there were reports that it doesn't happen with interpreter
(15:38:00) HdkR: JosJuice: Or that they didn't wait long enough between state loads to see it
(15:38:05) JosJuice: Also possible
(15:38:27) JosJuice: Yeah, now that I think about it, that doesn't sound unlikely at all
(15:38:38) HdkR: If it is something on the GPU then I wouldn't be surprised

It also seemed like the memory usage was increasing faster when EFB to RAM was used. Please correct me if I missed something important.

These unordered_maps should originally clear it after ClearCodeSpace().
sample patch (but very old): http://pastebin.com/qJXU5KBP
Sorry my bad english from japan.

hk.konpie wrote:

These unordered_maps should originally clear it after ClearCodeSpace().
sample patch (but very old): http://pastebin.com/qJXU5KBP
Sorry my bad english from japan.

English is fine, no need to worry.

Anyways, I've tested the patch and it mitigates the expanding RAM usage (clears it once it reaches a certain point) but doesn't solve why it's actually growing. However, it's better than nothing.

but doesn't solve why it's actually growing.

Do you try insert JitInterface::ClearCache() in State.cpp?
if (loadedSuccessfully)
{
JitInterface::ClearCache();

hk.konpie wrote:

but doesn't solve why it's actually growing.

Do you try insert JitInterface::ClearCache() in State.cpp?
if (loadedSuccessfully)
{
JitInterface::ClearCache();

I actually cleared the cache within JitInterface instead, just put my pull request up here: https://github.com/dolphin-emu/dolphin/pull/3535