Emulate GameCube memory card speeds #581

ghost · 2014-07-08T17:38:04Z

Fixes http://code.google.com/p/dolphin-emu/issues/detail?id=2284

Attempts to emulate GameCube memory card transfer speeds.

delroth · 2014-07-08T17:39:18Z

Source/Core/Core/HW/EXI_DeviceMemoryCard.cpp

@@ -227,26 +234,47 @@ void CEXIMemoryCard::SetCS(int cs)

 				//???

-				CmdDoneLater(5000);
+				if (address >= MC_DATA_FILES)


JMC47 · 2014-07-08T17:41:02Z

Did some rough and dirty testing with this build before PR went up and this is actually pretty nifty so far. I haven't tested enough games yet, but it fixes the mentioned issue in Wind Waker (J) and makes the saving/loading save times much, much, much more accurate in the games I tested.

Source/Core/Core/HW/EXI_DeviceMemoryCard.h

+		MC_DATA_DIRECTORY = 0x2000,
+		MC_DATA_DIRECTORYBACKUP = 0x4000,
+		MC_DATA_BLOCKALLOCMAP = 0x6000,
+		MC_DATA_BLOACKALLOCMAPBACKUP = 0x8000,


delroth · 2014-07-08T17:44:22Z

@LPFaint99 could you find some time to review this? AFAIK you're the local expert on everything memcards related :)

shuffle2 · 2014-07-08T17:53:12Z

I don't really know if the method of the fix is correct (hwtest?), but great job fixing that old bug @TotalNerd !

ghost · 2014-07-08T17:57:42Z

I'm hoping for somebody to make it more accurate. I originally did it cosmetically to satisfy my GameCube emulation authenticity but JMC47 pointed out it actually fixes an issue.

JMC47 · 2014-07-08T18:32:15Z

Tested a few more games to try and see if this breaks anything, as well as compared save/loading times to console. The memcard save and load times were consistent with console in Wrestlemania X8, Wind Waker, Luigi's Mansion, Worms 3D, and Zelda: Four Swords Adventures. I will test more games later.

Source/Core/Core/HW/EXI_DeviceMemoryCard.cpp

+// These are rough estimates based on GameCube video evidence of memory card "speeds"
+// This will need to be refined, and perhaps the idea behind the speed thing is wrong
+// But it works with (near?) perfect speed as far as saving goes
+static const float MC_TRANSFER_RATE_WRITE = 22.0f * 1024.0f;


ghost · 2014-07-13T05:28:50Z

Just wanted to point out with the latest commit, I am working on making this a little different and clean... was kinda a rushed result to see if it would work, but I'll see what I can do. Other than that, the read/write speed is not correct by any means.

JMC47 · 2014-07-13T17:36:53Z

Everything appears to be working with the new system for Reading/Writing... except GCI folder doesn't seem to have proper timings. As such, Wind Waker (J) only saves properly when you select memcard, not GCI folder. Weird, huh?

Other than that, it's all about tightening up the timings. First thing I noticed is that on console, different memory cards gave me different speeds (Metroid Prime took almost a quarter of a second longer to read my third party memcard vs a first party one.)

JMC47 · 2014-07-13T19:50:36Z

Did some serious testing on the latest revision (changed to 96.125 kb/s write, which will be added to the pull request soon, I'm sure)

I used a timer program plus my own reflexes for timing, which results in some of the variation I'm sure. At least it should be fairly consistent.

Console The Legend of Zelda: The Wind Waker - 1.41 - 1.55 seconds (2.33 - 2.5 unofficial card)
Pull Request The Legend of Zelda: The Wind Waker - 1.43 - 1.58 seconds
Master The Legend of Zelda: The Wind Waker - .3 seconds (or less, it's near instant)

Console Metroid Prime: .99 - 1.03 seconds (0.89 - 2 seconds with third party memory cards)
Pull Request Metroid Prime: .97 - 1.03 seconds
Master Metroid Prime: Less than .2 seconds; screen doesn't even stay up

For Metroid Prime, there was no good cue for timing, so I ended up doing a side by side test (wavebird hooked up to both console and emulator) and it came out identical (except the slight delay on T.V. tuner)

Console MVP 2004 Dynasty mode: 32.4-32.8 seconds (third party was over a minute)
Pull Request MVP 2004 Dynasty Mode: 32.8 - 33 seconds (the variation was pretty much none.)
Master MVP 2004 Dynasty Mode: 4.2 - 4.38 seconds

I will test more games, but it's a bit time consuming.

JMC47 · 2014-07-13T20:33:56Z

@LPFaint99 After testing this again, and seeing the new method of timing the memory card syncing, it should be ready for review.

shuffle2 · 2014-07-13T21:04:38Z

@dolphin-emu-bot rebuild

Source/Core/Core/HW/EXI_Device.cpp

@@ -102,7 +103,8 @@ IEXIDevice* EXIDevice_Create(TEXIDevices device_type, const int channel_num)
 	case EXIDEVICE_MEMORYCARDFOLDER:
 	{
 		bool gci_folder = (device_type == EXIDEVICE_MEMORYCARDFOLDER);
-		result = new CEXIMemoryCard(channel_num, gci_folder);
+		device_type = EXIDEVICE_MEMORYCARD;


phire · 2014-07-15T06:56:22Z

@dolphin-emu-bot rebuild

Parlane · 2014-07-15T08:11:54Z

@dolphin-emu-bot rebuild

JMC47 · 2014-07-15T09:11:52Z

Looks good to me. Retested GCI Folders and regular memory cards to make sure they were still working after the latest change. I can't do much code review, but functionally it's working as intended. Fixes issues with Netplay save timings in Gauntlet Dark Legacy, Wind Waker (NTSC-J) softlock when saving and, the bigger thing, is that it finally makes save and load times accurate,which was greatly lacking in Dolphin.

JMC47 · 2014-07-15T22:21:23Z

@LPFaint99 A few details emerged on issue 7481 about how memcards communicate. It's new to me, at least.

"The pull request isn't exactly correct since not all memory cards are created equal. For reading there's a latency amount (hardcoded as part of the card's ID) which necessitates a number of dummy reads to clock the EXI bus/nand chip before real data starts being returned. For writing there are certain cards (using samsung NANDs) that support "fast mode" which require no separate erase commands and can write 512 bytes at a time instead of 128."

My question is, since you're the memcard expert, is this more accurate implementation good enough to be merged (after code review of course,) and better than the current implementation? Or would it be better to stick with what's in Dolphin now?

LPFaint99 · 2014-07-16T03:58:00Z

As long as we are using timings from a nintendo brand card, it seems fine to me. deciding exactly which card/speed to use can be decided at any time.

3rd party card timing should be ignored.

as this solves several issues and is more accurate I would say merge is ok

Code LGTM, haven't done any functional testing, trusting JMC47's testing

Parlane · 2014-07-16T04:21:34Z

@dolphin-emu-bot rebuild

phire · 2014-07-16T04:34:10Z

@dolphin-emu-bot rebuild

Parlane · 2014-07-16T04:34:11Z

@dolphin-emu-bot rebuild

JMC47 · 2014-07-16T10:47:57Z

Retested the very latest build, everything's still working, I've never noticed any instability issues. Still fixes the issue mentioned throughout the PR.

LGTM (again :))

Source/Core/Core/HW/EXI.cpp

@@ -97,6 +97,11 @@ void ChangeDevice(const u8 channel, const TEXIDevices device_type, const u8 devi
 	CoreTiming::ScheduleEvent_Threadsafe(500000000, changeDevice, ((u64)channel << 32) | ((u64)device_type << 16) | device_num);
 }

+CEXIChannel* GetChannel(const u32 index)


JMC47 · 2014-07-17T17:08:10Z

Tested the latest version of the PR, everything still works.

Emulate GameCube memory card speeds

PatrickFerry · 2014-10-19T08:21:05Z

This PR introduced a bug. In "XIII" the main menu runs at 3 FPS

Edit: or very slowly, as long as you have ever saved that is

delroth reviewed Jul 8, 2014
View reviewed changes

Source/Core/Core/HW/EXI_DeviceMemoryCard.cpp

@@ -227,26 +234,47 @@ void CEXIMemoryCard::SetCS(int cs)

//???

CmdDoneLater(5000);

if (address >= MC_DATA_FILES)

This comment was marked as off-topic.

Sign in to view

This comment was marked as off-topic.

Sign in to view

delroth reviewed Jul 8, 2014
View reviewed changes

Source/Core/Core/HW/EXI_DeviceMemoryCard.h

MC_DATA_DIRECTORY = 0x2000,

MC_DATA_DIRECTORYBACKUP = 0x4000,

MC_DATA_BLOCKALLOCMAP = 0x6000,

MC_DATA_BLOACKALLOCMAPBACKUP = 0x8000,

This comment was marked as off-topic.

Sign in to view

delroth reviewed Jul 8, 2014
View reviewed changes

shuffle2 reviewed Jul 13, 2014
View reviewed changes

delroth reviewed Jul 16, 2014
View reviewed changes

TotalNerd added 2 commits July 17, 2014 10:57

Emulate GameCube memory card speeds

8a0093d

Redo timing in DMAWrite/DMARead

f602372

Revised channel access from EXI_DeviceMemoryCard

b2ac65b

delroth added a commit that referenced this pull request Jul 19, 2014

Merge pull request #581 from TotalNerd/memcard-emu

963e1a6

Emulate GameCube memory card speeds

delroth merged commit 963e1a6 into dolphin-emu:master Jul 19, 2014

ghost deleted the memcard-emu branch September 12, 2014 08:08

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Emulate GameCube memory card speeds #581

Emulate GameCube memory card speeds #581

ghost commented Jul 8, 2014

delroth commented Jul 8, 2014

This comment was marked as off-topic.

This comment was marked as off-topic.

JMC47 commented Jul 8, 2014

This comment was marked as off-topic.

delroth commented Jul 8, 2014

shuffle2 commented Jul 8, 2014

ghost commented Jul 8, 2014

JMC47 commented Jul 8, 2014

This comment was marked as off-topic.

This comment was marked as off-topic.

ghost commented Jul 13, 2014

JMC47 commented Jul 13, 2014

JMC47 commented Jul 13, 2014

JMC47 commented Jul 13, 2014

shuffle2 commented Jul 13, 2014

This comment was marked as off-topic.

This comment was marked as off-topic.

phire commented Jul 15, 2014

Parlane commented Jul 15, 2014

JMC47 commented Jul 15, 2014

JMC47 commented Jul 15, 2014

LPFaint99 commented Jul 16, 2014

Parlane commented Jul 16, 2014

phire commented Jul 16, 2014

Parlane commented Jul 16, 2014

JMC47 commented Jul 16, 2014

This comment was marked as off-topic.

JMC47 commented Jul 17, 2014

PatrickFerry commented Oct 19, 2014

Emulate GameCube memory card speeds #581

Emulate GameCube memory card speeds #581

Conversation

ghost commented Jul 8, 2014

delroth commented Jul 8, 2014

This comment was marked as off-topic.

This comment was marked as off-topic.

JMC47 commented Jul 8, 2014

This comment was marked as off-topic.

delroth commented Jul 8, 2014

shuffle2 commented Jul 8, 2014

ghost commented Jul 8, 2014

JMC47 commented Jul 8, 2014

This comment was marked as off-topic.

This comment was marked as off-topic.

ghost commented Jul 13, 2014

JMC47 commented Jul 13, 2014

JMC47 commented Jul 13, 2014

JMC47 commented Jul 13, 2014

shuffle2 commented Jul 13, 2014

This comment was marked as off-topic.

This comment was marked as off-topic.

phire commented Jul 15, 2014

Parlane commented Jul 15, 2014

JMC47 commented Jul 15, 2014

JMC47 commented Jul 15, 2014

LPFaint99 commented Jul 16, 2014

Parlane commented Jul 16, 2014

phire commented Jul 16, 2014

Parlane commented Jul 16, 2014

JMC47 commented Jul 16, 2014

This comment was marked as off-topic.

JMC47 commented Jul 17, 2014

PatrickFerry commented Oct 19, 2014