Search results

I just benchmarked it, and the r4300 emulation for this game is only 1% faster than the previous version. Maybe you are referring to the auto-frameskip? The 1% improvement is attributable to fixing the register cache and liveness analysis.

I'm not maintainer on repo because I have no way to test the PND. Craig still hasn't sent me the Pandora I ordered over 2 years ago.

Adventus mentioned that he was changing some things, but I never asked him for the patch.

Whoops. It was intended to be similar to the non-branch case below, but I forgot the loop. As written, the code doesn't do anything. (It clears a bit which was never set.)

Previous versions distributed as PNDs had a patch applied to put the save files on the SD card. The full source that I posted is based on the original mupen64plus and places saves in your home directory (ie on the NAND). I'm guessing sebt3 used this instead of applying the diffs. Your...

Updated dynarec: http://hotfile.com/dl/121021087/9688f9b/mupen64plus-arm-20110615.tar.gz.html Diffs: http://hotfile.com/dl/121021405/f8158af/patches-20110128-20110615.tar.gz.html Changes since 20110128: - Expand cache size for MIPS code (performance) - Avoid unnecessary saving/restoring of...

Let us know after you finish rewriting the x86 assembly code for ARM.

and bunnitude

Majora's Mask definitely hits the 32MB limit. Recompiling 64-bit MIPS code to a 32-bit instruction set results in output that's not all that well optimized for size. PowerPC has the same 32MB limit. MIPS can do 256MB jumps. The only platform where this isn't an issue is x86.

True for some games, though Majora's Mask is a counterexample. The code for the various events in clock town is rather large and complex. It's constantly running new code. With a large enough cache you can get most of it, but it requires a very large cache and playing through at least one...

For N64 and PSX, it's going to be modified, or at least have different code loaded for different parts of the game. Some emulators handle this better than others. Or you could save the recompiled code. It tends to be kinda large though.

Well, it is a cultural thing, there are a lot of scams in China so people there think it's normal.

Yes, and I don't really care that much about what Yongzh did. But here's an analogy... Suppose you get together for a friendly neighborhood ball game, and one guy gets caught cheating. Is it a big deal? No, not really. Now suppose there is one guy who is always cheating, and everyone...

Hey notaz, Since you removed pass 7 in the dynarec, you might want to re-add the part that marks the instruction after conditional branches as a branch target. This will avoid a recompile upon return from interrrupt (if the branch was not taken).

People tell me that it does compile and run on Android. AFAIK, the only things yongzh added was touchscreen input and a java frontend.

Is there something wrong with your git server? When I try to download the source, I get fatal: The remote end hung up unexpectedly

Looks like Google actually pulled it. More crap that Google needs to remove.

Several people are reporting that this is no longer available. Did google actually ban yongzh or is it just a glitch?

So the reason that the delay slot evicts the branch instruction's registers, but not vice versa, is because when it allocates registers for the delay slot, it does not distinguish between the case of a taken branch and a not-taken branch. So when analysing the delay slot instruction, it gives...

I think that's right. The basic idea is that the instructions are issued out-of-order, so it does the delay slot first, and then the branch. There are two register mappings, first the registers that are used by the delay slot instruction, which go into regs[i].regmap, then (under "branch...