04-25-2014, 01:10 AM -
1. A PPU recompiler (or two or three?) is planned, and an SPU recompiler is already in place. I am very certainly no expert on the matter, but going through an IL will just be extra indirection when you could optimize the target assembly anyway.
2. The only reason to do this would be for an academic research project. If you feel you are an academic, feel free to try doing it, but that is not the goal of most or all existing emulation projects.
3. Inline assembly is almost never worth it. rpcs3 already uses SSE intrinsics on some platforms. Assembly just convolutes the code while simultaneously making it unportable, for extremely little to no benefit whatsoever. You will have a much better time increasing cache coherence or, even better, fixing the larger architectural or implementation issues (like the memory manager as Bigpet points out.)
Thanks for the write-up. Anything else in that head of yours?
2. The only reason to do this would be for an academic research project. If you feel you are an academic, feel free to try doing it, but that is not the goal of most or all existing emulation projects.
3. Inline assembly is almost never worth it. rpcs3 already uses SSE intrinsics on some platforms. Assembly just convolutes the code while simultaneously making it unportable, for extremely little to no benefit whatsoever. You will have a much better time increasing cache coherence or, even better, fixing the larger architectural or implementation issues (like the memory manager as Bigpet points out.)
Thanks for the write-up. Anything else in that head of yours?