- GX2 utilizes TCL(.rpl) API for command submission instead of directly writing to an internal GPU fifo
- Submission & retire timestamps are correctly implemented as incremental counters
- Command buffering behaviour matches console
- Fixes race conditions on aarch64
This option still allows you to have proper stack traces for crashes or for the PPC debugger, but is a lot faster then the really slow interpreter (especially when loading into large games like BotW where there has to be a lot of asset decompressing). It makes memory access breakpoints much more brittle though, so it's not a perfect option.
Normal users should of course stick with using the recompiler.
Commit b089ae5b32 changed the PPCInterpreter struct that external plugins rely on to hook Cemu through e.g. the exported "osLib_registerHLEFunction". This commit moves some unused values down so that it keeps the same memory layout as before the PPC recompiler rework.
This reverts commit 372c314f06.
It broke formatting in an attempt to fix GCC builds.
Some other change (perhaps dependency updates) has resolved the issue.
We were using VK_EXT_DEPTH_CLIP_ENABLE but didn't actually request it.
Also fixed an assert when closing Cemu caused by incorrectly tracking the number of allocated pipelines
This allows a savy user, developer or modder to change the comment field of a logging breakpoint to include placeholders such as {r3} or {f3} to log the register values whenever that code is hit.
The result is treated as signed in most cases, but the calculation uses unsigned arithmetic.
As a concrete example where this matters, DS VC passes -1 (2^64-1) to OSWaitEventWithTimeout which internally causes an overflow. But only with unsigned arithmetic this will result in a large positive number that behaves like the intended infinite timeout. With signed arithmetic the result is negative and the events will timeout immediately.
oltolm