This adds a tool that writes a snapshot of the full PS3 guest address
space to disk for offline analysis.
This tool produces a flat image that any hex editor or script can index
directly by guest address. It dumps allocated PS3 guest memory only, not
RPCS3 host-process memory, host registers, or GPU-side state.
What it does:
1. Pauses emulation and waits up to five seconds for every PPU, SPU, and
RSX thread to settle. If they do not settle, nothing is written.
2. Copies every allocated guest page into an intermediate buffer on a
named worker thread, then writes it to a sparse `guest_memory.bin` whose
file offset equals the PS3 effective address. Reads go through
`vm::g_sudo_addr` so guest protected pages are captured too.
3. Saves each live SPU thread's 256 KB local store to its own file, with
its ID, LV2 ID, PC, and type recorded.
4. Writes `manifest.json`, listing every allocated region with its
addresses and permissions, the storage mode, and the SPU local-store
table.
5. Resumes emulation.
The menu entry is under Utilities next to Memory Viewer and is enabled
while a game is loaded. Free space is checked before writing. If sparse
file setup fails, the incomplete attempt is removed and the user can
choose another destination or cancel. Failed or cancelled dumps remove
their output folder instead of leaving a partial image.
Tested on Windows using an NTFS destination with a build from this
branch: dumped a running game with 479,535,104 allocated guest bytes
across 76 regions and five SPU local stores. The image was sparse, its
offsets matched guest addresses, and emulation resumed cleanly. The
sparse file failure path was also exercised with a temporary local test:
choosing another location reopened the destination picker, while
cancelling aborted the attempt without leaving partial output. Runtime
behavior on Linux and macOS is untested.
The first draft of this implementation was AI assisted. I reviewed and
adjusted the code and tested the Windows/NTFS path myself. I am happy to
rework anything or answer any questions.
Supports: when value is the immediate result of either a bitwise OR operation or a bitwise AND when either operands is a constant.
Prevents some false positives when the value has PHI nodes in its ancestors.
Fix accidentally clearing a binding during remap if the same button is assigned to rightclick in the pad navigation.
Only allow clearing a binding on release if the same button was also pressed while no remapping occured.
The `FMA` instruction conditionally zeros one multiplicand if the other
is zero/denormal to emulate the xfloat's extended range. This patch
replaces that with a `select(is_non_zero, fma(a, b, c), c)` which has a
shorter dependency chain in most situations and probably allows LLVM to
better optimize it with surrounding instructions.
I added a AVX512 path that uses `vfixupimmps` to prevent a pessimization
where LLVM transforms it into a strictly serial predicate chain. I
confirmed that it works with denormals and both zeros.
https://godbolt.org/z/hsP5G43Ye
LLVM 21 moved the KnownFPClass definition from `ValueTracking.h` to its own file. I verified that specific version moved it, with 20 being the last without it.
Replaced the denormal/zero check in `fma32x4` with LLVM's `KnownFPClass` analysis.
That method now includes a overload which expects the caller to provide the known struct. This is used for memorization with the FMA instructions, where the range clamping and instruction specific special value handling are combined. `FMS` and `FNMS` has a larger search depth as it can potentially shorten the chain by a lot.
Replace constant checks in `FM`, `FCEQ`, and `FCMEQ` with `KnownFPClass`. The equality compares now also optimize for when the operands is infinity, which is safe as it behaves identically to the default path.
Avoid clamping if the value is known not to be Inf/NaN (overflow). Includes a optional precalculated argument, which only calculates the known if it has not been done already.
Optimize the float multiply and compare methods by taking advantage of the `denormals-are-zero` flag, removing duplicated logic, and generally doing things more efficiently.
Refactors SELB by changing "division accuracy correction" pattern match to reduce nesting, include more comments, and use more descriptive variable names. Also replaced duplicated granularity-check loops with single loop which determines byte granularity as a integer.
Megamouse