mirror of
https://github.com/FEX-Emu/FEX.git
synced 2026-10-06 17:00:19 +02:00
Ryan Houdek
40a29ca9a7
This saves a whole bunch of memory. Cutting `Just Cause 2`'s title
screen from 1132MB anonymous FEX memory down to 438MB. 629MB in L2
alone.
L2 is primarily a means to reduce overhead in map queries, so it's all
about performance. But because it consumes a lot of people it's kind of
hard.
One idea is that the L2 lookups can be moved to shared data structures,
since we already pull the shared lock when doing an L2 lookup this is
already halfway there.
Side note, we're using unique locks even with read-only code paths
which we can't use the shared lock because this terrible recursive
mutex!
Instead of outright changing L2 behaviour and potentially wrecking
havoc, add a config option for now so testing can happen over time.
before:
```
Total FEX Anon memory resident: 1132 mB
JIT resident: 60 mB
OpDispatcher resident: 97 mB
Frontend resident: 37 mB
CPUBackend resident: 500 kB
Lookup cache resident: 629 mB
Lookup L1 cache resident: 108 mB
ThreadStates resident: 436 kB
```
after:
```
Total FEX Anon memory resident: 438 mB
JIT resident: 62 mB
OpDispatcher resident: 56 mB
Frontend resident: 22 mB
CPUBackend resident: 496 kB
Lookup cache resident: 0 (null)
Lookup L1 cache resident: 109 mB
ThreadStates resident: 436 kB
```
550 lines
20 KiB
Plaintext
550 lines
20 KiB
Plaintext
{
|
|
"Options": {
|
|
"CPU": {
|
|
"Multiblock": {
|
|
"Type": "bool",
|
|
"Default": "true",
|
|
"Desc": [
|
|
"Controls multiblock code compilation",
|
|
"Can cause long JIT compilation times and stutter"
|
|
]
|
|
},
|
|
"MaxInst": {
|
|
"Type": "int32",
|
|
"Default": "5000",
|
|
"Desc": [
|
|
"Maximum number of instruction to store in a block"
|
|
]
|
|
},
|
|
"HostFeatures": {
|
|
"Type": "strenum",
|
|
"Default": "FEXCore::Config::HostFeatures::OFF",
|
|
"Enums": {
|
|
"ENABLESVE": "enablesve",
|
|
"DISABLESVE": "disablesve",
|
|
"ENABLEAVX": "enableavx",
|
|
"DISABLEAVX": "disableavx",
|
|
"ENABLEAFP": "enableafp",
|
|
"DISABLEAFP": "disableafp",
|
|
"ENABLELRCPC": "enablelrcpc",
|
|
"DISABLELRCPC": "disablelrcpc",
|
|
"ENABLELRCPC2": "enablelrcpc2",
|
|
"DISABLELRCPC2": "disablelrcpc2",
|
|
"ENABLECSSC": "enablecssc",
|
|
"DISABLECSSC": "disablecssc",
|
|
"ENABLEPMULL128": "enablepmull128",
|
|
"DISABLEPMULL128": "disablepmull128",
|
|
"ENABLERNG": "enablerng",
|
|
"DISABLERNG": "disablerng",
|
|
"ENABLECLZERO": "enableclzero",
|
|
"DISABLECLZERO": "disableclzero",
|
|
"ENABLEATOMICS": "enableatomics",
|
|
"DISABLEATOMICS": "disableatomics",
|
|
"ENABLEFCMA": "enablefcma",
|
|
"DISABLEFCMA": "disablefcma",
|
|
"ENABLEFLAGM": "enableflagm",
|
|
"DISABLEFLAGM": "disableflagm",
|
|
"ENABLEFLAGM2": "enableflagm2",
|
|
"DISABLEFLAGM2": "disableflagm2",
|
|
"ENABLEFRINTTS": "enablefrintts",
|
|
"DISABLEFRINTTS": "disablefrintts",
|
|
"ENABLECRYPTO": "enablecrypto",
|
|
"DISABLECRYPTO": "disablecrypto",
|
|
"ENABLERPRES": "enablerpres",
|
|
"DISABLERPRES": "disablerpres",
|
|
"ENABLESVEBITPERM": "enablesvebitperm",
|
|
"DISABLESVEBITPERM": "disablesvebitperm",
|
|
"ENABLEPRESERVEALLABI": "enablepreserveallabi",
|
|
"DISABLEPRESERVEALLABI": "disablepreserveallabi",
|
|
"ENABLEWFXT": "enablewfxt",
|
|
"DISABLEWFXT": "disablewfxt",
|
|
"ENABLE3DNOW": "enable3dnow",
|
|
"DISABLE3DNOW": "disable3dnow",
|
|
"ENABLESSE4A": "enablesse4a",
|
|
"DISABLESSE4A": "disablesse4a"
|
|
},
|
|
"Desc": [
|
|
"Allows controlling of the CPU features in the JIT.",
|
|
"\toff: Default CPU features queried from CPU features",
|
|
"\t{enable,disable}sve: Will force enable or disable sve even if the host doesn't support it",
|
|
"\t{enable,disable}avx: Will force enable or disable avx even if the host doesn't support it",
|
|
"\t{enable,disable}afp: Will force enable or disable afp even if the host doesn't support it",
|
|
"\t{enable,disable}lrcpc: Will force enable or disable lrcpc even if the host doesn't support it",
|
|
"\t{enable,disable}lrcpc2: Will force enable or disable lrcpc2 even if the host doesn't support it",
|
|
"\t{enable,disable}cssc: Will force enable or disable cssc even if the host doesn't support it",
|
|
"\t{enable,disable}pmull128: Will force enable or disable pmull128 even if the host doesn't support it",
|
|
"\t{enable,disable}rng: Will force enable or disable rng even if the host doesn't support it",
|
|
"\t{enable,disable}clzero: Will force enable or disable clzero even if the host doesn't support it",
|
|
"\t{enable,disable}atomics: Will force enable or disable ARMv8.1 LSE atomics even if the host doesn't support it",
|
|
"\t{enable,disable}fcma: Will force enable or disable fcma even if the host doesn't support it",
|
|
"\t{enable,disable}flagm: Will force enable or disable flagm even if the host doesn't support it",
|
|
"\t{enable,disable}flagm2: Will force enable or disable flagm2 even if the host doesn't support it",
|
|
"\t{enable,disable}crypto: Will force enable or disable crypto extensions even if the host doesn't support it",
|
|
"\t{enable,disable}rpres: Will force enable or disable rpres even if the host doesn't support it",
|
|
"\t{enable,disable}svebitperm: Will force enable or disable svebitperm even if the host doesn't support it",
|
|
"\t{enable,disable}preserveallabi: Will force enable or disable preserve_all abi even if the host doesn't support it",
|
|
"\t{enable,disable}wfxt: Will force enable or disable wfxt even if the host doesn't support it",
|
|
"\t{enable,disable}3dnow: Will force enable or disable 3DNow! even if the host doesn't support it",
|
|
"\t{enable,disable}sse4a: Will force enable or disable SSE4a even if the host doesn't support it"
|
|
]
|
|
},
|
|
"SmallTSCScale": {
|
|
"Type": "bool",
|
|
"Default": "true",
|
|
"Desc": [
|
|
"Scales the cycle counter on systems that have low frequencies."
|
|
]
|
|
}
|
|
},
|
|
"Emulation": {
|
|
"RootFS": {
|
|
"Type": "str",
|
|
"Default": "",
|
|
"Desc": [
|
|
"Which Root filesystem prefix to use",
|
|
"This can be a filesystem path",
|
|
"\teg: ~/RootFS/Debian_x86_64",
|
|
"Or this can be a name of a rootfs",
|
|
"If the named rootfs exists in the FEX data folder then it will use that one",
|
|
"\teg: $HOME/.fex-emu/RootFS/<RootFS name>/",
|
|
"Or if you have XDG_DATA_HOME the config will search in that directory",
|
|
"\teg: $XDG_DATA_HOME/.fex-emu/RootFS/<RootFS name>/"
|
|
]
|
|
},
|
|
"ThunkHostLibs": {
|
|
"Type": "str",
|
|
"Default": "@CMAKE_INSTALL_FULL_LIBDIR@/fex-emu/HostThunks",
|
|
"Desc": [
|
|
"Folder to find the host-side thunking libraries."
|
|
]
|
|
},
|
|
"ThunkGuestLibs": {
|
|
"Type": "str",
|
|
"Default": "@CMAKE_INSTALL_PREFIX@/share/fex-emu/GuestThunks",
|
|
"Desc": [
|
|
"Folder to find the guest-side thunking libraries."
|
|
]
|
|
},
|
|
"ThunkConfig": {
|
|
"Type": "str",
|
|
"Default": "",
|
|
"Desc": [
|
|
"A json file specifying where to overlay the thunks.",
|
|
"This can be a filesystem path",
|
|
"\teg: ~/MyThunkConfig.json",
|
|
"Or this can be a named of a Thunk config file",
|
|
"If the named config file exists in the FEX data folder folder the it will use that one",
|
|
"\teg: $HOME/.fex-emu/ThunkConfigs/<ThunkConfig name>",
|
|
"Or if you have XDG_DATA_HOME the config will search in that directory",
|
|
"\teg: $XDG_DATA_HOME/.fex-emu/ThunkConfigs/<ThunkConfig name>"
|
|
]
|
|
},
|
|
"Env": {
|
|
"Type": "strarray",
|
|
"Default": "",
|
|
"Desc": [
|
|
"Adds an environment variable to the emulated environment."
|
|
]
|
|
},
|
|
"HostEnv": {
|
|
"Type": "strarray",
|
|
"Default": "",
|
|
"Desc": [
|
|
"Adds an environment variable to the host environment.",
|
|
"This can be useful for setting environment variables that thunks can pick up.",
|
|
"Typically isn't necessary since the guest libc isn't thunked. But is possible."
|
|
]
|
|
},
|
|
"AdditionalArguments": {
|
|
"Type": "strarray",
|
|
"Default": "",
|
|
"Desc": [
|
|
"Allows the user to pass additional arguments to the application"
|
|
]
|
|
},
|
|
"DisableL2Cache": {
|
|
"Type": "bool",
|
|
"Default": "false",
|
|
"Desc": [
|
|
"Disables FEXCore's JIT L2 cache lookup. Saving memory.",
|
|
"Can potentially introduce more stutters."
|
|
]
|
|
}
|
|
},
|
|
"Debug": {
|
|
"SingleStep": {
|
|
"Type": "bool",
|
|
"Default": "false",
|
|
"Desc": [
|
|
"Single stepping configuration."
|
|
]
|
|
},
|
|
"GdbServer": {
|
|
"Type": "bool",
|
|
"Default": "false",
|
|
"Desc": [
|
|
"Enables the GDB server."
|
|
]
|
|
},
|
|
"DumpIR": {
|
|
"Type": "str",
|
|
"Default": "no",
|
|
"Desc": [
|
|
"Folder to dump the IR in to.",
|
|
"[no, stdout, stderr, server, <Folder>]"
|
|
]
|
|
},
|
|
"PassManagerDumpIR": {
|
|
"Type": "strenum",
|
|
"Default": "FEXCore::Config::PassManagerDumpIR::OFF",
|
|
"Enums": {
|
|
"BEFOREOPT": "beforeopt",
|
|
"AFTEROPT": "afteropt",
|
|
"BEFOREPASS": "beforepass",
|
|
"AFTERPASS": "afterpass"
|
|
},
|
|
"Desc": [
|
|
"Allows controlling when FEX dumps its IR.",
|
|
"\toff: IR dumping will be disabled",
|
|
"\tbeforeopt: Dump IR before any optimizations",
|
|
"\tafteropt: Dump IR after all optimizations",
|
|
"\tbeforepass: Dump IR before every optimization pass",
|
|
"\tafterpass: Dump IR after every optimization pass"
|
|
]
|
|
},
|
|
"DumpGPRs": {
|
|
"Type": "bool",
|
|
"Default": "false",
|
|
"Desc": [
|
|
"When the test harness ends, print the GPR state."
|
|
]
|
|
},
|
|
"O0": {
|
|
"Type": "bool",
|
|
"Default": "false",
|
|
"Desc": [
|
|
"Disables optimizations passes for debugging."
|
|
]
|
|
},
|
|
"GlobalJITNaming": {
|
|
"Type": "bool",
|
|
"Default": "false",
|
|
"Desc": [
|
|
"Uses JITSymbols to name all JIT state as one symbol",
|
|
"Useful for querying how much time is spent inside of the JIT",
|
|
"Profiling tools will show JIT time as FEXJIT"
|
|
]
|
|
},
|
|
"LibraryJITNaming": {
|
|
"Type": "bool",
|
|
"Default": "false",
|
|
"Desc": [
|
|
"Uses JITSymbols to name JIT symbols grouped by library",
|
|
"Useful for querying how much time is spent in each guest library",
|
|
"Can be used to help guide thunk generation"
|
|
]
|
|
},
|
|
"BlockJITNaming": {
|
|
"Type": "bool",
|
|
"Default": "false",
|
|
"Desc": [
|
|
"Uses JITSymbols to name JIT symbols",
|
|
"Useful for determining hot blocks of code",
|
|
"Has some file writing overhead per JIT block"
|
|
]
|
|
},
|
|
"GDBSymbols": {
|
|
"Type": "bool",
|
|
"Default": "false",
|
|
"Desc": [
|
|
"Integrates with GDB using the JIT interface.",
|
|
"Needs the fex jit loader in GDB, which can be loaded via `jit-reader-load libFEXGDBReader.so.`",
|
|
"Also needs x86_64-linux-gnu-objdump in PATH.",
|
|
"Can be very slow."
|
|
]
|
|
},
|
|
"InjectLibSegFault": {
|
|
"Type": "bool",
|
|
"Default": "false",
|
|
"Desc": [
|
|
"Sets the environment variable LD_PRELOAD=libSegFault.so",
|
|
"This allows the user to very easily enable libSegFault without dealing with environment variables",
|
|
"Very useful for applications that have launch scripts that set the variable to nothing at launch",
|
|
"Set this in an application configuration for injecting in to only specific applications.",
|
|
"\tNote: If x86/x86_64 libSegFault.so isn't installed then this option won't work."
|
|
]
|
|
},
|
|
"Disassemble": {
|
|
"Type": "strenum",
|
|
"Default": "FEXCore::Config::Disassemble::OFF",
|
|
"Enums": {
|
|
"DISPATCHER": "dispatcher",
|
|
"BLOCKS": "blocks",
|
|
"STATS": "stats"
|
|
},
|
|
"Desc": [
|
|
"Allows controlling of the vixl disassembler.",
|
|
"\toff: No disassembly will be output",
|
|
"\tdispatcher: Will enable disassembly of the JIT dispatcher loop",
|
|
"\tblocks: Will enable disassembly of the translated instruction code blocks",
|
|
"\tstats: Will print stats when disassembling the code"
|
|
]
|
|
},
|
|
"ForceSVEWidth": {
|
|
"Type": "uint32",
|
|
"Default": "0",
|
|
"Desc": [
|
|
"Allows overriding the SVE width in the vixl simulator.",
|
|
"Useful as a debugging feature."
|
|
]
|
|
},
|
|
"DisableTelemetry": {
|
|
"Type": "bool",
|
|
"Default": "false",
|
|
"Desc": [
|
|
"Disables telemetry at runtime.",
|
|
"Useful for CI instcountCI mostly"
|
|
]
|
|
}
|
|
},
|
|
"Logging": {
|
|
"SilentLog": {
|
|
"Type": "bool",
|
|
"Default": "true",
|
|
"Desc": [
|
|
"Disables logging"
|
|
]
|
|
},
|
|
"OutputLog": {
|
|
"Type": "str",
|
|
"Default": "server",
|
|
"Desc": [
|
|
"File to write FEX output to.",
|
|
"[stdout, stderr, server, <Filename>]"
|
|
]
|
|
},
|
|
"TelemetryDirectory": {
|
|
"Type": "str",
|
|
"Default": "",
|
|
"Desc": [
|
|
"Redirects the telemetry folder that FEX usually writes to.",
|
|
"By default telemetry data is stored in {$FEX_APP_DATA_LOCATION,{$XDG_DATA_HOME,$HOME}/.fex-emu/Telemetry/}"
|
|
]
|
|
},
|
|
"ProfileStats": {
|
|
"Type": "bool",
|
|
"Default": "false",
|
|
"Desc": [
|
|
"Enables FEX's low-overhead sampling profile statistics.",
|
|
"Requires a supported version of Mangohud to see the results"
|
|
]
|
|
}
|
|
},
|
|
"Hacks": {
|
|
"SMCChecks": {
|
|
"Type": "uint8",
|
|
"Default": "FEXCore::Config::CONFIG_SMC_MTRACK",
|
|
"TextDefault": "mtrack",
|
|
"ArgumentHandler": "SMCCheckHandler",
|
|
"Desc": [
|
|
"Checks code for modification before execution.",
|
|
"\tnone: No checks",
|
|
"\tmtrack: Page tracking based invalidation (default)",
|
|
"\tfull: Validate code before every run (slow)"
|
|
]
|
|
},
|
|
"TSOEnabled": {
|
|
"Type": "bool",
|
|
"Default": "true",
|
|
"Desc": [
|
|
"Controls TSO IR ops.",
|
|
"Highly likely to break any multithreaded application if disabled."
|
|
]
|
|
},
|
|
"VectorTSOEnabled": {
|
|
"Type": "bool",
|
|
"Default": "false",
|
|
"Desc": [
|
|
"When TSO emulation is enabled, controls if vector loadstores should also be atomic."
|
|
]
|
|
},
|
|
"MemcpySetTSOEnabled": {
|
|
"Type": "bool",
|
|
"Default": "false",
|
|
"Desc": [
|
|
"When TSO emulation is enabled, controls if memcpy and memset should also be atomic.",
|
|
"Only affects REP MOVS and REP STOS instructions"
|
|
]
|
|
},
|
|
"HalfBarrierTSOEnabled": {
|
|
"Type": "bool",
|
|
"Default": "true",
|
|
"Desc": [
|
|
"When TSO emulation is enabled, controls if unaligned loads and stores should be backpatched to half-barrier atomics.",
|
|
"Can be dangerous due to aligned loadstores through the same code now become non-atomic."
|
|
]
|
|
},
|
|
"StrictInProcessSplitLocks": {
|
|
"Type": "bool",
|
|
"Default": "false",
|
|
"Desc": [
|
|
"Strict global lock when handling an unaligned atomic that crosses a 16-byte or cacheline granularity",
|
|
"This is required to ensure a split-lock doesn't tear inside the process"
|
|
]
|
|
},
|
|
"VolatileMetadata": {
|
|
"Type": "bool",
|
|
"Default": "true",
|
|
"Desc": [
|
|
"Use volatile metadata in PE files to inform TSO instructions when available.",
|
|
"When metadata is unavailable falls back to the currently enabled TSO options."
|
|
]
|
|
},
|
|
"X87ReducedPrecision": {
|
|
"Type": "bool",
|
|
"Default": "false",
|
|
"Desc": [
|
|
"Emulates X87 floating point using 64-bit precision. This reduces emulation accuracy and may result in rendering bugs."
|
|
]
|
|
},
|
|
"X87StrictReducedPrecision": {
|
|
"Type": "bool",
|
|
"Default": "false",
|
|
"Desc": [
|
|
"Enables stricter X87 floating point behavior when X87ReducedPrecision is enabled.",
|
|
"Adds additional checks and implementations like NaN propagation for better compatibility."
|
|
]
|
|
},
|
|
"ABILocalFlags": {
|
|
"Type": "bool",
|
|
"Default": "false",
|
|
"Desc": [
|
|
"When enabled enables an optimization around flags.",
|
|
"Assumes flags are not used across cals.",
|
|
"Hand-written assembly can violate this assumption."
|
|
]
|
|
},
|
|
"ParanoidTSO": {
|
|
"Type": "bool",
|
|
"Default": "false",
|
|
"Desc": [
|
|
"Makes TSO operations even more strict.",
|
|
"Forces vector loadstores to also become atomic."
|
|
]
|
|
},
|
|
"StallProcess": {
|
|
"Type": "bool",
|
|
"Default": "false",
|
|
"Desc": [
|
|
"Forces a process to stall out on initialization",
|
|
"Useful for a process that keeps restarting and doesn't work"
|
|
]
|
|
},
|
|
"HideHypervisorBit": {
|
|
"Type": "bool",
|
|
"Default": "false",
|
|
"Desc": [
|
|
"Hides the hypervisor CPUID bit when set.",
|
|
"Should only be used for applications that have issues with this set."
|
|
]
|
|
},
|
|
"StartupSleep": {
|
|
"Type": "uint32",
|
|
"Default": "0",
|
|
"Desc": [
|
|
"Sleeps the process at startup for a duration of seconds.",
|
|
"Useful if an application crashes too quickly to attach a debugger."
|
|
]
|
|
},
|
|
"StartupSleepProcName": {
|
|
"Type": "str",
|
|
"Default": "",
|
|
"Desc": [
|
|
"Contrains the startup sleep to only apply to processes that match this name."
|
|
]
|
|
},
|
|
"MonoHacks": {
|
|
"Type": "bool",
|
|
"Default": "true",
|
|
"Desc": [
|
|
"Permits a hook-based SMC approach and smaller JIT blocks when mono is detected."
|
|
]
|
|
}
|
|
},
|
|
"Misc": {
|
|
"ServerSocketPath": {
|
|
"Type": "str",
|
|
"Default": "",
|
|
"Desc": [
|
|
"Override for a FEXServer socket path. Only useful for chroots."
|
|
]
|
|
},
|
|
"NeedsSeccomp": {
|
|
"Type": "bool",
|
|
"Default": "false",
|
|
"Desc": [
|
|
"Disables inline syscalls in order to support seccomp handling"
|
|
]
|
|
},
|
|
"ExtendedVolatileMetadata": {
|
|
"Type": "str",
|
|
"Default": "",
|
|
"Desc": [
|
|
"Configuration provided volatile metadata. Only implemented for WoW64/arm64ec.",
|
|
"Limited in its use but can be handy.",
|
|
"Extends on top of what Microsoft has for volatile metadata, but also supported for WoW64.",
|
|
"Colon delimited modules, then semi-colon delimited instructions, then comma delimited ranges",
|
|
"Default disables TSO in the module, unless instructions overlap the range",
|
|
"<module>;<offset begin>-<offset-end>,...;<instruction offset to force TSO>,...:<another>",
|
|
"examples:",
|
|
" * Disable TSO for a full module: Just provide the module name:",
|
|
" `hl2_linux`",
|
|
" * Disable TSO for a part of the module:",
|
|
" `hl2_linux;<offset begin>-<offset-end>`",
|
|
" * Disable TSO for a part of the module, but enable TSO for some instructions within the module",
|
|
" `hl2_linux;<offset begin>-<offset-end>;<instruction offset>,<instruction offset>`",
|
|
" * Disable TSO for multiple modules",
|
|
" `hl2_linux:libsdl2.so`"
|
|
]
|
|
}
|
|
}
|
|
},
|
|
"UnnamedOptions": {
|
|
"Misc": {
|
|
"INTERPRETER_INSTALLED": {
|
|
"Type": "bool",
|
|
"Default": "false"
|
|
},
|
|
"APP_FILENAME": {
|
|
"Type": "str",
|
|
"Default": ""
|
|
},
|
|
"APP_CONFIG_NAME": {
|
|
"Type": "str",
|
|
"Default": "",
|
|
"Desc": [
|
|
"This is the application config name that has been loaded.",
|
|
"This differs from APP_FILENAME in two ways",
|
|
"Where APP_FILENAME always points to the executable path that FEX-Emu is executing.",
|
|
"This matches what is used to load the AppLayer configuration name.",
|
|
"When running through a compatibility layer like wine, this will only be the exe name, instead of wine full path."
|
|
]
|
|
},
|
|
"IS64BIT_MODE": {
|
|
"Type": "bool",
|
|
"Default": "false"
|
|
},
|
|
"DISABLE_VIXL_INDIRECT_RUNTIME_CALLS": {
|
|
"Type": "bool",
|
|
"Default": "true",
|
|
"Desc": [
|
|
"This option is used for the InstructionCountCI so it can generate the same codegen between Arm64 hosts and vixl simulator hosts.",
|
|
"Vixl simulator indirect runtime calls are a special hlt instruction with metadata after it. Effectively making a custom call instruction.",
|
|
"With visual simulator calls disabled, the code generation would be the same as on a native Arm64 host, but running the code is broken."
|
|
]
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|