LLI(1)

💅 NAME 💅

okay so welcome to lli(1), bestie — this is the manual page for lli 18. She's the LLVM interpreter & dynamic compiler, and yes, she's literally iconic. 🦋

lli — manual page for lli 18. No cap, this is the whole vibe. 💖

---

✨ DESCRIPTION ✨

OVERVIEW: llvm interpreter & dynamic compiler — she interprets, she dynamically compiles, she does the most. 💅

USAGE:

lli [options] <input bitcode> <program arguments>...

OPTIONS: buckle up, bestie, there are a LOT of them and they all matter. 🌸

🎨 Color Options 🎨

• --color — Use colors in output (default=autodetect). Let her serve some color, or let autodetect decide the vibe. 🌈

⚙️ General options ⚙️

• -O <char> — Optimization level. Pick from [-O0, -O1, -O2, or -O3] (default = '-O2'). Choose your chaos level wisely. 🔥
• --aarch64-neon-syntax=<value> — Choose the style of NEON code to emit from the AArch64 backend:
  • =generic — Emit generic NEON assembly
  • =apple — Emit Apple-style NEON assembly
• --aarch64-use-aa — Enable the use of AA during codegen. AA said yes. ✨
• --abort-on-max-devirt-iterations-reached — Abort when the max iterations for the devirtualization CGSCC repeat pass is reached. Dramatic exit unlocked. 💀
• --addrsig — Emit an address-significance table. Very signature, very slay. 🎀
• --align-loops=<uint> — Default alignment for loops. Keep those loops lined up. 🧚
• --allow-ginsert-as-artifact — Allow G_INSERT to be considered an artifact. Hack around AMDGPU test infinite loops. Respect the hack. 🛠️
• --amdgpu-atomic-optimizer-strategy=<value> — Select DPP or Iterative strategy for scan:
  • =DPP — Use DPP operations for scan
  • =Iterative — Use Iterative approach for scan
  • =None — Disable atomic optimizer
• --amdgpu-bypass-slow-div — Skip 64-bit divide for dynamic 32-bit values. Bye, slow div. 👋
• --amdgpu-disable-loop-alignment — Do not align and prefetch loops. Loops get zero alignment energy today. 🙈
• --amdgpu-dpp-combine — Enable DPP combiner. Combine the DPP. That's the whole note. 💜
• --amdgpu-dump-hsa-metadata — Dump AMDGPU HSA Metadata. Dump it, queen. 🗑️
• --amdgpu-enable-merge-m0 — Merge and hoist M0 initializations. M0 is getting merged and hoisted. 🌸
• --amdgpu-enable-power-sched — Enable scheduling to minimize mAI power bursts. Power-conscious scheduling, we love to see it. ⚡
• --amdgpu-kernarg-preload-count=<uint> — How many kernel arguments to preload onto SGPRs. Preload the args, bestie. 🫶
• --amdgpu-promote-alloca-to-vector-limit=<uint> — Maximum byte size to consider promote alloca to vector. Size limit for the alloca-to-vector promotion moment. 📏
• --amdgpu-sdwa-peephole — Enable SDWA peepholer. Peephole, but make it SDWA. 👀
• --amdgpu-use-aa-in-codegen — Enable the use of AA during codegen. AA is back and she's in codegen. ✨
• --amdgpu-verify-hsa-metadata — Verify AMDGPU HSA Metadata. Trust but verify, bestie. 🔍
• --amdgpu-vgpr-index-mode — Use GPR indexing mode instead of movrel for vector indexing. Index mode switch-up. 💅
• --arm-add-build-attributes — no description given here — she just exists, mysterious and iconic. 🎀
• --arm-implicit-it=<value> — Allow conditional instructions outside of an IT block:
  • =always — Accept in both ISAs, emit implicit ITs in Thumb
  • =never — Warn in ARM, reject in Thumb
  • =arm — Accept in ARM, reject in Thumb
  • =thumb — Warn in ARM, emit implicit ITs in Thumb
• --asm-show-inst — Emit internal instruction representation to assembly file. Show the internal representation, no hiding. 🫣
• --atomic-counter-update-promoted — Do counter update using atomic fetch add for promoted counters only. Atomic fetch add, but only for the promoted ones. 🔥
• --atomic-first-counter — Use atomic fetch add for first counter in a function (usually the entry counter). First counter gets the atomic treatment. 👑
• --basic-block-sections=<all | <function list (file)> | labels | none> — Emit basic blocks into separate sections. Pick your basic-block section situation. 🧩
• --bounds-checking-single-trap — Use one trap block per function. One trap, one function. 💀
• --bounds-checking-unique-traps — Always use one trap per check. Every check gets its own trap. 🙈
• --bpf-stack-size=<int> — Specify the BPF stack size limit. Stack size, specified. 📚
• --cfg-hide-cold-paths=<number> — Hide blocks with relative frequency below the given value. Cold paths? Hidden. 🧊
• --cfg-hide-deoptimize-paths — listed with no description here. She's giving mystery. 🕵️
• --cfg-hide-unreachable-paths — listed with no description here too. Also mysterious. 🌀
• --code-model=<value> — Choose code model:
  • =tiny — Tiny code model
  • =small — Small code model
  • =kernel — Kernel code model
  • =medium — Medium code model
  • =large — Large code model
• --compile-threads=<uint> — Choose the number of compile threads (jit-kind=orc-lazy only). Thread count, but only for jit-kind=orc-lazy. 🧵
• --cost-kind=<value> — Target cost kind:
  • =throughput — Reciprocal throughput
  • =latency — Instruction latency
  • =code-size — Code size
  • =size-latency — Code size and latency
• --cs-profile-generate — Perform context sensitive PGO instrumentation. Context sensitive PGO, periodt. 📈
• --cs-profile-path=<string> — Context sensitive profile file path. Drop the path. 🛣️
• --data-sections — Emit data into separate sections. Data said separate sections, actually. 🍓
• --debug-entry-values — Enable debug info for the debug entry values. Debug entry values get debug info. 🐛
• --debug-info-correlate — Use debug info to correlate profiles. Deprecated, use -profile-correlate=debug-info. She said what she said. ⚠️
• --debugger-tune=<value> — Tune debug info for a particular debugger:
  • =gdb — gdb
  • =lldb — lldb
  • =dbx — dbx
  • =sce — SCE targets (e.g. PS4)
• --debugify-func-limit=<ulong> — Set max number of processed functions per pass. Limit the functions per pass, bestie. 🔢
• --debugify-level=<value> — Kind of debug info to add:
  • =locations — Locations only
  • =location+variables — Locations and Variables
• --debugify-quiet — Suppress verbose debugify output. Quiet mode: activated. 🤫
• --denormal-fp-math=<value> — Select which denormal numbers the code is permitted to require:
  • =ieee — IEEE 754 denormal numbers
  • =preserve-sign — the sign of a flushed-to-zero number is preserved in the sign of 0
  • =positive-zero — denormals are flushed to positive zero
  • =dynamic — denormals have unknown treatment
• --denormal-fp-math-f32=<value> — Select which denormal numbers the code is permitted to require for float:
  • =ieee — IEEE 754 denormal numbers
  • =preserve-sign — the sign of a flushed-to-zero number is preserved in the sign of 0
  • =positive-zero — denormals are flushed to positive zero
  • =dynamic — denormals have unknown treatment
• --disable-auto-upgrade-debug-info — Disable autoupgrade of debug info. No autoupgrade today. 🚫
• --disable-i2p-p2i-opt — Disables inttoptr/ptrtoint roundtrip optimization. The roundtrip optimization is disabled. 😭
• --disable-lazy-compilation — Disable JIT lazy compilation. No lazy compilation, we're doing this now. 💅
• --disable-promote-alloca-to-lds — Disable promote alloca to LDS. Alloca-to-LDS promotion: off. 🙅
• --disable-promote-alloca-to-vector — Disable promote alloca to vector. Alloca-to-vector promotion: also off. 🙅
• --disable-tail-calls — Never emit tail calls. Never. Emit. Tail calls. 💀
• --dlopen=<string> — Dynamic libraries to load before linking. Load the libraries first, then link. 🫶
• --do-counter-promotion — Do counter register promotion. Promote the counter register. 👑
• --dot-cfg-mssa=<file name for generated dot file> — file name for generated dot file. Literally the file name for the generated dot file. 📄
• --dwarf-version=<int> — Dwarf version. Pick the number. 🧚
• --dwarf64 — Generate debugging info in the 64-bit DWARF format. 64-bit DWARF debug info, we love her. 🌟
• --emit-call-site-info — Emit call site debug information, if debug information is enabled. Only if debug info is enabled, bestie. 📞
• --emit-compact-unwind-non-canonical — Whether to try to emit Compact Unwind for non canonical entries. Non-canonical entries might get Compact Unwind. 🌀
• --emit-dwarf-unwind=<value> — Whether to emit DWARF EH frame entries:
  • =always — Always emit EH frame entries
  • =no-compact-unwind — Only emit EH frame entries when compact unwind is not available
  • =default — Use target platform default
• --emscripten-cxx-exceptions-allowed=<string> — The list of function names in which Emscripten-style exception handling is enabled (see emscripten EMSCRIPTEN_CATCHING_ALLOWED options). Function-name allowlist for the exception-handling vibe. 🌸
• --emulated-tls — Use emulated TLS model. TLS, but emulated. ✨
• --enable-approx-func-fp-math — Enable FP math optimizations that assume approx func. FP math optimizations with approx func assumptions. 🔥
• --enable-cache-manager — Use cache manager to save/load modules. Cache manager is handling save/load. 🗂️
• --enable-cse-in-irtranslator — Should enable CSE in irtranslator. CSE in irtranslator, if enabled. 🧠
• --enable-cse-in-legalizer — Should enable CSE in Legalizer. CSE in Legalizer, if enabled. 🧠
• --enable-emscripten-cxx-exceptions — WebAssembly Emscripten-style exception handling. Emscripten-style exceptions, enabled. 🦋
• --enable-emscripten-sjlj — WebAssembly Emscripten-style setjmp/longjmp handling. setjmp/longjmp, but Emscripten-style. 🎀
• --enable-gvn-hoist — turn on the GVN hoisting pass (default = off). she lifts stuff out of loops, she's that girl ✨
• --enable-gvn-memdep — no blurb on this one in the page, it's giving mysterious 💅
• --enable-gvn-sink — turn on the GVN sinking pass (default = off). opposite energy from hoisting, equally iconic 🫶
• --enable-jmc-instrument — instrument functions with a call to __CheckForDebuggerJustMyCode
• --enable-load-in-loop-pre — no blurb, she's just there 🔥
• --enable-load-pre — same deal, no description given 😭
• --enable-loop-simplifycfg-term-folding — no description in the page, lowkey a mystery 🦋
• --enable-name-compression — enable name/filename string compression ✨
• --enable-no-infs-fp-math — enable FP math optimizations that assume no +-Infs
• --enable-no-nans-fp-math — enable FP math optimizations that assume no NaNs
• --enable-no-signed-zeros-fp-math — enable FP math optimizations that assume the sign of 0 is insignificant
• --enable-no-trapping-fp-math — enable setting the FP exceptions build attribute not to use exceptions
• --enable-split-backedge-in-load-pre — no blurb given 🌸
• --enable-split-loopiv-heuristic — enable loop iv regalloc heuristic
• --enable-tlsdesc — enable the use of TLS Descriptors
• --enable-unsafe-fp-math — enable optimizations that may decrease FP precision. heads up bestie, "unsafe" is right there in the name 💀
• --entry-function=<function> — specify the entry function (default = 'main') of the executable
• --exception-model=<value> — exception model. pick your fighter:
  • =default — default exception handling model
  • =dwarf — DWARF-like CFI based exception handling
  • =sjlj — SjLj exception handling
  • =arm — ARM EHABI exceptions
  • =wineh — Windows exception model
  • =wasm — WebAssembly exception handling
• --experimental-debug-variable-locations — use experimental new value-tracking variable locations
• --experimental-debuginfo-iterators — enable communicating debuginfo positions through iterators, eliminating intrinsics
• --extra-archive=<input archive> — extra archive files to be loaded
• --extra-module=<input bitcode> — extra modules to be loaded
• --extra-object=<input object> — extra object files to be loaded
• --fake-argv0=<executable> — override the 'argv[0]' value passed into the executing program. catfish but make it processes 🐟
• --fatal-warnings — treat warnings as errors. no more "it's just a warning" coping 💅
• --filetype=<value> — choose a file type (not all types are supported by all targets):
  • =asm — emit an assembly ('.s') file
  • =obj — emit a native object ('.o') file
  • =null — emit nothing, for performance testing
• --float-abi=<value> — choose float ABI type:
  • =default — target default float ABI type
  • =soft — soft float ABI (implied by -soft-float)
  • =hard — hard float ABI (uses FP registers)
• --force-dwarf-frame-section — always emit a debug frame section.
• --force-interpreter — force interpretation: disable JIT
• --force-tail-folding-style=<value> — force the tail folding style:
  • =none — disable tail folding
  • =data — create lane mask for data only, using active.lane.mask intrinsic
  • =data-without-lane-mask — create lane mask with compare/stepvector
  • =data-and-control — create lane mask using active.lane.mask intrinsic, and use it for both data and control flow
  • =data-and-control-without-rt-check — similar to data-and-control, but remove the runtime check
• --fp-contract=<value> — enable aggressive formation of fused FP ops:
  • =fast — fuse FP ops whenever profitable
  • =on — only fuse 'blessed' FP ops.
  • =off — only fuse FP ops when the result won't be affected.
• --frame-pointer=<value> — specify frame pointer elimination optimization:
  • =all — disable frame pointer elimination
  • =non-leaf — disable frame pointer elimination for non-leaf frame
  • =none — enable frame pointer elimination
• --fs-profile-debug-bw-threshold=<uint> — only show debug message if the source branch weight is greater than this value.
• --fs-profile-debug-prob-diff-threshold=<uint> — only show debug message if the branch probability is greater than this value (in percentage).
• --function-sections — emit functions into separate sections
• --generate-merged-base-profiles — when generating nested context-sensitive profiles, always generate extra base profile for function with all its context profiles merged into it.
• --gpsize=<uint> — Global Pointer Addressing Size. the default size is 8.
• --hash-based-counter-split — rename counter variable of a comdat function based on cfg hash
• --hexagon-rdf-limit=<uint> — no description, she's mysterious 🎀
• --hot-cold-split — enable hot-cold splitting pass
• --ignore-xcoff-visibility — not emit the visibility attribute for asm in AIX OS or give all symbols 'unspecified' visibility in XCOFF object file
• --import-all-index — import all external functions in index.
• --incremental-linker-compatible — when used with filetype=obj, emit an object file which can be used with an incremental linker
• --instcombine-code-sinking — enable code sinking
• --instcombine-guard-widening-window=<uint> — how wide an instruction window to bypass looking for another guard
• --instcombine-max-num-phis=<uint> — maximum number phis to handle in intptr/ptrint folding
• --instcombine-max-sink-users=<uint> — maximum number of undroppable users for instruction sinking
• --instcombine-maxarray-size=<uint> — maximum array size considered when doing a combine
• --instcombine-negator-enabled — should we attempt to sink negations? (yes bestie we should 😌)
• --instcombine-negator-max-depth=<uint> — what is the maximal lookup depth when trying to check for viability of negation sinking.
• --instrprof-atomic-counter-update-all — make all profile counter updates atomic (for testing only)
• --internalize-public-api-file=<filename> — a file containing list of symbol names to preserve
• --internalize-public-api-list=<list> — a list of symbol names to preserve
• --iterative-counter-promotion — allow counter promotion across the whole loop nest.
• --jd=<string> — specifies the JITDylib to be used for any subsequent -extra-module arguments.
• --jit-kind=<value> — choose underlying JIT kind:
  • =mcjit — MCJIT
  • =orc — Orc JIT
  • =orc-lazy — Orc-based lazy JIT.
• --jit-linker=<value> — choose the dynamic linker/loader:
  • =default — default for platform and JIT-kind
  • =rtdyld — RuntimeDyld
  • =jitlink — Orc-specific linker
• --large-data-threshold=<ulong> — choose large data threshold for x86_64 medium code model
• --load=<pluginfilename> — load the specified plugin
• --lto-aix-system-assembler=<path> — path to a system assembler, picked up on AIX only
• --lto-embed-bitcode=<value> — embed LLVM bitcode in object files produced by LTO:
  • =none — do not embed
  • =optimized — embed after all optimization passes
  • =post-merge-pre-opt — embed post merge, but before optimizations
• --lto-pass-remarks-filter=<regex> — only record optimization remarks from passes whose names match the given regular expression
• --lto-pass-remarks-format=<format> — the format used for serializing remarks (default: YAML)
• --lto-pass-remarks-output=<filename> — output filename for pass remarks
• --march=<string> — architecture to generate code for (see --version)
• --matrix-default-layout=<value> — sets the default matrix layout:
  • =column-major — use column-major layout
  • =row-major — use row-major layout
• --matrix-print-after-transpose-opt — no blurb given, she speaks for herself 💜
• --mattr=<a1,+a2,-a3,...> — target specific attributes (-mattr=help for details)
• --max-counter-promotions=<int> — max number of allowed counter promotions
• --max-counter-promotions-per-loop=<uint> — caps how many counter promotions you're allowed per loop, so register pressure doesn't get out of control. little guardrail, big slay 💅
• --mc-relax-all — when you're using it with filetype=obj, this relaxes all the fixups in the object file you're emitting. no leftover tension, all relaxed ✨
• --mcabac — tbd. that's it. that's the whole description. the docs literally said "tbd" and left 😭 we respect the mystery tho 🫶
• --mcjit-remote-process=<filename> — gives you the filename of the process to launch for remote MCJIT execution. if you don't specify one? no worries bestie, remote execution just gets simulated in-process instead 🌀
• --mcpu=<cpu-name> — target one specific cpu type. and hey, -mcpu=help has the details if you wanna peek 👀
• --meabi=<value> — sets the EABI type. the default depends on triple, so it's giving ✨situational✨
  • =default — Triple default EABI version
  • =4 — EABI version 4
  • =5 — EABI version 5
  • =gnu — EABI GNU
• --merror-missing-parenthesis — throws an error when a parenthesis around predicate registers is missing. strict queen behavior 👑
• --merror-noncontigious-register — errors out on register names that aren't contigious (spelling straight from the source, we do not fix, we do not judge 💀)
• --mhvx — turns on Hexagon Vector eXtensions 🐝
• --mhvx=<value> — same thing, but you pick your version:
  • =v60 — Build for HVX v60
  • =v62 — Build for HVX v62
  • =v65 — Build for HVX v65
  • =v66 — Build for HVX v66
  • =v67 — Build for HVX v67
  • =v68 — Build for HVX v68
  • =v69 — Build for HVX v69
  • =v71 — Build for HVX v71
  • =v73 — Build for HVX v73
• --mips-compact-branches=<value> — MIPS specific! it's the compact branch policy, and yes there are three moods:
  • =never — Do not use compact branches if possible.
  • =optimal — Use compact branches where appropriate (default).
  • =always — Always use compact branches if possible.
• --mips16-constant-islands — enables mips16 constant islands 🏝️
• --mips16-hard-float — enables mips16 hard float.
• --mir-strip-debugify-only — controls whether mir-strip-debug only strips debug info from debugified modules, by default. very "only in this specific circumstance" energy 🙈
• --misexpect-tolerance=<uint> — stops diagnostics from being emitted while profile counts sit within N% of the threshold. (yes, the original really ends with a double period .. — iconique 💅)
• --mno-compound — disables looking for compound instructions for Hexagon.
• --mno-fixup — disables fixing up resolved relocations for Hexagon.
• --mno-ldc1-sdc1 — expands double precision loads and stores into their single precision counterparts. one becomes two, the math is mathing 🧮
• --mno-pairing — disables looking for duplex instructions for Hexagon.
• --mtriple=<string> — overrides the target triple for the module.
• --mwarn-missing-parenthesis — warns (instead of erroring) when the parenthesis around predicate registers is missing. gentle reminder vibes 🌸
• --mwarn-noncontigious-register — warns about register names that arent contigious (again: their spelling, not mine 🫶)
• --mwarn-sign-mismatch — warns when a signed and an unsigned value don't match. the red flag detector 🚩
• --mxcoff-roptr — when set to true, const objects with relocatable address values get put into the RO data section.
• --no-deprecated-warn — suppresses all deprecated warnings. living unbothered, we ignore the past 💀
• --no-discriminators — disables generation of discriminator information.
• --no-integrated-as — disables the integrated assembler.
• --no-process-syms — does not resolve lli process symbols in JIT'd code.
• --no-type-check — suppresses type errors (Wasm).
• --no-warn — suppresses all warnings. full peace and quiet mode 🔇
• --nozero-initialized-in-bss — doesn't place zero-initialized symbols into the bss section.
• --nvptx-sched4reg — NVPTX Specific: schedule for register pressue (their spelling again bestie, we're keeping it 💜)
• --object-cache-dir=<string> — the directory where cached object files go. it must be user writable — no writable, no cache, no slay 🔥
• --object-size-offset-visitor-max-visit-instructions=<uint> — the maximum number of instructions ObjectSizeOffsetVisitor is allowed to look at.
• --orc-runtime=<string> — uses the ORC runtime from the given path.
• --per-module-lazy — lazy compilation happens on whole module boundaries instead of on individual functions. zoom out, look at the whole picture 🦋
• --pgo-block-coverage — use this one to enable basic block coverage instrumentation.
• --pgo-temporal-instrumentation — use this one to enable temporal instrumentation.
• --pgo-view-block-coverage-graph — creates a dot file of CFGs with block coverage inference information. graph girlies rise 🎀
• --poison-checking-function-local — checks that returns are non-poison. (for testing!)
• --print-pipeline-passes — prints a -passes compatible string describing the pipeline. and it's best-effort only — no promises, we're doing our best 😭
• --profile-correlate=<value> — uses debug info or a binary file to correlate profiles:
  • =<empty> — No profile correlation
  • =debug-info — Use debug info to correlate
  • =binary — Use binary to correlate
• --r600-ir-structurize — use the StructurizeCFG IR pass.
• --relax-elf-relocations — emits GOTPCRELX/REX_GOTPCRELX instead of GOTPCREL on x86-64 ELF.
• --relocation-model=<value> — picks your relocation model, and the menu is long:
  • =static — Non-relocatable code
  • =pic — Fully relocatable, position independent code
  • =dynamic-no-pic — Relocatable external references, non-relocatable code
  • =ropi — Code and read-only data relocatable, accessed PC-relative
  • =rwpi — Read-write data relocatable, accessed relative to static base
  • =ropi-rwpi — Combination of ropi and rwpi
• --remote-mcjit — executes MCJIT'ed code in a separate process.
• --riscv-add-build-attributes — (no description given, it simply exists and it's iconic for that 👑)
• --riscv-use-aa — enables the use of AA during codegen.
• --runtime-counter-relocation — enables relocating counters at runtime.
• --safepoint-ir-verifier-print-only — (again, no description — it's mysterious, it's lowkey main character energy 🖤)
• --sample-profile-check-record-coverage=<N> — emits a warning if less than N% of records in the input profile are matched to the IR.
• --sample-profile-check-sample-coverage=<N> — emits a warning if less than N% of samples in the input profile are matched to the IR.
• --sample-profile-max-propagate-iterations=<uint> — the maximum number of iterations it'll go through when propagating sample block/edge weights through the CFG. ✨boundaries✨
• --skip-ret-exit-block — suppresses counter promotion if exit blocks contain ret.
• --soft-float — generates software floating point library calls.
• --speculative-counter-promotion-max-exiting=<uint> — the max number of exiting blocks of a loop that's allowed for speculative counter promotion.
• --speculative-counter-promotion-to-loop — when this option is false, and the target block is in a loop, promotion gets disallowed — unless the promoted counter update can be further/iteratively promoted into an acyclic region. yes it's a whole conditional, we love a plot twist 😭
• --split-machine-functions — splits out cold basic blocks from machine functions, based on profile information. the cold blocks get their own little section ❄️
• --stack-size-section — emits a section containing stack size metadata.
• --stack-symbol-ordering — orders local stack symbols.
• --stackrealign — forces the stack to align to the minimum alignment. get in line bestie 📏
• --strict-dwarf — use strict dwarf.
• --summary-file=<string> — the summary file to use for function importing.
• --sve-tail-folding=<string> — controls the use of vectorisation using tail-folding for SVE, and the option gets specified in the form (Initial)[+(Flag1|Flag2|...)]:
  • disabled — (Initial) No loop types will vectorize using tail-folding
  • default — (Initial) Uses the default tail-folding settings for the target CPU
  • all — (Initial) All legal loop types will vectorize using tail-folding
  • simple — (Initial) Use tail-folding for simple loops (not reductions or recurrences)
  • reductions — Use tail-folding for loops containing reductions
  • noreductions — Inverse of above
  • recurrences — Use tail-folding for loops containing fixed order recurrences
  • norecurrences — Inverse of above
  • reverse — Use tail-folding for loops requiring reversed predicates
  • noreverse — Inverse of above
• --swift-async-fp=<value> — tells the compiler when it should set the Swift async frame pointer bit 💅
  • =auto — just figure it out from the deployment target, queen decides
  • =always — always set the bit, no questions
  • =never — never set the bit, periodt
• --tail-predication=<value> — the MVE tail-predication pass options, aka loop glow-up settings ✨
  • =disabled — don't tail-predicate loops at all
  • =enabled-no-reductions — enable tail-predication, but not for reduction loops
  • =enabled — enable tail-predication, reduction loops included, we're going all in
  • =force-enabled-no-reductions — enable tail-predication but not for reduction loops, and force it, which might be unsafe 😭
  • =force-enabled — enable tail-predication including reduction loops, and force it, which might be unsafe 😭
• --tailcallopt — turns fastcc calls into tail calls by (potentially) changing the ABI, lowkey chaotic but ok
• --thinlto-assume-merged — assumes the input has already gone through ThinLTO function importing and the other pre-optimization pipeline changes, so don't come for it
• --thread-entry=<string> — calls the given entry-point on a new thread (only for jit-kind=orc-lazy btw)
• --thread-model=<value> — pick your threading model 🧵
  • =posix — POSIX thread model
  • =single — single thread model, main character energy only
• --threads=<int> — ok so the page literally gives no description for this one, it's giving mysterious 🙈
• --tls-size=<uint> — bit size of immediate TLS offsets
• --type-based-intrinsic-cost — calculate intrinsics cost based only on argument types, no cap
• --unique-basic-block-section-names — gives a unique name to every basic block section 💖
• --unique-section-names — gives a unique name to every section, everybody gets their own little name tag 🎀
• --use-ctors — use .ctors instead of .init_array.
• --vec-extabi — enable the AIX Extended Altivec ABI.
• --verify-region-info — verify region info (heads up: time consuming 🐌)
• --vp-counters-per-site=<number> — the average number of profile counters allocated per value profiling site
• --vp-static-alloc — do static counter allocation for value profiler
• --wasm-enable-eh — WebAssembly exception handling
• --wasm-enable-sjlj — WebAssembly setjmp/longjmp handling
• --x86-align-branch=<string> — specify types of branches to align (it's a plus-separated list of types, slay):
  • jcc — indicates conditional jumps
  • fused — indicates fused conditional jumps
  • jmp — indicates direct unconditional jumps
  • call — indicates direct and indirect calls
  • ret — indicates rets
  • indirect — indicates indirect unconditional jumps
• --x86-align-branch-boundary=<uint> — controls how the assembler should align branches with NOP. If the boundary's size is not 0, it must be a power of 2 and no less than 32. Branches get aligned so they're not across or against the boundary of the specified size. The default value 0 does not align branches.
• --x86-branches-within-32B-boundaries — aligns selected instructions to mitigate the negative performance impact of Intel's micro code update for errata skx102. ⚠️ This may break assumptions about labels corresponding to particular instructions, and should be used with caution.
• --x86-pad-max-prefix-size=<uint> — maximum number of prefixes to use for padding
• --xcoff-traceback-table — emit the XCOFF traceback table
• --xray-function-index — emit xray_fn_idx section

---

✨ Generic Options ✨

• --help — display available options (there's --help-hidden for more 👀)
• --help-list — display list of available options (again, --help-list-hidden for more 👀)
• --version — display the version of this program

---

🎀 Polly Options 🎀

Configure the polly loop optimizer, bestie 💜

• --polly — enable the polly optimizer (works with -O1, -O2 or -O3)
• --polly-2nd-level-tiling — enable a 2nd level loop of loop tiling, tiling on tiling fr fr
• --polly-ast-print-accesses — print memory access functions
• --polly-context=<isl parameter set> — provide additional constraints on the context parameters
• --polly-dce-precise-steps=<int> — the number of precise steps between two approximating iterations. (A value of -1 schedules another approximation stage before the actual dead code elimination.
• --polly-delicm-max-ops=<int> — maximum number of isl operations to invest for lifetime analysis; 0 means no limit 🌈
• --polly-detect-full-functions — allow the detection of full functions
• --polly-dump-after — dump the module after Polly transformations into a file suffixed with "-after"
• --polly-dump-after-file=<string> — dump the module after Polly transformations to the given file
• --polly-dump-before — dump the module before Polly transformations into a file suffixed with "-before"
• --polly-dump-before-file=<string> — dump the module before Polly transformations to the given file
• --polly-enable-simplify — simplify SCoP after optimizations
• --polly-ignore-func=<string> — ignore functions that match a regex. Multiple regexes can be comma separated. Scop detection will ignore all functions that match ANY of the regexes provided.
• --polly-isl-arg=<argument> — option passed to ISL
• --polly-matmul-opt — perform optimizations of matrix multiplications based on pattern matching
• --polly-on-isl-error-abort — abort if an isl error is encountered 💀
• --polly-only-func=<string> — only run on functions that match a regex. Multiple regexes can be comma separated. Scop detection will run on all functions that match ANY of the regexes provided.
• --polly-only-region=<identifier> — only run on certain regions (the provided identifier must appear in the name of the region's entry block
• --polly-only-scop-detection — only run scop detection, but no other optimizations
• --polly-optimized-scops — Polly - dump polyhedral description of Scops optimized with the isl scheduling optimizer and the set of post-scheduling transformations applied on the schedule tree
• --polly-parallel — generate thread parallel code (isl codegen only)
• --polly-parallel-force — force generation of thread parallel code, ignoring any cost model 😤
• --polly-pattern-matching-based-opts — perform optimizations based on pattern matching
• --polly-postopts — apply post-rescheduling optimizations such as tiling (requires -polly-reschedule)
• --polly-pragma-based-opts — apply user-directed transformation from metadata
• --polly-pragma-ignore-depcheck — skip the dependency check for pragma-based transformations
• --polly-process-unprofitable — process scops that are unlikely to benefit from Polly optimizations.
• --polly-register-tiling — enable register tiling
• --polly-report — print information about the activities of Polly
• --polly-reschedule — optimize SCoPs using ISL
• --polly-show — highlight the code regions that will be optimized in a (CFG BBs and LLVM-IR instructions)
• --polly-show-only — highlight the code regions that will be optimized in a (CFG only BBs)
• --polly-stmt-granularity=<value> — algorithm to use for splitting basic blocks into multiple statements
  • =bb — one statement per basic block
  • =scalar-indep — scalar independence heuristic
  • =store — store-level granularity
• --polly-tc-opt — perform optimizations of tensor contractions based on pattern matching
• --polly-tiling — enable loop tiling
• --polly-vectorizer=<value> — select the vectorization strategy
  • =none — no vectorization
  • =stripmine — strip-mine outer loops for the loop-vectorizer to trigger

---

And that's a wrap on the options, bestie 🫶 this page is from lli 18, dated June 2025, and it belongs to the LLI(1) manual. Go forth and flag responsibly 💖✨