USAGE: llvm-exegesis [options] OPTIONS: Color Options: --color - Use colors in output (default=autodetect) General options: --allow-ginsert-as-artifact - Allow G_INSERT to be considered an artifact. Hack around AMDGPU test infinite loops. --atomic-counter-update-promoted - Do counter update using atomic fetch add for promoted counters only --atomic-first-counter - Use atomic fetch add for first counter in a function (usually the entry counter) --bounds-checking-single-trap - Use one trap block per function --debug-info-correlate - Use debug info to correlate profiles. --debugify-level= - Kind of debug info to add =locations - Locations only =location+variables - Locations and Variables --debugify-quiet - Suppress verbose debugify output --disable-i2p-p2i-opt - Disables inttoptr/ptrtoint roundtrip optimization --do-counter-promotion - Do counter register promotion --dot-cfg-mssa= - file name for generated dot file --enable-cse-in-irtranslator - Should enable CSE in irtranslator --enable-cse-in-legalizer - Should enable CSE in Legalizer --enable-name-compression - Enable name/filename string compression --experimental-debug-variable-locations - Use experimental new value-tracking variable locations --fs-profile-debug-bw-threshold= - Only show debug message if the source branch weight is greater than this value. --fs-profile-debug-prob-diff-threshold= - Only show debug message if the branch probility is greater than this value (in percentage). --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. --hash-based-counter-split - Rename counter variable of a comdat function based on cfg hash --instrprof-atomic-counter-update-all - Make all profile counter updates atomic (for testing only) --iterative-counter-promotion - Allow counter promotion across the whole loop nest. --max-counter-promotions= - Max number of allowed counter promotions --max-counter-promotions-per-loop= - Max number counter promotions per loop to avoid increasing register pressure too much --mir-strip-debugify-only - Should mir-strip-debug only strip debug info from debugified modules by default --opaque-pointers - Use opaque pointers --runtime-counter-relocation - Enable relocating counters at runtime. --sample-profile-check-record-coverage= - Emit a warning if less than N% of records in the input profile are matched to the IR. --sample-profile-check-sample-coverage= - Emit a warning if less than N% of samples in the input profile are matched to the IR. --sample-profile-max-propagate-iterations= - Maximum number of iterations to go through when propagating sample block/edge weights through the CFG. --skip-ret-exit-block - Suppress counter promotion if exit blocks contain ret. --speculative-counter-promotion-max-exiting= - The max number of exiting blocks of a loop to allow speculative counter promotion --speculative-counter-promotion-to-loop - When the option is false, if the target block is in a loop, the promotion will be disallowed unless the promoted counter update can be further/iteratively promoted into an acyclic region. --verify-region-info - Verify region info (time consuming) --vp-counters-per-site= - The average number of profile counters allocated per value profiling site. --vp-static-alloc - Do static counter allocation for value profiler --x86-align-branch= - Specify types of branches to align (plus separated list of types): 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= - Control how the assembler should align branches with NOP. If the boundary's size is not 0, it should be a power of 2 and no less than 32. Branches will be aligned to prevent from being across or against the boundary of specified size. The default value 0 does not align branches. --x86-branches-within-32B-boundaries - Align selected instructions to mitigate negative performance impact of Intel's micro code update for errata skx102. May break assumptions about labels corresponding to particular instructions, and should be used with caution. --x86-pad-max-prefix-size= - Maximum number of prefixes to use for padding Generic Options: --help - Display available options (--help-hidden for more) --help-list - Display list of available options (--help-list-hidden for more) --version - Display the version of this program llvm-exegesis analysis options: --analysis-clustering= - the clustering algorithm to use =dbscan - use DBSCAN/OPTICS algorithm =naive - one cluster per opcode --analysis-clustering-epsilon= - epsilon for benchmark point clustering --analysis-clusters-output-file= - --analysis-display-unstable-clusters - if there is more than one benchmark for an opcode, said benchmarks may end up not being clustered into the same cluster if the measured performance characteristics are different. by default all such opcodes are filtered out. this flag will instead show only such unstable opcodes --analysis-inconsistencies-output-file= - --analysis-inconsistency-epsilon= - epsilon for detection of when the cluster is different from the LLVM schedule profile values --analysis-numpoints= - minimum number of points in an analysis cluster (dbscan only) llvm-exegesis benchmark options: --dump-object-to-disk - dumps the generated benchmark object to disk and prints a message to access it --ignore-invalid-sched-class - ignore instructions that do not define a sched class --loop-body-size= - when repeating the instruction snippet by looping over it, duplicate the snippet until the loop body contains at least this many instruction --max-configs-per-opcode= - allow to snippet generator to generate at most that many configs --num-repetitions= - number of time to repeat the asm snippet --opcode-index= - opcode to measure, by index, or -1 to measure all opcodes --opcode-name= - comma-separated list of opcodes to measure, by name --repetition-mode= - how to repeat the instruction snippet =duplicate - Duplicate the snippet =loop - Loop over the snippet =min - All of the above and take the minimum of measurements --snippets-file= - code snippets to measure llvm-exegesis benchmark x86-options: --x86-lbr-sample-period= - The sample period (nbranches/sample), used for LBR sampling llvm-exegesis options: --benchmarks-file= - File to read (analysis mode) or write (latency/uops/inverse_throughput modes) benchmark results. “-” uses stdin/stdout. --mcpu= - cpu name to use for pfm counters, leave empty to autodetect --mode= - the mode to run =latency - Instruction Latency =inverse_throughput - Instruction Inverse Throughput =uops - Uop Decomposition =analysis - Analysis --result-aggregation-mode= - How to aggregate multi-values result =min - Keep min reading =max - Keep max reading =mean - Compute mean of all readings =min-variance - Keep readings set with min-variance