diff options
| author | Cyril SIX <cyril.six@kalray.eu> | 2020-10-14 11:08:55 +0200 |
|---|---|---|
| committer | Cyril SIX <cyril.six@kalray.eu> | 2020-10-14 11:08:55 +0200 |
| commit | b09b9a2e3f3f2612582bbf7ee624a48ad0e0b40f (patch) | |
| tree | 2e685120a97c2cb77fa61d73bcb57603dcfc4639 /driver/Driver.ml | |
| parent | 7642f95b4a2c577316ae1b9696ab0d9ba4cf8b2d (diff) | |
| download | compcert-kvx-b09b9a2e3f3f2612582bbf7ee624a48ad0e0b40f.tar.gz compcert-kvx-b09b9a2e3f3f2612582bbf7ee624a48ad0e0b40f.zip | |
Updated --help
Diffstat (limited to 'driver/Driver.ml')
| -rw-r--r-- | driver/Driver.ml | 14 |
1 files changed, 5 insertions, 9 deletions
diff --git a/driver/Driver.ml b/driver/Driver.ml index b0b1cdea..fae1524f 100644 --- a/driver/Driver.ml +++ b/driver/Driver.ml @@ -207,15 +207,11 @@ Processing options: -fpostpass Perform postpass scheduling (only for K1 architecture) [on] -fpostpass= <optim> Perform postpass scheduling with the specified optimization [list] (<optim>=list: list scheduling, <optim>=ilp: ILP, <optim>=greedy: just packing bundles) - -fduplicate <nb_nodes> Perform tail duplication to form superblocks on predicted traces - nb_nodes control the heuristic deciding to duplicate or not - A value of -1 desactivates the entire pass (including branch prediction) - A value of 0 desactivates the duplication (but activates the branch prediction) - FIXME : this is desactivated by default for now - -finvertcond Invert conditions based on predicted paths (to prefer fallthrough). - Requires -fduplicate to be also activated [on] - -ftracelinearize Linearizes based on the traces identified by duplicate phase - It is heavily recommended to activate -finvertcond with this pass [off] + -fpredict Insert static branch prediction information [on] + Also swaps ifso/ifnot branches accordingly at RTL level + -ftailduplicate n Perform tail duplication for RTL code blocks of size n (not counting Inops) [0] + -ftracelinearize Uses branch prediction information to improve the Linearize [on] + -funrollsingle n Unrolls a single iteration of innermost loops of size n (not counting Inops) [0] -fforward-moves Forward moves after CSE -finline Perform inlining of functions [on] -finline-functions-called-once Integrate functions only required by their |