diff options
| author | Xavier Leroy <xavier.leroy@college-de-france.fr> | 2019-05-17 16:17:51 +0200 |
|---|---|---|
| committer | Xavier Leroy <xavier.leroy@college-de-france.fr> | 2019-06-07 09:08:40 +0200 |
| commit | b3bfdc3655479f7f8f1c5e3a7571473a72b421cb (patch) | |
| tree | 8fdc27ef9b7c186ee3af9c04327a28b7e8424fd1 /x86/Op.v | |
| parent | 8e3a73448c5ddfa4be3871d7f4fd80281a7549f4 (diff) | |
| download | compcert-b3bfdc3655479f7f8f1c5e3a7571473a72b421cb.tar.gz compcert-b3bfdc3655479f7f8f1c5e3a7571473a72b421cb.zip | |
x86 branchless implementation of unsigned int32 -> float conversion
Based on a suggestion by Rémi Hutin.
The new implementation produces smaller code (3 FP instructions and 2
integer "and") and has no conditional branch.
On average, it is a little slower than the old implementation.
Hence both implementations are supported:
- with -ffavor-branchless, use the new implementation
- by default, use the old, branching implementation
Diffstat (limited to 'x86/Op.v')
0 files changed, 0 insertions, 0 deletions