diff options
| author | Princeton Ferro <pferro@nvidia.com> | 2025-10-01 12:08:49 -0700 |
|---|---|---|
| committer | GitHub <noreply@github.com> | 2025-10-01 12:08:49 -0700 |
| commit | 9e04291fd20489882259625ec327b87ecef6fa8c (patch) | |
| tree | 6f84b2eeb54a49ce87c3d51ed1e4ae86b940dc46 /libcxx/test/selftest/pass.cpp/link-error.pass.cpp | |
| parent | e84dcba9246d696715fe1daa4ddb218182580a70 (diff) | |
| download | llvm-9e04291fd20489882259625ec327b87ecef6fa8c.zip llvm-9e04291fd20489882259625ec327b87ecef6fa8c.tar.gz llvm-9e04291fd20489882259625ec327b87ecef6fa8c.tar.bz2 | |
[InstCombine] linearize complexity of `findDemandedEltsByAllUsers()` (#161436)
Each call to `findemandedEltsBySingleUser()` returns a new APInt that
must be OR'd with the current APInt. For large vectors with many uses
this can be slow, if the total number of operations is `{# uses} x {size
of vector}`. Instead or OR'ing, use `setBit()` on the passed-in APInt.
Diffstat (limited to 'libcxx/test/selftest/pass.cpp/link-error.pass.cpp')
0 files changed, 0 insertions, 0 deletions