aboutsummaryrefslogtreecommitdiff
path: root/gcc/expr.cc
diff options
context:
space:
mode:
authorKugan Vivekanandarajah <kvivekananda@nvidia.com>2024-10-31 07:20:49 +1100
committerKugan Vivekanandarajah <kvivekananda@nvidia.com>2024-10-31 07:21:30 +1100
commit5be5c66071b407a767856b8fa300ede54fcf11b4 (patch)
tree3538c4326193cc5aa78fb5b5b0c1cfd1c757e174 /gcc/expr.cc
parent857145bf9a2168c8037820148b8894fbd74a823b (diff)
downloadgcc-5be5c66071b407a767856b8fa300ede54fcf11b4.zip
gcc-5be5c66071b407a767856b8fa300ede54fcf11b4.tar.gz
gcc-5be5c66071b407a767856b8fa300ede54fcf11b4.tar.bz2
[PATCH] Adjust param_vect_max_version_for_alias_checks
This patch sets param_vect_max_version_for_alias_checks to 15. This was causing GCC to miss vectorization opportunities for an application, making it slower than LLVM by about ~14%. Original default of 10 itself is arbitary. Given that, GCC's vectoriser does consideres cost of alias checks, increasing this param is reasonable. In this case we need a value of at teast 11 whereas the current default is 10. gcc/ChangeLog: * params.opt: Adjust param_vect_max_version_for_alias_checks gcc/testsuite/ChangeLog: * g++.dg/alias-checks.C: New test. Signed-off-by: Kugan Vivekanandarajah <kvivekananda@nvidia.com>
Diffstat (limited to 'gcc/expr.cc')
0 files changed, 0 insertions, 0 deletions