| commit | 37e4b35abb2a72ee9180413809d8903699d26db0 | |
| author | Paul Bissonnette <paulbiss@fb.com> | |
| Mon, 6 May 2019 17:42:48 +0000 (6 10:42 -0700) | ||
| committer | Hhvm Bot <hhvm-bot@users.noreply.github.com> | |
| Mon, 6 May 2019 18:06:07 +0000 (6 11:06 -0700) | ||
| tree | a62fc5588c28b9513970de3d4317f4f1e2372253 | treesnapshot (tar.gz zip) |
| parent | 2c2538a350952011e3ff624bfdde0aa5ff32f610 | commitdiff |
Improve memory effects for inlining instructions
Summary: I have a few more ideas for improving store-elim/load-elim leveraging the custom memory effects types for DefInlineFP/InlineReturn. I've noticed that even though we try to eliminate the frame entirely sometimes the memory effects for DefInlineFP and InlineReturn can mess with load/store elim in ways that prevent instructions that pin the frame from being elided.
Reviewed By: ricklavoie
Differential Revision: D15215193
fbshipit-source-id: 38637e25a44143ea1541cc121c3d37c68d891462
Summary: I have a few more ideas for improving store-elim/load-elim leveraging the custom memory effects types for DefInlineFP/InlineReturn. I've noticed that even though we try to eliminate the frame entirely sometimes the memory effects for DefInlineFP and InlineReturn can mess with load/store elim in ways that prevent instructions that pin the frame from being elided.
Reviewed By: ricklavoie
Differential Revision: D15215193
fbshipit-source-id: 38637e25a44143ea1541cc121c3d37c68d891462