| commit | 0075459d0ab88ff4a5b05ba2c008c39774dfa6cb | |
| author | Max Wang <mwang@fb.com> | |
| Wed, 27 Jan 2016 20:17:20 +0000 (27 12:17 -0800) | ||
| committer | hhvm-bot <hhvm-bot@fb.com> | |
| Wed, 27 Jan 2016 20:30:45 +0000 (27 12:30 -0800) | ||
| tree | 1f05c46b6adac202f54395960cbaadc744577a39 | treesnapshot (tar.gz zip) |
| parent | 737483672ea1c60efc26b9883ddbd3c99fab1fdd | commitdiff |
Use lea{} instead of addqi{}/subqi{} for rsp()
Summary:
This makes things a bit nicer for non-x64 land, where there are
sometimes further restrictions on the native stack pointer.
Two of these changes are in very slow (intercept-related) paths in
unique stubs, which is surely not perf-sensitive.
The other two are for lowering vcall{}. Interestingly, we already use
lea{} for pushing vcall{}'s args, just not for popping them. Probably
doing both won't be the end of the world, and if perflab is happy, I
think this is Fine (TM).
Reviewed By: markw65
Differential Revision: D2869189
fb-gh-sync-id: 649ea58f11e4e34556456b4c0c7ccf467a3e3f83
Summary:
This makes things a bit nicer for non-x64 land, where there are
sometimes further restrictions on the native stack pointer.
Two of these changes are in very slow (intercept-related) paths in
unique stubs, which is surely not perf-sensitive.
The other two are for lowering vcall{}. Interestingly, we already use
lea{} for pushing vcall{}'s args, just not for popping them. Probably
doing both won't be the end of the world, and if perflab is happy, I
think this is Fine (TM).
Reviewed By: markw65
Differential Revision: D2869189
fb-gh-sync-id: 649ea58f11e4e34556456b4c0c7ccf467a3e3f83
| hphp/runtime/vm/jit/unique-stubs-x64.cpp | diffblobblamehistory | |
| hphp/runtime/vm/jit/vasm-lower.cpp | diffblobblamehistory |