Sign Up
Log In
Log In
or
Sign Up
Places
All Projects
Status Monitor
Collapse sidebar
home:Ledest:erlang:18
erlang
0621-emulator-Document-the-futility-of-aligning...
Overview
Repositories
Revisions
Requests
Users
Attributes
Meta
File 0621-emulator-Document-the-futility-of-aligning-opcode-la.patch of Package erlang
From 013eaf3ea872048db1ffa5da5c6dcfba07b4141b Mon Sep 17 00:00:00 2001 From: Frej Drejhammar <frej.drejhammar@gmail.com> Date: Fri, 28 Aug 2020 10:15:45 +0200 Subject: [PATCH] emulator: Document the futility of aligning opcode labels The x86 and AMD64 optimization guides recommend that branch targets are paragraph aligned (16 bytes). Neither GCC nor Clang align the addresses of first class labels as used by the OpCase construction when the jump table is enabled. Aligning the OpCase labels looks like a worthwhile optimization, but despite the recommendation in the optimization guides, forcing alignment leads to a ~5% slowdown of the emulator. Therefore alignment of OpCase labels are not attempted. As at least two OTP developers have independently attempted this optimization and measured the same slowdown, this patch adds a comment documenting this fact. The comment is placed where the OpCase macro is defined, as a person trying to implement alignment will likely look there. --- erts/emulator/beam/beam_emu.c | 9 +++++++++ 1 file changed, 9 insertions(+) diff --git a/erts/emulator/beam/beam_emu.c b/erts/emulator/beam/beam_emu.c index 2ed144b9b6..b6b19b686b 100644 --- a/erts/emulator/beam/beam_emu.c +++ b/erts/emulator/beam/beam_emu.c @@ -48,6 +48,15 @@ /* #define HARDDEBUG 1 */ +/* The x86 and AMD64 optimization guides recommend that branch targets are + * paragraph aligned (16 bytes). Neither GCC nor Clang align the addresses of + * first class labels as used by the OpCase construction when the jump table is + * enabled. Aligning the OpCase labels looks like a worthwhile optimization, but + * despite the recommendation in the optimization guides, forcing alignment + * leads to a ~5% slowdown of the emulator. Therefore alignment of OpCase labels + * are not attempted. + */ + #if defined(NO_JUMP_TABLE) # define OpCase(OpCode) case op_##OpCode # define CountCase(OpCode) case op_count_##OpCode -- 2.26.2
Locations
Projects
Search
Status Monitor
Help
OpenBuildService.org
Documentation
API Documentation
Code of Conduct
Contact
Support
@OBShq
Terms
openSUSE Build Service is sponsored by
The Open Build Service is an
openSUSE project
.
Sign Up
Log In
Places
Places
All Projects
Status Monitor