Google Git
Sign in
fuchsia / integration / e98f7a96f249d4eab83fac78d7f584f02ee1b99c
commite98f7a96f249d4eab83fac78d7f584f02ee1b99c[log] [tgz]
authorYifei Teng <yifeit@fuchsia.infra.roller.google.com>Wed Aug 18 23:22:00 2021 +0000
committerCopybara-Service <copybara-worker@google.com>Wed Aug 18 16:23:21 2021 -0700
tree536917ee7f2d096e35a2ce04242d4a1413c636ac
parent091a0a5e1bc2f6be364f855afb72143f589ef15c [diff]
[roll] Roll fuchsia [fidl][llcpp] Avoid redundant buffer initialization

Disables automatic variable initialization [1] in cases where it's safe,
specifically in generated code and in FIDL message buffer types that is
only used for encoding/decoding.

Unfortunately the disabling is a bit less well encapsulated than we'd
like, since one has to apply it on a per-variable basis.

The macro definition is inspired by [2].

See [3] for a godbolt showing the difference in generated code.

LLCPP performance generally improved. Encoding a byte vector went from
>2000ns to ~200ns since we are no longer filling the 64 KiB buffer.

Examples perfcompare data:

LLCPP/Encode/BoolArray/256/Steps.Encode/WallTime                            faster            0.987-0.997    1720.3 +/- 8.0 ns             1706.56 +/- 0.49 ns
LLCPP/Encode/ByteArray/16/Steps.Encode/WallTime                             faster            0.856-0.882    130.8 +/- 1.2 ns              113.61 +/- 0.66 ns
LLCPP/Encode/ByteArray/16/Steps.Teardown/WallTime                           faster            0.962-0.996    14.31 +/- 0.13 ns             14.00 +/- 0.11 ns
LLCPP/Encode/ByteArray/256/Steps.Encode/WallTime                            faster            0.784-0.809    130.8 +/- 1.3 ns              104.12 +/- 0.61 ns
LLCPP/Encode/ByteArray/256/Steps.Teardown/WallTime                          faster            0.957-0.977    15.316 +/- 0.031 ns           14.81 +/- 0.12 ns
LLCPP/Encode/ByteVector/16/Steps.Encode/WallTime                            faster            0.106-0.115    2124 +/- 70 ns                235.1 +/- 1.8 ns
LLCPP/Encode/ByteVector/256/Steps.Encode/WallTime                           faster            0.107-0.117    2122 +/- 69 ns                237.5 +/- 2.5 ns
LLCPP/Encode/ByteVector/4096/Steps.Encode/WallTime                          faster            0.091-0.100    2079 +/- 70 ns                198.0 +/- 3.1 ns
LLCPP/Encode/ByteVector/4096/Steps.Setup/WallTime                           faster            0.983-0.999    2170 +/- 12 ns                2151.0 +/- 5.7 ns
LLCPP/Encode/ByteVector/65280/Steps.Encode/WallTime                         faster            0.091-0.098    2044 +/- 69 ns                192.6 +/- 1.3 ns
LLCPP/Encode/ByteVectorLimit/1/Steps.Encode/WallTime                        faster            0.821-0.925    145.4 +/- 3.4 ns              126.7 +/- 4.6 ns
LLCPP/Encode/DatagramSocket/RecvMsg/Request/Steps.Encode/WallTime           faster            0.813-0.899    132.0 +/- 2.8 ns              112.9 +/- 3.3 ns
LLCPP/Encode/DatagramSocket/RecvMsg/Request/Steps.Teardown/WallTime         slower            1.025-1.036    13.780 +/- 0.030 ns           14.200 +/- 0.042 ns
LLCPP/Encode/DatagramSocket/RecvMsg/Result/Steps.Encode/WallTime            faster            0.129-0.143    2155 +/- 77 ns                293.3 +/- 4.8 ns
LLCPP/Encode/DatagramSocket/SendMsg2/Request/Steps.Encode/WallTime          faster            0.105-0.119    2105 +/- 76 ns                235.4 +/- 6.5 ns
LLCPP/Encode/DatagramSocket/SendMsg2/Result/Steps.Encode/WallTime           faster            0.857-0.940    173.0 +/- 2.9 ns              155.2 +/- 4.5 ns

In particular, the time it takes to round-trip a sync call more
than halved, since we also avoid buffer filling in the async binding:

LLCPP/EchoCall/ByteVector/16/Steps.EchoCall/WallTime                        faster            0.404-0.426    13300 +/- 108 ns              5514 +/- 102 ns

See [4] for the full data.

[1]: https://reviews.llvm.org/D54604
[2]: https://chromium.googlesource.com/chromium/src/base/+/8f4dbc27d7a19efd35181ddd533523ae6371a03d/compiler_specific.h#276
[3]: https://godbolt.org/z/Tsjo3bf9W
[4]: https://logs.chromium.org/logs/fuchsia/buildbucket/cr-buildbucket.appspot.com/8838599435516804016/+/u/compare_perf_test_results_without_and_with_CL/stdout?format=raw

Original-Reviewed-on: https://fuchsia-review.googlesource.com/c/fuchsia/+/570303
Original-Revision: 2b581926a60660e8fd36ac01ffdc3c22e7cf8b2e
GitOrigin-RevId: 0d1f2a032bca25605cd742964e865625ad65732c
Change-Id: I7a68a134cf00b1a6f74d2af728be83efc5ceb306
1 file changed
tree: 536917ee7f2d096e35a2ce04242d4a1413c636ac
  1. garnet/
  2. infra/
  3. peridot/
  4. third_party/
  5. topaz/
  6. zircon/
  7. firmware
  8. flower
  9. jiri.lock
  10. minimal
  11. prebuilts
  12. README.md
  13. stem
  14. test_durations
README.md

Integration

This repository contains Fuchsia's Global Integration manifest files.

Making changes

All changes should be made to the internal version of this repository. Our infrastructure automatically updates this version when the internal one changes.

Currently all changes must be made by a Google employee. Non-Google employees wishing to make a change can ask for assistance via the IRC channel #fuchsia on Freenode.

Obtaining the source

First install Jiri.

Next run:

$ jiri init
$ jiri import minimal https://fuchsia.googlesource.com/integration
$ jiri update

Third party

Third party projects should have their own subdirectory in ./third_party.