Files
platform-external-webrtc/rtc_base/robo_caller.cc
Karl Wiberg d2c6967f1d Optimize RoboCaller::AddReceiver() for code size
Essentially, instead of having the inlined UntypedFunction::Create(f)
return an UntypedFunction which is then passed as an argument to
non-inlined RoboCallerReceivers::AddReceiverImpl(), we let
UntypedFunction::PrepareArgs(f) return a few different kinds of
trivial structs (depending on what sort of type f has) which are
passed as arguments to non-inlined RoboCallerReceivers::AddReceiver()
(which then converts them to UntypedFunction by calling
UntypedFunction::Create()). These structs are smaller than
UntypedFunction and optimized for argument passing, so many fewer
instructions are needed.

Example code:

  struct Foo {
    void Receive(int, float, int, float);
    void TestAddLambdaReceiver();
    webrtc::RoboCaller<int, float, int, float> rc;
  };

  void Foo::TestAddLambdaReceiver() {
    rc.AddReceiver([this](int a, float b, int c, float d){
        Receive(a, b, c, d);});
  }

On arm32, we get before this CL:

  Foo::TestAddLambdaReceiver():
        push    {r11, lr}
        mov     r11, sp
        sub     sp, sp, #24
        ldr     r1, .LCPI0_0
        mov     r2, #0
        stm     sp, {r0, r2}
        add     r1, pc, r1
        str     r2, [sp, #20]
        str     r1, [sp, #16]
        mov     r1, sp
        bl      RoboCallerReceivers::AddReceiverImpl
        mov     sp, r11
        pop     {r11, pc}
  .LCPI0_0:
        .long   CallInlineStorage<Foo::TestAddLambdaReceiver()::$_0>
  CallInlineStorage<Foo::TestAddLambdaReceiver()::$_0>:
        ldr     r0, [r0]
        b       Foo::Receive(int, float, int, float)

After this CL:

  Foo::TestAddLambdaReceiver():
        ldr     r3, .LCPI0_0
        mov     r2, r0
        add     r3, pc, r3
        b       RoboCallerReceivers::AddReceiver<1u>
  .LCPI0_0:
        .long   CallInlineStorage<Foo::TestAddLambdaReceiver()::$_0>
  CallInlineStorage<Foo::TestAddLambdaReceiver()::$_0>:
        ldr     r0, [r0]
        b       Foo::Receive(int, float, int, float)

(Symbol names abbreviated so that they'll fit on one line.)

So a reduction from 64 to 28 bytes. The improvements on arm64 and
x86_64 are similar.

Bug: webrtc:11943
Change-Id: I93fbba083be0235051c3279d3e3f6852a4a9fdad
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/185960
Commit-Queue: Karl Wiberg <kwiberg@webrtc.org>
Reviewed-by: Mirko Bonadei <mbonadei@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#32244}
2020-09-30 06:59:44 +00:00

41 lines
1.4 KiB
C++

/*
* Copyright 2020 The WebRTC Project Authors. All rights reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include "rtc_base/robo_caller.h"
namespace webrtc {
namespace robo_caller_impl {
RoboCallerReceivers::RoboCallerReceivers() = default;
RoboCallerReceivers::~RoboCallerReceivers() = default;
void RoboCallerReceivers::Foreach(
rtc::FunctionView<void(UntypedFunction&)> fv) {
for (auto& r : receivers_) {
fv(r);
}
}
template void RoboCallerReceivers::AddReceiver(
UntypedFunction::TrivialUntypedFunctionArgs<1>);
template void RoboCallerReceivers::AddReceiver(
UntypedFunction::TrivialUntypedFunctionArgs<2>);
template void RoboCallerReceivers::AddReceiver(
UntypedFunction::TrivialUntypedFunctionArgs<3>);
template void RoboCallerReceivers::AddReceiver(
UntypedFunction::TrivialUntypedFunctionArgs<4>);
template void RoboCallerReceivers::AddReceiver(
UntypedFunction::NontrivialUntypedFunctionArgs);
template void RoboCallerReceivers::AddReceiver(
UntypedFunction::FunctionPointerUntypedFunctionArgs);
} // namespace robo_caller_impl
} // namespace webrtc