citra-nightly/src/audio_core/interpolate.cpp
Subv d7459354f5 Audio: Use std::deque instead of std::vector for the audio buffer type (StereoBuffer16).
The current code inserts and deletes elements from the beginning of the audio buffer, which is very inefficient in an std::vector.

Profiling was done using VisualStudio2017's Performance Analyzer in Super Mario 3D Land.

Before this change: AudioInterp::Linear had 14.14% of the runtime (inclusive) and most of that time was spent in std::vector's insert implementation.
After this change: AudioInterp::Linear has 0.36% of the runtime (inclusive)
2017-09-25 18:31:37 -05:00

77 lines
2.8 KiB
C++

// Copyright 2016 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include "audio_core/interpolate.h"
#include "common/assert.h"
#include "common/math_util.h"
namespace AudioInterp {
// Calculations are done in fixed point with 24 fractional bits.
// (This is not verified. This was chosen for minimal error.)
constexpr u64 scale_factor = 1 << 24;
constexpr u64 scale_mask = scale_factor - 1;
/// Here we step over the input in steps of rate, until we consume all of the input.
/// Three adjacent samples are passed to fn each step.
template <typename Function>
static void StepOverSamples(State& state, StereoBuffer16& input, float rate,
DSP::HLE::StereoFrame16& output, size_t& outputi, Function fn) {
ASSERT(rate > 0);
if (input.empty())
return;
input.insert(input.begin(), {state.xn2, state.xn1});
const u64 step_size = static_cast<u64>(rate * scale_factor);
u64 fposition = state.fposition;
size_t inputi = 0;
while (outputi < output.size()) {
inputi = static_cast<size_t>(fposition / scale_factor);
if (inputi + 2 >= input.size()) {
inputi = input.size() - 2;
break;
}
u64 fraction = fposition & scale_mask;
output[outputi++] = fn(fraction, input[inputi], input[inputi + 1], input[inputi + 2]);
fposition += step_size;
}
state.xn2 = input[inputi];
state.xn1 = input[inputi + 1];
state.fposition = fposition - inputi * scale_factor;
input.erase(input.begin(), std::next(input.begin(), inputi + 2));
}
void None(State& state, StereoBuffer16& input, float rate, DSP::HLE::StereoFrame16& output,
size_t& outputi) {
StepOverSamples(
state, input, rate, output, outputi,
[](u64 fraction, const auto& x0, const auto& x1, const auto& x2) { return x0; });
}
void Linear(State& state, StereoBuffer16& input, float rate, DSP::HLE::StereoFrame16& output,
size_t& outputi) {
// Note on accuracy: Some values that this produces are +/- 1 from the actual firmware.
StepOverSamples(state, input, rate, output, outputi,
[](u64 fraction, const auto& x0, const auto& x1, const auto& x2) {
// This is a saturated subtraction. (Verified by black-box fuzzing.)
s64 delta0 = MathUtil::Clamp<s64>(x1[0] - x0[0], -32768, 32767);
s64 delta1 = MathUtil::Clamp<s64>(x1[1] - x0[1], -32768, 32767);
return std::array<s16, 2>{
static_cast<s16>(x0[0] + fraction * delta0 / scale_factor),
static_cast<s16>(x0[1] + fraction * delta1 / scale_factor),
};
});
}
} // namespace AudioInterp