mirror of
https://github.com/yuzu-mirror/yuzu
synced 2024-12-21 15:43:05 +00:00
d7459354f5
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)
127 lines
4.2 KiB
C++
127 lines
4.2 KiB
C++
// Copyright 2016 Citra Emulator Project
|
|
// Licensed under GPLv2 or any later version
|
|
// Refer to the license.txt file included.
|
|
|
|
#include <array>
|
|
#include <cstddef>
|
|
#include <cstring>
|
|
#include <vector>
|
|
#include "audio_core/codec.h"
|
|
#include "common/assert.h"
|
|
#include "common/common_types.h"
|
|
#include "common/math_util.h"
|
|
|
|
namespace Codec {
|
|
|
|
StereoBuffer16 DecodeADPCM(const u8* const data, const size_t sample_count,
|
|
const std::array<s16, 16>& adpcm_coeff, ADPCMState& state) {
|
|
// GC-ADPCM with scale factor and variable coefficients.
|
|
// Frames are 8 bytes long containing 14 samples each.
|
|
// Samples are 4 bits (one nibble) long.
|
|
|
|
constexpr size_t FRAME_LEN = 8;
|
|
constexpr size_t SAMPLES_PER_FRAME = 14;
|
|
constexpr std::array<int, 16> SIGNED_NIBBLES = {
|
|
{0, 1, 2, 3, 4, 5, 6, 7, -8, -7, -6, -5, -4, -3, -2, -1}};
|
|
|
|
const size_t ret_size =
|
|
sample_count % 2 == 0 ? sample_count : sample_count + 1; // Ensure multiple of two.
|
|
StereoBuffer16 ret(ret_size);
|
|
|
|
int yn1 = state.yn1, yn2 = state.yn2;
|
|
|
|
const size_t NUM_FRAMES =
|
|
(sample_count + (SAMPLES_PER_FRAME - 1)) / SAMPLES_PER_FRAME; // Round up.
|
|
for (size_t framei = 0; framei < NUM_FRAMES; framei++) {
|
|
const int frame_header = data[framei * FRAME_LEN];
|
|
const int scale = 1 << (frame_header & 0xF);
|
|
const int idx = (frame_header >> 4) & 0x7;
|
|
|
|
// Coefficients are fixed point with 11 bits fractional part.
|
|
const int coef1 = adpcm_coeff[idx * 2 + 0];
|
|
const int coef2 = adpcm_coeff[idx * 2 + 1];
|
|
|
|
// Decodes an audio sample. One nibble produces one sample.
|
|
const auto decode_sample = [&](const int nibble) -> s16 {
|
|
const int xn = nibble * scale;
|
|
// We first transform everything into 11 bit fixed point, perform the second order
|
|
// digital filter, then transform back.
|
|
// 0x400 == 0.5 in 11 bit fixed point.
|
|
// Filter: y[n] = x[n] + 0.5 + c1 * y[n-1] + c2 * y[n-2]
|
|
int val = ((xn << 11) + 0x400 + coef1 * yn1 + coef2 * yn2) >> 11;
|
|
// Clamp to output range.
|
|
val = MathUtil::Clamp(val, -32768, 32767);
|
|
// Advance output feedback.
|
|
yn2 = yn1;
|
|
yn1 = val;
|
|
return (s16)val;
|
|
};
|
|
|
|
size_t outputi = framei * SAMPLES_PER_FRAME;
|
|
size_t datai = framei * FRAME_LEN + 1;
|
|
for (size_t i = 0; i < SAMPLES_PER_FRAME && outputi < sample_count; i += 2) {
|
|
const s16 sample1 = decode_sample(SIGNED_NIBBLES[data[datai] >> 4]);
|
|
ret[outputi].fill(sample1);
|
|
outputi++;
|
|
|
|
const s16 sample2 = decode_sample(SIGNED_NIBBLES[data[datai] & 0xF]);
|
|
ret[outputi].fill(sample2);
|
|
outputi++;
|
|
|
|
datai++;
|
|
}
|
|
}
|
|
|
|
state.yn1 = yn1;
|
|
state.yn2 = yn2;
|
|
|
|
return ret;
|
|
}
|
|
|
|
static s16 SignExtendS8(u8 x) {
|
|
// The data is actually signed PCM8.
|
|
// We sign extend this to signed PCM16.
|
|
return static_cast<s16>(static_cast<s8>(x));
|
|
}
|
|
|
|
StereoBuffer16 DecodePCM8(const unsigned num_channels, const u8* const data,
|
|
const size_t sample_count) {
|
|
ASSERT(num_channels == 1 || num_channels == 2);
|
|
|
|
StereoBuffer16 ret(sample_count);
|
|
|
|
if (num_channels == 1) {
|
|
for (size_t i = 0; i < sample_count; i++) {
|
|
ret[i].fill(SignExtendS8(data[i]));
|
|
}
|
|
} else {
|
|
for (size_t i = 0; i < sample_count; i++) {
|
|
ret[i][0] = SignExtendS8(data[i * 2 + 0]);
|
|
ret[i][1] = SignExtendS8(data[i * 2 + 1]);
|
|
}
|
|
}
|
|
|
|
return ret;
|
|
}
|
|
|
|
StereoBuffer16 DecodePCM16(const unsigned num_channels, const u8* const data,
|
|
const size_t sample_count) {
|
|
ASSERT(num_channels == 1 || num_channels == 2);
|
|
|
|
StereoBuffer16 ret(sample_count);
|
|
|
|
if (num_channels == 1) {
|
|
for (size_t i = 0; i < sample_count; i++) {
|
|
s16 sample;
|
|
std::memcpy(&sample, data + i * sizeof(s16), sizeof(s16));
|
|
ret[i].fill(sample);
|
|
}
|
|
} else {
|
|
for (size_t i = 0; i < sample_count; ++i) {
|
|
std::memcpy(&ret[i], data + i * sizeof(s16) * 2, 2 * sizeof(s16));
|
|
}
|
|
}
|
|
|
|
return ret;
|
|
}
|
|
};
|