Andy Hung | 546734b | 2014-04-01 18:31:42 -0700 | [diff] [blame] | 1 | /* |
| 2 | * Copyright (C) 2014 The Android Open Source Project |
| 3 | * |
| 4 | * Licensed under the Apache License, Version 2.0 (the "License"); |
| 5 | * you may not use this file except in compliance with the License. |
| 6 | * You may obtain a copy of the License at |
| 7 | * |
| 8 | * http://www.apache.org/licenses/LICENSE-2.0 |
| 9 | * |
| 10 | * Unless required by applicable law or agreed to in writing, software |
| 11 | * distributed under the License is distributed on an "AS IS" BASIS, |
| 12 | * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
| 13 | * See the License for the specific language governing permissions and |
| 14 | * limitations under the License. |
| 15 | */ |
| 16 | |
| 17 | //#define LOG_NDEBUG 0 |
| 18 | #define LOG_TAG "audioflinger_resampler_tests" |
| 19 | |
| 20 | #include <unistd.h> |
| 21 | #include <stdio.h> |
| 22 | #include <stdlib.h> |
| 23 | #include <fcntl.h> |
| 24 | #include <string.h> |
| 25 | #include <sys/mman.h> |
| 26 | #include <sys/stat.h> |
| 27 | #include <errno.h> |
| 28 | #include <time.h> |
| 29 | #include <math.h> |
| 30 | #include <vector> |
| 31 | #include <utility> |
| 32 | #include <cutils/log.h> |
| 33 | #include <gtest/gtest.h> |
| 34 | #include <media/AudioBufferProvider.h> |
| 35 | #include "AudioResampler.h" |
| 36 | |
| 37 | #define ARRAY_SIZE(a) (sizeof(a) / sizeof((a)[0])) |
| 38 | |
| 39 | template<typename T, typename U> |
| 40 | struct is_same |
| 41 | { |
| 42 | static const bool value = false; |
| 43 | }; |
| 44 | |
| 45 | template<typename T> |
| 46 | struct is_same<T, T> // partial specialization |
| 47 | { |
| 48 | static const bool value = true; |
| 49 | }; |
| 50 | |
| 51 | template<typename T> |
| 52 | static inline T convertValue(double val) |
| 53 | { |
| 54 | if (is_same<T, int16_t>::value) { |
| 55 | return floor(val * 32767.0 + 0.5); |
| 56 | } else if (is_same<T, int32_t>::value) { |
| 57 | return floor(val * (1UL<<31) + 0.5); |
| 58 | } |
| 59 | return val; // assume float or double |
| 60 | } |
| 61 | |
| 62 | /* Creates a type-independent audio buffer provider from |
| 63 | * a buffer base address, size, framesize, and input increment array. |
| 64 | * |
| 65 | * No allocation or deallocation of the provided buffer is done. |
| 66 | */ |
| 67 | class TestProvider : public android::AudioBufferProvider { |
| 68 | public: |
| 69 | TestProvider(const void* addr, size_t frames, size_t frameSize, |
| 70 | const std::vector<size_t>& inputIncr) |
| 71 | : mAddr(addr), |
| 72 | mNumFrames(frames), |
| 73 | mFrameSize(frameSize), |
| 74 | mNextFrame(0), mUnrel(0), mInputIncr(inputIncr), mNextIdx(0) |
| 75 | { |
| 76 | } |
| 77 | |
| 78 | virtual android::status_t getNextBuffer(Buffer* buffer, int64_t pts __unused = kInvalidPTS ) |
| 79 | { |
| 80 | size_t requestedFrames = buffer->frameCount; |
| 81 | if (requestedFrames > mNumFrames - mNextFrame) { |
| 82 | buffer->frameCount = mNumFrames - mNextFrame; |
| 83 | } |
| 84 | if (!mInputIncr.empty()) { |
| 85 | size_t provided = mInputIncr[mNextIdx++]; |
| 86 | ALOGV("getNextBuffer() mValue[%d]=%u not %u", |
| 87 | mNextIdx-1, provided, buffer->frameCount); |
| 88 | if (provided < buffer->frameCount) { |
| 89 | buffer->frameCount = provided; |
| 90 | } |
| 91 | if (mNextIdx >= mInputIncr.size()) { |
| 92 | mNextIdx = 0; |
| 93 | } |
| 94 | } |
| 95 | ALOGV("getNextBuffer() requested %u frames out of %u frames available" |
| 96 | " and returned %u frames\n", |
| 97 | requestedFrames, mNumFrames - mNextFrame, buffer->frameCount); |
| 98 | mUnrel = buffer->frameCount; |
| 99 | if (buffer->frameCount > 0) { |
| 100 | buffer->raw = (char *)mAddr + mFrameSize * mNextFrame; |
| 101 | return android::NO_ERROR; |
| 102 | } else { |
| 103 | buffer->raw = NULL; |
| 104 | return android::NOT_ENOUGH_DATA; |
| 105 | } |
| 106 | } |
| 107 | |
| 108 | virtual void releaseBuffer(Buffer* buffer) |
| 109 | { |
| 110 | if (buffer->frameCount > mUnrel) { |
| 111 | ALOGE("releaseBuffer() released %u frames but only %u available " |
| 112 | "to release\n", buffer->frameCount, mUnrel); |
| 113 | mNextFrame += mUnrel; |
| 114 | mUnrel = 0; |
| 115 | } else { |
| 116 | |
| 117 | ALOGV("releaseBuffer() released %u frames out of %u frames available " |
| 118 | "to release\n", buffer->frameCount, mUnrel); |
| 119 | mNextFrame += buffer->frameCount; |
| 120 | mUnrel -= buffer->frameCount; |
| 121 | } |
| 122 | buffer->frameCount = 0; |
| 123 | buffer->raw = NULL; |
| 124 | } |
| 125 | |
| 126 | void reset() |
| 127 | { |
| 128 | mNextFrame = 0; |
| 129 | } |
| 130 | |
| 131 | size_t getNumFrames() |
| 132 | { |
| 133 | return mNumFrames; |
| 134 | } |
| 135 | |
| 136 | void setIncr(const std::vector<size_t> inputIncr) |
| 137 | { |
| 138 | mNextIdx = 0; |
| 139 | mInputIncr = inputIncr; |
| 140 | } |
| 141 | |
| 142 | protected: |
| 143 | const void* mAddr; // base address |
| 144 | size_t mNumFrames; // total frames |
| 145 | int mFrameSize; // frame size (# channels * bytes per sample) |
| 146 | size_t mNextFrame; // index of next frame to provide |
| 147 | size_t mUnrel; // number of frames not yet released |
| 148 | std::vector<size_t> mInputIncr; // number of frames provided per call |
| 149 | size_t mNextIdx; // index of next entry in mInputIncr to use |
| 150 | }; |
| 151 | |
| 152 | /* Creates a buffer filled with a sine wave. |
| 153 | * |
| 154 | * Returns a pair consisting of the sine signal buffer and the number of frames. |
| 155 | * The caller must delete[] the buffer when no longer needed (no shared_ptr<>). |
| 156 | */ |
| 157 | template<typename T> |
| 158 | static std::pair<T*, size_t> createSine(size_t channels, |
| 159 | double freq, double samplingRate, double time) |
| 160 | { |
| 161 | double tscale = 1. / samplingRate; |
| 162 | size_t frames = static_cast<size_t>(samplingRate * time); |
| 163 | T* buffer = new T[frames * channels]; |
| 164 | for (size_t i = 0; i < frames; ++i) { |
| 165 | double t = i * tscale; |
| 166 | double y = sin(2. * M_PI * freq * t); |
| 167 | T yt = convertValue<T>(y); |
| 168 | |
| 169 | for (size_t j = 0; j < channels; ++j) { |
| 170 | buffer[i*channels + j] = yt / (j + 1); |
| 171 | } |
| 172 | } |
| 173 | return std::make_pair(buffer, frames); |
| 174 | } |
| 175 | |
| 176 | /* Creates a buffer filled with a chirp signal (a sine wave sweep). |
| 177 | * |
| 178 | * Returns a pair consisting of the chirp signal buffer and the number of frames. |
| 179 | * The caller must delete[] the buffer when no longer needed (no shared_ptr<>). |
| 180 | * |
| 181 | * When creating the Chirp, note that the frequency is the true sinusoidal |
| 182 | * frequency not the sampling rate. |
| 183 | * |
| 184 | * http://en.wikipedia.org/wiki/Chirp |
| 185 | */ |
| 186 | template<typename T> |
| 187 | static std::pair<T*, size_t> createChirp(size_t channels, |
| 188 | double minfreq, double maxfreq, double samplingRate, double time) |
| 189 | { |
| 190 | double tscale = 1. / samplingRate; |
| 191 | size_t frames = static_cast<size_t>(samplingRate * time); |
| 192 | T *buffer = new T[frames * channels]; |
| 193 | // note the chirp constant k has a divide-by-two. |
| 194 | double k = (maxfreq - minfreq) / (2. * time); |
| 195 | for (size_t i = 0; i < frames; ++i) { |
| 196 | double t = i * tscale; |
| 197 | double y = sin(2. * M_PI * (k * t + minfreq) * t); |
| 198 | T yt = convertValue<T>(y); |
| 199 | |
| 200 | for (size_t j = 0; j < channels; ++j) { |
| 201 | buffer[i*channels + j] = yt / (j + 1); |
| 202 | } |
| 203 | } |
| 204 | return std::make_pair(buffer, frames); |
| 205 | } |
| 206 | |
| 207 | /* This derived class creates a buffer provider of datatype T, |
| 208 | * consisting of an input signal, e.g. from createChirp(). |
| 209 | * The number of frames can be obtained from the base class |
| 210 | * TestProvider::getNumFrames(). |
| 211 | */ |
| 212 | template <typename T> |
| 213 | class SignalProvider : public TestProvider { |
| 214 | public: |
| 215 | SignalProvider(const std::pair<T*, size_t>& bufferInfo, size_t channels, |
| 216 | const std::vector<size_t>& values) |
| 217 | : TestProvider(bufferInfo.first, bufferInfo.second, channels * sizeof(T), values), |
| 218 | mManagedPtr(bufferInfo.first) |
| 219 | { |
| 220 | } |
| 221 | |
| 222 | virtual ~SignalProvider() |
| 223 | { |
| 224 | delete[] mManagedPtr; |
| 225 | } |
| 226 | |
| 227 | protected: |
| 228 | T* mManagedPtr; |
| 229 | }; |
| 230 | |
| 231 | void resample(void *output, size_t outputFrames, const std::vector<size_t> &outputIncr, |
| 232 | android::AudioBufferProvider *provider, android::AudioResampler *resampler) |
| 233 | { |
| 234 | for (size_t i = 0, j = 0; i < outputFrames; ) { |
| 235 | size_t thisFrames = outputIncr[j++]; |
| 236 | if (j >= outputIncr.size()) { |
| 237 | j = 0; |
| 238 | } |
| 239 | if (thisFrames == 0 || thisFrames > outputFrames - i) { |
| 240 | thisFrames = outputFrames - i; |
| 241 | } |
| 242 | resampler->resample((int32_t*) output + 2*i, thisFrames, provider); |
| 243 | i += thisFrames; |
| 244 | } |
| 245 | } |
| 246 | |
| 247 | void buffercmp(const void *reference, const void *test, |
| 248 | size_t outputFrameSize, size_t outputFrames) |
| 249 | { |
| 250 | for (size_t i = 0; i < outputFrames; ++i) { |
| 251 | int check = memcmp((const char*)reference + i * outputFrameSize, |
| 252 | (const char*)test + i * outputFrameSize, outputFrameSize); |
| 253 | if (check) { |
| 254 | ALOGE("Failure at frame %d", i); |
| 255 | ASSERT_EQ(check, 0); /* fails */ |
| 256 | } |
| 257 | } |
| 258 | } |
| 259 | |
| 260 | void testBufferIncrement(size_t channels, unsigned inputFreq, unsigned outputFreq, |
| 261 | enum android::AudioResampler::src_quality quality) |
| 262 | { |
| 263 | // create the provider |
| 264 | std::vector<size_t> inputIncr; |
| 265 | SignalProvider<int16_t> provider(createChirp<int16_t>(channels, |
| 266 | 0., outputFreq/2., outputFreq, outputFreq/2000.), |
| 267 | channels, inputIncr); |
| 268 | |
| 269 | // calculate the output size |
| 270 | size_t outputFrames = ((int64_t) provider.getNumFrames() * outputFreq) / inputFreq; |
| 271 | size_t outputFrameSize = 2 * sizeof(int32_t); |
| 272 | size_t outputSize = outputFrameSize * outputFrames; |
| 273 | outputSize &= ~7; |
| 274 | |
| 275 | // create the resampler |
| 276 | const int volumePrecision = 12; /* typical unity gain */ |
| 277 | android::AudioResampler* resampler; |
| 278 | |
| 279 | resampler = android::AudioResampler::create(16, channels, outputFreq, quality); |
| 280 | resampler->setSampleRate(inputFreq); |
| 281 | resampler->setVolume(1 << volumePrecision, 1 << volumePrecision); |
| 282 | |
| 283 | // set up the reference run |
| 284 | std::vector<size_t> refIncr; |
| 285 | refIncr.push_back(outputFrames); |
| 286 | void* reference = malloc(outputSize); |
| 287 | resample(reference, outputFrames, refIncr, &provider, resampler); |
| 288 | |
| 289 | provider.reset(); |
| 290 | |
| 291 | #if 0 |
| 292 | /* this test will fail - API interface issue: reset() does not clear internal buffers */ |
| 293 | resampler->reset(); |
| 294 | #else |
| 295 | delete resampler; |
| 296 | resampler = android::AudioResampler::create(16, channels, outputFreq, quality); |
| 297 | resampler->setSampleRate(inputFreq); |
| 298 | resampler->setVolume(1 << volumePrecision, 1 << volumePrecision); |
| 299 | #endif |
| 300 | |
| 301 | // set up the test run |
| 302 | std::vector<size_t> outIncr; |
| 303 | outIncr.push_back(1); |
| 304 | outIncr.push_back(2); |
| 305 | outIncr.push_back(3); |
| 306 | void* test = malloc(outputSize); |
| 307 | resample(test, outputFrames, outIncr, &provider, resampler); |
| 308 | |
| 309 | // check |
| 310 | buffercmp(reference, test, outputFrameSize, outputFrames); |
| 311 | |
| 312 | free(reference); |
| 313 | free(test); |
| 314 | delete resampler; |
| 315 | } |
| 316 | |
| 317 | template <typename T> |
| 318 | inline double sqr(T v) |
| 319 | { |
| 320 | double dv = static_cast<double>(v); |
| 321 | return dv * dv; |
| 322 | } |
| 323 | |
| 324 | template <typename T> |
| 325 | double signalEnergy(T *start, T *end, unsigned stride) |
| 326 | { |
| 327 | double accum = 0; |
| 328 | |
| 329 | for (T *p = start; p < end; p += stride) { |
| 330 | accum += sqr(*p); |
| 331 | } |
| 332 | unsigned count = (end - start + stride - 1) / stride; |
| 333 | return accum / count; |
| 334 | } |
| 335 | |
| 336 | void testStopbandDownconversion(size_t channels, |
| 337 | unsigned inputFreq, unsigned outputFreq, |
| 338 | unsigned passband, unsigned stopband, |
| 339 | enum android::AudioResampler::src_quality quality) |
| 340 | { |
| 341 | // create the provider |
| 342 | std::vector<size_t> inputIncr; |
| 343 | SignalProvider<int16_t> provider(createChirp<int16_t>(channels, |
| 344 | 0., inputFreq/2., inputFreq, inputFreq/2000.), |
| 345 | channels, inputIncr); |
| 346 | |
| 347 | // calculate the output size |
| 348 | size_t outputFrames = ((int64_t) provider.getNumFrames() * outputFreq) / inputFreq; |
| 349 | size_t outputFrameSize = 2 * sizeof(int32_t); |
| 350 | size_t outputSize = outputFrameSize * outputFrames; |
| 351 | outputSize &= ~7; |
| 352 | |
| 353 | // create the resampler |
| 354 | const int volumePrecision = 12; /* typical unity gain */ |
| 355 | android::AudioResampler* resampler; |
| 356 | |
| 357 | resampler = android::AudioResampler::create(16, channels, outputFreq, quality); |
| 358 | resampler->setSampleRate(inputFreq); |
| 359 | resampler->setVolume(1 << volumePrecision, 1 << volumePrecision); |
| 360 | |
| 361 | // set up the reference run |
| 362 | std::vector<size_t> refIncr; |
| 363 | refIncr.push_back(outputFrames); |
| 364 | void* reference = malloc(outputSize); |
| 365 | resample(reference, outputFrames, refIncr, &provider, resampler); |
| 366 | |
| 367 | int32_t *out = reinterpret_cast<int32_t *>(reference); |
| 368 | |
| 369 | // check signal energy in passband |
| 370 | const unsigned passbandFrame = passband * outputFreq / 1000.; |
| 371 | const unsigned stopbandFrame = stopband * outputFreq / 1000.; |
| 372 | |
| 373 | // check each channel separately |
| 374 | for (size_t i = 0; i < channels; ++i) { |
| 375 | double passbandEnergy = signalEnergy(out, out + passbandFrame * channels, channels); |
| 376 | double stopbandEnergy = signalEnergy(out + stopbandFrame * channels, |
| 377 | out + outputFrames * channels, channels); |
| 378 | double dbAtten = -10. * log10(stopbandEnergy / passbandEnergy); |
| 379 | ASSERT_GT(dbAtten, 60.); |
| 380 | |
| 381 | #if 0 |
| 382 | // internal verification |
| 383 | printf("if:%d of:%d pbf:%d sbf:%d sbe: %f pbe: %f db: %.2f\n", |
| 384 | provider.getNumFrames(), outputFrames, |
| 385 | passbandFrame, stopbandFrame, stopbandEnergy, passbandEnergy, dbAtten); |
| 386 | for (size_t i = 0; i < 10; ++i) { |
| 387 | printf("%d\n", out[i+passbandFrame*channels]); |
| 388 | } |
| 389 | for (size_t i = 0; i < 10; ++i) { |
| 390 | printf("%d\n", out[i+stopbandFrame*channels]); |
| 391 | } |
| 392 | #endif |
| 393 | } |
| 394 | |
| 395 | free(reference); |
| 396 | delete resampler; |
| 397 | } |
| 398 | |
| 399 | /* Buffer increment test |
| 400 | * |
| 401 | * We compare a reference output, where we consume and process the entire |
| 402 | * buffer at a time, and a test output, where we provide small chunks of input |
| 403 | * data and process small chunks of output (which may not be equivalent in size). |
| 404 | * |
| 405 | * Two subtests - fixed phase (3:2 down) and interpolated phase (147:320 up) |
| 406 | */ |
| 407 | TEST(audioflinger_resampler, bufferincrement_fixedphase) { |
| 408 | // all of these work |
| 409 | static const enum android::AudioResampler::src_quality kQualityArray[] = { |
| 410 | android::AudioResampler::LOW_QUALITY, |
| 411 | android::AudioResampler::MED_QUALITY, |
| 412 | android::AudioResampler::HIGH_QUALITY, |
| 413 | android::AudioResampler::VERY_HIGH_QUALITY, |
| 414 | android::AudioResampler::DYN_LOW_QUALITY, |
| 415 | android::AudioResampler::DYN_MED_QUALITY, |
| 416 | android::AudioResampler::DYN_HIGH_QUALITY, |
| 417 | }; |
| 418 | |
| 419 | for (size_t i = 0; i < ARRAY_SIZE(kQualityArray); ++i) { |
| 420 | testBufferIncrement(2, 48000, 32000, kQualityArray[i]); |
| 421 | } |
| 422 | } |
| 423 | |
| 424 | TEST(audioflinger_resampler, bufferincrement_interpolatedphase) { |
| 425 | // all of these work except low quality |
| 426 | static const enum android::AudioResampler::src_quality kQualityArray[] = { |
| 427 | // android::AudioResampler::LOW_QUALITY, |
| 428 | android::AudioResampler::MED_QUALITY, |
| 429 | android::AudioResampler::HIGH_QUALITY, |
| 430 | android::AudioResampler::VERY_HIGH_QUALITY, |
| 431 | android::AudioResampler::DYN_LOW_QUALITY, |
| 432 | android::AudioResampler::DYN_MED_QUALITY, |
| 433 | android::AudioResampler::DYN_HIGH_QUALITY, |
| 434 | }; |
| 435 | |
| 436 | for (size_t i = 0; i < ARRAY_SIZE(kQualityArray); ++i) { |
| 437 | testBufferIncrement(2, 22050, 48000, kQualityArray[i]); |
| 438 | } |
| 439 | } |
| 440 | |
| 441 | /* Simple aliasing test |
| 442 | * |
| 443 | * This checks stopband response of the chirp signal to make sure frequencies |
| 444 | * are properly suppressed. It uses downsampling because the stopband can be |
| 445 | * clearly isolated by input frequencies exceeding the output sample rate (nyquist). |
| 446 | */ |
| 447 | TEST(audioflinger_resampler, stopbandresponse) { |
| 448 | // not all of these may work (old resamplers fail on downsampling) |
| 449 | static const enum android::AudioResampler::src_quality kQualityArray[] = { |
| 450 | //android::AudioResampler::LOW_QUALITY, |
| 451 | //android::AudioResampler::MED_QUALITY, |
| 452 | //android::AudioResampler::HIGH_QUALITY, |
| 453 | //android::AudioResampler::VERY_HIGH_QUALITY, |
| 454 | android::AudioResampler::DYN_LOW_QUALITY, |
| 455 | android::AudioResampler::DYN_MED_QUALITY, |
| 456 | android::AudioResampler::DYN_HIGH_QUALITY, |
| 457 | }; |
| 458 | |
| 459 | // in this test we assume a maximum transition band between 12kHz and 20kHz. |
| 460 | // there must be at least 60dB relative attenuation between stopband and passband. |
| 461 | for (size_t i = 0; i < ARRAY_SIZE(kQualityArray); ++i) { |
| 462 | testStopbandDownconversion(2, 48000, 32000, 12000, 20000, kQualityArray[i]); |
| 463 | } |
| 464 | |
| 465 | // in this test we assume a maximum transition band between 7kHz and 15kHz. |
| 466 | // there must be at least 60dB relative attenuation between stopband and passband. |
| 467 | // (the weird ratio triggers interpolative resampling) |
| 468 | for (size_t i = 0; i < ARRAY_SIZE(kQualityArray); ++i) { |
| 469 | testStopbandDownconversion(2, 48000, 22101, 7000, 15000, kQualityArray[i]); |
| 470 | } |
| 471 | } |