update sdl Merge commit '644725478f4de0f074a6834e8423ac36dce3974f'

This commit is contained in:
2023-09-23 18:53:11 +02:00
172 changed files with 7495 additions and 4062 deletions

View File

@ -176,13 +176,13 @@ static int audio_initOpenCloseQuitAudio(void *arg)
case 0:
/* Set standard desired spec */
desired.freq = 22050;
desired.format = SDL_AUDIO_S16SYS;
desired.format = SDL_AUDIO_S16;
desired.channels = 2;
case 1:
/* Set custom desired spec */
desired.freq = 48000;
desired.format = SDL_AUDIO_F32SYS;
desired.format = SDL_AUDIO_F32;
desired.channels = 2;
break;
}
@ -261,14 +261,14 @@ static int audio_pauseUnpauseAudio(void *arg)
case 0:
/* Set standard desired spec */
desired.freq = 22050;
desired.format = SDL_AUDIO_S16SYS;
desired.format = SDL_AUDIO_S16;
desired.channels = 2;
break;
case 1:
/* Set custom desired spec */
desired.freq = 48000;
desired.format = SDL_AUDIO_F32SYS;
desired.format = SDL_AUDIO_F32;
desired.channels = 2;
break;
}
@ -441,18 +441,34 @@ static int audio_printCurrentAudioDriver(void *arg)
}
/* Definition of all formats, channels, and frequencies used to test audio conversions */
static SDL_AudioFormat g_audioFormats[] = { SDL_AUDIO_S8, SDL_AUDIO_U8, SDL_AUDIO_S16LSB, SDL_AUDIO_S16MSB, SDL_AUDIO_S16SYS, SDL_AUDIO_S16,
SDL_AUDIO_S32LSB, SDL_AUDIO_S32MSB, SDL_AUDIO_S32SYS, SDL_AUDIO_S32,
SDL_AUDIO_F32LSB, SDL_AUDIO_F32MSB, SDL_AUDIO_F32SYS, SDL_AUDIO_F32 };
static const char *g_audioFormatsVerbose[] = { "SDL_AUDIO_S8", "SDL_AUDIO_U8", "SDL_AUDIO_S16LSB", "SDL_AUDIO_S16MSB", "SDL_AUDIO_S16SYS", "SDL_AUDIO_S16",
"SDL_AUDIO_S32LSB", "SDL_AUDIO_S32MSB", "SDL_AUDIO_S32SYS", "SDL_AUDIO_S32",
"SDL_AUDIO_F32LSB", "SDL_AUDIO_F32MSB", "SDL_AUDIO_F32SYS", "SDL_AUDIO_F32" };
static SDL_AudioFormat g_audioFormats[] = {
SDL_AUDIO_S8, SDL_AUDIO_U8,
SDL_AUDIO_S16LE, SDL_AUDIO_S16BE,
SDL_AUDIO_S32LE, SDL_AUDIO_S32BE,
SDL_AUDIO_F32LE, SDL_AUDIO_F32BE
};
static const char *g_audioFormatsVerbose[] = {
"SDL_AUDIO_S8", "SDL_AUDIO_U8",
"SDL_AUDIO_S16LE", "SDL_AUDIO_S16BE",
"SDL_AUDIO_S32LE", "SDL_AUDIO_S32BE",
"SDL_AUDIO_F32LE", "SDL_AUDIO_F32BE"
};
static const int g_numAudioFormats = SDL_arraysize(g_audioFormats);
static Uint8 g_audioChannels[] = { 1, 2, 4, 6 };
static const int g_numAudioChannels = SDL_arraysize(g_audioChannels);
static int g_audioFrequencies[] = { 11025, 22050, 44100, 48000 };
static const int g_numAudioFrequencies = SDL_arraysize(g_audioFrequencies);
/* Verify the audio formats are laid out as expected */
SDL_COMPILE_TIME_ASSERT(SDL_AUDIO_U8_FORMAT, SDL_AUDIO_U8 == SDL_AUDIO_BITSIZE(8));
SDL_COMPILE_TIME_ASSERT(SDL_AUDIO_S8_FORMAT, SDL_AUDIO_S8 == (SDL_AUDIO_BITSIZE(8) | SDL_AUDIO_MASK_SIGNED));
SDL_COMPILE_TIME_ASSERT(SDL_AUDIO_S16LE_FORMAT, SDL_AUDIO_S16LE == (SDL_AUDIO_BITSIZE(16) | SDL_AUDIO_MASK_SIGNED));
SDL_COMPILE_TIME_ASSERT(SDL_AUDIO_S16BE_FORMAT, SDL_AUDIO_S16BE == (SDL_AUDIO_S16LE | SDL_AUDIO_MASK_BIG_ENDIAN));
SDL_COMPILE_TIME_ASSERT(SDL_AUDIO_S32LE_FORMAT, SDL_AUDIO_S32LE == (SDL_AUDIO_BITSIZE(32) | SDL_AUDIO_MASK_SIGNED));
SDL_COMPILE_TIME_ASSERT(SDL_AUDIO_S32BE_FORMAT, SDL_AUDIO_S32BE == (SDL_AUDIO_S32LE | SDL_AUDIO_MASK_BIG_ENDIAN));
SDL_COMPILE_TIME_ASSERT(SDL_AUDIO_F32LE_FORMAT, SDL_AUDIO_F32LE == (SDL_AUDIO_BITSIZE(32) | SDL_AUDIO_MASK_FLOAT | SDL_AUDIO_MASK_SIGNED));
SDL_COMPILE_TIME_ASSERT(SDL_AUDIO_F32BE_FORMAT, SDL_AUDIO_F32BE == (SDL_AUDIO_F32LE | SDL_AUDIO_MASK_BIG_ENDIAN));
/**
* \brief Builds various audio conversion structures
*
@ -466,7 +482,7 @@ static int audio_buildAudioStream(void *arg)
int i, ii, j, jj, k, kk;
/* No conversion needed */
spec1.format = SDL_AUDIO_S16LSB;
spec1.format = SDL_AUDIO_S16LE;
spec1.channels = 2;
spec1.freq = 22050;
stream = SDL_CreateAudioStream(&spec1, &spec1);
@ -478,7 +494,7 @@ static int audio_buildAudioStream(void *arg)
spec1.format = SDL_AUDIO_S8;
spec1.channels = 1;
spec1.freq = 22050;
spec2.format = SDL_AUDIO_S16LSB;
spec2.format = SDL_AUDIO_S16LE;
spec2.channels = 2;
spec2.freq = 44100;
stream = SDL_CreateAudioStream(&spec1, &spec2);
@ -533,7 +549,7 @@ static int audio_buildAudioStreamNegative(void *arg)
spec1.format = SDL_AUDIO_S8;
spec1.channels = 1;
spec1.freq = 22050;
spec2.format = SDL_AUDIO_S16LSB;
spec2.format = SDL_AUDIO_S16LE;
spec2.channels = 2;
spec2.freq = 44100;
@ -546,7 +562,7 @@ static int audio_buildAudioStreamNegative(void *arg)
spec1.format = SDL_AUDIO_S8;
spec1.channels = 1;
spec1.freq = 22050;
spec2.format = SDL_AUDIO_S16LSB;
spec2.format = SDL_AUDIO_S16LE;
spec2.channels = 2;
spec2.freq = 44100;
@ -692,54 +708,65 @@ static int audio_convertAudio(void *arg)
} else {
Uint8 *dst_buf = NULL, *src_buf = NULL;
int dst_len = 0, src_len = 0, real_dst_len = 0;
int l = 64;
int src_samplesize, dst_samplesize;
int l = 64, m;
int src_framesize, dst_framesize;
int src_silence, dst_silence;
src_samplesize = (SDL_AUDIO_BITSIZE(spec1.format) / 8) * spec1.channels;
dst_samplesize = (SDL_AUDIO_BITSIZE(spec2.format) / 8) * spec2.channels;
src_framesize = SDL_AUDIO_FRAMESIZE(spec1);
dst_framesize = SDL_AUDIO_FRAMESIZE(spec2);
/* Create some random data to convert */
src_len = l * src_samplesize;
src_len = l * src_framesize;
SDLTest_Log("Creating dummy sample buffer of %i length (%i bytes)", l, src_len);
src_buf = (Uint8 *)SDL_malloc(src_len);
SDLTest_AssertCheck(dst_buf != NULL, "Check src data buffer to convert is not NULL");
SDLTest_AssertCheck(src_buf != NULL, "Check src data buffer to convert is not NULL");
if (src_buf == NULL) {
return TEST_ABORTED;
}
src_len = src_len & ~(src_samplesize - 1);
dst_len = dst_samplesize * (src_len / src_samplesize);
if (spec1.freq < spec2.freq) {
const double mult = ((double)spec2.freq) / ((double)spec1.freq);
dst_len *= (int) SDL_ceil(mult);
}
src_silence = SDL_GetSilenceValueForFormat(spec1.format);
SDL_memset(src_buf, src_silence, src_len);
dst_len = dst_len & ~(dst_samplesize - 1);
dst_buf = (Uint8 *)SDL_calloc(1, dst_len);
dst_len = ((int)((((Sint64)l * spec2.freq) - 1) / spec1.freq) + 1) * dst_framesize;
dst_buf = (Uint8 *)SDL_malloc(dst_len);
SDLTest_AssertCheck(dst_buf != NULL, "Check dst data buffer to convert is not NULL");
if (dst_buf == NULL) {
return TEST_ABORTED;
}
real_dst_len = SDL_GetAudioStreamAvailable(stream);
SDLTest_AssertCheck(0 == real_dst_len, "Verify available (pre-put); expected: %i; got: %i", 0, real_dst_len);
/* Run the audio converter */
if (SDL_PutAudioStreamData(stream, src_buf, src_len) < 0 ||
SDL_FlushAudioStream(stream) < 0) {
return TEST_ABORTED;
}
real_dst_len = SDL_GetAudioStreamAvailable(stream);
SDLTest_AssertCheck(dst_len == real_dst_len, "Verify available (post-put); expected: %i; got: %i", dst_len, real_dst_len);
real_dst_len = SDL_GetAudioStreamData(stream, dst_buf, dst_len);
SDLTest_AssertCheck(real_dst_len > 0, "Verify result value; expected: > 0; got: %i", real_dst_len);
if (real_dst_len < 0) {
SDLTest_AssertCheck(dst_len == real_dst_len, "Verify result value; expected: %i; got: %i", dst_len, real_dst_len);
if (dst_len != real_dst_len) {
return TEST_ABORTED;
}
real_dst_len = SDL_GetAudioStreamAvailable(stream);
SDLTest_AssertCheck(0 == real_dst_len, "Verify available (post-get); expected: %i; got: %i", 0, real_dst_len);
dst_silence = SDL_GetSilenceValueForFormat(spec2.format);
for (m = 0; m < dst_len; ++m) {
if (dst_buf[m] != dst_silence) {
SDLTest_LogError("Output buffer is not silent");
return TEST_ABORTED;
}
}
SDL_DestroyAudioStream(stream);
/* Free converted buffer */
SDL_free(src_buf);
SDL_free(dst_buf);
}
}
}
@ -794,24 +821,30 @@ static int audio_resampleLoss(void *arg)
double signal_to_noise;
double max_error;
} test_specs[] = {
{ 50, 440, 0, 44100, 48000, 60, 0.0025 },
{ 50, 5000, SDL_PI_D / 2, 20000, 10000, 65, 0.0010 },
{ 50, 440, 0, 44100, 48000, 80, 0.0009 },
{ 50, 5000, SDL_PI_D / 2, 20000, 10000, 999, 0.0001 },
{ 50, 440, 0, 22050, 96000, 79, 0.0120 },
{ 50, 440, 0, 96000, 22050, 80, 0.0002 },
{ 0 }
};
int spec_idx = 0;
int min_channels = 1;
int max_channels = 1 /*8*/;
int num_channels = min_channels;
for (spec_idx = 0; test_specs[spec_idx].time > 0; ++spec_idx) {
for (spec_idx = 0; test_specs[spec_idx].time > 0;) {
const struct test_spec_t *spec = &test_specs[spec_idx];
const int frames_in = spec->time * spec->rate_in;
const int frames_target = spec->time * spec->rate_out;
const int len_in = frames_in * (int)sizeof(float);
const int len_target = frames_target * (int)sizeof(float);
const int len_in = (frames_in * num_channels) * (int)sizeof(float);
const int len_target = (frames_target * num_channels) * (int)sizeof(float);
SDL_AudioSpec tmpspec1, tmpspec2;
Uint64 tick_beg = 0;
Uint64 tick_end = 0;
int i = 0;
int j = 0;
int ret = 0;
SDL_AudioStream *stream = NULL;
float *buf_in = NULL;
@ -826,10 +859,10 @@ static int audio_resampleLoss(void *arg)
spec->time, spec->freq, spec->phase, spec->rate_in, spec->rate_out);
tmpspec1.format = SDL_AUDIO_F32;
tmpspec1.channels = 1;
tmpspec1.channels = num_channels;
tmpspec1.freq = spec->rate_in;
tmpspec2.format = SDL_AUDIO_F32;
tmpspec2.channels = 1;
tmpspec2.channels = num_channels;
tmpspec2.freq = spec->rate_out;
stream = SDL_CreateAudioStream(&tmpspec1, &tmpspec2);
SDLTest_AssertPass("Call to SDL_CreateAudioStream(SDL_AUDIO_F32, 1, %i, SDL_AUDIO_F32, 1, %i)", spec->rate_in, spec->rate_out);
@ -846,7 +879,10 @@ static int audio_resampleLoss(void *arg)
}
for (i = 0; i < frames_in; ++i) {
*(buf_in + i) = (float)sine_wave_sample(i, spec->rate_in, spec->freq, spec->phase);
float f = (float)sine_wave_sample(i, spec->rate_in, spec->freq, spec->phase);
for (j = 0; j < num_channels; ++j) {
*(buf_in + (i * num_channels) + j) = f;
}
}
tick_beg = SDL_GetPerformanceCounter();
@ -877,9 +913,7 @@ static int audio_resampleLoss(void *arg)
len_out = SDL_GetAudioStreamData(stream, buf_out, len_target);
SDLTest_AssertPass("Call to SDL_GetAudioStreamData(stream, buf_out, %i)", len_target);
/** !!! FIXME: SDL_AudioStream does not return output of the same length as
** !!! FIXME: the input even if SDL_FlushAudioStream is called. */
SDLTest_AssertCheck(len_out <= len_target, "Expected output length to be no larger than %i, got %i.",
SDLTest_AssertCheck(len_out == len_target, "Expected output length to be no larger than %i, got %i.",
len_target, len_out);
SDL_DestroyAudioStream(stream);
if (len_out > len_target) {
@ -890,13 +924,15 @@ static int audio_resampleLoss(void *arg)
tick_end = SDL_GetPerformanceCounter();
SDLTest_Log("Resampling used %f seconds.", ((double)(tick_end - tick_beg)) / SDL_GetPerformanceFrequency());
for (i = 0; i < len_out / (int)sizeof(float); ++i) {
const float output = *(buf_out + i);
for (i = 0; i < frames_target; ++i) {
const double target = sine_wave_sample(i, spec->rate_out, spec->freq, spec->phase);
const double error = SDL_fabs(target - output);
max_error = SDL_max(max_error, error);
sum_squared_error += error * error;
sum_squared_value += target * target;
for (j = 0; j < num_channels; ++j) {
const float output = *(buf_out + (i * num_channels) + j);
const double error = SDL_fabs(target - output);
max_error = SDL_max(max_error, error);
sum_squared_error += error * error;
sum_squared_value += target * target;
}
}
SDL_free(buf_out);
signal_to_noise = 10 * SDL_log10(sum_squared_value / sum_squared_error); /* decibel */
@ -909,10 +945,303 @@ static int audio_resampleLoss(void *arg)
signal_to_noise, spec->signal_to_noise);
SDLTest_AssertCheck(max_error <= spec->max_error, "Maximum conversion error %f should be no more than %f.",
max_error, spec->max_error);
if (++num_channels > max_channels) {
num_channels = min_channels;
++spec_idx;
}
}
return TEST_COMPLETED;
}
/**
* \brief Check accuracy converting between audio formats.
*
* \sa SDL_ConvertAudioSamples
*/
static int audio_convertAccuracy(void *arg)
{
static SDL_AudioFormat formats[] = { SDL_AUDIO_S8, SDL_AUDIO_U8, SDL_AUDIO_S16, SDL_AUDIO_S32 };
static const char* format_names[] = { "S8", "U8", "S16", "S32" };
int src_num = 65537 + 2048 + 48 + 256 + 100000;
int src_len = src_num * sizeof(float);
float* src_data = SDL_malloc(src_len);
int i, j;
SDLTest_AssertCheck(src_data != NULL, "Expected source buffer to be created.");
if (src_data == NULL) {
return TEST_ABORTED;
}
j = 0;
/* Generate a uniform range of floats between [-1.0, 1.0] */
for (i = 0; i < 65537; ++i) {
src_data[j++] = ((float)i - 32768.0f) / 32768.0f;
}
/* Generate floats close to 1.0 */
const float max_val = 16777216.0f;
for (i = 0; i < 1024; ++i) {
float f = (max_val + (float)(512 - i)) / max_val;
src_data[j++] = f;
src_data[j++] = -f;
}
for (i = 0; i < 24; ++i) {
float f = (max_val + (float)(3u << i)) / max_val;
src_data[j++] = f;
src_data[j++] = -f;
}
/* Generate floats far outside the [-1.0, 1.0] range */
for (i = 0; i < 128; ++i) {
float f = 2.0f + (float) i;
src_data[j++] = f;
src_data[j++] = -f;
}
/* Fill the rest with random floats between [-1.0, 1.0] */
for (i = 0; i < 100000; ++i) {
src_data[j++] = SDLTest_RandomSint32() / 2147483648.0f;
}
/* Shuffle the data for good measure */
for (i = src_num - 1; i > 0; --i) {
float f = src_data[i];
j = SDLTest_RandomIntegerInRange(0, i);
src_data[i] = src_data[j];
src_data[j] = f;
}
for (i = 0; i < SDL_arraysize(formats); ++i) {
SDL_AudioSpec src_spec, tmp_spec;
Uint64 convert_begin, convert_end;
Uint8 *tmp_data, *dst_data;
int tmp_len, dst_len;
int ret;
SDL_AudioFormat format = formats[i];
const char* format_name = format_names[i];
/* Formats with > 23 bits can represent every value exactly */
float min_delta = 1.0f;
float max_delta = -1.0f;
/* Subtract 1 bit to account for sign */
int bits = SDL_AUDIO_BITSIZE(format) - 1;
float target_max_delta = (bits > 23) ? 0.0f : (1.0f / (float)(1 << bits));
float target_min_delta = -target_max_delta;
src_spec.format = SDL_AUDIO_F32;
src_spec.channels = 1;
src_spec.freq = 44100;
tmp_spec.format = format;
tmp_spec.channels = 1;
tmp_spec.freq = 44100;
convert_begin = SDL_GetPerformanceCounter();
tmp_data = NULL;
tmp_len = 0;
ret = SDL_ConvertAudioSamples(&src_spec, (const Uint8*) src_data, src_len, &tmp_spec, &tmp_data, &tmp_len);
SDLTest_AssertCheck(ret == 0, "Expected SDL_ConvertAudioSamples(F32->%s) to succeed", format_name);
if (ret != 0) {
SDL_free(src_data);
return TEST_ABORTED;
}
dst_data = NULL;
dst_len = 0;
ret = SDL_ConvertAudioSamples(&tmp_spec, tmp_data, tmp_len, &src_spec, &dst_data, &dst_len);
SDLTest_AssertCheck(ret == 0, "Expected SDL_ConvertAudioSamples(%s->F32) to succeed", format_name);
if (ret != 0) {
SDL_free(tmp_data);
SDL_free(src_data);
return TEST_ABORTED;
}
convert_end = SDL_GetPerformanceCounter();
SDLTest_Log("Conversion via %s took %f seconds.", format_name, ((double)(convert_end - convert_begin)) / SDL_GetPerformanceFrequency());
SDL_free(tmp_data);
for (j = 0; j < src_num; ++j) {
float x = src_data[j];
float y = ((float*)dst_data)[j];
float d = SDL_clamp(x, -1.0f, 1.0f) - y;
min_delta = SDL_min(min_delta, d);
max_delta = SDL_max(max_delta, d);
}
SDLTest_AssertCheck(min_delta >= target_min_delta, "%s has min delta of %+f, should be >= %+f", format_name, min_delta, target_min_delta);
SDLTest_AssertCheck(max_delta <= target_max_delta, "%s has max delta of %+f, should be <= %+f", format_name, max_delta, target_max_delta);
SDL_free(dst_data);
}
SDL_free(src_data);
return TEST_COMPLETED;
}
/**
* \brief Check accuracy when switching between formats
*
* \sa SDL_SetAudioStreamFormat
*/
static int audio_formatChange(void *arg)
{
int i;
SDL_AudioSpec spec1, spec2, spec3;
int frames_1, frames_2, frames_3;
int length_1, length_2, length_3;
int retval = 0;
int status = TEST_ABORTED;
float* buffer_1 = NULL;
float* buffer_2 = NULL;
float* buffer_3 = NULL;
SDL_AudioStream* stream = NULL;
double max_error = 0;
double sum_squared_error = 0;
double sum_squared_value = 0;
double signal_to_noise = 0;
double target_max_error = 0.02;
double target_signal_to_noise = 75.0;
int sine_freq = 500;
spec1.format = SDL_AUDIO_F32;
spec1.channels = 1;
spec1.freq = 20000;
spec2.format = SDL_AUDIO_F32;
spec2.channels = 1;
spec2.freq = 40000;
spec3.format = SDL_AUDIO_F32;
spec3.channels = 1;
spec3.freq = 80000;
frames_1 = spec1.freq;
frames_2 = spec2.freq;
frames_3 = spec3.freq * 2;
length_1 = (int)(frames_1 * sizeof(*buffer_1));
buffer_1 = (float*) SDL_malloc(length_1);
if (!SDLTest_AssertCheck(buffer_1 != NULL, "Expected buffer_1 to be created.")) {
goto cleanup;
}
length_2 = (int)(frames_2 * sizeof(*buffer_2));
buffer_2 = (float*) SDL_malloc(length_2);
if (!SDLTest_AssertCheck(buffer_2 != NULL, "Expected buffer_2 to be created.")) {
goto cleanup;
}
length_3 = (int)(frames_3 * sizeof(*buffer_3));
buffer_3 = (float*) SDL_malloc(length_3);
if (!SDLTest_AssertCheck(buffer_3 != NULL, "Expected buffer_3 to be created.")) {
goto cleanup;
}
for (i = 0; i < frames_1; ++i) {
buffer_1[i] = (float) sine_wave_sample(i, spec1.freq, sine_freq, 0.0f);
}
for (i = 0; i < frames_2; ++i) {
buffer_2[i] = (float) sine_wave_sample(i, spec2.freq, sine_freq, 0.0f);
}
stream = SDL_CreateAudioStream(NULL, NULL);
if (!SDLTest_AssertCheck(stream != NULL, "Expected SDL_CreateAudioStream to succeed")) {
goto cleanup;
}
retval = SDL_SetAudioStreamFormat(stream, &spec1, &spec3);
if (!SDLTest_AssertCheck(retval == 0, "Expected SDL_SetAudioStreamFormat(spec1, spec3) to succeed")) {
goto cleanup;
}
retval = SDL_GetAudioStreamAvailable(stream);
if (!SDLTest_AssertCheck(retval == 0, "Expected SDL_GetAudioStreamAvailable return 0")) {
goto cleanup;
}
retval = SDL_PutAudioStreamData(stream, buffer_1, length_1);
if (!SDLTest_AssertCheck(retval == 0, "Expected SDL_PutAudioStreamData(buffer_1) to succeed")) {
goto cleanup;
}
retval = SDL_FlushAudioStream(stream);
if (!SDLTest_AssertCheck(retval == 0, "Expected SDL_FlushAudioStream to succeed")) {
goto cleanup;
}
retval = SDL_SetAudioStreamFormat(stream, &spec2, &spec3);
if (!SDLTest_AssertCheck(retval == 0, "Expected SDL_SetAudioStreamFormat(spec2, spec3) to succeed")) {
goto cleanup;
}
retval = SDL_PutAudioStreamData(stream, buffer_2, length_2);
if (!SDLTest_AssertCheck(retval == 0, "Expected SDL_PutAudioStreamData(buffer_1) to succeed")) {
goto cleanup;
}
retval = SDL_FlushAudioStream(stream);
if (!SDLTest_AssertCheck(retval == 0, "Expected SDL_FlushAudioStream to succeed")) {
goto cleanup;
}
retval = SDL_GetAudioStreamAvailable(stream);
if (!SDLTest_AssertCheck(retval == length_3, "Expected SDL_GetAudioStreamAvailable to return %i, got %i", length_3, retval)) {
goto cleanup;
}
retval = SDL_GetAudioStreamData(stream, buffer_3, length_3);
if (!SDLTest_AssertCheck(retval == length_3, "Expected SDL_GetAudioStreamData to return %i, got %i", length_3, retval)) {
goto cleanup;
}
retval = SDL_GetAudioStreamAvailable(stream);
if (!SDLTest_AssertCheck(retval == 0, "Expected SDL_GetAudioStreamAvailable to return 0")) {
goto cleanup;
}
for (i = 0; i < frames_3; ++i) {
const float output = buffer_3[i];
const float target = (float) sine_wave_sample(i, spec3.freq, sine_freq, 0.0f);
const double error = SDL_fabs(target - output);
max_error = SDL_max(max_error, error);
sum_squared_error += error * error;
sum_squared_value += target * target;
}
signal_to_noise = 10 * SDL_log10(sum_squared_value / sum_squared_error); /* decibel */
SDLTest_AssertCheck(isfinite(sum_squared_value), "Sum of squared target should be finite.");
SDLTest_AssertCheck(isfinite(sum_squared_error), "Sum of squared error should be finite.");
/* Infinity is theoretically possible when there is very little to no noise */
SDLTest_AssertCheck(!isnan(signal_to_noise), "Signal-to-noise ratio should not be NaN.");
SDLTest_AssertCheck(isfinite(max_error), "Maximum conversion error should be finite.");
SDLTest_AssertCheck(signal_to_noise >= target_signal_to_noise, "Conversion signal-to-noise ratio %f dB should be no less than %f dB.",
signal_to_noise, target_signal_to_noise);
SDLTest_AssertCheck(max_error <= target_max_error, "Maximum conversion error %f should be no more than %f.",
max_error, target_max_error);
status = TEST_COMPLETED;
cleanup:
SDL_free(buffer_1);
SDL_free(buffer_2);
SDL_free(buffer_3);
SDL_DestroyAudioStream(stream);
return status;
}
/* ================= Test Case References ================== */
/* Audio test cases */
@ -952,12 +1281,8 @@ static const SDLTest_TestCaseReference audioTest9 = {
audio_lockUnlockOpenAudioDevice, "audio_lockUnlockOpenAudioDevice", "Locks and unlocks an open audio device.", TEST_ENABLED
};
/* TODO: enable test when SDL_ConvertAudio segfaults on cygwin have been fixed.
* TODO: re-check, since this was changer to AudioStream */
/* For debugging, test case can be run manually using --filter audio_convertAudio */
static const SDLTest_TestCaseReference audioTest10 = {
audio_convertAudio, "audio_convertAudio", "Convert audio using available formats.", TEST_DISABLED
audio_convertAudio, "audio_convertAudio", "Convert audio using available formats.", TEST_ENABLED
};
/* TODO: enable test when SDL_AudioDeviceConnected has been implemented. */
@ -986,11 +1311,20 @@ static const SDLTest_TestCaseReference audioTest16 = {
audio_resampleLoss, "audio_resampleLoss", "Check signal-to-noise ratio and maximum error of audio resampling.", TEST_ENABLED
};
static const SDLTest_TestCaseReference audioTest17 = {
audio_convertAccuracy, "audio_convertAccuracy", "Check accuracy converting between audio formats.", TEST_ENABLED
};
static const SDLTest_TestCaseReference audioTest18 = {
audio_formatChange, "audio_formatChange", "Check handling of format changes.", TEST_ENABLED
};
/* Sequence of Audio test cases */
static const SDLTest_TestCaseReference *audioTests[] = {
&audioTest1, &audioTest2, &audioTest3, &audioTest4, &audioTest5, &audioTest6,
&audioTest7, &audioTest8, &audioTest9, &audioTest10, &audioTest11,
&audioTest12, &audioTest13, &audioTest14, &audioTest15, &audioTest16, NULL
&audioTest12, &audioTest13, &audioTest14, &audioTest15, &audioTest16,
&audioTest17, &audioTest18, NULL
};
/* Audio test suite (global) */