mirror of
https://github.com/mackron/miniaudio.git
synced 2026-04-21 15:56:58 +02:00
Resampler: Optimization to f32, no LPF code path.
This commit is contained in:
David Reid
+202
-87
@@ -59325,6 +59325,109 @@ static MA_INLINE ma_result ma_linear_resampler_process_pcm_frames_s16_no_lpf(ma_
|
||||
return MA_SUCCESS;
|
||||
}
|
||||
|
||||
static MA_INLINE ma_result ma_linear_resampler_process_pcm_frames_f32_no_lpf(ma_linear_resampler* pResampler, const float* pFramesInF32, ma_uint64* pFrameCountIn, float* pFramesOutF32, ma_uint64* pFrameCountOut, float invSampleRateOut)
|
||||
{
|
||||
ma_uint64 frameCountIn;
|
||||
ma_uint64 frameCountOut;
|
||||
ma_uint64 framesProcessedIn;
|
||||
ma_uint64 framesProcessedOut;
|
||||
ma_uint32 c;
|
||||
|
||||
MA_ASSERT(pResampler != NULL);
|
||||
MA_ASSERT(pFramesInF32 != NULL);
|
||||
MA_ASSERT(pFrameCountIn != NULL);
|
||||
MA_ASSERT(pFramesOutF32 != NULL);
|
||||
MA_ASSERT(pFrameCountOut != NULL);
|
||||
|
||||
frameCountIn = *pFrameCountIn;
|
||||
frameCountOut = *pFrameCountOut;
|
||||
framesProcessedIn = 0;
|
||||
framesProcessedOut = 0;
|
||||
|
||||
/* If there's a cached frame we need to process it. */
|
||||
if (pResampler->inTimeInt == 0) {
|
||||
MA_ASSERT(pResampler->cachedFrameCount <= 1); /* There is at most one cached frame. */
|
||||
|
||||
while (pResampler->cachedFrameCount > 0 && frameCountIn > 0 && framesProcessedOut < frameCountOut) {
|
||||
float a = pResampler->inTimeFrac * invSampleRateOut;
|
||||
|
||||
for (c = 0; c < pResampler->channels; c += 1) {
|
||||
pFramesOutF32[c] = ma_mix_f32_fast(pResampler->x0.f32[c], pFramesInF32[c], a);
|
||||
}
|
||||
pFramesOutF32 += pResampler->channels;
|
||||
|
||||
framesProcessedOut += 1;
|
||||
|
||||
/* Advance time forward. */
|
||||
pResampler->inTimeInt += pResampler->inAdvanceInt;
|
||||
pResampler->inTimeFrac += pResampler->inAdvanceFrac;
|
||||
if (pResampler->inTimeFrac >= pResampler->sampleRateOut) {
|
||||
pResampler->inTimeFrac -= pResampler->sampleRateOut;
|
||||
pResampler->inTimeInt += 1;
|
||||
}
|
||||
|
||||
/* Subtract one from the time to account for the cached frame, but only if the entire frame was processed. */
|
||||
if (pResampler->inTimeInt > 0) {
|
||||
pResampler->inTimeInt -= 1;
|
||||
pResampler->cachedFrameCount = 0;
|
||||
}
|
||||
}
|
||||
} else {
|
||||
/* The rate must have changed between calls. Ignore the cached frame. */
|
||||
}
|
||||
|
||||
while (framesProcessedOut < frameCountOut && pResampler->inTimeInt < frameCountIn) {
|
||||
if (pResampler->inTimeInt + 1 < frameCountIn) {
|
||||
float a = pResampler->inTimeFrac * invSampleRateOut;
|
||||
|
||||
for (c = 0; c < pResampler->channels; c += 1) {
|
||||
pFramesOutF32[c] = ma_mix_f32_fast(pFramesInF32[(pResampler->inTimeInt * pResampler->channels) + c], pFramesInF32[((pResampler->inTimeInt + 1) * pResampler->channels) + c], a);
|
||||
}
|
||||
pFramesOutF32 += pResampler->channels;
|
||||
|
||||
framesProcessedOut += 1;
|
||||
|
||||
/* Advance time forward. */
|
||||
pResampler->inTimeInt += pResampler->inAdvanceInt;
|
||||
pResampler->inTimeFrac += pResampler->inAdvanceFrac;
|
||||
if (pResampler->inTimeFrac >= pResampler->sampleRateOut) {
|
||||
pResampler->inTimeFrac -= pResampler->sampleRateOut;
|
||||
pResampler->inTimeInt += 1;
|
||||
}
|
||||
} else {
|
||||
/*
|
||||
There is not enough input frames to interpolate. We'll need to stop here. But it's important that we cache
|
||||
the frame to ensure we make some forward progress.
|
||||
*/
|
||||
for (c = 0; c < pResampler->channels; c += 1) {
|
||||
pResampler->x0.f32[c] = pFramesInF32[(pResampler->inTimeInt * pResampler->channels) + c];
|
||||
}
|
||||
|
||||
pResampler->cachedFrameCount = 1;
|
||||
pResampler->inTimeInt += 1;
|
||||
|
||||
break;
|
||||
}
|
||||
}
|
||||
|
||||
/* The number of frames we processed is simply the difference between our current time and previous time, clamped. */
|
||||
framesProcessedIn = pResampler->inTimeInt;
|
||||
if (framesProcessedIn > frameCountIn) { /* Should never overshoot when upsampling. Downsampling could overshoot. */
|
||||
framesProcessedIn = frameCountIn;
|
||||
}
|
||||
|
||||
if (pResampler->inTimeInt >= framesProcessedIn) {
|
||||
pResampler->inTimeInt -= framesProcessedIn;
|
||||
} else {
|
||||
pResampler->inTimeInt = 0;
|
||||
}
|
||||
|
||||
*pFrameCountIn = framesProcessedIn;
|
||||
*pFrameCountOut = framesProcessedOut;
|
||||
|
||||
return MA_SUCCESS;
|
||||
}
|
||||
|
||||
static ma_result ma_linear_resampler_process_pcm_frames_s16_downsample(ma_linear_resampler* pResampler, const void* pFramesIn, ma_uint64* pFrameCountIn, void* pFramesOut, ma_uint64* pFrameCountOut)
|
||||
{
|
||||
const ma_int16* pFramesInS16;
|
||||
@@ -59528,60 +59631,66 @@ static ma_result ma_linear_resampler_process_pcm_frames_f32_downsample(ma_linear
|
||||
framesProcessedOut = 0;
|
||||
invSampleRateOut = 1.0f / pResampler->sampleRateOut;
|
||||
|
||||
while (framesProcessedOut < frameCountOut) {
|
||||
/* Before interpolating we need to load the buffers. When doing this we need to ensure we run every input sample through the filter. */
|
||||
while (pResampler->inTimeInt > 0 && frameCountIn > framesProcessedIn) {
|
||||
ma_uint32 iChannel;
|
||||
if (pResampler->lpfOrder == 0) {
|
||||
/* Fast path. No LPF needed. */
|
||||
return ma_linear_resampler_process_pcm_frames_f32_no_lpf(pResampler, pFramesInF32, pFrameCountIn, pFramesOutF32, pFrameCountOut, invSampleRateOut);
|
||||
} else {
|
||||
/* Slow path. Need LPF. */
|
||||
while (framesProcessedOut < frameCountOut) {
|
||||
/* Before interpolating we need to load the buffers. When doing this we need to ensure we run every input sample through the filter. */
|
||||
while (pResampler->inTimeInt > 0 && frameCountIn > framesProcessedIn) {
|
||||
ma_uint32 iChannel;
|
||||
|
||||
if (pFramesInF32 != NULL) {
|
||||
for (iChannel = 0; iChannel < pResampler->channels; iChannel += 1) {
|
||||
pResampler->x0.f32[iChannel] = pResampler->x1.f32[iChannel];
|
||||
pResampler->x1.f32[iChannel] = pFramesInF32[iChannel];
|
||||
if (pFramesInF32 != NULL) {
|
||||
for (iChannel = 0; iChannel < pResampler->channels; iChannel += 1) {
|
||||
pResampler->x0.f32[iChannel] = pResampler->x1.f32[iChannel];
|
||||
pResampler->x1.f32[iChannel] = pFramesInF32[iChannel];
|
||||
}
|
||||
pFramesInF32 += pResampler->channels;
|
||||
} else {
|
||||
for (iChannel = 0; iChannel < pResampler->channels; iChannel += 1) {
|
||||
pResampler->x0.f32[iChannel] = pResampler->x1.f32[iChannel];
|
||||
pResampler->x1.f32[iChannel] = 0;
|
||||
}
|
||||
}
|
||||
pFramesInF32 += pResampler->channels;
|
||||
} else {
|
||||
for (iChannel = 0; iChannel < pResampler->channels; iChannel += 1) {
|
||||
pResampler->x0.f32[iChannel] = pResampler->x1.f32[iChannel];
|
||||
pResampler->x1.f32[iChannel] = 0;
|
||||
|
||||
/* Filter. Do not apply filtering if sample rates are the same or else you'll get dangerous glitching. */
|
||||
if (pResampler->lpfOrder > 0 && (pResampler->inAdvanceInt != 1 || pResampler->inAdvanceFrac != 0)) {
|
||||
ma_lpf_process_pcm_frame_f32(&pResampler->lpf, pResampler->x1.f32, pResampler->x1.f32);
|
||||
}
|
||||
|
||||
framesProcessedIn += 1;
|
||||
pResampler->inTimeInt -= 1;
|
||||
}
|
||||
|
||||
/* Filter. Do not apply filtering if sample rates are the same or else you'll get dangerous glitching. */
|
||||
if (pResampler->lpfOrder > 0 && (pResampler->inAdvanceInt != 1 || pResampler->inAdvanceFrac != 0)) {
|
||||
ma_lpf_process_pcm_frame_f32(&pResampler->lpf, pResampler->x1.f32, pResampler->x1.f32);
|
||||
if (pResampler->inTimeInt > 0) {
|
||||
break; /* Ran out of input data. */
|
||||
}
|
||||
|
||||
framesProcessedIn += 1;
|
||||
pResampler->inTimeInt -= 1;
|
||||
/* Getting here means the frames have been loaded and filtered and we can generate the next output frame. */
|
||||
if (pFramesOutF32 != NULL) {
|
||||
MA_ASSERT(pResampler->inTimeInt == 0);
|
||||
ma_linear_resampler_interpolate_frame_f32(pResampler, invSampleRateOut, pFramesOutF32);
|
||||
|
||||
pFramesOutF32 += pResampler->channels;
|
||||
}
|
||||
|
||||
framesProcessedOut += 1;
|
||||
|
||||
/* Advance time forward. */
|
||||
pResampler->inTimeInt += pResampler->inAdvanceInt;
|
||||
pResampler->inTimeFrac += pResampler->inAdvanceFrac;
|
||||
if (pResampler->inTimeFrac >= pResampler->sampleRateOut) {
|
||||
pResampler->inTimeFrac -= pResampler->sampleRateOut;
|
||||
pResampler->inTimeInt += 1;
|
||||
}
|
||||
}
|
||||
|
||||
if (pResampler->inTimeInt > 0) {
|
||||
break; /* Ran out of input data. */
|
||||
}
|
||||
*pFrameCountIn = framesProcessedIn;
|
||||
*pFrameCountOut = framesProcessedOut;
|
||||
|
||||
/* Getting here means the frames have been loaded and filtered and we can generate the next output frame. */
|
||||
if (pFramesOutF32 != NULL) {
|
||||
MA_ASSERT(pResampler->inTimeInt == 0);
|
||||
ma_linear_resampler_interpolate_frame_f32(pResampler, invSampleRateOut, pFramesOutF32);
|
||||
|
||||
pFramesOutF32 += pResampler->channels;
|
||||
}
|
||||
|
||||
framesProcessedOut += 1;
|
||||
|
||||
/* Advance time forward. */
|
||||
pResampler->inTimeInt += pResampler->inAdvanceInt;
|
||||
pResampler->inTimeFrac += pResampler->inAdvanceFrac;
|
||||
if (pResampler->inTimeFrac >= pResampler->sampleRateOut) {
|
||||
pResampler->inTimeFrac -= pResampler->sampleRateOut;
|
||||
pResampler->inTimeInt += 1;
|
||||
}
|
||||
return MA_SUCCESS;
|
||||
}
|
||||
|
||||
*pFrameCountIn = framesProcessedIn;
|
||||
*pFrameCountOut = framesProcessedOut;
|
||||
|
||||
return MA_SUCCESS;
|
||||
}
|
||||
|
||||
static ma_result ma_linear_resampler_process_pcm_frames_f32_upsample(ma_linear_resampler* pResampler, const void* pFramesIn, ma_uint64* pFrameCountIn, void* pFramesOut, ma_uint64* pFrameCountOut)
|
||||
@@ -59606,60 +59715,66 @@ static ma_result ma_linear_resampler_process_pcm_frames_f32_upsample(ma_linear_r
|
||||
framesProcessedOut = 0;
|
||||
invSampleRateOut = 1.0f / pResampler->sampleRateOut;
|
||||
|
||||
while (framesProcessedOut < frameCountOut) {
|
||||
/* Before interpolating we need to load the buffers. */
|
||||
while (pResampler->inTimeInt > 0 && frameCountIn > framesProcessedIn) {
|
||||
ma_uint32 iChannel;
|
||||
if (pResampler->lpfOrder == 0) {
|
||||
/* Fast path. No LPF needed. */
|
||||
return ma_linear_resampler_process_pcm_frames_f32_no_lpf(pResampler, pFramesInF32, pFrameCountIn, pFramesOutF32, pFrameCountOut, invSampleRateOut);
|
||||
} else {
|
||||
/* Slow path. Need LPF. */
|
||||
while (framesProcessedOut < frameCountOut) {
|
||||
/* Before interpolating we need to load the buffers. */
|
||||
while (pResampler->inTimeInt > 0 && frameCountIn > framesProcessedIn) {
|
||||
ma_uint32 iChannel;
|
||||
|
||||
if (pFramesInF32 != NULL) {
|
||||
for (iChannel = 0; iChannel < pResampler->channels; iChannel += 1) {
|
||||
pResampler->x0.f32[iChannel] = pResampler->x1.f32[iChannel];
|
||||
pResampler->x1.f32[iChannel] = pFramesInF32[iChannel];
|
||||
}
|
||||
pFramesInF32 += pResampler->channels;
|
||||
} else {
|
||||
for (iChannel = 0; iChannel < pResampler->channels; iChannel += 1) {
|
||||
pResampler->x0.f32[iChannel] = pResampler->x1.f32[iChannel];
|
||||
pResampler->x1.f32[iChannel] = 0;
|
||||
if (pFramesInF32 != NULL) {
|
||||
for (iChannel = 0; iChannel < pResampler->channels; iChannel += 1) {
|
||||
pResampler->x0.f32[iChannel] = pResampler->x1.f32[iChannel];
|
||||
pResampler->x1.f32[iChannel] = pFramesInF32[iChannel];
|
||||
}
|
||||
pFramesInF32 += pResampler->channels;
|
||||
} else {
|
||||
for (iChannel = 0; iChannel < pResampler->channels; iChannel += 1) {
|
||||
pResampler->x0.f32[iChannel] = pResampler->x1.f32[iChannel];
|
||||
pResampler->x1.f32[iChannel] = 0;
|
||||
}
|
||||
}
|
||||
|
||||
framesProcessedIn += 1;
|
||||
pResampler->inTimeInt -= 1;
|
||||
}
|
||||
|
||||
framesProcessedIn += 1;
|
||||
pResampler->inTimeInt -= 1;
|
||||
}
|
||||
|
||||
if (pResampler->inTimeInt > 0) {
|
||||
break; /* Ran out of input data. */
|
||||
}
|
||||
|
||||
/* Getting here means the frames have been loaded and we can generate the next output frame. */
|
||||
if (pFramesOutF32 != NULL) {
|
||||
MA_ASSERT(pResampler->inTimeInt == 0);
|
||||
ma_linear_resampler_interpolate_frame_f32(pResampler, invSampleRateOut, pFramesOutF32);
|
||||
|
||||
/* Filter. Do not apply filtering if sample rates are the same or else you'll get dangerous glitching. */
|
||||
if (pResampler->lpfOrder > 0 && (pResampler->inAdvanceInt != 1 || pResampler->inAdvanceFrac != 0)) {
|
||||
ma_lpf_process_pcm_frame_f32(&pResampler->lpf, pFramesOutF32, pFramesOutF32);
|
||||
if (pResampler->inTimeInt > 0) {
|
||||
break; /* Ran out of input data. */
|
||||
}
|
||||
|
||||
pFramesOutF32 += pResampler->channels;
|
||||
/* Getting here means the frames have been loaded and we can generate the next output frame. */
|
||||
if (pFramesOutF32 != NULL) {
|
||||
MA_ASSERT(pResampler->inTimeInt == 0);
|
||||
ma_linear_resampler_interpolate_frame_f32(pResampler, invSampleRateOut, pFramesOutF32);
|
||||
|
||||
/* Filter. Do not apply filtering if sample rates are the same or else you'll get dangerous glitching. */
|
||||
if (pResampler->lpfOrder > 0 && (pResampler->inAdvanceInt != 1 || pResampler->inAdvanceFrac != 0)) {
|
||||
ma_lpf_process_pcm_frame_f32(&pResampler->lpf, pFramesOutF32, pFramesOutF32);
|
||||
}
|
||||
|
||||
pFramesOutF32 += pResampler->channels;
|
||||
}
|
||||
|
||||
framesProcessedOut += 1;
|
||||
|
||||
/* Advance time forward. */
|
||||
pResampler->inTimeInt += pResampler->inAdvanceInt;
|
||||
pResampler->inTimeFrac += pResampler->inAdvanceFrac;
|
||||
if (pResampler->inTimeFrac >= pResampler->sampleRateOut) {
|
||||
pResampler->inTimeFrac -= pResampler->sampleRateOut;
|
||||
pResampler->inTimeInt += 1;
|
||||
}
|
||||
}
|
||||
|
||||
framesProcessedOut += 1;
|
||||
*pFrameCountIn = framesProcessedIn;
|
||||
*pFrameCountOut = framesProcessedOut;
|
||||
|
||||
/* Advance time forward. */
|
||||
pResampler->inTimeInt += pResampler->inAdvanceInt;
|
||||
pResampler->inTimeFrac += pResampler->inAdvanceFrac;
|
||||
if (pResampler->inTimeFrac >= pResampler->sampleRateOut) {
|
||||
pResampler->inTimeFrac -= pResampler->sampleRateOut;
|
||||
pResampler->inTimeInt += 1;
|
||||
}
|
||||
return MA_SUCCESS;
|
||||
}
|
||||
|
||||
*pFrameCountIn = framesProcessedIn;
|
||||
*pFrameCountOut = framesProcessedOut;
|
||||
|
||||
return MA_SUCCESS;
|
||||
}
|
||||
|
||||
static ma_result ma_linear_resampler_process_pcm_frames_f32(ma_linear_resampler* pResampler, const void* pFramesIn, ma_uint64* pFrameCountIn, void* pFramesOut, ma_uint64* pFrameCountOut)
|
||||
|
||||
Reference in New Issue
Block a user