mirror/miniaudio
1
0
Fork 0
mirror of https://github.com/mackron/miniaudio.git synced 2026-04-22 00:06:59 +02:00
Code Issues Packages Projects Releases Wiki Activity

Resampler: Optimization to the no-LPF path.

Browse Source 
This moves some checks outside the loop. A bit more code duplication,
but does improve speed.
This commit is contained in:
David Reid
2026-02-14 13:00:52 +10:00
parent 0fe2f7effd
commit 0615ce28f1
1 changed files with 434 additions and 304 deletions
Download Patch File Download Diff File Expand all files Collapse all files
miniaudio.h
+152 -22
View File
@@ -59378,11 +59378,18 @@ static MA_INLINE ma_result ma_linear_resampler_process_pcm_frames_s16_no_lpf(ma_
/* Experimental loop unrolling to make it easier for SIMD-ification. */ /* Experimental loop unrolling to make it easier for SIMD-ification. */
#if 1 #if 1
{ {
if (channels == 1) {
while (framesProcessedOut + 4 <= frameCountOut) { while (framesProcessedOut + 4 <= frameCountOut) {
ma_uint32 inTimeIntTemp; ma_uint32 inTimeIntTemp;
ma_uint32 inTimeFracTemp; ma_uint32 inTimeFracTemp;
ma_uint32 inTimeInt[4]; ma_uint32 inTimeInt[4];
ma_uint32 inTimeFrac[4]; ma_uint32 inTimeFrac[4];
ma_int32 x[4];
ma_int32 y[4];
ma_int32 a[4];
ma_int32 d[4];
ma_int32 n[4];
ma_int32 r[4];
int i; int i;
inTimeIntTemp = pResampler->inTimeInt; inTimeIntTemp = pResampler->inTimeInt;
@@ -59409,16 +59416,6 @@ static MA_INLINE ma_result ma_linear_resampler_process_pcm_frames_s16_no_lpf(ma_
pResampler->inTimeInt = inTimeIntTemp; pResampler->inTimeInt = inTimeIntTemp;
pResampler->inTimeFrac = inTimeFracTemp; pResampler->inTimeFrac = inTimeFracTemp;
/* We should now be able to SIMD-ify the rest. For now I am trusting the compiler to vectorize this, but I'll experiment with some manual stuff later. */
{
if (channels == 1) {
ma_int32 x[4];
ma_int32 y[4];
ma_int32 a[4];
ma_int32 d[4];
ma_int32 n[4];
ma_int32 r[4];
x[0] = pFramesInS16[inTimeInt[0] + 0]; x[0] = pFramesInS16[inTimeInt[0] + 0];
x[1] = pFramesInS16[inTimeInt[1] + 0]; x[1] = pFramesInS16[inTimeInt[1] + 0];
x[2] = pFramesInS16[inTimeInt[2] + 0]; x[2] = pFramesInS16[inTimeInt[2] + 0];
@@ -59453,13 +59450,47 @@ static MA_INLINE ma_result ma_linear_resampler_process_pcm_frames_s16_no_lpf(ma_
pFramesOutS16[1] = r[1]; pFramesOutS16[1] = r[1];
pFramesOutS16[2] = r[2]; pFramesOutS16[2] = r[2];
pFramesOutS16[3] = r[3]; pFramesOutS16[3] = r[3];
pFramesOutS16 += 4;
framesProcessedOut += 4;
}
} else if (channels == 2) { } else if (channels == 2) {
while (framesProcessedOut + 4 <= frameCountOut) {
ma_uint32 inTimeIntTemp;
ma_uint32 inTimeFracTemp;
ma_uint32 inTimeInt[4];
ma_uint32 inTimeFrac[4];
ma_int32 x[8]; ma_int32 x[8];
ma_int32 y[8]; ma_int32 y[8];
ma_int32 a[8]; ma_int32 a[8];
ma_int32 d[8]; ma_int32 d[8];
ma_int32 n[8]; ma_int32 n[8];
ma_int32 r[8]; ma_int32 r[8];
int i;
inTimeIntTemp = pResampler->inTimeInt;
inTimeFracTemp = pResampler->inTimeFrac;
for (i = 0; i < 4; i += 1) {
inTimeInt[i] = inTimeIntTemp;
inTimeFrac[i] = inTimeFracTemp;
inTimeIntTemp += pResampler->inAdvanceInt;
inTimeFracTemp += pResampler->inAdvanceFrac;
if (inTimeFracTemp >= pResampler->sampleRateOut) {
inTimeFracTemp -= pResampler->sampleRateOut;
inTimeIntTemp += 1;
}
}
/* Check that we have one extra sample at the end for doing the interpolation. */
if (inTimeInt[3] + 1 >= frameCountIn) {
break; /* Not enough input frames. */
}
/* Advance the timer. */
pResampler->inTimeInt = inTimeIntTemp;
pResampler->inTimeFrac = inTimeFracTemp;
x[0] = pFramesInS16[((inTimeInt[0] + 0) * 2) + 0]; x[0] = pFramesInS16[((inTimeInt[0] + 0) * 2) + 0];
x[1] = pFramesInS16[((inTimeInt[0] + 0) * 2) + 1]; x[1] = pFramesInS16[((inTimeInt[0] + 0) * 2) + 1];
@@ -59523,7 +59554,42 @@ static MA_INLINE ma_result ma_linear_resampler_process_pcm_frames_s16_no_lpf(ma_
pFramesOutS16[5] = r[5]; pFramesOutS16[5] = r[5];
pFramesOutS16[6] = r[6]; pFramesOutS16[6] = r[6];
pFramesOutS16[7] = r[7]; pFramesOutS16[7] = r[7];
pFramesOutS16 += 8;
framesProcessedOut += 4;
}
} else { } else {
while (framesProcessedOut + 4 <= frameCountOut) {
ma_uint32 inTimeIntTemp;
ma_uint32 inTimeFracTemp;
ma_uint32 inTimeInt[4];
ma_uint32 inTimeFrac[4];
int i;
inTimeIntTemp = pResampler->inTimeInt;
inTimeFracTemp = pResampler->inTimeFrac;
for (i = 0; i < 4; i += 1) {
inTimeInt[i] = inTimeIntTemp;
inTimeFrac[i] = inTimeFracTemp;
inTimeIntTemp += pResampler->inAdvanceInt;
inTimeFracTemp += pResampler->inAdvanceFrac;
if (inTimeFracTemp >= pResampler->sampleRateOut) {
inTimeFracTemp -= pResampler->sampleRateOut;
inTimeIntTemp += 1;
}
}
/* Check that we have one extra sample at the end for doing the interpolation. */
if (inTimeInt[3] + 1 >= frameCountIn) {
break; /* Not enough input frames. */
}
/* Advance the timer. */
pResampler->inTimeInt = inTimeIntTemp;
pResampler->inTimeFrac = inTimeFracTemp;
for (c = 0; c < channels; c += 1) { for (c = 0; c < channels; c += 1) {
ma_int32 x[4]; ma_int32 x[4];
ma_int32 y[4]; ma_int32 y[4];
@@ -59567,7 +59633,6 @@ static MA_INLINE ma_result ma_linear_resampler_process_pcm_frames_s16_no_lpf(ma_
pFramesOutS16[(2 * channels) + c] = r[2]; pFramesOutS16[(2 * channels) + c] = r[2];
pFramesOutS16[(3 * channels) + c] = r[3]; pFramesOutS16[(3 * channels) + c] = r[3];
} }
}
pFramesOutS16 += 4 * channels; pFramesOutS16 += 4 * channels;
framesProcessedOut += 4; framesProcessedOut += 4;
@@ -59684,11 +59749,18 @@ static MA_INLINE ma_result ma_linear_resampler_process_pcm_frames_f32_no_lpf(ma_
/* Experimental loop unrolling to make it easier for SIMD-ification. */ /* Experimental loop unrolling to make it easier for SIMD-ification. */
#if 1 #if 1
{ {
if (channels == 1) {
while (framesProcessedOut + 4 <= frameCountOut) { while (framesProcessedOut + 4 <= frameCountOut) {
ma_uint32 inTimeIntTemp; ma_uint32 inTimeIntTemp;
ma_uint32 inTimeFracTemp; ma_uint32 inTimeFracTemp;
ma_uint32 inTimeInt[4]; ma_uint32 inTimeInt[4];
ma_uint32 inTimeFrac[4]; ma_uint32 inTimeFrac[4];
float x[4];
float y[4];
float a[4];
float d[4];
float n[4];
float r[4];
int i; int i;
inTimeIntTemp = pResampler->inTimeInt; inTimeIntTemp = pResampler->inTimeInt;
@@ -59715,16 +59787,6 @@ static MA_INLINE ma_result ma_linear_resampler_process_pcm_frames_f32_no_lpf(ma_
pResampler->inTimeInt = inTimeIntTemp; pResampler->inTimeInt = inTimeIntTemp;
pResampler->inTimeFrac = inTimeFracTemp; pResampler->inTimeFrac = inTimeFracTemp;
/* We should now be able to SIMD-ify the rest. For now I am trusting the compiler to vectorize this, but I'll experiment with some manual stuff later. */
{
if (channels == 1) {
float x[4];
float y[4];
float a[4];
float d[4];
float n[4];
float r[4];
x[0] = pFramesInF32[inTimeInt[0] + 0]; x[0] = pFramesInF32[inTimeInt[0] + 0];
x[1] = pFramesInF32[inTimeInt[1] + 0]; x[1] = pFramesInF32[inTimeInt[1] + 0];
x[2] = pFramesInF32[inTimeInt[2] + 0]; x[2] = pFramesInF32[inTimeInt[2] + 0];
@@ -59759,13 +59821,47 @@ static MA_INLINE ma_result ma_linear_resampler_process_pcm_frames_f32_no_lpf(ma_
pFramesOutF32[1] = r[1]; pFramesOutF32[1] = r[1];
pFramesOutF32[2] = r[2]; pFramesOutF32[2] = r[2];
pFramesOutF32[3] = r[3]; pFramesOutF32[3] = r[3];
pFramesOutF32 += 4;
framesProcessedOut += 4;
}
} else if (channels == 2) { } else if (channels == 2) {
while (framesProcessedOut + 4 <= frameCountOut) {
ma_uint32 inTimeIntTemp;
ma_uint32 inTimeFracTemp;
ma_uint32 inTimeInt[4];
ma_uint32 inTimeFrac[4];
float x[8]; float x[8];
float y[8]; float y[8];
float a[8]; float a[8];
float d[8]; float d[8];
float n[8]; float n[8];
float r[8]; float r[8];
int i;
inTimeIntTemp = pResampler->inTimeInt;
inTimeFracTemp = pResampler->inTimeFrac;
for (i = 0; i < 4; i += 1) {
inTimeInt[i] = inTimeIntTemp;
inTimeFrac[i] = inTimeFracTemp;
inTimeIntTemp += pResampler->inAdvanceInt;
inTimeFracTemp += pResampler->inAdvanceFrac;
if (inTimeFracTemp >= pResampler->sampleRateOut) {
inTimeFracTemp -= pResampler->sampleRateOut;
inTimeIntTemp += 1;
}
}
/* Check that we have one extra sample at the end for doing the interpolation. */
if (inTimeInt[3] + 1 >= frameCountIn) {
break; /* Not enough input frames. */
}
/* Advance the timer. */
pResampler->inTimeInt = inTimeIntTemp;
pResampler->inTimeFrac = inTimeFracTemp;
x[0] = pFramesInF32[((inTimeInt[0] + 0) * 2) + 0]; x[0] = pFramesInF32[((inTimeInt[0] + 0) * 2) + 0];
x[1] = pFramesInF32[((inTimeInt[0] + 0) * 2) + 1]; x[1] = pFramesInF32[((inTimeInt[0] + 0) * 2) + 1];
@@ -59829,7 +59925,42 @@ static MA_INLINE ma_result ma_linear_resampler_process_pcm_frames_f32_no_lpf(ma_
pFramesOutF32[5] = r[5]; pFramesOutF32[5] = r[5];
pFramesOutF32[6] = r[6]; pFramesOutF32[6] = r[6];
pFramesOutF32[7] = r[7]; pFramesOutF32[7] = r[7];
pFramesOutF32 += 8;
framesProcessedOut += 4;
}
} else { } else {
while (framesProcessedOut + 4 <= frameCountOut) {
ma_uint32 inTimeIntTemp;
ma_uint32 inTimeFracTemp;
ma_uint32 inTimeInt[4];
ma_uint32 inTimeFrac[4];
int i;
inTimeIntTemp = pResampler->inTimeInt;
inTimeFracTemp = pResampler->inTimeFrac;
for (i = 0; i < 4; i += 1) {
inTimeInt[i] = inTimeIntTemp;
inTimeFrac[i] = inTimeFracTemp;
inTimeIntTemp += pResampler->inAdvanceInt;
inTimeFracTemp += pResampler->inAdvanceFrac;
if (inTimeFracTemp >= pResampler->sampleRateOut) {
inTimeFracTemp -= pResampler->sampleRateOut;
inTimeIntTemp += 1;
}
}
/* Check that we have one extra sample at the end for doing the interpolation. */
if (inTimeInt[3] + 1 >= frameCountIn) {
break; /* Not enough input frames. */
}
/* Advance the timer. */
pResampler->inTimeInt = inTimeIntTemp;
pResampler->inTimeFrac = inTimeFracTemp;
for (c = 0; c < channels; c += 1) { for (c = 0; c < channels; c += 1) {
float x[4]; float x[4];
float y[4]; float y[4];
@@ -59873,7 +60004,6 @@ static MA_INLINE ma_result ma_linear_resampler_process_pcm_frames_f32_no_lpf(ma_
pFramesOutF32[(2 * channels) + c] = r[2]; pFramesOutF32[(2 * channels) + c] = r[2];
pFramesOutF32[(3 * channels) + c] = r[3]; pFramesOutF32[(3 * channels) + c] = r[3];
} }
}
pFramesOutF32 += 4 * channels; pFramesOutF32 += 4 * channels;
framesProcessedOut += 4; framesProcessedOut += 4;