improve SINC resampler performance
The improvement is about 60% by just tweaking a few
things to help the compiler generate better code.
It turns out that inlining too much stuff manually was hurting us.
Change-Id: I8068f0f75051f95ac600e50ce552572dd1e8c304
diff --git a/services/audioflinger/AudioResamplerSinc.cpp b/services/audioflinger/AudioResamplerSinc.cpp
index 5f25760..b478e8d 100644
--- a/services/audioflinger/AudioResamplerSinc.cpp
+++ b/services/audioflinger/AudioResamplerSinc.cpp
@@ -23,6 +23,7 @@
#include <cutils/properties.h>
#include <stdlib.h>
#include <utils/Log.h>
+#include <cutils/compiler.h>
namespace android {
// ----------------------------------------------------------------------------
@@ -305,38 +306,28 @@
// Always read-in the first samples from the input buffer
int16_t* head = impulse + c->halfNumCoefs*CHANNELS;
- head[0] = in[inputIndex*CHANNELS + 0];
- if (CHANNELS == 2)
- head[1] = in[inputIndex*CHANNELS + 1];
+ for (size_t i=0 ; i<CHANNELS ; i++) {
+ head[i] = in[inputIndex*CHANNELS + i];
+ }
// handle boundary case
int32_t l, r;
- while (outputIndex < outputSampleCount) {
- filterCoefficient<CHANNELS>(l, r, phaseFraction, impulse);
- out[outputIndex++] += 2 * mulRL(1, l, vRL);
- out[outputIndex++] += 2 * mulRL(0, r, vRL);
+ while (CC_LIKELY(outputIndex < outputSampleCount)) {
+ filterCoefficient<CHANNELS>(l, r, phaseFraction, impulse, vRL);
+ out[outputIndex++] += l;
+ out[outputIndex++] += r;
phaseFraction += phaseIncrement;
- const uint32_t phaseIndex = phaseFraction >> kNumPhaseBits;
- if (phaseIndex == 1) {
+ const size_t phaseIndex = phaseFraction >> kNumPhaseBits;
+ for (size_t i=0 ; i<phaseIndex ; i++) {
inputIndex++;
- if (inputIndex >= frameCount)
- break; // need a new buffer
- read<CHANNELS>(impulse, phaseFraction, in, inputIndex);
- } else if (phaseIndex == 2) { // maximum value
- inputIndex++;
- if (inputIndex >= frameCount)
- break; // 0 frame available, 2 frames needed
- // read first frame
- read<CHANNELS>(impulse, phaseFraction, in, inputIndex);
- inputIndex++;
- if (inputIndex >= frameCount)
- break; // 0 frame available, 1 frame needed
- // read second frame
+ if (inputIndex >= frameCount) {
+ goto done; // need a new buffer
+ }
read<CHANNELS>(impulse, phaseFraction, in, inputIndex);
}
}
-
+done:
// if done with buffer, save samples
if (inputIndex >= frameCount) {
inputIndex -= frameCount;
@@ -366,20 +357,20 @@
const uint32_t phaseIndex = phaseFraction >> kNumPhaseBits;
impulse += CHANNELS;
phaseFraction -= 1LU<<kNumPhaseBits;
- if (impulse >= mRingFull) {
+ if (CC_UNLIKELY(impulse >= mRingFull)) {
const size_t stateSize = (c->halfNumCoefs*2)*CHANNELS;
memcpy(mState, mState+stateSize, sizeof(int16_t)*stateSize);
impulse -= stateSize;
}
int16_t* head = impulse + c->halfNumCoefs*CHANNELS;
- head[0] = in[inputIndex*CHANNELS + 0];
- if (CHANNELS == 2)
- head[1] = in[inputIndex*CHANNELS + 1];
+ for (size_t i=0 ; i<CHANNELS ; i++) {
+ head[i] = in[inputIndex*CHANNELS + i];
+ }
}
template<int CHANNELS>
void AudioResamplerSinc::filterCoefficient(
- int32_t& l, int32_t& r, uint32_t phase, const int16_t *samples)
+ int32_t& l, int32_t& r, uint32_t phase, const int16_t *samples, uint32_t vRL)
{
const Constants *c = mConstants;
@@ -399,20 +390,15 @@
const int32_t* coefs = mFirCoefs;
const int16_t *sP = samples;
const int16_t *sN = samples+CHANNELS;
- for (unsigned int i=0 ; i < c->halfNumCoefs/4 ; i++) {
+ const size_t offset = 1 << c->coefsBits;
+ const size_t count = c->halfNumCoefs;
+ for (size_t i=0 ; i < count ; i++) {
interpolate<CHANNELS>(l, r, coefs+indexP, lerpP, sP);
interpolate<CHANNELS>(l, r, coefs+indexN, lerpN, sN);
- sP -= CHANNELS; sN += CHANNELS; coefs += 1 << c->coefsBits;
- interpolate<CHANNELS>(l, r, coefs+indexP, lerpP, sP);
- interpolate<CHANNELS>(l, r, coefs+indexN, lerpN, sN);
- sP -= CHANNELS; sN += CHANNELS; coefs += 1 << c->coefsBits;
- interpolate<CHANNELS>(l, r, coefs+indexP, lerpP, sP);
- interpolate<CHANNELS>(l, r, coefs+indexN, lerpN, sN);
- sP -= CHANNELS; sN += CHANNELS; coefs += 1 << c->coefsBits;
- interpolate<CHANNELS>(l, r, coefs+indexP, lerpP, sP);
- interpolate<CHANNELS>(l, r, coefs+indexN, lerpN, sN);
- sP -= CHANNELS; sN += CHANNELS; coefs += 1 << c->coefsBits;
+ sP -= CHANNELS; sN += CHANNELS; coefs += offset;
}
+ l = 2 * mulRL(1, l, vRL);
+ r = 2 * mulRL(0, r, vRL);
}
template<int CHANNELS>
diff --git a/services/audioflinger/AudioResamplerSinc.h b/services/audioflinger/AudioResamplerSinc.h
index 48bc747..3a6e356 100644
--- a/services/audioflinger/AudioResamplerSinc.h
+++ b/services/audioflinger/AudioResamplerSinc.h
@@ -50,7 +50,7 @@
template<int CHANNELS>
inline void filterCoefficient(
- int32_t& l, int32_t& r, uint32_t phase, const int16_t *samples);
+ int32_t& l, int32_t& r, uint32_t phase, const int16_t *samples, uint32_t vRL);
template<int CHANNELS>
inline void interpolate(