I'm trying to find a super fast way of getting the sign of each value in a vector. I was hoping to find a function in the accelerate framework to do this, but couldn't find one. Here's what it would do:
float *inputVector = .... // some audio vector
int length = ...// length of input vector.
float *outputVector = ....// result
for( int i = 0; i<length; i++ )
{
if( inputVector[i] >= 0 ) outputVector[i] = 1;
else outputVector[i] = -1;
}
Ok, I think I've found a way...
vvcopysignf() "Copies an array, setting the sign of each value based on a second array."
So, one method would be to make an array of 1s, then use this function to change the sign of the 1s based on an input array.
float *ones = ... // a vector filled with 1's
float *input = .... // an input vector
float *output = ... // an output vector
int bufferSize = ... // size of the vectors;
vvcopysignf(output, ones, input, &bufferSize);
//output now is an array of -1s and 1s based the sign of the input.
Related
I'm trying to write a function within a compute shader (HLSL) that accept an argument being an array on different size. The compiler always reject it.
Example (not working!):
void TestFunc(in uint SA[])
{
int K;
for (K = 0; SA[K] != 0; K++) {
// Some code using SA array
}
}
[numthreads(1, 1, 1)]
void CSMain(
uint S1[] = {1, 2, 3, 4 }; // Compiler happy and discover the array size
uint S2[] = {10, 20}; // Compiler happy and discover the array size
TestFunc(S1);
TestFunc(S2);
}
If I give an array size in TestFunc(), then the compiler is happy when calling TestFunc() passing that specific array size but refuse the call for another size.
You cannot have function parameters of indeterminate size.
You need to initialize an array of know length, and an int variable that holds the array length.
void TestFunc(in uint SA[4], in uint saCount)
{ int K;
for (K = 0; SA[K] != 0; K++)
{
// Some code using SA array, saCount is your array length;
}
}
[numthreads(1, 1, 1)]
void CSMain()
{
uint S1count = 4;
uint S1[] = {1, 2, 3, 4 };
uint S2count = 2;
uint S2[] = {10, 20,0,0};
TestFunc(S1, S1count);
TestFunc(S2, S2count);
}
In my example I have set your array max size as 4, but you can set it bigger if needed. You can also set multiple functions for different array lengths, of set up multiple passes if your data overflows your array max size.
Edit to answer comment
The issue is that array dimensions of function parameters must be explicit as the compiler error states. This cannot be avoided. What you can do however, is avoid passing the array at all. If you in-line your TestFunc in your CSMain, you avoid passing the array and your routine compiles and runs. I know it can make your code longer and harder to maintain, but it's the only way to do what you want with an array of unspecified length. The advantage is that this way you have access to array.Length that might make your code simpler.
I am re-writing the particle filter library of iOS in Swift from Objective C which is available on Bitbucket and I have a question regarding a syntax of Objective C which I cannot understand.
The code goes as follows:
- (void)setRssi:(NSInteger)rssi {
_rssi = rssi;
// Ignore zeros in average, StdDev -- we clear the value before setting it to
// prevent old values from hanging around if there's no reading
if (rssi == 0) {
self.meters = 0;
return;
}
self.meters = [self metersFromRssi:rssi];
NSInteger* pidx = self.rssiBuffer;
*(pidx+self.bufferIndex++) = rssi;
if (self.bufferIndex >= RSSIBUFFERSIZE) {
self.bufferIndex %= RSSIBUFFERSIZE;
self.bufferFull = YES;
}
if (self.bufferFull) {
// Only calculate trailing mean and Std Dev when we have enough data
double accumulator = 0;
for (NSInteger i = 0; i < RSSIBUFFERSIZE; i++) {
accumulator += *(pidx+i);
}
self.meanRssi = accumulator / RSSIBUFFERSIZE;
self.meanMeters = [self metersFromRssi:self.meanRssi];
accumulator = 0;
for (NSInteger i = 0; i < RSSIBUFFERSIZE; i++) {
NSInteger difference = *(pidx+i) - self.meanRssi;
accumulator += difference*difference;
}
self.stdDeviationRssi = sqrt( accumulator / RSSIBUFFERSIZE);
self.meanMetersVariance = ABS(
[self metersFromRssi:self.meanRssi]
- [self metersFromRssi:self.meanRssi+self.stdDeviationRssi]
);
}
}
The class continues with more code and functions which are not important and what I do not understand are these two lines
NSInteger* pidx = self.rssiBuffer;
*(pidx+self.bufferIndex++) = rssi;
Variable pidx is initialized to the size of a buffer which was previously defined and then in the next line the size of that buffer and buffer plus one is equal to the RSSI variable which is passed as a parameter in the function.
I assume that * has something to do with reference but I just can't figure out the purpose of this line. Variable pidx is used only in this function for calculating trailing mean and standard deviation.
Let explain those code:
NSInteger* pidx = self.rssiBuffer; means that you are getting pointer of the first value of the buffer.
*(pidx+self.bufferIndex++) = rssi; means that you are setting the value of the buffer at index 0+self.bufferIndex to rssiand then increase bufferIndex by 1. Thanks to #Jakub Vano point it out.
In C++, it will look like that
int self.rssiBuffer[1000]; // I assume we have buffer like that
self.rssiBuffer[self.bufferIndex++] = rssi
I am converting from the following format:
const int four_bytes_per_float = 4;
const int eight_bits_per_byte = 8;
_stereoGraphStreamFormat.mFormatID = kAudioFormatLinearPCM;
_stereoGraphStreamFormat.mFormatFlags = kAudioFormatFlagsNativeFloatPacked | kAudioFormatFlagIsNonInterleaved;
_stereoGraphStreamFormat.mBytesPerPacket = four_bytes_per_float;
_stereoGraphStreamFormat.mFramesPerPacket = 1;
_stereoGraphStreamFormat.mBytesPerFrame = four_bytes_per_float;
_stereoGraphStreamFormat.mChannelsPerFrame = 2;
_stereoGraphStreamFormat.mBitsPerChannel = eight_bits_per_byte * four_bytes_per_float;
_stereoGraphStreamFormat.mSampleRate = 44100;
to the following format:
interleavedAudioDescription.mFormatID = kAudioFormatLinearPCM;
interleavedAudioDescription.mFormatFlags = kAudioFormatFlagIsSignedInteger;
interleavedAudioDescription.mChannelsPerFrame = 2;
interleavedAudioDescription.mBytesPerPacket = sizeof(SInt16)*interleavedAudioDescription.mChannelsPerFrame;
interleavedAudioDescription.mFramesPerPacket = 1;
interleavedAudioDescription.mBytesPerFrame = sizeof(SInt16)*interleavedAudioDescription.mChannelsPerFrame;
interleavedAudioDescription.mBitsPerChannel = 8 * sizeof(SInt16);
interleavedAudioDescription.mSampleRate = 44100;
Using the following code:
int32_t availableBytes = 0;
void* tailL = TPCircularBufferTail(inputBufferL(), &availableBytes);
void* tailR = TPCircularBufferTail(inputBufferR(), &availableBytes);
// If we have no data in the buffer, we simply return
if (availableBytes <= 0)
{
return;
}
// ========== Non-Interleaved to Interleaved (Plus Samplerate Conversion) =========
// Get the number of frames available
UInt32 frames = availableBytes / this->mInputFormat.mBytesPerFrame;
pcmOutputBuffer->mBuffers[0].mDataByteSize = frames * interleavedAudioDescription.mBytesPerFrame;
struct complexInputDataProc_t data = (struct complexInputDataProc_t) { .self = this, .sourceL = tailL, .sourceR = tailR, .byteLength = availableBytes };
// Do the conversion
OSStatus result = AudioConverterFillComplexBuffer(interleavedAudioConverter,
complexInputDataProc,
&data,
&frames,
pcmOutputBuffer,
NULL);
// Tell the buffers how much data we consumed during the conversion so that it can be removed
TPCircularBufferConsume(inputBufferL(), availableBytes);
TPCircularBufferConsume(inputBufferR(), availableBytes);
// ========== Buffering Of Interleaved Samples =========
// If we got converted frames back from the converter, we want to add it to a separate buffer
if (frames > 0)
{
// Make sure we have enough space in the buffer to store the new data
TPCircularBufferHead(&pcmCircularBuffer, &availableBytes);
if (availableBytes > pcmOutputBuffer->mBuffers[0].mDataByteSize)
{
// Add the newly converted data to the buffer
TPCircularBufferProduceBytes(&pcmCircularBuffer, pcmOutputBuffer->mBuffers[0].mData, frames * interleavedAudioDescription.mBytesPerFrame);
}
else
{
printf("No Space in Buffer\n");
}
}
However I am getting the following output:
It should be a perfect sine wave, however as you can see it is not.
I have been working on this for days now and just can’t seem to find where it is going wrong.
Can anyone see something that I might be missing?
Edit:
Buffer initialisation:
TPCircularBuffer pcmCircularBuffer;
static SInt16 pcmOutputBuf[BUFFER_SIZE];
pcmOutputBuffer = (AudioBufferList*)malloc(sizeof(AudioBufferList));
pcmOutputBuffer->mNumberBuffers = 1;
pcmOutputBuffer->mBuffers[0].mNumberChannels = 2;
pcmOutputBuffer->mBuffers[0].mData = pcmOutputBuf;
TPCircularBufferInit(&pcmCircularBuffer, BUFFER_SIZE);
Complex input data proc:
static OSStatus complexInputDataProc(AudioConverterRef inAudioConverter,
UInt32 *ioNumberDataPackets,
AudioBufferList *ioData,
AudioStreamPacketDescription **outDataPacketDescription,
void *inUserData) {
struct complexInputDataProc_t *arg = (struct complexInputDataProc_t*)inUserData;
BroadcastingServices::MP3Encoder *self = (BroadcastingServices::MP3Encoder*)arg->self;
if ( arg->byteLength <= 0 )
{
*ioNumberDataPackets = 0;
return 100; //kNoMoreDataErr;
}
UInt32 framesAvailable = arg->byteLength / self->interleavedAudioDescription.mBytesPerFrame;
if (*ioNumberDataPackets > framesAvailable)
{
*ioNumberDataPackets = framesAvailable;
}
ioData->mBuffers[0].mData = arg->sourceL;
ioData->mBuffers[0].mDataByteSize = arg->byteLength;
ioData->mBuffers[1].mData = arg->sourceR;
ioData->mBuffers[1].mDataByteSize = arg->byteLength;
arg->byteLength = 0;
return noErr;
}
I see a few things that raise a red flag.
1) as mentioned in a comment above, the fact that you are overwriting availableBytes for the left input with that from the right:
void* tailL = TPCircularBufferTail(inputBufferL(), &availableBytes);
void* tailR = TPCircularBufferTail(inputBufferR(), &availableBytes);
If the two input streams are changing asynchronously to this code then most certainly you have a race condition.
2) Truncation errors: availableBytes is not necessarily a multiple of bytes per frame. If not then the following bit of code could cause you to consume more bytes than you actually converted.
void* tailL = TPCircularBufferTail(inputBufferL(), &availableBytes);
void* tailR = TPCircularBufferTail(inputBufferR(), &availableBytes);
...
UInt32 frames = availableBytes / this->mInputFormat.mBytesPerFrame;
...
TPCircularBufferConsume(inputBufferL(), availableBytes);
TPCircularBufferConsume(inputBufferR(), availableBytes);
3) If the output buffer is not ready to consume all of the input you just throw the input buffer away. That happens in this code.
if (availableBytes > pcmOutputBuffer->mBuffers[0].mDataByteSize)
{
...
}
else
{
printf("No Space in Buffer\n");
}
I'd be really curious if your seeing the print output.
Here's is how I would suggest doing it. It's going to be pseudo-codeish since I don't have anything necessary to compile and test it.
int32_t availableBytesInL = 0;
int32_t availableBytesInR = 0;
int32_t availableBytesOut = 0;
// figure out how many bytes are available in each buffer.
void* tailL = TPCircularBufferTail(inputBufferL(), &availableBytesInL);
void* tailR = TPCircularBufferTail(inputBufferR(), &availableBytesInR);
TPCircularBufferHead(&pcmCircularBuffer, &availableBytesOut);
// figure out how many full frames are available
UInt32 framesInL = availableBytesInL / mInputFormat.mBytesPerFrame;
UInt32 framesInR = availableBytesInR / mInputFormat.mBytesPerFrame;
UInt32 framesOut = availableBytesOut / interleavedAudioDescription.mBytesPerFrame;
// figure out how many frames to process this time.
UInt32 frames = min(min(framesInL, framesInL), framesOut);
if (frames == 0)
return;
int32_t bytesConsumed = frames * mInputFormat.mBytesPerFrame;
struct complexInputDataProc_t data = (struct complexInputDataProc_t) {
.self = this, .sourceL = tailL, .sourceR = tailR, .byteLength = bytesConsumed };
// Do the conversion
OSStatus result = AudioConverterFillComplexBuffer(interleavedAudioConverter,
complexInputDataProc,
&data,
&frames,
pcmOutputBuffer,
NULL);
int32_t bytesProduced = frames * interleavedAudioDescription.mBytesPerFrame;
// Tell the buffers how much data we consumed during the conversion so that it can be removed
TPCircularBufferConsume(inputBufferL(), bytesConsumed);
TPCircularBufferConsume(inputBufferR(), bytesConsumed);
TPCircularBufferProduceBytes(&pcmCircularBuffer, pcmOutputBuffer->mBuffers[0].mData, bytesProduced);
Basically what I've done here is to figure out up front how many frames should be processed making sure I'm only processing as many frames as the output buffer can handle. If it were me I'd also add some checking for buffer underruns on the output and buffer overruns on the input. Finally, I'm not exactly sure of the semantics of AudioConverterFillComplexBuffer wrt the frame parameter that is passing in and out. It's conceivable that the # frames out would be less or more than the number of frames in. Although, since your not doing sample rate conversion that's probably not going to happen. I've attempted to account for that condition by assigning bytesProduced after the conversion.
Hope this helps. If not you have 2 other clues. One is that the drop outs are periodic and two is that the size of the drop outs looks to be about the same. If you can figure out how many samples each are then you can look for those numbers in your code.
I don't think your output buffer, pcmCircularBuffer, is big enough.
Try replacing
TPCircularBufferInit(&pcmCircularBuffer, BUFFER_SIZE);
with
TPCircularBufferInit(&pcmCircularBuffer, sizeof(pcmOutputBuf));
Even if that is the solution, I think there are some problems with your code. I don't know exactly what you're doing, I guess encoding mp3 (which by itself is an uphill battle on iOS, why not use hardware AAC?), but unless you have realtime demands on both input and output, why use ring buffers at all? Also, I recommend using units to visually catch type frame/byte size mismatches: e.g. BUFFER_SIZE_IN_FRAMES
If it's not the solution, then I want to see the sine generator.
i am trying to build an iOS 7 application that detecting the sound/song pitch(or frequency), For example: 349.23Hz, 392.00Hz, 440.00Hz......
So, I download the "Auto Correllation" project (it's a Musician's ket http://musicianskit.com/developer.php), I run it on iOS 7 Simulator, it works fine, The "hanning fft window" have value (not NaN), and able get the frequency finally.
But, it doesn't work on iPhone device, it cannot has any value in "hanning fft window".
Can anybody have a look into these classes by Kevin Murphy and tell me how I could modify them to work on iPhone device(not the iOS simulator)?
Many many thanks~
I've pasted my code below:
// PitchDetector.m
-(id) initWithSampleRate: (float) rate lowBoundFreq: (int) low hiBoundFreq: (int) hi andDelegate: (id<PitchDetectorDelegate>) initDelegate {
self.lowBoundFrequency = low;
self.hiBoundFrequency = hi;
self.sampleRate = rate;
self.delegate = initDelegate;
bufferLength = self.sampleRate/self.lowBoundFrequency;
hann = (float*) malloc(sizeof(float)*bufferLength);
// applied the Hanning windows, the 'hann' is the Hanning fft Window
vDSP_hann_window(hann, bufferLength, vDSP_HANN_NORM);
sampleBuffer = (SInt16*) malloc(512);
samplesInSampleBuffer = 0;
result = (float*) malloc(sizeof(float)*bufferLength);
return self;
}
-(void) performWithNumFrames: (NSNumber*) numFrames;
{
int n = numFrames.intValue;
float freq = 0;
SInt16 *samples = sampleBuffer;
int returnIndex = 0;
float sum;
bool goingUp = false;
float normalize = 0;
for(int i = 0; i<n; i++) {
sum = 0;
for(int j = 0; j<n; j++) {
//here I found the hann[j] is NaN. seems doesn't have value in hann('hann' is the Hanning fft Window)
//if hann[j] is Not a Number (NaN), the value of sum also to be NaN.
sum += (samples[j]*samples[j+i])*hann[j];
}
if(i ==0 ) normalize = sum;
result[i] = sum/normalize;
}
......
......
}
I am using this same program from:
https://github.com/fotock/PitchDetectorExample/tree/1c68491f9c9bff2e851f5711c47e1efe4092f4de
Although I have not put this on an iPhone yet, only simulator, I was having problems from time time with the program crashing. I found that I needed to manually update it with from a "fork" of the code on github found here:
https://github.com/fotock/PitchDetectorExample/network
I added Jordan Liggitt's bug fixes manually and now the app does not crash. I hope this helps because if it does not, then I will be facing the same issues when I load this app on an iPhone.
Hope it works!
Update
I have now installed this on an iPhone vs the simulator and it works as it should without errors or crashing.
iam getting from c.dot net web service
byte[] data = new byte[] {-33, -96,0, 0, 0,0,0,0};
I want to convert this into long value
I tried this
long result = (long)ByteBuffer.wrap(index).getInt();
I am getting the result as -543162368 wheras actual value is 41183
First off you want to call getLong() instead of getInt() on the buffer.
However, the data you're receiving is little-endian, which means that it starts with the low order byte first. ByteBuffers are constructed as default with big endian order. You need to set the order to LITTLE_ENDIAN to get the correct value out.
ByteBuffer buffer = ByteBuffer.wrap(index)
buffer.order(ByteOrder.LITTLE_ENDIAN);
long result = buffer.getLong();
Since you apparently can't set the byte order or use getLong, you will need to do it like this:
// Reverse array
for (int i = 0; i < 4; ++i)
{
byte temp = data[i];
data[i] = data[8-i];
data[8-i] = temp;
}
// Get two ints and shift the first int into the high order bytes
// of the result.
ByteBuffer buffer = ByteBuffer.wrap(data);
long result = ((long)buffer.getInt()) << 32;
result |= (long)buffer.getInt();
result should now contain the value.