Archives Discussions

[anlongstar]

domian, stream, array

eg:

//--------------------------------------------------------

Kernel:

kernel void
domainAdd(double inCCCCCC,double inputStream[][], out double o[][])
{
       int2 position = instance().xy;
   int i=position.x;
   int j=position.y;
   o=inputStream+1.0;

}

kernel void
domainAddStream(double inCCCCCC,double inputStream<>, out double o<>
{
   o=inputStream+1.0;
}

//-------------------------------------------------

the main:

 int blockSize=128; 

#define real double
 
 const unsigned int width = blockSize;
 const unsigned int height = blockSize;
 
 // Specifying the size of the 2D stream
 unsigned int streamSize[] = {width, height};
 
 // Specifying the rank of the stream
 unsigned int rank = 2;
 
 real* outputBuffer = new real[width * height];
 memset(outputBuffer, 0, width * height * sizeof(real));
 real* inputBuffer = new real[width * height];
 memset(inputBuffer, 1, width * height * sizeof(real));
 
 // Create a 2D stream of specified size i.e. 64x64 floating-point values   
 brook::Stream<real> *outputStream;
 outputStream=new Stream<real>(rank, streamSize); 
 brook::Stream<real> inputStream(rank, streamSize);
 

 inputStream.read(inputBuffer);
 unsigned int start[] = {5,0};//(width,height)
 unsigned int end[] = {16,4};

 
 //domainAdd(1.0,inputStream.domain(start,end),outputStream->domain(start,end));// this will get the wrong answer
 domainAddStream(1.0,inputStream.domain(start,end),outputStream->domain(start,end));

[gaurav_garg]

kernel void domainAdd(double inCCCCCC,double inputStream[][], out double o[][]) {        int2 position = instance().xy;    int i=position.x;    int j=position.y;    o=inputStream+1.0;

}

kernel void domainAddStream(double inCCCCCC,double inputStream<>, out double o<>) {    o=inputStream+1.0; }

Both of your above kernels are doing different thing. instance().x points to the column of the kernel instance and instance().y points to the row.

When using C-style indexing for gather or scatter stream, you must index like this-

o[row][column]

Making this change, gives the same result for both the kernels.

[anlongstar]

Thank u!