BitonicSorter.compute

#pragma kernel BitonicInit
#pragma kernel BitonicMain

#define THREAD_GROUP_SIZE 256
#define MAX_PARALLEL_GROUPS 1024
#define MAX_PARALLEL_THREADS (THREAD_GROUP_SIZE * MAX_PARALLEL_GROUPS)

uint _Count;
int _Block, _Dimension;
RWStructuredBuffer<uint> _Keys;
StructuredBuffer<float> _Values;

[numthreads(THREAD_GROUP_SIZE, 1, 1)]
void BitonicInit (uint3 threadID : SV_DispatchThreadID) 
{
	uint i = threadID.x + threadID.y * MAX_PARALLEL_THREADS;
	if (i < _Count) _Keys[i] = i;
}

[numthreads(THREAD_GROUP_SIZE, 1, 1)]
void BitonicMain (uint3 threadID : SV_DispatchThreadID) 
{
	uint x = threadID.x + threadID.y * MAX_PARALLEL_THREADS;
	uint y = x ^ _Block;
	if (y < x || x >= _Count) return;
	uint a = _Keys[x];
	uint b = _Keys[y];
	bool shift = ((_Values[a] - _Values[b]) * ((x & _Dimension) == 0 ? 1 : -1)) > 0;
	_Keys[x] = shift ? b : a;
	_Keys[y] = shift ? a : b;
}