allocations: Make buffer kernel more efficient for multiple allocations

[lakshmih]

• Fix malloc for 'access_string' and 'kernel_string'.
• Fix typo in 'number_of_work_itmes'.

[lakshmih]

This is how the change would modify the kernel:

Old:

__kernel void sample_test( __global uint *buffer0,  __global uint *buffer1,  __global uint *buffer2,  __global uint *buffer3,  __global uint *result, __global ulong *array_sizes, uint per_item)
{
	int tid = get_global_id(0);
	uint r = 0;
	ulong i;
	for(i=(ulong)tid*(ulong)per_item; i<(ulong)(1+tid)*(ulong)per_item; i++) {
		if (i<array_sizes[0]) r += buffer0[i];
		if (i<array_sizes[1]) r += buffer1[i];
		if (i<array_sizes[2]) r += buffer2[i];
		if (i<array_sizes[3]) r += buffer3[i];
	}
	result[tid] = r;
}

New:

__kernel void sample_test( __global uint *buffer0,  __global uint *buffer1,  __global uint *buffer2,  __global uint *buffer3,  __global uint *result, __global ulong *array_sizes, uint per_item)
{
	int tid = get_global_id(0);
	uint r = 0;
	ulong i;
	ulong end0 = min((ulong)(1+tid)*(ulong)per_item, array_sizes[0]);
	for(i=(ulong)tid*(ulong)per_item; i<end0; i++) {
		r += buffer0[i];
	}
	ulong end1 = min((ulong)(1+tid)*(ulong)per_item, array_sizes[1]);
	for(i=(ulong)tid*(ulong)per_item; i<end1; i++) {
		r += buffer1[i];
	}
	ulong end2 = min((ulong)(1+tid)*(ulong)per_item, array_sizes[2]);
	for(i=(ulong)tid*(ulong)per_item; i<end2; i++) {
		r += buffer2[i];
	}
	ulong end3 = min((ulong)(1+tid)*(ulong)per_item, array_sizes[3]);
	for(i=(ulong)tid*(ulong)per_item; i<end3; i++) {
		r += buffer3[i];
	}
	result[tid] = r;
}