Achieving Max Performance

[EJain-Dev]

For any kernel running on an OpenCL enabled device, the following holds true to achieve optimal performance:

If each EU runs n warps concurrently, ideally the number of warps per EU should be equivalent to: 
n * ((mem_ops * (flops / bandwidth)) / compute_ops)

However, this requires 3 things that OpenCL does not provide:

1. FLOPS
2. Bandwidth
3. Register File Size

Is there any extension or future extension idea to provide these 3 details for a given device?