CREAMS : The Challenges Of ISA-Homogeneous Systems
1046 Words5 Pages
The paper then takes a brief detour from talking about CReAMS to talk about attempts that have previously been made to solve many of the same challenges that the authors hope to solve with CReAMS. The paper first talks about ISA-homogeneous systems, namely ReMAPP, Thread Warping, and big-LITTLE, and then the heterogeneous system KAHRISMA. For all four of these systems, the benefits and drawbacks are provided, which is meant to provide context for how the benefits of both homogeneous and heterogeneous systems were determined. The authors hoped to use what they learned from these systems to create CReAMS. CReAMS is a homogeneous system that achieves simulated heterogeneity dynamically using a binary translation mechanism. CReAMS also uses a…show more content… The system would have a four set associative address cache with 64 entries, a private 32 KB four-way set associative data cache, and a private 8 KB four-way set associative instruction cache. This section drew heavily on what we learned in class about pipelining, logical units, and memory caching. This made for a great way to apply knowledge learned in class.
The paper then presents the final aspect of the system, which is specifically of the authors’ design. This is what the authors call dynamic detection hardware, or DDH. This hardware is responsible for detecting instructions, as well as allocation in the datapath described above. It is a four stage pipelined circuit with the stages instruction decode, dependence verification, resource allocation, and update tables. The paper uses a simple loop in code to demonstrate the four modes of the hardware. The four stages are probing, detecting, reconfiguring, and accelerating. This a very interesting way of approaching allocation in the datapath and is quite different from anything we learned in class.
The paper finishes by presenting the experimental results of CReAMS against two different, more typical SparcV8-based systems and basically has a victory lap after showing greater performance across a wide variety of applications along with greater energy efficiency, especially in more complex applications.
Introduction Main memory is an absolutely essential part of computing systems. This fact is obvious, as there has to be