Intermediate variable elimination in a global context for a 3D multimedia application

Power Estimation Approach of Dynamic Data Storage on a Hardware Software Boundary Level

“…One of the results of this initial profiling is detailed timing information as shown in Figure 2, there it can be seen that intra-algorithm data dependencies with small but extremely accessed buffers exist in both cases. Next, this information is used to apply global transformations [9]. This results in the removal of 3 DDTs in the matching algorithm and 2 DDTs in the game engine, as Table 1 and Figure 3 show.…”

“…When these intermediate DDTs are not used for any other purpose and there is an injective relation in the data-flow, they can be removed [9]. Figure 3 shows the results obtained after these transformations for our multimedia drivers.…”

“…This will be illustrated based on an application demonstrator. The details of the optimisation approach itself are given in [9]. It allows to analyse the behaviour of the DDTs of the application, as Figure 2 shows.…”

“…A library provides standardised interfaces to the most commonly used N -layered implementations. source to source transformations are possible: Using structured profiling reports, a global source code optimisation is able to eliminate temporary buffers that introduce memory movements between intra-algorithmic phases without any useful processing of data [9].…”

“…intermediate variable elimination transformation [9]). Finally, the refinement of the DDTs can be performed with an exploration that uses the same profiling framework to evaluate possible trade-offs between power consumption, memory footprint and performance.…”