Using Heuristic Value Prediction and Dynamic Task Granularity Resizing to Improve Software Speculation

<table class="table-group" width="800px"><tr class="fig-image" id="a"><td><img alt="478013.fig.008a" src="https://static.hindawi.com/articles/tswj/volume-2014/478013/figures/478013.fig.008a.jpg"/></td></tr><tr class="fig-caption"><td align="center"><table><tr><td><b>(a) </b>Execution with DEPENDENCY = 0 (bigger task granularity wins)</td></tr></table></td></tr><tr class="fig-image" id="b"><td><img alt="478013.fig.008b" src="https://static.hindawi.com/articles/tswj/volume-2014/478013/figures/478013.fig.008b.jpg"/></td></tr><tr class="fig-caption"><td align="center"><table><tr><td><b>(b) </b>Execution with DEPENDENCY = 1 (smaller task granularity wins)</td></tr></table></td></tr></table>

<table>The execution flow of a code section under the conventional speculative model with static task granularity (the two parts of the figure share the same code section, in which the “#ifdef” part in the code section brings a dependency. The shadowed part in the task shows the control overhead brought by task creating and committing. The (a) part of the figure shows the execution flow with DEPENDENCY = 0, while the (b) part shows the execution flow with DEPENDENCY = 1. The bigger task granularity is 8, while the smaller task granularity is 2).</table>

The Scientific World Journal

fig8

Figure 8

Figure 8: Using Heuristic Value Prediction and Dynamic Task Granularity Resizing to Improve Software Speculation 