Running The RadeonSI NIR Back-End With Mesa 19.1 Git

Some of the synthetic OpenGL benchmarks were also helped by the NIR route.

When looking at the CPU usage between these RadeonSI configurations, overall it was nearly the same but under the default stack there were cases of higher CPU spikes than with the NIR mode.

The GPU usage for the most part was the same.

Just in select cases like BioShock Infinite were there some areas where using NIR led to lower GPU utilization while still delivering faster performance on advantage.

Arguably the most important factor though is no longer hitting any major problems like we encountered in past rounds of RadeonSI NIR testing. But as far as when the RadeonSI NIR mode will be enabled by default, that has yet to be determined but can be easily enabled via the R600_DEBUG=nir environment variable for more widespread testing.