Reducing The CPU Usage In Mesa To Improve Performance

[log0]

Those are the moments where I really wish GCC would implement processor-specific auto-dispatch like the Intel compiler, so that people don't have to write that stuff by hand.

[leonmaxx]

auto-dispatch like the Intel compiler

Yeah, and don't forget "Cripple AMD" feature - that is why i never used intel compilers.

[dungeon]

Hmm just tried, not impressed by that mesa patch... CPU usage goes down, but fps goes down too

[okias]

Hmm just tried, not impressed by that mesa patch... CPU usage goes down, but fps goes down too

This code only optimize, it can't affect FPS. FPS could affected only in case that CPU has been 80%+ and can't feed GPU fast enough. As you say, CPU usage decreased, so it can't have any effect on FPS.

What card and configuration do you use?

[Luke]

Why not? Phoronix Test Suite was one of their tools

No need to give them article coverage even, as long as the cpu and gpu usage can be found in openbenchmarking for the run.

For Phoronix tools to be used this way contributes more value to the Linux community as a whole than almost anything else. Phoronix is part of the story, thus has every reason to cover it. This is also a good example of why both game and hardware vendors should contribute both to Phoronix and to experimenters who do this kind of heavy lifting.

[nanonyme]

Michael: How about getting inspired and making it easier to run profilers with pts? According to the blog it at least used to be a tad cumbersome

[leonmaxx]

This code only optimize, it can't affect FPS. FPS could affected only in case that CPU has been 80%+ and can't feed GPU fast enough. As you say, CPU usage decreased, so it can't have any effect on FPS.

Actually it can. Lesser load on CPU - cpufreq governor could run it on lower frequency.

[curaga]

For Phoronix tools to be used this way contributes more value to the Linux community as a whole than almost anything else. Phoronix is part of the story, thus has every reason to cover it. This is also a good example of why both game and hardware vendors should contribute both to Phoronix and to experimenters who do this kind of heavy lifting.

If Michael measures "oh, no change & nothing interesting", he's better off not including the graph as he already does for such non-events. But the results would still be on OB as long as he enabled the monitors, and even a "boring" result might be useful to somebody.

[tarceri]

Hmm just tried, not impressed by that mesa patch... CPU usage goes down, but fps goes down too

Out of interest. What did you benchmark and what were the results?

Also what cpu/gpu did you test with?

[tarceri]

Did he finish ARB_arrays_of_arrays at last?

Its mostly done I've been waiting for reviews of my latest piglit tests before pushing to hard with the Mesa changes. However, the dev's are busy and are taking a while to getting around to review.