New TTM Code Can Yield 3~5x Faster Page Allocation For AMDGPU, Other Benefits

[phoronix]

Phoronix: New TTM Code Can Yield 3~5x Faster Page Allocation For AMDGPU, Other Benefits

The Linux kernel's TTM memory management code that is most notably used by the Radeon / AMDGPU kernel drivers but also Nouveau, QXL, VMWGFX, and others, is seeing a new back-end allocation pool that can yield 3~5x faster page allocation performance for video memory...

http://www.phoronix.com/scan.php?page=news_item&px=TTM-New-Page-Alloc-Pool

[oleid]

Uh, interesting! How would this be visible in real world applications and games? I mean, I would expect games to not constantly reallocate memory.

[caligula]

Nice, but why was the old code so bad? Who's to blame?

[Tomin]

Nice, but why was the old code so bad? Who's to blame?

My guess is that no one in particular. Like most code, it was probably designed for something simpler or otherwise for a different use case originally and then it has grown to cover more. It just was in need of a rewrite. It's quite common in software that some parts of code need rewrites to simplify them after time has passed.

Note that I didn't even look at the code so I don't know what's the truth here.

[Steffo]

I would like to see a comparison benchmark.

[oleid]

Nice, but why was the old code so bad? Who's to blame?

Obviously : your mama sorry, couldn't resist ^^

[smitty3268]

My guess is that no one in particular. Like most code, it was probably designed for something simpler or otherwise for a different use case originally and then it has grown to cover more. It just was in need of a rewrite. It's quite common in software that some parts of code need rewrites to simplify them after time has passed.

Note that I didn't even look at the code so I don't know what's the truth here.

I believe TTM may have initially been written by VMWare, so it would make sense that it probably got extended in various ways it wasn't designed for when hardware drivers started using it.

[rene]

Nice, but why was the old code so bad? Who's to blame?

Then one who asks and did not review nor write a patch before ;-)

[MrCooper]

I believe TTM may have initially been written by VMWare, so it would make sense that it probably got extended in various ways it wasn't designed for when hardware drivers started using it.

TTM was originally written by Tungsten Graphics, before it was acquired by VMware. It was originally designed primarily for Intel integrated GPUs. For whatever reasons, Intel decided to invent GEM instead of going with TTM (even though the latter was already working with comparable functionality).

[Deathsimple]

Uh, interesting! How would this be visible in real world applications and games? I mean, I would expect games to not constantly reallocate memory.

I don't expect much improvement for real world applications. They just try to avoid freeing and reallocating memory quite hard because of the overhead.