AMD's Kernel Graphics Code Is Approaching One Million Lines

[microcode]

Exhibit A why Linux desperately needs a working HAL. A stable driver ABI wouldn't hurt either.

A HAL would just make the headers even larger, I don't know what you think you know, but I don't think you know what you think you know.

[smitty3268]

Why are there 6 times as many lines in header files than source files?

Why does the AMDGPU driver contain more lines than all the others combined? I mean Intel graphics and radeon are full-featured (shader, video en-/decoding, display) drivers as well, so why does AMDGPU need that many additional lines? Is this a code quality issue like with DC, just to a lesser extent?

Header files = code comments + lots of register defines.

It's a non-issue. Nobody cares how big those header files get.

Note that if you look at lines of c/c++ code, the driver is in line with the others:
amd: 135k
radeon: 120k
intel: 110k
nouveau: 95k

[Marc Driftmeyer]

Exhibit A why Linux desperately needs a working HAL. A stable driver ABI wouldn't hurt either.

This has been an OEM desire for over 10 years but God told them all to shove it as he announces his next round of kernel okays.

[Marc Driftmeyer]

Header files = code comments + lots of register defines.

It's a non-issue. Nobody cares how big those header files get.

Note that if you look at lines of c/c++ code, the driver is in line with the others:
amd: 135k
radeon: 120k
intel: 110k
nouveau: 95k

Correct. This is something a first quarter/semester C student learns regarding code size.

[CrystalGamma]

If i want to buy a 1080TI or Fury X PCIe3.0-16X card and if i only consider this as a Full-Featured GPU i will not find any Intel product in this class of hardware. Full-Featured hardware means: PCIe 16X card.

I don't see what features have to do with performance …

[ldo17]

Exhibit A why Linux desperately needs a working HAL. A stable driver ABI wouldn't hurt either.

I can remember, some years ago, a comment by Greg Kroah-Hartman, along similar lines, about the Linux USB support stack.

At that point, both Windows and Linux had reworked their USB driver support three times. The difference was, Windows still carried around the debris from those previous attempts, for backward compatibility to avoid breaking older drivers. Linux didn’t have to, since all the in-tree drivers could be updated at the same time as the underlying architecture was redone.

And that is yet another reason why Linux can run so much more efficiently than Windows on the same hardware, and why Linux development can proceed so much faster than Windows.

[dungeon]

Anybody knows how much those guys have code lines?

If you ask MS user something like "How much CUs it have?" he we probably answer "dunno, got another yesterday"

[ldo17]

Anybody knows how much those guys have code lines?

Hundreds of millions. Of large chunks of which, nobody left at Microsoft understands what they do.

[CrystalGamma]

Correct. This is something a first quarter/semester C student learns regarding code size.

I guess my memories of C++ (where it's not unusual to have code in header files, which is bad for compile time and often for binary size) got the better of me.

[ldo17]

This is something a first quarter/semester C student learns regarding code size.

... at least in the days before inline ...