Watch David Airlie Talk About Vulkan & RADV From LCA2017

[jaxxed]

If NIR is converted to LLVM IR before optimisation is applied, why NIR is needed in the first place? It is better to convert SPIRV to LLVM IR directly, isn't it?

This question has been brought up before. Smarter people than I replied with the two following points:

1. IR conversion is considered a low cost operation (doing it many times doesn't consume too much resources compared to other operations)
2. Using a common IR format at some point in mesa allows developers from different drivers to cooperate on feature development and feature optimization, for cases where features can be developed outside of the kernel driver.

Basically, if the IR is good, then you can write common implementations that work across drivers, in a way that each driver can then figure out how to consume.

* I remember also some discussion about how one of the IRs used in Mesa (I think in Gallium) is not simple to convert to/from NIR as it is not SSA, while NIR is SSA. That might have been Ilia or Marek who mentioned that.

[starshipeleven]

The hypocrite. First he sides with ATI on destroying a proper open source display driver, then he acts like the big savior a few years down the road, when ATI is unpopular once again. Unbelievable. Powerplay all the way.

Stop this whining, all this only makes you look as a sore loser.

[libv]

Stop this whining, all this only makes you look as a sore loser.

Well, you're the one stuck with atombios thanks to people like Dave Airlie. I gladly stay very very away from AMD hw since my SuSE days.

[starshipeleven]

Well, you're the one stuck with atombios thanks to people like Dave Airlie. I gladly stay very very away from AMD hw since my SuSE days.

They act like men at least, and got something working at all even if it's half-closed.

You failed and keep acting like a fucking child.

Now man the fuck up and stop this whining.

[libv]

They act like men at least, and got something working at all even if it's half-closed.

You failed and keep acting like a fucking child.

Now man the fuck up and stop this whining.

I am not sure who you really are, but close to 4k posts in just over a year is pretty impressive. You smell like either an astroturfer or someone who used to be named differently. You probably know full well what you are doing here, but in case you need to go catch up on history:

In 2007/2008, there was real C code that was working _very_ well, with a massive community backing it. For no reason except for their own glorification, Dave Airlie and his friends sided with ATI to work to kill the real open source driver that forced ATIs hand. DAL and the lack of an AMD written vulkan top level is his own fault, and even if his actions today are honorable (i've seen him in action enough to doubt it), it can never make up for close of a decade of missing documentation and firmware dependency that he helped institute.

[starshipeleven]

but in case you need to go catch up on history: ....

Stop it. You failed, they won. They are going to be the "official heroes of opensource AMD drivers" even if they are not, but if you keep acting like this you will keep looking like a butthurt pre-schooler.

You can choose to stop whining and do something else, or keep whining and getting trolled by random guys on forums that are fed up with grown-up men acting like pre-schoolers also in real life.

[libv]

... getting trolled by random guys on forums...

Ah, full time troll it is then. Understood.

[starshipeleven]

Ah, full time troll it is then. Understood.

Glad you can read between the lines, lol.

[indepe]

Of course it is … it's royalty-free (though certification will cost you unless you are a significant OSS project like Mesa), it's portable, easy to implement (so less barrier for open source implementations as seen in the video with RADV) and leaves less room for drivers to hide „secret sauce” which generally benefits closed-source implementations.

Last but not least, those who have worked with it in depth, appear to think it accomplishes its technical goals. For me, that is/was one of the open questions.

[difron]

This question has been brought up before. Smarter people than I replied with the two following points:

1. IR conversion is considered a low cost operation (doing it many times doesn't consume too much resources compared to other operations)
2. Using a common IR format at some point in mesa allows developers from different drivers to cooperate on feature development and feature optimization, for cases where features can be developed outside of the kernel driver.

Basically, if the IR is good, then you can write common implementations that work across drivers, in a way that each driver can then figure out how to consume.

* I remember also some discussion about how one of the IRs used in Mesa (I think in Gallium) is not simple to convert to/from NIR as it is not SSA, while NIR is SSA. That might have been Ilia or Marek who mentioned that.

Thank you for your answer! Still cant say I am convinced by this reasoning. Amd already have driver built around llvm ir. Same llvm backend or close one us used in rocm project I think. And it is natural to expect them to develop spirv llvm frontend as it is part of opengl 2.1.