Announcement

Collapse
No announcement yet.

NIR Has Been Revised As A New IR For Mesa

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • NIR Has Been Revised As A New IR For Mesa

    Phoronix: NIR Has Been Revised As A New IR For Mesa

    Posted to the Mesa mailing list this week was a set of 123 patches that reintroduces NIR as a new IR for Mesa...

    Phoronix, Linux Hardware Reviews, Linux hardware benchmarks, Linux server benchmarks, Linux benchmarking, Desktop Linux, Linux performance, Open Source graphics, Linux How To, Ubuntu benchmarks, Ubuntu hardware, Phoronix Test Suite

  • #2
    Acronyms

    What is SSA?

    Michael, please use the <abbr> tag.

    Comment


    • #3
      Originally posted by uid313 View Post
      What is SSA?

      Michael, please use the <abbr> tag.

      Comment


      • #4
        And it's important because a lot of compiler optimizations become possible or a lot simpler when in this form. Most academic papers discussing new optimizations and algorithms assume everything will be in SSA form as well.

        Mesa already has lots of hardware backends that go to SSA form (such as the r600g sb-optimized backend) but it wasn't present in the common GLSL compiler before.
        Last edited by smitty3268; 18 December 2014, 03:17 PM.

        Comment


        • #5
          Okay, If here is some of the Mesa developers, please explain why we need another IR?
          Why LLVM as used by AMD is not suitable. LLVM is already in SSA form, and works quite well for AMD FOSS driver ( and closed source one too).
          Thanks.

          Comment


          • #6
            Originally posted by Drago View Post
            Okay, If here is some of the Mesa developers, please explain why we need another IR?
            Why LLVM as used by AMD is not suitable. LLVM is already in SSA form, and works quite well for AMD FOSS driver ( and closed source one too).
            Thanks.
            There was a discussion on this, and it basically came down to the fact that Intel didn't want to have LLVM as a dependency. It's really that simple, cut, and dry.

            This code is 99% written by Intel, so they can basically do whatever they want. Everyone else is just along for the free ride.

            Comment


            • #7
              Originally posted by smitty3268 View Post
              There was a discussion on this, and it basically came down to the fact that Intel didn't want to have LLVM as a dependency. It's really that simple, cut, and dry.

              This code is 99% written by Intel, so they can basically do whatever they want. Everyone else is just along for the free ride.
              And yet 99% of LLVM/Clang that Intel is leveraging they haven't written a single line of code.

              Perhaps it is one of the reasons that only Argonne Labs, Intel and a few others only give a rat's ass about OpenMP being in the LLVM/Clang project.

              Apple, Google, Sony, AMD, etc., really couldn't give a rats ass about it.

              Comment


              • #8
                Originally posted by Marc Driftmeyer View Post
                And yet 99% of LLVM/Clang that Intel is leveraging they haven't written a single line of code.
                Huh? Intel isn't using LLVM.

                At least not in Mesa. Intel seems pretty team-driven, in that different teams can choose what they want to use and what they don't want to use. Certain Mesa intel developers are very anti-LLVM, but that doesn't really mean anything to the rest of the company.

                Comment

                Working...
                X