Announcement

Collapse
No announcement yet.

Intel Updates Its Closed-Source OpenCL SDK For Linux

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

  • Intel Updates Its Closed-Source OpenCL SDK For Linux

    Phoronix: Intel Updates Its Closed-Source OpenCL SDK For Linux

    The 2014 release of Intel's OpenCL SDK for Windows and Linux is now available...

    http://www.phoronix.com/vr.php?view=MTY5Nzk

  • #2
    Hi,

    IMO, if you really got a chance to do some benchmark(luxmark or some luxrender cases) on the same IVB platform with beignet and the windows opencl sdk, you may change the following conclusion:
    "...but it's far from being on-par with the Intel OpenCL Windows support... "

    The beignet team have been working hard in the past months. And now, beignet could get about 99% pass rate on all the test suite we have now including opencv ocl test suite/piglit and some other internal test suite.
    And beignet is now about 10-20x faster than the last released version 0.8. The baytrail support is done in good shape, and the haswell support is also almost done.

    Comment


    • #3
      They should rather keep up with llvm development. beignet seems to never compile with the latest llvm revision.

      Comment


      • #4
        We really wanted to keep up with the latest LLVM version. But we just found that is not such an easy thing for the following reasons:
        1. LLVM change interfaces very freqently, even without a version bump.
        2. Sometime, new LLVM release introduces severity OpenCL related regressions , for example when we tried to update to clang/llvm 4.4.
        We submitted the related bugs into the llvm bug tracking system. But there are no responses for months.

        So we change to stick to one stable version which is LLVM/clang 3.3 currently. And then before next major release, we will try to make it
        work with latest LLVM/Clang if possible. Anyway, thanks for point this out, this is really one of the pain with llvm/clang.

        Originally posted by ChrisXY View Post
        They should rather keep up with llvm development. beignet seems to never compile with the latest llvm revision.

        Comment


        • #5
          If it's really in that good a shape, is there a release coming? You can't complain about people not knowing the current state if there is no stable release advertised around.

          Comment


          • #6
            Originally posted by curaga View Post
            If it's really in that good a shape, is there a release coming? You can't complain about people not knowing the current state if there is no stable release advertised around.
            Yes, we are going to release next stable version in about one month. Please just treat my "complain" as some sort of pre-advertisement for the new release and no offence.

            Comment


            • #7
              I'm using 3.5 (3.x, not 4.x) svn snapshots because of radeonsi. (hybrid intel ivy bridge / radeonsi graphics in a laptop)

              Of course I could just have one stable and one svn installed and do stuff with environment variables and configure switches, but it's kind of a hassle...

              I'm not sure what you're doing wrong. Tom Stellard seems be able to push things upstream with not much resistance if he is confident it's good and has test cases and such. Maybe ask on the mailing list, that seems to be the most active channel?

              Comment


              • #8
                Originally posted by ChrisXY View Post
                I'm using 3.5 (3.x, not 4.x) svn snapshots because of radeonsi. (hybrid intel ivy bridge / radeonsi graphics in a laptop)

                Of course I could just have one stable and one svn installed and do stuff with environment variables and configure switches, but it's kind of a hassle...

                I'm not sure what you're doing wrong. Tom Stellard seems be able to push things upstream with not much resistance if he is confident it's good and has test cases and such. Maybe ask on the mailing list, that seems to be the most active channel?
                Right, 4.4 was a typo, it should be 3.4. Thanks for correct that. I think the story is different from mesa/clover, as the major issue with llvm/clang is in the compiler backend not the libcl runtime driver part. Beignet itself has a standalone backend for GEN IR. And mesa/clover just includes a runtime driver. If I am wrong, please correct me. Thanks.

                Comment


                • #9
                  Originally posted by ChrisXY View Post
                  I'm using 3.5 (3.x, not 4.x) svn snapshots because of radeonsi. (hybrid intel ivy bridge / radeonsi graphics in a laptop)

                  Of course I could just have one stable and one svn installed and do stuff with environment variables and configure switches, but it's kind of a hassle...

                  I'm not sure what you're doing wrong. Tom Stellard seems be able to push things upstream with not much resistance if he is confident it's good and has test cases and such. Maybe ask on the mailing list, that seems to be the most active channel?
                  Right, 4.4 was a typo. It should be llvm/clang 3.4. Thanks for correct that.
                  As to the mesa/clover, IMO, the biggest difference here is that beignet has a complete llvm backend for Gen IR, but mesa/clover only has a runtime driver. The major compatible issue with LLVM/CLANG is related to the compiler backend. The runtime library has little or no issue with different llvm/clang versions. If I am wrong, please correct me. Thanks.

                  Comment

                  Working...
                  X