intel/compute-runtime 2018ww19-010806

on GitHub

Components revisions included in the package:

• intel/compute-runtime@ b59a5f1
• intel/gmmlib@a501534
• intel/intel-graphics-compiler@858e183
• intel/llvm-patches@0ba9deb
• intel/opencl-clang@fa75886

Additional components revisions used in build:

• intel/libva@c9bb65b
• KhronosGroup/OpenCL-Headers@f039db6

QUALITY EXPECTATIONS

Unless noted otherwise, the driver was validated on Ubuntu 16.04.4 LTS with stock kernel.

Quality levels:

• Pre-Release - major issues or regressions; suitable for preliminary checkout
• Beta - no known major issues or regressions; suitable for integration and broad testing
• Production - Beta + meets Khronos certification requirements; suitable for production

ISSUES & REGRESSIONS OF NOTE

• Fine grained SVM is not supported in this release. Please see issue #34 for details.
• Broadwell - device enqueue not fully operational – you may experience slowness in kernel execution or GPU hang
• In our weekly validation runs we have switched from Ubuntu 16.04.3 (default kernel 4.10) to Ubuntu 16.04.4 (default kernel 4.13). Functional delta:
  • GLK and CFL platforms are supported, we will no longer use vanilla kernel 4.14 LTS to validate these platforms. For CFL i915.alpha_support=1 kernel parameter is required.
  • One functional issue on GLK under debug – GLK downgraded to “Beta” quality
• Since the last release, the driver was changed to create an asynchronous thread in DrmGemCloseWorker class which may cause segfaults when application or ICD doesn't link pthread library. It's known issue https://sourceware.org/bugzilla/show_bug.cgi?id=16628 when thread is created in library loaded by dlopen in application which is not linked with pthread library. We are implementing a solution to this.
  In the meantime, you can use one of the following workarounds:
  • compile application with flag "-pthread"
  • set environment variable LD_PRELOAD=<path/to/libthread>
  • use an ICD which links pthread library