> -----Original Message----- > From: > intel-gfx-bounces+zhigang.gong=linux.intel.com at lists.freedesktop.org > [mailto:intel-gfx-bounces+zhigang.gong=linux.intel.com at lists.freedesktop. > org] On Behalf Of Zou, Nanhai > Sent: Tuesday, April 16, 2013 2:47 PM > To: Dave Airlie > Cc: intel-gfx at lists.freedesktop.org > Subject: Re: FW: Release Notes for Beignet Version 0.1 > > Hi, > > > Again not really important, we use LLVM and TGSI backends in the > radeon drivers now. The gallium drivers currently use GPU specific LLVM > IRs produced from specific LLVM backends, but that isn't strictly necessary > just leads to better optimising opportunities. > > Because you now have a whole bunch of code that is useless to anyone > else. Distro want to ship this sort of thing, we don't want 5 Mesa like > implementations for OpenCL, we want one we can actually distribute and > manage, and >maybe in 5 years support. > > The point is at the current stage, we don't see the necessary of > introduce all those complexity and dependencies. > Considering we can share very few code with other part of > Mesa(expect some GPU related define in header files). > > We want to keep this project tiny and self-contained at this point, > focus on make OpenCL really fast and useful on IVB+ GPU. > > We can merge or hook the backend into gallium when there are > other mature gallium based OpenCL implementations. > That should not be a lot of task, consider the common code are very > few. > > > > Some other questions: > > a) have you got an ivybridge LLVM backend? are you going to upstream > this, I heard it isn't even written in LLVM machine description format. > > No, we convert llvm IR to Gen IR then to IVB assembly. > There are many choices when implement an execute model of IVB. E.g > 8 way or 16 way, SOA mode or AOS mode. > We choose 16 way AOS mode in the driver. > I don't know if it make sense to upstream only this particular > combination. > > > > b) have you looked at pocl, libclc etc? Maybe you guys want to run on > CPU as well at GPU at some point, in which case maybe again looking > around before jumping into implementing stuff might help. > > We do have looked at libclc. > Again, I don't see the necessary to introduce the complexity at this > point > Almost all the functions in libclc maps to 1 instructions on IVB. We do > not see the necessary of introducing a library to wrap those. > > > c) does this use the open source ICD at least? > No, we will check that. [Gong, Zhigang] As I know, Simon implemented ICD for Beignet and it may need some time to rebase to the latest beignet code base. If you are interested, this is the link http://psi5.com/~geier/beignet.git. > > > Dave. > _______________________________________________ > Intel-gfx mailing list > Intel-gfx at lists.freedesktop.org > http://lists.freedesktop.org/mailman/listinfo/intel-gfx
