On Wed, 2014-06-18 at 08:45 +0200, ext Michal Babej wrote:
> Hi,
>
> So, about merging to master... I did a few changes in the cmake branch
> that touch the code as well. If we want to keep both autotools and
> cmake, i need to revert them (or adapt autotools to them).
At least for a transition period we want to keep both. How about in the
long run? Both systems have their pros and cons, and for me they come
out pretty even.
> - replace CLFLAGS with KERNEL_CL_FLAGS, and CLANGXX_FLAGS with
> KERNEL_CLANGXX_FLAGS - they seem more fitting and less confusing
> (with LLVM_{CXX,C,CL}FLAGS being values from llvm-config)
Less confusion is always good. A lot of confusion has come from the
pthread and basic devices always being the same as host.
> - changed TARGET_{LLC,CLANG}_FLAGS into
> {CELL,TCE}_TARGET_{LLC,CLANG}_FLAGS - seems more consistent
> wrt >1 ocl "targets"
I think there is no point in dragging the CellSPU device along anymore.
Its pretty dead by now, and should be removed before pocl 0.10 release,
IMHO.
> Looking at the last item (SIZEOF), seems only used in
> lib/CL/devices/pthread/pthread.c to enable/disable half and double
> extensions...
This was due to clang reporting sizeof(long) as 4 on 32 bit platforms,
even when compiling OpenCL-C (where it is defined as 8). This error then
cascaded to the handling of doubles. Haven't checked if the bug is
fixed, but please leave this check in, or the ARM build will fail at
least on older LLVM/Clangs.
kalle
------------------------------------------------------------------------------
HPCC Systems Open Source Big Data Platform from LexisNexis Risk Solutions
Find What Matters Most in Your Big Data with HPCC Systems
Open Source. Fast. Scalable. Simple. Ideal for Dirty Data.
Leverages Graph Analysis for Fast Processing & Easy Data Exploration
http://p.sf.net/sfu/hpccsystems
_______________________________________________
pocl-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/pocl-devel
