[dpdk-dev] [PATCH v3 00/10] split architecture specific operations
chaozhu at linux.vnet.ibm.com
Wed Nov 5 03:39:56 CET 2014
> The set of patches split x86 architecture specific operations from DPDK and put
> them to x86 arch directory.
> This will make the adoption of DPDK much easier on other computer architecture.
> For a new architecture, just add an architecture specific directory and
> necessary building configuration files, then DPDK eal library can support it.
> Reviewing patchset from Chao, I ended up modifying it along the way,
> so here is a new iteration of this patchset.
> Changes since Chao v2 patchset :
> - added a preliminary patch for moving rte_atomic.h (for better readability)
> - fixed documentation generation
> - implemented a generic header for each arch specific header (cpuflags, memcpy,
> prefetch were missing)
> - removed C++ stuff from generic headers
> - centralised all doxygen stuff in generic headers (no need to have duplicates)
> - refactored rte_cycles functions
> - moved vmware tsc stuff to arch rte_cycles.h headers
> - finished x86 factorisation
> Little summary of current state :
> - all applications continue to include the eal headers as before, these headers
> are the arch-specific ones
> - the arch specific headers always include the generic ones. The generic headers
> contain the doxygen documentation and code common to all architectures
> - a x86 architecture has been defined which handles both 32bits and 64bits
> It builds fine for 32/64 bits (w and w/o "force intrinsics"), but I really would
> like a lot of eyes on this (and I would say, especially, rte_cycles, rte_memcpy
> and rte_cpuflags).
> I still have some concerns about the use of intrinsics for architecture != x86
> but I think Chao will be the best to look at this.
Acked-by: Chao Zhu <bjzhuc at cn.ibm.com>
More information about the dev