Open shamisp opened 10 years ago
There is no real pack/unpack in the lmi implementation. Since we are using the key/value store to send data around, I would assume that the PMI implementation takes care of data integrity. However, I fixed the unpack to memcpy the data into the user buffer and to return the offset.
As to unpacking multiple rte_iovecs, do we want to change rte_unpack or introduce an additional function?
PMI - this is not issue of integrity but support for heterogeneous system. PMI doesn't provide this. Since the datatypes are basic, probably host-to-network translation has to do the trick.
rte_unpack - seems like it is much more easier to extend existing API ?
Also, probably we have to define clarify the documentation to make sure that user allocate memory properly for the unpack destination
So the proposed change would look something like this: int rte_unpack (rte_iovec_t *data, int iov_count, void src, uint32_t *offset)?
Looks good.
On Dec 5, 2013, at 10:41 AM, Swen Boehm notifications@github.com<mailto:notifications@github.com> wrote:
So the proposed change would look something like this: int rte_unpack (rte_iovec_t *data, int iov_count, void src, uint32_t *offset)?
— Reply to this email directly or view it on GitHubhttps://github.com/uccs/librte/issues/7#issuecomment-29908148.