Clarification on the Capabilities of Xeon Phi

[Joe_C_5]

I have purchased 3x Xeon Phi 3120P.
Now I realize they are totally useless without the compiler, which is another hidden cost of 700 USD. Then, what I am wondering is whether I will have the same regret once I purchase the compilers as well. This is my situation:
- I have an application that I have build myself in C++ and that I want to run as a native application on the device. However, the application has many dependencies from other libraries, who are used as such, i.e. the shared libraries are already compiled either for 32-bit or 64-bit x86 architecture. I dont have their source code to recompile them for the MIC. My biggest worry is whether I can port those libraries on the XEON PHI?

Most of this libraries are socket communication interfaces for a database data pull. The database is located on the host, I pull data from the host, process the data, and retrieve back a result to the database. I repeat the same thing in a cycle. They do have plenty of pointers and structs. Please explain to me how can I port that application to the XEON PHI? My application works perfectly with MPI on the CPU, but I need greater level of parallelism, hence I use the PHI. It has more cores than my CPUs. However, I am already dissapointed to hear that you are selling a device without the compilers. What I am worried to hear that there will be limitations on the portability too, only to realize NVIDIA GPU is far better choice. Unless I am mistaking?!

I am looking to hear from you, as soon as possible.

[TimP]

https://software.intel.com/sites/default/files/650_Intel_R__Xeon_Phi_tm__Cluster_configuration-v081.pdf

is a valuable reference on porting libraries which can be built with the gcc which comes for free with the MPSS.  I don't know whether Intel has a public site where the pre-built libraries referred to there may be downloaded.

Many of those libraries could also be built with the Intel compilers; where they don't provide for cross-compilation, you may find the setup awkward.

More complete gcc support appears to be coming for the future KNL version of Intel(r) Xeon Phi(tm).

[Joe_C_5]

Tim, are you saying that Intel or gcc cannot cross-compile those already compiled shared libraries?

[Joe_C_5]

Also, I am not starting with a cluster. Lets say, I want to port the application to one of the Phis. How do I proceed with the porting, with gcc? Can you give me reference? Intel support on this is total crap.

[TimP]

Joe C. wrote:

Tim, are you saying that Intel or gcc cannot cross-compile those already compiled shared libraries?

Yes, you are pretty much dead for porting to KNC if you don't have source code.  KNL will offer some x86_64 instruction set compatibility, at a price in performance.

I don't see how you intended to link against x86 compiled code and run on Nvidia GPU.