Intel Software Conference 2015 – c’t

H igh Performance Computing, so Intel’s “chief evangelist” James Reinders in his keynote address to the now tenth ISTEP Conference in Europe, is a focus, the Intel addressed not only to the specially designated Xeon Phi, but also with compilers and tools. These offer many assistance in the parallelization and vectorization. Soon after all wait here Xeon Phi Knights Landing processors AVX512 and up to 288 logical cores.

But besides HPC are there are now also other important areas, for the Intel Software offers holds – partly free, with costs partially. These include the areas of multimedia and mobile apps.

Before Intel introduced its new products more accurately, were Janel Gavin, founder and head of Evans Data Corporation (EDC), an overview of what your Corporation has specialized on the market for software development. And who sees in Europe for developers after their presentation quite well. EDC sees 24 per cent growth in the next four years to 2019. North America, however, remains far behind with 11 percent. Leader is not surprising, the Asian continent including Pacific appendage with a projected increase of 53 percent. The largest developer crowd in Europe is currently working in Russia with around 890 thousand, followed by Germany (458 thousand) and the UK (438 thousand).

The often maligned Windows remains clear as development environment with more than 80 percent of the first choice, way ahead of Linux with 8 or OS X with 4 percent. Desktop PCs are still the most common target, indicated 64 percent of the developers surveyed. The boom of mobile apps is a little subsided. In Europe, about the proportion of those who develop mobile apps, stops at 40 percent for years. In other regions it is higher but also stagnant. Android as a target, as Evans loses lately, iOS have overtaken it again. As a programming language Java is just ahead of C ++ at the top, C #, which once came up almost to the top two programming languages ​​in 2009, is further dropped.

parallel study in 2016

A new, highly practical guide: the Vectorization Advisor

Intel Parallel Studio XE Edition 2016 can now already be downloaded as a beta. It includes C ++ – and Fortran compilers, all libraries and tools such as the Math Kernel Library (MKL 11.3), Threading Buildings Blocks (TBB 4.4) Integrated Performance Primitives (IPP 9.0) and in the Professional Edition also Inspector, Advisor and Amplifier. A new feature is the Intel Data Analytic Acceleration Library (DAAL) with an extensive set of features as they need the data analysts. Another new feature is a Vectorization Adviser, the clear all the loops and their Vektorisierungsmöglichkeiten lists and gives useful advice. As usual, there are also still the Cluster Edition, which additionally contains MPI and Cluster Tools

The C ++ -. Compiler in the next edition to support the full extent OpenMP 4.0, albeit with smaller restrictions on the data types for User Defined Reductions (only “Plain Old Data”, no SIMD). But added some extensions already under OpenMP 4.1. Also new are pragmas that simplify the tiling or blocking, as #pragma block_loop factor (250) Level (2) . This avoids the so far usual unsightly decomposing the loops “on foot”.

For more beauty and the feature “Annotated Source Listing provides ( ASL), “by which the compiler its Optimization Reports lays clearly in a copy of the source code (as text or HTML).

Cilk wakes up

Intel C ++ – extension for parallel programming, Cilk Plus, has been extended to SIMD capability. Improves also was feasible with the Cilk cooperation between CPU and HD graphics (GFX). With upcoming 48 execution units, each with 7 hardware threads, using the GFX processing power is becoming more interesting for normal computing, with at least up to 845 gigaflops (single precision) at the Iris 6100, it offers in this area more power <"possible_gallery asset_position images" div class = id = "asset_26241">

than 8 Haswell cores.

parallelize loops with Cilk and access Intel HD graphics – nothing easier than that

<. / div>

The GFX-integration goes via OpenCL, but with Cilk that is orders of magnitude easier. To date, the community of OpenSouce transmitted Cilk has more gefristet a wallflower. But presented with the now gcc 5.1, at the Cilk is one of the most important new features that may soon be different. OpenMP is indeed far more flexible – but Cilk is where you can use it far simpler and more elegant. Simply add the pragma offload target (gfx) and pin () designate the addresses shared by CPU and GFX – that’s it , The rest is Cilk. Data transfers as in the PCIe cards are thanks to the shared memory space is not needed, the offloading then often pays even with small amounts of data.

CPU and GFX together provide an up to six times more computing power.

The pretty little picture Sphereflake-you can then on a Core i7 let 4770 3.5 GHz with cleverly chosen division into 20 percent CPU and 80 percent GFX-load by up to a factor of 6 to run faster.

Also OpenMP 4.0 now provides support for offload HD Graphics: – qopenmp offload = GFX and Speicherpinning #pragma omp target map ()

To date could you do not use a common virtual memory that is to Version 2016 change (not in the current beta).

Enrolled students can not apply commercial use now for free C ++ – Download compilers and tools for Windows and OS X, not only the free C ++ – Linux version as before. Anyone actively involved in open source projects, the Linux-suites also gets nothing, as academics – unless they have a seat in the US. For teachers there are special conditions.

for the embedded area, the system is intended Studio 2015 and for Linux, Android, Windows – limited to C ++ – Compiler and IPP library – also VxWorks. Otherwise, there are also here the three versions Composer, Professional and Ultimate with C ++ – compilers and different libraries, analyzers and debuggers. Sorry, no math library MKL for Android here.

For ARM

Intel’s cross-platform project “Integrated Native Developer Experience (INDE),” is now out of the beta phase. It provides C ++ – and Java tools and libraries for Android and Windows. <"Possible_gallery asset_position images" div class = id = "asset_1"> The special thing about it is that it is supported on Android not only Intel hardware, but to a limited extent, ARM.

The Intel compiler, debugger and the Platform Analyzer missing, otherwise you can take the free Home Edition of INDE already start something.