Q-NOTE QN-7000HX Technical Information Page 1

XAPP1206 v1.1 June 12, 2014 www.xilinx.com 1© Copyright 2014 Xilinx, Inc. Xilinx, the Xilinx logo, Artix, ISE, Kintex, Spartan, Virtex, Vivado, Zynq,

Software Performance Optimization MethodsXAPP1206 v1.1 June 12, 2014 www.xilinx.com 10All techniques for programing with NEON listed above are discus

Software Performance Optimization MethodsXAPP1206 v1.1 June 12, 2014 www.xilinx.com 11In addition to these open source libraries, NEON is also popula

Software Performance Optimization MethodsXAPP1206 v1.1 June 12, 2014 www.xilinx.com 12code from standard C or C++. However, in practice, this optimiz

Software Performance Optimization MethodsXAPP1206 v1.1 June 12, 2014 www.xilinx.com 13Figure 4 shows the setting optimization flags.The compiler migh

Software Performance Optimization MethodsXAPP1206 v1.1 June 12, 2014 www.xilinx.com 14• Use suitable data typesAn example of a standard dot product a

Software Performance Optimization MethodsXAPP1206 v1.1 June 12, 2014 www.xilinx.com 15Using the techniques above, you can modify the C source code to

Software Performance Optimization MethodsXAPP1206 v1.1 June 12, 2014 www.xilinx.com 16produce a result different from non-NEON optimized code for flo

Software Performance Optimization MethodsXAPP1206 v1.1 June 12, 2014 www.xilinx.com 17Tab l e 6 can give developers some basic ideas about NEON types

Software Performance Optimization MethodsXAPP1206 v1.1 June 12, 2014 www.xilinx.com 18Accessing Two D-registers of a Q-registerThis can be done using

Software Performance Optimization MethodsXAPP1206 v1.1 June 12, 2014 www.xilinx.com 19The disadvantage is obvious. First, it is difficult to maintain

Software Examples and LabsXAPP1206 v1.1 June 12, 2014 www.xilinx.com 2only options. So, in some situations, you might need to re-write time-critical

Boost NEON Performance by Improving Memory Access EfficiencyXAPP1206 v1.1 June 12, 2014 www.xilinx.com 20Therefore, when writing assembly instruction

Boost NEON Performance by Improving Memory Access EfficiencyXAPP1206 v1.1 June 12, 2014 www.xilinx.com 21Generally, loading and storing multiple inst

Boost NEON Performance by Improving Memory Access EfficiencyXAPP1206 v1.1 June 12, 2014 www.xilinx.com 22A graphical demonstration of VST3 is shown i

Boost NEON Performance by Improving Memory Access EfficiencyXAPP1206 v1.1 June 12, 2014 www.xilinx.com 23The algorithm used is still the dot product

Boost NEON Performance by Improving Memory Access EfficiencyXAPP1206 v1.1 June 12, 2014 www.xilinx.com 24this is at line 67 of the source file benchm

Boost NEON Performance by Improving Memory Access EfficiencyXAPP1206 v1.1 June 12, 2014 www.xilinx.com 25In Figure 6, you can see the 16-byte arrays

Boost NEON Performance by Improving Memory Access EfficiencyXAPP1206 v1.1 June 12, 2014 www.xilinx.com 26Fully associative cache can solve this issue

SummaryXAPP1206 v1.1 June 12, 2014 www.xilinx.com 27ii < 8; ii++, rresult += N, ra += N) for (ki = 0, rb = &b[ko][jo];ki < 8

Revision HistoryXAPP1206 v1.1 June 12, 2014 www.xilinx.com 287. Cortex™-A Series Programmer's Guidesilver.arm.com/download/download.tm?pv=129601

Before You Begin: Important ConceptsXAPP1206 v1.1 June 12, 2014 www.xilinx.com 3Before You Begin: Important ConceptsBefore addressing specific optimi

Before You Begin: Important ConceptsXAPP1206 v1.1 June 12, 2014 www.xilinx.com 4Figure 1 shows the single core Cortex-A9 processor block diagram.One

Before You Begin: Important ConceptsXAPP1206 v1.1 June 12, 2014 www.xilinx.com 5number can be different from the overall performance. The best way to

Before You Begin: Important ConceptsXAPP1206 v1.1 June 12, 2014 www.xilinx.com 6cores, and one private timer (32-bit) for each core. The timers have

Before You Begin: Important ConceptsXAPP1206 v1.1 June 12, 2014 www.xilinx.com 7Figure 2 compares SIMD parallel add with 32-bit scalar add.To achieve

Before You Begin: Important ConceptsXAPP1206 v1.1 June 12, 2014 www.xilinx.com 8Cortex-A9 processor. For floating-point operation, VFP can support bo

Before You Begin: Important ConceptsXAPP1206 v1.1 June 12, 2014 www.xilinx.com 9NEON Performance LimitWhen using NEON to optimize software algorithms