Products
Accelerating C Code, No Assembly Required
Product Briefs
White Papers
The What, Why and How of Configurable Processors
How to Increase ASICs and SOC Computational Performance with Long-Word Processors
Processor Ports and Queues: Easily Overcome I/O Bandwidth Obstacles in Your Next ASIC or SOC Design
Videos
Processor Configuration with Chris Rowen
Accelerating Existing C Code Using TIE
One of the reasons for resorting to assembly language use is because conventional, fixed-ISA processors often cannot achieve performance goals using C or C++ alone. However, even assembly code is limited to the native performance of the fixed-ISA processor. Tensilica’s Xtensa processor is extensible. This extensibility allows the processor’s resources and capabilities to be tailored to specific tasks.
Program optimization using TIE (Tensilica’s Instruction Extension language for the Xtensa processor) is first accomplished by identifying hot spots in the C code – performance-sensitive regions of application code usually residing in the inner loops of algorithms. The execution profiler allows a system designer to quickly analyze code efficiency and to identify where custom instructions (implemented with TIE) can best enhance performance. The software /firmware team or the system designer then defines and implements one or more new TIE instructions to accelerate the hot-spot code.
Stay in C, No Assembly Coding Required
The new TIE instructions appear as intrinsic function calls in C-code routines. The need for assembly coding disappears. Xtensa compilers automatically perform all the necessary optimizations, register allocations, and scheduling of the assembly code generated by the compilation. To get more performance, the system designer can rebuild the processor with additional TIE instructions and re-profile the application to see the performance gain.
The designer can iteratively develop a sequence of processor builds while profiling each new tailored processor to weigh the benefit of adding instructions and pipeline-acceleration hardware through the TIE language. Aggressive use of function-unit parallelism and other acceleration techniques (such as overlapping loads with instruction execution) can often deliver 10X, 100X, or even greater performance increases.
For general information on Xtensa software development tools, click here
For more information on specific software development tools available for the Xtensa processor, click on the links below:
