To: Larry Brew who wrote (2715 ) 1/4/1998 7:37:00 PM From: pat mudge Read Replies (1) | Respond to of 6180
[Broncos and DSP reportcard] Well, Larry, you got your wish on the Broncos. Now, I'll expect $500/share by the end of the year. :)) Getting back to business, here's a C6X report card from last June:ti.com 'C6x Report Card An evaluation of the performance of TI's 'C6201 processor and compiler Submitted by Loughborough Sound Images Editor's Note: Loughborough Sound Images supplies board-level products based on DSP processors. This document presents results LSI achieved in evaluations of TI's TMS320C6x DSP family and a study of the quality of the development tools used. The complete paper is available at: lsi-dsp.co.uk . The TMS320C6x is the latest digital signal processor (DSP) from Texas Instruments. It represents their ninth generation of DSP and is a breakthrough in terms of system performance. Offering up to 1600 RISC-style MIPS, it is an order of magnitude faster than current-generation DSPs. A comparative evaluation to assess the performance of compiled C code on the 'C6x, TMS320C4x and SHARCT processors was undertaken. This investigation centered around two tests, a 2D convolution and a Fast Fourier Transform (FFT) benchmark. With no optimization of the original C code and using 32 bit data, the 'C6x quite significantly out performs both of the other processors in the FFT benchmark. Its performance is further enhanced by subsequent optimization of the C code, the enhancement being more pronounced when using 16 bit data. In some cases, the 'C6x compiler produced code that is over 95 percent efficient compared to theoretical limits. In the convolution benchmark, using optimized C code, the 'C6x considerably outperforms the SHARC. For a 3x3 kernel on a 64x64 array, the 'C6x performed the convolution over 30 times faster than the SHARC processor. The 'C6x compiler produces code that is over three times more efficient than the equivalent from the SHARC or 'C4x compilers on the convolution benchmark. >>>>