Computing an inner product is a very common procedure

Info iconThis preview shows page 1. Sign up to view the full content.

View Full Document Right Arrow Icon
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: cles, of which 4 are loop overhead (decrementing the loop counter and branching back at the end of the loop). Each loop computes two products, so each product requires 5 cycles. With loop unrolling (replicating the code to compute many more products in a single loop iteration) the cost for each product reduces towards 3 cycles. The best an ARM9TDMI can achieve is one product in 10 cycles, over three times slower. The difference is accounted for partly by the slower multiplier on the ARM9TDMI, partly by its less efficient handling of 16-bit operands, and the remainder by the extra instructions required to test and correct for saturation. 8.10 Example and exercises Estimate the proportion of the number of test vectors required to test an ARM core via the JTAG and AMBA interfaces. An ARM core has in the region of 100 interface connections (32 data, 32 address, control, clock, bus, mode, and so on). The JTAG interface is serial. If the tester allows one vector to specify a pulse on TCK it will take 100 vectors to apply a parallel pattern to the ARM core. The AMBA test interface accesses the ARM periphery in five sections (see 'Test interface' on page 219), requiring five vectors on a standard tester. The JTAG interface therefore appears to require 20 times the number of vectors. (Remember that JTAG is intended for PCB testing, not production VLSI testing.) In fact both the JTAG-based EmbeddedlCE and AMBA interfaces include optimizations to improve the efficiency of getting instructions into the ARM core, which is the dominant requirement in testing it, but a very detailed analysis would be required to take these into account in the estimate. Summarize the problem areas in the production VLSI testing of complex macrocellbased system chips and discuss the relative merits of the various solutions. Describe and differentiate between production VLSI testing, printed circuit board testing and system debugging, and describe how a JTAG test port may be used to address each of these. Where is the JTAG approach most effective and where is it least effective? Example 8.1 Exercise 8.1.1 Exercise 8.1.2 246 Architectural Support for System Development Exercise 8.1.3 What problem do...
View Full Document

This document was uploaded on 10/30/2011 for the course CSE 378 380 at SUNY Buffalo.

Ask a homework question - tutors are online