ARM Design Philosophy

  • Physical features

    • embedded systems require battery power

      • ARM is designed to be small to reduce power consumption and extend battery operation

  • High code density is another major require

    • embedded systems have limited memory due to cost and/or size restrictions

      • high code density is useful for apps that have limited on-board memory, e.g. mobile devices or mass storage

  • ARM has incorporated hardware debug tech, allowing engineers to view what is happening while the processor executing code

  • ARM core is not pure RISC because of the constriants of embedded systems

    • often viewed positively bc ARM doesn't take RISC concepts too far.

Variable cycle execution for certain instructions - Not all ARM instructions execute in one cycle - load-store-multiple instructions vary in number of cycles, which depends on number of registers transferred - Transfers can occur in sequential memory addresses - increases performnace -> sequenctial access is faster than random - code density is also improved since multiple register transfers are common at beginning/end of function


Code density (Also called: byte efficiency, code efficiency, operation length reduction) - refers to all of the instructions needed to completed a particular task - whether or not the majority of instructions fit into instruction cache, has major implications on performance - CISC introduced more powerful instructions to the ISA, requiring fewer total instructions required to complete common tasks - This generally improved code density - RISC saught to eliminate complex instructions by breaking them into simple ones that can exec in 1 cycle - RISC prioritized other CPU performance metrics over code density - Because CISC uses Variable length instructions (variable-length opcode) commonly used instructions should be short - analogous to the 80/20 rule - a small number of instructions from the ISA are used for the majority of a workload. - RISC uses fixed-length (or fixed-width) instructions. It's known that 16-bit optimizes code density for fixed-width instructions, compared to 8 and 32-bit width


  • Inline barrel shifter

    • (hardware component) preprocesses one of the input registers before it is used by an instruction

      • improves core performance and code density by expanding the cability of many instructions

        • SEE ch 2, 3, 4 (ARM System Developer's Guide)

*** VERY IMPORTANT ***

  • THUMB ISA

    • 16-bit fixed length ISA

    • An enhancement to preprocessor core

      • an addition the the original ISA

    • Permits ARM core to execute 16 or 32-bit instructions

      • 30% optimization to code density

  • Conditional execution

    • instruction only executed when specific condition is met

  • Enchanced instructions

    • DSP (digital signal processor) instructions

      • allows faster-performing ARM processor in cases permitting the replacement of traditional combinations of processor + DSP

        • supports 16x16-bit multiplier operations and saturation