4.9 Comparative Architectures

4.9 Comparative Architectures

Please note that all of the processors have a common subset of instructions. ADD, AND, OR, SUB, and XOR. All processors can load and store values to memory, and use constants. They may have other instructions depending on the hardware support: MUL, CALL, RET, and DIV.

4.9.1 Web Sites

1. Advanced Micro Devices, AMD Opteron^TM (NUMA).
2. Analog Devices DSP.
3. Intel Pentium^TM.
4. IBM RISC.
5. Microchip Embedded & DSP.
6. Freescale Embedded & DSP.
7. Texas Instruments Embedded & DSP.

4.9.2 Notes - Processor types

• CISC processors
  This is what we have been programming on all semester. The Intel Pentium^TM and the AMD Athlon^TM processors are good examples.
• RISC processors
  IBM/Apple power processors IBM PowerPC^TM 750GX-750FX
• DSP processors
  The primary feature of this architecture is the ability to do a single cycle multiply accumulate for Finite Impulse Response (FIR) filters. It may have several other features including a reverse carry to assist Fast Fourier Transforms (FFT). Texas Instruments (TMS320C62x), Freescale(DSP563XX), and Analog Devices (ADSP-BF561 - Blackfin^TM) all offer Digital Signal Processors. Microchip has a low end, low power DSP (dsPIC30Fxxx) in their current product line.
• Embedded systems processors.
  We will look at the PIC10F200 series of processors from Microchip. Theses come in a 6 pin package.
• Dual/multi-processor/Dual core/Hyperthreading^TM (Intel) systems -- SMP
• Dual/multi-processor systems -- NUMA
    Opteron^TM from AMD. Non-Uniform Memory Access.

4.9.3 Notes - Performance issues

• Harvard architecture
• Symmetric multiprocessing, (SMP) bottlenecks.
• Cache issues
  • Single cache
  • Instruction/Data cache
  • Bus snooping
  • Translation Look-aside Buffer, TLB.
  • Branch cache
• Speculative execution
• Single Instruction, Multiple Data, (SIMD)
• Vector processing
• Multiple execution units
• Pipelining & pipeline flushes.

4.9 Comparative Architectures