Data Manipulation-計算機概論筆記

一定愛配于天立教授開放課程食用:課程連結

Data Manipulation

The communication between CPU and memory

Register -> CPU Cache

• Adding values in memory
  1. Get value A from the memory and put it in a register.
  2. Get value B from the memory and put it in another register.
  3. Activate the additional circuit and put the final value into another register.
  4. Store the result the memory.
• Machine instruction
  1. Data transfer
     • load, store, I/O
  2. Arithmetic/logic
     • and, or, add, sub...
     • shift, halt
  3. Control
     • jump, halt
  4. RISC -> Reduced instruction set computing
     • Ex: PRC, SPRC
  5. CISC -> Complex instruction set computing
     • Ex: x86，x64

Example of machine instruction

• 3 -> store
• 5 -> register No.5
• A7 -> memory cell A7
• all in all -> Store the content of register No.5 to memory cell A7.

The machine instruction of adding two values

Program execution

• Machine cycle

• Fetch

• Clock -> how many machine cycle a cpu is able to do in one second

Arithmetic and logic unit(ALU)

• Arithmetic operations
• Logic/bit operations

Shift/Rotation

• Logic shift
  • 10100000 -> 01010000(right)
  • 10100000 -> 01000000(left)
  • Both moving left and right add a 0
• Arithmetic shift

  • 10100000 -> 11010000(right, the number is divided by 2) Moving right，copy the leftest bit.

  • 10100000 -> 11000000(left, the number is multiplied by 2) Moving left，add a 0 to the right.

• Rotation
  • 10100000 -> 01010000(right)
  • 10100000 -> 01000001(left)

Controllers and peripheral devices

• Specialized -> SATA...
• General -> USB, HDMI...

Memory-mapped I/O

• The periphral devices are recognized as part of the memory.

Communication

• DMA(direct memory access)
  • Authorized controllers can access data directly from main memory without notifying CPU
• Hand shaking
  • Two way communication
  • Coordinating activities
• Parallel/Serial
• Transfer rate unit -> bit per second

Pipelining(accelerating the process)

ex -> pre-fetching

• Throughput -> Total amount of work accomplished in a given amount of time
• Pre-fetching issus -> conditional jump

Parallel/distributed computing

• Parallel
  • Multiprocessor
  • MIMD，SISD，SIMD，MISD(M->Multiple，S->Single，I->Instruction, D->Dataset)
• Distributed
  • Linking computers via network
  • Seperate processors and memory
• Issues
  • Data dependency
  • Load balancing
  • Synchronization
  • Reliability

How to parallelize?

Speed up and Scaling

• Amdahl's law(阿姆達爾定律)
• Gain = (P -> proportion that is parallelizable, S -> proportion that is serial only)