Combine CS and EE to design and implement an optimized computing system.

• the processor gets instructions and data from the memory
• input writes data to the memory
• output reads data from the memory
• control sends the signal that determind the operations of the…
• datapath, which does computations

$\textrm{control} + \textrm{datapath} = \textrm{processor}$

History §

Babbage §

• diff engine
  • George Scheutz
• analytical engine
  • Scheutz

Binary §

Unsigned binary integers §

$x = x_{n - 1} 2^{n - 1} + x_{n - 2} 2^{n - 2} + \cdots + x_0 2^0$

Range: $0$ to $2^n - 1$

2’s complement signed integers §

$x = -x_{n - 1}2^{n - 1} + x_{n-2} 2^{n-2} + \cdots + x_0 2^0$

MSB is the sign bit.

Range: $-2^{n-1}$ to $+2^{n-1} - 1$

• $0 = {0000 \cdots 0000}_2$
• $-1 = {1111 \cdots 1111}_2$
• Most positive: ${0111 \cdots 1111}_2$
• Most negative: ${1000 \cdots 0000}_2$

Overflow §

Sum requires more bits than the input data width.

Sign extension §

move 2’s complement signed int from smaller to larger container: replicate the sign bit to the left

Floating Points §

• normalized
• not normalized

IEEE 754-1985 standard.

Types:

• single precision — 32 bits
• double precision — 64 bits

Components:

• sign (1 bit)

• exponent (8 vs 11 bits)

  implicit bias 😏 (127 vs 1023)

• fraction (23 vs 52 bits)

$x = (-1)^{\mathrm{sign}} \times (1 + \mathrm{fraction}) \times 2^{\mathrm{exponent} - \mathrm{bias}}$

special cases

• $\mathrm{exponent} = 0 \cdots 0$

  $x = (-1)^\mathrm{sign} \times (0 + \mathrm{fraction}) \times 2^{-\textrm{bias}}$

  • $\mathrm{fraction} \neq 0\cdots 0$

    subnormal/denormal numbers.

  • $\mathrm{fraction} = 0 \cdots 0$

    literally $\pm 0$

• $\mathrm{exponent} = 1 \cdots 1$

  • $\mathrm{fraction} \neq 0 \cdots 0$

    NaN aka Not a Number.

  • $\mathrm{fraction} = 0 \cdots 0$

    $\pm \infty$

Memory §

Byte ordering §

• Big Endian — LSB has highest address
• Little Endian — LSB has lowest address

Example (0x07AB9DCE):

Addresses, data.

• $2^{10}$: Kibibyte (KiB)

• $2^{20}$: Mebibyte (Mib)

• …

• Microprocessor

  CPU components. No memory, no I/O.

• Microcomputer

  Has memory, some I/O.

• Microcontroller

  Many types of memory, sophisticated I/O.

TM4C123GH6PM §

Memory:

• Flash 256 KiB
• SRAM 32 KiB
• ROM

C §

Pointers §

components:

• value: address location
• type: the type of the data at the stored address

1. read operands only from the register file
2. write operations only to the register file
3. only one way to move data between the register file and data memory

• data addresses
• instruction addresses

Bitwise operators §

• ~a — not

• a & b — and

  x & 0 = 0; x & 1 = x. clear

• a | b — or

  x | 0 = x; x | 1 = 1. set

• a ^ b — xor

  x ^ 0 = x; x ^ 1 = ~x

• a >> n — right shift

• a << n — left shift

RISC V §

Reduced Instruction Set Computing.

RISC-V has 32 registers. x0 through x31. Have no type.

R-format — register-register arithmetic §

I-format — register-immediate arithmetic §

S-format — stores §

B-format — branches §

U-format — 20-bit upper immediate §

J-format — jumps §

Instructions §

• add

• lw — load word

  lw dest,offset(source)
  

• sw — store word

  sw dest,offset(source)
  

• lb — load byte

  lb dest,offset(source)
  

• sb — store byte

  sb dest,offset(source)
  

• slli — shift left (logical)

  slli dest,source,bits
  

• srai — shift right (arithmetic)

• beq — branch (if) equal

• bne — branch (if) not equal

• blt — branch (if) less than

• bge — branch (if) less than

• j — jump

Analog to Digital §

input: continous time signal

output: discrete sequence

Logic Design §

Demultiplexer

Multiplexer

Canonical forms:

• POS
• SOP

Transistors:

• n
• p

Cache §

memory hierarchy.

trade-off between speed and price/size. small, fast vs large, slow.

locality:

• temporal locality
• spatial locality

types:

• direct-mapped cache
• set associative cache

$C = S \times E \times B$ data bytes

• $S = 2^s$
• $E = 2^e$. when direct-mapped, $e = 1$
• $B = 2^b$

Word length: $t + s + b$