ECEN 350 Exam 1, Prof Mi Lu

Compiler

High-level language → Assembly

Assembler

Assembly → Machine language

5 classic components of a computer

input, output, memory, datapath, control

Memory

where programs and data are kept

datapath

performs arithmetic operations

control

tells datapath, memory, and i/o devices what to do according to program instructions

processor

active part of computer. adds numbers, tests numbers, signal i/o.

datapath and control combined

also called cpu

DRAM

dynamic random access memory.

takes same amount of time to access the memory no matter what portion of memory is read

cache

built using sram - static ram.

consists of small, fast memory that acts as buffer for DRAM

implementation

hardware that obeys the architecture design of principle technique

instruction set architecture

includes anything programmers need to know to make a binary machine language program work correctly, including instructions, i/o devices, etc.

integrated circuit

combines dozens to millions of transistors into a single chip

vlsi

very large scale integrated circuit, integrates hundreds of thousands to millions of transistors into a single chip

manufacturing of a chip

silicon crystal ingot → sliced into wafers → chopped into dies

defect

microscopic flaw in wafer.

any defect can cause failure

true

why chop wafer into dies?

allows discarding only defects rather than entire wafer

yield

percentage of good dies from total dies on wafer

each wafer same price, meaning fewer dies →

higher cost per die

each wafer same price, meaning

2x yield → half the cost per die

cost per die =

cost of wafer / (dies per wafer * die yield)

dies per wafer =

wafer area / die area

die yield =

(1 + (defects per unit area * die area) / alpha) ^ -alpha

what is alpha in die yield equation?

number of critical processing steps in manufacturing process. eg. MOS: alpha = 2

cpu execution time_1 =

cpu clock cycles * clock cycle time

cpu execution time definition

actual time the cpu spends computing for a specific task

cpu (execution) time_2 =

instruction count * cpi * clock cycle time

clock cycle time =

1/clock rate

performance =

1/execution time

A is n times as fast as B, n(performance) =

performace(A)/performance(B)

A is n times as fast as B, n(execution time) =

execution time(B)/execution time(A)

MIPS definition

millions of instructions per second

MIPS_1 =

instruction count / (execution time * 10^6)

MIPS_2 =

clock rate / (cpi * 10^6)

issues with MIPS

1. doesnt account for different instruction sets

2. varies between programs on same computer

3. if more instructions, but each instruction faster, mips increases but program may run slower

MFLOPS =

number floating point operations / (execution time * 10^6)

MFLOPS definition

millions of floating point operations per second

average execution time =

1/n * sum(execution time)

weighted arithmetic mean =

sum(w*execution time)

geometric mean =

nth root(product( execution time / execution time reference))

when to use geometric mean?

when averaging speedups

computer design principles

1. make common case fast

2. amdahl’s law

amdahl’s law =

exec time after improvement = execution time affected / amount of improvement + exec time unaffected

how many registers?

32

how many bits are instructions?

32

how many bits in each register?

64

how many bytes in memory space?

2^64

how many bits in a byte?

8

how many bits in a halfword

16

how many bits in a word?

32

how many bits in a doubleword?

64

arm is byte addressable

true

little endian

least significant byte at lowest address

instructions must be aligned to multiple of

4

B and CB type instruction branch offsets are

PC relative

PC relative

number of instructions from current instruction to label. + → foward, - → backward (before multiplying by 4)

assembly order of arguments is _ of machine code order

reverse

R type stands for

Register format

R type bit allocation

11:opcode, 5:second source, 6:shift amount, 5: first source, 5: destinition

R type examples

ADD, ORR, LSL, BR

I type stands for

Immediate arithmetic

I type bit allocation

10: opcode, 12: immediate, 5: source, 5: destination

I type examples

ADDI, ANDI

D type stands for

Data

D type bit allocation

11: opcode, 9: offset (signed), 2: op2, 5: base address, 5: load/store reg

D type examples

LDUR, STUR, LDURW, LDURSB

B type stands for

unconditional branch

B type bit allocation

6:opcode, 26: branch offset

B type examples

B, BL

CB type stands for

conditional branch

CB type bit allocation

8: opcode, 19: branch offset, 5: tested reg.

CB type examples

CBZ, CBNZ

describe MOVK

moves 16 bit immediate into destination register, can be LSL by multiple of 16, keeps other bits unchanged

describe MOVZ

moves 16 bit immediate into destination register, can be LSL by multiple of 16, zeros other bits

AND, ORR, EOR, ANDI, ORRI, EORI, etc. are bitwise operations

true

PC stands for what and explain what PC is

program counter, address of instruction to be executed, incremented by 4 after each execution, written over during branch instruction

which direction and operation are performed on stack pointer when allocating additional memory

sp growns down, SUB performed

what register is stack pointer SP in?

X28

what register is frame pointer FP in?

X29

what register is link register LR in?

X30

what register is the zero constant XZR in?

X31

For LDUR operations only, what is suffix to sign extend instead of default zero extend?

S

for LDUR and STUR operations, what is suffix to load/store a byte

B

for LDUR and STUR operations, what is suffix to load/store a halfword

H

for LDUR and STUR operations, what is suffix to load/store a word

W

for LDUR and STUR operations, what is suffix to load/store a doubleword

nothing

immediate addressing definition

operand is a constant within instruction

register addressing definition

operand is a register

base or displacement addressing definition

operand is at the memory location whose address is the sum of a register and a constant in instruction

PC-relative addressing definition

branch address is sum of PC and constant in instruction*4

how to get representation of a negative number in 2’s comp?

1. write as binary

2. invert all bits

3. add 1

how to extending numbers in 2’s comp representation?

extend sign bit

how to detect overflow?

when adding same sign, if result is different sign, overflow occured

how to do integer multiplication?

multiplicand * multipler

start: 0…0 | multiplier

1. check LSB of Product/Multiplier register

   1. if 1: add multiplicand to product register

   2. if 0: do nothing

2. shift P/M register right 1 bit

3. repeat register size amount of times

end: entire P/M register is result

note: only works on unsigned numbers

how to do integer division?

dividend / divisor

start: 0…0 | dividend

1. shift R/Q register left 1 bit, new bit shifted in unknown

2. subtract remainder register by divisor (add negative of divisor)

3. if remainder register nonneg: keep result, set new bit = 1

   1. if remainder neg: restore before subtraction, set new bit = 0

4. repeat register size amount of times

end: remainder | quotient

IEEE floating point representation single precision bias

127

IEEE floating point representation single precision bit allocation

1: sign, 8: exponent, 23: fraction

IEEE floating point representation fraction should be normalized with hidden 1

true

given IEEE floating point representation, convert to value formula

(-1)^s * (1+fraction) * 2^(exponent - bias)

IEEE rounding: round to nearest

even