Zusammenfassung der Ressource
Advanced Computer Architecture
- Introduction
- Hardware components
- ALU
- Control Unit
- Registers
- Program execution
- Instruction FETCH
- Instruction DECODE
- Execute
- Memory Access
- Write Back
- Performance metric
- execution time = cpu time = IC x CPI x Tc
- clock period
- semiconductor
technology
- datapath
design and
implemetation
- instruction count
- program
- instruction set
architecture
- clocks per instuction
- isa
- datapath design
- parallel and pipelined HW
- indicator: execution time
- performance = 1/execution time
- total no. of cycles = sum(IC k x CPI k), k is a category
- average CPI = total cycles/IC = sum (wk x IC k)
- datapath
- multicycle datapath: CPI > 1
- pipelined datapth: CPI can bereduced up to 1
- parallel
datapath: CPI
<1
- VLIW/Superscalar processors
- SIMD and multicore processors
- speedup
- speedup =
execution time of A/
execution time of B
Anmerkungen:
- 1. Speed up of computer B over A
2. Computer B is the enhanced version of computer A
3. The one being evaluated as divisor
- throughput rate
- response time/latency
- MIPS
- native
- relative
- peak
- FLOPS
- Enhancement
- reduce Tc ⇒
increase clock
frequency
- power
consumption ↑
- CPI may ↑
- memory wall problem
- reduce CPI
- instruction
pipelining
- multicore
processing
- reduce IC
- more
complex
instructions
- Tc may ↑
- multi-issue
processor
- power dissipation
- dynamic
- stati
- ISA
- Instruction format
- R-format
- I-format
- J-format
- adressing modes
- register file
- types of operations
- ALU
- arithmetic
- logical
- shift
- Data Transfer
- load-store
- Control-Flow
- unconditional
- conditional
- memory address space