CSCI 370: Final Review

Instruction Level Parallelism

• Loop Unrolling
• Sequential vs Parallel computation
• Functional efficiencies

Code Optimization

• Static Analysis
  • Alias, Pointer, Shape, Escape, Array Access, Dependence, Control Flow, Data Flow
  • Be able to describe TWO
• Intermediate Representation
  • What is it?
  • Why is it useful?

Memory Hierarchy

• DRAM Structure / Access pattern (RAS, CAS, cell, supercell)
• Bus structure
  • { CPU } <==( system bus )==> { I/O bridge } <==( memory bus )==> { memory }
• Memory read vs write
• Principles of Locality
  • Temporal vs Spatial
• Example Hierarchy (registers, L1, L2, L3, Memory, Storage)

Caches

• Definition
• Terms: cacheline, set, associativity, offset, block size
• Hits vs Misses
  • Compulsory vs Capacity vs Conflict (misses)
• Direct vs K-way associative vs Fully associative
• Simulating a sequence of memory accesses
• Miss rate + miss penalty

Cache Performance

• Memory Mountain (analyze a picture / compare two figures)
• loop transformations (permutation, blocking)

Virtual Memory

• Definition
• Terms: Physical vs Virtual Address, Page Table, Page Fault
• Translation Lookaside Buffer (TLB) -- what is it? how does it work?
• Mapping from VA to PA -- steps taken
• Simulating memory accesses with TLB, configuration, and Page Table
• Multi-level page tables -- justification?

Exceptions

• definition
• synchronous vs asynchronous
• interrupt vs trap vs fault vs abort

Linking

• Types of Object files
• Why use linkers?
• What do linkers do?
• ELF Binary: .text, .data, .bss, .rodata
• Global vs External vs Local symbols
• Strong vs Weak symbols

x87 and SIMD:

• x87 stack-based example
• definition of SIMD
• SSE registers + instruction naming conventions
  • packed vs scalar
  • float vs double

Virtualization and System IO

• gang scheduling vs hypervisors
• USB / Thunderbolt: trends over time
• PCIe: trends over time; how is data transferred

Multicore

• Snoopy cache (dirty bit)
• False sharing
• Hyperthreading vs Multicore
  • what is replicated?
  • what isn't replicated?

Low Power

• ARM ISA
• Barrel Shifter
• Instruction Format
  • conditional execution
  • setting condition codes