ASL at ETHZ - ETH Zurich

The parts of a processor design that one needs to understand to write assembly code.

This usually consists of instruction set specification and registers.

Example ISAs are x86, MIPS, ARM, ...

We will be using x86 in this course.

Yes

It depends on compiler flags.

Extension of the ISA. Data types and instructions for the parallel

computation on short (length 2-8) vectors of integers or floats

1. Useful: Many applications have the necessary fine-grain parallelism. Then: speedup by a factor close to vector length
2. Doable: Chip designers have enough transistors to play with; easy to build with replication

X = X+ V • Z

An add and a multiply done as one operation, i.e., involves only one rounding step which gives better accuracy than sequence of mult and add

This is a natural pattern in many algorithms

16-way single is a vector with 16 floats

8-way double is a vector with 8 doubles

A double is 8 bytes = 64 bit.

So you need 512 bit registers to hold one of these vectors.

Implementation of the architecture

Examples: Caches, cache structure, CPU frequency, details of the virtual

memory system

So those are either Intel processors or AMD processors.

Because data can also be stored in registers.

Because if the index computations are simple - if they are affine then they don't constitute a bottleneck.

A function/piece of code is:

Compute bound if it has high operational intensity

Memory bound if it has low operational intensity

The cache is split into sets and if the cache contains two lines/blocks per set it is said to be 2-way associative.