What do we have??

TODO: Intro

• Make your experiments reproducable and gain insights in differences in reproduced results from others.

TODO: How to get hardware info

• lscpu (clock speed, architecture)
• lshw (cache/ram info)

TODO: Latency of instructions

• Arithmatic instructions low latency
• Memory reads and writes more complex
  • (use comparison of real time adding from ComputerFile: https://www.youtube.com/watch?v=PpaQrzoDW2I)

TODO: How to get software info

• gcc -v

TODO: Provide scripts on how to get this information

Chapter 1: Scoping

What do we want to know?

What do we want to measure? Why do we want to measure it? Do we need to measure it? Main takeaway: do not benchmark everything! (talk about tools like valgrind etc)

TODO: Make Flowchart ending in benchmark or don’t benchmark.

Chapter 2: Theoretical speed

What do we expect??

TODO: Intro

• Get insights on how an algorithm performs relative to theory

TODO: Introduce Complexity

• Big O theoretical concept
  • Time complexity
  • Space complexity
• Big O practical concept
  • A for loop where we do one multiplication and one addition:

for (int r1 = 0; r1 < LEN_R1; r1++) {               /* n^1*add + n^1*lt  */
    for (int c2 = 0; c2 < LEN_C2; c2++) {           /* n^2*add + n^2*lt  */
        res[r1][c2] = 0;
        for (int c1 = 0; c1 < LEN_C1; c1++) {       /* n^3*add + n^3*lt  */ 
            res[r1][c2] += m1[r1][c1] * m2[c1][c2]; /* n^3*add + n^3*mul */
        }
    }
}

NOTE: assume the compiler does not optimize anything.

Time: O(add x (2n^3+n^2+n^1) + lt x (n^3+n^2+n^1)+ mul x n^3)

TODO: Calculate theoretical speed TODO: Simple case study

• Why is this off?
  • optimizations of the compiler
  • memory reads and writes

Chapter 3: Validity

What are we measuring?

TODO: Testing with pre and post conditions

• Have these be the same when comparing versions or programs. This is to ensure we dont compare apples with oranges.

TODO: Use Scoping

• Find a subsection of code that answers the question that has been Scoped

SOURCES

algorithmica: Algorithms for Modern Hardware

Ulrich Drepper: What Every Programmer Should Know About Memory