Exploring performance across 3 problem variations with solutions implemented without explicit vectorization, with SSE intrinsics, or AVX2 intrinsics.

Important: References to "non-vectorized" solutions means that the solutions are not explicitly vectorized, the compiler will still implicitly utilize vectorization. TODO: more detail

Default (Advent of Code 2021 Day 7 Part 1):
Given an input string containing a comma-separated list of integers in the range [0,9999], return the SADS (sum of absolute differences) of every number in the input relative to the rounded median of the input values\

Input parsing: done inside solution function
\

Implemented Solutions: SSE (sol::default_sse), non-vectorized (sol::default)\

Relaxed (variant of Default):

Sum of absolute differences between each element in input and first element in input input value range: [0,9999]

Input parsing: is done outside solution function
\

Implemented Solutions: AVX2 (sol::relaxed_avx2), SSE (sol::relaxed_sse), non-vectorized (sol::relaxed)

Ideal (variant of Relaxed):

Sum of absolute differences between each element in input and first element in input input value range: [0,255]

Input parsing: done outside solution function

Implemented Solutions: AVX2 (sol::ideal_avx2, sol::relaxed_avx2), SSE (sol::relaxed_sse), non-vectorized (sol::relaxed)

Intel(R) Core(TM) i7-9750H CPU @ 2.60GHz

Note: The Default implementations for both non-vectorized and SSE solutions are naïve approaches to the problem and not intended to be a meaningful comparison of the differences between non-vectorized implementations and implementations explicitly using intrinsics.

In practice: parsing and calculating the median of input data dominates the running time for both solutions.

See either the Relaxed or Ideal variants of the Default problem for

(1x input = 1000 integers [0,9999])

(1x input = 1000 integers [0,9999])

(1x input = 1000 integers [0,256]) 'AVX2' refers to the general AVX2 solution for the Relaxed problem 'AVX2 optimized' refers to a solution which exploits the constrained value range of the Ideal problem

1. Clone this Repository.
2. cd $CLONE_DIR/
3. cargo run --release -- -v ideal_avx2 -n 1000x

-h, --help                 Print help information
-n, --name <NAME>          Input file name: "input/{NAME}.txt" [default: ]
-v, --variant <VARIANT>    Variant: ["default", "default_sse", "relaxed", "relaxed_sse",
                           "relaxed_avx2", "ideal_avx2"] [default: ]

1. cargo bench -- --quiet

1. cargo test