Introduction (required): Title and brief overview of what the poster reports on. Title: Pipeline architectures in C++: overloaded pipe operator | std::expected and its monadic operations Brief programmers. One of its most characteristic patterns is composition of functions in the form of a pipeline pattern. Since C++20 we can use the ranges library with its characteristic function composition abilities thanks to the overloaded pipe operator. In this poster I show how to implement a custom pipeline framework that employs std::expected, available since C++23. An overloaded custom pipe operator

49Fern! 50 manya227 June 2024 Pipeline pipeline({ vadd(a, b, len, out_1), vadd(out_1, c, len, out_2), }); pipeline.constructPipeline(); pipeline = pipeline.finalize(); void my_fused_impl(const size;Fern! 50 manya227 June 2024 Pipeline pipeline({ vadd(a, b, len, out_1), vadd(out_1, c, len, out_2), }); pipeline.constructPipeline(); pipeline = pipeline.finalize(); void my_fused_impl(const len1); for(int64 t x2 = out 2 idx; x2 < out 2 idx + out 2 size;Fern! 50 pipeline.constructPipeline(); pipeline = pipeline.finalize(); void my_fused_impl(const Array a, const Array b

) pipeline1 transform(f) >>= ( ) pipeline2 inputs >>= ( ) sends inputs to pipeline2 lvalue16 >>= transform(f) >>= filter(p) >>= push_back(results) ( ) pipeline1 transform(f) >>= ( ) pipeline2 (getInputs() (getInputs() >>= ) sends inputs to pipeline2 rvalue getInputs() Also, because it is right associative. Right associativity allows rvalues as input17 inputs | transform(f) | filter(p) | push_back(results) OF THE REST OF THE TALK What gets in a pipeline ranges STL algorithms Multiple inputs The pipes Inside a pipe Branching out pipes The output of a pipeline Simple container adapters Complex container

results of one operation affect another operation ● Data dependency: a = (v1 + v2)*(v1 – v2) ● Pipeline increases CPU utilization ● Multiple instruction streams run in parallel – Dependencies within streams USE MORE OF THE HARDWAREBranchless Computing 21 BRANCHES: BANE OF THE PIPELINES ● Hard to pipeline code: a+=(v1[i]>v2[i])?v1[i]:v2[i] ● Pipelining relies on a continuous stream of instructions jump a[i]:a+=v1[i] ...Branchless Computing 22 BRANCHES: BANE OF THE PIPELINES ● Not hard to pipeline code: a+=(v1[i]>v2[i])?v1[i]:v2[i] Performance and Efficiency load:v1[i] load:v2[i] cmp[i]:v1[i]>v2[i]

equal to n>>1. •Each instruction that is fetched from the memory is pushed into a pipeline, one of the steps in the pipeline is execution, execution may be piped as well. WHY DID THIS HAPPEN:Alex Dathskovsky equal to n>>1. •Each instruction that is fetched from the memory is pushed into a pipeline, one of the steps in the pipeline is execution, execution may be piped as well. •Each execution has its own unit equal to n>>1. •Each instruction that is fetched from the memory is pushed into a pipeline, one of the steps in the pipeline is execution, execution may be piped as well. •Each execution has its own unit

C++20 Nico Josuttis C++ Filter View @cppcon 2024-09-17 3©2024 by josuttis.com 7 C++ Example of Pipeline of Range Adaptors int main() { std::map composers { {"Bach", 1685}, {"Mozart" '\n'; } } Output: - Beethoven - Chopin - Mozart C++20 ©2024 by josuttis.com 8 C++ Using a Pipeline of Views // view 4th to 11th value that are multiples of 3 with suffix "s": auto v = std::views::iota(1) C++20 Nico Josuttis C++ Filter View @cppcon 2024-09-17 4©2024 by josuttis.com 9 C++ Using a Pipeline of Views // view 4th to 11th value that are multiples of 3 with suffix "s": auto v = std::views::iota(1)

pipelines ▸ A pipeline begins with some sequence (the source), passes through zero or more adaptors, and then finishes with a call to an algorithm 30ITERATION REVISITED FLUX PIPELINE EXAMPLE std�::vector .value_or(0); �// handle empty sequences 31ITERATION REVISITED FLUX PIPELINE EXAMPLE std�::vector vec = get_vector(); int max_even_sq = flux�::ref(vec) �//

it is effectively computing the same lockfile that existed before. It is also possible, and a CI pipeline could do it, to directly check that libb/1.1 is defined inside the new lockfiles. If it is not there necessary. • conan-build: a repository mainly for CI purposes. When CI is creating packages in a pipeline, it can put those packages in this repository, so they can still be used in the CI pipelines, be the form of a Pull Request, wanting to merge those changes in the main “develop” branch. Package pipeline The first thing the CI will do is to build libb/0.2@user/testing package, containing the developer

it is effectively computing the same lockfile that existed before. It is also possible, and a CI pipeline could do it, to directly check that libb/1.1 is defined inside the new lockfiles. If it is not there necessary. • conan-build: a repository mainly for CI purposes. When CI is creating packages in a pipeline, it can put those packages in this repository, so they can still be used in the CI pipelines, be the form of a Pull Request, wanting to merge those changes in the main “develop” branch. Package pipeline The first thing the CI will do is to build libb/0.2@user/testing package, containing the developer

necessary. • conan-build: a repository mainly for CI purposes. When CI is creating packages in a pipeline, it can put those packages in this repository, so they can still be used in the CI pipelines, be the form of a Pull Request, wanting to merge those changes in the main “develop” branch. Package pipeline The first thing the CI will do is to build libb/0.2@user/testing package, containing the developer /locks/libb_base.lock 138 Chapter 9. Versioning Conan Documentation, Release 1.28.2 Products pipeline There is an important question to be addressed: when a package changes, what other packages consuming