JuliaLang/julia

better calling convention for large numbers of arguments

Open

#44,417 opened on Mar 2, 2022

View on GitHub
 (4 comments) (0 reactions) (0 assignees)Julia (5,773 forks)batch import
compiler:codegengood first issuelatency

Repository metrics

Stars
 (48,709 stars)
PR merge metrics
 (Avg merge 20d 6h) (157 merged PRs in 30d)

Description

Currently, we mostly map julia function arguments directly to LLVM arguments regardless of how many there are. That probably causes us to hit some pathological cases, since LLVM is not likely to care about or optimize handling 1000 arguments, since most front-ends do not generate that.

For example:

julia> fields = [:($(Symbol("var$i"))::Int) for i in 1:1000];

julia> @eval struct x
           $(fields...)
           end

julia> @time @eval x(fill(1,1000)...)
  0.482531 seconds (39.92 k allocations: 2.195 MiB, 99.91% compilation time)

... which goes down to 0.16 seconds with -O0. There may be multiple issues (for example there seems to be too much time in dead store elimination), but we should probably use an array for the tail of the argument list.

This is not exactly easy, but could be a good relatively small project for somebody who wants to get started on the compiler.

Contributor guide

Research direction
Study the current argument passing in Julia's code generation and implement a change to pass large argument lists via an array or stack instead of direct LLVM arguments.
Tech stack
julia
Domain
compilers
Issue type
Refactor
DifficultyEstimated implementation difficulty for a new contributor, from 1 for very small changes to 5 for expert-level work.
3
Estimated timeA rough time range for an experienced contributor to investigate, implement, test, and prepare a pull request.
Half day
Activity statusHow available the issue appears right now: fresh, active, stale, blocked, or waiting on maintainer input.
Active
ClarityHow clearly the issue explains the expected change, acceptance criteria, and next step.
Clear
Prerequisites
JuliaLLVMGit
Newbie friendlinessA 1-100 score estimating how approachable this issue is for first-time contributors.
60