JuliaGPU / JuliaGPU/GPUCompiler.jl

Slow kernel compilation

Open
#65 13 comments 0 reactions 0 assignees View on GitHub
Dominant language
Julia
Stars
187
Forks
68
Avg merge
1d 12h
Merged PRs (30d)
28

Description

IIRC it used to be 200us or so, and @vtjnash mentioned Base compilation taking ~1us~ 1ms or so for the simplest functions. This is pretty slow then:

```
julia> kernel() = return
kernel (generic function with 1 method)

julia> include("pkg/GPUCompiler/test/definitions/native.jl")
native_code_execution (generic function with 1 method)

julia> job, kwargs = native_job(kernel, Tuple{})
(CompilerJob of function kernel() for NativeCompilerTarget, Base.Iterators.Pairs{Union{},Union{},Tuple{},NamedTuple{(),Tuple{}}}())

julia> @benchmark GPUCompiler.compile(:asm, job)
BenchmarkTools.Trial:
memory estimate: 389.70 KiB
allocs estimate: 7019
--------------
minimum time: 7.027 ms (0.00% GC)
median time: 7.784 ms (0.00% GC)
mean time: 7.805 ms (0.28% GC)
maximum time: 11.175 ms (24.24% GC)
--------------
samples: 640
evals/sample: 1
```

Contributor guide

No contributing guide indexed for this repository

Research direction

Start with the benchmark in the issue and inspect pkg/GPUCompiler/test/definitions/native.jl, native_job(kernel, Tuple{}), and GPUCompiler.compile(:asm, job). Reproduce the reported timing and trace where compilation time and allocations are spent. Done means identifying and addressing the slowdown, with benchmark results showing improved compilation time.

Written by the indexing model from the issue text.

Assessment

Tech stack
julia
Domain
compilers, performance
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
25/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.