JuliaGPU/CUDA.jl

Wrong source paths in debug info

Open

#2,460 创建于 2024年8月6日

在 GitHub 查看
 (0 评论) (0 反应) (0 负责人)Julia (1,408 star) (274 fork)batch import
bugcuda kernelsgood first issue

描述

julia> @device_code_ptx raw=true CuArray([1]).+1

	.file	1 "/home/tim/.julia/packages/GPUArrays/8Y80U/src/host/broadcast.jl"
	.file	2 "./int.jl"
	.file	3 "/home/tim/.julia/packages/GPUArrays/8Y80U/src/device/indexing.jl"
	.file	4 "./essentials.jl"
	.file	5 "/home/tim/.julia/packages/LLVM/5DlHM/src/interop/base.jl"
	.file	6 "./broadcast.jl"
	.file	7 "/home/tim/Julia/pkg/CUDA/src/device/array.jl"
	.file	8 "/home/tim/.julia/packages/LLVM/5DlHM/src/interop/pointer.jl"
	.file	9 "./none"

Only package paths seem to be reported correctly. We should try to resolve these to the Julia BINDIR for a better profiling experience.

贡献者指南

技术栈
无
领域
developer experienceobservability
议题类型
bug
难度面向新贡献者的预计实现难度，1 表示很小改动，5 表示专家级工作。
4
预计时间有经验贡献者完成调查、实现、测试并准备 pull request 的粗略时间范围。
1-2 days
活动状态议题当前的可参与程度：新鲜、活跃、陈旧、阻塞或等待维护者输入。
fresh
清晰度议题是否清楚说明期望改动、验收标准和下一步。
clear
前置要求
Julia programmingLLVM debug infoCUDA toolchain
新手友好度1-100 的估计分数，表示该议题对首次贡献者的友好程度。
30
研究方向
Investigate how source file paths are generated in Julia's LLVM pass for CUDA kernels by examining files like host/broadcast.jl and device/indexing.jl in GPUArrays, and related code in CUDA.jl. Look for how the path mapping to Julia's BINDIR can be implemented. Check the LLVM.interop modules for any existing path resolution logic. The issue suggests resolving paths to the Julia BINDIR for a better profiling experience.