Comments (11)
nsailor
commented on June 12, 2024
1
Just tested with Julia 1.5.1:
Julia Version 1.5.1
Commit 697e782ab8 (2020-08-25 20:08 UTC)
Platform Info:
OS: Linux (x86_64-pc-linux-gnu)
CPU: AMD Ryzen 5 2600 Six-Core Processor
WORD_SIZE: 64
LIBM: libopenlibm
LLVM: libLLVM-9.0.1 (ORCJIT, znver1)
and the example seems to work, despite build AMDGPU warning me that the build failed.
from amdgpu.jl.
jpsamaroo
commented on June 12, 2024
1
LLVM.jl needs to be at least 3.0, which is the release that has the appropriate changes. You should be able to run the following code if your LLVM.jl version if correct:
using LLVM, LLVM.Interop
JuliaContext() do ctx
end
from amdgpu.jl.
jpsamaroo
commented on June 12, 2024
This is one of two possible things:
- Your ROCR library isn't installed, or isn't found by AMDGPU.jl (check the build logs to see if this is the case, it's called "HSA")
- The build logic is crap and needs fixing
I can push a fix for 2. in a few minutes.
from amdgpu.jl.
Cvikli
commented on June 12, 2024
I found a code that could show I don't see any "agent" which is interesting in this case after the tutorial.
using AMDGPU
@show AMDGPU.agents()
Results:
AMDGPU.agents() = HSAAgent[]
I will look after this ROCR, didn't know it is something I need yet. Will be back with the results.
Thank you for the fast reply!
from amdgpu.jl.
jpsamaroo
commented on June 12, 2024
AMDGPU.agents() only shows the agents that have been detected during HSA initialization; however, if HSA init fails or is skipped, then it will return an empty vector. I suspect HSA initialization is either failing or being skipped for you. #44 should fix HSA init being skipped incorrectly.
from amdgpu.jl.
Cvikli
commented on June 12, 2024
With 0.2.0. Runnin the test code:
using AMDGPU
@show AMDGPU.agents()
N=32
a = rand(Float64, N)
b = rand(Float64, N)
c_cpu = a + b
a_d = AMDGPU.HSAArray(a)
b_d = AMDGPU.HSAArray(b)
c_d = similar(a_d)
function vadd!(c, a, b)
i = workitemIdx().x
c[i] = a[i] + b[i]
return
end
wait(@roc groupsize=N vadd!(c_d, a_d, b_d))
The output:
AMDGPU.agents() = HSAAgent[GPU: Vega 20 [Radeon VII] (gfx906), GPU: Vega 20 [Radeon VII] (gfx906), GPU: Vega 20 [Radeon VII] (gfx906), CPU: AMD Ryzen Threadripper 1950X 16-Core Processor]
ERROR: LoadError: MethodError: no method matching JuliaContext(::AMDGPU.var"#231#233"{LLVM.Module})
Closest candidates are:
JuliaContext() at /home/user/.julia/packages/LLVM/T8ZBA/src/interop/base.jl:8
Stacktrace:
[1] emit_exception_user!(::LLVM.Module) at /home/user/.julia/packages/AMDGPU/nnddY/src/exceptions.jl:34
[2] process_module! at /home/user/.julia/packages/AMDGPU/nnddY/src/compiler.jl:19 [inlined]
[3] irgen(::GPUCompiler.CompilerJob{GPUCompiler.GCNCompilerTarget,AMDGPU.ROCCompilerParams}, ::Core.MethodInstance, ::UInt64) at /home/user/.julia/packages/GPUCompiler/rABm5/src/irgen.jl:366
[4] macro expansion at /home/user/.julia/packages/TimerOutputs/dVnaw/src/TimerOutput.jl:206 [inlined]
[5] macro expansion at /home/user/.julia/packages/GPUCompiler/rABm5/src/driver.jl:101 [inlined]
[6] macro expansion at /home/user/.julia/packages/TimerOutputs/dVnaw/src/TimerOutput.jl:206 [inlined]
[7] codegen(::Symbol, ::GPUCompiler.CompilerJob{GPUCompiler.GCNCompilerTarget,AMDGPU.ROCCompilerParams}; libraries::Bool, deferred_codegen::Bool, optimize::Bool, strip::Bool, validate::Bool, only_entry::Bool) at /home/user/.julia/packages/GPUCompiler/rABm5/src/driver.jl:100
[8] emit_function!(::LLVM.Module, ::GPUCompiler.CompilerJob{GPUCompiler.GCNCompilerTarget,AMDGPU.ROCCompilerParams}, ::Function, ::GPUCompiler.Runtime.RuntimeMethodInstance) at /home/user/.julia/packages/GPUCompiler/rABm5/src/rtlib.jl:77
[9] build_runtime(::GPUCompiler.CompilerJob{GPUCompiler.GCNCompilerTarget,AMDGPU.ROCCompilerParams}) at /home/user/.julia/packages/GPUCompiler/rABm5/src/rtlib.jl:117
[10] (::GPUCompiler.var"#65#68"{GPUCompiler.CompilerJob{GPUCompiler.GCNCompilerTarget,AMDGPU.ROCCompilerParams},String,String})() at /home/user/.julia/packages/GPUCompiler/rABm5/src/rtlib.jl:159
[11] get!(::GPUCompiler.var"#65#68"{GPUCompiler.CompilerJob{GPUCompiler.GCNCompilerTarget,AMDGPU.ROCCompilerParams},String,String}, ::Dict{String,LLVM.Module}, ::String) at ./dict.jl:450
[12] load_runtime(::GPUCompiler.CompilerJob{GPUCompiler.GCNCompilerTarget,AMDGPU.ROCCompilerParams}) at /home/user/.julia/packages/GPUCompiler/rABm5/src/rtlib.jl:151
[13] codegen(::Symbol, ::GPUCompiler.CompilerJob{GPUCompiler.GCNCompilerTarget,AMDGPU.ROCCompilerParams}; libraries::Bool, deferred_codegen::Bool, optimize::Bool, strip::Bool, validate::Bool, only_entry::Bool) at /home/user/.julia/packages/GPUCompiler/rABm5/src/driver.jl:96
[14] compile(::Symbol, ::GPUCompiler.CompilerJob{GPUCompiler.GCNCompilerTarget,AMDGPU.ROCCompilerParams}; libraries::Bool, deferred_codegen::Bool, optimize::Bool, strip::Bool, validate::Bool, only_entry::Bool) at /home/user/.julia/packages/GPUCompiler/rABm5/src/driver.jl:39
[15] compile at /home/user/.julia/packages/GPUCompiler/rABm5/src/driver.jl:35 [inlined]
[16] _rocfunction(::GPUCompiler.FunctionSpec{typeof(vadd!),Tuple{ROCDeviceArray{Float64,1,AMDGPU.AS.Global},ROCDeviceArray{Float64,1,AMDGPU.AS.Global},ROCDeviceArray{Float64,1,AMDGPU.AS.Global}}}; device::AMDGPU.RuntimeDevice{HSAAgent}, queue::AMDGPU.RuntimeQueue{HSAQueue}, kwargs::Base.Iterators.Pairs{Union{},Union{},Tuple{},NamedTuple{(),Tuple{}}}) at /home/user/.julia/packages/AMDGPU/nnddY/src/execution.jl:324
[17] _rocfunction at /home/user/.julia/packages/AMDGPU/nnddY/src/execution.jl:321 [inlined]
[18] check_cache(::typeof(AMDGPU._rocfunction), ::GPUCompiler.FunctionSpec{typeof(vadd!),Tuple{ROCDeviceArray{Float64,1,AMDGPU.AS.Global},ROCDeviceArray{Float64,1,AMDGPU.AS.Global},ROCDeviceArray{Float64,1,AMDGPU.AS.Global}}}, ::UInt64; kwargs::Base.Iterators.Pairs{Union{},Union{},Tuple{},NamedTuple{(),Tuple{}}}) at /home/user/.julia/packages/GPUCompiler/rABm5/src/cache.jl:24
[19] vadd! at /home/user/repo/amd_test/tests/test_AMD.jl:20 [inlined]
[20] cached_compilation(::typeof(AMDGPU._rocfunction), ::GPUCompiler.FunctionSpec{typeof(vadd!),Tuple{ROCDeviceArray{Float64,1,AMDGPU.AS.Global},ROCDeviceArray{Float64,1,AMDGPU.AS.Global},ROCDeviceArray{Float64,1,AMDGPU.AS.Global}}}, ::UInt64; kwargs::Base.Iterators.Pairs{Union{},Union{},Tuple{},NamedTuple{(),Tuple{}}}) at /home/user/.julia/packages/GPUCompiler/rABm5/src/cache.jl:0
[21] cached_compilation(::Function, ::GPUCompiler.FunctionSpec{typeof(vadd!),Tuple{ROCDeviceArray{Float64,1,AMDGPU.AS.Global},ROCDeviceArray{Float64,1,AMDGPU.AS.Global},ROCDeviceArray{Float64,1,AMDGPU.AS.Global}}}, ::UInt64) at /home/user/.julia/packages/GPUCompiler/rABm5/src/cache.jl:40 (repeats 2 times)
[22] rocfunction(::Function, ::Type{T} where T; name::Nothing, kwargs::Base.Iterators.Pairs{Union{},Union{},Tuple{},NamedTuple{(),Tuple{}}}) at /home/user/.julia/packages/AMDGPU/nnddY/src/execution.jl:315
[23] rocfunction(::Function, ::Type{T} where T) at /home/user/.julia/packages/AMDGPU/nnddY/src/execution.jl:314
[24] top-level scope at /home/user/.julia/packages/AMDGPU/nnddY/src/execution.jl:184
[25] include_string(::Function, ::Module, ::String, ::String) at ./loading.jl:1088
[26] include_string(::Module, ::String, ::String) at ./loading.jl:1096
in expression starting at /home/user/repo/amd_test/tests/test_AMD.jl:23
Which is already a little bit better.
What do you think what do I do wrong?
from amdgpu.jl.
jpsamaroo
commented on June 12, 2024
That error is due to having an old LLVM version, which shouldn't happen. Can you try upgrading your packages and make sure your LLVM version is >= 3.0.0?
from amdgpu.jl.
Cvikli
commented on June 12, 2024
That is interesting, I had 3.0 but now I downloaded the "#master" from github:
pkg> status
Status `~/repo/amd_test/Project.toml`
[21141c5a] AMDGPU v0.2.0 `https://github.com/JuliaGPU/AMDGPU.jl.git#master`
[6e4b80f9] BenchmarkTools v0.5.0
[336ed68f] CSV v0.7.7
[159f3aea] Cairo v1.0.5
[81a5f4ea] CatViews v1.0.0
[944b1d66] CodecZlib v0.7.0
[5ae59095] Colors v0.12.4
[a81c6b42] Compose v0.9.0
[864edb3b] DataStructures v0.17.20
[31c24e10] Distributions v0.23.10
[da5c29d0] EllipsisNotation v0.4.0
[587475ba] Flux v0.8.3
[186bb1d3] Fontconfig v0.4.0
[59287772] Formatting v0.4.1
[28b8d3ca] GR v0.51.0
[2535ab7d] JSON2 v0.3.2
[0f8b85d8] JSON3 v1.1.1
[1902f260] Knet v1.2.3
[929cbde3] LLVM v3.0.0 `https://github.com/maleadt/LLVM.jl.git#master`
[093fc24a] LightGraphs v1.3.3
[c03570c3] Memoize v0.4.3
[47be7bcc] ORCA v0.5.0
[e4faabce] PProf v1.1.0
[d96e819e] Parameters v0.12.1
[ccf2f8ad] PlotThemes v2.0.0
[58dd65bb] Plotly v0.3.0
[f0f68f2c] PlotlyJS v0.14.0
[91a5bcdd] Plots v1.6.0
[c46f51b8] ProfileView v0.6.5
[d330b81b] PyPlot v2.9.0
[295af30f] Revise v2.7.3
[aa65fe97] SnoopCompile v2.1.1
[2913bbd2] StatsBase v0.33.1
[37f6aa50] TikzPictures v3.2.0
[bdfc003b] TimesDates v0.2.6
[37b6cedf] Traceur v0.3.1
[9f7883ad] Tracker v0.2.11
[e88e6eb3] Zygote v0.5.5
[10745b16] Statistics
Some more:
julia> versioninfo()
Julia Version 1.5.1
Commit 697e782ab8 (2020-08-25 20:08 UTC)
Platform Info:
OS: Linux (x86_64-pc-linux-gnu)
CPU: AMD Ryzen Threadripper 1950X 16-Core Processor
WORD_SIZE: 64
LIBM: libopenlibm
LLVM: libLLVM-9.0.1 (ORCJIT, znver1)
Environment:
JULIA_EDITOR = "/snap/code/42/usr/share/code/code"
JULIA_NUM_THREADS = 8
The error is unchanged for me. Any other idea?
Maybe the problem is that I built the clang and installed it as the description statet somewhere to build it:
➜ build git:(5ff5319) clang --version
clang version 12.0.0 (https://github.com/RadeonOpenCompute/llvm-project.git 0e84ebbde2430d9c8628e67e957e98e46ed32df1)
Target: x86_64-unknown-linux-gnu
Thread model: posix
InstalledDir: /opt/rocm/llvm/bin
Do you think this can cause this issue?
Hmm I have two clang on my computer now: One in /opt/rocm/llvm/bin and one standard. I will check if this can cause the issue.
The standard one.
➜ ~ clang --version
clang version 10.0.0-4ubuntu1
Target: x86_64-pc-linux-gnu
Thread model: posix
InstalledDir: /usr/bin
It seems to me there are many different LLVM.
What is your clang and julia version?
from amdgpu.jl.
Cvikli
commented on June 12, 2024
Which LLVM version should I update, what do you think? The julia's llvm seems to be the same to me.
from amdgpu.jl.
Cvikli
commented on June 12, 2024
You are right, I can't run this 3 liner example code!
from amdgpu.jl.
jpsamaroo
commented on June 12, 2024
@Cvikli I would recommend you retry with a fresh Julia environment, since this appears to be an issue with the version of LLVM being pulled in. Since we lower-bound LLVM to 3, and no one else can reproduce this behavior, I'm closing this issue.
from amdgpu.jl.
Related Issues (20)
- Multiple workers hang test suite on Julia 1.10
- Support for ROCm 5.7.1 HOT 9
- Support for multi-GPU nodes broken in 0.7 HOT 3
- AMDGPU 0.7.x target error on Frontier HOT 6
- [rocSPARSE] ILU(0) and IC(0) preconditioners are not working HOT 1
- Mixed Device Libs Not Detected if Not in Project HOT 2
- error: ran out of registers during register allocation HOT 2
- Hostcall tests hang HOT 29
- KA tests trigger assertion if julia is built with them HOT 3
- Rely on Atomix.jl for atomics
- Invalid `Os::currentStackPtr()`
- Device printing is broken on Julia 1.11
- Performance gap on a 7-point stencil Laplacian kernel on Frontier MI250x GPUs HOT 1
- rocWMMA support HOT 6
- has_rocm_gpu() fails HOT 4
- @atomic is slow within AMDGPU.jl HOT 30
- AMDGPU fails test and crashes when initialized HOT 6
- Update rocSPARSE to ROCm 6.0 HOT 4
- Trying to import AMDGPU fails with (an lvm?) error HOT 7
- `reduce(f, A)` fails to compile if Julia is started with `--check-bounds=no`
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from amdgpu.jl.