Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Issues with forward evaluation of conv1x1 on GPU with CUDA 4 #73

Closed
alisiahkoohi opened this issue Feb 22, 2023 · 8 comments
Closed

Issues with forward evaluation of conv1x1 on GPU with CUDA 4 #73

alisiahkoohi opened this issue Feb 22, 2023 · 8 comments

Comments

@alisiahkoohi
Copy link
Collaborator

I run into errors when evaluating the GLOW network (more percisely conv1x1 layer) on GPU (e.g., when running this line).

The error originates in https://github.com/slimgroup/InvertibleNetworks.jl/blob/master/src/utils/compute_utils.jl#L20. Specifically, this line raises ERROR: DivideError: integer division error (can be circuvmented by replacing norm(v) with sum(v.^2)) and this line raises CUBLASError: an invalid value was used as an argument (code 7, CUBLAS_STATUS_INVALID_VALUE), which I was unable to fix.

This issue does not exists when using [email protected] (possible any version before 4? not sure.)
@mloubout
Copy link
Member

For the first one dot(v, v) is in general faster than norm so should swap it, I'll have a look at the gemm error

mloubout added a commit that referenced this issue Feb 27, 2023
@mloubout
fix #73
00ed516
@mloubout
Copy link
Member

I can seem to be able to replicate either issue with CUDA 4 on my desktop. which version exactly leads to this error?

@alisiahkoohi
Copy link
Collaborator Author

I also get a slightly different error with a fresh instalation:

$ julia --project examples/networks/network_glow.jl 
  Downloaded artifact: CUDA_Driver
  Downloaded artifact: CUDA_Runtime
  Downloaded artifact: CUDNN
ERROR: LoadError: CUBLASError: the requested feature is not supported (code 15, CUBLAS_STATUS_NOT_SUPPORTED)
Stacktrace:
  [1] throw_api_error(res::CUDA.CUBLAS.cublasStatus_t)
    @ CUDA.CUBLAS /data/as356/.julia/packages/CUDA/ZdCxS/lib/cublas/libcublas.jl:11
  [2] macro expansion
    @ /data/as356/.julia/packages/CUDA/ZdCxS/lib/cublas/libcublas.jl:24 [inlined]
  [3] cublasSgemv_v2(handle::Ptr{CUDA.CUBLAS.cublasContext}, trans::Char, m::Int64, n::Int64, alpha::Bool, A::CUDA.CuArray{Float32, 2, CUDA.Mem.DeviceBuffer}, lda::Int64, x::CUDA.CuArray{Float32, 1, CUDA.Mem.DeviceBuffer}, incx::Int64, beta::Bool, y::CUDA.CuArray{Float32, 1, CUDA.Mem.DeviceBuffer}, incy::Int64)
    @ CUDA.CUBLAS /data/as356/.julia/packages/CUDA/ZdCxS/lib/utils/call.jl:26
  [4] gemv!
    @ /data/as356/.julia/packages/CUDA/ZdCxS/lib/cublas/wrappers.jl:331 [inlined]
  [5] gemv_dispatch!(Y::CUDA.CuArray{Float32, 1, CUDA.Mem.DeviceBuffer}, A::CUDA.CuArray{Float32, 2, CUDA.Mem.DeviceBuffer}, B::CUDA.CuArray{Float32, 1, CUDA.Mem.DeviceBuffer}, alpha::Bool, beta::Bool)
    @ CUDA.CUBLAS /data/as356/.julia/packages/CUDA/ZdCxS/lib/cublas/linalg.jl:179
  [6] mul!
    @ /data/as356/.julia/packages/CUDA/ZdCxS/lib/cublas/linalg.jl:188 [inlined]
  [7] mul!
    @ /opt2/software/julia/1.7.1/share/julia/stdlib/v1.7/LinearAlgebra/src/matmul.jl:275 [inlined]
  [8] chain_lr(::Base.ReshapedArray{Float32, 2, SubArray{Float32, 3, CUDA.CuArray{Float32, 4, CUDA.Mem.DeviceBuffer}, Tuple{Base.Slice{Base.OneTo{Int64}}, Base.Slice{Base.OneTo{Int64}}, Base.Slice{Base.OneTo{Int64}}, Int64}, true}, Tuple{}}, ::CUDA.CuArray{Float32, 1, CUDA.Mem.DeviceBuffer}, ::CUDA.CuArray{Float32, 1, CUDA.Mem.DeviceBuffer}, ::CUDA.CuArray{Float32, 1, CUDA.Mem.DeviceBuffer})
    @ InvertibleNetworks /data/as356/InvertibleNetworks.jl/src/utils/compute_utils.jl:25
  [9] forward(X::CUDA.CuArray{Float32, 4, CUDA.Mem.DeviceBuffer}, C::Conv1x1; logdet::Nothing)
    @ InvertibleNetworks /data/as356/InvertibleNetworks.jl/src/layers/invertible_layer_conv1x1.jl:184
 [10] forward(X::CUDA.CuArray{Float32, 4, CUDA.Mem.DeviceBuffer}, C::Conv1x1)
    @ InvertibleNetworks /data/as356/InvertibleNetworks.jl/src/layers/invertible_layer_conv1x1.jl:174
 [11] _predefined_mode(obj::Conv1x1, sym::Symbol, args::CUDA.CuArray{Float32, 4, CUDA.Mem.DeviceBuffer}; kwargs::Base.Pairs{Symbol, Union{}, Tuple{}, NamedTuple{(), Tuple{}}})
    @ InvertibleNetworks /data/as356/InvertibleNetworks.jl/src/utils/neuralnet.jl:29
 [12] _predefined_mode
    @ /data/as356/InvertibleNetworks.jl/src/utils/neuralnet.jl:28 [inlined]
 [13] #134#135
    @ /data/as356/InvertibleNetworks.jl/src/utils/neuralnet.jl:40 [inlined]
 [14] #134
    @ /data/as356/InvertibleNetworks.jl/src/utils/neuralnet.jl:40 [inlined]
 [15] forward(X::CUDA.CuArray{Float32, 4, CUDA.Mem.DeviceBuffer}, L::CouplingLayerGlow)
    @ InvertibleNetworks /data/as356/InvertibleNetworks.jl/src/layers/invertible_layer_glow.jl:96
 [16] _predefined_mode(obj::CouplingLayerGlow, sym::Symbol, args::CUDA.CuArray{Float32, 4, CUDA.Mem.DeviceBuffer}; kwargs::Base.Pairs{Symbol, Union{}, Tuple{}, NamedTuple{(), Tuple{}}})
    @ InvertibleNetworks /data/as356/InvertibleNetworks.jl/src/utils/neuralnet.jl:29
 [17] _predefined_mode
    @ /data/as356/InvertibleNetworks.jl/src/utils/neuralnet.jl:28 [inlined]
 [18] #134#135
    @ /data/as356/InvertibleNetworks.jl/src/utils/neuralnet.jl:40 [inlined]
 [19] (::InvertibleNetworks.var"#134#136"{InvertibleNetworks.var"#134#135#137"{CouplingLayerGlow}})(args::CUDA.CuArray{Float32, 4, CUDA.Mem.DeviceBuffer})
    @ InvertibleNetworks /data/as356/InvertibleNetworks.jl/src/utils/neuralnet.jl:40
 [20] forward(X::CUDA.CuArray{Float32, 4, CUDA.Mem.DeviceBuffer}, G::NetworkGlow)
    @ InvertibleNetworks /data/as356/InvertibleNetworks.jl/src/networks/invertible_network_glow.jl:110
 [21] _predefined_mode(obj::NetworkGlow, sym::Symbol, args::CUDA.CuArray{Float32, 4, CUDA.Mem.DeviceBuffer}; kwargs::Base.Pairs{Symbol, Union{}, Tuple{}, NamedTuple{(), Tuple{}}})
    @ InvertibleNetworks /data/as356/InvertibleNetworks.jl/src/utils/neuralnet.jl:29
 [22] _predefined_mode
    @ /data/as356/InvertibleNetworks.jl/src/utils/neuralnet.jl:28 [inlined]
 [23] #134#135
    @ /data/as356/InvertibleNetworks.jl/src/utils/neuralnet.jl:40 [inlined]
 [24] (::InvertibleNetworks.var"#134#136"{InvertibleNetworks.var"#134#135#137"{NetworkGlow}})(args::CUDA.CuArray{Float32, 4, CUDA.Mem.DeviceBuffer})
    @ InvertibleNetworks /data/as356/InvertibleNetworks.jl/src/utils/neuralnet.jl:40
 [25] loss(X::CUDA.CuArray{Float32, 4, CUDA.Mem.DeviceBuffer})
    @ Main /data/as356/InvertibleNetworks.jl/examples/networks/network_glow.jl:27
 [26] top-level scope
    @ /data/as356/InvertibleNetworks.jl/examples/networks/network_glow.jl:34
in expression starting at /data/as356/InvertibleNetworks.jl/examples/networks/network_glow.jl:3

Current versions are:

(InvertibleNetworks) pkg> status
     Project InvertibleNetworks v2.2.3
      Status `/data/as356/InvertibleNetworks.jl/Project.toml`
  [052768ef] CUDA v4.0.1
  [d360d2e6] ChainRulesCore v1.15.7
  [587475ba] Flux v0.13.13
  [bb331ad6] JOLI v0.8.4
  [872c559c] NNlib v0.8.19
  [29a6e085] Wavelets v0.9.5
  [37e2e46d] LinearAlgebra
  [9a3f8284] Random
  [10745b16] Statistics

@mloubout
Copy link
Member

What's your CUDA version? CUDA.versioninfo()

@alisiahkoohi
Copy link
Collaborator Author 

julia> CUDA.versioninfo()
CUDA runtime 11.8, artifact installation
CUDA driver 11.8
NVIDIA driver 470.103.1, originally for CUDA 11.4

Libraries: 
- CUBLAS: 11.8.1
- CURAND: 10.3.0
- CUFFT: 10.9.0
- CUSOLVER: 11.4.1
- CUSPARSE: 11.7.5
- CUPTI: 18.0.0
- NVML: 11.0.0+470.103.1

@mloubout
Copy link
Member

Hum really weird

@mloubout
Copy link
Member

mloubout commented Mar 3, 2023

What's the CUDA version inside the InvertibleNetworks? I.e julia --project then status

@alisiahkoohi
Copy link
Collaborator Author

I posted this information here but below is the outcome for a fresh installation (from master branch):

$ julia --project                                                                                                          
               _                                                                                                                        
   _       _ _(_)_     |  Documentation: https://docs.julialang.org                                                                     
  (_)     | (_) (_)    |                                                                                                                
   _ _   _| |_  __ _   |  Type "?" for help, "]?" for Pkg help.                                                                         
  | | | | | | |/ _` |  |                                                                                                                
  | | |_| | | | (_| |  |  Version 1.7.1 (2021-12-22)                                                                                    
 _/ |\__'_|_|_|\__'_|  |  Official https://julialang.org/ release                                                                       
|__/                   |                                                                                                                
                                                                                                                                        
(InvertibleNetworks) pkg> status                                                                                                        
     Project InvertibleNetworks v2.2.4                                                                                                  
      Status `/data/as356/InvertibleNetworks.jl/Project.toml`                                                                           
  [052768ef] CUDA v4.0.1                                                                                                                
  [d360d2e6] ChainRulesCore v1.15.7                                                                                                     
  [587475ba] Flux v0.13.13                                                                                                              
  [bb331ad6] JOLI v0.8.4                                                                                                                
  [872c559c] NNlib v0.8.19                                                                                                              
  [29a6e085] Wavelets v0.9.5                                                                                                            
  [37e2e46d] LinearAlgebra                                                                                                              
  [9a3f8284] Random                                                                                                                     
  [10745b16] Statistics

Same error as above: LoadError: CUBLASError: the requested feature is not supported (code 15, CUBLAS_STATUS_NOT_SUPPORTED) originated at

    [8] chain_lr(::Base.ReshapedArray{Float32, 2, SubArray{Float32, 3, CUDA.CuArray{Float32, 4, CUDA.Mem.DeviceBuffer}, Tuple{Base.Slice{Base.OneTo{Int64}}, Base.Slice{Base.OneTo{Int64}}, Base.Slice{Base.OneTo{Int64}}, Int64}, true}, Tuple{}}, ::CUDA.CuArray{Float32, 1, CUDA.Mem.DeviceBuffer}, ::CUDA.CuArray{Float32, 1, CUDA.Mem.DeviceBuffer}, ::CUDA.CuArray{Float32, 1, CUDA.Mem.DeviceBuffer})
    @ InvertibleNetworks /data/as356/InvertibleNetworks.jl/src/utils/compute_utils.jl:25

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

No branches or pull requests
2 participants
@mloubout @alisiahkoohi