Fix ones_like on GPUs

[adrhill]

Closes #59.

Benchmarks

I had to run tests on a GPU with only 2 GB of VRAM, so I only benchmarked very small batchsizes:

using CUDA, cuDNN
CUDA.functional() || error("CUDA not functional")
CUDA.versioninfo()
CUDA.allowscalar(false)

using Flux
using ExplainableAI
using Metalhead
using Test 

# Load model
model = VGG(16, pretrain=true).layers
model = strip_softmax(model)

# Create analyzer
composite = EpsilonPlusFlat()
analyzer = LRP(model, composite)

model_gpu = model |> gpu;
analyzer_gpu = LRP(model_gpu, composite)

function time_analyzer(batchsize)
    input = rand(Float32, 224, 224, 3, batchsize)
    input_gpu = input |> gpu

    @info "Timing CPU with batchsize $batchsize..."
    analyze(input, analyzer)
    @time analyze(input, analyzer)
	
    @info "Timing GPU with batchsize $batchsize..."
    analyze(input_gpu, analyzer_gpu)
    @time analyze(input_gpu, analyzer_gpu)
    return nothing
end

julia> time_analyzer(1)
[ Info: Timing CPU with batchsize 1...
  2.211998 seconds (4.32 k allocations: 3.342 GiB, 6.10% gc time)
[ Info: Timing GPU with batchsize 1...
  0.810851 seconds (13.94 k allocations: 778.767 KiB, 4.53% gc time)

julia> time_analyzer(2)
[ Info: Timing CPU with batchsize 2...
  3.751037 seconds (4.36 k allocations: 3.998 GiB, 1.97% gc time)
[ Info: Timing GPU with batchsize 2...
  1.433466 seconds (12.08 k allocations: 657.875 KiB, 2.17% gc time)

julia> time_analyzer(5)
[ Info: Timing CPU with batchsize 5...
  9.260670 seconds (4.37 k allocations: 5.964 GiB, 1.66% gc time)
[ Info: Timing GPU with batchsize 5...
  3.676279 seconds (16.99 k allocations: 994.595 KiB, 9.84% gc time)

julia> CUDA.versioninfo()
CUDA runtime 12.1, artifact installation
CUDA driver 12.2
NVIDIA driver 535.104.5

CUDA libraries: 
- CUBLAS: 12.1.3
- CURAND: 10.3.2
- CUFFT: 11.0.2
- CUSOLVER: 11.4.5
- CUSPARSE: 12.1.0
- CUPTI: 18.0.0
- NVML: 12.0.0+535.104.5

Julia packages: 
- CUDA: 4.4.1
- CUDA_Driver_jll: 0.5.0+1
- CUDA_Runtime_jll: 0.6.0+0

Toolchain:
- Julia: 1.9.3
- LLVM: 14.0.6
- PTX ISA support: 3.2, 4.0, 4.1, 4.2, 4.3, 5.0, 6.0, 6.1, 6.3, 6.4, 6.5, 7.0, 7.1, 7.2, 7.3, 7.4, 7.5
- Device capability support: sm_37, sm_50, sm_52, sm_53, sm_60, sm_61, sm_62, sm_70, sm_72, sm_75, sm_80, sm_86

1 device:
  0: NVIDIA GeForce 940MX (sm_50, 1.953 GiB / 2.000 GiB available)

[codecov]

Codecov Report

Patch coverage: 100.00% and project coverage change: +0.27% 🎉

Comparison is base (3aa10ba) 94.92% compared to head (52f6dc6) 95.19%.
Report is 3 commits behind head on master.

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #142      +/-   ##
==========================================
+ Coverage   94.92%   95.19%   +0.27%     
==========================================
  Files          18       19       +1     
  Lines         689      687       -2     
==========================================
  Hits          654      654              
+ Misses         35       33       -2     

Files Changed	Coverage Δ
src/ExplainableAI.jl	100.00% <ø> (ø)
src/lrp/rules.jl	94.16% <ø> (+0.96%)	⬆️
ext/TullioLRPRulesExt.jl	100.00% <100.00%> (ø)
src/lrp/canonize.jl	100.00% <100.00%> (ø)
src/utils.jl	95.45% <100.00%> (ø)

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.