Native profiler output is limited to around 100 columns when printing to a file

[johnbcoughlin]

Describe the bug

The native profiler table output is truncated for no good reason when writing to a file.

To reproduce

open("example.txt", "w") do filehandle
    CUDA.@profile io=filehandle CUDA.rand(10, 10)
end

$ cat example.txt
Profiler ran for 443.47 ms, capturing 536 events.

Host-side activity: calling CUDA APIs took 88.12 ms (19.87% of the trace)
┌──────────┬───────────┬───────┬───────────┬───────────┬───────────┬────────────
│ Time (%) │      Time │ Calls │  Avg time │  Min time │  Max time │ Name      ⋯
├──────────┼───────────┼───────┼───────────┼───────────┼───────────┼────────────
│   19.64% │  87.08 ms │     2 │  43.54 ms │   2.23 ms │  84.85 ms │ cudaLaunc ⋯
│    0.09% │ 380.99 µs │     1 │ 380.99 µs │ 380.99 µs │ 380.99 µs │ cudaFree  ⋯
│    0.02% │  80.11 µs │     1 │  80.11 µs │  80.11 µs │  80.11 µs │ cudaDevic ⋯
│    0.01% │  37.67 µs │     1 │  37.67 µs │  37.67 µs │  37.67 µs │ cudaMallo ⋯
│    0.00% │  19.79 µs │     1 │  19.79 µs │  19.79 µs │  19.79 µs │ cuDeviceG ⋯
│    0.00% │  10.73 µs │     1 │  10.73 µs │  10.73 µs │  10.73 µs │ cuMemAllo ⋯
│    0.00% │  10.49 µs │     1 │  10.49 µs │  10.49 µs │  10.49 µs │ cudaGetDe ⋯
│    0.00% │   5.01 µs │     3 │   1.67 µs │ 238.42 ns │   4.29 µs │ cudaDevic ⋯
│    0.00% │    3.1 µs │     4 │ 774.86 ns │ 238.42 ns │   1.43 µs │ cudaGetLa ⋯
│    0.00% │   2.38 µs │     1 │   2.38 µs │   2.38 µs │   2.38 µs │ cuInit    ⋯
│    0.00% │ 715.26 ns │     1 │ 715.26 ns │ 715.26 ns │ 715.26 ns │ cuDeviceG ⋯
│    0.00% │ 476.84 ns │     1 │ 476.84 ns │ 476.84 ns │ 476.84 ns │ cuDriverG ⋯
│    0.00% │ 476.84 ns │     1 │ 476.84 ns │ 476.84 ns │ 476.84 ns │ cuDeviceT ⋯
│    0.00% │ 238.42 ns │     1 │ 238.42 ns │ 238.42 ns │ 238.42 ns │ cuModuleG ⋯
│    0.00% │ 238.42 ns │     1 │ 238.42 ns │ 238.42 ns │ 238.42 ns │ cuDeviceG ⋯
│    0.00% │    0.0 ns │     1 │    0.0 ns │    0.0 ns │    0.0 ns │ cuDeviceG ⋯
└──────────┴───────────┴───────┴───────────┴───────────┴───────────┴────────────
                                                                1 column omitted

Device-side activity: GPU was busy for 94.41 µs (0.02% of the trace)
┌──────────┬──────────┬───────┬──────────┬──────────┬──────────┬────────────────
│ Time (%) │     Time │ Calls │ Avg time │ Min time │ Max time │ Name          ⋯
├──────────┼──────────┼───────┼──────────┼──────────┼──────────┼────────────────
│    0.02% │ 92.51 µs │     1 │ 92.51 µs │ 92.51 µs │ 92.51 µs │ _Z20generate_ ⋯
│    0.00% │  1.91 µs │     1 │  1.91 µs │  1.91 µs │  1.91 µs │ _Z13gen_seque ⋯
└──────────┴──────────┴───────┴──────────┴──────────┴──────────┴────────────────
                                                                1 column omitted

Paste your Manifest.toml here, or accurately describe which version of CUDA.jl and its dependencies (GPUArrays.jl, GPUCompiler.jl, LLVM.jl) you are using.

Expected behavior

A clear and concise description of what you expected to happen.

Version info

Details on Julia:

julia> versioninfo()
Julia Version 1.9.0
Commit 8e630552924 (2023-05-07 11:25 UTC)
Platform Info:
  OS: Linux (x86_64-linux-gnu)
  CPU: 80 × Intel(R) Xeon(R) Gold 6230 CPU @ 2.10GHz
  WORD_SIZE: 64
  LIBM: libopenlibm
  LLVM: libLLVM-14.0.6 (ORCJIT, cascadelake)
  Threads: 2 on 80 virtual cores
Environment:
  LD_LIBRARY_PATH = /sw/contrib/stf-src/petsc/mpich/3.15.1/lib:/sw/contrib/stf-src/hdf5/mpich/1.12.0/lib:/sw/contrib/stf-src/mpich/4.0a2/lib:/sw/contrib/stf-src/mpich/4.0a2/include:/sw/ucx/1.10.0/lib:/sw/contrib/stf-src/gdb/10.2/lib:/sw/gcc/10.2.0/lib64:/sw/gcc/10.2.0/lib
  JULIA_REVISE_POLL = 1
  JULIA_DEPOT_PATH = /gscratch/aaplasma/johnbc/.julia
  JULIA_PROJECT = .

Details on CUDA:

julia> CUDA.versioninfo()
CUDA runtime 12.2, artifact installation
CUDA driver 12.2
NVIDIA driver 535.104.12

CUDA libraries:
- CUBLAS: 12.2.5
- CURAND: 10.3.3
- CUFFT: 11.0.8
- CUSOLVER: 11.5.2
- CUSPARSE: 12.1.2
- CUPTI: 20.0.0
- NVML: 12.0.0+535.104.12

Julia packages:
- CUDA: 5.0.0
- CUDA_Driver_jll: 0.6.0+4
- CUDA_Runtime_jll: 0.9.2+4

Toolchain:
- Julia: 1.9.0
- 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: Quadro RTX 6000 (sm_75, 19.855 GiB / 24.000 GiB available)

[maleadt]

I guess we shouldn't limit the nocrop case to IOBuffer:

CUDA.jl/src/profile.jl

Lines 573 to 579 in 50136fb

	crop = if io isa IOBuffer
	# when emitting to a string, render all content (e.g., for the tests)
	:none
	else
	:horizontal
	end

As a workaround, render it to an IOBuffer and write that to your file.