Multidimensional reverse

[piever]

Describe the bug

The reverse! API supports both a single value or a tuple, eg reverse!(A, dims=(1, 2)). I've noticed that CUDA errors for the tuple case.

To reproduce

julia> using CUDA

julia> A = CUDA.rand(3, 3, 3);

julia> reverse!(A, dims=(1, 2))
ERROR: TypeError: in keyword argument dims, expected Integer, got a value of type Tuple{Int64, Int64}
Stacktrace:
 [1] top-level scope
   @ REPL[37]:1
 [2] top-level scope
   @ C:\Users\pietro\.julia\packages\CUDA\9T5Sq\src\initialization.jl:66

I'm on CUDA v3.4.2 (latest release), but it seems that even on the master branch this problem exists.

Relevant version info:

[052768ef] CUDA v3.4.2
[0c68f7d7] GPUArrays v8.0.2
[61eb1bfa] GPUCompiler v0.12.9
[929cbde3] LLVM v4.4.0

Expected behavior

I imagined that reverse!(A, dims::Tuple{Vararg{Int}}) would work. I suspect a simple solution would be to just define it by calling the single dimension method for each dimension in dims, but I'm not sure how wasteful this would be.

Version info

Details on Julia:

julia> versioninfo()
Julia Version 1.6.2
Commit 1b93d53fc4 (2021-07-14 15:36 UTC)
Platform Info:
  OS: Windows (x86_64-w64-mingw32)
  CPU: Intel(R) Core(TM) i7-7700HQ CPU @ 2.80GHz
  WORD_SIZE: 64
  LIBM: libopenlibm
  LLVM: libLLVM-11.0.1 (ORCJIT, skylake)

Details on CUDA:

julia> CUDA.versioninfo()
CUDA toolkit 11.4.1, artifact installation
CUDA driver 11.2.0
NVIDIA driver 462.30.0

Libraries:
- CUBLAS: 11.5.4
- CURAND: 10.2.5
- CUFFT: 10.5.1
- CUSOLVER: 11.2.0
- CUSPARSE: 11.6.0
- CUPTI: 14.0.0
- NVML: 11.0.0+462.30
- CUDNN: 8.20.2 (for CUDA 11.4.0)
- CUTENSOR: 1.3.0 (for CUDA 11.2.0)

Toolchain:
- Julia: 1.6.2
- LLVM: 11.0.1
- 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
- Device capability support: sm_35, sm_37, sm_50, sm_52, sm_53, sm_60, sm_61, sm_62, sm_70, sm_72, sm_75, sm_80

1 device:
  0: Quadro M1200 (sm_50, 2.307 GiB / 4.000 GiB available)

[maleadt]

Yeah, the kernel is currently written with only a single dimension in mind, and would need to be adapted. Reversing multiple times is pretty wasteful, and we should just need to update the index calculation in the current kernels.

[RainerHeintzmann]

... just ran into the same issue. Any progress on this?

[maleadt]

No. Feel free to take a stab at it! Happy to help you with any issues.

[RainerHeintzmann]

I made a revised version, which now should be able to handle this. Also found a way to kick out the empty threads. Should I make a pull request?

[maleadt]

#1899