Inconsistency of == and isequal for CuArray

[JeffFessler]

Describe the bug

Inconsistent behavior between == and isequal for CuArray

To reproduce

The Minimal Working Example (MWE) for this bug:

using CUDA: CuArray
import CUDA

CUDA.allowscalar(false)

x = ones(2,3)
y = ones(6)
xc = CuArray(x)
yc = CuArray(y)
@assert !isequal(x, y)
@assert x !== y
@assert !isequal(xc, yc)
@assert xc == yc # huh?

Manifest.toml

CUDA v3.10.0
Julia 1.7.1

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

I expect two arrays of different dimensions to fail both isequal and ==.
Regular arrays have that property, so should CuArrays, right?

Version info

Details on Julia:

Julia Version 1.7.1
Commit ac5cc99908 (2021-12-22 19:35 UTC)
Platform Info:
  OS: Linux (x86_64-pc-linux-gnu)
  CPU: Intel(R) Xeon(R) Silver 4214 CPU @ 2.20GHz
  WORD_SIZE: 64
  LIBM: libopenlibm
  LLVM: libLLVM-12.0.1 (ORCJIT, cascadelake)
Environment:
  JULIA_NUM_THREADS = 48

Details on CUDA:

CUDA toolkit 11.7, artifact installation
NVIDIA driver 510.73.5, for CUDA 11.6
CUDA driver 11.7

Libraries: 
- CUBLAS: 11.10.1
- CURAND: 10.2.10
- CUFFT: 10.7.2
- CUSOLVER: 11.3.5
- CUSPARSE: 11.7.3
- CUPTI: 17.0.0
- NVML: 11.0.0+510.73.5
- CUDNN: 8.30.2 (for CUDA 11.5.0)
- CUTENSOR: 1.4.0 (for CUDA 11.5.0)

Toolchain:
- Julia: 1.7.1
- LLVM: 12.0.1
- 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

4 devices:
  0: Quadro RTX 5000 (sm_75, 12.149 GiB / 16.000 GiB available)
  1: Quadro RTX 5000 (sm_75, 15.745 GiB / 16.000 GiB available)
  2: Quadro RTX 5000 (sm_75, 15.745 GiB / 16.000 GiB available)
  3: Quadro RTX 5000 (sm_75, 15.745 GiB / 16.000 GiB available)

Additional context

I am writing unit tests for another package to make it CUDA compatible
JeffFessler/LinearMapsAA.jl#43

[JeffFessler]

Possibly related to this comment in the code:

CUDA.jl/lib/cudadrv/context.jl

Line 137 in 2656e15

# NOTE: we don't implement `isequal` or `hash` in order to fall back to `===` and `objectid`

[maleadt]

That's comparison on contexts, unrelated to arrays.

[JeffFessler]

I just realized I can use @which to help determine the source of the difference.
isequal for two CuArrays reverts to the usual isequal for AbstractArrays in Base, which does an axes check as expected:
https://github.com/JuliaLang/julia/blob/742b9abb4dd4621b667ec5bb3434b8b3602f96fd/base/abstractarray.jl#L2525

But == for two CuArrays is calling a method in GPUArrays that uses mapreduce with no axes check:
https://github.com/JuliaGPU/GPUArrays.jl/blob/fc0d327ecc2fd0b3b73427cf6f491591aa096b75/src/host/mapreduce.jl#L88

So this issue really belongs with GPUArrays and is related to JuliaGPU/GPUArrays.jl#405.

After reading the manual more closely, I learned that == and isequal differ for arrays that have missing elements. https://docs.julialang.org/en/v1/base/math/#Base.:==