Try re-enabling the convolution tests

[kshyatt]

I ran these tests on a loop overnight and didn't see any of the errors reported on #725. This PR is a trial balloon to see if perhaps things have been fixed since the newest CUDA version out when that issue was opened is no longer supported.

[github-actions]

Your PR requires formatting changes to meet the project's style guidelines.
Please consider running Runic (git runic master) to apply these changes.

Click here to view the suggested changes.

diff --git a/lib/cudnn/test/convolution.jl b/lib/cudnn/test/convolution.jl
index 23b5ab7fa..af2abb450 100644
--- a/lib/cudnn/test/convolution.jl
+++ b/lib/cudnn/test/convolution.jl
@@ -134,9 +134,10 @@ function convtest(;
         ay2 = act.(ay .+ beta * ay0)
     end
 
-    d = cudnnConvolutionDescriptor(convdims(padding,size(ax),1),
-                                    convdims(stride,size(ax),1),
-                                    convdims(dilation,size(ax),1), mode,
+    d = cudnnConvolutionDescriptor(
+        convdims(padding, size(ax), 1),
+        convdims(stride, size(ax), 1),
+        convdims(dilation, size(ax), 1), mode,
                                     cudnnDataType(dataType), mathType, reorderType,
                                     Cint(group))
     @test ay1 ≈ cudnnConvolutionForward(cw0, cx; bias, activation, mode, padding,

[kshyatt]

If it does turn out these are still flaky, perhaps we could enable them but mark passing optional, like julia-nightly is right now?

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 82.08%. Comparing base (b1cabe8) to head (db653d7).
Report is 1 commits behind head on master.

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #2678      +/-   ##
==========================================
+ Coverage   81.15%   82.08%   +0.93%     
==========================================
  Files         154      154              
  Lines       13662    13661       -1     
==========================================
+ Hits        11087    11214     +127     
+ Misses       2575     2447     -128     

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

[github-actions]

CUDA.jl Benchmarks

Benchmark suite	Current: db653d7	Previous: b1cabe8	Ratio
latency/precompile	46398876023 ns	46290695219 ns	1.00
latency/ttfp	6964832797 ns	6948229451 ns	1.00
latency/import	3634082110 ns	3635808334 ns	1.00
integration/volumerhs	9624305.5 ns	9610186 ns	1.00
integration/byval/slices=1	146718 ns	146887 ns	1.00
integration/byval/slices=3	425305 ns	425144 ns	1.00
integration/byval/reference	145002 ns	144989 ns	1.00
integration/byval/slices=2	286132 ns	286091 ns	1.00
integration/cudadevrt	103310 ns	103314 ns	1.00
kernel/indexing	14090 ns	14040 ns	1.00
kernel/indexing_checked	14695 ns	14692.5 ns	1.00
kernel/occupancy	641.6904761904761 ns	645.9763313609468 ns	0.99
kernel/launch	2026.1 ns	2006.2 ns	1.01
kernel/rand	16835 ns	14653 ns	1.15
array/reverse/1d	19565 ns	19753 ns	0.99
array/reverse/2d	22919 ns	24740 ns	0.93
array/reverse/1d_inplace	9686.333333333334 ns	11286 ns	0.86
array/reverse/2d_inplace	11485 ns	13173 ns	0.87
array/copy	20675 ns	20947 ns	0.99
array/iteration/findall/int	157734 ns	158277 ns	1.00
array/iteration/findall/bool	138882 ns	139096 ns	1.00
array/iteration/findfirst/int	152928 ns	153472 ns	1.00
array/iteration/findfirst/bool	153558.5 ns	154188 ns	1.00
array/iteration/scalar	71726 ns	71437 ns	1.00
array/iteration/logical	206480 ns	213413 ns	0.97
array/iteration/findmin/1d	40936 ns	41271.5 ns	0.99
array/iteration/findmin/2d	93305 ns	93923 ns	0.99
array/reductions/reduce/1d	35354 ns	35851 ns	0.99
array/reductions/reduce/2d	40668 ns	41097.5 ns	0.99
array/reductions/mapreduce/1d	32988 ns	33307 ns	0.99
array/reductions/mapreduce/2d	40551 ns	40963 ns	0.99
array/broadcast	20596 ns	20794 ns	0.99
array/copyto!/gpu_to_gpu	13502 ns	13538 ns	1.00
array/copyto!/cpu_to_gpu	208355 ns	208466 ns	1.00
array/copyto!/gpu_to_cpu	244260 ns	243822 ns	1.00
array/accumulate/1d	108169 ns	108685 ns	1.00
array/accumulate/2d	79704 ns	80636 ns	0.99
array/construct	1259.9 ns	1271.5 ns	0.99
array/random/randn/Float32	42962 ns	43772 ns	0.98
array/random/randn!/Float32	26124.5 ns	26451 ns	0.99
array/random/rand!/Int64	26977 ns	27104 ns	1.00
array/random/rand!/Float32	8893.666666666666 ns	8878.333333333334 ns	1.00
array/random/rand/Int64	29761 ns	29965 ns	0.99
array/random/rand/Float32	12825 ns	13254 ns	0.97
array/permutedims/4d	61208 ns	61539 ns	0.99
array/permutedims/2d	55082 ns	55658 ns	0.99
array/permutedims/3d	55840.5 ns	56143.5 ns	0.99
array/sorting/1d	2775929 ns	2760870 ns	1.01
array/sorting/by	3368065.5 ns	3370232 ns	1.00
array/sorting/2d	1084874.5 ns	1085291 ns	1.00
cuda/synchronization/stream/auto	1021.4 ns	1003.1333333333333 ns	1.02
cuda/synchronization/stream/nonblocking	6392.6 ns	6318.2 ns	1.01
cuda/synchronization/stream/blocking	789.9803921568628 ns	805.3536585365854 ns	0.98
cuda/synchronization/context/auto	1175.8 ns	1184.2 ns	0.99
cuda/synchronization/context/nonblocking	6636.6 ns	6601.6 ns	1.01
cuda/synchronization/context/blocking	902.8 ns	926.2857142857143 ns	0.97

This comment was automatically generated by workflow using github-action-benchmark.

[maleadt]

Looks like it works? Let's remove the code, re-run CI, and merge if it still works