Support cuTENSOR contractors for 1D views

[kshyatt]

Should address #2407

Relatedly we should consider what to do in the more generic case of a reshaped n-dim CuArray where n > 1, but I think that is a separate issue.

[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/cutensor/src/types.jl b/lib/cutensor/src/types.jl
index 68b7865d2..b014a76bd 100644
--- a/lib/cutensor/src/types.jl
+++ b/lib/cutensor/src/types.jl
@@ -202,7 +202,8 @@ mutable struct CuTensorDescriptor
     handle::cutensorTensorDescriptor_t
     # inner constructor handles creation and finalizer of the descriptor
     function CuTensorDescriptor(sz::Vector{Int64}, st::Vector{Int64}, eltype::DataType,
-                                alignmentRequirement::UInt32=CUTENSOR_ALIGNMENT)
+            alignmentRequirement::UInt32 = CUTENSOR_ALIGNMENT
+        )
         desc = Ref{cutensorTensorDescriptor_t}()
         length(st) == (N = length(sz)) || throw(ArgumentError("size and stride vectors must have the same length"))
         cutensorCreateTensorDescriptor(handle(), desc, N, sz, st, eltype, alignmentRequirement)
diff --git a/lib/cutensor/test/contractions.jl b/lib/cutensor/test/contractions.jl
index 5a14f3d74..e9f8ec082 100644
--- a/lib/cutensor/test/contractions.jl
+++ b/lib/cutensor/test/contractions.jl
@@ -164,33 +164,33 @@ eltypes = [(Float32, Float32, Float32, Float32),
     end
 end
 
-# https://github.com/JuliaGPU/CUDA.jl/issues/2407
-@testset "contractions of views" begin
-    @testset for (eltyA, eltyB, eltyC, eltyCompute) in eltypes
-        dimsA = (16,)
-        dimsB = (4,)
-        dimsC = (8,)
-        A = rand(eltyA, dimsA)
-        B = rand(eltyB, dimsB)
-        C = rand(eltyC, dimsC)
-        dA = CuArray(A)
-        dB = CuArray(B)
-        dC = CuArray(C)
-        dD = CuArray(C)
-        vA = @view dA[1:4]
-        vB = @view dB[4:4]
-        vC = @view dC[3:6]
-        vD = @view dD[3:6]
-        tA = CuTensor(reshape(vA, (4, 1)), [1, 2])
-        tB = CuTensor(reshape(vB, (1, 1)), [3, 2])
-        tC = CuTensor(reshape(vC, (1, 4)), [3, 1])
-        mul!(tC, tA, tB)
-        tA2 = CuTensor(copy(vA), [1, 2])
-        tB2 = CuTensor(copy(vB), [3, 2])
-        tD = CuTensor(copy(vD), [3, 1])
-        mul!(tD, tA2, tB2)
-        @test tD.data ≈ tD.data
+    # https://github.com/JuliaGPU/CUDA.jl/issues/2407
+    @testset "contractions of views" begin
+        @testset for (eltyA, eltyB, eltyC, eltyCompute) in eltypes
+            dimsA = (16,)
+            dimsB = (4,)
+            dimsC = (8,)
+            A = rand(eltyA, dimsA)
+            B = rand(eltyB, dimsB)
+            C = rand(eltyC, dimsC)
+            dA = CuArray(A)
+            dB = CuArray(B)
+            dC = CuArray(C)
+            dD = CuArray(C)
+            vA = @view dA[1:4]
+            vB = @view dB[4:4]
+            vC = @view dC[3:6]
+            vD = @view dD[3:6]
+            tA = CuTensor(reshape(vA, (4, 1)), [1, 2])
+            tB = CuTensor(reshape(vB, (1, 1)), [3, 2])
+            tC = CuTensor(reshape(vC, (1, 4)), [3, 1])
+            mul!(tC, tA, tB)
+            tA2 = CuTensor(copy(vA), [1, 2])
+            tB2 = CuTensor(copy(vB), [3, 2])
+            tD = CuTensor(copy(vD), [3, 1])
+            mul!(tD, tA2, tB2)
+            @test tD.data ≈ tD.data
+        end
     end
-end
 
 end

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 73.50%. Comparing base (5461475) to head (bc175aa).
Report is 4 commits behind head on master.

Additional details and impacted files

@@           Coverage Diff           @@
##           master    #2650   +/-   ##
=======================================
  Coverage   73.50%   73.50%           
=======================================
  Files         158      158           
  Lines       15321    15324    +3     
=======================================
+ Hits        11261    11264    +3     
  Misses       4060     4060           

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

[github-actions]

CUDA.jl Benchmarks

Benchmark suite	Current: bc175aa	Previous: f62af73	Ratio
latency/precompile	46455682564.5 ns	46667053506.5 ns	1.00
latency/ttfp	6952404648 ns	6954689688 ns	1.00
latency/import	3610907756 ns	3631856123 ns	0.99
integration/volumerhs	9625805.5 ns	9624743.5 ns	1.00
integration/byval/slices=1	146896 ns	146953 ns	1.00
integration/byval/slices=3	425201 ns	425334 ns	1.00
integration/byval/reference	144890 ns	145208 ns	1.00
integration/byval/slices=2	286139 ns	286016 ns	1.00
integration/cudadevrt	103259 ns	103424 ns	1.00
kernel/indexing	14102 ns	14214 ns	0.99
kernel/indexing_checked	14572 ns	14910 ns	0.98
kernel/occupancy	637.8 ns	637.5449101796407 ns	1.00
kernel/launch	2016.1 ns	2102 ns	0.96
kernel/rand	15027 ns	18239 ns	0.82
array/reverse/1d	19976 ns	19474 ns	1.03
array/reverse/2d	25056 ns	23910 ns	1.05
array/reverse/1d_inplace	10901.333333333334 ns	10670 ns	1.02
array/reverse/2d_inplace	11355 ns	12291 ns	0.92
array/copy	21126 ns	20955 ns	1.01
array/iteration/findall/int	159837 ns	155336 ns	1.03
array/iteration/findall/bool	140579 ns	133979 ns	1.05
array/iteration/findfirst/int	155148.5 ns	154049 ns	1.01
array/iteration/findfirst/bool	156103 ns	153056 ns	1.02
array/iteration/scalar	71680 ns	61530 ns	1.16
array/iteration/logical	217035.5 ns	202309 ns	1.07
array/iteration/findmin/1d	42575 ns	37878 ns	1.12
array/iteration/findmin/2d	94370.5 ns	93537 ns	1.01
array/reductions/reduce/1d	45523 ns	37060.5 ns	1.23
array/reductions/reduce/2d	50983.5 ns	50765 ns	1.00
array/reductions/mapreduce/1d	39076.5 ns	36727 ns	1.06
array/reductions/mapreduce/2d	51964 ns	42618.5 ns	1.22
array/broadcast	21183 ns	20743 ns	1.02
array/copyto!/gpu_to_gpu	11902.5 ns	13730.5 ns	0.87
array/copyto!/cpu_to_gpu	211185 ns	207788 ns	1.02
array/copyto!/gpu_to_cpu	245806 ns	243117 ns	1.01
array/accumulate/1d	109476 ns	108517 ns	1.01
array/accumulate/2d	80318 ns	79641 ns	1.01
array/construct	1315.5 ns	1306.5 ns	1.01
array/random/randn/Float32	49329.5 ns	43234.5 ns	1.14
array/random/randn!/Float32	26838 ns	26328 ns	1.02
array/random/rand!/Int64	27100 ns	27074 ns	1.00
array/random/rand!/Float32	8577.666666666666 ns	8647.666666666666 ns	0.99
array/random/rand/Int64	30079 ns	29948 ns	1.00
array/random/rand/Float32	13234 ns	13039 ns	1.01
array/permutedims/4d	61252 ns	60777 ns	1.01
array/permutedims/2d	55497 ns	55571 ns	1.00
array/permutedims/3d	56650 ns	55866 ns	1.01
array/sorting/1d	2777654 ns	2764795 ns	1.00
array/sorting/by	3368720 ns	3367795 ns	1.00
array/sorting/2d	1085641.5 ns	1084334 ns	1.00
cuda/synchronization/stream/auto	1059.9 ns	1052.3 ns	1.01
cuda/synchronization/stream/nonblocking	6416 ns	6404.4 ns	1.00
cuda/synchronization/stream/blocking	796.6116504854369 ns	810.0736842105263 ns	0.98
cuda/synchronization/context/auto	1177.7 ns	1185.6 ns	0.99
cuda/synchronization/context/nonblocking	6656.8 ns	6726.6 ns	0.99
cuda/synchronization/context/blocking	899.6458333333334 ns	925.975 ns	0.97

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