Add some matrix functions for symm/herm

[kshyatt]

No description provided.

[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/cusolver/linalg.jl b/lib/cusolver/linalg.jl
index 4ed2a372d..3aad894f3 100644
--- a/lib/cusolver/linalg.jl
+++ b/lib/cusolver/linalg.jl
@@ -143,11 +143,11 @@ end
 # matrix functions
 for func in (:(Base.exp), :(Base.cos), :(Base.sin), :(Base.tan), :(Base.cosh), :(Base.sinh), :(Base.tanh), :(Base.atan), :(Base.asinh), :(Base.atanh), :(Base.cbrt))
     @eval begin
-        function ($func)(A::Symmetric{T, <:StridedCuMatrix}) where {T<:BlasReal}
+        function ($func)(A::Symmetric{T, <:StridedCuMatrix}) where {T <: BlasReal}
             F = eigen(A)
             return Symmetric((F.vectors * Diagonal(($func).(F.values))) * F.vectors')
         end
-        function ($func)(A::Hermitian{T, <:StridedCuMatrix}) where {T<:BlasReal}
+        function ($func)(A::Hermitian{T, <:StridedCuMatrix}) where {T <: BlasReal}
             F = eigen(A)
             return Hermitian((F.vectors * Diagonal(($func).(F.values))) * F.vectors')
         end

[github-actions]

CUDA.jl Benchmarks

Benchmark suite	Current: 2e337d6	Previous: 1e35ff7	Ratio
latency/precompile	56526512594 ns	57073264350 ns	0.99
latency/ttfp	8372733969 ns	8361242115.5 ns	1.00
latency/import	4495232086 ns	4520871512 ns	0.99
integration/volumerhs	9623228.5 ns	9609145.5 ns	1.00
integration/byval/slices=1	146712 ns	146784 ns	1.00
integration/byval/slices=3	425560 ns	425930 ns	1.00
integration/byval/reference	144948 ns	144913 ns	1.00
integration/byval/slices=2	286219 ns	286275 ns	1.00
integration/cudadevrt	103443 ns	103477 ns	1.00
kernel/indexing	14259 ns	14088 ns	1.01
kernel/indexing_checked	15019 ns	14920 ns	1.01
kernel/occupancy	670.0696202531645 ns	670.5283018867924 ns	1.00
kernel/launch	2147.5555555555557 ns	2192.1111111111113 ns	0.98
kernel/rand	15861 ns	18597.5 ns	0.85
array/reverse/1d	19880 ns	19990 ns	0.99
array/reverse/2dL_inplace	66766 ns	66851 ns	1.00
array/reverse/1dL	69999 ns	70214 ns	1.00
array/reverse/2d	21834 ns	21764 ns	1.00
array/reverse/1d_inplace	9955 ns	9644 ns	1.03
array/reverse/2d_inplace	13229 ns	11083 ns	1.19
array/reverse/2dL	73829 ns	73680.5 ns	1.00
array/reverse/1dL_inplace	66902 ns	66780 ns	1.00
array/copy	20788 ns	20656 ns	1.01
array/iteration/findall/int	157366 ns	157234 ns	1.00
array/iteration/findall/bool	139781 ns	139637.5 ns	1.00
array/iteration/findfirst/int	162437.5 ns	161491 ns	1.01
array/iteration/findfirst/bool	162944.5 ns	161981.5 ns	1.01
array/iteration/scalar	73361 ns	72914 ns	1.01
array/iteration/logical	216385.5 ns	215503 ns	1.00
array/iteration/findmin/1d	51485 ns	52893.5 ns	0.97
array/iteration/findmin/2d	96578 ns	96673.5 ns	1.00
array/reductions/reduce/Int64/1d	43801 ns	43374 ns	1.01
array/reductions/reduce/Int64/dims=1	48298 ns	44924.5 ns	1.08
array/reductions/reduce/Int64/dims=2	61637 ns	61289 ns	1.01
array/reductions/reduce/Int64/dims=1L	88743 ns	89013 ns	1.00
array/reductions/reduce/Int64/dims=2L	88050 ns	88275 ns	1.00
array/reductions/reduce/Float32/1d	37741.5 ns	37043 ns	1.02
array/reductions/reduce/Float32/dims=1	42392.5 ns	43018 ns	0.99
array/reductions/reduce/Float32/dims=2	60016 ns	59774 ns	1.00
array/reductions/reduce/Float32/dims=1L	52204 ns	52409 ns	1.00
array/reductions/reduce/Float32/dims=2L	72239 ns	72278 ns	1.00
array/reductions/mapreduce/Int64/1d	43587 ns	43540 ns	1.00
array/reductions/mapreduce/Int64/dims=1	52873.5 ns	45057.5 ns	1.17
array/reductions/mapreduce/Int64/dims=2	61427 ns	61470 ns	1.00
array/reductions/mapreduce/Int64/dims=1L	88867 ns	88923 ns	1.00
array/reductions/mapreduce/Int64/dims=2L	88334.5 ns	88349 ns	1.00
array/reductions/mapreduce/Float32/1d	37950.5 ns	36698 ns	1.03
array/reductions/mapreduce/Float32/dims=1	41864.5 ns	41442 ns	1.01
array/reductions/mapreduce/Float32/dims=2	59855 ns	59908 ns	1.00
array/reductions/mapreduce/Float32/dims=1L	52441 ns	52585 ns	1.00
array/reductions/mapreduce/Float32/dims=2L	72155 ns	72014 ns	1.00
array/broadcast	19857 ns	20078 ns	0.99
array/copyto!/gpu_to_gpu	11329 ns	12908 ns	0.88
array/copyto!/cpu_to_gpu	214313.5 ns	213437 ns	1.00
array/copyto!/gpu_to_cpu	283907 ns	283206 ns	1.00
array/accumulate/Int64/1d	124824 ns	124198 ns	1.01
array/accumulate/Int64/dims=1	83387 ns	83165 ns	1.00
array/accumulate/Int64/dims=2	157802 ns	157631 ns	1.00
array/accumulate/Int64/dims=1L	1709895.5 ns	1709733 ns	1.00
array/accumulate/Int64/dims=2L	966630 ns	966057.5 ns	1.00
array/accumulate/Float32/1d	109134 ns	108414 ns	1.01
array/accumulate/Float32/dims=1	80525 ns	79731.5 ns	1.01
array/accumulate/Float32/dims=2	147654.5 ns	146657 ns	1.01
array/accumulate/Float32/dims=1L	1618976.5 ns	1616606.5 ns	1.00
array/accumulate/Float32/dims=2L	698299.5 ns	697417 ns	1.00
array/construct	1267.7 ns	1271.5 ns	1.00
array/random/randn/Float32	48182 ns	45612 ns	1.06
array/random/randn!/Float32	24994 ns	24822 ns	1.01
array/random/rand!/Int64	27284 ns	27264 ns	1.00
array/random/rand!/Float32	8850 ns	8854 ns	1.00
array/random/rand/Int64	31309.5 ns	29823 ns	1.05
array/random/rand/Float32	13005 ns	13073 ns	0.99
array/permutedims/4d	60165 ns	59525 ns	1.01
array/permutedims/2d	53938.5 ns	53919 ns	1.00
array/permutedims/3d	54676 ns	54583 ns	1.00
array/sorting/1d	2757835 ns	2757051 ns	1.00
array/sorting/by	3344388.5 ns	3344047 ns	1.00
array/sorting/2d	1081134 ns	1080794 ns	1.00
cuda/synchronization/stream/auto	1037.1 ns	1034 ns	1.00
cuda/synchronization/stream/nonblocking	8373.400000000001 ns	8105 ns	1.03
cuda/synchronization/stream/blocking	833.2760416666667 ns	796.4842105263158 ns	1.05
cuda/synchronization/context/auto	1172.2 ns	1198.2 ns	0.98
cuda/synchronization/context/nonblocking	7213 ns	8018.6 ns	0.90
cuda/synchronization/context/blocking	916.9090909090909 ns	918.6428571428571 ns	1.00

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

[codecov]

Codecov Report

❌ Patch coverage is 92.85714% with 1 line in your changes missing coverage. Please review.
✅ Project coverage is 89.49%. Comparing base (11bed80) to head (a4de770).
⚠️ Report is 1 commits behind head on master.

Files with missing lines	Patch %	Lines
lib/cusolver/linalg.jl	92.85%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #2962      +/-   ##
==========================================
+ Coverage   89.36%   89.49%   +0.12%     
==========================================
  Files         150      150              
  Lines       13119    13133      +14     
==========================================
+ Hits        11724    11753      +29     
+ Misses       1395     1380      -15     

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

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[maleadt]

LGTM.

Isn't this generic enough for GPUArrays.jl though?

[kshyatt]

Do all the GPUArray backends have a symmetric eigensolver? That’s what this relies on

…

On Wed, Nov 19, 2025 at 1:17 PM Tim Besard ***@***.***> wrote: ***@***.**** approved this pull request. LGTM. Isn't this generic enough for GPUArrays.jl though? — Reply to this email directly, view it on GitHub <#2962 (review)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAGKJY4UYEKDDKQLZJY2AML35RNWHAVCNFSM6AAAAACLPGH4ACVHI2DSMVQWIX3LMV43YUDVNRWFEZLROVSXG5CSMV3GSZLXHMZTIOBSGQYDCOJSGU> . You are receiving this because you authored the thread.Message ID: ***@***.***>