v5.8.3

CUDA v5.8.3

Diff since v5.8.2

Merged pull requests:

• More tests for diagm (#2791) (@kshyatt)
• Add JLD2 to test env (#2792) (@christiangnrd)
• cuTENSOR: Destroy plan description and preference after construction. (#2794) (@maleadt)
• More tests for sparse matrix dimension checks (#2796) (@kshyatt)
• Better error messages and tests for sm2 (#2797) (@kshyatt)
• Reorganize interfaces tests, lower allocations (#2799) (@kshyatt)
• Cleanup and less memory use for cusparse linalg tests (#2800) (@kshyatt)
• Separately version the CUDA compiler (#2801) (@maleadt)
• Remove shape-preserving Diagonal conversion constructors. (#2805) (@maleadt)
• More accumulation and reduction benchmarks (#2808) (@christiangnrd)
• Rationalize and try to fix failing ldiv tests (#2809) (@kshyatt)
• Simplify specifying benchmark output file (#2814) (@christiangnrd)
• Add KA unified memory support (#2819) (@christiangnrd)
• Augment docs about setting runtime version (#2822) (@david-macmahon)
• Fix a list numbering problem in docs (#2824) (@david-macmahon)
• Move things to GPUToolbox. (#2826) (@maleadt)
• Initial compatibility with CUDA 13 (#2834) (@maleadt)

Closed issues:

• Array constructors for ones, zeros, rand, ... (#159)
• CuSparse documentation (#135)
• gemm_strided_batched throws error on windows (#132)
• Documentation: An example for allocating Unified Memory arrays (#33)
• norm function errors on big arrays (#598)
• CuSparse factorizations (#1396)
• opnorm(::CuMatrix, p) for p = (1, Inf) (#1533)
• CUSPARSE: support broadcasting for CuSparseVectors (#2699)
• Remove erroneous CuArray(::Diagonal) methods (#2734)
• Matrix-Matrix-Multiplication fails with CuSparseMatrixBSR. (#2745)
• Possible CPU memory leak in cuTENSOR plans (#2793)
• Test fail for libraries/cublas/level1 (#2810)
• CI is failing due to CUSPARSEVEC (#2817)
• Support for CUDA 13 (#2831)