v4.1.0
CUDA v4.1.0
Closed issues:
- ERROR: LoadError: bin\cublas64_11.dll when installing CUDA (#1750)
- System-wide CUDA in LD_LIBRARY_PATH breaks CUBLAS (#1755)
- CuDeviceTexture getindex breaks when executed on the CPU (#1757)
- cuDNN.version can cause Julia to crash, missing
cudnn_ops_infer64_8.dll
(#1777) - cuDNN compile error "ERROR: LoadError: ArgumentError: invalid version string: local" (#1783)
- "Error: No CUDA Runtime library found" for ≥v4.0.0 (#1808)
- sqrt broken in kernels 'Format of __nvvm__reflect function not recognized' (#1817)
Merged pull requests:
- Add support for CUDA 12.0. (#1742) (@maleadt)
- Add more fixes and tests for CUDA toolkit 12.0 (#1756) (@amontoison)
- Update manifest (#1758) (@github-actions[bot])
- Fix test/cusparse/interfaces.jl (#1762) (@amontoison)
- Simplify the function sig. (#1763) (@N5N3)
- Update manifest (#1770) (@github-actions[bot])
- Make versioninfo() resilient against NVML EPERM. (#1771) (@maleadt)
- Move CUDAKernels to CUDA.jl (#1772) (@vchuravy)
- [CUSPARSE] Improve conversion and tests between sparse matrices (#1774) (@amontoison)
- Use geam for + and - operations with CuMatrix{<:CublasFloat} (#1775) (@amontoison)
- Update manifest (#1776) (@github-actions[bot])
- Update manifest (#1781) (@github-actions[bot])
- Update manifest (#1784) (@github-actions[bot])
- [CUSPARSE] Update preconditioners.jl (#1785) (@amontoison)
- [CUSOLVER] Avoid the conversion to CSR format for reordering routines (#1786) (@amontoison)
- Bump GPUCompiler. (#1787) (@maleadt)
- Remove unneeded variable. (#1788) (@maleadt)
- [CUSPARSE] Update conversions.jl (#1791) (@amontoison)
- Update to CUDNN 8.8.1 for CUDA 12 compatibility. (#1792) (@maleadt)
- Add support for CUDA 12.1 (#1793) (@maleadt)
- [CUSPARSE] Interface color reordering (#1794) (@amontoison)
- [CUSPARSE] Interface gtsv2 (#1795) (@amontoison)
- Update manifest (#1796) (@github-actions[bot])
- Adapt to GPUCompiler 0.18 (#1799) (@maleadt)
- Follow
Array
's behavior when initializing (#1800) (@lcw) - [CUSOLVER] Support A \ b for rectangular matrices (#1802) (@amontoison)
- Use symbols instead of values when emitting code, when possible. (#1804) (@maleadt)
- Refactor CI pipeline a little. (#1805) (@maleadt)
- [CUSOLVER] Improve the dispatch for LAPACK routines (#1806) (@amontoison)
- Diagonal for lower triangular of LU decomposition set incorrectly (#1813) (@tgymnich)
- CompatHelper: add new compat entry for "KernelAbstractions" at version "0.9" (#1824) (@github-actions[bot])
- Rebuild CUPTI API with support for STRUCT_SIZE (#1827) (@vchuravy)
- Release CUDA 4.1 (#1828) (@vchuravy)