Search results

[BUG] Compile time grows much faster than numba-cuda when a kernel composes many non-trivial device functions

bugSomething isn't workingSomething isn't working

Status: Open.

#249 In NVIDIA/numba-cuda-mlir;

· raphaelchinchilla opened on Aug 1, 2026

[BUG] A closure-captured array is copied to the device heap per thread, per launch, and never freed — kernels die with CUDA_ERROR_ILLEGAL_ADDRESS after N launches

bugSomething isn't workingSomething isn't working

Status: Open.

#247 In NVIDIA/numba-cuda-mlir;

· raphaelchinchilla opened on Aug 1, 2026

[FEA] Enhance Numba-CUDA MLIR Type Inferencing

Status: Open.

#237 In NVIDIA/numba-cuda-mlir;

· atmnp opened on Jul 21, 2026

Built-in float min and max return NaN when either operand is NaN

Status: Open.

#229 In NVIDIA/numba-cuda-mlir;

· ccam80 opened on Jul 14, 2026

lineinfo attributes lines from inlined cross-file device functions to the wrong source file

Status: Open.

#224 In NVIDIA/numba-cuda-mlir;

· ccam80 opened on Jul 13, 2026

RecursionError compiling kernels with large control-flow graphs

Status: Open.

#222 In NVIDIA/numba-cuda-mlir;

· ccam80 opened on Jul 13, 2026

fastmath has almost no effect on generated code, and selective flag sets are rejected

Status: Open.

#220 In NVIDIA/numba-cuda-mlir;

· ccam80 opened on Jul 12, 2026

llvm70-translate emits nvvmir.version debug metadata 0.0, which recent libnvvm rejects

Status: Open.

#218 In NVIDIA/numba-cuda-mlir;

· ccam80 opened on Jul 12, 2026

Port remaining Numba-CUDA test modules to numba_cuda_tests for coverage parity (follow-up to numba-cuda#515)

Status: Open.

#215 In NVIDIA/numba-cuda-mlir;

· CodersAcademy006 opened on Jul 11, 2026

[BUG] Record types still yield illegal memory access

bugSomething isn't workingSomething isn't working

could-not-reproduceReported problem no longer reproduces on current main; verify before closingReported problem no longer reproduces on current main; verify before closing

Status: Open.

#214 In NVIDIA/numba-cuda-mlir;

· c200chromebook opened on Jul 10, 2026

[FEA] Ufunc support

Status: Open.

#210 In NVIDIA/numba-cuda-mlir;

· gmarkall opened on Jul 9, 2026

[FEA] Allow libraries that implement __array_ufunc__ to override CUDAUFunc

Status: Open.

#209 In NVIDIA/numba-cuda-mlir;

· gmarkall opened on Jul 9, 2026