Issues

• array_scan reports CUDA error 402 during graph capture

  bug
  NVIDIA/warp#2319 · Filter by author shi-eric opened on Nov 13, 2026

• CUDA kernels without shared tiles reserve static shared memory for the tile allocator

  bugtile
  Bug

  NVIDIA/warp#2318 · Filter by author adenzler-nvidia opened on Nov 13, 2026

• Functions passed to wp.tile_map are not hashed: kernels differing only in a captured constant reuse stale code

  bugtile
  Bug

  NVIDIA/warp#2317 · Filter by author adenzler-nvidia opened on Nov 13, 2026

• tile_assign() gives wrong results when its destination is read between assignments

  bugtile
  Bug

  NVIDIA/warp#2316 · Filter by author shi-eric opened on Nov 13, 2026 · 1.21.0

• verify_autograd_array_access ignores arrays read or written by tile loads and stores

  NVIDIA/warp#2314 · Filter by author sohailshk opened on Nov 13, 2026

• verify_autograd_array_access does not flag arrays read as the left operand of a binary operator

  NVIDIA/warp#2313 · Filter by author sohailshk opened on Nov 13, 2026

• Document builtin overloads that jet spaces add at runtime, such as wp.value() on scalar jets

  docs
  Documentation

  NVIDIA/warp#2311 · Filter by author alecjacobson opened on Nov 13, 2026 · 1.20.0
  1

• Make deterministic sort-reduce scale with emitted records during CUDA graph replay

  NVIDIA/warp#2308 · Filter by author johnnynunez opened on Nov 12, 2026

• Element assignment and in-place tile solves give wrong results right after the tile is read

  bugtile
  Bug

  NVIDIA/warp#2306 · Filter by author shi-eric opened on Nov 12, 2026

• Tape.backward() starts a struct argument's value-field adjoints at the field values

  bug
  Bug

  NVIDIA/warp#2302 · Filter by author sohailshk opened on Nov 12, 2026

• Writing one component of a struct's vector or matrix field drops its gradient

  bug
  Bug

  NVIDIA/warp#2301 · Filter by author sohailshk opened on Nov 12, 2026

• Kernel requests too little shared memory when it moves a tile returned by a wp.func to shared storage

  bugtile
  Bug

  NVIDIA/warp#2294 · Filter by author adenzler-nvidia opened on Nov 11, 2026 · 1.20.0
  

• Support union type annotations on kernel and function parameters

  feature request
  Enhancement

  NVIDIA/warp#2293 · Filter by author cube1324 opened on Nov 11, 2026

• Support wp.handle in framework and DLPack converters

  feature request
  Enhancement

  NVIDIA/warp#2292 · Filter by author c0d1f1ed opened on Nov 11, 2026 · 1.21.0
  

• wp.sqrt() gradient is inf at x == 0, unlike wp.length()

  bug
  Bug

  NVIDIA/warp#2290 · Filter by author eric-heiden opened on Nov 11, 2026 · 1.20.0
  

• [BUG] wp.Mesh crashes on out-of-range triangle indices

  bugspatial acceleration
  Bug

  NVIDIA/warp#2288 · Filter by author christophercrouzet opened on Nov 11, 2026
  

• [REQ] Make wp.sparse.bsr_mm differentiable (tape.backward currently crashes)

  autodifffeature requestwarp.sparse
  Enhancement

  NVIDIA/warp#2287 · Filter by author alecjacobson opened on Nov 11, 2026
  

• [REQ] Make wp.sparse.bsr_mv differentiable

  autodifffeature requestwarp.sparse
  Enhancement

  NVIDIA/warp#2286 · Filter by author alecjacobson opened on Nov 11, 2026
  

• Add wp.no_grad(), Tape.get_active() and nested tapes

  feature request
  Enhancement

  NVIDIA/warp#2285 · Filter by author eric-heiden opened on Nov 10, 2026
  

• Loaded APIC graphs crash on CPU and return zeros on CUDA when a kernel samples a texture

  apicbug
  Bug

  NVIDIA/warp#2284 · Filter by author shi-eric opened on Nov 10, 2026

• Loaded APIC graph crashes when a kernel reads through a struct array's wp.array field

  apicbug
  Bug

  NVIDIA/warp#2283 · Filter by author shi-eric opened on Nov 10, 2026

• Register and indexed tile operations produce wrong results for padded vector and matrix arrays

  bugtile
  Bug

  NVIDIA/warp#2282 · Filter by author shi-eric opened on Nov 10, 2026 · 1.21.0
  

• Gradients of wp.frac() and vector wp.mod() are always zero

  autodiffbug
  Bug

  NVIDIA/warp#2280 · Filter by author Lynn-hh opened on Nov 10, 2026 · 1.20.0
  

• wp.autograd.jacobian() returns zero gradients for kernels whose adjoint reads an output (atomic_min/atomic_max)

  bug
  Bug

  NVIDIA/warp#2279 · Filter by author dalyathan opened on Nov 10, 2026
  1

  

• [REQ] Reserve less memory by default for rebuildable volumes

  feature request
  Enhancement

  NVIDIA/warp#2278 · Filter by author christophercrouzet opened on Nov 10, 2026