KernelInterface-v0.4.2

KernelInterface KernelInterface-v0.4.2

Diff since KernelInterface-v0.4.1

Closed issues:

• Add generic print infrastructure (#17)
• Support generic error infrastructure (#18)
• weird issue with index macro (#30)
• Better stacktraces (#44)
• Stencil example (#49)
• Compilation error with dynamic ndrange (#54)
• wait some times fails (#90)
• Illegal memory access was encountered (#91)
• @print functionality inconsistent with CUDADevice() (#112)
• Broadcasting fails with CUDA: unsupported dynamic function invocation (call to overdub) (#113)
• Degraded performance using kwargs in kernels (#114)
• User-facing API like vmap (#117)
• Error on CPU dynamic dispatch (#119)
• eps(::AbstractFloat) inside a kernel doesn't compile on Julia 1.5.1 (#127)
• uniform memory and @synchronize inside an if statement (#132)
• Use with nested StructArrays (#138)
• Broken link to @uniform in docs (#151)
• Suggestion: Better ergonomics around collecting multiple kernels (#162)
• specify ngroups as alternative to ndrange (#173)
• Pull @unroll out into its own tiny package (#199)
• nested task error: unsupported or misplaced expression aliasscope when using julia --compile=min (#232)
• Reduction interface (#234)
• Docs for AMD support. (#235)
• Inconsistent KernelError on the GPU telling me to use return in my kernel (#248)
• Methods for GPU not properly invalidated upon redefinition (#250)
• 3d convolutions (#259)
• @synchronize inside while loops doesn't work on CPUs (#262)
• Function calls in kernels not called (#263)
• CI on nigthly is broken due to AMDGPU (#271)
• Synchronization in a nested while loop causes loss of threadID (#274)
• creation of a subarray for function calls results in an error with throw_checksize_error (#275)
• @uniforms dependent on @index values fail (#277)
• ROCMKernels doesn't support 1.7 (#284)
• Types as arguments fail on GPU (#285)
• tuples with multiple functions cannot be accessed on the GPU with a complicated symbol (#286)
• Function redefinion on GPU fails with indirect call (#287)
• Inconsistent behavior with multiple GPU (#289)
• wait(kernel(...) performs much worse on Julia v1.7 (#290)
• Unnecessary recompilations using CUDAKernels (#293)
• InvalidIRError: Reason: unsupported dynamic function invocation (call to print_to_string(xs...) (#295)
• Enzyme gradient always seems to stall (#298)
• Include overview in the README (#302)
• Add back support for force-inlining (#313)
• Roadmap to 1.0 (#319)
• Doc: what does CPU() do ? (#324)
• StackOverflow for get_device on ROCArray (#325)
• Simple Block Reduce Fails when using while loops (#330)
• Missing docstring for KernelAbstractions.partition (#333)
• @print not functional on AMDGPU (#339)
• Using RuntimeGeneratedFunctions gives a GPU compilation error (#347)
• Const on 1.9 is failing on CI (#350)
• KernelAbstractions.@spawn backend ... for easy Julia task based concurrency (#358)
• StackOverflow Error on get_backend(cu(rand(3))) (#366)
• Offsetting kernel indices (#384)
• How do I detect what GPU is installed on a host? (#385)
• Add API for device selection (#395)
• Index type (#396)
• Possible error due to KernelAbstractions while using VSCode debugger (#405)
• Variable scoping issue leading to unexpected UndefVarError on CPU (#413)
• exposing warp-level semantics (#420)
• Julia 1.11 private tests are failing (#433)
• Looping over high-dimensional arrays (#438)
• Allow return or return nothing in kernels (#443)
• BUG: @synchronize inside loop in CPU is not working (#465)
• Inconsistent results across platforms and datatypes on CPU backend (#468)
• Significant perf drop when using dynamic ranges in GPU kernel (#470)
• @index cannot be used in the CPU if is not a direct rhs of an assignment (#471)
• simple batched dot kernel is ~1.7x slower with Const on Titan RTX (#479)
• issue with using enzyme on kernel abstractions on CUDA backend (#480)
• LLVM error when using Atomix.@atomic with FP16 (#486)
• Enzyme autodiff produces out-of-bounds error for some kernels. (#493)
• Atomic operations on complex numbers (#497)
• Error when using @Const macro with Julia --check-bounds=no startup option (#498)
• How (if possible) can we get rid of SparseArrays and StaticArrays? (#506)
• amd gpu give different results when nested loop is used (#517)
• 1.11: reduce_private test generates gcframe under --check-bounds=yes (#528)
• Qualified KA.@index macro not detected. (#542)
• localmem does not work for cpu backend? (#544)
• Kernel only gives proper result when adding @print() (#547)
• Complex struct as input to a kernel (#574)
• ndrange semantics and performance surprise (#585)
• Codecov failure (#586)
• Vendor-agnostic device information (#617)
• get_backend support for MVector (#620)
• options for more aggressive inlining (#628)
• API for # of cores/multiprocessors (#631)
• Unexpected behavior for KA.device!(backend, num) (#636)
• WaterLily performance regression on CPU multi-threading backend for Julia 1.12.0 (#644)
• UndefVarError when variable is defined (#647)
• @Const can lead to wrong results on CPU in Julia 1.11/1.12 (#652)
• How to query a backend's native/recommended groupsize? (#740)
• show of GeneratedKernelError{Msg} throws for an unbound Msg (#897)