v0.10.11

OpenCL v0.10.11

Diff since v0.10.10

Merged pull requests:

• Bump actions/upload-artifact from 4 to 7 (#451) (@dependabot[bot])
• Add floating-point atomic add/sub and min/max intrinsics (#453) (@michel2323)
• handle missing vendor info in OpenCL version string gracefully (#457) (@simeonschaub)
• Update version to 0.10.10 (#461) (@christiangnrd)
• Float64 KA (#462) (@christiangnrd)
• Update Statistics requirement to 1 (#466) (@dependabot[bot])
• Update IOCapture requirement to 1.0.0 (#467) (@dependabot[bot])
• don't return reshaped array from mapreducedim! (#468) (@simeonschaub)
• Bump version (#470) (@christiangnrd)

Closed issues:

• Support subgroups (#284)
• AcceleratedKernels.cumsum runs endlessly when using pocl's cpu backend on Intel CPU (#388)

v0.10.10

OpenCL v0.10.10

Diff since v0.10.9

Closed issues:

• Native codegen for non-SPIRV back-ends (#234)
• PoCL: Float16 support (#312)
• PoCL: Occasional segfault on Windows when using USM buffers (#317)
• Codegen using SPIRV-LLVM Translator regressed (#349)
• Fix and slim down CI (#366)
• LLVM SPIR-V backend crashes on functions returning small static arrays (e.g. SMatrix{1,1,Float32}) (#422)

v0.10.9

OpenCL v0.10.9

Diff since v0.10.8

Merged pull requests:

• Add atomic float support (#399) (@michel2323)
• [SPIRVIntrinsics] Fix @device_override (#404) (@christiangnrd)
• Disable broken vectorization tests on Windows and use windows-2022 in CI (#405) (@christiangnrd)
• Fix macos CI (#406) (@christiangnrd)
• Bump actions/checkout from 5 to 6 (#407) (@dependabot[bot])
• Add relevant package versions to versioninfo (#409) (@christiangnrd)
• Fix CI for locally built POCL (#412) (@christiangnrd)
• [OpenCL] Subgroup support (#413) (@christiangnrd)
• Fixes found with JETLS (#416) (@christiangnrd)
• [SPIRVIntrinsics] Sub-group shuffle & barrier (#420) (@christiangnrd)

Closed issues:

• PoCL/Windows: undefined symbol: _initterm_e (#393)
• [CI] Local POCL builds failing (#408)

v0.10.8

OpenCL v0.10.8

Diff since v0.10.7

Merged pull requests:

• SIMD support for math intrinsics (#379) (@simeonschaub)
• [KA] support unified memory API (#394) (@simeonschaub)

Closed issues:

• support vectorized math intrinsics (#376)

v0.10.7

OpenCL v0.10.7

Diff since v0.10.6

Merged pull requests:

• change mangling of pointers (#395) (@simeonschaub)
• Reduce macOS runs on CI (#397) (@christiangnrd)
• Bump peter-evans/find-comment from 3 to 4 (#400) (@dependabot[bot])
• Bump peter-evans/create-or-update-comment from 4 to 5 (#401) (@dependabot[bot])
• fix rand(::AbstractRange{Int64}) (#402) (@simeonschaub)

Closed issues:

• Updating SPIRV_Tools_jll breaks Rusticl backend (#398)

v0.10.6

OpenCL v0.10.6

Diff since v0.10.5

Merged pull requests:

• implement device-side RNG (#365) (@simeonschaub)
• add Float16 math functions (#368) (@simeonschaub)
• Bump actions/checkout from 4 to 5 (#372) (@dependabot[bot])
• [SPIRVIntrinsics] GPUToolbox compat (#373) (@christiangnrd)
• Use macos-15-intel (#375) (@christiangnrd)
• GPUToolbox compat again (#378) (@christiangnrd)
• Fix return_type test failure in 1.12 (#380) (@christiangnrd)
• Fix typo in Float64 unsupported error (#381) (@christiangnrd)
• Various test improvements (#382) (@christiangnrd)
• Test improvement follow-up (#387) (@christiangnrd)
• 1.12 (#391) (@christiangnrd)
• Use --quickfail in CI (#392) (@christiangnrd)
• Version 0.10.6 (#396) (@simeonschaub)

Closed issues:

• wrong definition for rem (#374)
• findmax broken in Julia 1.12 (#384)

v0.10.5

OpenCL v0.10.5

Diff since v0.10.4

Merged pull requests:

• TagBot: use DOCUMENTER_KEY (#362) (@simeonschaub)
• add profiling docs (#363) (@simeonschaub)
• minor README fixes (#364) (@simeonschaub)
• CI: Build PoCL without hwloc. (#367) (@maleadt)
• allow disabling validation (#369) (@simeonschaub)
• Version 0.10.5 (#370) (@simeonschaub)

v0.10.4

OpenCL v0.10.4

Diff since v0.10.3

Merged pull requests:

• Add sources for SPIRVIntrinsics (#321) (@vchuravy)
• Use stacked method tables (#325) (@vchuravy)
• Allow direct host-interactions with SVM-backed arrays (#336) (@VarLad)
• Track memory type when converting arguments to pointers (#341) (@maleadt)
• Port control barriers to SPIR-V, add memory barriers. (#343) (@maleadt)
• fix broadcasting for differently backed CLArrays of different dimensions (#345) (@simeonschaub)
• Int64 atomics (#347) (@simeonschaub)
• Disable broken hardware-based linear KA index. (#348) (@maleadt)
• pass extensions argument through @opencl (#355) (@simeonschaub)
• set up Documenter (#357) (@simeonschaub)
• add backend to COMPILER_KWARGS (#358) (@simeonschaub)
• Bump actions/checkout from 4 to 5 (#359) (@dependabot[bot])
• fix docs setup (#360) (@simeonschaub)
• Version 0.10.4 (#361) (@simeonschaub)

Closed issues:

• Invalid device buffer to CPU pointer conversions are allowed (#294)
• PoCL: Occasional __init__ failure during wrapper generation (#318)
• SVM pointer conversion and other related issues (#337)
• Behaviour discrepancy between the REPL and the testing environment, as well as between logically similar kernels. (#339)
• issues with Int64 atomic_cmpxchg! (#353)
• add some proper documentation (#354)

v0.10.3

OpenCL v0.10.3

Diff since v0.10.2

Merged pull requests:

• Switch to the LLVM SPIR-V back-end. (#285) (@maleadt)
• CI: Fixed arch for macOS jobs (#286) (@stemann)
• Avoid promition to Int32 in work_items functions (#287) (@vchuravy)
• Remove access keyword from examples (#288) (@blegat)
• Bump OpenCL_jll to 2024.10.24 (#289) (@blegat)
• Update jupyter notebooks (#291) (@blegat)
• Typo fix : pivate -> private (#292) (@blegat)
• cl.LocalMemory -> CLPtr (#293) (@blegat)
• Add warning about using pocl_jll on top of script (#295) (@blegat)
• Ptr -> CLPtr in examples (#296) (@blegat)
• Add pocl upstream as a CI (#298) (@vchuravy)
• [OpenCLKernel] Use barrier with LOCAL_MEM_FENCE (#300) (@vchuravy)
• Pass actually pointer typed things to printf (#302) (@vchuravy)
• Bump SPIRVIntrinsics (#306) (@christiangnrd)
• Fix typo in intrinsic mangling, and always test using latest subpackage. (#307) (@maleadt)
• Upgrade and switch to the non-unified SPIRV LLVM Translator. (#308) (@maleadt)
• Update PoCL to v7. (#309) (@maleadt)
• Silence 1.12 warnings (#310) (@christiangnrd)
• Add Buffer Device Address Backend (#311) (@VarLad)
• Windows support (#313) (@maleadt)
• CompatHelper: bump compat for SPIRVIntrinsics to 0.2, (keep existing compat) (#320) (@github-actions[bot])
• Cast sizeof(::LocalMem) to Int (#323) (@blegat)
• [NFC] Docstring typo (#327) (@christiangnrd)
• Add JLD2 to test env (#329) (@christiangnrd)
• Fix OpenCL on 1.12 (#333) (@VarLad)
• KA.__synchronize, add GLOBAL_MEM_FENCE semantics (#338) (@vchuravy)

Closed issues:

• Buffers aren't released (#208)
• OOB trap is fatal for CPU back-ends (#249)
• Fails to create array on Rusticl driver (#255)
• Atlernative allocation strategies (#256)
• Support Atomix.jl (#280)
• Support cl_mem based buffers using cl_ext_buffer_device_address extension (#281)
• ndrange provided in KernelAbstractions kernels is broken (#283)
• Invalid instruction with pocl package from ArchLinux (#290)
• Failure for POCL without inbounds (#299)
• PoCL: Linking can fail due to missing libc (#303)
• PoCL's CPU platform gives isgpu(backend) == true (#304)
• PoCL 7.0 (#305)
• PoCL: Ship additional system libraries on macOS (#315)
• Printing failure for LocalMem (#322)
• does not support USM or coarse-grained SVM (#324)
• Look into Vulkan shader compilers (#331)
• TypeError in add_ptr in Julia 1.12 (#332)

v0.10.2

OpenCL v0.10.2

Diff since v0.10.1

Merged pull requests:

• Revamp memory management, and add USM support. (#264) (@VarLad)
• Elide bounds checks when kernels contains manual ones. (#276) (@maleadt)
• Fix CI again. (#277) (@maleadt)

Closed issues:

• Simplify use of OpenCL extensions (#257)