v4.0.1

GPUArrays v4.0.1

Diff since v4.0.0

Closed issues:

• generic mul! should support wrappers (#276)
• Base.copyto! override from CuArrays not currently in CUDA.jl (#294)

Merged pull requests:

• Deprecate the synchronize host API. (#298) (@maleadt)

v4.0.0

GPUArrays v4.0.0

Diff since v3.4.1

Closed issues:

• RNG issues (#277)
• Error running GPUArrays on Jupyter (#287)

Merged pull requests:

• Add randn function. (#280) (@Ellipse0934)
• Rework copyto!: support GPU wrappers, remove support for CPU wrappers (#284) (@maleadt)
• Fix compatibility of fallback matmatmul with wrapper types. (#289) (@maleadt)
• Bail out when performing empty broadcast. (#291) (@maleadt)
• Require an implementation of Base.copy. (#292) (@maleadt)
• Fix check for empty indexing. (#293) (@maleadt)

v3.4.1

GPUArrays v3.4.1

Diff since v3.4.0

Merged pull requests:

• Simplify use of Adapt (#281) (@maleadt)
• Use oneunit. (#282) (@maleadt)
• Bump Adapt. (#283) (@maleadt)

v3.4.0

GPUArrays v3.4.0

Diff since v3.3.0

Closed issues:

• mean doesn't work on 3.3.0 (#271)

Merged pull requests:

• Add inv for Triangular types and test. (#274) (@aterenin)
• Split the tests into suites using some metaprogramming. (#279) (@maleadt)

v3.3.0

GPUArrays v3.3.0

Diff since v3.2.0

Closed issues:

• Equality broken on 3.2.0 (#266)
• Reduce a broadcast (#268)

Merged pull requests:

• Implement + and - for UniformScaling and various wrappers (#256) (@haampie)
• Add copyto! for GPU to GPU for UpperTriangular. (#269) (@aterenin)
• Support and use broadcast with mapreduce. (#270) (@maleadt)

v3.2.0

GPUArrays v3.2.0

Diff since v3.1.0

Closed issues:

• Low-quality RNG (#257)
• Indexing with empty array fails (#264)

Merged pull requests:

• Add a function for temporarily allowing scalar indexing. (#255) (@maleadt)
• Specialize varargs version of mapreduce. (#258) (@maleadt)
• Avoid calling copyto kernel with no threads. (#259) (@maleadt)
• Test for implicit singleton dims with mapreducedim. (#260) (@maleadt)
• Fix #257 (#261) (@chrstphrbrns)
• Add copytri! and tests. (#262) (@aterenin)
• Return the original matrix from generic matmul (#263) (@haampie)
• Generalize _unsafe_getindex! to work for empty indexing in any dimension (#265) (@DrChainsaw)

v3.1.0

GPUArrays v3.1.0

Diff since v3.0.1

Merged pull requests:

• Fix 5 arg mul! (#253) (@haampie)
• Fix UniformScaling constructor + copyto! (#254) (@haampie)

v3.0.1

GPUArrays v3.0.1

Diff since v3.0.0

Merged pull requests:

• Add tests for the warning/erroring behavior of indexing. (#251) (@maleadt)
• generic_matmul! fixes (#252) (@maleadt)

v3.0.0

GPUArrays v3.0.0

Diff since v2.0.1

Closed issues:

• ERROR_ILLEGAL_ADDRESS (#17)
• Question about B-S benchmark (#19)
• logreg example does not want to work. + few qestions. (#26)
• Macro for loops (#33)
• Transpiler having troubles with broadcast (#41)
• opencl: broadcast complex64 fail (osX) (#46)
• free memory, check memory status (#51)
• ArrayFire backend support (#60)
• Execution difference between opencl backend and native Julia (#63)
• available_devices and all_devices don't work currently (#76)
• Broadcast still doesn't catch many cases (#107)
• gpu_call blocks + threads (#133)
• Improve adaption of GPUArrays (#142)
• cached_state not defined (#174)
• Incorrect results for adjoint of copyto! (#177)
• mapreduce et.al (maximum) with dims (#180)
• Different result for Array and GPUArrays with isapprox (#191)
• stable doc link is dead (#192)
• isapprox errors when given a transposed CuArray (#199)
• ishermitian causes dynamic call to simple_broadcast_index on Julia v1.2 (#201)
• 0D broadcast always return a container: ArrayStyle{CuArray} breaks dispatch? (#204)
• Broadcasting on nested ReshapedArrays and Adjoints falls back to DefaultArrayStyle (#244)

Merged pull requests:

• Improvements to array materialization (#224) (@maleadt)
• Properly handle reducing complex numbers with abs. (#226) (@maleadt)
• Properly handle reducing complex numbers with abs2 (#227) (@Sleort)
• Remove broken vector functionality. (#228) (@maleadt)
• Large clean-up (#229) (@maleadt)
• Remove non-1D indexing (#230) (@maleadt)
• Add Base.clamp! (#231) (@janEbert)
• Update testsuite docs (#232) (@janEbert)
• Simplify gpu_call (#233) (@maleadt)
• Remove BLAS methods/tests (#234) (@maleadt)
• Port some CuArrays linalg kernels. (#235) (@maleadt)
• Clean-ups. (#236) (@maleadt)
• Port functionality from CuArrays (#237) (@maleadt)
• Fixes to broadcast and indexing (#240) (@maleadt)
• Install TagBot as a GitHub Action (#241) (@JuliaTagBot)
• Add ability to seed the RNG, and do so initially. (#242) (@maleadt)
• Breaking gpu_call interface changes (#243) (@maleadt)
• Simplify map dispatch. (#245) (@maleadt)
• Remove mapreduce implementation, expect users to provide mapreducedim! implementation. (#246) (@maleadt)
• Disallow 0D scalar iteration. (#247) (@maleadt)
• Remove CUDAnative workaround. (#249) (@maleadt)
• Support GPU-like array types with map. (#250) (@maleadt)

v2.0.1

v2.0.1 (2019-12-05)

Diff since v2.0.0

Closed issues:

• Incompatibility with FillArrays 0.7 (#216)

Merged pull requests:

• Use Adapt to support setindex with CPU slices. (#223) (maleadt)
• CompatHelper: bump compat for "AbstractFFTs" to "0.5" (#222) (github-actions[bot])