Tullio.jl Versions Save

Tullio v0.2.12

Diff since v0.2.11

Closed issues:

• Spurious printing of "DiffRules._abs_deriv" when LoopVectorization.jl loaded (#26)
• Migrate to travis-ci.com (#39)
• Migrate GPU CI to Buildkite? (#56)
• CI is failing (timing out) on Julia nightly (#61)
• TODO: Install the Buildkite GitHub App on this repo (#64)
• Idea: on Buildkite, only run the GPU-related tests? (#67)

Merged pull requests:

• Create CompatHelper.yml (#52) (@CarloLucibello)
• Several tweaks to the CI and testing configurations (#55) (@DilumAluthge)
• More LoopVectorization tests & checks (#57) (@mcabbott)
• Split up the test suite into three test groups (#62) (@DilumAluthge)
• GPU CI: Add the config file for the Buildkite pipeline (#65) (@DilumAluthge)
• use autotuning for CUDA kernels by default (#82) (@simeonschaub)

Tullio v0.2.11

Diff since v0.2.10

Closed issues:

• error on julia 1.6 with LoopVectorization on MacOS (#38)
• Default Thread settings (threads=true) cause bad performance on CPU (#45)
• Macro hygiene issue ndims (#46)
• Allow div in range inference? (#48)
• Massive performance degredation between v0.2.8 and v0.2.9 with LoopVectorization.jl (#50)
• Error requiring LoopVectorization from Tullio (#51)

Merged pull requests:

• Add tests for LoopVectorization (#53) (@mcabbott)

Tullio v0.2.10

Diff since v0.2.9

Tullio v0.2.9

Diff since v0.2.8

Closed issues:

• Unexpected behaviour when summing with different sign (#44)

Tullio v0.2.8

Diff since v0.2.7

Closed issues:

• Slowdown when multiple variables looped (#40)
• Warning when loaded with Reversediff (#41)
• Not sure what to do with x + a + c (#43)

Tullio v0.2.7

Diff since v0.2.6

Closed issues:

• Scalar reductions (#29)

Merged pull requests:

• Better scalar reductions (#31) (@mcabbott)
• Add mod, clamp, and a magic underscore (#34) (@mcabbott)

Tullio v0.2.6

Diff since v0.2.5

Merged pull requests:

• Improvements to KernelAbstractions code? (#32) (@mcabbott)

Tullio v0.2.5

Diff since v0.2.4

Tullio v0.2.4

Diff since v0.2.3

Tullio v0.2.3

Diff since v0.2.2

Closed issues:

• @tullio seems slower than equivalent for loop? (#22)
• The initialization for user defined reduction function is always zero(T) (#24)
• Force avx to "avx = false" when use LHS scattering, unless the index is checked to be unique. (#27)

Merged pull requests:

• Better choice of tile sizes? (#23) (@mcabbott)
• Allow init=0.0, and updating with any reduction (#25) (@mcabbott)
• Simplify the generated code in ex_nored, and use first() to collect scalar output instead of sum() (#28) (@N3N5)