v0.6.1

BFloat16s v0.6.1

Diff since v0.6.0

Merged pull requests:

• Bump actions/checkout from 5 to 6 (#115) (@dependabot[bot])
• Bump actions/cache from 4 to 5 (#116) (@dependabot[bot])
• Require LLVM 19 for llvm_storage on aarch64 (#117) (@AntonOresten)

Closed issues:

• BFloat16 arithmetic fails when used with CUDA on Julia v1.11 (#110)
• -(::BFloat16) causes LLVM error on 1.12 with aarch64 (#113)

v0.6.0

BFloat16s v0.6.0

Diff since v0.5.1

Breaking changes

No breaking changes.

New features

• parse/tryparse
• nightly support
• read/write

Merged pull requests:

• Add read/write support for BFloat16 values (#98) (@kengruven)
• Implement parse and tryparse (#104) (@mkitti)
• remove outdated performance warning from readme (#105) (@oscardssmith)
• Bump actions/checkout from 4 to 5 (#109) (@dependabot[bot])
• Fixes for nightly (#111) (@christiangnrd)

Closed issues:

• Segfault using Julia 1.11-alpha2 on AMD EPYC 9554 (#68)
• tag new version? (#90)
• IO (#96)
• Is there something like ComplexBF16? (#101)
• Fails to precompile on Julia nightly (#112)

v0.5.1

BFloat16s v0.5.1

Diff since v0.5.0

Merged pull requests:

• BFloat16(::BigFloat) and BigFloat(::BFloat16) (#63) (@milankl)
• Bump actions/cache from 3 to 4 (#65) (@dependabot[bot])
• Update CI.yml (#66) (@ViralBShah)
• Bump codecov/codecov-action from 3 to 4 (#67) (@dependabot[bot])
• Remove Test dependency (#69) (@christiangnrd)
• Bump julia-actions/setup-julia from 1 to 2 (#70) (@dependabot[bot])
• rand BFloat16 sampling without conversion (#73) (@milankl)
• Update CI.yml (#74) (@ViralBShah)
• Update README.md (#75) (@ViralBShah)
• Update README.md (#76) (@ViralBShah)
• Enable AArch64 arithmetic on LLVM 19. (#77) (@maleadt)
• Update CI.yml to test on pre-releases (#80) (@ViralBShah)
• rand via scaled random integers (#82) (@milankl)
• CI: Test on more systems (#83) (@eschnett)
• Bump codecov/codecov-action from 4 to 5 (#87) (@dependabot[bot])
• Silence warning in 1.12+ (#91) (@christiangnrd)
• Improve test coverage (#92) (@christiangnrd)
• Remove redundant operator definitions (#93) (@christiangnrd)
• Add definitions for 128-bit unsafe_trunc when llvm_arithmetic == true (#94) (@christiangnrd)
• Add BFloat(Irrational) constructor (#99) (@kengruven)
• Define Base.issubnormal(x::BFloat16) (take 2) (#100) (@christiangnrd)

Closed issues:

• Support new Float8 and NormedFloat4 formats? (#47)
• Segfault using Julia 1.11-alpha2 on AMD EPYC 9554 (#68)
• rand(BFloat16, n) appears shifted from [0, 1) into (0, 1] (#71)
• Improve precision of rand sampling without conversion (#79)
• Base.issubnormal(::BFloat16) is missing (#86)
• Creating sysimage fails on Windows (#88)
• Segmentation fault when converting BFloat16 to Float32 (#89)
• Irrationals (#97)

v0.5.0

BFloat16s v0.5.0

Diff since v0.4.2

This is a breaking release, changing the way code is generated for operations with BFloat16 numbers. On Julia 1.10, this will now rely on LLVM, potentially generating code that performs BFloat16 operations natively. On older versions of Julia, the existing software implementation is still used.

Although care has been taken to make sure both implementations behave identically, it is possible that this change has introduced differences, which is why this release has been tagged as a breaking release. If you rely on this package, it is recommended to test your code using both Julia 1.10 on a platform where LLVM supports bfloat (i.e., x86 or x86_64), and using any older version of Julia. Differences in behavior are likely bugs.

Merged pull requests:

• enable dependabot for GitHub actions (#46) (@ranocha)
• next/prevfloat(::BFloat16,::Integer) with tests (#49) (@milankl)
• Base.decompose(::BFloat16) (#50) (@milankl)
• Adapt to upstream changes wrt. native support for BFloat16 (#51) (@maleadt)
• Bump actions/checkout from 2 to 4 (#54) (@dependabot[bot])
• Bump codecov/codecov-action from 1 to 3 (#55) (@dependabot[bot])
• Bump actions/cache from 1 to 3 (#56) (@dependabot[bot])
• Test more Julia versions. (#57) (@maleadt)
• Remove unused printf functionality. (#58) (@maleadt)
• extending the readme a bit (#59) (@milankl)
• add maxintfloat(::Type{BFloat16}) (#60) (@milankl)
• MassInstallAction: Install the CompatHelper workflow on this repository (#61) (@DilumAluthge)
• Bump to v0.5.0 (#62) (@milankl)

Closed issues:

• Fix nonexistent overload of Printf functionality (#40)
• nextfloat/prevfloat(::BFloat16,::Integer) missing (#48)
• please give Tim Besard write permission (#52)
• is BFloat16s.jl up to date with JuliaLang/julia#51470 (#53)

updated info on BFloat16 hardware

from GunnarFarneback

v0.4.0

Update mathfuncs.jl

v0.3.0

4 PRs merged

v0.1.0

v0.1.0 (2019-10-21)

Diff since 5f5260caf88d8e071859a8dba84f7f8766decff6

Closed issues:

• Conversion to and from Float64 (#9)
• Constructors (#7)
• Missing promotion for comparison operators (#6)
• Missing sqrt (#5)

Merged pull requests:

• Functions for BFloat16 (#11) (amontoison)
• using linalg (#1) (ChrisRackauckas)