Codename: Perdita Quiescent
What's Changed
A lot of changes this release with The One Big News: SVE SUPPORT
🎉 🎉 🎉 🎉 🎉 🎉 🎉 🎉 🎉 🎉 🎉 🎉
Second huge change : We now take Frequency Scaling on AVX512 machines in algorithm implementations.
THis means that algorithms will by default avoid big frequency scaling on AVX512 by using 32 byte registers unless specified otherwise.
Removal and Depreciation
- The proba module as been removed. (See #1490) It will be reworked as a separate project later on with a proper API.
Architectures/Compilers Support & Fixes
- SVE with fixed size, power of 2 cardinal is now supported for most of our API. Some more optimizations are on the way but the support is functional. (See #1432, #1435, #1436, #1437, #1438, #1448, #1453, #1455, #1456, #1463, #1464, #1472, #1473, #1474,#1487, #1489, #1491, #1494, #1533, #1556)
Other Fixes
- EVE now compiles on ARM M1 with Homebrew g++. (See #1471)
- EVE now compiles with Apple Clang (See #1479, #1485, #1502, #1530, #1531)
- Various efforts have been made to advance the MSVC situation. Help still welcomed.
- ARM all and any are now more efficient (See #1500)
- Integral sign/signnz functions are now more efficient on x86 (See #1499)
- Implementation for X86 AVX2/AVX512 gather and masked gather are now optimized. (See #1526)
Features
- @jtlap implemented a large amount of new functions for
eve::complex. - Functions like exp, log or sqrt can now be called with a real entry and a complex output. (See #1528)
- Binary functions for which a n-ary extensions is available now support being called with a tuple-like parameter instead of a dynamic range (See #1422, #1509)
- The minmax function is now available. (See #1507)
- @DenisYaroshevskiy implemented new traits for algorithms to take care of costly kernels, no alignment and to support fused operations. (See #1535, #1543)
Bug Fixes
- Convert is now more efficient and don't generate piecewise evaluation in some scenario involving logicals. (See #1447, #1428)
if_elsenow uses the proper constant generator in optimized cases. (See #1529)- Prevent constant to erroneously be callable with non-specific product types. (See #1540)
- A large cleanup of old traits and concepts has been done. The basics concepts around vectorizable and vectorized types has been therefore simplified and streamlined. (See #1468, #1477, #1527, #1488, #1545)
- Fix issue with dynamic SIMD extension detection that were broken by accident. (See #1504)