V0.2.1

Changed

• Let CMake install libraries and header files (#57)
• Update CI tune_cuda workflow for latest Kernel Tuner (#57)

V0.2.0

Added

• Add support for FP8 e4m3 data type (#24)
• Add hardware support checks for FP8 (#35)
• Implement scaling with alpha/beta (#31)
• Add support for bfloat16 (#20)
• Add start-to-end pipeline class (#17)
• Add compatibility with HIP 7 (#26)
• Set PATH and CXX for HIP test workflow (#28)
• Guard tf32 for architectures that don't support it (#23)
• Add CI for tests on GPU hardware (#1)

Changed

• Move direction and complex axis location arguments from packing.Run to constructor (#25)
• Release preparation (#34)
• Implement syncwarp for HIP<7 (#29)
• Make type_selector compatible with sm70 (Tesla V100) (#21)
• Cleanup type selector (#11)
• Fix compilation of benchmark in HIP mode when host compiler != hipcc (#18)
• Fix benchmark (#12)

V0.1.0

First release, matching code base at the time of paper submission.