8 bit floating point#1910

Open

rbuchner-aril wants to merge 19 commits into

riscv-software-src:masterfrom

rbuchner-aril:rbuchner/FP8

Commits on Feb 7, 2025

Implement f8_to_f16 and f16_to_f8 conversion functions
rbuchner-aril
committed
Implement F8 multiply
rbuchner-aril
committed
Extract logic for emulating f8 function using f16 from f8_mul into f8_emulation module
rbuchner-aril
committed
Implement f8_sqrt, f8_rsqrte7, f8_recip7 in softfloat
rbuchner-aril
committed
Implement f8_div in softfloat
rbuchner-aril
committed
Implement f8_add/f8_sub in softfloat
rbuchner-aril
committed
Implement f8_madd instruction in softfloat
rbuchner-aril
committed
Implement f8_to_(u)i16 in softfloat
rbuchner-aril
committed
Implement f8_to_(u)i8 in softfloat
rbuchner-aril
committed
Implement (u)i32_to_f8 in softfloat
rbuchner-aril
committed
Update vector FP classify to support FP8
rbuchner-aril
committed
Implement F8 min/max
rbuchner-aril
committed
Implement f8 comparsion functions
rbuchner-aril
committed
Add thread local softfloat_fp8Mode variable to control the 8 bit floating point format
rbuchner-aril
committed
Extract expF8UI and fracF8UI based on fp8Mode

Omar Almatov
authored and
rbuchner-aril
committed
(1/4) Add OCP e4m3 and e5m2 to list of f8 modes

Omar Almatov
authored and
rbuchner-aril
committed
(2/4) Add macros to support new OCP f8 modes

Omar Almatov
authored and
rbuchner-aril
committed
(3/4) Support e4m3/e5m2 modes in f8 to f16 conversion

Omar Almatov
authored and
rbuchner-aril
committed
(4/4) Support e4m3/e5m2 modes in f16 to f8 conversion

Omar Almatov
authored and
rbuchner-aril
committed