[Libre-soc-dev] pysvp64asm: opcode setvli not supported
Luke Kenneth Casson Leighton
lkcl at lkcl.net
Tue Jun 15 19:19:56 BST 2021
i replaced the fmuls and fadds with fmadds, and also took the
liberty of using fmsubs in appropriate locations, which also gave
the opportunity to remove tmpsum and tmpsum2 because with
fmsubs / fmadds both sum and sum2 can be directly used as
accumulators, carrying on the subtract (or add).
now, at the end (outside the loop) i used fmsubs and it produces
+ve 0 so i think we're ok with that, no need for fsel or branch
etc. or fnegs sum sum
i then moved on to buf1000 - interestingly it fails. and buf2000.
so i ran the *scalar* version, and that *also* fails.
i therefore suspect some sort of overflow / underflow in the
underlying FPMUL32 routine which has gone undetected
by the buf0 test.
luckily, joy-of-joys, i can run qemu on the scalar buf1000 test.
More information about the Libre-soc-dev