akirilov-arm commented on Issue #2310:
@cfallin A side question - I noticed that
Bconst
generates 1 for the true value instead of -1 (e.g. 0xFFFFFFFF forbconst.b32 true
; same on x86), but then I remembered your changes in #2058. Is this the expected behaviour? And then the other question - what would be the correct value for the vector counterpart? Currently we generate a vector with 1s in every lane.
akirilov-arm commented on Issue #2310:
@cfallin A side note about possible future work - the scalar boolean handling could be improved by using the
CSETM
(instead ofCSET
) andSBFX
/SBFM
instructions.
cfallin commented on Issue #2310:
Perfect, thanks!
Last updated: Dec 23 2024 at 12:05 UTC