Skip to content

neon: Enable Flush-to-zero mode for float programs

Doug Nazar requested to merge dougnazar/orc:neon into master

Some of the instructions are not NEON (which always uses FTZ), but are actually VFP, which requires enabling FTZ mode.

Only 2 opcodes now fail, divf & sqrtf but they're both multi-step (Newton-Raphson) so don't handle denormals & 0 very well.

Merge request reports