On 10/28/26 19:36, Gareth Lloyd wrote:
These helpers move bytes rather than interpret them, so a host SSE2 instruction gives bit-exact results and needs no gate on the float environment. A host without the routines keeps the portable loops.tests/tcg/i386/test-avx.c and test-mmx.c already cover these instructions.
It would be useful to see it also for another host.Also, have you checked if compiler vectorization can produce PACK/UNPACK instructions?
Thanks, Paolo
