amd64 back end: generate better code for 2x64<-->V128 and 4x64<-->V256 transfers ..
commitb17d5ffdb844cf081c86d7df9489f61b4392ca47
authorJulian Seward <jseward@acm.org>
Sat, 22 Dec 2018 17:04:42 +0000 (22 18:04 +0100)
committerJulian Seward <jseward@acm.org>
Sat, 22 Dec 2018 17:04:42 +0000 (22 18:04 +0100)
treed0ab4b604959d712d63af58c806683e2e6499e2e
parentdda0d80f3db1632b204b522a1dbb009490265b0e
amd64 back end: generate better code for 2x64<-->V128 and 4x64<-->V256 transfers ..

.. by adding support for MOVQ xmm/ireg and using that to implement 64HLtoV128,
4x64toV256 and their inverses.  This reduces the number of instructions,
removes the use of memory as an intermediary, and avoids store-forwarding
stalls.
VEX/priv/host_amd64_defs.c
VEX/priv/host_amd64_defs.h
VEX/priv/host_amd64_isel.c