alexcrichton opened PR #6058 from x64-more-avx to main:
This PR refactors some gpr-to-xmm-and-back pieces of the x64 backend along with some new AVX instructions, namely:
- The
movdandmovqinstructions now have dedicated helpers with AVX special cases like all other vector-related instructions.- The
mov_rmi_to_xmmfunction is no longer an auto-conversion betweenRegMemImmandXmmMemImmsince it performs conversions via register moves while all other register-to-register conversions were simply asserting that the register was of the right type.- AVX lowerings were added for
movd,movq,vcvtsi2s{s,d}, andv{,p}movmsk{b,ps,pd}
alexcrichton updated PR #6058 from x64-more-avx to main.
alexcrichton updated PR #6058 from x64-more-avx to main.
alexcrichton updated PR #6058 from x64-more-avx to main.
cfallin assigned PR #6058 to cfallin.
cfallin submitted PR review.
cfallin submitted PR review.
cfallin created PR review comment:
s/ot/to/
(also, wtf, this is the weirdest encoding quirk ever)
alexcrichton updated PR #6058 (assigned to cfallin) from x64-more-avx to main.
alexcrichton has enabled auto merge for PR #6058.
alexcrichton updated PR #6058 (assigned to cfallin) from x64-more-avx to main.
alexcrichton updated PR #6058 (assigned to cfallin) from x64-more-avx to main.
alexcrichton merged PR #6058.
Last updated: Dec 13 2025 at 21:03 UTC