[PATCH] D44365: [X86] Improve codegen of bitcasts of BUILD_VECTOR to MMX register.

[Craig Topper via Phabricator via llvm-commits]

craig.topper created this revision.
craig.topper added reviewers: spatel, RKSimon.

The 64 bit BUILD_VECTOR is normally going to be promoted to 128 bits by widening each element. This changes the size of each element and requires a shuffle to compress it back down to 64-bits. We also do the copy from XMM to MMX through memory.

This patch pads the vector with undef elements and emits and explicit XMM->MMX copy operation.


Repository:
  rL LLVM

https://reviews.llvm.org/D44365

Files:
  lib/Target/X86/X86ISelLowering.cpp
  test/CodeGen/X86/mmx-build-vector.ll
  test/CodeGen/X86/pr29222.ll
  test/CodeGen/X86/vec_insert-mmx.ll

-------------- next part --------------
A non-text attachment was scrubbed...
Name: D44365.137944.patch
Type: text/x-patch
Size: 49635 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20180311/f5ed398c/attachment.bin>