[PATCH] D58419: [GISel]: Allow G_EXTRACT_VEC_ELT's result to be larger the source element type

[Aditya Nandakumar via Phabricator via llvm-commits]

aditya_nandakumar created this revision.
aditya_nandakumar added reviewers: arsenm, aemerson, dsanders, volkan, bogner, paquette, qcolombet.
Herald added a subscriber: wdng.
Herald added a project: LLVM.

It's quite inconvenient when dealing with G_EXTRACT_VEC_ELT of <4xs8>,0  where s8 is not legal in the target - typically one ends up with G_UNMERGE_VALUES + G_LSHR by 8 which expands to three instructions.
Instead allow the destination type to be larger than source scalar type and it's higher bits are any extended. This is similar to DAG's opcode.
Note - I couldn't really find a use in tree - so I added a unit test.


Repository:
  rL LLVM

https://reviews.llvm.org/D58419

Files:
  include/llvm/Target/GenericOpcodes.td
  lib/CodeGen/GlobalISel/MachineIRBuilder.cpp
  lib/CodeGen/MachineVerifier.cpp
  test/MachineVerifier/test_g_extract_vec_elt.mir
  unittests/CodeGen/GlobalISel/MachineIRBuilderTest.cpp

-------------- next part --------------
A non-text attachment was scrubbed...
Name: D58419.187475.patch
Type: text/x-patch
Size: 4504 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20190220/0a4644ca/attachment.bin>