projects / platform / upstream / llvm.git / commit
? search:
summary | shortlog | log | commit | commitdiff | tree
(parent: 6ca7eb2) | patch
[SystemZ] Bugfix in SystemZTargetLowering::combineINT_TO_FP()
authorJonas Paulsson <paulsson@linux.vnet.ibm.com>
Wed, 18 May 2022 12:20:30 +0000 (14:20 +0200)
committerJonas Paulsson <paulsson@linux.vnet.ibm.com>
Wed, 18 May 2022 14:32:37 +0000 (16:32 +0200)
commit4273e616e573f4854e9c659c2e6d8a31f3e3c595
tree590453b15676b1743daa319b960ebce65f57f97etree | snapshot
parent6ca7eb2c6d7da2264f410f270633bf698ab5d87dcommit | diff
[SystemZ] Bugfix in SystemZTargetLowering::combineINT_TO_FP()

Make sure to also handle extended value types to avoid crashing.

Resulting integers greater than 64 bits are not optimized (i128 is not a
legal type), and vectorizing seems to result in libcalls instead of just
scalarization.

Other extended vector types like <10 x float> are however now handled and
should result in vectorized conversions.

Reviewed By: Ulrich Weigand

Differential Revision: https://reviews.llvm.org/D125881
llvm/lib/Target/SystemZ/SystemZISelLowering.cpp diff | blob | history
llvm/test/CodeGen/SystemZ/vec-move-23.ll diff | blob | history
Domain: System / Toolchain;