Explicitly zero out parts of a vector which are required to be zero by the algorithm...
authorEli Friedman <eli.friedman@gmail.com>
Mon, 29 Aug 2011 21:15:46 +0000 (21:15 +0000)
committerEli Friedman <eli.friedman@gmail.com>
Mon, 29 Aug 2011 21:15:46 +0000 (21:15 +0000)
commitf3704769bb2e8170769840528869ae5696545481
treef943b23cbca8d59227bb97dd924b42e32d25f587
parentd33b276bcb1efd4fa14c32fec68ae24baf30eb28
Explicitly zero out parts of a vector which are required to be zero by the algorithm in LowerUINT_TO_FP_i32.  This only has a substantial effect on the generated code when the input is extracted from a vector register; other ways of loading an i32 do the appropriate zeroing implicitly.  Fixes PR10802.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@138768 91177308-0d34-0410-b5e6-96231b3b80d8
lib/Target/X86/X86ISelLowering.cpp
test/CodeGen/X86/uint_to_fp-2.ll