projects / platform / upstream / llvm.git / commit
? search:
summary | shortlog | log | commit | commitdiff | tree
(parent: 790d0fa) | patch
[AVX-512][InstCombine] Teach InstCombineCalls how to simplify demanded for scalar...
authorCraig Topper <craig.topper@gmail.com>
Sun, 11 Dec 2016 07:42:04 +0000 (07:42 +0000)
committerCraig Topper <craig.topper@gmail.com>
Sun, 11 Dec 2016 07:42:04 +0000 (07:42 +0000)
commitd96395365a3cf8383625e54f429ec66d21ac5e31
tree7fe686cb06ff64f4efbdb45d1fc712ac67066ea8tree | snapshot
parent790d0fa569cc43126313d30f4fc1a9b9e8d77977commit | diff
[AVX-512][InstCombine] Teach InstCombineCalls how to simplify demanded for scalar cmp intrinsics with masking and rounding.

These intrinsics don't read the upper elements of their first and second input. These are slightly different the the SSE version which does use the upper bits of its first element as passthru bits since the result goes to an XMM register. For AVX-512 the result goes to a mask register instead.

llvm-svn: 289371
llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp diff | blob | history
Domain: System / Toolchain;