R600/SI: Don't promote f32 select to i32
authorMatt Arsenault <Matthew.Arsenault@amd.com>
Fri, 12 Dec 2014 02:30:29 +0000 (02:30 +0000)
committerMatt Arsenault <Matthew.Arsenault@amd.com>
Fri, 12 Dec 2014 02:30:29 +0000 (02:30 +0000)
commit477b178276921913bc282195723b684c225692e5
treec8fe46825c20314d96b88b4061e880894597cef9
parent8970390f81c67fb444bb850a8c8eea0b7364fc2a
R600/SI: Don't promote f32 select to i32

This is nice for the instruction patterns, but it complicates
min / max matching. The select doesn't have the correct type and would
require looking through the bitcasts for the real float operands.

llvm-svn: 224092
llvm/lib/Target/R600/SIISelLowering.cpp
llvm/lib/Target/R600/SIInstructions.td
llvm/test/CodeGen/R600/seto.ll
llvm/test/CodeGen/R600/setuo.ll
llvm/test/CodeGen/R600/sint_to_fp.ll
llvm/test/CodeGen/R600/uint_to_fp.ll