When materializing an i32, SExt vs ZExt doesn't matter when we're trying to fit

in a 16-bit immediate.  However, for the shorter non-legal types (i.e., i1, i8,
i16) we should not sign-extend.  This prevents us from materializing things
such as 'true' (i.e., i1 1).


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@143743 91177308-0d34-0410-b5e6-96231b3b80d8
This commit is contained in:
Chad Rosier 2011-11-04 23:09:49 +00:00
parent a54c8ea8cf
commit a4e07270bc

View File

@ -551,7 +551,7 @@ unsigned ARMFastISel::ARMMaterializeInt(const Constant *C, EVT VT) {
// If we can do this in a single instruction without a constant pool entry
// do so now.
const ConstantInt *CI = cast<ConstantInt>(C);
if (Subtarget->hasV6T2Ops() && isUInt<16>(CI->getSExtValue())) {
if (Subtarget->hasV6T2Ops() && isUInt<16>(CI->getZExtValue())) {
unsigned Opc = isThumb ? ARM::t2MOVi16 : ARM::MOVi16;
unsigned ImmReg = createResultReg(TLI.getRegClassFor(VT));
AddOptionalDefs(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL,