Mark tADDrSPi as having side effects again.
It really doesn't, but when r141929 removed the hasSideEffects flag from
this instruction, it caused miscompilations. I am guessing that it got
moved across a stack pointer update.
Also clear isRematerializable after checking that this instruction is
in fact never rematerialized in the nightly test suite.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@142030 91177308-0d34-0410-b5e6-96231b3b80d8
diff --git a/lib/Target/ARM/ARMInstrThumb.td b/lib/Target/ARM/ARMInstrThumb.td
index 3e81516..cedb547 100644
--- a/lib/Target/ARM/ARMInstrThumb.td
+++ b/lib/Target/ARM/ARMInstrThumb.td
@@ -310,9 +310,9 @@
}
// ADD <Rd>, sp, #<imm8>
-// This is rematerializable, which is particularly useful for taking the
-// address of locals.
-let isReMaterializable = 1, neverHasSideEffects = 1 in
+// FIXME: This should not be marked as having side effects, and it should be
+// rematerializable. Clearing the side effect bit causes miscompilations,
+// probably because the instruction can be moved around.
def tADDrSPi : T1pI<(outs tGPR:$dst), (ins GPRsp:$sp, t_imm0_1020s4:$imm),
IIC_iALUi, "add", "\t$dst, $sp, $imm", []>,
T1Encoding<{1,0,1,0,1,?}> {
