[AVR] Make use of the constant value 0 in R1

The register R1 is defined to have the constant value 0 in the avr-gcc
calling convention (which we follow). Unfortunately, we don't really
make use of it. This patch replaces `LDI 0` instructions with a copy
from R1.

This reduces code size: my AVR build of compiler-rt goes from 50660 to
50240 bytes of code size, which is a 0.8% reduction. Presumably it will
also improve execution speed, although I didn't measure this.

Differential Revision: https://reviews.llvm.org/D117425

llvm/lib/Target/AVR/AVRISelLowering.cpp

@@ -1695,6 +1695,18 @@ MachineBasicBlock *AVRTargetLowering::insertMul(MachineInstr &MI,
   return BB;
 }
 
+// Insert a read from R1, which almost always contains the value 0.
+MachineBasicBlock *
+AVRTargetLowering::insertCopyR1(MachineInstr &MI, MachineBasicBlock *BB) const {
+  const TargetInstrInfo &TII = *Subtarget.getInstrInfo();
+  MachineBasicBlock::iterator I(MI);
+  BuildMI(*BB, I, MI.getDebugLoc(), TII.get(AVR::COPY))
+      .add(MI.getOperand(0))
+      .addReg(AVR::R1);
+  MI.eraseFromParent();
+  return BB;
+}
+
 MachineBasicBlock *
 AVRTargetLowering::EmitInstrWithCustomInserter(MachineInstr &MI,
                                                MachineBasicBlock *MBB) const {
@@ -1717,6 +1729,8 @@ AVRTargetLowering::EmitInstrWithCustomInserter(MachineInstr &MI,
   case AVR::MULRdRr:
   case AVR::MULSRdRr:
     return insertMul(MI, MBB);
+  case AVR::CopyR1:
+    return insertCopyR1(MI, MBB);
   }
 
   assert((Opc == AVR::Select16 || Opc == AVR::Select8) &&

llvm/lib/Target/AVR/AVRISelLowering.h

@@ -187,6 +187,8 @@ protected:
 private:
   MachineBasicBlock *insertShift(MachineInstr &MI, MachineBasicBlock *BB) const;
   MachineBasicBlock *insertMul(MachineInstr &MI, MachineBasicBlock *BB) const;
+  MachineBasicBlock *insertCopyR1(MachineInstr &MI,
+                                  MachineBasicBlock *BB) const;
 };
 
 } // end namespace llvm

llvm/lib/Target/AVR/AVRInstrInfo.td

@@ -2390,6 +2390,10 @@ def Asr16 : ShiftPseudo<(outs DREGS
                                                      : $src, i8
                                                      : $cnt))]>;
 
+// lowered to a copy from R1, which contains the value zero.
+let usesCustomInserter=1 in
+def CopyR1 : Pseudo<(outs GPR8:$rd), (ins), "clrz\t$rd", [(set i8:$rd, 0)]>;
+
 //===----------------------------------------------------------------------===//
 // Non-Instruction Patterns
 //===----------------------------------------------------------------------===//

llvm/test/CodeGen/AVR/smul-with-overflow.ll

@@ -18,7 +18,7 @@ entry-block:
 ; CHECK: ldi    [[RET:r[0-9]+]], 1
 ; CHECK: cp     {{.*}}[[HIGH]], {{.*}}[[LOW]]
 ; CHECK: brne   [[LABEL:.LBB[_0-9]+]]
-; CHECK: ldi    {{.*}}[[RET]], 0
+; CHECK: mov    {{.*}}[[RET]], r1
 ; CHECK: {{.*}}[[LABEL]]
 ; CHECK: ret
 }

llvm/test/CodeGen/AVR/store-undef.ll

@@ -6,8 +6,7 @@
 ; CHECK-LABEL: foo
 define void @foo() {
 
-  ; CHECK:      ldi [[SRC:r[0-9]+]], 0
-  ; CHECK-NEXT: st [[PTRREG:X|Y|Z]], [[SRC]]
+  ; CHECK: st [[PTRREG:X|Y|Z]], r1
   store i8 0, i8* undef, align 4
   ret void
 }

llvm/test/CodeGen/AVR/umul-with-overflow.ll

@@ -14,7 +14,7 @@ entry-block:
 ; CHECK: ldi    [[RET:r[0-9]+]], 1
 ; CHECK: cpi    {{.*}}[[HIGH]], 0
 ; CHECK: brne   [[LABEL:.LBB[_0-9]+]]
-; CHECK: ldi    {{.*}}[[RET]], 0
+; CHECK: mov    {{.*}}[[RET]], r1
 ; CHECK: {{.*}}[[LABEL]]
 ; CHECK: ret
 }