[M68k] Introduce more MOVI cases (#98377)

Add three more special cases for loading registers with immediates.

The first allows values in the range of [-255, 255] to be loaded with
MOVEQ, even if the register is more than 8 bits and the sign extention
is unwanted. This is done by loading the bitwise complement of the
desired value, then performing a NOT instruction on the loaded register.

This special case is only used when a simple MOVEQ cannot be used, and
is only used for 32 bit data registers. Address registers cannot support
MOVEQ, and the two-instruction sequence is no faster or smaller than a
plain MOVE instruction when loading 16 bit immediates on the 68000, and
likely slower for more sophisticated microarchitectures. However, the
instruction sequence is both smaller and faster than the corresponding
MOVE instruction for 32 bit register widths.

The second special case is for zeroing address registers. This simply
expands to subtracting a register with itself, consuming one instruction
word rather than 2-3, with a small improvement in speed as well.

The last special case is for assigning sign-extended 16-bit values to a
full address register. This takes advantage of the fact that the movea.w
instruction sign extends the output, permitting the immediate to be
smaller. This is similar to using lea with a 16-bit address, which is
not added in this patch as 16-bit absolute addressing is not yet
implemented.

This is a v2 submission of #90817. It also creates a 'Data' test
directory to better align with the backend's tablegen layout.

Author: n8pjl

llvm/lib/Target/M68k/M68kInstrInfo.cpp

@@ -346,8 +346,8 @@ void M68kInstrInfo::AddZExt(MachineBasicBlock &MBB,
   BuildMI(MBB, I, DL, get(And), Reg).addReg(Reg).addImm(Mask);
 }
 
-// Convert MOVI to MOVQ if the target is a data register and the immediate
-// fits in a sign-extended i8, otherwise emit a plain MOV.
+// Convert MOVI to the appropriate instruction (sequence) for setting
+// the register to an immediate value.
 bool M68kInstrInfo::ExpandMOVI(MachineInstrBuilder &MIB, MVT MVTSize) const {
   Register Reg = MIB->getOperand(0).getReg();
   int64_t Imm = MIB->getOperand(1).getImm();
@@ -360,18 +360,66 @@ bool M68kInstrInfo::ExpandMOVI(MachineInstrBuilder &MIB, MVT MVTSize) const {
   if (AR16->contains(Reg) || AR32->contains(Reg))
     IsAddressReg = true;
 
+  // We need to assign to the full register to make IV happy
+  Register SReg =
+      MVTSize == MVT::i32
+          ? Reg
+          : Register(RI.getMatchingMegaReg(Reg, IsAddressReg ? AR32 : DR32));
+  assert(SReg && "No viable MEGA register available");
+
   LLVM_DEBUG(dbgs() << "Expand " << *MIB.getInstr() << " to ");
 
+  // Sign extention doesn't matter if we only use the bottom 8 bits
   if (MVTSize == MVT::i8 || (!IsAddressReg && Imm >= -128 && Imm <= 127)) {
     LLVM_DEBUG(dbgs() << "MOVEQ\n");
 
-    // We need to assign to the full register to make IV happy
-    Register SReg =
-        MVTSize == MVT::i32 ? Reg : Register(RI.getMatchingMegaReg(Reg, DR32));
-    assert(SReg && "No viable MEGA register available");
-
     MIB->setDesc(get(M68k::MOVQ));
     MIB->getOperand(0).setReg(SReg);
+
+    // Counter the effects of sign-extension with a bitwise not.
+    // This is only faster and smaller for 32 bit values.
+  } else if (DR32->contains(Reg) && isUInt<8>(Imm)) {
+    LLVM_DEBUG(dbgs() << "MOVEQ and NOT\n");
+
+    MachineBasicBlock &MBB = *MIB->getParent();
+    DebugLoc DL = MIB->getDebugLoc();
+
+    unsigned SubReg = RI.getSubReg(Reg, M68k::MxSubRegIndex8Lo);
+    assert(SubReg && "No viable SUB register available");
+
+    BuildMI(MBB, MIB.getInstr(), DL, get(M68k::MOVQ), SReg).addImm(~Imm & 0xFF);
+    BuildMI(MBB, MIB.getInstr(), DL, get(M68k::NOT8d), SubReg).addReg(SubReg);
+
+    MIB->removeFromParent();
+
+    // Special case for setting address register to NULL (0)
+  } else if (IsAddressReg && Imm == 0) {
+    LLVM_DEBUG(dbgs() << "SUBA\n");
+
+    MachineBasicBlock &MBB = *MIB->getParent();
+    DebugLoc DL = MIB->getDebugLoc();
+
+    BuildMI(MBB, MIB.getInstr(), DL, get(M68k::SUB32ar), SReg)
+        .addReg(SReg, RegState::Undef)
+        .addReg(SReg, RegState::Undef);
+
+    MIB->removeFromParent();
+
+    // movea.w implicitly sign extends to the full register width,
+    // so exploit that if the immediate fits in the correct range.
+    //
+    // TODO: use lea imm.w, %an for further constants when 16-bit
+    // absolute addressing is implemented.
+  } else if (AR32->contains(Reg) && isUInt<16>(Imm)) {
+    LLVM_DEBUG(dbgs() << "MOVEA w/ implicit extend\n");
+
+    unsigned SubReg = RI.getSubReg(Reg, M68k::MxSubRegIndex16Lo);
+    assert(SubReg && "No viable SUB register available");
+
+    MIB->setDesc(get(M68k::MOV16ai));
+    MIB->getOperand(0).setReg(SubReg);
+
+    // Fall back to a move with immediate
   } else {
     LLVM_DEBUG(dbgs() << "MOVE\n");
     MIB->setDesc(get(MVTSize == MVT::i16 ? M68k::MOV16ri : M68k::MOV32ri));

llvm/test/CodeGen/M68k/Arith/add.ll

@@ -65,7 +65,8 @@ define fastcc void @test4(ptr inreg %a) nounwind {
 ; CHECK-NEXT:    movem.l %d2, (0,%sp) ; 8-byte Folded Spill
 ; CHECK-NEXT:    move.l (%a0), %d0
 ; CHECK-NEXT:    moveq #0, %d1
-; CHECK-NEXT:    move.l #128, %d2
+; CHECK-NEXT:    moveq #127, %d2
+; CHECK-NEXT:    not.b %d2
 ; CHECK-NEXT:    add.l (4,%a0), %d2
 ; CHECK-NEXT:    addx.l %d0, %d1
 ; CHECK-NEXT:    move.l %d2, (4,%a0)

llvm/test/CodeGen/M68k/CConv/fastcc-call.ll

@@ -13,8 +13,8 @@ define i32 @foo1() nounwind uwtable {
 ; CHECK-NEXT:    move.l #5, (%sp)
 ; CHECK-NEXT:    moveq #1, %d0
 ; CHECK-NEXT:    moveq #2, %d1
-; CHECK-NEXT:    move.l #3, %a0
-; CHECK-NEXT:    move.l #4, %a1
+; CHECK-NEXT:    move.w #3, %a0
+; CHECK-NEXT:    move.w #4, %a1
 ; CHECK-NEXT:    jsr (bar1@PLT,%pc)
 ; CHECK-NEXT:    moveq #0, %d0
 ; CHECK-NEXT:    adda.l #4, %sp

llvm/test/CodeGen/M68k/link-unlnk.ll renamed to llvm/test/CodeGen/M68k/Data/link-unlnk.ll

@@ -0,0 +1,149 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 2
+; RUN: llc < %s -mtriple=m68k-linux -verify-machineinstrs | FileCheck %s
+
+define i1 @return_true() {
+; CHECK-LABEL: return_true:
+; CHECK:         .cfi_startproc
+; CHECK-NEXT:  ; %bb.0:
+; CHECK-NEXT:    moveq #1, %d0
+; CHECK-NEXT:    rts
+  ret i1 true
+}
+
+define i8 @return_0_i8() {
+; CHECK-LABEL: return_0_i8:
+; CHECK:         .cfi_startproc
+; CHECK-NEXT:  ; %bb.0:
+; CHECK-NEXT:    moveq #0, %d0
+; CHECK-NEXT:    rts
+  ret i8 0
+}
+
+define i16 @return_0_i16() {
+; CHECK-LABEL: return_0_i16:
+; CHECK:         .cfi_startproc
+; CHECK-NEXT:  ; %bb.0:
+; CHECK-NEXT:    moveq #0, %d0
+; CHECK-NEXT:    rts
+  ret i16 0
+}
+
+define i32 @return_0_i32() {
+; CHECK-LABEL: return_0_i32:
+; CHECK:         .cfi_startproc
+; CHECK-NEXT:  ; %bb.0:
+; CHECK-NEXT:    moveq #0, %d0
+; CHECK-NEXT:    rts
+  ret i32 0
+}
+
+define i64 @return_0_i64() {
+; CHECK-LABEL: return_0_i64:
+; CHECK:         .cfi_startproc
+; CHECK-NEXT:  ; %bb.0:
+; CHECK-NEXT:    moveq #0, %d0
+; CHECK-NEXT:    move.l %d0, %d1
+; CHECK-NEXT:    rts
+  ret i64 0
+}
+
+define i16 @return_neg1_i16() {
+; CHECK-LABEL: return_neg1_i16:
+; CHECK:         .cfi_startproc
+; CHECK-NEXT:  ; %bb.0:
+; CHECK-NEXT:    moveq #-1, %d0
+; CHECK-NEXT:    rts
+  ret i16 -1
+}
+
+define i32 @return_neg1_i32() {
+; CHECK-LABEL: return_neg1_i32:
+; CHECK:         .cfi_startproc
+; CHECK-NEXT:  ; %bb.0:
+; CHECK-NEXT:    moveq #-1, %d0
+; CHECK-NEXT:    rts
+  ret i32 -1
+}
+
+define i8 @return_160_i8() {
+; CHECK-LABEL: return_160_i8:
+; CHECK:         .cfi_startproc
+; CHECK-NEXT:  ; %bb.0:
+; CHECK-NEXT:    moveq #-96, %d0
+; CHECK-NEXT:    rts
+  ret i8 160
+}
+
+define i16 @return_160_i16() {
+; CHECK-LABEL: return_160_i16:
+; CHECK:         .cfi_startproc
+; CHECK-NEXT:  ; %bb.0:
+; CHECK-NEXT:    move.w #160, %d0
+; CHECK-NEXT:    rts
+  ret i16 160
+}
+
+define i32 @return_160_i32() {
+; CHECK-LABEL: return_160_i32:
+; CHECK:         .cfi_startproc
+; CHECK-NEXT:  ; %bb.0:
+; CHECK-NEXT:    moveq #95, %d0
+; CHECK-NEXT:    not.b %d0
+; CHECK-NEXT:    rts
+  ret i32 160
+}
+
+define i16 @return_14281_i16() {
+; CHECK-LABEL: return_14281_i16:
+; CHECK:         .cfi_startproc
+; CHECK-NEXT:  ; %bb.0:
+; CHECK-NEXT:    move.w #14281, %d0
+; CHECK-NEXT:    rts
+  ret i16 14281
+}
+
+define i32 @return_14281_i32() {
+; CHECK-LABEL: return_14281_i32:
+; CHECK:         .cfi_startproc
+; CHECK-NEXT:  ; %bb.0:
+; CHECK-NEXT:    move.l #14281, %d0
+; CHECK-NEXT:    rts
+  ret i32 14281
+}
+
+define i64 @return_14281_i64() {
+; CHECK-LABEL: return_14281_i64:
+; CHECK:         .cfi_startproc
+; CHECK-NEXT:  ; %bb.0:
+; CHECK-NEXT:    moveq #0, %d0
+; CHECK-NEXT:    move.l #14281, %d1
+; CHECK-NEXT:    rts
+  ret i64 14281
+}
+
+define ptr @return_null() {
+; CHECK-LABEL: return_null:
+; CHECK:         .cfi_startproc
+; CHECK-NEXT:  ; %bb.0:
+; CHECK-NEXT:    suba.l %a0, %a0
+; CHECK-NEXT:    rts
+  ret ptr null
+}
+
+define ptr @return_nonnull() {
+; CHECK-LABEL: return_nonnull:
+; CHECK:         .cfi_startproc
+; CHECK-NEXT:  ; %bb.0:
+; CHECK-NEXT:    move.w #200, %a0
+; CHECK-NEXT:    rts
+  ret ptr inttoptr (i32 200 to ptr)
+}
+
+define ptr @return_large_nonnull() {
+; CHECK-LABEL: return_large_nonnull:
+; CHECK:         .cfi_startproc
+; CHECK-NEXT:  ; %bb.0:
+; CHECK-NEXT:    move.l #74281, %a0
+; CHECK-NEXT:    rts
+  ret ptr inttoptr (i32 74281 to ptr)
+}