[mips][microMIPS] Use call instructions with short delay slots

[sstankovic]

LGTM, with two changes.

lib/Target/Mips/MipsDelaySlotFiller.cpp

550 ↗

(On Diff #16426)

Typo "chenge". Also, I think it would improve readability if you add parenthesis, since code related to "if" occupies 3 lines.

test/CodeGen/Mips/micromips-delay-slot.ll

16 ↗

(On Diff #16426)

Since jals allows only 16-bits delay slot instructions, you should also CHECK that instruction in delay slot is 16-bits.

Closed by commit rL222752 (authored by zjovanovic).

llvm/trunk/lib/Target/Mips/MipsDelaySlotFiller.cpp

Iter Filler::replaceWithCompactBranch(MachineBasicBlock &MBB,
 
   Iter tmpIter = Branch;
   Branch = std::prev(Branch);
   MBB.erase(tmpIter);
 
   return Branch;
 }
 
+// For given opcode returns opcode of corresponding instruction with short
+// delay slot.
+static int getEquivalentCallShort(int Opcode) {
+  switch (Opcode) {
+  case Mips::BGEZAL:
+    return Mips::BGEZALS_MM;
+  case Mips::BLTZAL:
+    return Mips::BLTZALS_MM;
+  case Mips::JAL:
+    return Mips::JALS_MM;
+  case Mips::JALR:
+    return Mips::JALRS_MM;
+  case Mips::JALR16_MM:
+    return Mips::JALRS16_MM;
+  default:
+    llvm_unreachable("Unexpected call instruction for microMIPS.");
+  }
+}
+
 /// runOnMachineBasicBlock - Fill in delay slots for the given basic block.
 /// We assume there is only one delay slot per delayed instruction.
 bool Filler::runOnMachineBasicBlock(MachineBasicBlock &MBB) {
   bool Changed = false;
   bool InMicroMipsMode = TM.getSubtarget<MipsSubtarget>().inMicroMipsMode();
+  const MipsInstrInfo *TII =
+      static_cast<const MipsInstrInfo *>(TM.getSubtargetImpl()->getInstrInfo());
 
   for (Iter I = MBB.begin(); I != MBB.end(); ++I) {
     if (!hasUnoccupiedSlot(&*I))
       continue;
 
-    // For microMIPS, at the moment, do not fill delay slots of call
-    // instructions.
-    //
-    // TODO: Support for replacing regular call instructions with corresponding
-    // short delay slot instructions should be implemented.
-    if (!InMicroMipsMode || !I->isCall()) {
-      ++FilledSlots;
-      Changed = true;
 
-      // Delay slot filling is disabled at -O0.
-      if (!DisableDelaySlotFiller && (TM.getOptLevel() != CodeGenOpt::None)) {
-        if (searchBackward(MBB, I))
-          continue;
+      bool Filled = false;
 
-        if (I->isTerminator()) {
-          if (searchSuccBBs(MBB, I))
-            continue;
-        } else if (searchForward(MBB, I)) {
-          continue;
+      if (searchBackward(MBB, I)) {
+        Filled = true;
+      } else if (I->isTerminator()) {
+        if (searchSuccBBs(MBB, I)) {
+          Filled = true;
+        }
+        Filled = true;
+      }
+
+      if (Filled) {
+        // Get instruction with delay slot.
+        MachineBasicBlock::instr_iterator DSI(I);
+
+        if (InMicroMipsMode && TII->GetInstSizeInBytes(std::next(DSI)) == 2 &&
+            DSI->isCall()) {
+          // If instruction in delay slot is 16b change opcode to
+          // corresponding instruction with short delay slot.
+          DSI->setDesc(TII->get(getEquivalentCallShort(DSI->getOpcode())));
         }
+
+        continue;
       }
     }
 
     // If instruction is BEQ or BNE with one ZERO register, then instead of
     // adding NOP replace this instruction with the corresponding compact
     // branch instruction, i.e. BEQZC or BNEZC.
     unsigned Opcode = I->getOpcode();
     if (InMicroMipsMode &&
         (Opcode == Mips::BEQ || Opcode == Mips::BNE) &&
         ((unsigned) I->getOperand(1).getReg()) == Mips::ZERO) {
 
       I = replaceWithCompactBranch(MBB, I, I->getDebugLoc());
 
     } else {
       // Bundle the NOP to the instruction with the delay slot.
-      const MipsInstrInfo *TII = static_cast<const MipsInstrInfo *>(
-          TM.getSubtargetImpl()->getInstrInfo());
       BuildMI(MBB, std::next(I), I->getDebugLoc(), TII->get(Mips::NOP));
       MIBundleBuilder(MBB, I, std::next(I, 2));
     }
   }
 
   return Changed;
 }
 

llvm/trunk/test/CodeGen/Mips/micromips-delay-slot.ll

 ; RUN: llc -march=mipsel -mcpu=mips32r2 -mattr=+micromips \
-; RUN:   -relocation-model=pic -O3 < %s | FileCheck %s
+; RUN:   -relocation-model=static -O2 < %s | FileCheck %s
 
-; Function Attrs: nounwind uwtable
-define i32 @foo(i32 %a) #0 {
+; Function Attrs: nounwind
+define i32 @foo(i32 signext %a) #0 {
 entry:
   %a.addr = alloca i32, align 4
   store i32 %a, i32* %a.addr, align 4
   %0 = load i32* %a.addr, align 4
   %shl = shl i32 %0, 2
-  %call = call i32 @bar(i32 %shl)
+  %call = call i32 @bar(i32 signext %shl)
   ret i32 %call
 }
 
-declare i32 @bar(i32) #1
-
-; CHECK: nop
+declare i32 @bar(i32 signext) #1
 
+; CHECK:      jals
+; CHECK-NEXT: sll16