AMDGPU/R600: Use machine operands instead of ints to track literals

[jvesely]

Summary

This will be used for global addresses

[• tstellarAMD]

lib/Target/AMDGPU/R600ControlFlowFinalizer.cpp

432

What is this value for?

[jvesely]

lib/Target/AMDGPU/R600ControlFlowFinalizer.cpp

432

Just a forgotten magic value during development (I wanted to make it different from 0 used below). It's replaced in the next patch (D19792).

[jvesely]

lib/Target/AMDGPU/R600ControlFlowFinalizer.cpp

432

should I change it to zero?

[• tstellarAMD]

LGTM.

lib/Target/AMDGPU/R600ControlFlowFinalizer.cpp

432

Sure.

[jvesely]

remove debugging magic value

lib/Target/AMDGPU/R600ControlFlowFinalizer.cpp

MakeFetchClause(MachineBasicBlock &MBB, MachineBasicBlock::iterator &I)
     }
     MachineInstr *MIb = BuildMI(MBB, ClauseHead, MBB.findDebugLoc(ClauseHead),
         getHWInstrDesc(IsTex?CF_TC:CF_VC))
         .addImm(0) // ADDR
         .addImm(AluInstCount - 1); // COUNT
     return ClauseFile(MIb, std::move(ClauseContent));
   }
 
-  void getLiteral(MachineInstr *MI, std::vector<int64_t> &Lits) const {
+  void getLiteral(MachineInstr *MI, std::vector<MachineOperand *> &Lits) const {
     static const unsigned LiteralRegs[] = {
       AMDGPU::ALU_LITERAL_X,
       AMDGPU::ALU_LITERAL_Y,
       AMDGPU::ALU_LITERAL_Z,
       AMDGPU::ALU_LITERAL_W
     };
     const SmallVector<std::pair<MachineOperand *, int64_t>, 3 > Srcs =
         TII->getSrcs(MI);
-    for (unsigned i = 0, e = Srcs.size(); i < e; ++i) {
-      if (Srcs[i].first->getReg() != AMDGPU::ALU_LITERAL_X)
+    for (const auto &Src:Srcs) {
+      if (Src.first->getReg() != AMDGPU::ALU_LITERAL_X)
         continue;
-      int64_t Imm = Srcs[i].second;
-      std::vector<int64_t>::iterator It =
-          std::find(Lits.begin(), Lits.end(), Imm);
+      int64_t Imm = Src.second;
+      std::vector<MachineOperand*>::iterator It =
+          std::find_if(Lits.begin(), Lits.end(),
+                    [&](MachineOperand* val)
+                        { return val->isImm() && (val->getImm() == Imm);});
+
+      // Get corresponding Operand
+      MachineOperand &Operand = MI->getOperand(
+              TII->getOperandIdx(MI->getOpcode(), AMDGPU::OpName::literal));
+
       if (It != Lits.end()) {
+        // Reuse existing literal reg
         unsigned Index = It - Lits.begin();
-        Srcs[i].first->setReg(LiteralRegs[Index]);
+        Src.first->setReg(LiteralRegs[Index]);
       } else {
+        // Allocate new literal reg
         assert(Lits.size() < 4 && "Too many literals in Instruction Group");
-        Srcs[i].first->setReg(LiteralRegs[Lits.size()]);
-        Lits.push_back(Imm);
+        Src.first->setReg(LiteralRegs[Lits.size()]);
+        Lits.push_back(&Operand);
       }
     }
   }
 
   MachineBasicBlock::iterator insertLiterals(
       MachineBasicBlock::iterator InsertPos,
       const std::vector<unsigned> &Literals) const {
     MachineBasicBlock *MBB = InsertPos->getParent();
MakeALUClause(MachineBasicBlock &MBB, MachineBasicBlock::iterator &I)
     I++;
     for (MachineBasicBlock::instr_iterator E = MBB.instr_end(); I != E;) {
       if (IsTrivialInst(I)) {
         ++I;
         continue;
       }
       if (!I->isBundle() && !TII->isALUInstr(I->getOpcode()))
         break;
-      std::vector<int64_t> Literals;
+      std::vector<MachineOperand *>Literals;
       if (I->isBundle()) {
         MachineInstr *DeleteMI = I;
         MachineBasicBlock::instr_iterator BI = I.getInstrIterator();
         while (++BI != E && BI->isBundledWithPred()) {
           BI->unbundleFromPred();
-          for (unsigned i = 0, e = BI->getNumOperands(); i != e; ++i) {
-            MachineOperand &MO = BI->getOperand(i);
+          for (MachineOperand &MO : BI->operands()) {
             if (MO.isReg() && MO.isInternalRead())
               MO.setIsInternalRead(false);
           }
           getLiteral(&*BI, Literals);
           ClauseContent.push_back(&*BI);
         }
         I = BI;
         DeleteMI->eraseFromParent();
       } else {
         getLiteral(I, Literals);
         ClauseContent.push_back(I);
         I++;
       }
-      for (unsigned i = 0, e = Literals.size(); i < e; i+=2) {
-        unsigned literal0 = Literals[i];
-        unsigned literal2 = (i + 1 < e)?Literals[i + 1]:0;
-        MachineInstr *MILit = BuildMI(MBB, I, I->getDebugLoc(),
-            TII->get(AMDGPU::LITERALS))
-            .addImm(literal0)
-            .addImm(literal2);
+      for (unsigned i = 0, e = Literals.size(); i < e; i += 2) {
+        MachineInstrBuilder MILit = BuildMI(MBB, I, I->getDebugLoc(),
+            TII->get(AMDGPU::LITERALS));
+        if (Literals[i]->isImm()) {
+            MILit.addImm(Literals[i]->getImm());
+        } else {
+            MILit.addImm(9876543);
+        }
+        if (i + 1 < e) {
+          if (Literals[i + 1]->isImm()) {
+            MILit.addImm(Literals[i + 1]->getImm());
+          } else {
+            MILit.addImm(9876543);
+          }
+        } else
+          MILit.addImm(0);
         ClauseContent.push_back(MILit);
       }
     }
     assert(ClauseContent.size() < 128 && "ALU clause is too big");
     ClauseHead->getOperand(7).setImm(ClauseContent.size() - 1);
     return ClauseFile(ClauseHead, std::move(ClauseContent));
   }