AMDGPU/SI: Enable frame index scavenging during PrologEpilogueInserter

Summary:
This allows us to use virtual registers when we need extra registers
for inserting spill instructions in SIRegisterInfo:eliminateFrameIndex().

Once all the frame indices have been eliminated, the
PrologEpilogueInserter does an extra pass over the program to replace
all virtual registers with physical ones.

This allows us to make more efficient use of our emergency spill slots,
so we only need to create one.

Reviewers: arsenm

Subscribers: arsenm, llvm-commits

Differential Revision: http://reviews.llvm.org/D17591

llvm-svn: 262728

diff --git a/llvm/lib/Target/AMDGPU/SIRegisterInfo.cpp b/llvm/lib/Target/AMDGPU/SIRegisterInfo.cpp
index a1eb185cf096f7efdc95597125b445fdc444b786..80b446939e4ebcd3e3246883d79cb472eb7deee6 100644 (file)
--- a/llvm/lib/Target/AMDGPU/SIRegisterInfo.cpp
+++ b/llvm/lib/Target/AMDGPU/SIRegisterInfo.cpp
@@ -182,6 +182,11 @@ bool SIRegisterInfo::requiresRegisterScavenging(const MachineFunction &Fn) const
   return Fn.getFrameInfo()->hasStackObjects();
 }
 
+bool
+SIRegisterInfo::requiresFrameIndexScavenging(const MachineFunction &MF) const {
+  return MF.getFrameInfo()->hasStackObjects();
+}
+
 static unsigned getNumSubRegsForSpillOp(unsigned Op) {
 
   switch (Op) {
@@ -222,11 +227,11 @@ void SIRegisterInfo::buildScratchLoadStore(MachineBasicBlock::iterator MI,
                                            unsigned Value,
                                            unsigned ScratchRsrcReg,
                                            unsigned ScratchOffset,
-                                           int64_t Offset,
-                                           RegScavenger *RS) const {
+                                           int64_t Offset) const {
 
   MachineBasicBlock *MBB = MI->getParent();
-  const MachineFunction *MF = MI->getParent()->getParent();
+  MachineFunction *MF = MI->getParent()->getParent();
+  MachineRegisterInfo &MRI = MF->getRegInfo();
   const SIInstrInfo *TII =
       static_cast<const SIInstrInfo *>(MF->getSubtarget().getInstrInfo());
   LLVMContext &Ctx = MF->getFunction()->getContext();
@@ -241,7 +246,7 @@ void SIRegisterInfo::buildScratchLoadStore(MachineBasicBlock::iterator MI,
   unsigned Size = NumSubRegs * 4;
 
   if (!isUInt<12>(Offset + Size)) {
-    SOffset = RS->scavengeRegister(&AMDGPU::SGPR_32RegClass, MI, 0);
+    SOffset = MRI.createVirtualRegister(&AMDGPU::SGPR_32RegClass);
     if (SOffset == AMDGPU::NoRegister) {
       RanOutOfSGPRs = true;
       SOffset = AMDGPU::SGPR0;
@@ -283,6 +288,7 @@ void SIRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator MI,
                                         int SPAdj, unsigned FIOperandNum,
                                         RegScavenger *RS) const {
   MachineFunction *MF = MI->getParent()->getParent();
+  MachineRegisterInfo &MRI = MF->getRegInfo();
   MachineBasicBlock *MBB = MI->getParent();
   SIMachineFunctionInfo *MFI = MF->getInfo<SIMachineFunctionInfo>();
   MachineFrameInfo *FrameInfo = MF->getFrameInfo();
@@ -375,7 +381,7 @@ void SIRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator MI,
             TII->getNamedOperand(*MI, AMDGPU::OpName::src)->getReg(),
             TII->getNamedOperand(*MI, AMDGPU::OpName::scratch_rsrc)->getReg(),
             TII->getNamedOperand(*MI, AMDGPU::OpName::scratch_offset)->getReg(),
-             FrameInfo->getObjectOffset(Index), RS);
+             FrameInfo->getObjectOffset(Index));
       MI->eraseFromParent();
       break;
     case AMDGPU::SI_SPILL_V32_RESTORE:
@@ -388,7 +394,7 @@ void SIRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator MI,
             TII->getNamedOperand(*MI, AMDGPU::OpName::dst)->getReg(),
             TII->getNamedOperand(*MI, AMDGPU::OpName::scratch_rsrc)->getReg(),
             TII->getNamedOperand(*MI, AMDGPU::OpName::scratch_offset)->getReg(),
-            FrameInfo->getObjectOffset(Index), RS);
+            FrameInfo->getObjectOffset(Index));
       MI->eraseFromParent();
       break;
     }
@@ -397,7 +403,7 @@ void SIRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator MI,
       int64_t Offset = FrameInfo->getObjectOffset(Index);
       FIOp.ChangeToImmediate(Offset);
       if (!TII->isImmOperandLegal(MI, FIOperandNum, FIOp)) {
-        unsigned TmpReg = RS->scavengeRegister(&AMDGPU::VGPR_32RegClass, MI, SPAdj);
+        unsigned TmpReg = MRI.createVirtualRegister(&AMDGPU::VGPR_32RegClass);
         BuildMI(*MBB, MI, MI->getDebugLoc(),
                 TII->get(AMDGPU::V_MOV_B32_e32), TmpReg)
                 .addImm(Offset);

diff --git a/llvm/lib/Target/AMDGPU/SIRegisterInfo.h b/llvm/lib/Target/AMDGPU/SIRegisterInfo.h
index 76eaa2cdc60b5f8d8e315af6751b5105feae48ef..0916de02e2b6d97e89f2ef97616a262e0fadcec9 100644 (file)
--- a/llvm/lib/Target/AMDGPU/SIRegisterInfo.h
+++ b/llvm/lib/Target/AMDGPU/SIRegisterInfo.h
@@ -49,6 +49,8 @@ public:
 
   bool requiresRegisterScavenging(const MachineFunction &Fn) const override;
 
+  bool requiresFrameIndexScavenging(const MachineFunction &MF) const override;
+
   void eliminateFrameIndex(MachineBasicBlock::iterator MI, int SPAdj,
                            unsigned FIOperandNum,
                            RegScavenger *RS) const override;
@@ -162,7 +164,7 @@ private:
   void buildScratchLoadStore(MachineBasicBlock::iterator MI,
                              unsigned LoadStoreOp, unsigned Value,
                              unsigned ScratchRsrcReg, unsigned ScratchOffset,
-                             int64_t Offset, RegScavenger *RS) const;
+                             int64_t Offset) const;
 };
 
 } // End namespace llvm