R600/SI: Teach SIInstrInfo::moveToVALU() how to move S_LOAD_*_IMM instructions

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@208432 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/R600/SIInstrInfo.cpp b/lib/Target/R600/SIInstrInfo.cpp
index 454b7c2d55f878e80364420ee8fbbabf7abb3ee6..029b97ee36a97abffd6ed7a02035912157bcf616 100644 (file)
--- a/lib/Target/R600/SIInstrInfo.cpp
+++ b/lib/Target/R600/SIInstrInfo.cpp
@@ -661,8 +661,11 @@ unsigned SIInstrInfo::getVALUOp(const MachineInstr &MI) {
   case AMDGPU::S_CMP_GE_I32: return AMDGPU::V_CMP_GE_I32_e32;
   case AMDGPU::S_CMP_LT_I32: return AMDGPU::V_CMP_LT_I32_e32;
   case AMDGPU::S_CMP_LE_I32: return AMDGPU::V_CMP_LE_I32_e32;
+  case AMDGPU::S_LOAD_DWORD_IMM:
   case AMDGPU::S_LOAD_DWORD_SGPR: return AMDGPU::BUFFER_LOAD_DWORD_ADDR64;
+  case AMDGPU::S_LOAD_DWORDX2_IMM:
   case AMDGPU::S_LOAD_DWORDX2_SGPR: return AMDGPU::BUFFER_LOAD_DWORDX2_ADDR64;
+  case AMDGPU::S_LOAD_DWORDX4_IMM:
   case AMDGPU::S_LOAD_DWORDX4_SGPR: return AMDGPU::BUFFER_LOAD_DWORDX4_ADDR64;
   }
 }
@@ -1029,15 +1032,39 @@ void SIInstrInfo::legalizeOperands(MachineInstr *MI) const {
 void SIInstrInfo::moveSMRDToVALU(MachineInstr *MI, MachineRegisterInfo &MRI) const {
   MachineBasicBlock *MBB = MI->getParent();
   switch (MI->getOpcode()) {
+    case AMDGPU::S_LOAD_DWORD_IMM:
     case AMDGPU::S_LOAD_DWORD_SGPR:
+    case AMDGPU::S_LOAD_DWORDX2_IMM:
     case AMDGPU::S_LOAD_DWORDX2_SGPR:
+    case AMDGPU::S_LOAD_DWORDX4_IMM:
     case AMDGPU::S_LOAD_DWORDX4_SGPR:
       unsigned NewOpcode = getVALUOp(*MI);
-      unsigned Offset = MI->getOperand(2).getReg();
+      unsigned RegOffset;
+      unsigned ImmOffset;
 
+      if (MI->getOperand(2).isReg()) {
+        RegOffset = MI->getOperand(2).getReg();
+        ImmOffset = 0;
+      } else {
+        assert(MI->getOperand(2).isImm());
+        // SMRD instructions take a dword offsets and MUBUF instructions
+        // take a byte offset.
+        ImmOffset = MI->getOperand(2).getImm() << 2;
+        RegOffset = MRI.createVirtualRegister(&AMDGPU::SGPR_32RegClass);
+        if (isUInt<12>(ImmOffset)) {
+          BuildMI(*MBB, MI, MI->getDebugLoc(), get(AMDGPU::S_MOV_B32),
+                  RegOffset)
+                  .addImm(0);
+        } else {
+          BuildMI(*MBB, MI, MI->getDebugLoc(), get(AMDGPU::S_MOV_B32),
+                  RegOffset)
+                  .addImm(ImmOffset);
+          ImmOffset = 0;
+        }
+      }
 
       unsigned SRsrc = MRI.createVirtualRegister(&AMDGPU::SReg_128RegClass);
-      unsigned DWord0 = Offset;
+      unsigned DWord0 = RegOffset;
       unsigned DWord1 = MRI.createVirtualRegister(&AMDGPU::SGPR_32RegClass);
       unsigned DWord2 = MRI.createVirtualRegister(&AMDGPU::SGPR_32RegClass);
       unsigned DWord3 = MRI.createVirtualRegister(&AMDGPU::SGPR_32RegClass);
@@ -1058,9 +1085,13 @@ void SIInstrInfo::moveSMRDToVALU(MachineInstr *MI, MachineRegisterInfo &MRI) con
               .addReg(DWord3)
               .addImm(AMDGPU::sub3);
      MI->setDesc(get(NewOpcode));
-     MI->getOperand(2).setReg(MI->getOperand(1).getReg());
+     if (MI->getOperand(2).isReg()) {
+       MI->getOperand(2).setReg(MI->getOperand(1).getReg());
+     } else {
+       MI->getOperand(2).ChangeToRegister(MI->getOperand(1).getReg(), false);
+     }
      MI->getOperand(1).setReg(SRsrc);
-     MI->addOperand(*MBB->getParent(), MachineOperand::CreateImm(0));
+     MI->addOperand(*MBB->getParent(), MachineOperand::CreateImm(ImmOffset));
   }
 }
 

diff --git a/test/CodeGen/R600/salu-to-valu.ll b/test/CodeGen/R600/salu-to-valu.ll
index 3fd0db34598c9c569bb0498452c2c102a6ac029c..e7719b6feb3b933521060895ada90cd257d8b18b 100644 (file)
--- a/test/CodeGen/R600/salu-to-valu.ll
+++ b/test/CodeGen/R600/salu-to-valu.ll
@@ -74,3 +74,17 @@ endif:
   store i32 %6, i32 addrspace(1)* %out
   ret void
 }
+
+; Test moving ann SMRD with an immediate offset to the VALU
+
+; CHECK-LABEL: @smrd_valu2
+; CHECK: BUFFER_LOAD_DWORD
+define void @smrd_valu2(i32 addrspace(1)* %out, [8 x i32] addrspace(2)* %in) {
+entry:
+  %0 = call i32 @llvm.r600.read.tidig.x() nounwind readnone
+  %1 = add i32 %0, 4
+  %2 = getelementptr [8 x i32] addrspace(2)* %in, i32 %0, i32 4
+  %3 = load i32 addrspace(2)* %2
+  store i32 %3, i32 addrspace(1)* %out
+  ret void
+}