AMDGPU: add +xnack feature

[oota-llvm.git] / lib / Target / AMDGPU / R600ISelLowering.cpp

diff --git a/lib/Target/AMDGPU/R600ISelLowering.cpp b/lib/Target/AMDGPU/R600ISelLowering.cpp
index ecda332ed6647ca9ba5f49e609f258c7eba64f96..124a9c6e0f564046af0d26244e912c0174c52099 100644 (file)
--- a/lib/Target/AMDGPU/R600ISelLowering.cpp
+++ b/lib/Target/AMDGPU/R600ISelLowering.cpp
@@ -190,6 +190,10 @@ R600TargetLowering::R600TargetLowering(TargetMachine &TM,
   setSchedulingPreference(Sched::Source);
 }
 
+static inline bool isEOP(MachineBasicBlock::iterator I) {
+  return std::next(I)->getOpcode() == AMDGPU::RETURN;
+}
+
 MachineBasicBlock * R600TargetLowering::EmitInstrWithCustomInserter(
     MachineInstr * MI, MachineBasicBlock * BB) const {
   MachineFunction * MF = BB->getParent();
@@ -276,12 +280,18 @@ MachineBasicBlock * R600TargetLowering::EmitInstrWithCustomInserter(
   case AMDGPU::RAT_WRITE_CACHELESS_32_eg:
   case AMDGPU::RAT_WRITE_CACHELESS_64_eg:
   case AMDGPU::RAT_WRITE_CACHELESS_128_eg: {
-    unsigned EOP = (std::next(I)->getOpcode() == AMDGPU::RETURN) ? 1 : 0;
-
     BuildMI(*BB, I, BB->findDebugLoc(I), TII->get(MI->getOpcode()))
             .addOperand(MI->getOperand(0))
             .addOperand(MI->getOperand(1))
-            .addImm(EOP); // Set End of program bit
+            .addImm(isEOP(I)); // Set End of program bit
+    break;
+  }
+  case AMDGPU::RAT_STORE_TYPED_eg: {
+    BuildMI(*BB, I, BB->findDebugLoc(I), TII->get(MI->getOpcode()))
+            .addOperand(MI->getOperand(0))
+            .addOperand(MI->getOperand(1))
+            .addOperand(MI->getOperand(2))
+            .addImm(isEOP(I)); // Set End of program bit
     break;
   }
 
@@ -539,7 +549,7 @@ MachineBasicBlock * R600TargetLowering::EmitInstrWithCustomInserter(
         }
       }
     }
-    bool EOP = (std::next(I)->getOpcode() == AMDGPU::RETURN) ? 1 : 0;
+    bool EOP = isEOP(I);
     if (!EOP && !isLastInstructionOfItsType)
       return BB;
     unsigned CfInst = (MI->getOpcode() == AMDGPU::EG_ExportSwz)? 84 : 40;
@@ -946,6 +956,8 @@ SDValue R600TargetLowering::LowerTrig(SDValue Op, SelectionDAG &DAG) const {
   EVT VT = Op.getValueType();
   SDValue Arg = Op.getOperand(0);
   SDLoc DL(Op);
+
+  // TODO: Should this propagate fast-math-flags?
   SDValue FractPart = DAG.getNode(AMDGPUISD::FRACT, DL, VT,
       DAG.getNode(ISD::FADD, DL, VT,
         DAG.getNode(ISD::FMUL, DL, VT, Arg,