R600: Use function inputs to represent data stored in gpr

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194425 91177308-0d34-0410-b5e6-96231b3b80d8

4 files changed, 39 insertions, 41 deletions

diff --git a/lib/Target/R600/AMDGPUCallingConv.td b/lib/Target/R600/AMDGPUCallingConv.td
index fbcdc6aec4..65cdb24673 100644
--- a/lib/Target/R600/AMDGPUCallingConv.td
+++ b/lib/Target/R600/AMDGPUCallingConv.td
@@ -42,6 +42,17 @@ def CC_SI : CallingConv<[
 
 ]>;
 
+// Calling convention for R600
+def CC_R600 : CallingConv<[
+  CCIfInReg<CCIfType<[v4f32, v4i32] , CCAssignToReg<[
+    T0_XYZW, T1_XYZW, T2_XYZW, T3_XYZW, T4_XYZW, T5_XYZW, T6_XYZW, T7_XYZW,
+    T8_XYZW, T9_XYZW, T10_XYZW, T11_XYZW, T12_XYZW, T13_XYZW, T14_XYZW, T15_XYZW,
+    T16_XYZW, T17_XYZW, T18_XYZW, T19_XYZW, T20_XYZW, T21_XYZW, T22_XYZW,
+    T23_XYZW, T24_XYZW, T25_XYZW, T26_XYZW, T27_XYZW, T28_XYZW, T29_XYZW,
+    T30_XYZW, T31_XYZW, T32_XYZW
+  ]>>>
+]>;
+
 // Calling convention for compute kernels
 def CC_AMDGPU_Kernel : CallingConv<[
   CCCustom<"allocateStack">
@@ -57,5 +68,7 @@ def CC_AMDGPU : CallingConv<[
        "State.getMachineFunction().getInfo<R600MachineFunctionInfo>()->"
        "ShaderType == ShaderType::COMPUTE", CCDelegateTo<CC_AMDGPU_Kernel>>,
   CCIf<"State.getTarget().getSubtarget<AMDGPUSubtarget>()"#
-       ".getGeneration() >= AMDGPUSubtarget::SOUTHERN_ISLANDS", CCDelegateTo<CC_SI>>
+       ".getGeneration() >= AMDGPUSubtarget::SOUTHERN_ISLANDS", CCDelegateTo<CC_SI>>,
+  CCIf<"State.getTarget().getSubtarget<AMDGPUSubtarget>()"#
+       ".getGeneration() < AMDGPUSubtarget::SOUTHERN_ISLANDS", CCDelegateTo<CC_R600>>
 ]>;
diff --git a/lib/Target/R600/R600ISelLowering.cpp b/lib/Target/R600/R600ISelLowering.cpp
index a7522b2771..336a2991d1 100644
--- a/lib/Target/R600/R600ISelLowering.cpp
+++ b/lib/Target/R600/R600ISelLowering.cpp
@@ -554,51 +554,23 @@ SDValue R600TargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG) const
     SDLoc DL(Op);
     switch(IntrinsicID) {
     default: return AMDGPUTargetLowering::LowerOperation(Op, DAG);
-    case AMDGPUIntrinsic::R600_load_input: {
-      int64_t RegIndex = cast<ConstantSDNode>(Op.getOperand(1))->getZExtValue();
-      unsigned Reg = AMDGPU::R600_TReg32RegClass.getRegister(RegIndex);
-      MachineFunction &MF = DAG.getMachineFunction();
-      MachineRegisterInfo &MRI = MF.getRegInfo();
-      MRI.addLiveIn(Reg);
-      return DAG.getCopyFromReg(DAG.getEntryNode(),
-          SDLoc(DAG.getEntryNode()), Reg, VT);
-    }
-
-    case AMDGPUIntrinsic::R600_interp_input: {
+    case AMDGPUIntrinsic::R600_interp_xy:
+    case AMDGPUIntrinsic::R600_interp_zw: {
       int slot = cast<ConstantSDNode>(Op.getOperand(1))->getZExtValue();
-      int ijb = cast<ConstantSDNode>(Op.getOperand(2))->getSExtValue();
       MachineSDNode *interp;
-      if (ijb < 0) {
-        const MachineFunction &MF = DAG.getMachineFunction();
-        const R600InstrInfo *TII =
-          static_cast<const R600InstrInfo*>(MF.getTarget().getInstrInfo());
-        interp = DAG.getMachineNode(AMDGPU::INTERP_VEC_LOAD, DL,
-            MVT::v4f32, DAG.getTargetConstant(slot / 4 , MVT::i32));
-        return DAG.getTargetExtractSubreg(
-            TII->getRegisterInfo().getSubRegFromChannel(slot % 4),
-            DL, MVT::f32, SDValue(interp, 0));
-      }
+      SDValue RegisterINode = Op.getOperand(2);
+      SDValue RegisterJNode = Op.getOperand(3);
 
-      MachineFunction &MF = DAG.getMachineFunction();
-      MachineRegisterInfo &MRI = MF.getRegInfo();
-      unsigned RegisterI = AMDGPU::R600_TReg32RegClass.getRegister(2 * ijb);
-      unsigned RegisterJ = AMDGPU::R600_TReg32RegClass.getRegister(2 * ijb + 1);
-      MRI.addLiveIn(RegisterI);
-      MRI.addLiveIn(RegisterJ);
-      SDValue RegisterINode = DAG.getCopyFromReg(DAG.getEntryNode(),
-          SDLoc(DAG.getEntryNode()), RegisterI, MVT::f32);
-      SDValue RegisterJNode = DAG.getCopyFromReg(DAG.getEntryNode(),
-          SDLoc(DAG.getEntryNode()), RegisterJ, MVT::f32);
-
-      if (slot % 4 < 2)
+      if (IntrinsicID == AMDGPUIntrinsic::R600_interp_xy)
         interp = DAG.getMachineNode(AMDGPU::INTERP_PAIR_XY, DL,
-            MVT::f32, MVT::f32, DAG.getTargetConstant(slot / 4 , MVT::i32),
+            MVT::f32, MVT::f32, DAG.getTargetConstant(slot, MVT::i32),
             RegisterJNode, RegisterINode);
       else
         interp = DAG.getMachineNode(AMDGPU::INTERP_PAIR_ZW, DL,
-            MVT::f32, MVT::f32, DAG.getTargetConstant(slot / 4 , MVT::i32),
+            MVT::f32, MVT::f32, DAG.getTargetConstant(slot, MVT::i32),
             RegisterJNode, RegisterINode);
-      return SDValue(interp, slot % 2);
+      return DAG.getNode(ISD::BUILD_VECTOR, DL, MVT::v2f32,
+          SDValue(interp, 0), SDValue(interp, 1));
     }
     case AMDGPUIntrinsic::R600_tex:
     case AMDGPUIntrinsic::R600_texc:
@@ -1339,6 +1311,8 @@ SDValue R600TargetLowering::LowerFormalArguments(
   SmallVector<CCValAssign, 16> ArgLocs;
   CCState CCInfo(CallConv, isVarArg, DAG.getMachineFunction(),
                  getTargetMachine(), ArgLocs, *DAG.getContext());
+  MachineFunction &MF = DAG.getMachineFunction();
+  unsigned ShaderType = MF.getInfo<R600MachineFunctionInfo>()->ShaderType;
 
   SmallVector<ISD::InputArg, 8> LocalIns;
 
@@ -1352,6 +1326,13 @@ SDValue R600TargetLowering::LowerFormalArguments(
     EVT VT = Ins[i].VT;
     EVT MemVT = LocalIns[i].VT;
 
+    if (ShaderType != ShaderType::COMPUTE) {
+      unsigned Reg = MF.addLiveIn(VA.getLocReg(), &AMDGPU::R600_Reg128RegClass);
+      SDValue Register = DAG.getCopyFromReg(Chain, DL, Reg, VT);
+      InVals.push_back(Register);
+      continue;
+    }
+
     PointerType *PtrTy = PointerType::get(VT.getTypeForEVT(*DAG.getContext()),
                                                    AMDGPUAS::CONSTANT_BUFFER_0);
 
diff --git a/lib/Target/R600/R600Instructions.td b/lib/Target/R600/R600Instructions.td
index 0734cca48b..0b2e6ec2bc 100644
--- a/lib/Target/R600/R600Instructions.td
+++ b/lib/Target/R600/R600Instructions.td
@@ -418,7 +418,7 @@ def INTERP_VEC_LOAD :  AMDGPUShaderInst <
   (outs R600_Reg128:$dst),
   (ins i32imm:$src0),
   "INTERP_LOAD $src0 : $dst",
-  []>;
+  [(set R600_Reg128:$dst, (int_R600_interp_const imm:$src0))]>;
 
 def INTERP_XY : R600_2OP <0xD6, "INTERP_XY", []> {
   let bank_swizzle = 5;
diff --git a/lib/Target/R600/R600Intrinsics.td b/lib/Target/R600/R600Intrinsics.td
index b5cb3698f1..cd0b4193c3 100644
--- a/lib/Target/R600/R600Intrinsics.td
+++ b/lib/Target/R600/R600Intrinsics.td
@@ -39,10 +39,14 @@ let TargetPrefix = "R600", isTarget = 1 in {
       llvm_i32_ty // coord_type_w
     ], [IntrNoMem]>;
 
-  def int_R600_load_input :
-    Intrinsic<[llvm_float_ty], [llvm_i32_ty], [IntrNoMem]>;
   def int_R600_interp_input :
     Intrinsic<[llvm_float_ty], [llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>;
+  def int_R600_interp_const :
+    Intrinsic<[llvm_v4f32_ty], [llvm_i32_ty], [IntrNoMem]>;
+def int_R600_interp_xy :
+    Intrinsic<[llvm_v2f32_ty], [llvm_i32_ty, llvm_float_ty, llvm_float_ty], [IntrNoMem]>;
+def int_R600_interp_zw :
+    Intrinsic<[llvm_v2f32_ty], [llvm_i32_ty, llvm_float_ty, llvm_float_ty], [IntrNoMem]>;
   def int_R600_load_texbuf :
     Intrinsic<[llvm_v4f32_ty], [llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>;
   def int_R600_tex : TextureIntrinsicFloatInput;