R600/SI: add proper formal parameter handling for SI

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
llvm-svn: 176623

diff --git a/llvm/lib/Target/R600/AMDGPU.td b/llvm/lib/Target/R600/AMDGPU.td
index 40f4741..1a26c77 100644 (file)
--- a/llvm/lib/Target/R600/AMDGPU.td
+++ b/llvm/lib/Target/R600/AMDGPU.td
@@ -38,3 +38,4 @@ include "AMDGPUInstrInfo.td"
 include "AMDGPUIntrinsics.td"
 include "AMDGPURegisterInfo.td"
 include "AMDGPUInstructions.td"
+include "AMDGPUCallingConv.td"

diff --git a/llvm/lib/Target/R600/AMDGPUCallingConv.td b/llvm/lib/Target/R600/AMDGPUCallingConv.td
new file mode 100644 (file)
index 0000000..45ae37e
--- /dev/null
+++ b/llvm/lib/Target/R600/AMDGPUCallingConv.td
@@ -0,0 +1,42 @@
+//===---- AMDCallingConv.td - Calling Conventions for Radeon GPUs ---------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This describes the calling conventions for the AMD Radeon GPUs.
+//
+//===----------------------------------------------------------------------===//
+
+// Inversion of CCIfInReg
+class CCIfNotInReg<CCAction A> : CCIf<"!ArgFlags.isInReg()", A> {}
+
+// Calling convention for SI
+def CC_SI : CallingConv<[
+
+  CCIfInReg<CCIfType<[f32, i32] , CCAssignToReg<[
+    SGPR0, SGPR1, SGPR2, SGPR3, SGPR4, SGPR5, SGPR6, SGPR7,
+    SGPR8, SGPR9, SGPR10, SGPR11, SGPR12, SGPR13, SGPR14, SGPR15
+  ]>>>,
+
+  CCIfInReg<CCIfType<[i64] , CCAssignToRegWithShadow<
+    [ SGPR0, SGPR2, SGPR4, SGPR6, SGPR8, SGPR10, SGPR12, SGPR14 ],
+    [ SGPR1, SGPR3, SGPR5, SGPR7, SGPR9, SGPR11, SGPR12, SGPR15 ]
+  >>>,
+
+  CCIfNotInReg<CCIfType<[f32, i32] , CCAssignToReg<[
+    VGPR0, VGPR1, VGPR2, VGPR3, VGPR4, VGPR5, VGPR6, VGPR7,
+    VGPR8, VGPR9, VGPR10, VGPR11, VGPR12, VGPR13, VGPR14, VGPR15,
+    VGPR16, VGPR17, VGPR18, VGPR19, VGPR20, VGPR21, VGPR22, VGPR23,
+    VGPR24, VGPR25, VGPR26, VGPR27, VGPR28, VGPR29, VGPR30, VGPR31
+  ]>>>
+
+]>;
+
+def CC_AMDGPU : CallingConv<[
+  CCIf<"State.getTarget().getSubtarget<AMDGPUSubtarget>().device()"#
+       "->getGeneration() == AMDGPUDeviceInfo::HD7XXX", CCDelegateTo<CC_SI>>
+]>;

diff --git a/llvm/lib/Target/R600/AMDGPUISelLowering.cpp b/llvm/lib/Target/R600/AMDGPUISelLowering.cpp
index 0a33264..5995b6f 100644 (file)
--- a/llvm/lib/Target/R600/AMDGPUISelLowering.cpp
+++ b/llvm/lib/Target/R600/AMDGPUISelLowering.cpp
@@ -14,7 +14,10 @@
 //===----------------------------------------------------------------------===//
 
 #include "AMDGPUISelLowering.h"
+#include "AMDGPURegisterInfo.h"
 #include "AMDILIntrinsicInfo.h"
+#include "AMDGPUSubtarget.h"
+#include "llvm/CodeGen/CallingConvLower.h"
 #include "llvm/CodeGen/MachineFunction.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
 #include "llvm/CodeGen/SelectionDAG.h"
@@ -22,6 +25,8 @@
 
 using namespace llvm;
 
+#include "AMDGPUGenCallingConv.inc"
+
 AMDGPUTargetLowering::AMDGPUTargetLowering(TargetMachine &TM) :
   TargetLowering(TM, new TargetLoweringObjectFileELF()) {
 
@@ -64,17 +69,10 @@ AMDGPUTargetLowering::AMDGPUTargetLowering(TargetMachine &TM) :
 // TargetLowering Callbacks
 //===---------------------------------------------------------------------===//
 
-SDValue AMDGPUTargetLowering::LowerFormalArguments(
-                                      SDValue Chain,
-                                      CallingConv::ID CallConv,
-                                      bool isVarArg,
-                                      const SmallVectorImpl<ISD::InputArg> &Ins,
-                                      DebugLoc DL, SelectionDAG &DAG,
-                                      SmallVectorImpl<SDValue> &InVals) const {
-  for (unsigned i = 0, e = Ins.size(); i < e; ++i) {
-    InVals.push_back(SDValue());
-  }
-  return Chain;
+void AMDGPUTargetLowering::AnalyzeFormalArguments(CCState &State,
+                             const SmallVectorImpl<ISD::InputArg> &Ins) const {
+
+  State.AnalyzeFormalArguments(Ins, CC_AMDGPU);
 }
 
 SDValue AMDGPUTargetLowering::LowerReturn(

diff --git a/llvm/lib/Target/R600/AMDGPUISelLowering.h b/llvm/lib/Target/R600/AMDGPUISelLowering.h
index 9e7d997..f31b646 100644 (file)
--- a/llvm/lib/Target/R600/AMDGPUISelLowering.h
+++ b/llvm/lib/Target/R600/AMDGPUISelLowering.h
@@ -39,15 +39,12 @@ protected:
   bool isHWTrueValue(SDValue Op) const;
   bool isHWFalseValue(SDValue Op) const;
 
+  void AnalyzeFormalArguments(CCState &State,
+                              const SmallVectorImpl<ISD::InputArg> &Ins) const;
+
 public:
   AMDGPUTargetLowering(TargetMachine &TM);
 
-  virtual SDValue LowerFormalArguments(SDValue Chain, CallingConv::ID CallConv,
-                             bool isVarArg,
-                             const SmallVectorImpl<ISD::InputArg> &Ins,
-                             DebugLoc DL, SelectionDAG &DAG,
-                             SmallVectorImpl<SDValue> &InVals) const;
-
   virtual SDValue LowerReturn(SDValue Chain, CallingConv::ID CallConv,
                               bool isVarArg,
                               const SmallVectorImpl<ISD::OutputArg> &Outs,

diff --git a/llvm/lib/Target/R600/AMDILISelLowering.cpp b/llvm/lib/Target/R600/AMDILISelLowering.cpp
index f65e1f3..922cac1 100644 (file)
--- a/llvm/lib/Target/R600/AMDILISelLowering.cpp
+++ b/llvm/lib/Target/R600/AMDILISelLowering.cpp
@@ -33,11 +33,6 @@
 
 using namespace llvm;
 //===----------------------------------------------------------------------===//
-// Calling Convention Implementation
-//===----------------------------------------------------------------------===//
-#include "AMDGPUGenCallingConv.inc"
-
-//===----------------------------------------------------------------------===//
 // TargetLowering Implementation Help Functions End
 //===----------------------------------------------------------------------===//
 

diff --git a/llvm/lib/Target/R600/SIISelLowering.cpp b/llvm/lib/Target/R600/SIISelLowering.cpp
index a9eca31..8b46cb0 100644 (file)
--- a/llvm/lib/Target/R600/SIISelLowering.cpp
+++ b/llvm/lib/Target/R600/SIISelLowering.cpp
@@ -18,6 +18,8 @@
 #include "SIInstrInfo.h"
 #include "SIMachineFunctionInfo.h"
 #include "SIRegisterInfo.h"
+#include "llvm/IR/Function.h"
+#include "llvm/CodeGen/CallingConvLower.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
 #include "llvm/CodeGen/SelectionDAG.h"
@@ -74,6 +76,105 @@ SITargetLowering::SITargetLowering(TargetMachine &TM) :
   setTargetDAGCombine(ISD::SETCC);
 }
 
+SDValue SITargetLowering::LowerFormalArguments(
+                                      SDValue Chain,
+                                      CallingConv::ID CallConv,
+                                      bool isVarArg,
+                                      const SmallVectorImpl<ISD::InputArg> &Ins,
+                                      DebugLoc DL, SelectionDAG &DAG,
+                                      SmallVectorImpl<SDValue> &InVals) const {
+
+  const TargetRegisterInfo *TRI = getTargetMachine().getRegisterInfo();
+
+  MachineFunction &MF = DAG.getMachineFunction();
+  FunctionType *FType = MF.getFunction()->getFunctionType();
+
+  assert(CallConv == CallingConv::C);
+
+  SmallVector<ISD::InputArg, 16> Splits;
+  for (unsigned i = 0, e = Ins.size(); i != e; ++i) {
+    const ISD::InputArg &Arg = Ins[i];
+   
+    // Split vertices into their elements
+    if (Arg.VT.isVector()) {
+      ISD::InputArg NewArg = Arg;
+      NewArg.Flags.setSplit();
+      NewArg.VT = Arg.VT.getVectorElementType();
+
+      // We REALLY want the ORIGINAL number of vertex elements here, e.g. a
+      // three or five element vertex only needs three or five registers,
+      // NOT four or eigth.
+      Type *ParamType = FType->getParamType(Arg.OrigArgIndex);
+      unsigned NumElements = ParamType->getVectorNumElements();
+
+      for (unsigned j = 0; j != NumElements; ++j) {
+        Splits.push_back(NewArg);
+        NewArg.PartOffset += NewArg.VT.getStoreSize();
+      }
+
+    } else {
+      Splits.push_back(Arg);
+    }
+  }
+
+  SmallVector<CCValAssign, 16> ArgLocs;
+  CCState CCInfo(CallConv, isVarArg, DAG.getMachineFunction(),
+                 getTargetMachine(), ArgLocs, *DAG.getContext());
+
+  AnalyzeFormalArguments(CCInfo, Splits);
+
+  for (unsigned i = 0, e = Ins.size(), ArgIdx = 0; i != e; ++i) {
+
+    CCValAssign &VA = ArgLocs[ArgIdx++];
+    assert(VA.isRegLoc() && "Parameter must be in a register!");
+
+    unsigned Reg = VA.getLocReg();
+    MVT VT = VA.getLocVT();
+
+    if (VT == MVT::i64) {
+      // For now assume it is a pointer
+      Reg = TRI->getMatchingSuperReg(Reg, AMDGPU::sub0,
+                                     &AMDGPU::SReg_64RegClass);
+      Reg = MF.addLiveIn(Reg, &AMDGPU::SReg_64RegClass);
+      InVals.push_back(DAG.getCopyFromReg(Chain, DL, Reg, VT));
+      continue;
+    }
+
+    const TargetRegisterClass *RC = TRI->getMinimalPhysRegClass(Reg, VT);
+
+    Reg = MF.addLiveIn(Reg, RC);
+    SDValue Val = DAG.getCopyFromReg(Chain, DL, Reg, VT);
+
+    const ISD::InputArg &Arg = Ins[i];
+    if (Arg.VT.isVector()) {
+
+      // Build a vector from the registers
+      Type *ParamType = FType->getParamType(Arg.OrigArgIndex);
+      unsigned NumElements = ParamType->getVectorNumElements();
+
+      SmallVector<SDValue, 4> Regs;
+      Regs.push_back(Val);
+      for (unsigned j = 1; j != NumElements; ++j) {
+        Reg = ArgLocs[ArgIdx++].getLocReg();
+        Reg = MF.addLiveIn(Reg, RC);
+        Regs.push_back(DAG.getCopyFromReg(Chain, DL, Reg, VT));
+      }
+
+      // Fill up the missing vector elements
+      NumElements = Arg.VT.getVectorNumElements() - NumElements;
+      for (unsigned j = 0; j != NumElements; ++j)
+        Regs.push_back(DAG.getUNDEF(VT));
+ 
+      InVals.push_back(DAG.getNode(ISD::BUILD_VECTOR, DL, Arg.VT,
+                                   Regs.data(), Regs.size()));
+      continue;
+    }
+
+    InVals.push_back(Val);
+  }
+  return Chain;
+}
+
 MachineBasicBlock * SITargetLowering::EmitInstrWithCustomInserter(
     MachineInstr * MI, MachineBasicBlock * BB) const {
   MachineRegisterInfo & MRI = BB->getParent()->getRegInfo();

diff --git a/llvm/lib/Target/R600/SIISelLowering.h b/llvm/lib/Target/R600/SIISelLowering.h
index 737162f..6d77d88 100644 (file)
--- a/llvm/lib/Target/R600/SIISelLowering.h
+++ b/llvm/lib/Target/R600/SIISelLowering.h
@@ -43,6 +43,13 @@ class SITargetLowering : public AMDGPUTargetLowering {
 
 public:
   SITargetLowering(TargetMachine &tm);
+
+  SDValue LowerFormalArguments(SDValue Chain, CallingConv::ID CallConv,
+                               bool isVarArg,
+                               const SmallVectorImpl<ISD::InputArg> &Ins,
+                               DebugLoc DL, SelectionDAG &DAG,
+                               SmallVectorImpl<SDValue> &InVals) const;
+
   virtual MachineBasicBlock * EmitInstrWithCustomInserter(MachineInstr * MI,
                                               MachineBasicBlock * BB) const;
   virtual EVT getSetCCResultType(EVT VT) const;