R600: Use the same compute kernel calling convention for all GPUs

A side-effect of this is that now the compiler expects kernel arguments
to be 4-byte aligned.

Reviewed-by: Vincent Lejeune <vljn at ovi.com>
llvm-svn: 186916

3 files changed, 27 insertions, 11 deletions

diff --git a/llvm/lib/Target/R600/AMDGPUCallingConv.td b/llvm/lib/Target/R600/AMDGPUCallingConv.td
index 29a0326143b..5c9a3e425dd 100644
--- a/llvm/lib/Target/R600/AMDGPUCallingConv.td
+++ b/llvm/lib/Target/R600/AMDGPUCallingConv.td
@@ -36,9 +36,9 @@ def CC_SI : CallingConv<[
 
 ]>;
 
-// Calling convention for SI compute kernels
-def CC_SI_Kernel : CallingConv<[
-  CCIfType<[v4i32, v4f32], CCAssignToStack <16, 4>>,
+// Calling convention for compute kernels
+def CC_AMDGPU_Kernel : CallingConv<[
+  CCIfType<[v4i32, v4f32], CCAssignToStack <16, 16>>,
   CCIfType<[i64, f64],     CCAssignToStack < 8, 4>>,
   CCIfType<[i32, f32],     CCAssignToStack < 4, 4>>,
   CCIfType<[i16],          CCAssignToStack < 2, 4>>,
@@ -46,8 +46,14 @@ def CC_SI_Kernel : CallingConv<[
 ]>;
 
 def CC_AMDGPU : CallingConv<[
-  CCIf<"State.getMachineFunction().getInfo<SIMachineFunctionInfo>()->"#
-       "ShaderType == ShaderType::COMPUTE", CCDelegateTo<CC_SI_Kernel>>,
+  CCIf<"State.getTarget().getSubtarget<AMDGPUSubtarget>().getGeneration() == "
+       "AMDGPUSubtarget::SOUTHERN_ISLANDS && "
+       "State.getMachineFunction().getInfo<SIMachineFunctionInfo>()->"#
+       "ShaderType == ShaderType::COMPUTE", CCDelegateTo<CC_AMDGPU_Kernel>>,
+  CCIf<"State.getTarget().getSubtarget<AMDGPUSubtarget>().getGeneration() < "
+       "AMDGPUSubtarget::SOUTHERN_ISLANDS && "
+       "State.getMachineFunction().getInfo<R600MachineFunctionInfo>()->"
+       "ShaderType == ShaderType::COMPUTE", CCDelegateTo<CC_AMDGPU_Kernel>>,
   CCIf<"State.getTarget().getSubtarget<AMDGPUSubtarget>()"#
        ".getGeneration() == AMDGPUSubtarget::SOUTHERN_ISLANDS", CCDelegateTo<CC_SI>>
 ]>;
diff --git a/llvm/lib/Target/R600/AMDGPUISelLowering.cpp b/llvm/lib/Target/R600/AMDGPUISelLowering.cpp
index d74d9f89c37..c90176b8f88 100644
--- a/llvm/lib/Target/R600/AMDGPUISelLowering.cpp
+++ b/llvm/lib/Target/R600/AMDGPUISelLowering.cpp
@@ -18,6 +18,7 @@
 #include "AMDGPURegisterInfo.h"
 #include "AMDGPUSubtarget.h"
 #include "AMDILIntrinsicInfo.h"
+#include "R600MachineFunctionInfo.h"
 #include "SIMachineFunctionInfo.h"
 #include "llvm/CodeGen/CallingConvLower.h"
 #include "llvm/CodeGen/MachineFunction.h"
diff --git a/llvm/lib/Target/R600/R600ISelLowering.cpp b/llvm/lib/Target/R600/R600ISelLowering.cpp
index ac4a81c9ac5..7f93f23f280 100644
--- a/llvm/lib/Target/R600/R600ISelLowering.cpp
+++ b/llvm/lib/Target/R600/R600ISelLowering.cpp
@@ -16,6 +16,7 @@
 #include "R600Defines.h"
 #include "R600InstrInfo.h"
 #include "R600MachineFunctionInfo.h"
+#include "llvm/CodeGen/CallingConvLower.h"
 #include "llvm/CodeGen/MachineFrameInfo.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
@@ -1212,11 +1213,17 @@ SDValue R600TargetLowering::LowerFormalArguments(
                                       const SmallVectorImpl<ISD::InputArg> &Ins,
                                       SDLoc DL, SelectionDAG &DAG,
                                       SmallVectorImpl<SDValue> &InVals) const {
-  unsigned ParamOffsetBytes = 36;
+  SmallVector<CCValAssign, 16> ArgLocs;
+  CCState CCInfo(CallConv, isVarArg, DAG.getMachineFunction(),
+                 getTargetMachine(), ArgLocs, *DAG.getContext());
+
+  AnalyzeFormalArguments(CCInfo, Ins);
+
   Function::const_arg_iterator FuncArg =
                             DAG.getMachineFunction().getFunction()->arg_begin();
   for (unsigned i = 0, e = Ins.size(); i < e; ++i, ++FuncArg) {
-    EVT VT = Ins[i].VT;
+    CCValAssign &VA = ArgLocs[i];
+    EVT VT = VA.getLocVT();
     Type *ArgType = FuncArg->getType();
     unsigned ArgSizeInBits = ArgType->isPointerTy() ?
                              32 : ArgType->getPrimitiveSizeInBits();
@@ -1239,12 +1246,14 @@ SDValue R600TargetLowering::LowerFormalArguments(
 
     PointerType *PtrTy = PointerType::get(VT.getTypeForEVT(*DAG.getContext()),
                                                     AMDGPUAS::PARAM_I_ADDRESS);
+
+    // The first 36 bytes of the input buffer contains information about
+    // thread group and global sizes.
     SDValue Arg = DAG.getExtLoad(LoadType, DL, VT, DAG.getRoot(),
-                                DAG.getConstant(ParamOffsetBytes, MVT::i32),
-                                       MachinePointerInfo(UndefValue::get(PtrTy)),
-                                       ArgVT, false, false, ArgBytes);
+                           DAG.getConstant(36 + VA.getLocMemOffset(), MVT::i32),
+                           MachinePointerInfo(UndefValue::get(PtrTy)),
+                           ArgVT, false, false, ArgBytes);
     InVals.push_back(Arg);
-    ParamOffsetBytes += ArgBytes;
   }
   return Chain;
 }