Index: lib/Target/AMDGPU/AMDGPUCallingConv.td
===================================================================
--- lib/Target/AMDGPU/AMDGPUCallingConv.td
+++ lib/Target/AMDGPU/AMDGPUCallingConv.td
@@ -25,14 +25,10 @@
     SGPR32, SGPR33, SGPR34, SGPR35, SGPR36, SGPR37, SGPR38, SGPR39
   ]>>>,
 
-  CCIfInReg<CCIfType<[i64] , CCAssignToRegWithShadow<
-    [ SGPR0, SGPR2, SGPR4, SGPR6, SGPR8, SGPR10, SGPR12, SGPR14,
-      SGPR16, SGPR18, SGPR20, SGPR22, SGPR24, SGPR26, SGPR28, SGPR30,
-      SGPR32, SGPR34, SGPR36, SGPR38 ],
-    [ SGPR1, SGPR3, SGPR5, SGPR7, SGPR9, SGPR11, SGPR13, SGPR15,
-      SGPR17, SGPR19, SGPR21, SGPR23, SGPR25, SGPR27, SGPR29, SGPR31,
-      SGPR33, SGPR35, SGPR37, SGPR39 ]
-  >>>,
+  // We have no way of referring to the generated register tuples
+  // here, so use a custom function.
+  CCIfInReg<CCIfType<[i64], CCCustom<"allocateSGPRTuple">>>,
+  CCIfByVal<CCIfType<[i64], CCCustom<"allocateSGPRTuple">>>,
 
   // 32*4 + 4 is the minimum for a fetch shader consumer with 32 inputs.
   CCIfNotInReg<CCIfType<[f32, i32, f16] , CCAssignToReg<[
@@ -53,17 +49,7 @@
     VGPR112, VGPR113, VGPR114, VGPR115, VGPR116, VGPR117, VGPR118, VGPR119,
     VGPR120, VGPR121, VGPR122, VGPR123, VGPR124, VGPR125, VGPR126, VGPR127,
     VGPR128, VGPR129, VGPR130, VGPR131, VGPR132, VGPR133, VGPR134, VGPR135
-  ]>>>,
-
-  CCIfByVal<CCIfType<[i64] , CCAssignToRegWithShadow<
-    [ SGPR0, SGPR2, SGPR4, SGPR6, SGPR8, SGPR10, SGPR12, SGPR14,
-      SGPR16, SGPR18, SGPR20, SGPR22, SGPR24, SGPR26, SGPR28, SGPR30,
-      SGPR32, SGPR34, SGPR36, SGPR38 ],
-    [ SGPR1, SGPR3, SGPR5, SGPR7, SGPR9, SGPR11, SGPR13, SGPR15,
-      SGPR17, SGPR19, SGPR21, SGPR23, SGPR25, SGPR27, SGPR29, SGPR31,
-      SGPR33, SGPR35, SGPR37, SGPR39 ]
-  >>>
-
+  ]>>>
 ]>;
 
 def RetCC_SI : CallingConv<[
Index: lib/Target/AMDGPU/AMDGPUISelLowering.cpp
===================================================================
--- lib/Target/AMDGPU/AMDGPUISelLowering.cpp
+++ lib/Target/AMDGPU/AMDGPUISelLowering.cpp
@@ -44,6 +44,37 @@
   return true;
 }
 
+static bool allocateCCRegs(unsigned ValNo, MVT ValVT, MVT LocVT,
+                           CCValAssign::LocInfo LocInfo,
+                           ISD::ArgFlagsTy ArgFlags, CCState &State,
+                           const TargetRegisterClass *RC,
+                           unsigned NumRegs) {
+  ArrayRef<MCPhysReg> RegList = makeArrayRef(RC->begin(), NumRegs);
+  unsigned RegResult = State.AllocateReg(RegList);
+  if (!RegResult)
+    return false;
+
+  State.addLoc(CCValAssign::getReg(ValNo, ValVT, RegResult, LocVT, LocInfo));
+  return true;
+}
+
+static bool allocateSGPRTuple(unsigned ValNo, MVT ValVT, MVT LocVT,
+                              CCValAssign::LocInfo LocInfo,
+                              ISD::ArgFlagsTy ArgFlags, CCState &State) {
+  switch (LocVT.SimpleTy) {
+  case MVT::i64:
+  case MVT::f64:
+  case MVT::v2i32:
+  case MVT::v2f32: {
+    // Up to SGPR0-SGPR39
+    return allocateCCRegs(ValNo, ValVT, LocVT, LocInfo, ArgFlags, State,
+                          &AMDGPU::SGPR_64RegClass, 20);
+  }
+  default:
+    return false;
+  }
+}
+
 #include "AMDGPUGenCallingConv.inc"
 
 // Find a larger type to do a load / store of a vector with.
Index: lib/Target/AMDGPU/SIISelLowering.cpp
===================================================================
--- lib/Target/AMDGPU/SIISelLowering.cpp
+++ lib/Target/AMDGPU/SIISelLowering.cpp
@@ -1087,17 +1087,6 @@
     assert(VA.isRegLoc() && "Parameter must be in a register!");
 
     unsigned Reg = VA.getLocReg();
-
-    if (VT == MVT::i64) {
-      // For now assume it is a pointer
-      Reg = TRI->getMatchingSuperReg(Reg, AMDGPU::sub0,
-                                     &AMDGPU::SGPR_64RegClass);
-      Reg = MF.addLiveIn(Reg, &AMDGPU::SGPR_64RegClass);
-      SDValue Copy = DAG.getCopyFromReg(Chain, DL, Reg, VT);
-      InVals.push_back(Copy);
-      continue;
-    }
-
     const TargetRegisterClass *RC = TRI->getMinimalPhysRegClass(Reg, VT);
 
     Reg = MF.addLiveIn(Reg, RC);