AMDGPU: Decompose all values to 32-bit pieces for calling conventions

This is the more natural lowering, and presents more opportunities to
reduce 64-bit ops to 32-bit.

This should also help avoid issues graphics shaders have had with
64-bit values, and simplify argument lowering in globalisel.

llvm-svn: 366578

1 files changed, 0 insertions, 74 deletions

diff --git a/llvm/lib/Target/AMDGPU/AMDGPUISelLowering.cpp b/llvm/lib/Target/AMDGPU/AMDGPUISelLowering.cpp
index 39016ed3719..73aff33f2e0 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUISelLowering.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUISelLowering.cpp
@@ -39,80 +39,6 @@
 #include "llvm/Support/KnownBits.h"
 using namespace llvm;
 
-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 == AMDGPU::NoRegister)
-    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:
-  case MVT::v4i16:
-  case MVT::v4f16: {
-    // Up to SGPR0-SGPR105
-    return allocateCCRegs(ValNo, ValVT, LocVT, LocInfo, ArgFlags, State,
-                          &AMDGPU::SGPR_64RegClass, 53);
-  }
-  default:
-    return false;
-  }
-}
-
-// Allocate up to VGPR31.
-//
-// TODO: Since there are no VGPR alignent requirements would it be better to
-// split into individual scalar registers?
-static bool allocateVGPRTuple(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:
-  case MVT::v4i16:
-  case MVT::v4f16: {
-    return allocateCCRegs(ValNo, ValVT, LocVT, LocInfo, ArgFlags, State,
-                          &AMDGPU::VReg_64RegClass, 31);
-  }
-  case MVT::v4i32:
-  case MVT::v4f32:
-  case MVT::v2i64:
-  case MVT::v2f64: {
-    return allocateCCRegs(ValNo, ValVT, LocVT, LocInfo, ArgFlags, State,
-                          &AMDGPU::VReg_128RegClass, 29);
-  }
-  case MVT::v8i32:
-  case MVT::v8f32: {
-    return allocateCCRegs(ValNo, ValVT, LocVT, LocInfo, ArgFlags, State,
-                          &AMDGPU::VReg_256RegClass, 25);
-
-  }
-  case MVT::v16i32:
-  case MVT::v16f32: {
-    return allocateCCRegs(ValNo, ValVT, LocVT, LocInfo, ArgFlags, State,
-                          &AMDGPU::VReg_512RegClass, 17);
-
-  }
-  default:
-    return false;
-  }
-}
-
 #include "AMDGPUGenCallingConv.inc"
 
 // Find a larger type to do a load / store of a vector with.