AMDGPU: Use SGPR_128 instead of SReg_128 for vregs

SGPR_128 only includes the real allocatable SGPRs, and SReg_128 adds
the additional non-allocatable TTMP registers. There's no point in
allocating SReg_128 vregs. This shrinks the size of the classes
regalloc needs to consider, which is usually good.

llvm-svn: 374284
diff --git a/llvm/lib/Target/AMDGPU/SIMachineFunctionInfo.cpp b/llvm/lib/Target/AMDGPU/SIMachineFunctionInfo.cpp
index 58f4590..73115af 100644
--- a/llvm/lib/Target/AMDGPU/SIMachineFunctionInfo.cpp
+++ b/llvm/lib/Target/AMDGPU/SIMachineFunctionInfo.cpp
@@ -189,7 +189,7 @@
   const SIRegisterInfo &TRI) {
   ArgInfo.PrivateSegmentBuffer =
     ArgDescriptor::createRegister(TRI.getMatchingSuperReg(
-    getNextUserSGPR(), AMDGPU::sub0, &AMDGPU::SReg_128RegClass));
+    getNextUserSGPR(), AMDGPU::sub0, &AMDGPU::SGPR_128RegClass));
   NumUserSGPRs += 4;
   return ArgInfo.PrivateSegmentBuffer.getRegister();
 }