Index: llvm/trunk/lib/Target/AMDGPU/SIISelLowering.cpp
===================================================================
--- llvm/trunk/lib/Target/AMDGPU/SIISelLowering.cpp
+++ llvm/trunk/lib/Target/AMDGPU/SIISelLowering.cpp
@@ -10464,6 +10464,8 @@
       }
       break;
     case 'a':
+      if (!Subtarget->hasMAIInsts())
+        break;
       switch (VT.getSizeInBits()) {
       default:
         return std::make_pair(0U, nullptr);
Index: llvm/trunk/lib/Target/AMDGPU/SIRegisterInfo.cpp
===================================================================
--- llvm/trunk/lib/Target/AMDGPU/SIRegisterInfo.cpp
+++ llvm/trunk/lib/Target/AMDGPU/SIRegisterInfo.cpp
@@ -220,6 +220,14 @@
     reserveRegisterTuples(Reserved, Reg);
   }
 
+  // Reserve all the rest AGPRs if there are no instructions to use it.
+  if (!ST.hasMAIInsts()) {
+    for (unsigned i = 0; i < MaxNumVGPRs; ++i) {
+      unsigned Reg = AMDGPU::AGPR_32RegClass.getRegister(i);
+      reserveRegisterTuples(Reserved, Reg);
+    }
+  }
+
   const SIMachineFunctionInfo *MFI = MF.getInfo<SIMachineFunctionInfo>();
 
   unsigned ScratchWaveOffsetReg = MFI->getScratchWaveOffsetReg();
Index: llvm/trunk/test/CodeGen/AMDGPU/spill-vgpr-to-agpr.ll
===================================================================
--- llvm/trunk/test/CodeGen/AMDGPU/spill-vgpr-to-agpr.ll
+++ llvm/trunk/test/CodeGen/AMDGPU/spill-vgpr-to-agpr.ll
@@ -1,5 +1,5 @@
 ; RUN: llc -march=amdgcn -mcpu=gfx908 -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GFX908 %s
-; RUN: llc -march=amdgcn -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GFX900 %s
+; RUN: not llc -march=amdgcn -mcpu=gfx900 -verify-machineinstrs < %s 2>&1 | FileCheck -check-prefixes=GCN,GFX900 %s
 
 ; GCN-LABEL: {{^}}max_10_vgprs:
 ; GFX900-DAG: s_mov_b32 s{{[0-9]+}}, SCRATCH_RSRC_DWORD0
@@ -57,23 +57,21 @@
 }
 
 ; GCN-LABEL: {{^}}max_10_vgprs_used_9a:
-; GCN-DAG:    s_mov_b32 s{{[0-9]+}}, SCRATCH_RSRC_DWORD0
-; GCN-DAG:    s_mov_b32 s{{[0-9]+}}, SCRATCH_RSRC_DWORD1
+; GFX908-DAG: s_mov_b32 s{{[0-9]+}}, SCRATCH_RSRC_DWORD0
+; GFX908-DAG: s_mov_b32 s{{[0-9]+}}, SCRATCH_RSRC_DWORD1
 ; GFX908:     v_accvgpr_write_b32 a9, v{{[0-9]}}
-; GCN:        buffer_store_dword v{{[0-9]}},
-; GFX900:     buffer_store_dword v{{[0-9]}},
-; GFX900:     buffer_load_dword v{{[0-9]}},
-; GFX900:     buffer_load_dword v{{[0-9]}},
+; GFX908:     buffer_store_dword v{{[0-9]}},
 ; GFX908-NOT: buffer_
 ; GFX908:     v_accvgpr_read_b32 v{{[0-9]}}, a9
 ; GFX908:     buffer_load_dword v{{[0-9]}},
 ; GFX908-NOT: buffer_
 
-; GCN:    NumVgprs: 10
-; GFX900: ScratchSize: 12
+; GFX900:     couldn't allocate input reg for constraint 'a'
+
+; GFX908: NumVgprs: 10
 ; GFX908: ScratchSize: 8
-; GCN:    VGPRBlocks: 2
-; GCN:    NumVGPRsForWavesPerEU: 10
+; GFX908: VGPRBlocks: 2
+; GFX908: NumVGPRsForWavesPerEU: 10
 define amdgpu_kernel void @max_10_vgprs_used_9a(i32 addrspace(1)* %p) #0 {
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   call void asm sideeffect "", "a,a,a,a,a,a,a,a,a"(i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9)