AMDGPU/SI: Use flat for global load/store when targeting HSA

Summary:
  For some reason doing executing an MUBUF instruction with the addr64
  bit set and a zero base pointer in the resource descriptor causes
  the memory operation to be dropped when the shader is executed using
  the HSA runtime.

  This kind of MUBUF instruction is commonly used when the pointer is
  stored in VGPRs.  The base pointer field in the resource descriptor
  is set to zero and and the pointer is stored in the vaddr field.

  This patch resolves the issue by only using flat instructions for
  global memory operations when targeting HSA. This is an overly
  conservative fix as all other configurations of MUBUF instructions
  appear to work.

  NOTE: re-commit by fixing a failure in Codegen/AMDGPU/llvm.dbg.value.ll

Reviewers: tstellarAMD

Subscribers: arsenm, llvm-commits

Differential Revision: http://reviews.llvm.org/D15543

llvm-svn: 256282

13 files changed, 131 insertions, 82 deletions

diff --git a/llvm/lib/Target/AMDGPU/AMDGPU.td b/llvm/lib/Target/AMDGPU/AMDGPU.td
index 68b50504ee4..5913bf7f773 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPU.td
+++ b/llvm/lib/Target/AMDGPU/AMDGPU.td
@@ -108,6 +108,11 @@ def FeatureEnableUnsafeDSOffsetFolding : SubtargetFeature <"unsafe-ds-offset-fol
         "true",
         "Force using DS instruction immediate offsets on SI">;
 
+def FeatureFlatForGlobal : SubtargetFeature<"flat-for-global",
+        "FlatForGlobal",
+        "true",
+        "Force to generate flat instruction for global">;
+
 def FeatureFlatAddressSpace : SubtargetFeature<"flat-address-space",
         "FlatAddressSpace",
         "true",
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp b/llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp
index ea7c6429b7d..b33040b4d06 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp
@@ -95,7 +95,7 @@ private:
   bool SelectDS1Addr1Offset(SDValue Ptr, SDValue &Base, SDValue &Offset) const;
   bool SelectDS64Bit4ByteAligned(SDValue Ptr, SDValue &Base, SDValue &Offset0,
                                  SDValue &Offset1) const;
-  void SelectMUBUF(SDValue Addr, SDValue &SRsrc, SDValue &VAddr,
+  bool SelectMUBUF(SDValue Addr, SDValue &SRsrc, SDValue &VAddr,
                    SDValue &SOffset, SDValue &Offset, SDValue &Offen,
                    SDValue &Idxen, SDValue &Addr64, SDValue &GLC, SDValue &SLC,
                    SDValue &TFE) const;
@@ -920,12 +920,16 @@ static bool isLegalMUBUFImmOffset(const ConstantSDNode *Imm) {
   return isUInt<12>(Imm->getZExtValue());
 }
 
-void AMDGPUDAGToDAGISel::SelectMUBUF(SDValue Addr, SDValue &Ptr,
+bool AMDGPUDAGToDAGISel::SelectMUBUF(SDValue Addr, SDValue &Ptr,
                                      SDValue &VAddr, SDValue &SOffset,
                                      SDValue &Offset, SDValue &Offen,
                                      SDValue &Idxen, SDValue &Addr64,
                                      SDValue &GLC, SDValue &SLC,
                                      SDValue &TFE) const {
+  // Subtarget prefers to use flat instruction
+  if (Subtarget->useFlatForGlobal())
+    return false;
+
   SDLoc DL(Addr);
 
   GLC = CurDAG->getTargetConstant(0, DL, MVT::i1);
@@ -958,14 +962,14 @@ void AMDGPUDAGToDAGISel::SelectMUBUF(SDValue Addr, SDValue &Ptr,
 
     if (isLegalMUBUFImmOffset(C1)) {
         Offset = CurDAG->getTargetConstant(C1->getZExtValue(), DL, MVT::i16);
-        return;
+        return true;
     } else if (isUInt<32>(C1->getZExtValue())) {
       // Illegal offset, store it in soffset.
       Offset = CurDAG->getTargetConstant(0, DL, MVT::i16);
       SOffset = SDValue(CurDAG->getMachineNode(AMDGPU::S_MOV_B32, DL, MVT::i32,
                    CurDAG->getTargetConstant(C1->getZExtValue(), DL, MVT::i32)),
                         0);
-      return;
+      return true;
     }
   }
 
@@ -977,13 +981,15 @@ void AMDGPUDAGToDAGISel::SelectMUBUF(SDValue Addr, SDValue &Ptr,
     Ptr = N0;
     VAddr = N1;
     Offset = CurDAG->getTargetConstant(0, DL, MVT::i16);
-    return;
+    return true;
   }
 
   // default case -> offset
   VAddr = CurDAG->getTargetConstant(0, DL, MVT::i32);
   Ptr = Addr;
   Offset = CurDAG->getTargetConstant(0, DL, MVT::i16);
+
+  return true;
 }
 
 bool AMDGPUDAGToDAGISel::SelectMUBUFAddr64(SDValue Addr, SDValue &SRsrc,
@@ -996,8 +1002,9 @@ bool AMDGPUDAGToDAGISel::SelectMUBUFAddr64(SDValue Addr, SDValue &SRsrc,
   if (Subtarget->getGeneration() >= AMDGPUSubtarget::VOLCANIC_ISLANDS)
     return false;
 
-  SelectMUBUF(Addr, Ptr, VAddr, SOffset, Offset, Offen, Idxen, Addr64,
-              GLC, SLC, TFE);
+  if (!SelectMUBUF(Addr, Ptr, VAddr, SOffset, Offset, Offen, Idxen, Addr64,
+              GLC, SLC, TFE))
+    return false;
 
   ConstantSDNode *C = cast<ConstantSDNode>(Addr64);
   if (C->getSExtValue()) {
@@ -1063,8 +1070,9 @@ bool AMDGPUDAGToDAGISel::SelectMUBUFOffset(SDValue Addr, SDValue &SRsrc,
   const SIInstrInfo *TII =
     static_cast<const SIInstrInfo *>(Subtarget->getInstrInfo());
 
-  SelectMUBUF(Addr, Ptr, VAddr, SOffset, Offset, Offen, Idxen, Addr64,
-              GLC, SLC, TFE);
+  if (!SelectMUBUF(Addr, Ptr, VAddr, SOffset, Offset, Offen, Idxen, Addr64,
+              GLC, SLC, TFE))
+    return false;
 
   if (!cast<ConstantSDNode>(Offen)->getSExtValue() &&
       !cast<ConstantSDNode>(Idxen)->getSExtValue() &&
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUSubtarget.cpp b/llvm/lib/Target/AMDGPU/AMDGPUSubtarget.cpp
index 0aee09bab59..44e0c47877a 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUSubtarget.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUSubtarget.cpp
@@ -45,6 +45,8 @@ AMDGPUSubtarget::initializeSubtargetDependencies(const Triple &TT,
   // disable it.
 
   SmallString<256> FullFS("+promote-alloca,+fp64-denormals,");
+  if (isAmdHsaOS()) // Turn on FlatForGlobal for HSA.
+    FullFS += "+flat-for-global,";
   FullFS += FS;
 
   if (GPU == "" && TT.getArch() == Triple::amdgcn)
@@ -68,9 +70,9 @@ AMDGPUSubtarget::AMDGPUSubtarget(const Triple &TT, StringRef GPU, StringRef FS,
       DumpCode(false), R600ALUInst(false), HasVertexCache(false),
       TexVTXClauseSize(0), Gen(AMDGPUSubtarget::R600), FP64(false),
       FP64Denormals(false), FP32Denormals(false), FastFMAF32(false),
-      CaymanISA(false), FlatAddressSpace(false), EnableIRStructurizer(true),
-      EnablePromoteAlloca(false), EnableIfCvt(true), EnableLoadStoreOpt(false),
-      EnableUnsafeDSOffsetFolding(false),
+      CaymanISA(false), FlatAddressSpace(false), FlatForGlobal(false),
+      EnableIRStructurizer(true), EnablePromoteAlloca(false), EnableIfCvt(true),
+      EnableLoadStoreOpt(false), EnableUnsafeDSOffsetFolding(false),
       WavefrontSize(0), CFALUBug(false), LocalMemorySize(0),
       EnableVGPRSpilling(false), SGPRInitBug(false), IsGCN(false),
       GCN1Encoding(false), GCN3Encoding(false), CIInsts(false), LDSBankCount(0),
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUSubtarget.h b/llvm/lib/Target/AMDGPU/AMDGPUSubtarget.h
index 971b5179b13..9c7bb88f8f4 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUSubtarget.h
+++ b/llvm/lib/Target/AMDGPU/AMDGPUSubtarget.h
@@ -70,6 +70,7 @@ private:
   bool FastFMAF32;
   bool CaymanISA;
   bool FlatAddressSpace;
+  bool FlatForGlobal;
   bool EnableIRStructurizer;
   bool EnablePromoteAlloca;
   bool EnableIfCvt;
@@ -159,6 +160,10 @@ public:
     return FlatAddressSpace;
   }
 
+  bool useFlatForGlobal() const {
+    return FlatForGlobal;
+  }
+
   bool hasBFE() const {
     return (getGeneration() >= EVERGREEN);
   }
diff --git a/llvm/lib/Target/AMDGPU/CIInstructions.td b/llvm/lib/Target/AMDGPU/CIInstructions.td
index 7b8cb125dad..afbd155fee5 100644
--- a/llvm/lib/Target/AMDGPU/CIInstructions.td
+++ b/llvm/lib/Target/AMDGPU/CIInstructions.td
@@ -234,3 +234,63 @@ def : Pat <
 >;
 
 } // End Predicates = [isCI]
+
+
+//===----------------------------------------------------------------------===//
+// Patterns to generate flat for global
+//===----------------------------------------------------------------------===//
+
+def useFlatForGlobal : Predicate <
+  "Subtarget->useFlatForGlobal() || "
+  "Subtarget->getGeneration() >= AMDGPUSubtarget::VOLCANIC_ISLANDS">;
+
+let Predicates = [useFlatForGlobal] in {
+
+// 1. Offset as 20bit DWORD immediate
+def : Pat <
+  (SIload_constant v4i32:$sbase, IMM20bit:$offset),
+  (S_BUFFER_LOAD_DWORD_IMM $sbase, (as_i32imm $offset))
+>;
+
+// Patterns for global loads with no offset
+class FlatLoadPat <FLAT inst, SDPatternOperator node, ValueType vt> : Pat <
+  (vt (node i64:$addr)),
+  (inst $addr, 0, 0, 0)
+>;
+
+def : FlatLoadPat <FLAT_LOAD_UBYTE, az_extloadi8_global, i32>;
+def : FlatLoadPat <FLAT_LOAD_SBYTE, sextloadi8_global, i32>;
+def : FlatLoadPat <FLAT_LOAD_USHORT, az_extloadi16_global, i32>;
+def : FlatLoadPat <FLAT_LOAD_SSHORT, sextloadi16_global, i32>;
+def : FlatLoadPat <FLAT_LOAD_DWORD, global_load, i32>;
+def : FlatLoadPat <FLAT_LOAD_DWORDX2, global_load, v2i32>;
+def : FlatLoadPat <FLAT_LOAD_DWORDX4, global_load, v4i32>;
+
+class FlatStorePat <FLAT inst, SDPatternOperator node, ValueType vt> : Pat <
+  (node vt:$data, i64:$addr),
+  (inst $data, $addr, 0, 0, 0)
+>;
+
+def : FlatStorePat <FLAT_STORE_BYTE, truncstorei8_global, i32>;
+def : FlatStorePat <FLAT_STORE_SHORT, truncstorei16_global, i32>;
+def : FlatStorePat <FLAT_STORE_DWORD, global_store, i32>;
+def : FlatStorePat <FLAT_STORE_DWORDX2, global_store, v2i32>;
+def : FlatStorePat <FLAT_STORE_DWORDX4, global_store, v4i32>;
+
+class FlatAtomicPat <FLAT inst, SDPatternOperator node, ValueType vt> : Pat <
+  (vt (node i64:$addr, vt:$data)),
+  (inst $addr, $data, 0, 0)
+>;
+
+def : FlatAtomicPat <FLAT_ATOMIC_ADD_RTN, atomic_add_global, i32>;
+def : FlatAtomicPat <FLAT_ATOMIC_AND_RTN, atomic_and_global, i32>;
+def : FlatAtomicPat <FLAT_ATOMIC_SUB_RTN, atomic_sub_global, i32>;
+def : FlatAtomicPat <FLAT_ATOMIC_SMAX_RTN, atomic_max_global, i32>;
+def : FlatAtomicPat <FLAT_ATOMIC_UMAX_RTN, atomic_umax_global, i32>;
+def : FlatAtomicPat <FLAT_ATOMIC_SMIN_RTN, atomic_min_global, i32>;
+def : FlatAtomicPat <FLAT_ATOMIC_UMIN_RTN, atomic_umin_global, i32>;
+def : FlatAtomicPat <FLAT_ATOMIC_OR_RTN, atomic_or_global, i32>;
+def : FlatAtomicPat <FLAT_ATOMIC_SWAP_RTN, atomic_swap_global, i32>;
+def : FlatAtomicPat <FLAT_ATOMIC_XOR_RTN, atomic_xor_global, i32>;
+
+} // End Predicates = [useFlatForGlobal]
diff --git a/llvm/lib/Target/AMDGPU/VIInstructions.td b/llvm/lib/Target/AMDGPU/VIInstructions.td
index 9d29e4700f1..20a026a822e 100644
--- a/llvm/lib/Target/AMDGPU/VIInstructions.td
+++ b/llvm/lib/Target/AMDGPU/VIInstructions.td
@@ -101,58 +101,3 @@ def S_DCACHE_WB_VOL : SMEM_Inval <0x23,
 
 } // End SIAssemblerPredicate = DisableInst, SubtargetPredicate = isVI
 
-//===----------------------------------------------------------------------===//
-// SMEM Patterns
-//===----------------------------------------------------------------------===//
-
-let Predicates = [isVI] in {
-
-// 1. Offset as 20bit DWORD immediate
-def : Pat <
-  (SIload_constant v4i32:$sbase, IMM20bit:$offset),
-  (S_BUFFER_LOAD_DWORD_IMM $sbase, (as_i32imm $offset))
->;
-
-// Patterns for global loads with no offset
-class FlatLoadPat <FLAT inst, SDPatternOperator node, ValueType vt> : Pat <
-  (vt (node i64:$addr)),
-  (inst $addr, 0, 0, 0)
->;
-
-def : FlatLoadPat <FLAT_LOAD_UBYTE, az_extloadi8_global, i32>;
-def : FlatLoadPat <FLAT_LOAD_SBYTE, sextloadi8_global, i32>;
-def : FlatLoadPat <FLAT_LOAD_USHORT, az_extloadi16_global, i32>;
-def : FlatLoadPat <FLAT_LOAD_SSHORT, sextloadi16_global, i32>;
-def : FlatLoadPat <FLAT_LOAD_DWORD, global_load, i32>;
-def : FlatLoadPat <FLAT_LOAD_DWORDX2, global_load, v2i32>;
-def : FlatLoadPat <FLAT_LOAD_DWORDX4, global_load, v4i32>;
-
-class FlatStorePat <FLAT inst, SDPatternOperator node, ValueType vt> : Pat <
-  (node vt:$data, i64:$addr),
-  (inst $data, $addr, 0, 0, 0)
->;
-
-def : FlatStorePat <FLAT_STORE_BYTE, truncstorei8_global, i32>;
-def : FlatStorePat <FLAT_STORE_SHORT, truncstorei16_global, i32>;
-def : FlatStorePat <FLAT_STORE_DWORD, global_store, i32>;
-def : FlatStorePat <FLAT_STORE_DWORDX2, global_store, v2i32>;
-def : FlatStorePat <FLAT_STORE_DWORDX4, global_store, v4i32>;
-
-class FlatAtomicPat <FLAT inst, SDPatternOperator node, ValueType vt> : Pat <
-  (vt (node i64:$addr, vt:$data)),
-  (inst $addr, $data, 0, 0)
->;
-
-def : FlatAtomicPat <FLAT_ATOMIC_ADD_RTN, atomic_add_global, i32>;
-def : FlatAtomicPat <FLAT_ATOMIC_AND_RTN, atomic_and_global, i32>;
-def : FlatAtomicPat <FLAT_ATOMIC_SUB_RTN, atomic_sub_global, i32>;
-def : FlatAtomicPat <FLAT_ATOMIC_SMAX_RTN, atomic_max_global, i32>;
-def : FlatAtomicPat <FLAT_ATOMIC_UMAX_RTN, atomic_umax_global, i32>;
-def : FlatAtomicPat <FLAT_ATOMIC_SMIN_RTN, atomic_min_global, i32>;
-def : FlatAtomicPat <FLAT_ATOMIC_UMIN_RTN, atomic_umin_global, i32>;
-def : FlatAtomicPat <FLAT_ATOMIC_OR_RTN, atomic_or_global, i32>;
-def : FlatAtomicPat <FLAT_ATOMIC_SWAP_RTN, atomic_swap_global, i32>;
-def : FlatAtomicPat <FLAT_ATOMIC_XOR_RTN, atomic_xor_global, i32>;
-
-
-} // End Predicates = [isVI]
diff --git a/llvm/test/CodeGen/AMDGPU/ci-use-flat-for-global.ll b/llvm/test/CodeGen/AMDGPU/ci-use-flat-for-global.ll
new file mode 100644
index 00000000000..1a37e3c75fa
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/ci-use-flat-for-global.ll
@@ -0,0 +1,15 @@
+; RUN: llc < %s -mtriple=amdgcn--amdhsa -mcpu=kaveri | FileCheck -check-prefix=HSA-DEFAULT %s
+; RUN: llc < %s -mtriple=amdgcn--amdhsa -mcpu=kaveri -mattr=-flat-for-global | FileCheck -check-prefix=HSA-NODEFAULT %s
+; RUN: llc < %s -mtriple=amdgcn -mcpu=kaveri | FileCheck -check-prefix=NOHSA-DEFAULT %s
+; RUN: llc < %s -mtriple=amdgcn -mcpu=kaveri -mattr=+flat-for-global | FileCheck -check-prefix=NOHSA-NODEFAULT %s
+
+
+; HSA-DEFAULT: flat_store_dword
+; HSA-NODEFAULT: buffer_store_dword
+; NOHSA-DEFAULT: buffer_store_dword
+; NOHSA-NODEFAULT: flat_store_dword
+define void @test(i32 addrspace(1)* %out) {
+entry:
+  store i32 0, i32 addrspace(1)* %out
+  ret void
+}
diff --git a/llvm/test/CodeGen/AMDGPU/hsa.ll b/llvm/test/CodeGen/AMDGPU/hsa.ll
index d9bb586163d..abc89b7fd83 100644
--- a/llvm/test/CodeGen/AMDGPU/hsa.ll
+++ b/llvm/test/CodeGen/AMDGPU/hsa.ll
@@ -1,6 +1,8 @@
-; RUN: llc < %s -mtriple=amdgcn--amdhsa -mcpu=kaveri | FileCheck --check-prefix=HSA-CI --check-prefix=HSA %s
-; RUN: llc < %s -mtriple=amdgcn--amdhsa -mcpu=carrizo | FileCheck --check-prefix=HSA-VI --check-prefix=HSA %s
-; RUN: llc < %s -mtriple=amdgcn--amdhsa -mcpu=kaveri -filetype=obj  | llvm-readobj -symbols -s -sd | FileCheck --check-prefix=ELF %s
+; RUN: llc < %s -mtriple=amdgcn--amdhsa -mcpu=kaveri | FileCheck --check-prefix=HSA %s
+; RUN: llc < %s -mtriple=amdgcn--amdhsa -mcpu=kaveri -mattr=-flat-for-global | FileCheck --check-prefix=HSA-CI %s
+; RUN: llc < %s -mtriple=amdgcn--amdhsa -mcpu=carrizo  | FileCheck --check-prefix=HSA %s
+; RUN: llc < %s -mtriple=amdgcn--amdhsa -mcpu=carrizo -mattr=-flat-for-global | FileCheck --check-prefix=HSA-VI %s
+; RUN: llc < %s -mtriple=amdgcn--amdhsa -mcpu=kaveri -filetype=obj | llvm-readobj -symbols -s -sd | FileCheck --check-prefix=ELF %s
 ; RUN: llc < %s -mtriple=amdgcn--amdhsa -mcpu=kaveri | llvm-mc -filetype=obj -triple amdgcn--amdhsa -mcpu=kaveri | llvm-readobj -symbols -s -sd | FileCheck %s --check-prefix=ELF
 
 ; The SHT_NOTE section contains the output from the .hsa_code_object_*
@@ -47,7 +49,8 @@
 ; HSA-CI: s_mov_b32 s[[HI:[0-9]]], 0x100f000
 ; On VI+ we also need to set MTYPE = 2
 ; HSA-VI: s_mov_b32 s[[HI:[0-9]]], 0x1100f000
-; HSA: buffer_store_dword v{{[0-9]+}}, s[0:[[HI]]], 0
+; Make sure we generate flat store for HSA
+; HSA: flat_store_dword v{{[0-9]+}}
 
 define void @simple(i32 addrspace(1)* %out) {
 entry:
diff --git a/llvm/test/CodeGen/AMDGPU/large-alloca-compute.ll b/llvm/test/CodeGen/AMDGPU/large-alloca-compute.ll
index de802c49ed4..8347b8c96ec 100644
--- a/llvm/test/CodeGen/AMDGPU/large-alloca-compute.ll
+++ b/llvm/test/CodeGen/AMDGPU/large-alloca-compute.ll
@@ -1,7 +1,7 @@
 ; RUN: llc -march=amdgcn -mcpu=bonaire < %s | FileCheck -check-prefix=GCN -check-prefix=CI -check-prefix=ALL %s
 ; RUN: llc -march=amdgcn -mcpu=tonga < %s | FileCheck -check-prefix=GCN -check-prefix=VI -check-prefix=ALL %s
-; RUN: llc -march=amdgcn -mcpu=bonaire -mtriple=amdgcn-unknown-amdhsa < %s | FileCheck -check-prefix=GCNHSA -check-prefix=CIHSA -check-prefix=ALL %s
-; RUN: llc -march=amdgcn -mcpu=tonga -mtriple=amdgcn-unknown-amdhsa < %s | FileCheck -check-prefix=GCNHSA -check-prefix=VIHSA -check-prefix=ALL %s
+; RUN: llc -march=amdgcn -mcpu=bonaire -mtriple=amdgcn-unknown-amdhsa < %s -mattr=-flat-for-global | FileCheck -check-prefix=GCNHSA -check-prefix=CIHSA -check-prefix=ALL %s
+; RUN: llc -march=amdgcn -mcpu=tonga -mtriple=amdgcn-unknown-amdhsa -mattr=-flat-for-global < %s | FileCheck -check-prefix=GCNHSA -check-prefix=VIHSA -check-prefix=ALL %s
 
 ; FIXME: align on alloca seems to be ignored for private_segment_alignment
 
diff --git a/llvm/test/CodeGen/AMDGPU/llvm.dbg.value.ll b/llvm/test/CodeGen/AMDGPU/llvm.dbg.value.ll
index cc109327d92..b01f8ab2bdf 100644
--- a/llvm/test/CodeGen/AMDGPU/llvm.dbg.value.ll
+++ b/llvm/test/CodeGen/AMDGPU/llvm.dbg.value.ll
@@ -1,4 +1,4 @@
-; RUN: llc -O0 -march=amdgcn -mtriple=amdgcn-unknown-amdhsa -verify-machineinstrs < %s | FileCheck %s
+; RUN: llc -O0 -march=amdgcn -mtriple=amdgcn-unknown-amdhsa -verify-machineinstrs -mattr=-flat-for-global < %s | FileCheck %s
 
 ; CHECK-LABEL: {{^}}test_debug_value:
 ; CHECK: s_load_dwordx2 s[4:5]
diff --git a/llvm/test/CodeGen/AMDGPU/move-addr64-rsrc-dead-subreg-writes.ll b/llvm/test/CodeGen/AMDGPU/move-addr64-rsrc-dead-subreg-writes.ll
index f4fa6211210..e9f641b736d 100644
--- a/llvm/test/CodeGen/AMDGPU/move-addr64-rsrc-dead-subreg-writes.ll
+++ b/llvm/test/CodeGen/AMDGPU/move-addr64-rsrc-dead-subreg-writes.ll
@@ -1,4 +1,4 @@
-; RUN: llc -march=amdgcn -mcpu=kaveri -mtriple=amdgcn-unknown-amdhsa < %s | FileCheck -check-prefix=GCN %s
+; RUN: llc -march=amdgcn -mcpu=kaveri -mtriple=amdgcn-unknown-amdhsa -mattr=-flat-for-global < %s | FileCheck -check-prefix=GCN %s
 
 ; Check that when mubuf addr64 instruction is handled in moveToVALU
 ; from the pointer, dead register writes are not emitted.
diff --git a/llvm/test/CodeGen/AMDGPU/register-count-comments.ll b/llvm/test/CodeGen/AMDGPU/register-count-comments.ll
index 8c104d9e34b..4bb315049be 100644
--- a/llvm/test/CodeGen/AMDGPU/register-count-comments.ll
+++ b/llvm/test/CodeGen/AMDGPU/register-count-comments.ll
@@ -1,5 +1,5 @@
 ; RUN: llc -march=amdgcn -verify-machineinstrs -asm-verbose < %s | FileCheck -check-prefix=SI %s
-; RUN: llc -march=amdgcn -mtriple=amdgcn-unknown-amdhsa -verify-machineinstrs -asm-verbose < %s | FileCheck -check-prefix=SI %s
+; RUN: llc -march=amdgcn -mtriple=amdgcn-unknown-amdhsa -verify-machineinstrs -asm-verbose -mattr=-flat-for-global < %s | FileCheck -check-prefix=SI %s
 
 declare i32 @llvm.SI.tid() nounwind readnone
 
diff --git a/llvm/test/CodeGen/AMDGPU/work-item-intrinsics.ll b/llvm/test/CodeGen/AMDGPU/work-item-intrinsics.ll
index f420ec9c7d2..e7fcd1ff365 100644
--- a/llvm/test/CodeGen/AMDGPU/work-item-intrinsics.ll
+++ b/llvm/test/CodeGen/AMDGPU/work-item-intrinsics.ll
@@ -129,7 +129,8 @@ entry:
 
 ; GCN-NOHSA: v_mov_b32_e32 [[VVAL:v[0-9]+]], s2{{$}}
 ; HSA: v_mov_b32_e32 [[VVAL:v[0-9]+]], s6{{$}}
-; GCN: buffer_store_dword [[VVAL]]
+; GCN-NOHSA: buffer_store_dword [[VVAL]]
+; HSA: flat_store_dword [[VVAL]]
 
 ; HSA: COMPUTE_PGM_RSRC2:USER_SGPR: 6
 ; GCN-NOHSA: COMPUTE_PGM_RSRC2:USER_SGPR: 2
@@ -155,7 +156,8 @@ entry:
 ; HSA: enable_sgpr_grid_workgroup_count_z = 0
 ; GCN-NOHSA: v_mov_b32_e32 [[VVAL:v[0-9]+]], s3
 ; GCN-HSA: v_mov_b32_e32 [[VVAL:v[0-9]+]], s7
-; GCN: buffer_store_dword [[VVAL]]
+; GCN-NOHSA: buffer_store_dword [[VVAL]]
+; HSA: flat_store_dword [[VVAL]]
 
 ; HSA: COMPUTE_PGM_RSRC2:USER_SGPR: 6
 ; GCN-NOHSA: COMPUTE_PGM_RSRC2:USER_SGPR: 2
@@ -190,7 +192,8 @@ entry:
 
 ; GCN-NOHSA: v_mov_b32_e32 [[VVAL:v[0-9]+]], s3{{$}}
 ; HSA: v_mov_b32_e32 [[VVAL:v[0-9]+]], s7{{$}}
-; GCN: buffer_store_dword [[VVAL]]
+; GCN-NOHSA: buffer_store_dword [[VVAL]]
+; HSA: flat_store_dword [[VVAL]]
 
 ; HSA: COMPUTE_PGM_RSRC2:USER_SGPR: 6
 ; GCN-NOHSA: COMPUTE_PGM_RSRC2:USER_SGPR: 2
@@ -211,7 +214,8 @@ entry:
 
 ; FUNC-LABEL: {{^}}tidig_x:
 ; HSA: compute_pgm_rsrc2_tidig_comp_cnt = 0
-; GCN: buffer_store_dword v0
+; GCN-NOHSA: buffer_store_dword v0
+; HSA: flat_store_dword v0
 define void @tidig_x(i32 addrspace(1)* %out) {
 entry:
   %0 = call i32 @llvm.r600.read.tidig.x() #0
@@ -226,7 +230,8 @@ entry:
 ; FUNC-LABEL: {{^}}tidig_y:
 
 ; HSA: compute_pgm_rsrc2_tidig_comp_cnt = 1
-; GCN: buffer_store_dword v1
+; GCN-NOHSA: buffer_store_dword v1
+; HSA: flat_store_dword v1
 define void @tidig_y(i32 addrspace(1)* %out) {
 entry:
   %0 = call i32 @llvm.r600.read.tidig.y() #0
@@ -240,7 +245,8 @@ entry:
 
 ; FUNC-LABEL: {{^}}tidig_z:
 ; HSA: compute_pgm_rsrc2_tidig_comp_cnt = 2
-; GCN: buffer_store_dword v2
+; GCN-NOHSA: buffer_store_dword v2
+; HSA: flat_store_dword v2
 define void @tidig_z(i32 addrspace(1)* %out) {
 entry:
   %0 = call i32 @llvm.r600.read.tidig.z() #0