AMDGPU: Make auto waitcnt before barrier a feature

[arsenm]

LGTM

lib/Target/AMDGPU/AMDGPU.td

 // instructions.
 
 def FeatureFlatForGlobal : SubtargetFeature<"flat-for-global",
   "FlatForGlobal",
   "true",
   "Force to generate flat instruction for global"
 >;
 
+def FeatureAutoWaitcntBeforeBarrier : SubtargetFeature <
+  "auto-waitcnt-before-barrier",
+  "AutoWaitcntBeforeBarrier",
+  "true",
+  "Hardware automatically inserts waitcnt before barrier"
+>;
+
 // Dummy feature used to disable assembler instructions.
 def FeatureDisable : SubtargetFeature<"",
   "FeatureDisable","true",
   "Dummy feature to disable assembler instructions"
 >;
 
 class SubtargetFeatureGeneration <string Value,
                                   list<SubtargetFeature> Implies> :

lib/Target/AMDGPU/AMDGPUSubtarget.h

protected:
   bool HalfRate64Ops;
 
   // Dynamially set bits that enable features.
   bool FP32Denormals;
   bool FP64FP16Denormals;
   bool FPExceptions;
   bool DX10Clamp;
   bool FlatForGlobal;
+  bool AutoWaitcntBeforeBarrier;
   bool UnalignedScratchAccess;
   bool UnalignedBufferAccess;
   bool HasApertureRegs;
   bool EnableXNACK;
   bool TrapHandler;
   bool DebuggerInsertNops;
   bool DebuggerReserveRegs;
   bool DebuggerEmitPrologue;
public:
   bool enableIEEEBit(const MachineFunction &MF) const {
     return AMDGPU::isCompute(MF.getFunction()->getCallingConv());
   }
 
   bool useFlatForGlobal() const {
     return FlatForGlobal;
   }
 
+  bool hasAutoWaitcntBeforeBarrier() const {
+    return AutoWaitcntBeforeBarrier;
+  }
+
   bool hasUnalignedBufferAccess() const {
     return UnalignedBufferAccess;
   }
 
   bool hasUnalignedScratchAccess() const {
     return UnalignedScratchAccess;
   }
 
public:
   unsigned getKernArgSegmentSize(const MachineFunction &MF, unsigned ExplictArgBytes) const;
 
   /// Return the maximum number of waves per SIMD for kernels using \p SGPRs SGPRs
   unsigned getOccupancyWithNumSGPRs(unsigned SGPRs) const;
 
   /// Return the maximum number of waves per SIMD for kernels using \p VGPRs VGPRs
   unsigned getOccupancyWithNumVGPRs(unsigned VGPRs) const;
 
-  /// \returns True if waitcnt instruction is needed before barrier instruction,
-  /// false otherwise.
-  bool needWaitcntBeforeBarrier() const {
-    return true;
-  }
-
   /// \returns true if the flat_scratch register should be initialized with the
   /// pointer to the wave's scratch memory rather than a size and offset.
   bool flatScratchIsPointer() const {
     return getGeneration() >= GFX9;
   }
 
   /// \returns SGPR allocation granularity supported by the subtarget.
   unsigned getSGPRAllocGranule() const {

lib/Target/AMDGPU/AMDGPUSubtarget.cpp

: AMDGPUGenSubtargetInfo(TT, GPU, FS),
     FastFMAF32(false),
     HalfRate64Ops(false),
 
     FP32Denormals(false),
     FP64FP16Denormals(false),
     FPExceptions(false),
     DX10Clamp(false),
     FlatForGlobal(false),
+    AutoWaitcntBeforeBarrier(false),
     UnalignedScratchAccess(false),
     UnalignedBufferAccess(false),
 
     HasApertureRegs(false),
     EnableXNACK(false),
     TrapHandler(false),
     DebuggerInsertNops(false),
     DebuggerReserveRegs(false),

lib/Target/AMDGPU/SIInsertWaitcnts.cpp

#endif
 
   // TODO: Tie force zero to a compiler triage option.
   bool ForceZero = false;
 
   // Check to see if this is an S_BARRIER, and if an implicit S_WAITCNT 0
   // occurs before the instruction. Doing it here prevents any additional
   // S_WAITCNTs from being emitted if the instruction was marked as
   // requiring a WAITCNT beforehand.
-  if (MI.getOpcode() == AMDGPU::S_BARRIER && ST->needWaitcntBeforeBarrier()) {
+  if (MI.getOpcode() == AMDGPU::S_BARRIER &&
+      !ST->hasAutoWaitcntBeforeBarrier()) {
     EmitSwaitcnt |=
         ScoreBrackets->updateByWait(VM_CNT, ScoreBrackets->getScoreUB(VM_CNT));
     EmitSwaitcnt |= ScoreBrackets->updateByWait(
         EXP_CNT, ScoreBrackets->getScoreUB(EXP_CNT));
     EmitSwaitcnt |= ScoreBrackets->updateByWait(
         LGKM_CNT, ScoreBrackets->getScoreUB(LGKM_CNT));
   }
 

lib/Target/AMDGPU/SIInsertWaits.cpp

for (MachineBasicBlock::iterator I = MBB.begin(), E = MBB.end();
       Counters Required;
 
       // Wait for everything before a barrier.
       //
       // S_SENDMSG implicitly waits for all outstanding LGKM transfers to finish,
       // but we also want to wait for any other outstanding transfers before
       // signalling other hardware blocks
       if ((I->getOpcode() == AMDGPU::S_BARRIER &&
-               ST->needWaitcntBeforeBarrier()) ||
+               !ST->hasAutoWaitcntBeforeBarrier()) ||
            I->getOpcode() == AMDGPU::S_SENDMSG ||
            I->getOpcode() == AMDGPU::S_SENDMSGHALT)
         Required = LastIssued;
       else
         Required = handleOperands(*I);
 
       Counters Increment = getHwCounts(*I);
 

test/CodeGen/AMDGPU/llvm.amdgcn.s.barrier.ll

-; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=GFX8 %s
-; RUN: llc -march=amdgcn -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=GFX9 %s
+; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=GFX8 -check-prefix=NOAUTO %s
+; RUN: llc -march=amdgcn -mattr=+auto-waitcnt-before-barrier -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=GFX8 -check-prefix=AUTO %s
+; RUN: llc -march=amdgcn -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=GFX9 -check-prefix=NOAUTO %s
+; RUN: llc -march=amdgcn -mcpu=gfx900 -mattr=+auto-waitcnt-before-barrier -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=GFX9 -check-prefix=AUTO %s
 
 ; GCN-LABEL: {{^}}test_barrier:
 ; GFX8: buffer_store_dword
 ; GFX9: flat_store_dword
-; GCN: s_waitcnt
+; NOAUTO: s_waitcnt
+; AUTO-NOT: s_waitcnt
 ; GCN: s_barrier
 define amdgpu_kernel void @test_barrier(i32 addrspace(1)* %out, i32 %size) #0 {
 entry:
   %tmp = call i32 @llvm.amdgcn.workitem.id.x()
   %tmp1 = getelementptr i32, i32 addrspace(1)* %out, i32 %tmp
   store i32 %tmp, i32 addrspace(1)* %tmp1
   call void @llvm.amdgcn.s.barrier()
   %tmp3 = sub i32 %size, 1