AMDGPU: Allow sinking of addressing modes for atomic_inc/dec

[• tstellarAMD]

LGTM.

[arsenm]

r297913

lib/Target/AMDGPU/SIISelLowering.h

class SITargetLowering final : public AMDGPUTargetLowering {
   /// global value \p GV, false otherwise.
   bool shouldEmitPCReloc(const GlobalValue *GV) const;
 
 public:
   SITargetLowering(const TargetMachine &tm, const SISubtarget &STI);
 
   const SISubtarget *getSubtarget() const;
 
+  bool isShuffleMaskLegal(const SmallVectorImpl<int> &/*Mask*/,
+                          EVT /*VT*/) const override;
+
   bool getTgtMemIntrinsic(IntrinsicInfo &, const CallInst &,
                           unsigned IntrinsicID) const override;
 
-  bool isShuffleMaskLegal(const SmallVectorImpl<int> &/*Mask*/,
-                          EVT /*VT*/) const override;
+  bool GetAddrModeArguments(IntrinsicInst * /*I*/,
+                            SmallVectorImpl<Value*> &/*Ops*/,
+                            Type *&/*AccessTy*/) const override;
 
   bool isLegalAddressingMode(const DataLayout &DL, const AddrMode &AM, Type *Ty,
                              unsigned AS) const override;
 
   bool allowsMisalignedMemoryAccesses(EVT VT, unsigned AS,
                                       unsigned Align,
                                       bool *IsFast) const override;
 

lib/Target/AMDGPU/SIISelLowering.cpp

 #include "llvm/IR/DebugLoc.h"
 #include "llvm/IR/DerivedTypes.h"
 #include "llvm/IR/DiagnosticInfo.h"
 #include "llvm/IR/Function.h"
 #include "llvm/IR/GlobalValue.h"
 #include "llvm/IR/InstrTypes.h"
 #include "llvm/IR/Instruction.h"
 #include "llvm/IR/Instructions.h"
+#include "llvm/IR/IntrinsicInst.h"
 #include "llvm/IR/Type.h"
 #include "llvm/Support/Casting.h"
 #include "llvm/Support/CodeGen.h"
 #include "llvm/Support/CommandLine.h"
 #include "llvm/Support/Compiler.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/MathExtras.h"
 #include "llvm/Target/TargetCallingConv.h"

 const SISubtarget *SITargetLowering::getSubtarget() const {
   return static_cast<const SISubtarget *>(Subtarget);
 }
 
 //===----------------------------------------------------------------------===//
 // TargetLowering queries
 //===----------------------------------------------------------------------===//
 
+bool SITargetLowering::isShuffleMaskLegal(const SmallVectorImpl<int> &,
+                                          EVT) const {
+  // SI has some legal vector types, but no legal vector operations. Say no
+  // shuffles are legal in order to prefer scalarizing some vector operations.
+  return false;
+}
+
 bool SITargetLowering::getTgtMemIntrinsic(IntrinsicInfo &Info,
                                           const CallInst &CI,
                                           unsigned IntrID) const {
   switch (IntrID) {
   case Intrinsic::amdgcn_atomic_inc:
   case Intrinsic::amdgcn_atomic_dec:
     Info.opc = ISD::INTRINSIC_W_CHAIN;
     Info.memVT = MVT::getVT(CI.getType());
     Info.ptrVal = CI.getOperand(0);
     Info.align = 0;
     Info.vol = false;
     Info.readMem = true;
     Info.writeMem = true;
     return true;
   default:
     return false;
   }
 }
 
-bool SITargetLowering::isShuffleMaskLegal(const SmallVectorImpl<int> &,
-                                          EVT) const {
-  // SI has some legal vector types, but no legal vector operations. Say no
-  // shuffles are legal in order to prefer scalarizing some vector operations.
+bool SITargetLowering::GetAddrModeArguments(IntrinsicInst *II,
+                                            SmallVectorImpl<Value*> &Ops,
+                                            Type *&AccessTy) const {
+  switch (II->getIntrinsicID()) {
+  case Intrinsic::amdgcn_atomic_inc:
+  case Intrinsic::amdgcn_atomic_dec: {
+    Value *Ptr = II->getArgOperand(0);
+    AccessTy = II->getType();
+    Ops.push_back(Ptr);
+    return true;
+  }
+  default:
     return false;
   }
+}
 
 bool SITargetLowering::isLegalFlatAddressingMode(const AddrMode &AM) const {
   // Flat instructions do not have offsets, and only have the register
   // address.
   return AM.BaseOffs == 0 && (AM.Scale == 0 || AM.Scale == 1);
 }
 
 bool SITargetLowering::isLegalMUBUFAddressingMode(const AddrMode &AM) const {

test/CodeGen/AMDGPU/cgp-addressing-modes.ll

endif:
   %x = phi i32 addrspace(3)* [ %tmp1, %if ], [ null, %entry ]
   store i32 addrspace(3)* %x, i32 addrspace(3)* addrspace(3)* %out.gep
   br label %done
 
 done:
   ret void
 }
 
+; OPT-LABEL: @test_sink_local_small_offset_atomic_inc_i32(
+; OPT: %sunkaddr = ptrtoint i32 addrspace(3)* %in to i32
+; OPT: %sunkaddr1 = add i32 %sunkaddr, 28
+; OPT: %sunkaddr2 = inttoptr i32 %sunkaddr1 to i32 addrspace(3)*
+; OPT: %tmp1 = call i32 @llvm.amdgcn.atomic.inc.i32.p3i32(i32 addrspace(3)* %sunkaddr2, i32 2)
+define void @test_sink_local_small_offset_atomic_inc_i32(i32 addrspace(3)* %out, i32 addrspace(3)* %in) {
+entry:
+  %out.gep = getelementptr i32, i32 addrspace(3)* %out, i32 999999
+  %in.gep = getelementptr i32, i32 addrspace(3)* %in, i32 7
+  %tid = call i32 @llvm.amdgcn.mbcnt.lo(i32 -1, i32 0) #0
+  %tmp0 = icmp eq i32 %tid, 0
+  br i1 %tmp0, label %endif, label %if
+
+if:
+  %tmp1 = call i32 @llvm.amdgcn.atomic.inc.i32.p3i32(i32 addrspace(3)* %in.gep, i32 2)
+  br label %endif
+
+endif:
+  %x = phi i32 [ %tmp1, %if ], [ 0, %entry ]
+  store i32 %x, i32 addrspace(3)* %out.gep
+  br label %done
+
+done:
+  ret void
+}
+
+; OPT-LABEL: @test_sink_local_small_offset_atomic_dec_i32(
+; OPT: %sunkaddr = ptrtoint i32 addrspace(3)* %in to i32
+; OPT: %sunkaddr1 = add i32 %sunkaddr, 28
+; OPT: %sunkaddr2 = inttoptr i32 %sunkaddr1 to i32 addrspace(3)*
+; OPT: %tmp1 = call i32 @llvm.amdgcn.atomic.dec.i32.p3i32(i32 addrspace(3)* %sunkaddr2, i32 2)
+define void @test_sink_local_small_offset_atomic_dec_i32(i32 addrspace(3)* %out, i32 addrspace(3)* %in) {
+entry:
+  %out.gep = getelementptr i32, i32 addrspace(3)* %out, i32 999999
+  %in.gep = getelementptr i32, i32 addrspace(3)* %in, i32 7
+  %tid = call i32 @llvm.amdgcn.mbcnt.lo(i32 -1, i32 0) #0
+  %tmp0 = icmp eq i32 %tid, 0
+  br i1 %tmp0, label %endif, label %if
+
+if:
+  %tmp1 = call i32 @llvm.amdgcn.atomic.dec.i32.p3i32(i32 addrspace(3)* %in.gep, i32 2)
+  br label %endif
+
+endif:
+  %x = phi i32 [ %tmp1, %if ], [ 0, %entry ]
+  store i32 %x, i32 addrspace(3)* %out.gep
+  br label %done
+
+done:
+  ret void
+}
+
 declare i32 @llvm.amdgcn.mbcnt.lo(i32, i32) #0
+declare i32 @llvm.amdgcn.atomic.inc.i32.p3i32(i32 addrspace(3)* nocapture, i32) #2
+declare i32 @llvm.amdgcn.atomic.dec.i32.p3i32(i32 addrspace(3)* nocapture, i32) #2
 
 attributes #0 = { nounwind readnone }
 attributes #1 = { nounwind }
+attributes #2 = { nounwind argmemonly }