Added GroupMemoryBarrierWithGroupSync intrinsic for DXIL

Author: adam-yang

llvm/include/llvm/IR/IntrinsicsDirectX.td

@@ -88,4 +88,6 @@ def int_dx_wave_is_first_lane : DefaultAttrsIntrinsic<[llvm_i1_ty], [], [IntrCon
 def int_dx_sign : DefaultAttrsIntrinsic<[LLVMScalarOrSameVectorWidth<0, llvm_i32_ty>], [llvm_any_ty], [IntrNoMem]>;
 def int_dx_step : DefaultAttrsIntrinsic<[LLVMMatchType<0>], [llvm_anyfloat_ty, LLVMMatchType<0>], [IntrNoMem]>;
 def int_dx_radians : DefaultAttrsIntrinsic<[llvm_anyfloat_ty], [LLVMMatchType<0>], [IntrNoMem]>;
+
+def int_dx_groupMemoryBarrierWithGroupSync : DefaultAttrsIntrinsic<[], [], []>;
 }

llvm/lib/Target/DirectX/DXIL.td

@@ -277,6 +277,7 @@ def IsFeedback : DXILAttribute;
 def IsWave : DXILAttribute;
 def NeedsUniformInputs : DXILAttribute;
 def IsBarrier : DXILAttribute;
+def NoDuplicate : DXILAttribute;
 
 class Overloads<Version ver, list<DXILOpParamType> ols> {
   Version dxil_version = ver;
@@ -810,3 +811,11 @@ def WaveGetLaneIndex : DXILOp<111, waveGetLaneIndex> {
   let stages = [Stages<DXIL1_0, [all_stages]>];
   let attributes = [Attributes<DXIL1_0, [ReadNone]>];
 }
+
+def Barrier : DXILOp<80, barrier> {
+  let Doc = "inserts a memory barrier in the shader";
+  let arguments = [Int32Ty];
+  let result = VoidTy;
+  let stages = [Stages<DXIL1_0, [compute, library]>];
+  let attributes = [Attributes<DXIL1_0, [NoDuplicate]>];
+}

llvm/lib/Target/DirectX/DXILConstants.h

@@ -30,6 +30,13 @@ enum class OpParamType : unsigned {
 #include "DXILOperation.inc"
 };
 
+enum class BarrierMode : unsigned {
+  SyncThreadGroup = 0x00000001,
+  UAVFenceGlobal = 0x00000002,
+  UAVFenceThreadGroup = 0x00000004,
+  TGSMFence = 0x00000008,
+};
+
 } // namespace dxil
 } // namespace llvm
 

llvm/lib/Target/DirectX/DXILOpLowering.cpp

@@ -210,6 +210,34 @@ class OpLowerer {
     });
   }
 
+  [[nodiscard]] bool lowerBarrier(Function& F, Intrinsic::ID IntrId) {
+    IRBuilder<> &IRB = OpBuilder.getIRB();
+    return replaceFunction(F, [&](CallInst *CI) -> Error {
+      unsigned BarrierMode = 0;
+      switch (IntrId) {
+      default:
+        report_fatal_error("Unhandled barrier operation type.");
+        break;
+      case Intrinsic::dx_groupMemoryBarrierWithGroupSync:
+        BarrierMode = (unsigned)dxil::BarrierMode::TGSMFence | (unsigned)dxil::BarrierMode::SyncThreadGroup;
+        break;
+      }
+
+      std::array<Value *, 1> Args{IRB.getInt32(BarrierMode)};
+
+      IRB.SetInsertPoint(CI);
+      Expected<CallInst *> OpCall =
+          OpBuilder.tryCreateOp(OpCode::Barrier, Args, CI->getName());
+      if (Error E = OpCall.takeError())
+        return E;
+
+      CI->replaceAllUsesWith(OpCall.get());
+      CI->eraseFromParent();
+
+      return Error::success();
+    });
+  }
+
   [[nodiscard]] bool lowerToBindAndAnnotateHandle(Function &F) {
     IRBuilder<> &IRB = OpBuilder.getIRB();
 
@@ -477,6 +505,9 @@ class OpLowerer {
     HasErrors |= replaceFunctionWithOp(F, OpCode);                             \
     break;
 #include "DXILOperation.inc"
+      case Intrinsic::dx_groupMemoryBarrierWithGroupSync:
+        HasErrors |= lowerBarrier(F, ID);
+        break;
       case Intrinsic::dx_handle_fromBinding:
         HasErrors |= lowerHandleFromBinding(F);
         break;

llvm/test/CodeGen/DirectX/GroupMemoryBarrierWithGroupSync.ll

@@ -0,0 +1,8 @@
+; RUN: opt -S  -dxil-op-lower -mtriple=dxil-pc-shadermodel6.3-library < %s | FileCheck %s --check-prefix=CHECK
+
+define void @test_group_memory_barrier_with_group_sync() {
+entry:
+  ; CHECK: call void @dx.op.barrier(i32 80, i32 9)
+  call void @llvm.dx.groupMemoryBarrierWithGroupSync()
+  ret void
+}