Added another intermediate memory barrier intrinsic that maps directly to the barrier dxil op

Author: adam-yang

llvm/include/llvm/IR/IntrinsicsDirectX.td

@@ -108,5 +108,6 @@ def int_dx_discard : DefaultAttrsIntrinsic<[], [llvm_i1_ty], []>;
 def int_dx_firstbituhigh : DefaultAttrsIntrinsic<[LLVMScalarOrSameVectorWidth<0, llvm_i32_ty>], [llvm_anyint_ty], [IntrNoMem]>;
 def int_dx_firstbitshigh : DefaultAttrsIntrinsic<[LLVMScalarOrSameVectorWidth<0, llvm_i32_ty>], [llvm_anyint_ty], [IntrNoMem]>;
 
+def int_dx_memory_barrier : DefaultAttrsIntrinsic<[], [llvm_i32_ty], []>;
 def int_dx_group_memory_barrier_with_group_sync : DefaultAttrsIntrinsic<[], [], []>;
 }

llvm/lib/Target/DirectX/DXIL.td

@@ -916,7 +916,8 @@ def WaveAllBitCount : DXILOp<135, waveAllOp> {
 
 def Barrier : DXILOp<80, barrier> {
   let Doc = "inserts a memory barrier in the shader";
-  let arguments = [];
+  let LLVMIntrinsic = int_dx_memory_barrier;
+  let arguments = [Int32Ty];
   let result = VoidTy;
   let stages = [Stages<DXIL1_0, [compute, library]>];
   let attributes = [Attributes<DXIL1_0, []>];

llvm/lib/Target/DirectX/DXILIntrinsicExpansion.cpp

@@ -10,6 +10,7 @@
 //  opcodes in DirectX Intermediate Language (DXIL).
 //===----------------------------------------------------------------------===//
 
+#include "DXILConstants.h"
 #include "DXILIntrinsicExpansion.h"
 #include "DirectX.h"
 #include "llvm/ADT/STLExtras.h"
@@ -67,6 +68,7 @@ static bool isIntrinsicExpansion(Function &F) {
   case Intrinsic::dx_sign:
   case Intrinsic::dx_step:
   case Intrinsic::dx_radians:
+  case Intrinsic::dx_group_memory_barrier_with_group_sync:
     return true;
   }
   return false;
@@ -453,6 +455,27 @@ static Value *expandRadiansIntrinsic(CallInst *Orig) {
   return Builder.CreateFMul(X, PiOver180);
 }
 
+static Value *expandMemoryBarrier(CallInst *Orig, Intrinsic::ID IntrinsicId) {
+  assert(IntrinsicId == Intrinsic::dx_group_memory_barrier_with_group_sync);
+  unsigned BarrierMode = 0;
+  switch (IntrinsicId) {
+  case Intrinsic::dx_group_memory_barrier_with_group_sync:
+    BarrierMode = (unsigned)dxil::BarrierMode::TGSMFence |
+                  (unsigned)dxil::BarrierMode::SyncThreadGroup;
+    break;
+  default:
+    report_fatal_error(Twine("Unexpected memory barrier intrinsic."),
+                       /* gen_crash_diag=*/false);
+    break;
+  }
+
+  IRBuilder<> Builder(Orig);
+  return Builder.CreateIntrinsic(
+      Builder.getVoidTy(), Intrinsic::dx_memory_barrier,
+      ArrayRef<Value *>{Builder.getInt32(BarrierMode)}, nullptr,
+      Orig->getName());
+}
+
 static Intrinsic::ID getMaxForClamp(Intrinsic::ID ClampIntrinsic) {
   if (ClampIntrinsic == Intrinsic::dx_uclamp)
     return Intrinsic::umax;
@@ -580,6 +603,9 @@ static bool expandIntrinsic(Function &F, CallInst *Orig) {
   case Intrinsic::dx_radians:
     Result = expandRadiansIntrinsic(Orig);
     break;
+  case Intrinsic::dx_group_memory_barrier_with_group_sync:
+    Result = expandMemoryBarrier(Orig, IntrinsicId);
+    break;
   }
   if (Result) {
     Orig->replaceAllUsesWith(Result);

llvm/lib/Target/DirectX/DXILOpLowering.cpp

@@ -260,29 +260,6 @@ class OpLowerer {
     });
   }
 
-  [[nodiscard]] bool lowerBarrier(Function &F, Intrinsic::ID IntrId,
-                                  ArrayRef<dxil::BarrierMode> BarrierModes) {
-    unsigned BarrierMode = 0;
-    for (const dxil::BarrierMode B : BarrierModes) {
-      BarrierMode |= (unsigned)B;
-    }
-    IRBuilder<> &IRB = OpBuilder.getIRB();
-    return replaceFunction(F, [&](CallInst *CI) -> Error {
-      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();
     Type *Int32Ty = IRB.getInt32Ty();
@@ -669,11 +646,6 @@ class OpLowerer {
     HasErrors |= replaceFunctionWithOp(F, OpCode);                             \
     break;
 #include "DXILOperation.inc"
-      case Intrinsic::dx_group_memory_barrier_with_group_sync:
-        HasErrors |= lowerBarrier(
-            F, ID,
-            {dxil::BarrierMode::TGSMFence, dxil::BarrierMode::SyncThreadGroup});
-        break;
       case Intrinsic::dx_handle_fromBinding:
         HasErrors |= lowerHandleFromBinding(F);
         break;

llvm/test/CodeGen/DirectX/group_memory_barrier_with_group_sync.ll

@@ -1,4 +1,4 @@
-; RUN: opt -S  -dxil-op-lower -mtriple=dxil-pc-shadermodel6.3-library < %s | FileCheck %s --check-prefix=CHECK
+; RUN: opt -S -dxil-intrinsic-expansion -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: