[mlir][NVVM] Add support for barrier0-reduction operation (llvm#167036)

Add support for `nvvm.barrier0.[popc|and|or]` operation. It is added as
a separate operation since `Barrier0Op` has no result.

https://docs.nvidia.com/cuda/nvvm-ir-spec/#barrier-and-memory-fence

This will be used in CUDA Fortran lowering: 

https://github.com/llvm/llvm-project/blob/49f55f4991227f3c7a2b8161bbf45c74b7023944/flang/lib/Optimizer/Builder/CUDAIntrinsicCall.cpp#L1081

And could be used later in the CUDA C/C++ with CIR

https://github.com/llvm/llvm-project/blob/49f55f4991227f3c7a2b8161bbf45c74b7023944/clang/lib/Headers/__clang_cuda_device_functions.h#L524

---------

Co-authored-by: Guray Ozen <[email protected]>

Author: 2 people

mlir/include/mlir/Dialect/LLVMIR/NVVMOps.td

@@ -921,6 +921,23 @@ def NVVM_Barrier0Op : NVVM_Op<"barrier0"> {
   }];
 }
 
+// Attrs describing the reduction operations for the barrier operation.
+def BarrierReductionPopc : I32EnumAttrCase<"POPC", 0, "popc">;
+def BarrierReductionAnd : I32EnumAttrCase<"AND", 1, "and">;
+def BarrierReductionOr : I32EnumAttrCase<"OR", 2, "or">;
+
+def BarrierReduction
+    : I32EnumAttr<"BarrierReduction", "NVVM barrier reduction operation",
+                  [BarrierReductionPopc, BarrierReductionAnd,
+                   BarrierReductionOr]> {
+  let genSpecializedAttr = 0;
+  let cppNamespace = "::mlir::NVVM";
+}
+def BarrierReductionAttr
+    : EnumAttr<NVVM_Dialect, BarrierReduction, "reduction"> {
+  let assemblyFormat = "`<` $value `>`";
+}
+
 def NVVM_BarrierOp : NVVM_Op<"barrier", [AttrSizedOperandSegments]> {
   let summary = "CTA Barrier Synchronization Op";
   let description = [{
@@ -935,6 +952,9 @@ def NVVM_BarrierOp : NVVM_Op<"barrier", [AttrSizedOperandSegments]> {
     - `numberOfThreads`: Specifies the number of threads participating in the barrier. 
       When specified, the value must be a multiple of the warp size. If not specified, 
       all threads in the CTA participate in the barrier.
+    - `reductionOp`: specifies the reduction operation (`popc`, `and`, `or`).
+    - `reductionPredicate`: specifies the predicate to be used with the
+      `reductionOp`. 
 
     The barrier operation guarantees that when the barrier completes, prior memory 
     accesses requested by participating threads are performed relative to all threads 
@@ -951,31 +971,37 @@ def NVVM_BarrierOp : NVVM_Op<"barrier", [AttrSizedOperandSegments]> {
     [For more information, see PTX ISA](https://docs.nvidia.com/cuda/parallel-thread-execution/index.html#parallel-synchronization-and-communication-instructions-bar)
   }];
 
-  let arguments = (ins     
-    Optional<I32>:$barrierId,
-    Optional<I32>:$numberOfThreads);
+  let extraClassDeclaration = [{
+    static mlir::NVVM::IDArgPair
+      getIntrinsicIDAndArgs(Operation &op, LLVM::ModuleTranslation &mt,
+                            llvm::IRBuilderBase& builder);
+  }];
+
+  let arguments = (ins Optional<I32>:$barrierId, Optional<I32>:$numberOfThreads,
+      OptionalAttr<BarrierReductionAttr>:$reductionOp,
+      Optional<I32>:$reductionPredicate);
   string llvmBuilder = [{
-    llvm::Value *id = $barrierId ? $barrierId : builder.getInt32(0);
-    if ($numberOfThreads)
-      createIntrinsicCall(
-          builder, llvm::Intrinsic::nvvm_barrier_cta_sync_aligned_count,
-          {id, $numberOfThreads});
-    else
-      createIntrinsicCall(
-          builder, llvm::Intrinsic::nvvm_barrier_cta_sync_aligned_all, {id});
+    auto [id, args] = NVVM::BarrierOp::getIntrinsicIDAndArgs(
+                        *op, moduleTranslation, builder);
+    if ($reductionOp)
+      $res = createIntrinsicCall(builder, id, args);
+    else 
+      createIntrinsicCall(builder, id, args);
   }];
+  let results = (outs Optional<I32>:$res);
+
   let hasVerifier = 1;
 
-  let assemblyFormat = "(`id` `=` $barrierId^)? (`number_of_threads` `=` $numberOfThreads^)? attr-dict";
+  let assemblyFormat =
+      "(`id` `=` $barrierId^)? (`number_of_threads` `=` $numberOfThreads^)? "
+      "($reductionOp^ $reductionPredicate)? (`->` type($res)^)? attr-dict";
 
-  let builders = [
-    OpBuilder<(ins), [{
-      return build($_builder, $_state, Value{}, Value{});
+  let builders = [OpBuilder<(ins), [{
+      return build($_builder, $_state, TypeRange{}, Value{}, Value{}, {}, Value{});
     }]>,
-    OpBuilder<(ins "Value":$barrierId), [{
-      return build($_builder, $_state, barrierId, Value{});
-    }]>
-  ];
+                  OpBuilder<(ins "Value":$barrierId), [{
+      return build($_builder, $_state, TypeRange{}, barrierId, Value{}, {}, Value{});
+    }]>];
 }
 
 def NVVM_BarrierArriveOp : NVVM_PTXBuilder_Op<"barrier.arrive"> 

mlir/lib/Dialect/LLVMIR/IR/NVVMDialect.cpp

@@ -1517,6 +1517,15 @@ LogicalResult NVVM::BarrierOp::verify() {
   if (getNumberOfThreads() && !getBarrierId())
     return emitOpError(
         "barrier id is missing, it should be set between 0 to 15");
+
+  if (getBarrierId() && (getReductionOp() || getReductionPredicate()))
+    return emitOpError("reduction are only available when id is 0");
+
+  if ((getReductionOp() && !getReductionPredicate()) ||
+      (!getReductionOp() && getReductionPredicate()))
+    return emitOpError("reduction predicate and reduction operation must be "
+                       "specified together");
+
   return success();
 }
 
@@ -1785,6 +1794,39 @@ std::string NVVM::MBarrierTryWaitParityOp::getPtx() {
 // getIntrinsicID/getIntrinsicIDAndArgs methods
 //===----------------------------------------------------------------------===//
 
+mlir::NVVM::IDArgPair NVVM::BarrierOp::getIntrinsicIDAndArgs(
+    Operation &op, LLVM::ModuleTranslation &mt, llvm::IRBuilderBase &builder) {
+  auto thisOp = cast<NVVM::BarrierOp>(op);
+  llvm::Value *barrierId = thisOp.getBarrierId()
+                               ? mt.lookupValue(thisOp.getBarrierId())
+                               : builder.getInt32(0);
+  llvm::Intrinsic::ID id;
+  llvm::SmallVector<llvm::Value *> args;
+  if (thisOp.getNumberOfThreads()) {
+    id = llvm::Intrinsic::nvvm_barrier_cta_sync_aligned_count;
+    args.push_back(barrierId);
+    args.push_back(mt.lookupValue(thisOp.getNumberOfThreads()));
+  } else if (thisOp.getReductionOp()) {
+    switch (*thisOp.getReductionOp()) {
+    case NVVM::BarrierReduction::AND:
+      id = llvm::Intrinsic::nvvm_barrier0_and;
+      break;
+    case NVVM::BarrierReduction::OR:
+      id = llvm::Intrinsic::nvvm_barrier0_or;
+      break;
+    case NVVM::BarrierReduction::POPC:
+      id = llvm::Intrinsic::nvvm_barrier0_popc;
+      break;
+    }
+    args.push_back(mt.lookupValue(thisOp.getReductionPredicate()));
+  } else {
+    id = llvm::Intrinsic::nvvm_barrier_cta_sync_aligned_all;
+    args.push_back(barrierId);
+  }
+
+  return {id, std::move(args)};
+}
+
 mlir::NVVM::IDArgPair MBarrierInitOp::getIntrinsicIDAndArgs(
     Operation &op, LLVM::ModuleTranslation &mt, llvm::IRBuilderBase &builder) {
   auto thisOp = cast<NVVM::MBarrierInitOp>(op);

mlir/test/Target/LLVMIR/nvvm/barrier.mlir

@@ -0,0 +1,20 @@
+// RUN: mlir-translate -mlir-to-llvmir %s  -split-input-file --verify-diagnostics | FileCheck %s
+
+// CHECK-LABEL: @llvm_nvvm_barrier(
+// CHECK-SAME: i32 %[[barId:.*]], i32 %[[numThreads:.*]], i32 %[[redOperand:.*]])
+llvm.func @llvm_nvvm_barrier(%barID : i32, %numberOfThreads : i32, %redOperand : i32) {
+  // CHECK: call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0)
+  nvvm.barrier
+  // CHECK: call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 %[[barId]])
+  nvvm.barrier id = %barID
+  // CHECK: call void @llvm.nvvm.barrier.cta.sync.aligned.count(i32 %[[barId]], i32 %[[numThreads]])
+  nvvm.barrier id = %barID number_of_threads = %numberOfThreads
+  // CHECK: %{{.*}} = call i32 @llvm.nvvm.barrier0.and(i32 %[[redOperand]])
+  %0 = nvvm.barrier #nvvm.reduction<and> %redOperand -> i32
+  // CHECK: %{{.*}} = call i32 @llvm.nvvm.barrier0.or(i32 %[[redOperand]])
+  %1 = nvvm.barrier #nvvm.reduction<or> %redOperand -> i32
+  // CHECK: %{{.*}} = call i32 @llvm.nvvm.barrier0.popc(i32 %[[redOperand]])
+  %2 = nvvm.barrier #nvvm.reduction<popc> %redOperand -> i32
+
+  llvm.return
+}

mlir/test/Target/LLVMIR/nvvmir.mlir

@@ -166,25 +166,6 @@ llvm.func @nvvm_rcp(%0: f32) -> f32 {
   llvm.return %1 : f32
 }
 
-// CHECK-LABEL: @llvm_nvvm_barrier0
-llvm.func @llvm_nvvm_barrier0() {
-  // CHECK: call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0)
-  nvvm.barrier0
-  llvm.return
-}
-
-// CHECK-LABEL: @llvm_nvvm_barrier(
-// CHECK-SAME: i32 %[[barId:.*]], i32 %[[numThreads:.*]])
-llvm.func @llvm_nvvm_barrier(%barID : i32, %numberOfThreads : i32) {
-  // CHECK: call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0)
-  nvvm.barrier
-  // CHECK: call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 %[[barId]])
-  nvvm.barrier id = %barID
-  // CHECK: call void @llvm.nvvm.barrier.cta.sync.aligned.count(i32 %[[barId]], i32 %[[numThreads]])
-  nvvm.barrier id = %barID number_of_threads = %numberOfThreads
-  llvm.return
-}
-
 // CHECK-LABEL: @llvm_nvvm_cluster_arrive
 llvm.func @llvm_nvvm_cluster_arrive() {
   // CHECK: call void @llvm.nvvm.barrier.cluster.arrive()