[flang][openacc] Added acc::RecipeInterface for getting alloca insertion point. (#68464)

Conversion of `hlfir.assign` operations inside OpenACC recipe operations
may result in `fir.alloca` insertion. FIRBuilder can only handle
alloca insertion inside FuncOp's and outlineable OpenMP operations.
I added a simple interface for OpenACC recipe operations that have
executable code inside all their regions, and alloca may be inserted
into the entry blocks of those regions always.

With our current approach the OptimizedBufferization pass is supposed
to lower these `hlfir.assign` operations into loops, because there
should not be conflicts between lhs/rhs. The pass is currently
only working on FuncOp, and this is why it does not optimize
`hlfir.assign` inside the recipes. I will fix it in a separate commit.

Since we run OptimizedBufferization only at >O0, these changes
should still be useful.

Note that the OpenACC codegen that applies the recipes should be aware
of potential alloca operations and produce appropriate stack clean-ups.

Author: vzakhari

flang/include/flang/Optimizer/HLFIR/Passes.td

@@ -13,6 +13,9 @@ include "mlir/Pass/PassBase.td"
 def ConvertHLFIRtoFIR : Pass<"convert-hlfir-to-fir", "::mlir::ModuleOp"> {
   let summary = "Lower High-Level FIR to FIR";
   let constructor = "hlfir::createConvertHLFIRtoFIRPass()";
+  let dependentDialects = [
+    "mlir::func::FuncDialect",
+  ];
 }
 
 def BufferizeHLFIR : Pass<"bufferize-hlfir", "::mlir::ModuleOp"> {

flang/lib/Optimizer/Builder/FIRBuilder.cpp

@@ -18,6 +18,7 @@
 #include "flang/Optimizer/Dialect/FIROpsSupport.h"
 #include "flang/Optimizer/Support/FatalError.h"
 #include "flang/Optimizer/Support/InternalNames.h"
+#include "mlir/Dialect/OpenACC/OpenACC.h"
 #include "mlir/Dialect/OpenMP/OpenMPDialect.h"
 #include "llvm/ADT/ArrayRef.h"
 #include "llvm/ADT/StringExtras.h"
@@ -200,9 +201,17 @@ mlir::Value fir::FirOpBuilder::allocateLocal(
 
 /// Get the block for adding Allocas.
 mlir::Block *fir::FirOpBuilder::getAllocaBlock() {
-  auto iface =
-      getRegion().getParentOfType<mlir::omp::OutlineableOpenMPOpInterface>();
-  return iface ? iface.getAllocaBlock() : getEntryBlock();
+  if (auto ompOutlineableIface =
+          getRegion()
+              .getParentOfType<mlir::omp::OutlineableOpenMPOpInterface>()) {
+    return ompOutlineableIface.getAllocaBlock();
+  }
+  if (auto accRecipeIface =
+          getRegion().getParentOfType<mlir::acc::RecipeInterface>()) {
+    return accRecipeIface.getAllocaBlock(getRegion());
+  }
+
+  return getEntryBlock();
 }
 
 mlir::Value fir::FirOpBuilder::createTemporaryAlloc(

flang/test/HLFIR/convert-assign-inside-openacc-recipe.fir

@@ -0,0 +1,51 @@
+// Check that hlfir.assign codegen is able to insert fir.alloca's inside
+// the regions of the OpenACC recipe.
+// RUN: fir-opt %s -convert-hlfir-to-fir | FileCheck %s
+
+acc.reduction.recipe @reduction_add_box_heap_Uxi32 : !fir.box<!fir.heap<!fir.array<?xi32>>> reduction_operator <add> init {
+^bb0(%arg0: !fir.box<!fir.heap<!fir.array<?xi32>>>):
+  %c0_i32 = arith.constant 0 : i32
+  %c0 = arith.constant 0 : index
+  %0:3 = fir.box_dims %arg0, %c0 : (!fir.box<!fir.heap<!fir.array<?xi32>>>, index) -> (index, index, index)
+  %1 = fir.shape %0#1 : (index) -> !fir.shape<1>
+  %2 = fir.allocmem !fir.array<?xi32>, %0#1 {bindc_name = ".tmp", uniq_name = ""}
+  %3:2 = hlfir.declare %2(%1) {uniq_name = ".tmp"} : (!fir.heap<!fir.array<?xi32>>, !fir.shape<1>) -> (!fir.box<!fir.array<?xi32>>, !fir.heap<!fir.array<?xi32>>)
+  hlfir.assign %c0_i32 to %3#0 : i32, !fir.box<!fir.array<?xi32>>
+  acc.yield %3#0 : !fir.box<!fir.array<?xi32>>
+} combiner {
+^bb0(%arg0: !fir.box<!fir.heap<!fir.array<?xi32>>>, %arg1: !fir.box<!fir.heap<!fir.array<?xi32>>>, %arg2: index, %arg3: index, %arg4: index):
+  %c1 = arith.constant 1 : index
+  %c0 = arith.constant 0 : index
+  %0 = arith.subi %arg3, %arg2 : index
+  %1 = arith.addi %0, %c1 : index
+  %2 = arith.divsi %1, %arg4 : index
+  %3 = arith.cmpi sgt, %2, %c0 : index
+  %4 = arith.select %3, %2, %c0 : index
+  %5 = fir.shape %4 : (index) -> !fir.shape<1>
+  %6 = hlfir.designate %arg0 (%arg2:%arg3:%arg4)  shape %5 : (!fir.box<!fir.heap<!fir.array<?xi32>>>, index, index, index, !fir.shape<1>) -> !fir.box<!fir.heap<!fir.array<?xi32>>>
+  %7 = hlfir.designate %arg1 (%arg2:%arg3:%arg4)  shape %5 : (!fir.box<!fir.heap<!fir.array<?xi32>>>, index, index, index, !fir.shape<1>) -> !fir.box<!fir.heap<!fir.array<?xi32>>>
+  %8 = fir.allocmem !fir.array<?xi32>, %4 {bindc_name = ".tmp.array", uniq_name = ""}
+  %9:2 = hlfir.declare %8(%5) {uniq_name = ".tmp.array"} : (!fir.heap<!fir.array<?xi32>>, !fir.shape<1>) -> (!fir.box<!fir.array<?xi32>>, !fir.heap<!fir.array<?xi32>>)
+  %true = arith.constant true
+  %c1_0 = arith.constant 1 : index
+  fir.do_loop %arg5 = %c1_0 to %4 step %c1_0 unordered {
+    %13 = hlfir.designate %6 (%arg5)  : (!fir.box<!fir.heap<!fir.array<?xi32>>>, index) -> !fir.ref<i32>
+    %14 = hlfir.designate %7 (%arg5)  : (!fir.box<!fir.heap<!fir.array<?xi32>>>, index) -> !fir.ref<i32>
+    %15 = fir.load %13 : !fir.ref<i32>
+    %16 = fir.load %14 : !fir.ref<i32>
+    %17 = arith.addi %15, %16 : i32
+    %18 = hlfir.designate %9#0 (%arg5)  : (!fir.box<!fir.array<?xi32>>, index) -> !fir.ref<i32>
+    hlfir.assign %17 to %18 temporary_lhs : i32, !fir.ref<i32>
+  }
+  %10 = fir.undefined tuple<!fir.box<!fir.array<?xi32>>, i1>
+  %11 = fir.insert_value %10, %true, [1 : index] : (tuple<!fir.box<!fir.array<?xi32>>, i1>, i1) -> tuple<!fir.box<!fir.array<?xi32>>, i1>
+  %12 = fir.insert_value %11, %9#0, [0 : index] : (tuple<!fir.box<!fir.array<?xi32>>, i1>, !fir.box<!fir.array<?xi32>>) -> tuple<!fir.box<!fir.array<?xi32>>, i1>
+  hlfir.assign %9#0 to %arg0 : !fir.box<!fir.array<?xi32>>, !fir.box<!fir.heap<!fir.array<?xi32>>>
+  acc.yield %arg0 : !fir.box<!fir.heap<!fir.array<?xi32>>>
+}
+// CHECK-LABEL:   acc.reduction.recipe @reduction_add_box_heap_Uxi32 : !fir.box<!fir.heap<!fir.array<?xi32>>> reduction_operator <add> init {
+// CHECK:         ^bb0(%[[VAL_0:.*]]: !fir.box<!fir.heap<!fir.array<?xi32>>>):
+// CHECK:           %[[VAL_1:.*]] = fir.alloca !fir.box<!fir.array<?xi32>>
+// CHECK-LABEL:   } combiner {
+// CHECK:         ^bb0(%[[VAL_0:.*]]: !fir.box<!fir.heap<!fir.array<?xi32>>>, %[[VAL_1:.*]]: !fir.box<!fir.heap<!fir.array<?xi32>>>, %[[VAL_2:.*]]: index, %[[VAL_3:.*]]: index, %[[VAL_4:.*]]: index):
+// CHECK:           %[[VAL_5:.*]] = fir.alloca !fir.box<!fir.heap<!fir.array<?xi32>>>

mlir/include/mlir/Dialect/OpenACC/CMakeLists.txt

@@ -24,3 +24,5 @@ mlir_tablegen(OpenACCTypeInterfaces.h.inc -gen-type-interface-decls)
 mlir_tablegen(OpenACCTypeInterfaces.cpp.inc -gen-type-interface-defs)
 add_public_tablegen_target(MLIROpenACCTypeInterfacesIncGen)
 add_dependencies(mlir-headers MLIROpenACCTypeInterfacesIncGen)
+
+add_mlir_interface(OpenACCOpsInterfaces)

mlir/include/mlir/Dialect/OpenACC/OpenACC.h

@@ -33,6 +33,8 @@
 #define GET_ATTRDEF_CLASSES
 #include "mlir/Dialect/OpenACC/OpenACCOpsAttributes.h.inc"
 
+#include "mlir/Dialect/OpenACC/OpenACCInterfaces.h"
+
 #define GET_OP_CLASSES
 #include "mlir/Dialect/OpenACC/OpenACCOps.h.inc"
 

mlir/include/mlir/Dialect/OpenACC/OpenACCInterfaces.h

@@ -0,0 +1,20 @@
+//===- OpenACCInterfaces.h - MLIR Interfaces for OpenACC --------*- C++ -*-===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+//
+// This file declares OpenACC Interface implementations for the OpenACC dialect.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef MLIR_DIALECT_OPENACC_OPENACCINTERFACES_H_
+#define MLIR_DIALECT_OPENACC_OPENACCINTERFACES_H_
+
+#include "mlir/IR/OpDefinition.h"
+
+#include "mlir/Dialect/OpenACC/OpenACCOpsInterfaces.h.inc"
+
+#endif // MLIR_DIALECT_OPENACC_OPENACCINTERFACES_H_

mlir/include/mlir/Dialect/OpenACC/OpenACCOps.td

@@ -21,6 +21,7 @@ include "mlir/IR/OpBase.td"
 include "mlir/IR/SymbolInterfaces.td"
 include "mlir/Dialect/OpenACC/OpenACCBase.td"
 include "mlir/Dialect/OpenACC/OpenACCOpsTypes.td"
+include "mlir/Dialect/OpenACC/OpenACCOpsInterfaces.td"
 include "mlir/Dialect/OpenACC/OpenACCTypeInterfaces.td"
 include "mlir/Dialect/OpenACCMPCommon/Interfaces/AtomicInterfaces.td"
 
@@ -519,7 +520,7 @@ def OpenACC_UpdateHostOp : OpenACC_DataExitOp<"update_host",
 //===----------------------------------------------------------------------===//
 
 def OpenACC_PrivateRecipeOp : OpenACC_Op<"private.recipe",
-    [IsolatedFromAbove, Symbol]> {
+    [IsolatedFromAbove, Symbol, RecipeInterface]> {
   let summary = "privatization recipe";
 
   let description = [{
@@ -576,7 +577,7 @@ def OpenACC_PrivateRecipeOp : OpenACC_Op<"private.recipe",
 //===----------------------------------------------------------------------===//
 
 def OpenACC_FirstprivateRecipeOp : OpenACC_Op<"firstprivate.recipe",
-    [IsolatedFromAbove, Symbol]> {
+    [IsolatedFromAbove, Symbol, RecipeInterface]> {
   let summary = "privatization recipe";
 
   let description = [{
@@ -642,7 +643,7 @@ def OpenACC_FirstprivateRecipeOp : OpenACC_Op<"firstprivate.recipe",
 //===----------------------------------------------------------------------===//
 
 def OpenACC_ReductionRecipeOp : OpenACC_Op<"reduction.recipe",
-    [IsolatedFromAbove, Symbol]> {
+    [IsolatedFromAbove, Symbol, RecipeInterface]> {
   let summary = "reduction recipe";
 
   let description = [{

mlir/include/mlir/Dialect/OpenACC/OpenACCOpsInterfaces.td

@@ -0,0 +1,41 @@
+//===-- OpenACCOpsInterfaces.td - OpenACC op interfaces ----*- tablegen -*-===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+//
+// This is the OpenACC Dialect interfaces definition file.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef OPENACC_OPS_INTERFACES
+#define OPENACC_OPS_INTERFACES
+
+include "mlir/IR/OpBase.td"
+
+def RecipeInterface : OpInterface<"RecipeInterface"> {
+    let description = [{
+      OpenACC operations with one or more regions holding executable code.
+    }];
+    let cppNamespace = "::mlir::acc";
+    let methods = [
+      InterfaceMethod<
+        /*description=*/[{
+          For the given region of the operation return the block
+          inside the region, where an alloca-like operation should be inserted.
+          The default implementation returns the entry block of the region.
+        }],
+        /*retTy*/"::mlir::Block *",
+        /*methodName=*/"getAllocaBlock",
+        /*args=*/(ins "::mlir::Region &":$region),
+        /*methodBody=*/"",
+        /*defaultImplementation=*/[{
+          return &region.front();
+        }]
+      >,
+    ];
+}
+
+#endif // OPENACC_OPS_INTERFACES

mlir/lib/Dialect/OpenACC/CMakeLists.txt

@@ -8,6 +8,7 @@ add_mlir_dialect_library(MLIROpenACCDialect
   MLIROpenACCOpsIncGen
   MLIROpenACCEnumsIncGen
   MLIROpenACCAttributesIncGen
+  MLIROpenACCOpsInterfacesIncGen
   MLIROpenACCTypeInterfacesIncGen
 
   LINK_LIBS PUBLIC

mlir/lib/Dialect/OpenACC/IR/OpenACC.cpp

@@ -23,6 +23,7 @@ using namespace acc;
 
 #include "mlir/Dialect/OpenACC/OpenACCOpsDialect.cpp.inc"
 #include "mlir/Dialect/OpenACC/OpenACCOpsEnums.cpp.inc"
+#include "mlir/Dialect/OpenACC/OpenACCOpsInterfaces.cpp.inc"
 #include "mlir/Dialect/OpenACC/OpenACCTypeInterfaces.cpp.inc"
 
 namespace {