[mlir][openacc] Update acc.loop to be a proper loop like operation

[clementval]

Summary

The initial design of the acc.loop was to be an operation that encapsulates
a loop like operation. This was an early design and we now want to change it
so the acc.loop operation becomes a real loop-like operation by implementing
the LoopLikeInterface.

Note that flang lowering will fail with this new design and a patch to update the lowering will be posted alongside this one.

[clementval]

Update example and lit tests for collapse case

[razvanlupusoru]

Looks great to me!

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

1139

The source language may have both inclusive and exclusive upper bounds - so in a multiple loop scenario I am not sure if a single attribute will be enough.

[clementval]

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

1139

I'm gonna update this with a DenseBoolArrayAttr so we can have a vlue for each "loop".

[clementval]

Switch inclusiveUpperbound to DenseBoolArrayAttr

[clementval]

Make range optional

[clementval]

Move to github https://github.com/llvm/llvm-project/pull/67355

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

 #include "mlir/IR/OpDefinition.h"
 #include "mlir/IR/SymbolTable.h"
 
 #include "mlir/Bytecode/BytecodeOpInterface.h"
 #include "mlir/Dialect/OpenACC/OpenACCOpsDialect.h.inc"
 #include "mlir/Dialect/OpenACC/OpenACCOpsEnums.h.inc"
 #include "mlir/Dialect/OpenACC/OpenACCTypeInterfaces.h.inc"
 #include "mlir/Interfaces/ControlFlowInterfaces.h"
+#include "mlir/Interfaces/LoopLikeInterface.h"
 #include "mlir/Interfaces/SideEffectInterfaces.h"
 
 #define GET_TYPEDEF_CLASSES
 #include "mlir/Dialect/OpenACC/OpenACCOpsTypes.h.inc"
 
 #define GET_ATTRDEF_CLASSES
 #include "mlir/Dialect/OpenACC/OpenACCOpsAttributes.h.inc"
 

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

 //===- OpenACCOps.td - OpenACC operation definitions -------*- tablegen -*-===//
 //
 // Part of the MLIR 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
 //
 // =============================================================================
 //
 // Defines MLIR OpenACC operations.
 //
 //===----------------------------------------------------------------------===//
 
 #ifndef OPENACC_OPS
 #define OPENACC_OPS
 
 include "mlir/Interfaces/ControlFlowInterfaces.td"
+include "mlir/Interfaces/LoopLikeInterface.td"
 include "mlir/Interfaces/SideEffectInterfaces.td"
 include "mlir/IR/BuiltinTypes.td"
 include "mlir/IR/EnumAttr.td"
 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/OpenACCTypeInterfaces.td"
def OpenACC_HostDataOp : OpenACC_Op<"host_data", [AttrSizedOperandSegments]> {
   let hasCanonicalizer = 1;
 }
 
 //===----------------------------------------------------------------------===//
 // 2.9 loop Construct
 //===----------------------------------------------------------------------===//
 
 def OpenACC_LoopOp : OpenACC_Op<"loop",
-    [AttrSizedOperandSegments, RecursiveMemoryEffects]> {
+    [AttrSizedOperandSegments, RecursiveMemoryEffects,
+     DeclareOpInterfaceMethods<LoopLikeOpInterface>]> {
   let summary = "loop construct";
 
   let description = [{
-    The "acc.loop" operation represents the OpenACC loop construct.
+    The "acc.loop" operation represents the OpenACC loop construct. The lower
+    and upper bounds specify a half-open range: the range includes the lower
+    bound but does not include the upper bound. If the `inclusive` attribute is
+    set then the upper bound is included.
 
     Example:
 
     ```mlir
-    acc.loop gang vector {
-      scf.for %arg3 = %c0 to %c10 step %c1 {
-        scf.for %arg4 = %c0 to %c10 step %c1 {
-          scf.for %arg5 = %c0 to %c10 step %c1 {
-            // ... body
-          }
-        }
-      }
+    acc.loop gang() vector() (%arg3 : index) = (%c0 : index) to (%c10 : index) step (%c1 : step) {
+      // Loop body
       acc.yield
     } attributes { collapse = 3 }
     ```
   }];
 
-  let arguments = (ins OptionalAttr<I64Attr>:$collapse,
-                       Optional<IntOrIndex>:$gangNum,
-                       Optional<IntOrIndex>:$gangDim,
-                       Optional<IntOrIndex>:$gangStatic,
-                       Optional<IntOrIndex>:$workerNum,
-                       Optional<IntOrIndex>:$vectorLength,
-                       UnitAttr:$seq,
-                       UnitAttr:$independent,
-                       UnitAttr:$auto_,
-                       UnitAttr:$hasGang,
-                       UnitAttr:$hasWorker,
-                       UnitAttr:$hasVector,
-                       Variadic<IntOrIndex>:$tileOperands,
-                       Variadic<OpenACC_PointerLikeTypeInterface>:$privateOperands,
-                       OptionalAttr<SymbolRefArrayAttr>:$privatizations,
-                       Variadic<AnyType>:$reductionOperands,
-                       OptionalAttr<SymbolRefArrayAttr>:$reductionRecipes);
+  let arguments = (ins
+    Variadic<IntOrIndex>:$lowerbound,
+    Variadic<IntOrIndex>:$upperbound,
+    Variadic<IntOrIndex>:$step,
+    UnitAttr:$inclusiveUpperbound,
+    OptionalAttr<I64Attr>:$collapse,
+    OptionalAttr<SymbolRefArrayAttr>:$reductionRecipes
+  );
 
   let results = (outs Variadic<AnyType>:$results);
 
   let regions = (region AnyRegion:$region);
 
   let extraClassDeclaration = [{
     static StringRef getAutoAttrStrName() { return "auto"; }
     static StringRef getGangNumKeyword() { return "num"; }
     static StringRef getGangDimKeyword() { return "dim"; }
     static StringRef getGangStaticKeyword() { return "static"; }
   }];
 
   let hasCustomAssemblyFormat = 1;
   let assemblyFormat = [{
     oilist(
-        `gang` `` custom<GangClause>($gangNum, type($gangNum), $gangDim, type($gangDim), $gangStatic, type($gangStatic), $hasGang)
-      | `worker` `` custom<WorkerClause>($workerNum, type($workerNum), $hasWorker)
-      | `vector` `` custom<VectorClause>($vectorLength, type($vectorLength), $hasVector)
+        `gang` `` custom<GangClause>($gangNum, type($gangNum), $gangDim,
+            type($gangDim), $gangStatic, type($gangStatic), $hasGang)
+      | `worker` ``
+          custom<WorkerClause>($workerNum, type($workerNum), $hasWorker)
+      | `vector` ``
+          custom<VectorClause>($vectorLength, type($vectorLength), $hasVector)
       | `private` `(` custom<SymOperandList>(
             $privateOperands, type($privateOperands), $privatizations)
         `)`
       | `tile` `(` $tileOperands `:` type($tileOperands) `)`
       | `reduction` `(` custom<SymOperandList>(
             $reductionOperands, type($reductionOperands), $reductionRecipes)
         `)`
     )
-    $region
+    custom<LoopControl>($region, $lowerbound, type($lowerbound), $upperbound,
+        type($upperbound), $step, type($step))
     ( `(` type($results)^ `)` )?
     attr-dict-with-keyword
   }];
 
   let hasVerifier = 1;
 }
 
 // Yield operation for the acc.loop and acc.parallel operations.

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

static ParseResult parseGangClause(
   hasGang = UnitAttr::get(parser.getBuilder().getContext());
   gangNum = std::nullopt;
   gangDim = std::nullopt;
   gangStatic = std::nullopt;
   bool needComa = false;
 
   // optional gang operands
   if (succeeded(parser.parseOptionalLParen())) {
+    // Allow empty parenthesis.
+    if (succeeded(parser.parseOptionalRParen()))
+      return success();
     while (true) {
       bool newValue = false;
       bool needValue = false;
       if (needComa) {
         if (succeeded(parser.parseOptionalComma()))
           needValue = true; // expect a new value after comma.
         else
           break;
if (failed(parser.parseRParen()))
       return failure();
   }
   return success();
 }
 
 void printGangClause(OpAsmPrinter &p, Operation *op, Value gangNum,
                      Type gangNumType, Value gangDim, Type gangDimType,
                      Value gangStatic, Type gangStaticType, UnitAttr hasGang) {
-  if (gangNum || gangStatic || gangDim) {
-    p << "(";
+  if (gangNum || gangStatic || gangDim) {
     if (gangNum) {
       p << LoopOp::getGangNumKeyword() << "=" << gangNum << " : "
         << gangNumType;
       if (gangStatic || gangDim)
         p << ", ";
     }
     if (gangDim) {
       p << LoopOp::getGangDimKeyword() << "=" << gangDim << " : "
         << gangDimType;
       if (gangStatic)
         p << ", ";
     }
     if (gangStatic)
       p << LoopOp::getGangStaticKeyword() << "=" << gangStatic << " : "
         << gangStaticType;
-    p << ")";
   }
+  p << ")";
 }
 
 static ParseResult
 parseWorkerClause(OpAsmParser &parser,
                   std::optional<OpAsmParser::UnresolvedOperand> &workerNum,
                   Type &workerNumType, UnitAttr &hasWorker) {
   hasWorker = UnitAttr::get(parser.getBuilder().getContext());
   if (succeeded(parser.parseOptionalLParen())) {
+    // Allow empty parenthesis.
+    if (succeeded(parser.parseOptionalRParen()))
+      return success();
     workerNum = OpAsmParser::UnresolvedOperand{};
     if (parser.parseOperand(*workerNum) ||
         parser.parseColonType(workerNumType) || parser.parseRParen())
       return failure();
   }
   return success();
 }
 
 void printWorkerClause(OpAsmPrinter &p, Operation *op, Value workerNum,
                        Type workerNumType, UnitAttr hasWorker) {
+  p << "(";
   if (workerNum)
-    p << "(" << workerNum << " : " << workerNumType << ")";
+    p << workerNum << " : " << workerNumType;
+  p << ")";
 }
 
 static ParseResult
 parseVectorClause(OpAsmParser &parser,
                   std::optional<OpAsmParser::UnresolvedOperand> &vectorLength,
                   Type &vectorLengthType, UnitAttr &hasVector) {
   hasVector = UnitAttr::get(parser.getBuilder().getContext());
   if (succeeded(parser.parseOptionalLParen())) {
+    // Allow empty parenthesis.
+    if (succeeded(parser.parseOptionalRParen()))
+      return success();
     vectorLength = OpAsmParser::UnresolvedOperand{};
     if (parser.parseOperand(*vectorLength) ||
         parser.parseColonType(vectorLengthType) || parser.parseRParen())
       return failure();
   }
   return success();
 }
 
 void printVectorClause(OpAsmPrinter &p, Operation *op, Value vectorLength,
                        Type vectorLengthType, UnitAttr hasVector) {
+  p << "(";
   if (vectorLength)
-    p << "(" << vectorLength << " : " << vectorLengthType << ")";
+    p << vectorLength << " : " << vectorLengthType;
+  p << ")";
 }
 
 LogicalResult acc::LoopOp::verify() {
   // auto, independent and seq attribute are mutually exclusive.
   if ((getAuto_() && (getIndependent() || getSeq())) ||
       (getIndependent() && getSeq())) {
     return emitError() << "only one of \"" << acc::LoopOp::getAutoAttrStrName()
                        << "\", " << getIndependentAttrName() << ", "
LogicalResult acc::LoopOp::verify() {
 
   // Check non-empty body().
   if (getRegion().empty())
     return emitError("expected non-empty body.");
 
   return success();
 }
 
+Region &acc::LoopOp::getLoopBody() { return getRegion(); }
+
+/// loop-control ::= `(` ssa-id-and-type-list `)` `=` `(` ssa-id-and-type-list
+/// `)` `to` `(` ssa-id-and-type-list `)` `step` `(` ssa-id-and-type-list `)`
+ParseResult
+parseLoopControl(OpAsmParser &parser, Region &region,
+                 SmallVectorImpl<OpAsmParser::UnresolvedOperand> &lowerbound,
+                 SmallVectorImpl<Type> &lowerboundType,
+                 SmallVectorImpl<OpAsmParser::UnresolvedOperand> &upperbound,
+                 SmallVectorImpl<Type> &upperboundType,
+                 SmallVectorImpl<OpAsmParser::UnresolvedOperand> &step,
+                 SmallVectorImpl<Type> &stepType) {
+
+  SmallVector<OpAsmParser::Argument> inductionVars;
+  if (parser.parseArgumentList(inductionVars, OpAsmParser::Delimiter::Paren,
+                               /*allowType=*/true) ||
+      parser.parseEqual() || parser.parseLParen() ||
+      parser.parseOperandList(lowerbound, inductionVars.size(),
+                              OpAsmParser::Delimiter::None) ||
+      parser.parseColonTypeList(lowerboundType) || parser.parseRParen() ||
+      parser.parseKeyword("to") || parser.parseLParen() ||
+      parser.parseOperandList(upperbound, inductionVars.size(),
+                              OpAsmParser::Delimiter::None) ||
+      parser.parseColonTypeList(upperboundType) || parser.parseRParen() ||
+      parser.parseKeyword("step") || parser.parseLParen() ||
+      parser.parseOperandList(step, inductionVars.size(),
+                              OpAsmParser::Delimiter::None) ||
+      parser.parseColonTypeList(stepType) || parser.parseRParen())
+    return failure();
+  return parser.parseRegion(region, inductionVars);
+}
+
+void printLoopControl(OpAsmPrinter &p, Operation *op, Region &region,
+                      ValueRange lowerbound, TypeRange lowerboundType,
+                      ValueRange upperbound, TypeRange upperboundType,
+                      ValueRange steps, TypeRange stepType) {
+  ValueRange regionArgs = region.front().getArguments();
+  p << "(" << regionArgs << " : " << regionArgs.getTypes() << ") = ("
+    << lowerbound << " : " << lowerboundType << ") to (" << upperbound << " : "
+    << upperboundType << ") "
+    << " step (" << steps << " : " << stepType << ")";
+  p.printRegion(region, /*printEntryBlockArgs=*/false);
+}
+
 //===----------------------------------------------------------------------===//
 // DataOp
 //===----------------------------------------------------------------------===//
 
 LogicalResult acc::DataOp::verify() {
   // 2.6.5. Data Construct restriction
   // At least one copy, copyin, copyout, create, no_create, present, deviceptr,
   // attach, or default clause must appear on a data construct.

mlir/test/Dialect/OpenACC/canonicalize.mlir

 
 // CHECK:  func @testupdateop(%{{.*}}: memref<f32>, [[IFCOND:%.*]]: i1)
 // CHECK:    acc.update if(%{{.*}}) dataOperands(%{{.*}} : memref<f32>)
 
 // -----
 
 func.func @testhostdataop(%a: memref<f32>, %ifCond: i1) -> () {
   %0 = acc.use_device varPtr(%a : memref<f32>) -> memref<f32>
+  %1 = arith.constant 1 : i32
+  %2 = arith.constant 10 : i32
   %false = arith.constant false
   acc.host_data dataOperands(%0 : memref<f32>) if(%false) {
-    acc.loop {
+    acc.loop (%iv : i32) = (%1 : i32) to (%2 : i32) step (%1 : i32) {
       acc.yield
     }
-    acc.loop {
+    acc.loop (%iv : i32) = (%1 : i32) to (%2 : i32) step (%1 : i32) {
       acc.yield
     }
     acc.terminator
   }
   return
 }
 
 // CHECK-LABEL: func.func @testhostdataop

mlir/test/Dialect/OpenACC/invalid.mlir

 // RUN: mlir-opt -split-input-file -verify-diagnostics %s
 
+%1 = arith.constant 1 : i32
+%2 = arith.constant 10 : i32
 // expected-error@+1 {{gang, worker or vector cannot appear with the seq attr}}
-acc.loop gang {
+acc.loop gang() (%iv : i32) = (%1 : i32) to (%2 : i32) step (%1 : i32) {
   "test.openacc_dummy_op"() : () -> ()
   acc.yield
 } attributes {seq}
 
 // -----
 
+%1 = arith.constant 1 : i32
+%2 = arith.constant 10 : i32
 // expected-error@+1 {{gang, worker or vector cannot appear with the seq attr}}
-acc.loop worker {
+acc.loop worker() (%iv : i32) = (%1 : i32) to (%2 : i32) step (%1 : i32) {
   "test.openacc_dummy_op"() : () -> ()
   acc.yield
 } attributes {seq}
 
 // -----
 
+%1 = arith.constant 1 : i32
+%2 = arith.constant 10 : i32
 // expected-error@+1 {{gang, worker or vector cannot appear with the seq attr}}
-acc.loop vector {
+acc.loop vector() (%iv : i32) = (%1 : i32) to (%2 : i32) step (%1 : i32) {
   "test.openacc_dummy_op"() : () -> ()
   acc.yield
 } attributes {seq}
 
 // -----
 
+%1 = arith.constant 1 : i32
+%2 = arith.constant 10 : i32
 // expected-error@+1 {{gang, worker or vector cannot appear with the seq attr}}
-acc.loop gang worker {
+acc.loop gang() worker() (%iv : i32) = (%1 : i32) to (%2 : i32) step (%1 : i32) {
   "test.openacc_dummy_op"() : () -> ()
   acc.yield
 } attributes {seq}
 
 // -----
 
+%1 = arith.constant 1 : i32
+%2 = arith.constant 10 : i32
 // expected-error@+1 {{gang, worker or vector cannot appear with the seq attr}}
-acc.loop gang vector {
+acc.loop gang() vector() (%iv : i32) = (%1 : i32) to (%2 : i32) step (%1 : i32) {
   "test.openacc_dummy_op"() : () -> ()
   acc.yield
 } attributes {seq}
 
 // -----
 
+%1 = arith.constant 1 : i32
+%2 = arith.constant 10 : i32
 // expected-error@+1 {{gang, worker or vector cannot appear with the seq attr}}
-acc.loop worker vector {
+acc.loop worker() vector() (%iv : i32) = (%1 : i32) to (%2 : i32) step (%1 : i32) {
   "test.openacc_dummy_op"() : () -> ()
   acc.yield
 } attributes {seq}
 
 // -----
 
+%1 = arith.constant 1 : i32
+%2 = arith.constant 10 : i32
 // expected-error@+1 {{gang, worker or vector cannot appear with the seq attr}}
-acc.loop gang worker vector {
+acc.loop gang() worker() vector() (%iv : i32) = (%1 : i32) to (%2 : i32) step (%1 : i32) {
   "test.openacc_dummy_op"() : () -> ()
   acc.yield
 } attributes {seq}
 
 // -----
 
-// expected-error@+1 {{expected non-empty body.}}
-acc.loop {
-}
-
-// -----
-
+%1 = arith.constant 1 : i32
+%2 = arith.constant 10 : i32
 // expected-error@+1 {{only one of "auto", "independent", "seq" can be present at the same time}}
-acc.loop {
+acc.loop (%iv : i32) = (%1 : i32) to (%2 : i32) step (%1 : i32) {
   acc.yield
 } attributes {auto_, seq}
 
 // -----
 
 // expected-error@+1 {{at least one operand or the default attribute must appear on the data operation}}
 acc.data {
   acc.yield

 acc.parallel {
 // expected-error@+1 {{'acc.init' op cannot be nested in a compute operation}}
   acc.init
   acc.yield
 }
 
 // -----
 
-acc.loop {
+%1 = arith.constant 1 : i32
+%2 = arith.constant 10 : i32
+acc.loop (%iv : i32) = (%1 : i32) to (%2 : i32) step (%1 : i32){
 // expected-error@+1 {{'acc.init' op cannot be nested in a compute operation}}
   acc.init
   acc.yield
 }
 
 // -----
 
 acc.parallel {
 // expected-error@+1 {{'acc.shutdown' op cannot be nested in a compute operation}}
   acc.shutdown
   acc.yield
 }
 
 // -----
 
-acc.loop {
+%1 = arith.constant 1 : i32
+%2 = arith.constant 10 : i32
+acc.loop (%iv : i32) = (%1 : i32) to (%2 : i32) step (%1 : i32) {
 // expected-error@+1 {{'acc.shutdown' op cannot be nested in a compute operation}}
   acc.shutdown
   acc.yield
 }
 
 // -----
 
-acc.loop {
+%1 = arith.constant 1 : i32
+%2 = arith.constant 10 : i32
+acc.loop (%iv : i32) = (%1 : i32) to (%2 : i32) step (%1 : i32) {
   "test.openacc_dummy_op"() ({
     // expected-error@+1 {{'acc.shutdown' op cannot be nested in a compute operation}}
     acc.shutdown
   }) : () -> ()
   acc.yield
 }
 
 // -----
^bb0(%arg0 : i32, %arg1 : !llvm.ptr<i32>):
   acc.yield
 } destroy {
 ^bb0(%arg0 : f32):
   acc.yield
 }
 
 // -----
 
+%1 = arith.constant 1 : i32
+%2 = arith.constant 10 : i32
 // expected-error@+1 {{expected ')'}}
-acc.loop gang(static=%i64Value: i64, num=%i64Value: i64 {
+acc.loop gang(static=%i64Value: i64, num=%i64Value: i64 (%iv : i32) = (%1 : i32) to (%2 : i32) step (%1 : i32) {
   "test.openacc_dummy_op"() : () -> ()
   acc.yield
 }
 
 // -----
 
 // expected-error@+1 {{expects non-empty init region}}
 acc.reduction.recipe @reduction_i64 : i64 reduction_operator<add> init {

 } combiner {
 ^bb0(%0: i64, %1: i64):
   %2 = arith.constant 0 : i32
   acc.yield %2 : i32
 }
 
 // -----
 
+%1 = arith.constant 1 : i32
+%2 = arith.constant 10 : i32
 // expected-error@+1 {{new value expected after comma}}
-acc.loop gang(static=%i64Value: i64, ) {
+acc.loop gang(static=%i64Value: i64, ) (%iv : i32) = (%1 : i32) to (%2 : i32) step (%1 : i32) {
   "test.openacc_dummy_op"() : () -> ()
   acc.yield
 }
 
 // -----
 
 func.func @fct1(%0 : !llvm.ptr<i32>) -> () {
   // expected-error@+1 {{expected symbol reference @privatization_i32 to point to a private declaration}}
   acc.serial private(@privatization_i32 -> %0 : !llvm.ptr<i32>) {
   }
   return
 }
 
 // -----
 
-// expected-error@+1 {{expect at least one of num, dim or static values}}
-acc.loop gang() {
-  "test.openacc_dummy_op"() : () -> ()
-  acc.yield
-}
-
-// -----
-
 %i64value = arith.constant 1 : i64
 // expected-error@+1 {{num_gangs expects a maximum of 3 values}}
 acc.parallel num_gangs(%i64value, %i64value, %i64value, %i64value : i64, i64, i64, i64) {
 }
 
 // -----
 
 %i64value = arith.constant 1 : i64

mlir/test/Dialect/OpenACC/ops.mlir

 // RUN: mlir-opt -split-input-file %s | FileCheck %s
 // Verify the printed output can be parsed.
 // RUN: mlir-opt -split-input-file %s | mlir-opt  -split-input-file  | FileCheck %s
 // Verify the generic form can be parsed.
 // RUN: mlir-opt -split-input-file -mlir-print-op-generic %s | mlir-opt -split-input-file | FileCheck %s
 
 func.func @compute1(%A: memref<10x10xf32>, %B: memref<10x10xf32>, %C: memref<10x10xf32>) -> memref<10x10xf32> {
   %c0 = arith.constant 0 : index
   %c10 = arith.constant 10 : index
   %c1 = arith.constant 1 : index
   %async = arith.constant 1 : i64
 
   acc.parallel async(%async: i64) {
-    acc.loop gang vector {
-      scf.for %arg3 = %c0 to %c10 step %c1 {
-        scf.for %arg4 = %c0 to %c10 step %c1 {
-          scf.for %arg5 = %c0 to %c10 step %c1 {
-            %a = memref.load %A[%arg3, %arg5] : memref<10x10xf32>
-            %b = memref.load %B[%arg5, %arg4] : memref<10x10xf32>
-            %cij = memref.load %C[%arg3, %arg4] : memref<10x10xf32>
-            %p = arith.mulf %a, %b : f32
-            %co = arith.addf %cij, %p : f32
-            memref.store %co, %C[%arg3, %arg4] : memref<10x10xf32>
-          }
-        }
-      }
+    acc.loop gang() vector() (%arg3 : index) = (%c0 : index) to (%c10 : index) step (%c1 : index) {
       acc.yield
-    } attributes { collapse = 3 }
+    } attributes { collapse = 3, inclusiveUpperbound}
     acc.yield
   }
 
   return %C : memref<10x10xf32>
 }
 
 // CHECK-LABEL: func @compute1(
 //  CHECK-NEXT:   %{{.*}} = arith.constant 0 : index
 //  CHECK-NEXT:   %{{.*}} = arith.constant 10 : index
 //  CHECK-NEXT:   %{{.*}} = arith.constant 1 : index
 //  CHECK-NEXT:   [[ASYNC:%.*]] = arith.constant 1 : i64
 //  CHECK-NEXT:   acc.parallel async([[ASYNC]] : i64) {
-//  CHECK-NEXT:     acc.loop gang vector {
-//  CHECK-NEXT:       scf.for %{{.*}} = %{{.*}} to %{{.*}} step %{{.*}} {
-//  CHECK-NEXT:         scf.for %{{.*}} = %{{.*}} to %{{.*}} step %{{.*}} {
-//  CHECK-NEXT:           scf.for %{{.*}} = %{{.*}} to %{{.*}} step %{{.*}} {
-//  CHECK-NEXT:             %{{.*}} = memref.load %{{.*}}[%{{.*}}, %{{.*}}] : memref<10x10xf32>
-//  CHECK-NEXT:             %{{.*}} = memref.load %{{.*}}[%{{.*}}, %{{.*}}] : memref<10x10xf32>
-//  CHECK-NEXT:             %{{.*}} = memref.load %{{.*}}[%{{.*}}, %{{.*}}] : memref<10x10xf32>
-//  CHECK-NEXT:             %{{.*}} = arith.mulf %{{.*}}, %{{.*}} : f32
-//  CHECK-NEXT:             %{{.*}} = arith.addf %{{.*}}, %{{.*}} : f32
-//  CHECK-NEXT:             memref.store %{{.*}}, %{{.*}}[%{{.*}}, %{{.*}}] : memref<10x10xf32>
-//  CHECK-NEXT:           }
-//  CHECK-NEXT:         }
-//  CHECK-NEXT:       }
+//  CHECK-NEXT:     acc.loop gang() vector() (%{{.*}}) = (%{{.*}}) to (%{{.*}}) step (%{{.*}})
+//  CHECK-NEXT:       scf.for %{{.*}} = %{{.*}} to %{{.*}} step %{{.*}} {
+//  CHECK-NEXT:         scf.for %{{.*}} = %{{.*}} to %{{.*}} step %{{.*}} {
+//  CHECK-NEXT:           %{{.*}} = memref.load %{{.*}}[%{{.*}}, %{{.*}}] : memref<10x10xf32>
+//  CHECK-NEXT:           %{{.*}} = memref.load %{{.*}}[%{{.*}}, %{{.*}}] : memref<10x10xf32>
+//  CHECK-NEXT:           %{{.*}} = memref.load %{{.*}}[%{{.*}}, %{{.*}}] : memref<10x10xf32>
+//  CHECK-NEXT:           %{{.*}} = arith.mulf %{{.*}}, %{{.*}} : f32
+//  CHECK-NEXT:           %{{.*}} = arith.addf %{{.*}}, %{{.*}} : f32
+//  CHECK-NEXT:           memref.store %{{.*}}, %{{.*}}[%{{.*}}, %{{.*}}] : memref<10x10xf32>
+//  CHECK-NEXT:         }
+//  CHECK-NEXT:       }
 //  CHECK-NEXT:       acc.yield
-//  CHECK-NEXT:     } attributes {collapse = 3 : i64}
+//  CHECK-NEXT:     } attributes {collapse = 3 : i64, inclusiveUpperbound}
 //  CHECK-NEXT:     acc.yield
 //  CHECK-NEXT:   }
 //  CHECK-NEXT:   return %{{.*}} : memref<10x10xf32>
 //  CHECK-NEXT: }
 
 // -----
 
 func.func @compute2(%A: memref<10x10xf32>, %B: memref<10x10xf32>, %C: memref<10x10xf32>) -> memref<10x10xf32> {
   %c0 = arith.constant 0 : index
   %c10 = arith.constant 10 : index
   %c1 = arith.constant 1 : index
 
   acc.parallel {
-    acc.loop {
-      scf.for %arg3 = %c0 to %c10 step %c1 {
-        scf.for %arg4 = %c0 to %c10 step %c1 {
-          scf.for %arg5 = %c0 to %c10 step %c1 {
-            %a = memref.load %A[%arg3, %arg5] : memref<10x10xf32>
-            %b = memref.load %B[%arg5, %arg4] : memref<10x10xf32>
-            %cij = memref.load %C[%arg3, %arg4] : memref<10x10xf32>
-            %p = arith.mulf %a, %b : f32
-            %co = arith.addf %cij, %p : f32
-            memref.store %co, %C[%arg3, %arg4] : memref<10x10xf32>
-          }
-        }
-      }
+    acc.loop (%arg3 : index) = (%c0 : index) to (%c10 : index) step (%c1 : index) {
       acc.yield
     } attributes {seq}
     acc.yield
   }
 
   return %C : memref<10x10xf32>
 }
 
 // CHECK-LABEL: func @compute2(
 //  CHECK-NEXT:   %{{.*}} = arith.constant 0 : index
 //  CHECK-NEXT:   %{{.*}} = arith.constant 10 : index
 //  CHECK-NEXT:   %{{.*}} = arith.constant 1 : index
 //  CHECK-NEXT:   acc.parallel {
-//  CHECK-NEXT:     acc.loop {
-//  CHECK-NEXT:       scf.for %{{.*}} = %{{.*}} to %{{.*}} step %{{.*}} {
+//  CHECK-NEXT:     acc.loop  (%{{.*}}) = (%{{.*}}) to (%{{.*}}) step (%{{.*}})
 //  CHECK-NEXT:         scf.for %{{.*}} = %{{.*}} to %{{.*}} step %{{.*}} {
 //  CHECK-NEXT:           scf.for %{{.*}} = %{{.*}} to %{{.*}} step %{{.*}} {
 //  CHECK-NEXT:             %{{.*}} = memref.load %{{.*}}[%{{.*}}, %{{.*}}] : memref<10x10xf32>
 //  CHECK-NEXT:             %{{.*}} = memref.load %{{.*}}[%{{.*}}, %{{.*}}] : memref<10x10xf32>
 //  CHECK-NEXT:             %{{.*}} = memref.load %{{.*}}[%{{.*}}, %{{.*}}] : memref<10x10xf32>
 //  CHECK-NEXT:             %{{.*}} = arith.mulf %{{.*}}, %{{.*}} : f32
 //  CHECK-NEXT:             %{{.*}} = arith.addf %{{.*}}, %{{.*}} : f32
 //  CHECK-NEXT:             memref.store %{{.*}}, %{{.*}}[%{{.*}}, %{{.*}}] : memref<10x10xf32>
 //  CHECK-NEXT:           }
 //  CHECK-NEXT:         }
-//  CHECK-NEXT:       }
 //  CHECK-NEXT:       acc.yield
 //  CHECK-NEXT:     } attributes {seq}
 //  CHECK-NEXT:     acc.yield
 //  CHECK-NEXT:   }
 //  CHECK-NEXT:   return %{{.*}} : memref<10x10xf32>
 //  CHECK-NEXT: }
 
 // -----
func.func @compute3(%a: memref<10x10xf32>, %b: memref<10x10xf32>, %c: memref<10xf32>, %d: memref<10xf32>) -> memref<10xf32> {
 
   %pa = acc.present varPtr(%a : memref<10x10xf32>) -> memref<10x10xf32>
   %pb = acc.present varPtr(%b : memref<10x10xf32>) -> memref<10x10xf32>
   %pc = acc.present varPtr(%c : memref<10xf32>) -> memref<10xf32>
   %pd = acc.present varPtr(%d : memref<10xf32>) -> memref<10xf32>
   acc.data dataOperands(%pa, %pb, %pc, %pd: memref<10x10xf32>, memref<10x10xf32>, memref<10xf32>, memref<10xf32>) {
     %private = acc.private varPtr(%c : memref<10xf32>) -> memref<10xf32>
     acc.parallel num_gangs(%numGangs: i64) num_workers(%numWorkers: i64) private(@privatization_memref_10_f32 -> %private : memref<10xf32>) {
-      acc.loop gang {
-        scf.for %x = %lb to %c10 step %st {
-          acc.loop worker {
-            scf.for %y = %lb to %c10 step %st {
-              %axy = memref.load %a[%x, %y] : memref<10x10xf32>
-              %bxy = memref.load %b[%x, %y] : memref<10x10xf32>
-              %tmp = arith.addf %axy, %bxy : f32
-              memref.store %tmp, %c[%y] : memref<10xf32>
-            }
-            acc.yield
-          }
-
-          acc.loop {
-            // for i = 0 to 10 step 1
-            //   d[x] += c[i]
-            scf.for %i = %lb to %c10 step %st {
-              %ci = memref.load %c[%i] : memref<10xf32>
-              %dx = memref.load %d[%x] : memref<10xf32>
-              %z = arith.addf %ci, %dx : f32
-              memref.store %z, %d[%x] : memref<10xf32>
-            }
-            acc.yield
-          } attributes {seq}
-        }
+      acc.loop gang() (%x : index) = (%lb : index) to (%c10 : index) step (%st : index) {
+        acc.loop worker() (%y : index) = (%lb : index) to (%c10 : index) step (%st : index) {
+        acc.loop (%i : index) = (%lb : index) to (%c10 : index) step (%st : index) {
         acc.yield
       }
       acc.yield
     }
     acc.terminator
   }
 
   return %d : memref<10xf32>
 }
 
 // CHECK:      func @compute3({{.*}}: memref<10x10xf32>, {{.*}}: memref<10x10xf32>, [[ARG2:%.*]]: memref<10xf32>, {{.*}}: memref<10xf32>) -> memref<10xf32> {
 // CHECK-NEXT:   [[C0:%.*]] = arith.constant 0 : index
 // CHECK-NEXT:   [[C1:%.*]] = arith.constant 1 : index
 // CHECK-NEXT:   [[C10:%.*]] = arith.constant 10 : index
 // CHECK-NEXT:   [[NUMGANG:%.*]] = arith.constant 10 : i64
 // CHECK-NEXT:   [[NUMWORKERS:%.*]] = arith.constant 10 : i64
 // CHECK:        acc.data dataOperands(%{{.*}}, %{{.*}}, %{{.*}}, %{{.*}} : memref<10x10xf32>, memref<10x10xf32>, memref<10xf32>, memref<10xf32>) {
 // CHECK-NEXT:     %[[P_ARG2:.*]] = acc.private varPtr([[ARG2]] : memref<10xf32>) -> memref<10xf32> 
 // CHECK-NEXT:     acc.parallel num_gangs([[NUMGANG]] : i64) num_workers([[NUMWORKERS]] : i64) private(@privatization_memref_10_f32 -> %[[P_ARG2]] : memref<10xf32>) {
-// CHECK-NEXT:       acc.loop gang {
-// CHECK-NEXT:         scf.for %{{.*}} = [[C0]] to [[C10]] step [[C1]] {
-// CHECK-NEXT:           acc.loop worker {
-// CHECK-NEXT:             scf.for %{{.*}} = [[C0]] to [[C10]] step [[C1]] {
+// CHECK-NEXT:       acc.loop gang() (%{{.*}}) = (%{{.*}}) to (%{{.*}}) step (%{{.*}})
+// CHECK-NEXT:           acc.loop worker() (%{{.*}}) = (%{{.*}}) to (%{{.*}}) step (%{{.*}})
 // CHECK-NEXT:               %{{.*}} = memref.load %{{.*}}[%{{.*}}, %{{.*}}] : memref<10x10xf32>
 // CHECK-NEXT:               %{{.*}} = memref.load %{{.*}}[%{{.*}}, %{{.*}}] : memref<10x10xf32>
 // CHECK-NEXT:               %{{.*}} = arith.addf %{{.*}}, %{{.*}} : f32
 // CHECK-NEXT:               memref.store %{{.*}}, %{{.*}}[%{{.*}}] : memref<10xf32>
-// CHECK-NEXT:             }
 // CHECK-NEXT:             acc.yield
 // CHECK-NEXT:           }
-// CHECK-NEXT:           acc.loop {
-// CHECK-NEXT:             scf.for %{{.*}} = [[C0]] to [[C10]] step [[C1]] {
+// CHECK-NEXT:           acc.loop (%{{.*}}) = (%{{.*}}) to (%{{.*}}) step (%{{.*}})
 // CHECK-NEXT:               %{{.*}} = memref.load %{{.*}}[%{{.*}}] : memref<10xf32>
 // CHECK-NEXT:               %{{.*}} = memref.load %{{.*}}[%{{.*}}] : memref<10xf32>
 // CHECK-NEXT:               %{{.*}} = arith.addf %{{.*}}, %{{.*}} : f32
 // CHECK-NEXT:               memref.store %{{.*}}, %{{.*}}[%{{.*}}] : memref<10xf32>
-// CHECK-NEXT:             }
 // CHECK-NEXT:             acc.yield
 // CHECK-NEXT:           } attributes {seq}
-// CHECK-NEXT:         }
 // CHECK-NEXT:         acc.yield
 // CHECK-NEXT:       }
 // CHECK-NEXT:       acc.yield
 // CHECK-NEXT:     }
 // CHECK-NEXT:     acc.terminator
 // CHECK-NEXT:   }
 // CHECK-NEXT:   return %{{.*}} : memref<10xf32>
 // CHECK-NEXT: }
 
 // -----
 
 func.func @testloopop() -> () {
   %i64Value = arith.constant 1 : i64
   %i32Value = arith.constant 128 : i32
   %idxValue = arith.constant 8 : index
+  %c0 = arith.constant 0 : index
+  %c10 = arith.constant 10 : index
+  %c1 = arith.constant 1 : index
 
-  acc.loop gang worker vector {
+  acc.loop gang() worker() vector() (%iv : index) = (%c0 : index) to (%c10 : index) step (%c1 : index) {
     "test.openacc_dummy_op"() : () -> ()
     acc.yield
   }
-  acc.loop gang(num=%i64Value: i64) {
+  acc.loop gang(num=%i64Value: i64) (%iv : index) = (%c0 : index) to (%c10 : index) step (%c1 : index) {
     "test.openacc_dummy_op"() : () -> ()
     acc.yield
   }
-  acc.loop gang(static=%i64Value: i64) {
+  acc.loop gang(static=%i64Value: i64) (%iv : index) = (%c0 : index) to (%c10 : index) step (%c1 : index) {
     "test.openacc_dummy_op"() : () -> ()
     acc.yield
   }
-  acc.loop worker(%i64Value: i64) {
+  acc.loop worker(%i64Value: i64) (%iv : index) = (%c0 : index) to (%c10 : index) step (%c1 : index) {
     "test.openacc_dummy_op"() : () -> ()
     acc.yield
   }
-  acc.loop worker(%i32Value: i32) {
+  acc.loop worker(%i32Value: i32) (%iv : index) = (%c0 : index) to (%c10 : index) step (%c1 : index) {
     "test.openacc_dummy_op"() : () -> ()
     acc.yield
   }
-  acc.loop worker(%idxValue: index) {
+  acc.loop worker(%idxValue: index) (%iv : index) = (%c0 : index) to (%c10 : index) step (%c1 : index) {
     "test.openacc_dummy_op"() : () -> ()
     acc.yield
   }
-  acc.loop vector(%i64Value: i64) {
+  acc.loop vector(%i64Value: i64) (%iv : index) = (%c0 : index) to (%c10 : index) step (%c1 : index) {
     "test.openacc_dummy_op"() : () -> ()
     acc.yield
   }
-  acc.loop vector(%i32Value: i32) {
+  acc.loop vector(%i32Value: i32) (%iv : index) = (%c0 : index) to (%c10 : index) step (%c1 : index) {
     "test.openacc_dummy_op"() : () -> ()
     acc.yield
   }
-  acc.loop vector(%idxValue: index) {
+  acc.loop vector(%idxValue: index) (%iv : index) = (%c0 : index) to (%c10 : index) step (%c1 : index) {
     "test.openacc_dummy_op"() : () -> ()
     acc.yield
   }
-  acc.loop gang(num=%i64Value: i64) worker vector {
+  acc.loop gang(num=%i64Value: i64) worker() vector() (%iv : index) = (%c0 : index) to (%c10 : index) step (%c1 : index) {
     "test.openacc_dummy_op"() : () -> ()
     acc.yield
   }
-  acc.loop gang(num=%i64Value: i64, static=%i64Value: i64) worker(%i64Value: i64) vector(%i64Value: i64) {
+  acc.loop gang(num=%i64Value: i64, static=%i64Value: i64) worker(%i64Value: i64) vector(%i64Value: i64) (%iv : index) = (%c0 : index) to (%c10 : index) step (%c1 : index) {
     "test.openacc_dummy_op"() : () -> ()
     acc.yield
   }
-  acc.loop gang(num=%i32Value: i32, static=%idxValue: index) {
+  acc.loop gang(num=%i32Value: i32, static=%idxValue: index) (%iv : index) = (%c0 : index) to (%c10 : index) step (%c1 : index) {
     "test.openacc_dummy_op"() : () -> ()
     acc.yield
   }
-  acc.loop tile(%i64Value, %i64Value : i64, i64) {
+  acc.loop tile(%i64Value, %i64Value : i64, i64) (%iv : index) = (%c0 : index) to (%c10 : index) step (%c1 : index) {
     "test.openacc_dummy_op"() : () -> ()
     acc.yield
   }
-  acc.loop tile(%i32Value, %i32Value : i32, i32) {
+  acc.loop tile(%i32Value, %i32Value : i32, i32) (%iv : index) = (%c0 : index) to (%c10 : index) step (%c1 : index) {
     "test.openacc_dummy_op"() : () -> ()
     acc.yield
   }
-  acc.loop gang(static=%i64Value: i64, num=%i64Value: i64) {
+  acc.loop gang(static=%i64Value: i64, num=%i64Value: i64) (%iv : index) = (%c0 : index) to (%c10 : index) step (%c1 : index) {
     "test.openacc_dummy_op"() : () -> ()
     acc.yield
   }
-  acc.loop gang(dim=%i64Value : i64, static=%i64Value: i64) {
+  acc.loop gang(dim=%i64Value : i64, static=%i64Value: i64) (%iv : index) = (%c0 : index) to (%c10 : index) step (%c1 : index) {
     "test.openacc_dummy_op"() : () -> ()
     acc.yield
   }
   return
 }
 
 // CHECK:      [[I64VALUE:%.*]] = arith.constant 1 : i64
 // CHECK-NEXT: [[I32VALUE:%.*]] = arith.constant 128 : i32
 // CHECK-NEXT: [[IDXVALUE:%.*]] = arith.constant 8 : index
-// CHECK:      acc.loop gang worker vector {
+// CHECK:      acc.loop gang() worker() vector()
 // CHECK-NEXT:   "test.openacc_dummy_op"() : () -> ()
 // CHECK-NEXT:   acc.yield
-// CHECK-NEXT: }
-// CHECK:      acc.loop gang(num=[[I64VALUE]] : i64) {
+// CHECK:      acc.loop gang(num=[[I64VALUE]] : i64)
 // CHECK-NEXT:   "test.openacc_dummy_op"() : () -> ()
 // CHECK-NEXT:   acc.yield
-// CHECK-NEXT: }
-// CHECK:      acc.loop gang(static=[[I64VALUE]] : i64) {
+// CHECK:      acc.loop gang(static=[[I64VALUE]] : i64)
 // CHECK-NEXT:   "test.openacc_dummy_op"() : () -> ()
 // CHECK-NEXT:   acc.yield
-// CHECK-NEXT: }
-// CHECK:      acc.loop worker([[I64VALUE]] : i64) {
+// CHECK:      acc.loop worker([[I64VALUE]] : i64)
 // CHECK-NEXT:   "test.openacc_dummy_op"() : () -> ()
 // CHECK-NEXT:   acc.yield
-// CHECK-NEXT: }
-// CHECK:      acc.loop worker([[I32VALUE]] : i32) {
+// CHECK:      acc.loop worker([[I32VALUE]] : i32)
 // CHECK-NEXT:   "test.openacc_dummy_op"() : () -> ()
 // CHECK-NEXT:   acc.yield
-// CHECK-NEXT: }
-// CHECK:      acc.loop worker([[IDXVALUE]] : index) {
+// CHECK:      acc.loop worker([[IDXVALUE]] : index)
 // CHECK-NEXT:   "test.openacc_dummy_op"() : () -> ()
 // CHECK-NEXT:   acc.yield
-// CHECK-NEXT: }
-// CHECK:      acc.loop vector([[I64VALUE]] : i64) {
+// CHECK:      acc.loop vector([[I64VALUE]] : i64)
 // CHECK-NEXT:   "test.openacc_dummy_op"() : () -> ()
 // CHECK-NEXT:   acc.yield
-// CHECK-NEXT: }
-// CHECK:      acc.loop vector([[I32VALUE]] : i32) {
+// CHECK:      acc.loop vector([[I32VALUE]] : i32)
 // CHECK-NEXT:   "test.openacc_dummy_op"() : () -> ()
 // CHECK-NEXT:   acc.yield
-// CHECK-NEXT: }
-// CHECK:      acc.loop vector([[IDXVALUE]] : index) {
+// CHECK:      acc.loop vector([[IDXVALUE]] : index)
 // CHECK-NEXT:   "test.openacc_dummy_op"() : () -> ()
 // CHECK-NEXT:   acc.yield
-// CHECK-NEXT: }
-// CHECK:      acc.loop gang(num=[[I64VALUE]] : i64) worker vector {
+// CHECK:      acc.loop gang(num=[[I64VALUE]] : i64) worker() vector()
 // CHECK-NEXT:   "test.openacc_dummy_op"() : () -> ()
 // CHECK-NEXT:   acc.yield
-// CHECK-NEXT: }
-// CHECK:      acc.loop gang(num=[[I64VALUE]] : i64, static=[[I64VALUE]] : i64) worker([[I64VALUE]] : i64) vector([[I64VALUE]] : i64) {
+// CHECK:      acc.loop gang(num=[[I64VALUE]] : i64, static=[[I64VALUE]] : i64) worker([[I64VALUE]] : i64) vector([[I64VALUE]] : i64)
 // CHECK-NEXT:   "test.openacc_dummy_op"() : () -> ()
 // CHECK-NEXT:   acc.yield
-// CHECK-NEXT: }
-// CHECK:      acc.loop gang(num=[[I32VALUE]] : i32, static=[[IDXVALUE]] : index) {
+// CHECK:      acc.loop gang(num=[[I32VALUE]] : i32, static=[[IDXVALUE]] : index)
 // CHECK-NEXT:   "test.openacc_dummy_op"() : () -> ()
 // CHECK-NEXT:   acc.yield
-// CHECK-NEXT: }
-// CHECK:      acc.loop tile([[I64VALUE]], [[I64VALUE]] : i64, i64) {
+// CHECK:      acc.loop tile([[I64VALUE]], [[I64VALUE]] : i64, i64)
 // CHECK-NEXT:   "test.openacc_dummy_op"() : () -> ()
 // CHECK-NEXT:   acc.yield
-// CHECK-NEXT: }
-// CHECK:      acc.loop tile([[I32VALUE]], [[I32VALUE]] : i32, i32) {
+// CHECK:      acc.loop tile([[I32VALUE]], [[I32VALUE]] : i32, i32)
 // CHECK-NEXT:   "test.openacc_dummy_op"() : () -> ()
 // CHECK-NEXT:   acc.yield
-// CHECK-NEXT: }
-// CHECK:      acc.loop gang(num=[[I64VALUE]] : i64, static=[[I64VALUE]] : i64) {
+// CHECK:      acc.loop gang(num=[[I64VALUE]] : i64, static=[[I64VALUE]] : i64)
 // CHECK-NEXT:   "test.openacc_dummy_op"() : () -> ()
 // CHECK-NEXT:   acc.yield
-// CHECK-NEXT: }
-// CHECK:      acc.loop gang(dim=[[I64VALUE]] : i64, static=[[I64VALUE]] : i64) {
+// CHECK:      acc.loop gang(dim=[[I64VALUE]] : i64, static=[[I64VALUE]] : i64)
 // CHECK-NEXT:   "test.openacc_dummy_op"() : () -> ()
 // CHECK-NEXT:   acc.yield
-// CHECK-NEXT: }
 
 // -----
 
 func.func @acc_loop_multiple_block() {
-  acc.parallel {
-    acc.loop {
-      %c1 = arith.constant 1 : index
-      cf.br ^bb1(%c1 : index)
+  %c0 = arith.constant 0 : index
+  %c10 = arith.constant 10 : index
+  acc.parallel {
+    acc.loop  (%iv : index) = (%c0 : index) to (%c10 : index) step (%c1 : index) {
+      %c1_1 = arith.constant 1 : index
+      cf.br ^bb1(%c1_1 : index)
     ^bb1(%9: index):
-      %c0 = arith.constant 0 : index
-      %12 = arith.cmpi sgt, %9, %c0 : index
+      %c0_1 = arith.constant 0 : index
+      %12 = arith.cmpi sgt, %9, %c0_1 : index
       cf.cond_br %12, ^bb2, ^bb3
     ^bb2:
       %c1_0 = arith.constant 1 : index
-      %c10 = arith.constant 10 : index
-      %22 = arith.subi %c10, %c1_0 : index
+      %c10_1 = arith.constant 10 : index
+      %22 = arith.subi %c10_1, %c1_0 : index
       cf.br ^bb1(%22 : index)
     ^bb3:
       acc.yield
     }
     acc.yield
   }
   return
 }

 // CHECK:         acc.yield %[[C0]] : i64
 // CHECK:       } combiner {
 // CHECK:       ^bb0(%[[ARG0:.*]]: i64, %[[ARG1:.*]]: i64):
 // CHECK:         %[[RES:.*]] = arith.addi %[[ARG0]], %[[ARG1]] : i64
 // CHECK:         acc.yield %[[RES]] : i64
 // CHECK:       }
 
 func.func @acc_reduc_test(%a : i64) -> () {
+  %c0 = arith.constant 0 : index
+  %c10 = arith.constant 10 : index
+  %c1 = arith.constant 1 : index
   acc.parallel reduction(@reduction_add_i64 -> %a : i64) {
-    acc.loop reduction(@reduction_add_i64 -> %a : i64) {
+    acc.loop reduction(@reduction_add_i64 -> %a : i64) (%iv : index) = (%c0 : index) to (%c10 : index) step (%c1 : index) {
       acc.yield
     }
     acc.yield
   }
   return
 }
 
 // CHECK-LABEL: func.func @acc_reduc_test(