[MLIR][OpenMP] Make omp.taskloop into a loop wrapper (llvm#87253)

This patch updates the definition of `omp.taskloop` to enforce the
restrictions of a wrapper operation.

mlir/include/mlir/Dialect/OpenMP/OpenMPClauseOperands.h

@@ -284,11 +284,10 @@ using TaskgroupClauseOps =
     detail::Clauses<AllocateClauseOps, TaskReductionClauseOps>;
 
 using TaskloopClauseOps =
-    detail::Clauses<AllocateClauseOps, CollapseClauseOps, FinalClauseOps,
-                    GrainsizeClauseOps, IfClauseOps, InReductionClauseOps,
-                    LoopRelatedOps, MergeableClauseOps, NogroupClauseOps,
-                    NumTasksClauseOps, PriorityClauseOps, PrivateClauseOps,
-                    ReductionClauseOps, UntiedClauseOps>;
+    detail::Clauses<AllocateClauseOps, FinalClauseOps, GrainsizeClauseOps,
+                    IfClauseOps, InReductionClauseOps, MergeableClauseOps,
+                    NogroupClauseOps, NumTasksClauseOps, PriorityClauseOps,
+                    PrivateClauseOps, ReductionClauseOps, UntiedClauseOps>;
 
 using TaskwaitClauseOps = detail::Clauses<DependClauseOps, NowaitClauseOps>;
 

mlir/include/mlir/Dialect/OpenMP/OpenMPOps.td

@@ -1030,32 +1030,30 @@ def TaskOp : OpenMP_Op<"task", [AttrSizedOperandSegments,
 }
 
 def TaskloopOp : OpenMP_Op<"taskloop", [AttrSizedOperandSegments,
-                           AutomaticAllocationScope, RecursiveMemoryEffects,
-                           AllTypesMatch<["lowerBound", "upperBound", "step"]>,
+                           AutomaticAllocationScope,
                            DeclareOpInterfaceMethods<LoopWrapperInterface>,
-                           ReductionClauseInterface]> {
+                           RecursiveMemoryEffects, ReductionClauseInterface,
+                           SingleBlockImplicitTerminator<"TerminatorOp">]> {
   let summary = "taskloop construct";
   let description = [{
     The taskloop construct specifies that the iterations of one or more
     associated loops will be executed in parallel using explicit tasks. The
     iterations are distributed across tasks generated by the construct and
     scheduled to be executed.
 
-    The `lowerBound` and `upperBound` specify a half-open range: the range
-    includes the lower bound but does not include the upper bound. If the
-    `inclusive` attribute is specified then the upper bound is also included.
-    The `step` specifies the loop step.
-
-    The body region can contain any number of blocks.
+    The body region can contain a single block which must contain a single
+    operation and a terminator. The operation must be another compatible loop
+    wrapper or an `omp.loop_nest`.
 
     ```
-    omp.taskloop <clauses>
-    for (%i1, %i2) : index = (%c0, %c0) to (%c10, %c10) step (%c1, %c1) {
-      %a = load %arrA[%i1, %i2] : memref<?x?xf32>
-      %b = load %arrB[%i1, %i2] : memref<?x?xf32>
-      %sum = arith.addf %a, %b : f32
-      store %sum, %arrC[%i1, %i2] : memref<?x?xf32>
-      omp.terminator
+    omp.taskloop <clauses> {
+      omp.loop_nest (%i1, %i2) : index = (%c0, %c0) to (%c10, %c10) step (%c1, %c1) {
+        %a = load %arrA[%i1, %i2] : memref<?x?xf32>
+        %b = load %arrB[%i1, %i2] : memref<?x?xf32>
+        %sum = arith.addf %a, %b : f32
+        store %sum, %arrC[%i1, %i2] : memref<?x?xf32>
+        omp.yield
+      }
     }
     ```
 
@@ -1132,11 +1130,7 @@ def TaskloopOp : OpenMP_Op<"taskloop", [AttrSizedOperandSegments,
     created.
   }];
 
-  let arguments = (ins Variadic<IntLikeType>:$lowerBound,
-                       Variadic<IntLikeType>:$upperBound,
-                       Variadic<IntLikeType>:$step,
-                       UnitAttr:$inclusive,
-                       Optional<I1>:$if_expr,
+  let arguments = (ins Optional<I1>:$if_expr,
                        Optional<I1>:$final_expr,
                        UnitAttr:$untied,
                        UnitAttr:$mergeable,
@@ -1179,8 +1173,7 @@ def TaskloopOp : OpenMP_Op<"taskloop", [AttrSizedOperandSegments,
           |`grain_size` `(` $grain_size `:` type($grain_size) `)`
           |`num_tasks` `(` $num_tasks `:` type($num_tasks) `)`
           |`nogroup` $nogroup
-    ) `for` custom<LoopControl>($region, $lowerBound, $upperBound, $step,
-                                  type($step), $inclusive) attr-dict
+    ) $region attr-dict
   }];
 
   let extraClassDeclaration = [{

mlir/lib/Dialect/OpenMP/IR/OpenMPDialect.cpp

@@ -1829,9 +1829,8 @@ void TaskloopOp::build(OpBuilder &builder, OperationState &state,
   MLIRContext *ctx = builder.getContext();
   // TODO Store clauses in op: reductionByRefAttr, privateVars, privatizers.
   TaskloopOp::build(
-      builder, state, clauses.loopLBVar, clauses.loopUBVar, clauses.loopStepVar,
-      clauses.loopInclusiveAttr, clauses.ifVar, clauses.finalVar,
-      clauses.untiedAttr, clauses.mergeableAttr, clauses.inReductionVars,
+      builder, state, clauses.ifVar, clauses.finalVar, clauses.untiedAttr,
+      clauses.mergeableAttr, clauses.inReductionVars,
       makeArrayAttr(ctx, clauses.inReductionDeclSymbols), clauses.reductionVars,
       makeArrayAttr(ctx, clauses.reductionDeclSymbols), clauses.priorityVar,
       clauses.allocateVars, clauses.allocatorVars, clauses.grainsizeVar,
@@ -1870,6 +1869,16 @@ LogicalResult TaskloopOp::verify() {
         "the grainsize clause and num_tasks clause are mutually exclusive and "
         "may not appear on the same taskloop directive");
   }
+
+  if (!isWrapper())
+    return emitOpError() << "must be a loop wrapper";
+
+  if (LoopWrapperInterface nested = getNestedWrapper()) {
+    // Check for the allowed leaf constructs that may appear in a composite
+    // construct directly after TASKLOOP.
+    if (!isa<SimdLoopOp>(nested))
+      return emitError() << "only supported nested wrapper is 'omp.simdloop'";
+  }
   return success();
 }
 

mlir/test/Dialect/OpenMP/invalid.mlir

@@ -1580,10 +1580,11 @@ func.func @omp_cancellationpoint2() {
 func.func @taskloop(%lb: i32, %ub: i32, %step: i32) {
   %testmemref = "test.memref"() : () -> (memref<i32>)
   // expected-error @below {{expected equal sizes for allocate and allocator variables}}
-  "omp.taskloop"(%lb, %ub, %ub, %lb, %step, %step, %testmemref) ({
-  ^bb0(%arg3: i32, %arg4: i32):
-    "omp.terminator"() : () -> ()
-  }) {operandSegmentSizes = array<i32: 2, 2, 2, 0, 0, 0, 0, 0, 1, 0, 0, 0>} : (i32, i32, i32, i32, i32, i32, memref<i32>) -> ()
+  "omp.taskloop"(%testmemref) ({
+    omp.loop_nest (%i, %j) : i32 = (%lb, %ub) to (%ub, %lb) step (%step, %step) {
+      omp.yield
+    }
+  }) {operandSegmentSizes = array<i32: 0, 0, 0, 0, 0, 1, 0, 0, 0>} : (memref<i32>) -> ()
   return
 }
 
@@ -1593,23 +1594,24 @@ func.func @taskloop(%lb: i32, %ub: i32, %step: i32) {
   %testf32 = "test.f32"() : () -> (!llvm.ptr)
   %testf32_2 = "test.f32"() : () -> (!llvm.ptr)
   // expected-error @below {{expected as many reduction symbol references as reduction variables}}
-  "omp.taskloop"(%lb, %ub, %ub, %lb, %step, %step, %testf32, %testf32_2) ({
-  ^bb0(%arg3: i32, %arg4: i32):
-    "omp.terminator"() : () -> ()
-  }) {operandSegmentSizes = array<i32: 2, 2, 2, 0, 0, 0, 2, 0, 0, 0, 0, 0>, reductions = [@add_f32]} : (i32, i32, i32, i32, i32, i32, !llvm.ptr, !llvm.ptr) -> ()
+  "omp.taskloop"(%testf32, %testf32_2) ({
+    omp.loop_nest (%i, %j) : i32 = (%lb, %ub) to (%ub, %lb) step (%step, %step) {
+      omp.yield
+    }
+  }) {operandSegmentSizes = array<i32: 0, 0, 0, 2, 0, 0, 0, 0, 0>, reductions = [@add_f32]} : (!llvm.ptr, !llvm.ptr) -> ()
   return
 }
 
 // -----
 
 func.func @taskloop(%lb: i32, %ub: i32, %step: i32) {
   %testf32 = "test.f32"() : () -> (!llvm.ptr)
-  %testf32_2 = "test.f32"() : () -> (!llvm.ptr)
   // expected-error @below {{expected as many reduction symbol references as reduction variables}}
-  "omp.taskloop"(%lb, %ub, %ub, %lb, %step, %step, %testf32) ({
-  ^bb0(%arg3: i32, %arg4: i32):
-    "omp.terminator"() : () -> ()
-  }) {operandSegmentSizes = array<i32: 2, 2, 2, 0, 0, 0, 1, 0, 0, 0, 0, 0>, reductions = [@add_f32, @add_f32]} : (i32, i32, i32, i32, i32, i32, !llvm.ptr) -> ()
+  "omp.taskloop"(%testf32) ({
+    omp.loop_nest (%i, %j) : i32 = (%lb, %ub) to (%ub, %lb) step (%step, %step) {
+      omp.yield
+    }
+  }) {operandSegmentSizes = array<i32: 0, 0, 0, 1, 0, 0, 0, 0, 0>, reductions = [@add_f32, @add_f32]} : (!llvm.ptr) -> ()
   return
 }
 
@@ -1619,23 +1621,24 @@ func.func @taskloop(%lb: i32, %ub: i32, %step: i32) {
   %testf32 = "test.f32"() : () -> (!llvm.ptr)
   %testf32_2 = "test.f32"() : () -> (!llvm.ptr)
   // expected-error @below {{expected as many reduction symbol references as reduction variables}}
-  "omp.taskloop"(%lb, %ub, %ub, %lb, %step, %step, %testf32, %testf32_2) ({
-  ^bb0(%arg3: i32, %arg4: i32):
-    "omp.terminator"() : () -> ()
-  }) {in_reductions = [@add_f32], operandSegmentSizes = array<i32: 2, 2, 2, 0, 0, 2, 0, 0, 0, 0, 0, 0>} : (i32, i32, i32, i32, i32, i32, !llvm.ptr, !llvm.ptr) -> ()
+  "omp.taskloop"(%testf32, %testf32_2) ({
+    omp.loop_nest (%i, %j) : i32 = (%lb, %ub) to (%ub, %lb) step (%step, %step) {
+      omp.yield
+    }
+  }) {in_reductions = [@add_f32], operandSegmentSizes = array<i32: 0, 0, 2, 0, 0, 0, 0, 0, 0>} : (!llvm.ptr, !llvm.ptr) -> ()
   return
 }
 
 // -----
 
 func.func @taskloop(%lb: i32, %ub: i32, %step: i32) {
   %testf32 = "test.f32"() : () -> (!llvm.ptr)
-  %testf32_2 = "test.f32"() : () -> (!llvm.ptr)
   // expected-error @below {{expected as many reduction symbol references as reduction variables}}
-  "omp.taskloop"(%lb, %ub, %ub, %lb, %step, %step, %testf32_2) ({
-  ^bb0(%arg3: i32, %arg4: i32):
-    "omp.terminator"() : () -> ()
-  }) {in_reductions = [@add_f32, @add_f32], operandSegmentSizes = array<i32: 2, 2, 2, 0, 0, 1, 0, 0, 0, 0, 0, 0>} : (i32, i32, i32, i32, i32, i32, !llvm.ptr) -> ()
+  "omp.taskloop"(%testf32) ({
+    omp.loop_nest (%i, %j) : i32 = (%lb, %ub) to (%ub, %lb) step (%step, %step) {
+      omp.yield
+    }
+  }) {in_reductions = [@add_f32, @add_f32], operandSegmentSizes = array<i32: 0, 0, 1, 0, 0, 0, 0, 0, 0>} : (!llvm.ptr) -> ()
   return
 }
 
@@ -1657,9 +1660,10 @@ func.func @taskloop(%lb: i32, %ub: i32, %step: i32) {
   %testf32 = "test.f32"() : () -> (!llvm.ptr)
   %testf32_2 = "test.f32"() : () -> (!llvm.ptr)
   // expected-error @below {{if a reduction clause is present on the taskloop directive, the nogroup clause must not be specified}}
-  omp.taskloop reduction(@add_f32 -> %testf32 : !llvm.ptr, @add_f32 -> %testf32_2 : !llvm.ptr) nogroup
-  for (%i, %j) : i32 = (%lb, %ub) to (%ub, %lb) step (%step, %step) {
-    omp.terminator
+  omp.taskloop reduction(@add_f32 -> %testf32 : !llvm.ptr, @add_f32 -> %testf32_2 : !llvm.ptr) nogroup {
+    omp.loop_nest (%i, %j) : i32 = (%lb, %ub) to (%ub, %lb) step (%step, %step) {
+      omp.yield
+    }
   }
   return
 }
@@ -1681,9 +1685,10 @@ combiner {
 func.func @taskloop(%lb: i32, %ub: i32, %step: i32) {
   %testf32 = "test.f32"() : () -> (!llvm.ptr)
   // expected-error @below {{the same list item cannot appear in both a reduction and an in_reduction clause}}
-  omp.taskloop reduction(@add_f32 -> %testf32 : !llvm.ptr) in_reduction(@add_f32 -> %testf32 : !llvm.ptr)
-  for (%i, %j) : i32 = (%lb, %ub) to (%ub, %lb) step (%step, %step) {
-    omp.terminator
+  omp.taskloop reduction(@add_f32 -> %testf32 : !llvm.ptr) in_reduction(@add_f32 -> %testf32 : !llvm.ptr) {
+    omp.loop_nest (%i, %j) : i32 = (%lb, %ub) to (%ub, %lb) step (%step, %step) {
+      omp.yield
+    }
   }
   return
 }
@@ -1693,15 +1698,42 @@ func.func @taskloop(%lb: i32, %ub: i32, %step: i32) {
 func.func @taskloop(%lb: i32, %ub: i32, %step: i32) {
   %testi64 = "test.i64"() : () -> (i64)
   // expected-error @below {{the grainsize clause and num_tasks clause are mutually exclusive and may not appear on the same taskloop directive}}
-  omp.taskloop grain_size(%testi64: i64) num_tasks(%testi64: i64)
-  for (%i, %j) : i32 = (%lb, %ub) to (%ub, %lb) step (%step, %step) {
+  omp.taskloop grain_size(%testi64: i64) num_tasks(%testi64: i64) {
+    omp.loop_nest (%i, %j) : i32 = (%lb, %ub) to (%ub, %lb) step (%step, %step) {
+      omp.yield
+    }
+  }
+  return
+}
+
+// -----
+
+func.func @taskloop(%lb: i32, %ub: i32, %step: i32) {
+  // expected-error @below {{op must be a loop wrapper}}
+  omp.taskloop {
+    %0 = arith.constant 0 : i32
     omp.terminator
   }
   return
 }
 
 // -----
 
+func.func @taskloop(%lb: i32, %ub: i32, %step: i32) {
+  // expected-error @below {{only supported nested wrapper is 'omp.simdloop'}}
+  omp.taskloop {
+    omp.distribute {
+      omp.loop_nest (%iv) : i32 = (%lb) to (%ub) step (%step) {
+        omp.yield
+      }
+      omp.terminator
+    }
+  }
+  return
+}
+
+// -----
+
 func.func @omp_threadprivate() {
   %1 = llvm.mlir.addressof @_QFsubEx : !llvm.ptr
   // expected-error @below {{op failed to verify that all of {sym_addr, tls_addr} have same type}}