MLIR

Multi-Level IR Compiler Framework

Dialect 'loop' definition

Operation definition

loop.for (ForOp)

for operation

Description:

The “loop.for” operation represents a loop taking 3 SSA value as operands that represent the lower bound, upper bound and step respectively. The operation defines an SSA value for its induction variable. It has one region capturing the loop body. The induction variable is represented as an argument of this region. This SSA value always has type index, which is the size of the machine word. The step is a value of type index, required to be positive. The lower and upper bounds specify a half-open range: the range includes the lower bound but does not include the upper bound.

The body region must contain exactly one block that terminates with “loop.yield”. Calling ForOp::build will create such a region and insert the terminator implicitly if none is defined, so will the parsing even in cases when it is absent from the custom format. For example:

   loop.for %iv = %lb to %ub step %step {
     ... // body
   }

“loop.for” can also operate on loop-carried variables and returns the final values after loop termination. The initial values of the variables are passed as additional SSA operands to the “loop.for” following the 3 loop control SSA values mentioned above (lower bound, upper bound and step). The operation region has equivalent arguments for each variable representing the value of the variable at the current iteration.

The region must terminate with a “loop.yield” that passes all the current iteration variables to the next iteration, or to the “loop.for” result, if at the last iteration. “loop.for” results hold the final values after the last iteration.

For example, to sum-reduce a memref:

func @reduce(%buffer: memref<1024xf32>, %lb: index, %ub: index, %step: index) -> (f32) {
  // Initial sum set to 0.
  %sum_0 = constant 0.0 : f32
  // iter_args binds initial values to the loop's region arguments.
  %sum = loop.for %iv = %lb to %ub step %step iter_args(%sum_iter = %sum_0) -> (f32) {
   %t = load %buffer[%iv] : memref<1024xf32>
    %sum_next = addf %sum_iter, %t : f32
    // Yield current iteration sum to next iteration %sum_iter or to %sum if final iteration.
    loop.yield %sum_next : f32
  }
  return %sum : f32
}   

If the “loop.for” defines any values, a yield must be explicitly present. The number and types of the “loop.for” results must match the initial values in the “iter_args” binding and the yield operands.

Another example with a nested “loop.if” (see “loop.if” for details) to perform conditional reduction:

func @conditional_reduce(%buffer: memref<1024xf32>, %lb: index, %ub: index, %step: index) -> (f32) {
  %sum_0 = constant 0.0 : f32
  %c0 = constant 0.0 : f32
  %sum = loop.for %iv = %lb to %ub step %step iter_args(%sum_iter = %sum_0) -> (f32) {
    %t = load %buffer[%iv] : memref<1024xf32>
    %cond = cmpf "ugt", %t, %c0 : f32
    %sum_next = loop.if %cond -> (f32) {
     %new_sum = addf %sum_iter, %t : f32
      loop.yield %new_sum : f32
   } else {
          loop.yield %sum_iter : f32
   }
    loop.yield %sum_next : f32
  }
  return %sum : f32
}

Operands:

  1. lowerBound: index
  2. upperBound: index
  3. step: index
  4. initArgs: any type

Attributes:

Results:

  1. results: any type

loop.if (IfOp)

if-then-else operation

Description:

The “loop.if” operation represents an if-then-else construct for conditionally executing two regions of code. The operand to an if operation is a boolean value. For example:

   loop.if %b  {
     ...
   } else {
     ...
   }

“loop.if” may also return results that are defined in its regions. The values defined are determined by which execution path is taken. For example:

   %x, %y = loop.if %b -> (f32, f32) {
     %x_true = ...
     %y_true = ...
     loop.yield %x_true, %y_true : f32, f32
   } else {
     %x_false = ... 
     %y_false = ... 
     loop.yield %x_false, %y_false : f32, f32
   }

“loop.if” regions are always terminated with “loop.yield”. If “loop.if” defines no values, the “loop.yield” can be left out, and will be inserted implicitly. Otherwise, it must be explicit. Also, if “loop.if” defines one or more values, the ‘else’ block cannot be omitted.

For example:

   loop.if %b  {
     ...
   }

Operands:

  1. condition: 1-bit integer

Attributes:

Results:

  1. results: any type

loop.parallel (ParallelOp)

parallel for operation

Description:

The “loop.parallel” operation represents a loop nest taking 3 groups of SSA values as operands that represent the lower bounds, upper bounds and steps, respectively. The operation defines a variadic number of SSA values for its induction variables. It has one region capturing the loop body. The induction variables are represented as an argument of this region. These SSA values always have type index, which is the size of the machine word. The steps are values of type index, required to be positive. The lower and upper bounds specify a half-open range: the range includes the lower bound but does not include the upper bound.

Semantically we require that the iteration space can be iterated in any order, and the loop body can be executed in parallel. If there are data races, the behavior is undefined.

The parallel loop operation supports reduction of values produced by individual iterations into a single result. This is modeled using the loop.reduce operation (see loop.reduce for details). Each result of a loop.parallel operation is associated with a reduce operation that is an immediate child. Reduces are matched to result values in order of their appearance in the body. Consequently, we require that the body region has the same number of results as it has reduce operations.

The body region must contain exactly one block that terminates with “loop.yield” without operands. Parsing ParallelOp will create such a region and insert the terminator when it is absent from the custom format. For example:

   loop.parallel (%iv) = (%lb) to (%ub) step (%step) {
     %zero = constant 0.0 : f32
     loop.reduce(%zero) {
       ^bb0(%lhs : f32, %rhs: f32):
         %res = addf %lhs, %rhs : f32
         loop.reduce.return %res : f32
     } : f32
   }

Operands:

  1. lowerBound: index
  2. upperBound: index
  3. step: index

Attributes:

Results:

  1. results: any type

loop.reduce (ReduceOp)

reduce operation for parallel for

Description:

“loop.reduce” is an operation occurring inside “loop.parallel” operations. It consists of one block with two arguments which have the same type as the operand of “loop.reduce”.

“loop.reduce” is used to model the value for reduction computations of a “loop.parallel” operation. It has to appear as an immediate child of a “loop.parallel” and is associated with a result value of its parent operation.

Association is in the order of appearance in the body where the first result of a parallel loop operation corresponds to the first “loop.reduce” in the operation’s body region. The reduce operation takes a single operand, which is the value to be used in the reduction.

The reduce operation contains a region whose entry block expects two arguments of the same type as the operand. As the iteration order of the parallel loop and hence reduction order is unspecified, the result of reduction may be non-deterministic unless the operation is associative and commutative.

The result of the reduce operation’s body must have the same type as the operands and associated result value of the parallel loop operation. Example:

  %operand = constant 1.0 : f32
  loop.reduce(%operand) {
    ^bb0(%lhs : f32, %rhs: f32):
      %res = addf %lhs, %rhs : f32
      loop.reduce.return %res : f32
  } : f32

Operands:

  1. operand: any type

Attributes:

Results:

loop.reduce.return (ReduceReturnOp)

terminator for reduce operation

Description:

“loop.reduce.return” is a special terminator operation for the block inside “loop.reduce”. It terminates the region. It should have the same type as the operand of “loop.reduce”. Example for the custom format:

  loop.reduce.return %res : f32

Operands:

  1. result: any type

Attributes:

Results:

loop.yield (YieldOp)

loop yield and termination operation

Description:

“loop.yield” yields an SSA value from a loop dialect op region and terminates the regions. The semantics of how the values are yielded is defined by the parent operation. If “loop.yield” has any operands, the operands must match the parent operation’s results. If the parent operation defines no values, then the “loop.yield” may be left out in the custom syntax and the builders will insert one implicitly. Otherwise, it has to be present in the syntax to indicate which values are yielded.

Operands:

  1. results: any type

Attributes:

Results: