1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
|
// RUN: mlir-opt %s \
// RUN: -async-parallel-for=async-dispatch=true \
// RUN: -canonicalize -inline -symbol-dce \
// RUN: | FileCheck %s
// RUN: mlir-opt %s \
// RUN: -async-parallel-for=async-dispatch=false \
// RUN: -canonicalize -inline -symbol-dce \
// RUN: | FileCheck %s
// Check that if we statically know that the parallel operation has a single
// block then all async operations will be canonicalized away and we will
// end up with a single synchonous compute function call.
// CHECK-LABEL: @loop_1d(
// CHECK: %[[MEMREF:.*]]: memref<?xf32>
func @loop_1d(%arg0: memref<?xf32>) {
// CHECK-DAG: %[[C0:.*]] = constant 0 : index
// CHECK-DAG: %[[C1:.*]] = constant 1 : index
// CHECK-DAG: %[[C100:.*]] = constant 100 : index
// CHECK-DAG: %[[ONE:.*]] = constant 1.000000e+00 : f32
// CHECK: scf.for %[[I:.*]] = %[[C0]] to %[[C100]] step %[[C1]]
// CHECK: memref.store %[[ONE]], %[[MEMREF]][%[[I]]]
%lb = constant 0 : index
%ub = constant 100 : index
%st = constant 1 : index
scf.parallel (%i) = (%lb) to (%ub) step (%st) {
%one = constant 1.0 : f32
memref.store %one, %arg0[%i] : memref<?xf32>
}
return
}
|
