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
34
35
36
37
38
39
|
// RUN: mlir-opt %s -scf-for-loop-specialization -split-input-file | FileCheck %s
#map0 = affine_map<()[s0, s1] -> (1024, s0 - s1)>
#map1 = affine_map<()[s0, s1] -> (64, s0 - s1)>
func.func @for(%outer: index, %A: memref<?xf32>, %B: memref<?xf32>,
%C: memref<?xf32>, %result: memref<?xf32>) {
%c0 = arith.constant 0 : index
%c1 = arith.constant 1 : index
%d0 = memref.dim %A, %c0 : memref<?xf32>
%b0 = affine.min #map0()[%d0, %outer]
scf.for %i0 = %c0 to %b0 step %c1 {
%B_elem = memref.load %B[%i0] : memref<?xf32>
%C_elem = memref.load %C[%i0] : memref<?xf32>
%sum_elem = arith.addf %B_elem, %C_elem : f32
memref.store %sum_elem, %result[%i0] : memref<?xf32>
}
return
}
// CHECK-LABEL: func @for(
// CHECK-SAME: [[ARG0:%.*]]: index, [[ARG1:%.*]]: memref<?xf32>, [[ARG2:%.*]]: memref<?xf32>, [[ARG3:%.*]]: memref<?xf32>, [[ARG4:%.*]]: memref<?xf32>) {
// CHECK: [[CST_0:%.*]] = arith.constant 0 : index
// CHECK: [[CST_1:%.*]] = arith.constant 1 : index
// CHECK: [[DIM_0:%.*]] = memref.dim [[ARG1]], [[CST_0]] : memref<?xf32>
// CHECK: [[MIN:%.*]] = affine.min #{{.*}}(){{\[}}[[DIM_0]], [[ARG0]]]
// CHECK: [[CST_1024:%.*]] = arith.constant 1024 : index
// CHECK: [[PRED:%.*]] = arith.cmpi eq, [[MIN]], [[CST_1024]] : index
// CHECK: scf.if [[PRED]] {
// CHECK: scf.for [[IDX0:%.*]] = [[CST_0]] to [[CST_1024]] step [[CST_1]] {
// CHECK: memref.store
// CHECK: }
// CHECK: } else {
// CHECK: scf.for [[IDX0:%.*]] = [[CST_0]] to [[MIN]] step [[CST_1]] {
// CHECK: memref.store
// CHECK: }
// CHECK: }
// CHECK: return
// CHECK: }
|
