xref: /llvm-project/mlir/test/Dialect/GPU/mapping.mlir (revision 7bdd3722f20c68ce14d153939bd8aa11d1c0d482) 
1// RUN: mlir-opt -gpu-map-parallel-loops -split-input-file %s | FileCheck %s
2
3func.func @parallel_loop(%arg0 : index, %arg1 : index, %arg2 : index,
4                    %arg3 : index) {
5  %zero = arith.constant 0 : index
6  %one = arith.constant 1 : index
7  %four = arith.constant 4 : index
8  scf.parallel (%i0, %i1) = (%arg0, %arg1) to (%arg2, %arg3)
9                                          step (%four, %four)  {
10    scf.parallel (%si0, %si1) = (%zero, %zero) to (%four, %four)
11                                            step (%one, %one)  {
12    }
13  }
14  return
15}
16
17// CHECK-LABEL:   func @parallel_loop(
18// CHECK:           scf.parallel
19// CHECK:             scf.parallel
20// CHECK:      {mapping = [#gpu.loop_dim_map<processor = thread_x, map = (d0) -> (d0), bound = (d0) -> (d0)>,
21// CHECK-SAME:             #gpu.loop_dim_map<processor = thread_y, map = (d0) -> (d0), bound = (d0) -> (d0)>]}
22// CHECK:      {mapping = [#gpu.loop_dim_map<processor = block_x, map = (d0) -> (d0), bound = (d0) -> (d0)>,
23// CHECK-SAME:             #gpu.loop_dim_map<processor = block_y, map = (d0) -> (d0), bound = (d0) -> (d0)>]}
24// CHECK-NOT: mapping
25
26// -----
27
28func.func @parallel_loop_4d(%arg0 : index, %arg1 : index, %arg2 : index,
29                       %arg3 : index) {
30  %zero = arith.constant 0 : index
31  %one = arith.constant 1 : index
32  %four = arith.constant 4 : index
33  scf.parallel (%i0, %i1, %i2, %i3) = (%zero, %zero, %zero, %zero) to (%arg0, %arg1, %arg2, %arg3)
34                                       step (%four, %four, %four, %four)  {
35    scf.parallel (%si0, %si1, %si2, %si3) = (%zero, %zero, %zero, %zero) to (%four, %four, %four, %four)
36                                             step (%one, %one, %one, %one)  {
37      scf.parallel (%ti0, %ti1, %ti2, %ti3) = (%zero, %zero, %zero, %zero) to (%four, %four, %four, %four)
38                                               step (%one, %one, %one, %one)  {
39      }
40    }
41  }
42  return
43}
44
45// CHECK-LABEL:   func @parallel_loop_4d(
46// CHECK:           scf.parallel
47// CHECK:             scf.parallel
48// CHECK:               scf.parallel
49// CHECK:      {mapping = [#gpu.loop_dim_map<processor = sequential, map = (d0) -> (d0), bound = (d0) -> (d0)>,
50// CHECK-SAME:             #gpu.loop_dim_map<processor = sequential, map = (d0) -> (d0), bound = (d0) -> (d0)>,
51// CHECK-SAME:             #gpu.loop_dim_map<processor = sequential, map = (d0) -> (d0), bound = (d0) -> (d0)>,
52// CHECK-SAME:             #gpu.loop_dim_map<processor = sequential, map = (d0) -> (d0), bound = (d0) -> (d0)>]}
53// CHECK:      {mapping = [#gpu.loop_dim_map<processor = thread_x, map = (d0) -> (d0), bound = (d0) -> (d0)>,
54// CHECK-SAME:             #gpu.loop_dim_map<processor = thread_y, map = (d0) -> (d0), bound = (d0) -> (d0)>,
55// CHECK-SAME:             #gpu.loop_dim_map<processor = thread_z, map = (d0) -> (d0), bound = (d0) -> (d0)>,
56// CHECK-SAME:             #gpu.loop_dim_map<processor = sequential, map = (d0) -> (d0), bound = (d0) -> (d0)>]}
57// CHECK:      {mapping = [#gpu.loop_dim_map<processor = block_x, map = (d0) -> (d0), bound = (d0) -> (d0)>,
58// CHECK-SAME:             #gpu.loop_dim_map<processor = block_y, map = (d0) -> (d0), bound = (d0) -> (d0)>,
59// CHECK-SAME:             #gpu.loop_dim_map<processor = block_z, map = (d0) -> (d0), bound = (d0) -> (d0)>,
60// CHECK-SAME:             #gpu.loop_dim_map<processor = sequential, map = (d0) -> (d0), bound = (d0) -> (d0)>]}
61// CHECK-NOT: mapping
62