1// RUN: mlir-opt %s \
2// RUN:   -gpu-kernel-outlining \
3// RUN:   -pass-pipeline='gpu.module(strip-debuginfo,convert-gpu-to-nvvm,gpu-to-cubin)' \
4// RUN:   -gpu-to-llvm \
5// RUN: | mlir-cpu-runner \
6// RUN:   --shared-libs=%linalg_test_lib_dir/libmlir_cuda_runtime%shlibext \
7// RUN:   --shared-libs=%linalg_test_lib_dir/libmlir_runner_utils%shlibext \
8// RUN:   --entry-point-result=void \
9// RUN: | FileCheck %s
10
11// CHECK: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12]
12func @main() {
13  %arg = memref.alloc() : memref<13xi32>
14  %dst = memref.cast %arg : memref<13xi32> to memref<?xi32>
15  %one = constant 1 : index
16  %c0 = constant 0 : index
17  %sx = memref.dim %dst, %c0 : memref<?xi32>
18  %cast_dst = memref.cast %dst : memref<?xi32> to memref<*xi32>
19  gpu.host_register %cast_dst : memref<*xi32>
20  gpu.launch blocks(%bx, %by, %bz) in (%grid_x = %one, %grid_y = %one, %grid_z = %one)
21             threads(%tx, %ty, %tz) in (%block_x = %sx, %block_y = %one, %block_z = %one) {
22    %t0 = index_cast %tx : index to i32
23    memref.store %t0, %dst[%tx] : memref<?xi32>
24    gpu.terminator
25  }
26  gpu.launch blocks(%bx, %by, %bz) in (%grid_x = %one, %grid_y = %one, %grid_z = %one)
27             threads(%tx, %ty, %tz) in (%block_x = %sx, %block_y = %one, %block_z = %one) {
28    %t0 = index_cast %tx : index to i32
29    memref.store %t0, %dst[%tx] : memref<?xi32>
30    gpu.terminator
31  }
32  call @print_memref_i32(%cast_dst) : (memref<*xi32>) -> ()
33  return
34}
35
36func private @print_memref_i32(%memref : memref<*xi32>)
37