1// RUN: mlir-cuda-runner %s --shared-libs=%cuda_wrapper_library_dir/libcuda-runtime-wrappers%shlibext,%linalg_test_lib_dir/libmlir_runner_utils%shlibext --entry-point-result=void | FileCheck %s
2
3// CHECK-COUNT-8: [{{(5356, ){12}5356}}]
4func @main() {
5  %arg = alloc() : memref<2x4x13xf32>
6  %dst = memref_cast %arg : memref<2x4x13xf32> to memref<?x?x?xf32>
7  %c0 = constant 0 : index
8  %c1 = constant 1 : index
9  %c2 = constant 2 : index
10  %sx = dim %dst, %c2 : memref<?x?x?xf32>
11  %sy = dim %dst, %c1 : memref<?x?x?xf32>
12  %sz = dim %dst, %c0 : memref<?x?x?xf32>
13  %cast_dst = memref_cast %dst : memref<?x?x?xf32> to memref<*xf32>
14  gpu.host_register %cast_dst : memref<*xf32>
15  gpu.launch blocks(%bx, %by, %bz) in (%grid_x = %c1, %grid_y = %c1, %grid_z = %c1)
16             threads(%tx, %ty, %tz) in (%block_x = %sx, %block_y = %sy, %block_z = %sz) {
17    %t0 = muli %tz, %block_y : index
18    %t1 = addi %ty, %t0 : index
19    %t2 = muli %t1, %block_x : index
20    %idx = addi %tx, %t2 : index
21    %t3 = index_cast %idx : index to i32
22    %val = sitofp %t3 : i32 to f32
23    %sum = "gpu.all_reduce"(%val) ({}) { op = "add" } : (f32) -> (f32)
24    store %sum, %dst[%tz, %ty, %tx] : memref<?x?x?xf32>
25    gpu.terminator
26  }
27  call @print_memref_f32(%cast_dst) : (memref<*xf32>) -> ()
28  return
29}
30
31func private @print_memref_f32(%ptr : memref<*xf32>)
32