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
|
// RUN: mlir-cuda-runner %s --print-ir-after-all --shared-libs=%cuda_wrapper_library_dir/libcuda-runtime-wrappers%shlibext,%linalg_test_lib_dir/libmlir_runner_utils%shlibext --entry-point-result=void | FileCheck %s --dump-input=always
// CHECK: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12]
func @main() {
%arg = alloc() : memref<13xi32>
%dst = memref_cast %arg : memref<13xi32> to memref<?xi32>
%one = constant 1 : index
%c0 = constant 0 : index
%sx = dim %dst, %c0 : memref<?xi32>
%cast_dst = memref_cast %dst : memref<?xi32> to memref<*xi32>
call @mcuMemHostRegisterInt32(%cast_dst) : (memref<*xi32>) -> ()
gpu.launch blocks(%bx, %by, %bz) in (%grid_x = %one, %grid_y = %one, %grid_z = %one)
threads(%tx, %ty, %tz) in (%block_x = %sx, %block_y = %one, %block_z = %one) {
%t0 = index_cast %tx : index to i32
store %t0, %dst[%tx] : memref<?xi32>
gpu.terminator
}
gpu.launch blocks(%bx, %by, %bz) in (%grid_x = %one, %grid_y = %one, %grid_z = %one)
threads(%tx, %ty, %tz) in (%block_x = %sx, %block_y = %one, %block_z = %one) {
%t0 = index_cast %tx : index to i32
store %t0, %dst[%tx] : memref<?xi32>
gpu.terminator
}
call @print_memref_i32(%cast_dst) : (memref<*xi32>) -> ()
return
}
func @mcuMemHostRegisterInt32(%memref : memref<*xi32>)
func @print_memref_i32(%memref : memref<*xi32>)
|
