1// RUN: mlir-opt %s \ 2// RUN: | mlir-opt -gpu-lower-to-nvvm-pipeline="cubin-format=%gpu_compilation_format" \ 3// RUN: | mlir-runner \ 4// RUN: --shared-libs=%mlir_cuda_runtime \ 5// RUN: --shared-libs=%mlir_runner_utils \ 6// RUN: --entry-point-result=void \ 7// RUN: | FileCheck %s 8 9// CHECK: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12] 10func.func @main() { 11 %arg = memref.alloc() : memref<13xi32> 12 %dst = memref.cast %arg : memref<13xi32> to memref<?xi32> 13 %one = arith.constant 1 : index 14 %c0 = arith.constant 0 : index 15 %sx = memref.dim %dst, %c0 : memref<?xi32> 16 %cast_dst = memref.cast %dst : memref<?xi32> to memref<*xi32> 17 gpu.host_register %cast_dst : memref<*xi32> 18 gpu.launch blocks(%bx, %by, %bz) in (%grid_x = %one, %grid_y = %one, %grid_z = %one) 19 threads(%tx, %ty, %tz) in (%block_x = %sx, %block_y = %one, %block_z = %one) { 20 %t0 = arith.index_cast %tx : index to i32 21 memref.store %t0, %dst[%tx] : memref<?xi32> 22 gpu.terminator 23 } 24 gpu.launch blocks(%bx, %by, %bz) in (%grid_x = %one, %grid_y = %one, %grid_z = %one) 25 threads(%tx, %ty, %tz) in (%block_x = %sx, %block_y = %one, %block_z = %one) { 26 %t0 = arith.index_cast %tx : index to i32 27 memref.store %t0, %dst[%tx] : memref<?xi32> 28 gpu.terminator 29 } 30 call @printMemrefI32(%cast_dst) : (memref<*xi32>) -> () 31 return 32} 33 34func.func private @printMemrefI32(%memref : memref<*xi32>) 35