1// RUN: mlir-opt %s \ 2// RUN: | mlir-opt -gpu-kernel-outlining \ 3// RUN: | mlir-opt -pass-pipeline='builtin.module(gpu.module(strip-debuginfo,convert-gpu-to-rocdl),rocdl-attach-target{chip=%chip})' \ 4// RUN: | mlir-opt -gpu-to-llvm -reconcile-unrealized-casts -gpu-module-to-binary \ 5// RUN: | mlir-runner \ 6// RUN: --shared-libs=%mlir_rocm_runtime \ 7// RUN: --shared-libs=%mlir_runner_utils \ 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.func @main() { 13 %arg = memref.alloc() : memref<13xi32> 14 %dst = memref.cast %arg : memref<13xi32> to memref<?xi32> 15 %c0 = arith.constant 0 : index 16 %c1 = arith.constant 1 : 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 %dst_device = call @mgpuMemGetDeviceMemRef1dInt32(%dst) : (memref<?xi32>) -> (memref<?xi32>) 21 gpu.launch blocks(%bx, %by, %bz) in (%grid_x = %c1, %grid_y = %c1, %grid_z = %c1) 22 threads(%tx, %ty, %tz) in (%block_x = %sx, %block_y = %c1, %block_z = %c1) { 23 %t0 = arith.index_cast %tx : index to i32 24 memref.store %t0, %dst_device[%tx] : memref<?xi32> 25 gpu.terminator 26 } 27 gpu.launch blocks(%bx, %by, %bz) in (%grid_x = %c1, %grid_y = %c1, %grid_z = %c1) 28 threads(%tx, %ty, %tz) in (%block_x = %sx, %block_y = %c1, %block_z = %c1) { 29 %t0 = arith.index_cast %tx : index to i32 30 memref.store %t0, %dst_device[%tx] : memref<?xi32> 31 gpu.terminator 32 } 33 call @printMemrefI32(%cast_dst) : (memref<*xi32>) -> () 34 return 35} 36 37func.func private @mgpuMemGetDeviceMemRef1dInt32(%ptr : memref<?xi32>) -> (memref<?xi32>) 38func.func private @printMemrefI32(%ptr : memref<*xi32>) 39