ace69e6b94
This PR improves the documentation for the `gpu-lower-to-nvvm-pipeline` (as it was remaning item for #75775) - Changes pipeline `gpu-lower-to-nvvm` -> `gpu-lower-to-nvvm-pipeline` - Adds a section in GPU Dialect in website. It clarifies the pipeline's functionality in lowering primary dialects to NVVM targets.
35 lines
1.4 KiB
MLIR
35 lines
1.4 KiB
MLIR
// RUN: mlir-opt %s \
|
|
// RUN: | mlir-opt -gpu-lower-to-nvvm-pipeline="cubin-format=%gpu_compilation_format" \
|
|
// RUN: | mlir-cpu-runner \
|
|
// RUN: --shared-libs=%mlir_cuda_runtime \
|
|
// RUN: --shared-libs=%mlir_runner_utils \
|
|
// RUN: --entry-point-result=void \
|
|
// RUN: | FileCheck %s
|
|
|
|
// CHECK: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12]
|
|
func.func @main() {
|
|
%arg = memref.alloc() : memref<13xi32>
|
|
%dst = memref.cast %arg : memref<13xi32> to memref<?xi32>
|
|
%one = arith.constant 1 : index
|
|
%c0 = arith.constant 0 : index
|
|
%sx = memref.dim %dst, %c0 : memref<?xi32>
|
|
%cast_dst = memref.cast %dst : memref<?xi32> to memref<*xi32>
|
|
gpu.host_register %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 = arith.index_cast %tx : index to i32
|
|
memref.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 = arith.index_cast %tx : index to i32
|
|
memref.store %t0, %dst[%tx] : memref<?xi32>
|
|
gpu.terminator
|
|
}
|
|
call @printMemrefI32(%cast_dst) : (memref<*xi32>) -> ()
|
|
return
|
|
}
|
|
|
|
func.func private @printMemrefI32(%memref : memref<*xi32>)
|