Created by: Chgtaxihe

复现用的PaddlePaddle模型:

from paddle import fluid
from paddle.fluid import layers

x = fluid.data(name='value', shape=[1], dtype='float32')

arr = layers.create_array('float32')
step = layers.fill_constant(shape=[1], dtype='int64', value=1)
max_step = layers.fill_constant(shape=[1], dtype='int64', value=5)

zero = layers.fill_constant(shape=[1], dtype='int64', value=0)
layers.array_write(x+1.0, step, array=arr)

condition = layers.less_than(step, max_step)
while_op = layers.While(condition)
with while_op.block():
    layers.array_write(x+1, step, array=arr)
    layers.increment(step)
    layers.less_than(step, max_step, cond=condition)

idx = layers.fill_constant([1], dtype='int64', value=4)
result = layers.array_read(arr, idx)

exe = fluid.Executor(fluid.CPUPlace())
exe.run(fluid.default_startup_program())
fluid.io.save_inference_model('./model_output',
                              feeded_var_names=['value'],
                              target_vars=[result],
                              executor=exe,
                              main_program=fluid.default_main_program())

if __name__ == '__main__':
    pass

转换命令:

./model_optimize_tool --model_dir=./infer --valid_targets=arm --optimize_out_type=naive_buffer --optimize_out=infer_out

错误提示:

Can't find a Cast kernel for Cast op: Tensor<arm,int64_t,NCHW,0>:fill_constant_0.tmp_0->Tensor<arm,float,NCHW,0>:write_to_array

按照提示，我应该把step 改为 float 类型，但是按照PaddlePaddle的文档，此处的step应当为 int64 类型。

请问应该如何解决？