import torch
import numpy as np
import torch_npu
from torch_npu.testing.testcase import TestCase, run_tests
from torch_npu.testing.common_utils import create_common_tensor
class TestMaxBackward(TestCase):
def cpu_op_other_exec(self, input1, input2):
input1.requires_grad = True
output = torch.max(input1, input2)
output.backward(torch.ones_like(output))
out = input1.grad
return out
def npu_op_other_exec(self, input1, input2):
input1.requires_grad = True
output = torch.max(input1, input2)
output.backward(torch.ones_like(output))
out = input1.grad
out = out.to('cpu')
return out
def max_result_other(self, shape_format):
for item in shape_format:
cpu_input1, npu_input1 = create_common_tensor(item[0], 0, 100)
cpu_input2, npu_input2 = create_common_tensor(item[0], 0, 7)
if cpu_input1.dtype == torch.float16:
cpu_input1 = cpu_input1.to(torch.float32)
cpu_input2 = cpu_input2.to(torch.float32)
cpu_output_other = self.cpu_op_other_exec(cpu_input1, cpu_input2)
npu_output_other = self.npu_op_other_exec(npu_input1, npu_input2)
if npu_output_other.dtype == torch.float16:
npu_output_other = npu_output_other.float()
self.assertRtolEqual(cpu_output_other, npu_output_other)
def test_max_other_shape_format_fp16_1d(self):
format_list = [0, 3, 4]
keepdim_list = [True, False]
shape_format = [[[np.float16, i, [8]], np.random.randint(0, 1), j] for i in format_list for j in keepdim_list
]
self.max_result_other(shape_format)
def test_max_other_shape_format_fp32_1d(self):
format_list = [0, 3, 4]
keepdim_list = [True, False]
shape_format = [[[np.float32, i, [8]], np.random.randint(0, 1), j] for i in format_list for j in
keepdim_list
]
self.max_result_other(shape_format)
def test_max_other_shape_format_fp16_2d(self):
format_list = [0, 3, 4, 29]
keepdim_list = [True, False]
shape_format = [[[np.float16, i, [8, 7]], np.random.randint(0, 2), j] for i in format_list for j in
keepdim_list
]
self.max_result_other(shape_format)
def test_max_other_shape_format_fp32_2d(self):
format_list = [0, 3, 4, 29]
keepdim_list = [True, False]
shape_format = [[[np.float32, i, [8, 7]], np.random.randint(0, 2), j] for i in format_list for j in
keepdim_list
]
self.max_result_other(shape_format)
def test_max_other_shape_format_fp16_3d(self):
format_list = [0, 3, 4, 29]
keepdim_list = [True, False]
shape_format = [[[np.float16, i, [8, 7, 9]], np.random.randint(0, 3), j] for i in format_list for j in
keepdim_list
]
self.max_result_other(shape_format)
def test_max_other_shape_format_fp32_3d(self):
format_list = [0, 3, 4, 29]
keepdim_list = [True, False]
shape_format = [[[np.float32, i, [8, 7, 9]], np.random.randint(0, 3), j] for i in format_list for j in
keepdim_list
]
self.max_result_other(shape_format)
def test_max_other_shape_format_fp16_4d(self):
format_list = [0, 3, 4, 29]
keepdim_list = [True, False]
shape_format = [[[np.float16, i, [8, 7, 9, 10]], np.random.randint(0, 4), j] for i in format_list for j
in keepdim_list
]
self.max_result_other(shape_format)
def test_max_other_shape_format_fp32_4d(self):
format_list = [0, 3, 4, 29]
keepdim_list = [True, False]
shape_format = [[[np.float32, i, [8, 7, 9, 10]], np.random.randint(0, 4), j] for i in format_list for j
in
keepdim_list
]
self.max_result_other(shape_format)
if __name__ == "__main__":
run_tests()