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 TestAdd(TestCase):
def cpu_op_out_exec(self, input1, input2, output):
torch.add(input1, input2, alpha=1, out=output)
output = output.numpy()
return output
def npu_op_out_exec_new(self, input1, input2, output):
torch.add(input1, input2, alpha=1, out=output)
output = output.to("cpu")
output = output.numpy()
return output
def cpu_op_exec(self, input1, input2):
output = torch.add(input1, input2, alpha=1)
output = output.numpy()
return output
def cpu_tensor_op_exec(self, input1, input2):
output = input1.add(input2, alpha=1)
output = output.numpy()
return output
def npu_tensor_op_exec(self, input1, input2):
output = input1.add(input2, alpha=1)
output = output.to("cpu")
output = output.numpy()
return output
def npu_op_exec_new(self, input1, input2):
output = torch.add(input1, input2, alpha=1)
output = output.to("cpu")
output = output.numpy()
return output
def cpu_op_exec_alpha(self, input1, input2):
output = torch.add(input1, input2, alpha=3)
output = output.numpy()
return output
def npu_op_exec_new_alpha(self, input1, input2):
output = torch.add(input1, input2, alpha=3)
output = output.to("cpu")
output = output.numpy()
return output
def cpu_op_scalar_exec(self, input1, scalar):
output = torch.add(input1, scalar, alpha=1)
output = output.numpy()
return output
def npu_op_scalar_exec_new(self, input1, scalar):
output = torch.add(input1, scalar, alpha=1)
output = output.to("cpu")
output = output.numpy()
return output
def cpu_op_scalar_exec_alpha(self, input1, scalar):
output = torch.add(input1, scalar, alpha=3)
output = output.numpy()
return output
def npu_op_scalar_exec_new_alpha(self, input1, scalar):
output = torch.add(input1, scalar, alpha=3)
output = output.to("cpu")
output = output.numpy()
return output
def add_scalar_result(self, shape_format):
for item in shape_format:
cpu_input, npu_input = create_common_tensor(item[0], 0, 100)
if cpu_input.dtype == torch.float16:
cpu_input = cpu_input.to(torch.float32)
cpu_output = self.cpu_op_scalar_exec(cpu_input, item[1])
npu_output = self.npu_op_exec_new(npu_input, item[1])
cpu_output = cpu_output.astype(npu_output.dtype)
self.assertRtolEqual(cpu_output, npu_output)
def add_scalar_alpha_result(self, shape_format):
for item in shape_format:
cpu_input, npu_input = create_common_tensor(item[0], 0, 100)
if cpu_input.dtype == torch.float16:
cpu_input = cpu_input.to(torch.float32)
cpu_output = self.cpu_op_scalar_exec_alpha(cpu_input, item[1])
npu_output = self.npu_op_scalar_exec_new_alpha(npu_input, item[1])
cpu_output = cpu_output.astype(npu_output.dtype)
self.assertRtolEqual(cpu_output, npu_output)
def add_result(self, shape_format):
for item in shape_format:
cpu_input1, npu_input1 = create_common_tensor(item, 0, 100)
cpu_input2, npu_input2 = create_common_tensor(item, 0, 100)
if cpu_input1.dtype == torch.float16:
cpu_input1 = cpu_input1.to(torch.float32)
cpu_input2 = cpu_input2.to(torch.float32)
cpu_output = self.cpu_op_exec(cpu_input1, cpu_input2)
npu_output = self.npu_op_exec_new(npu_input1, npu_input2)
cpu_output = cpu_output.astype(npu_output.dtype)
self.assertRtolEqual(cpu_output, npu_output)
def add_out_result(self, shape_format):
for item in shape_format:
cpuout = torch.randn(3)
npuout = torch.randn(3).to("npu")
cpu_input1, npu_input1 = create_common_tensor(item, 0, 100)
cpu_input2, npu_input2 = create_common_tensor(item, 0, 100)
if cpu_input1.dtype == torch.float16:
cpu_input1 = cpu_input1.to(torch.float32)
cpu_input2 = cpu_input2.to(torch.float32)
cpu_output = self.cpu_op_out_exec(cpu_input1, cpu_input2, cpuout)
npu_output = self.npu_op_out_exec_new(npu_input1, npu_input2, npuout)
cpu_output = cpu_output.astype(npu_output.dtype)
self.assertRtolEqual(cpu_output, npu_output)
def add_alpha_result(self, shape_format):
for item in shape_format:
cpu_input1, npu_input1 = create_common_tensor(item, 0, 100)
cpu_input2, npu_input2 = create_common_tensor(item, 0, 100)
if cpu_input1.dtype == torch.float16:
cpu_input1 = cpu_input1.to(torch.float32)
cpu_input2 = cpu_input2.to(torch.float32)
cpu_output = self.cpu_op_exec_alpha(cpu_input1, cpu_input2)
npu_output = self.npu_op_exec_new_alpha(npu_input1, npu_input2)
cpu_output = cpu_output.astype(npu_output.dtype)
self.assertRtolEqual(cpu_output, npu_output)
def test_add_scalar_shape_format_fp16_1d(self):
format_list = [0, 3]
scalar_list = [0, 1]
shape_format = [
[[np.float16, i, [18]], k] for i in format_list for k in scalar_list
]
self.add_scalar_result(shape_format)
def test_add_scalar_shape_format_fp32_1d(self):
format_list = [0, 3]
scalar_list = [0, 1]
shape_format = [
[[np.float32, i, [18]], k] for i in format_list for k in scalar_list
]
self.add_scalar_result(shape_format)
def test_add_scalar_shape_format_fp16_2d(self):
format_list = [0, 3, 29]
scalar_list = [0, 1]
shape_format = [
[[np.float16, i, [5, 256]], k] for i in format_list for k in scalar_list
]
self.add_scalar_result(shape_format)
def test_add_scalar_shape_format_fp32_2d(self):
format_list = [0, 3, 29]
scalar_list = [0, 1]
shape_format = [
[[np.float32, i, [5, 256]], k] for i in format_list for k in scalar_list
]
self.add_scalar_result(shape_format)
def test_add_scalar_shape_format_fp16_3d(self):
format_list = [0, 3, 29]
scalar_list = [0, 1]
shape_format = [
[[np.float16, i, [32, 3, 3]], k] for i in format_list for k in scalar_list
]
self.add_scalar_result(shape_format)
def test_add_scalar_shape_format_fp32_3d(self):
format_list = [0, 3, 29]
scalar_list = [0, 1]
shape_format = [
[[np.float32, i, [32, 3, 3]], k] for i in format_list for k in scalar_list
]
self.add_scalar_result(shape_format)
def test_add_scalar_shape_format_fp16_4d(self):
format_list = [0, 3, 29]
scalar_list = [0, 1]
shape_format = [
[[np.float16, i, [64, 112, 7, 7]], k]
for i in format_list
for k in scalar_list
]
self.add_scalar_result(shape_format)
def test_add_scalar_shape_format_fp32_4d(self):
format_list = [0, 3, 29]
scalar_list = [0, 1]
shape_format = [
[[np.float32, i, [64, 112, 7, 7]], k]
for i in format_list
for k in scalar_list
]
self.add_scalar_result(shape_format)
def test_add_scalar_shape_format_fp16_1d(self):
format_list = [0, 3]
scalar_list = [0, 1]
shape_format = [
[[np.float16, i, [18]], k] for i in format_list for k in scalar_list
]
self.add_scalar_alpha_result(shape_format)
def test_add_scalar_shape_format_fp32_1d(self):
format_list = [0, 3]
scalar_list = [0, 1]
shape_format = [
[[np.float32, i, [18]], k] for i in format_list for k in scalar_list
]
self.add_scalar_alpha_result(shape_format)
def test_add_scalar_shape_format_fp16_2d(self):
format_list = [0, 3, 29]
scalar_list = [0, 1]
shape_format = [
[[np.float16, i, [5, 256]], k] for i in format_list for k in scalar_list
]
self.add_scalar_alpha_result(shape_format)
def test_add_scalar_shape_format_fp32_2d(self):
format_list = [0, 3, 29]
scalar_list = [0, 1]
shape_format = [
[[np.float32, i, [5, 256]], k] for i in format_list for k in scalar_list
]
self.add_scalar_alpha_result(shape_format)
def test_add_scalar_shape_format_fp16_3d(self):
format_list = [0, 3, 29]
scalar_list = [0, 1]
shape_format = [
[[np.float16, i, [32, 3, 3]], k] for i in format_list for k in scalar_list
]
self.add_scalar_alpha_result(shape_format)
def test_add_scalar_shape_format_fp32_3d(self):
format_list = [0, 3, 29]
scalar_list = [0, 1]
shape_format = [
[[np.float32, i, [32, 3, 3]], k] for i in format_list for k in scalar_list
]
self.add_scalar_alpha_result(shape_format)
def test_add_scalar_shape_format_fp16_4d(self):
format_list = [0, 3, 29]
scalar_list = [0, 1]
shape_format = [
[[np.float16, i, [64, 112, 7, 7]], k]
for i in format_list
for k in scalar_list
]
self.add_scalar_alpha_result(shape_format)
def test_add_scalar_shape_format_fp32_4d(self):
format_list = [0, 3, 29]
scalar_list = [0, 1]
shape_format = [
[[np.float32, i, [64, 112, 7, 7]], k]
for i in format_list
for k in scalar_list
]
self.add_scalar_alpha_result(shape_format)
def test_add_shape_format_fp16_1d(self):
format_list = [0, 3]
shape_format = [[np.float16, i, [64]] for i in format_list]
self.add_result(shape_format)
def test_add_shape_format_fp32_1d(self):
format_list = [0, 3]
shape_format = [[np.float32, i, [64]] for i in format_list]
self.add_result(shape_format)
def test_add_shape_format_fp16_2d(self):
format_list = [0, 3, 29]
shape_format = [[np.float16, i, [5, 256]] for i in format_list]
self.add_result(shape_format)
def test_add_shape_format_fp32_2d(self):
format_list = [0, 3, 29]
shape_format = [[np.float32, i, [5, 256]] for i in format_list]
self.add_result(shape_format)
def test_add_shape_format_fp16_3d(self):
format_list = [0, 3, 29]
shape_format = [[np.float16, i, [32, 3, 3]] for i in format_list]
self.add_result(shape_format)
def test_add_shape_format_fp32_3d(self):
format_list = [0, 3, 29]
shape_format = [[np.float32, i, [32, 3, 3]] for i in format_list]
self.add_result(shape_format)
def test_add_shape_format_fp16_4d(self):
format_list = [0, 3, 29]
shape_format = [[np.float16, i, [64, 112, 7, 7]] for i in format_list]
self.add_result(shape_format)
def test_add_shape_format_fp32_4d(self):
format_list = [0, 3, 29]
shape_format = [[np.float32, i, [64, 112, 7, 7]] for i in format_list]
self.add_result(shape_format)
def test_add_shape_format_fp16_1d(self):
format_list = [0, 3]
shape_format = [[np.float16, i, [64]] for i in format_list]
self.add_alpha_result(shape_format)
def test_add_shape_format_fp32_1d(self):
format_list = [0, 3]
shape_format = [[np.float32, i, [64]] for i in format_list]
self.add_alpha_result(shape_format)
def test_add_shape_format_fp16_2d(self):
format_list = [0, 3, 29]
shape_format = [[np.float16, i, [5, 256]] for i in format_list]
self.add_alpha_result(shape_format)
def test_add_shape_format_fp32_2d(self):
format_list = [0, 3, 29]
shape_format = [[np.float32, i, [5, 256]] for i in format_list]
self.add_alpha_result(shape_format)
def test_add_shape_format_fp16_3d(self):
format_list = [0, 3, 29]
shape_format = [[np.float16, i, [32, 3, 3]] for i in format_list]
self.add_alpha_result(shape_format)
def test_add_shape_format_fp32_3d(self):
format_list = [0, 3, 29]
shape_format = [[np.float32, i, [32, 3, 3]] for i in format_list]
self.add_alpha_result(shape_format)
def test_add_shape_format_fp16_4d(self):
format_list = [0, 3, 29]
shape_format = [[np.float16, i, [64, 112, 7, 7]] for i in format_list]
self.add_alpha_result(shape_format)
def test_add_shape_format_fp32_4d(self):
format_list = [0, 3, 29]
shape_format = [[np.float32, i, [64, 112, 7, 7]] for i in format_list]
self.add_alpha_result(shape_format)
def test_add_mix_dtype(self):
cpu_input1, npu_input1 = create_common_tensor([np.int32, 0, (2, 3)], 1, 100)
cpu_input2, npu_input2 = create_common_tensor([np.float32, 0, (2, 3)], 1, 100)
cpu_output = torch.add(cpu_input1, cpu_input2)
npu_output = torch.add(npu_input1, npu_input2)
npu_output = npu_output.to("cpu")
self.assertRtolEqual(cpu_output, npu_output)
def test_add_scalar_check_5d_5d_match(self):
ca = torch.randn(4)
cb = ca.view(2, 2).transpose(1, 0)
na = ca.npu()
nb = cb.npu()
caout = torch.add(ca, 1)
cbout = torch.add(cb, 1)
naout = torch.add(na, 1)
nbout = torch.add(nb, 1)
naout = naout.to("cpu")
nbout = nbout.to("cpu")
self.assertRtolEqual(caout, naout)
self.assertRtolEqual(cbout, nbout)
def test_add_different_dtype(self):
cpu_x1 = torch.rand(2, 3, 4)
cpu_other1 = torch.rand(2, 3, 4).uniform_(1, 10).long()
npu_x1 = cpu_x1.npu()
npu_other1 = cpu_other1.npu()
cpu_out1 = cpu_x1 + cpu_other1
npu_out1 = npu_x1 + npu_other1
cpu_x2 = 1.5
cpu_other2 = torch.rand(2, 3, 4).uniform_(1, 10).long()
npu_x2 = 1.5
npu_other2 = cpu_other2.npu()
cpu_out2 = cpu_x2 + cpu_other2
npu_out2 = npu_x2 + npu_other2
cpu_x3 = torch.rand(2, 3, 4).int()
cpu_other3 = 3
npu_x3 = cpu_x3.npu()
npu_other3 = 3
cpu_out3 = cpu_x3 + cpu_other3
npu_out3 = npu_x3 + npu_other3
self.assertRtolEqual(cpu_out1, npu_out1.cpu())
self.assertRtolEqual(cpu_out2, npu_out2.cpu())
self.assertRtolEqual(cpu_out3, npu_out3.cpu())
def test_add_inplace_and_out_mix_dtype(self):
dtype_list = [
[np.int32, np.int64, np.int64],
[np.int64, np.int32, np.int64],
[np.float32, np.float16, np.float32],
[np.float16, np.float32, np.float32],
[np.int64, np.float32, np.float32],
]
for item in dtype_list:
cpu_input1, npu_input1 = create_common_tensor(
[item[0], 0, (2, 3, 4)], -100, 100
)
cpu_input2, npu_input2 = create_common_tensor(
[item[1], 0, (2, 3, 4)], -100, 100
)
_, npu_output = create_common_tensor([item[2], 0, (2, 3, 4)], 1, 100)
if item[0] == np.int64 and item[1] == np.float32:
try:
npu_input1.add_(npu_input2)
except RuntimeError as e:
self.assertRegex(
str(e),
"can't be cast to the desired output type",
)
else:
cpu_input1.add_(cpu_input2)
npu_input1.add_(npu_input2)
self.assertRtolEqual(cpu_input1, npu_input1.cpu())
cpu_output = torch.add(cpu_input1, cpu_input2)
torch.add(npu_input1, npu_input2, out=npu_output)
self.assertRtolEqual(cpu_output, npu_output.cpu())
def test_tensor_add_fp16(self):
format_list = [0, 3, 29]
shape_format = [[np.float16, i, [5, 256]] for i in format_list]
for item in shape_format:
cpu_input1, npu_input1 = create_common_tensor(item, 0, 100)
cpu_input2, npu_input2 = create_common_tensor(item, 0, 100)
if cpu_input1.dtype == torch.float16:
cpu_input1 = cpu_input1.to(torch.float32)
cpu_input2 = cpu_input2.to(torch.float32)
cpu_output = self.cpu_tensor_op_exec(cpu_input1, cpu_input2)
npu_output = self.npu_tensor_op_exec(npu_input1, npu_input2)
cpu_output = cpu_output.astype(npu_output.dtype)
self.assertRtolEqual(cpu_output, npu_output)
def test_tensor_add_fp32(self):
format_list = [0, 3, 29]
shape_format = [[np.float32, i, [5, 256]] for i in format_list]
for item in shape_format:
cpu_input1, npu_input1 = create_common_tensor(item, 0, 100)
cpu_input2, npu_input2 = create_common_tensor(item, 0, 100)
cpu_output = self.cpu_tensor_op_exec(cpu_input1, cpu_input2)
npu_output = self.npu_tensor_op_exec(npu_input1, npu_input2)
cpu_output = cpu_output.astype(npu_output.dtype)
self.assertRtolEqual(cpu_output, npu_output)
def test_add_dtype_check(self):
tensor_cpu = torch.tensor([1.0, 2.0, 3.0, 4.0])
result_ge = tensor_cpu.ge(tensor_cpu)
result_add = result_ge.add(tensor_cpu, alpha=2.0)
result_mul = result_add.mul(tensor_cpu)
result_cpu = result_mul.sub(tensor_cpu)
tensor_npu = torch.tensor([1.0, 2.0, 3.0, 4.0]).npu()
result_ge = tensor_npu.ge(tensor_npu)
result_add = result_ge.add(tensor_npu, alpha=2.0)
result_mul = result_add.mul(tensor_npu)
result_npu = result_mul.sub(tensor_npu)
self.assertRtolEqual(result_cpu, result_npu)
if __name__ == "__main__":
run_tests()