|
- # Copyright 2019 Huawei Technologies Co., Ltd
- #
- # Licensed under the Apache License, Version 2.0 (the "License");
- # you may not use this file except in compliance with the License.
- # You may obtain a copy of the License at
- #
- # http://www.apache.org/licenses/LICENSE-2.0
- #
- # Unless required by applicable law or agreed to in writing, software
- # distributed under the License is distributed on an "AS IS" BASIS,
- # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- # See the License for the specific language governing permissions and
- # limitations under the License.
-
- import numpy as np
- import itertools
- from akg.utils import kernel_exec as utils
- from akg.ops.nn import maxpool_grad
- from akg.utils.dsl_create import cal_pad_shapes_by_strategy
- from tests.common.tensorio import compare_tensor
- from tests.common.base import get_rtol_atol
- from tests.common.gen_random import random_gaussian
-
-
- def maxpool_benchmark(input, kernel, stride, pad):
- sh, sw = stride
- [ph_h, ph_t, pw_h, pw_t], [out_size_h, out_size_w] = \
- cal_pad_shapes_by_strategy(input.shape, kernel, stride, pad)
- N, C1, H, W, C0 = input.shape
- KH, KW = kernel
-
- out_shape = (N, C1, out_size_h, out_size_w, C0)
-
- out = np.zeros(out_shape)
-
- inputpad = np.full((N, C1, H + ph_h + ph_t, W + pw_h + pw_t, C0),
- np.finfo(input.dtype).min, dtype=input.dtype)
- inputpad[:, :, ph_h: ph_h + H, pw_h: pw_h + W, :] = input
-
- for i in range(out_size_h):
- for j in range(out_size_w):
- out[:, :, i, j, :] = np.max(inputpad[:, :,
- i * sh:i * sh + KH, j * sw:j * sw + KW, :], axis=(2, 3))
- return out
-
-
- # behaviour
- # - 0: return dy only for the first maximum within one kernel
- # - 1: return dy for all maximums within one kernel
- def benchmark(x, y, dy, kernel, stride, pad, behaviour=0):
-
- kernel_h, kernel_w = kernel
- stride_h, stride_w = stride
- [pad_h_head, pad_h_tail, pad_w_head, pad_w_tail], _ = cal_pad_shapes_by_strategy(x.shape, kernel, stride, pad)
- N, C1, H, W, C0 = x.shape
- pad_shape = (N, C1, H + pad_h_tail + pad_h_head, W + pad_w_tail + pad_w_head, C0)
-
- padx = np.full(pad_shape, 0, dtype=x.dtype)
- padx[:, :, pad_h_head:(pad_h_head + H), pad_w_head:(pad_w_head + W), :] = x
-
- dx = np.zeros(padx.shape, dtype=x.dtype)
- _, _, yH, yW, _ = y.shape
-
- if behaviour == 0:
- for n in range(N):
- for c1 in range(C1):
- for yh in range(yH):
- for yw in range(yW):
- for c0 in range(C0):
- out_maxpool1 = y[n, c1, yh, yw, c0]
- head_maxpool1 = dy[n, c1, yh, yw, c0]
- for kh,kw in itertools.product(range(kernel_h), range(kernel_w)):
- if padx[n, c1, yh*stride_h + kh, yw*stride_w + kw, c0] == out_maxpool1:
- dx[n, c1, yh*stride_h + kh, yw*stride_w + kw, c0] += head_maxpool1
- break
- elif behaviour == 1:
- for n in range(N):
- for c1 in range(C1):
- for yh in range(yH):
- for yw in range(yW):
- for c0 in range(C0):
- out_maxpool1 = y[n, c1, yh, yw, c0]
- head_maxpool1 = dy[n, c1, yh, yw, c0]
- for kh in range(kernel_h):
- for kw in range(kernel_w):
- if padx[n, c1, yh*stride_h + kh, yw*stride_w + kw, c0] == out_maxpool1:
- dx[n, c1, yh*stride_h + kh, yw*stride_w + kw, c0] += head_maxpool1
-
- return dx[:, :, pad_h_head:(pad_h_head + H), pad_w_head:(pad_w_head + W), :]
-
-
- def maxpool_grad_run(shape, kernel, stride, pad, dtype, attrs):
- # Create op
- _, [yh, yw] = cal_pad_shapes_by_strategy(shape, kernel, stride, pad)
- y_shape = (shape[0], shape[1], yh, yw, shape[4])
-
- if 'tuning' in attrs.keys():
- t = attrs.get("tuning", False)
- kernel_name = attrs.get("kernel_name", False)
- mod = utils.op_build_test(maxpool_grad.maxpool_grad,
- [shape, y_shape, y_shape], [dtype, dtype, dtype],
- op_attrs=[kernel, stride, pad],
- kernel_name=kernel_name, attrs=attrs, dump_code=True, tuning=t)
- if t:
- dy, expect, output, x, y = \
- gen_data(dtype, kernel, pad, shape, stride, y_shape)
- return mod, expect, (x, y, dy, output)
- else:
- return mod
- else:
- mod = utils.op_build_test(maxpool_grad.maxpool_grad,
- [shape, y_shape, y_shape], [dtype, dtype, dtype],
- op_attrs=[kernel, stride, pad],
- kernel_name='maxpool_grad', attrs=attrs, dump_code=True)
- dy, expect, output, x, y = \
- gen_data(dtype, kernel, pad, shape, stride, y_shape)
- output = utils.mod_launch(mod, (x, y, dy, output), expect=expect)
- rtol, atol = get_rtol_atol("maxpool_grad", dtype)
- return (x, y, dy), output, expect, \
- compare_tensor(output, expect, rtol=rtol, atol=atol, equal_nan=True)
-
-
- def gen_data(dtype, kernel, pad, shape, stride, y_shape):
- # Generate data for testing the op
- x = random_gaussian(shape, miu=1, sigma=0.1).astype(dtype)
- y = maxpool_benchmark(x, kernel, stride, pad)
- y = y.astype(dtype)
- assert y.shape == y_shape, "y shape wrong"
- dy = random_gaussian(y_shape, miu=1, sigma=0.1).astype(dtype)
- expect = benchmark(x, y, dy, kernel, stride, pad)
- output_shape = shape
- # output tensor must be initialized to zero
- output = np.full(output_shape, 0, dtype)
- return dy, expect, output, x, y
|
