1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
/*******************************************************************************
* Copyright 2022 Intel Corporation
*
* 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.
*******************************************************************************/
#ifndef GPU_INTEL_CONV_JIT_NORMALIZATION_HPP
#define GPU_INTEL_CONV_JIT_NORMALIZATION_HPP
#include "gpu/intel/jit/ir/gemm_schedule.hpp"
#include "gpu/intel/jit/ir/post_ops.hpp"
#include "gpu/intel/jit/ir/tensor.hpp"
namespace dnnl {
namespace impl {
namespace gpu {
namespace intel {
namespace conv {
namespace jit {
using namespace intel::jit;
class problem_t;
class post_op_view_mapper_t : public intel::jit::post_op_view_mapper_t {
public:
post_op_view_mapper_t(const gemm_schedule_t &schedule, const problem_t &prb,
const zero_points_config_t &zp_cfg, const layout_t &zp_dst)
: intel::jit::post_op_view_mapper_t(schedule.c_view())
, has_external_src_zps_(zp_cfg.needs_src_conv_precalc
|| zp_cfg.needs_src_reorder_precalc)
, schedule_(schedule)
, prb_(prb)
, zp_dst_(zp_dst) {}
view_t create_view(const dsl::type_t &type, uint32_t mask) const override {
return intel::jit::post_op_view_mapper_t::create_view(
type, normalize_mask(mask));
}
view_t create_view(const memory_desc_t &md) const override;
view_t create_src_zp_view(uint32_t mask) const override;
view_t try_create_bias_view(uint32_t mask) const override;
// Checks if convolution computes output elements that are out of bound in
// the output tensor. This can happen due to spatial padding.
//
// For example for forward convolution OW is padded to OW_PADDED. Then if
// ow >= OW (out of bounds) and iw = ow * SW - PW + kw * (DW + 1) < IW (in
// bounds) convolution computes an out-of-bound element which is not
// generally zero. This requires special handling if there are post-ops
// followed the convolution.
bool is_spurious_spatial(const pvar_t &dim) const override;
bool need_to_restore_zero_padding() const override;
bool use_dst_in_sum_post_op() const override;
bool can_use_scales() const override;
bool can_use_simple_src_zps() const override {
return has_external_src_zps_;
}
private:
uint32_t normalize_mask(uint32_t orig_mask) const;
const bool has_external_src_zps_;
const gemm_schedule_t &schedule_;
const problem_t &prb_;
const layout_t &zp_dst_;
};
void normalize_layouts(layout_t &src_layout, layout_t &wei_layout,
layout_t &dst_layout, layout_t &bia_layout, bool with_groups, dim_t g,
dim_t ic, dim_t oc, bool is_dw, const std::array<int, 3> &dhw_map,
bool add_groups);
} // namespace jit
} // namespace conv
} // namespace intel
} // namespace gpu
} // namespace impl
} // namespace dnnl
#endif