1 | /******************************************************************************* |
2 | * Copyright 2016-2022 Intel Corporation |
3 | * |
4 | * Licensed under the Apache License, Version 2.0 (the "License"); |
5 | * you may not use this file except in compliance with the License. |
6 | * You may obtain a copy of the License at |
7 | * |
8 | * http://www.apache.org/licenses/LICENSE-2.0 |
9 | * |
10 | * Unless required by applicable law or agreed to in writing, software |
11 | * distributed under the License is distributed on an "AS IS" BASIS, |
12 | * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
13 | * See the License for the specific language governing permissions and |
14 | * limitations under the License. |
15 | *******************************************************************************/ |
16 | |
17 | #ifndef CPU_X64_JIT_AVX2_1X1_CONV_KERNEL_F32_HPP |
18 | #define CPU_X64_JIT_AVX2_1X1_CONV_KERNEL_F32_HPP |
19 | |
20 | #include "common/c_types_map.hpp" |
21 | #include "common/memory.hpp" |
22 | #include "common/memory_tracking.hpp" |
23 | |
24 | #include "cpu/x64/injectors/jit_uni_postops_injector.hpp" |
25 | #include "cpu/x64/jit_generator.hpp" |
26 | #include "cpu/x64/jit_primitive_conf.hpp" |
27 | |
28 | namespace dnnl { |
29 | namespace impl { |
30 | namespace cpu { |
31 | namespace x64 { |
32 | |
33 | struct jit_avx2_1x1_conv_kernel_f32 : public jit_generator { |
34 | DECLARE_CPU_JIT_AUX_FUNCTIONS(jit_avx2_1x1_conv_kernel_f32) |
35 | |
36 | jit_avx2_1x1_conv_kernel_f32(const jit_1x1_conv_conf_t &ajcp, |
37 | const primitive_attr_t &attr, const memory_desc_t &dst_md); |
38 | |
39 | static status_t init_conf(jit_1x1_conv_conf_t &jcp, |
40 | const convolution_desc_t &cd, const memory_desc_wrapper &src_d, |
41 | const memory_desc_wrapper &weights_d, |
42 | const memory_desc_wrapper &dst_d, const primitive_attr_t &attr); |
43 | |
44 | static void init_scratchpad(memory_tracking::registrar_t &scratchpad, |
45 | const jit_1x1_conv_conf_t &jcp); |
46 | |
47 | jit_1x1_conv_conf_t jcp; |
48 | const primitive_attr_t &attr_; |
49 | |
50 | private: |
51 | std::unique_ptr<injector::jit_uni_postops_injector_t<avx2>> |
52 | postops_injector_; |
53 | |
54 | constexpr static int isa_simd_width_ |
55 | = cpu_isa_traits<avx2>::vlen / sizeof(float); |
56 | using reg64_t = const Xbyak::Reg64; |
57 | using ymm_t = const Xbyak::Ymm; |
58 | |
59 | reg64_t reg_bcast_data = rax; |
60 | reg64_t reg_load_data = rsi; |
61 | reg64_t reg_output_data = rbx; |
62 | reg64_t aux_reg_bcast_data = rdx; |
63 | reg64_t aux1_reg_bcast_data = abi_not_param1; |
64 | reg64_t aux_reg_load_data = abi_param1; |
65 | reg64_t aux_reg_output_data = rbp; |
66 | reg64_t reg_load_loop_work = r9; |
67 | reg64_t reg_bcast_loop_work = r10; |
68 | reg64_t reg_reduce_loop_work = r11; |
69 | reg64_t load_loop_iter = r13; |
70 | reg64_t bcast_loop_iter = r14; |
71 | reg64_t reduce_loop_iter = r15; |
72 | reg64_t reg_reduce_pos_flag = r8; |
73 | reg64_t reg_output_stride = r12; |
74 | reg64_t reg_bias_data = r12; |
75 | reg64_t reg_diff_bias_data = bcast_loop_iter; |
76 | reg64_t reg_tmp_output_stride = reg_bcast_data; |
77 | reg64_t reg_tmp = aux_reg_bcast_data; |
78 | reg64_t reg_output_stride_scale = load_loop_iter; |
79 | reg64_t reg_long_offt = reg_bcast_data; |
80 | |
81 | constexpr static int reg64_size_ = sizeof(int64_t); |
82 | constexpr static int reg_diff_bias_data_stack_offt = 0; |
83 | constexpr static int reg_binary_post_op_acc_off = 1 * reg64_size_; |
84 | constexpr static int reg_abi_param1_backup = 2 * reg64_size_; |
85 | constexpr static int reg_bcast_data_off = 3 * reg64_size_; |
86 | constexpr static int stack_space_needed = 4 * reg64_size_; |
87 | |
88 | ymm_t vreg_bcast = ymm_t(15); |
89 | ymm_t vtmp = ymm_t(14); |
90 | |
91 | void apply_postops( |
92 | const int load_loop_blk, const int ur, const int load_dim_tail); |
93 | void generate_bcast_loop(int load_loop_blk); |
94 | void generate_reduce_loop(int load_loop_blk, int ur); |
95 | void generate_diff_bias_loop(int load_loop_blk); |
96 | |
97 | void generate() override; |
98 | }; |
99 | |
100 | } // namespace x64 |
101 | } // namespace cpu |
102 | } // namespace impl |
103 | } // namespace dnnl |
104 | |
105 | #endif |
106 | |