1 /*******************************************************************************
2 * Copyright 2018 Intel Corporation
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
8 * http://www.apache.org/licenses/LICENSE-2.0
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 *******************************************************************************/
17 #ifndef JIT_UNI_X8S8S32X_1x1_CONV_KERNEL_HPP
18 #define JIT_UNI_X8S8S32X_1x1_CONV_KERNEL_HPP
20 #include "c_types_map.hpp"
21 #include "type_helpers.hpp"
22 #include "jit_generator.hpp"
23 #include "jit_primitive_conf.hpp"
33 template <cpu_isa_t isa>
34 struct jit_uni_x8s8s32x_1x1_conv_fwd_kernel: public jit_generator {
35 DECLARE_CPU_JIT_AUX_FUNCTIONS(jit_uni_x8s8s32x_1x1_conv_fwd_kernel)
37 jit_uni_x8s8s32x_1x1_conv_fwd_kernel(jit_1x1_conv_conf_t ajcp,
38 const primitive_attr_t &attr): jcp(ajcp), attr_(attr)
41 jit_ker = (void (*)(jit_1x1_conv_call_s *))this->getCode();
44 static bool post_ops_ok(jit_1x1_conv_conf_t &jcp,
45 const primitive_attr_t &attr);
46 static status_t init_conf(jit_1x1_conv_conf_t &jcp,
47 const convolution_desc_t &cd, const memory_desc_wrapper &src_d,
48 const memory_desc_wrapper &weights_d,
49 const memory_desc_wrapper &dst_d,
50 const memory_desc_wrapper &bias_pd,
51 const primitive_attr_t &attr,
52 bool with_relu = false, float relu_negative_slope = 0.f);
54 jit_1x1_conv_conf_t jcp;
55 const primitive_attr_t &attr_;
56 void (*jit_ker)(jit_1x1_conv_call_s *);
59 using Vmm = typename utils::conditional3<isa == sse42, Xbyak::Xmm,
60 isa == avx2, Xbyak::Ymm, Xbyak::Zmm>::type;
62 Reg64 reg_weight_data = rsi;
63 Reg64 reg_src_data = abi_not_param1;
64 Reg64 reg_dst_data = rbx;
65 Reg64 reg_bias_data = r12;
67 Reg64 reg_scales = rdx;
68 Reg64 aux_reg_src_data = rdx;
69 Reg64 aux_reg_weight_data = rax;
70 Reg64 aux_reg_dst_data = rbp;
71 Reg64 reg_oc_loop_work = r9;
72 Reg64 reg_ow_loop_work = r10;
73 Reg64 reg_loop_os_iter = r14;
74 Reg64 reg_loop_ic_iter = r15;
76 Reg64 reg_scratch = r14;
78 Vmm vreg_sum_0 = Vmm(15);
79 Vmm vreg_src = Vmm(14);
80 Vmm vmm_bias = Vmm(15);
81 Vmm vmm_zero = Vmm(14);
82 Vmm vmm_one = Vmm(13);
83 Xmm xmm_one = Xmm(13);
85 void loop_os(int oc_loop_blk);
86 void ic_loop(int oc_loop_blk, int ur);
90 bool maybe_relu(int position);
91 void cvt2ps(data_type_t type_in, Vmm vmm_in, const Xbyak::Operand &op);