1 /**
2 * Copyright 2022 Huawei Technologies Co., Ltd
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 #include "src/extendrt/delegate/tensorrt/cuda_impl/logical.cuh"
18 #include "src/extendrt/delegate/tensorrt/cuda_impl/cuda_helper.h"
19
20 template <typename T>
LogicalNotKernel(const T * input1,T * output,int element_cnt)21 __global__ void LogicalNotKernel(const T *input1, T *output, int element_cnt) {
22 for (int pos = blockIdx.x * blockDim.x + threadIdx.x; pos < element_cnt; pos += blockDim.x * gridDim.x) {
23 output[pos] = static_cast<T>(input1[pos] == 0);
24 }
25 }
26
27 template <typename T>
LogicalAndKernel(const T * input_addr1,const T * input_addr2,T * output,int size)28 __global__ void LogicalAndKernel(const T *input_addr1, const T *input_addr2, T *output, int size) {
29 for (int pos = blockIdx.x * blockDim.x + threadIdx.x; pos < size; pos += blockDim.x * gridDim.x) {
30 output[pos] = input_addr1[pos] * input_addr2[pos];
31 }
32 }
33
34 template <typename T>
LogicalOrKernel(const T * input_addr1,const T * input_addr2,T * output,int size)35 __global__ void LogicalOrKernel(const T *input_addr1, const T *input_addr2, T *output, int size) {
36 for (int pos = blockIdx.x * blockDim.x + threadIdx.x; pos < size; pos += blockDim.x * gridDim.x) {
37 T sum = input_addr1[pos] + input_addr2[pos];
38 output[pos] = static_cast<T>(sum > 0);
39 }
40 }
41
42 template <typename T>
GreaterOrEqualKernal(const T * input1,const T * input2,T * output,int element_cnt)43 __global__ void GreaterOrEqualKernal(const T *input1, const T *input2, T *output, int element_cnt) {
44 for (int pos = blockIdx.x * blockDim.x + threadIdx.x; pos < element_cnt; pos += blockDim.x * gridDim.x) {
45 output[pos] = (input1[pos] >= input2[pos]);
46 }
47 }
48
49 template <typename T>
LessOrEqualKernal(const T * input1,const T * input2,T * output,int element_cnt)50 __global__ void LessOrEqualKernal(const T *input1, const T *input2, T *output, int element_cnt) {
51 for (int pos = blockIdx.x * blockDim.x + threadIdx.x; pos < element_cnt; pos += blockDim.x * gridDim.x) {
52 output[pos] = (input1[pos] <= input2[pos]);
53 }
54 }
55
56 template <typename T>
LogicalNot(const T * input1,T * output,int element_cnt,cudaStream_t stream)57 void LogicalNot(const T *input1, T *output, int element_cnt, cudaStream_t stream) {
58 LogicalNotKernel<<<GET_BLOCKS(element_cnt), GET_THREADS, 0, stream>>>(input1, output, element_cnt);
59 }
60
61 template <typename T>
LogicalAnd(const T * input1,const T * input2,T * output,int element_cnt,cudaStream_t stream)62 void LogicalAnd(const T *input1, const T *input2, T *output, int element_cnt, cudaStream_t stream) {
63 LogicalAndKernel<<<GET_BLOCKS(element_cnt), GET_THREADS, 0, stream>>>(input1, input2, output, element_cnt);
64 }
65
66 template <typename T>
LogicalOr(const T * input1,const T * input2,T * output,int element_cnt,cudaStream_t stream)67 void LogicalOr(const T *input1, const T *input2, T *output, int element_cnt, cudaStream_t stream) {
68 LogicalOrKernel<<<GET_BLOCKS(element_cnt), GET_THREADS, 0, stream>>>(input1, input2, output, element_cnt);
69 }
70
71 template <typename T>
GreaterOrEqual(const T * input1,const T * input2,T * output,int element_cnt,cudaStream_t stream)72 void GreaterOrEqual(const T *input1, const T *input2, T *output, int element_cnt, cudaStream_t stream) {
73 GreaterOrEqualKernal<<<GET_BLOCKS(element_cnt), GET_THREADS, 0, stream>>>(input1, input2, output, element_cnt);
74 }
75
76 template <typename T>
LessOrEqual(const T * input1,const T * input2,T * output,int element_cnt,cudaStream_t stream)77 void LessOrEqual(const T *input1, const T *input2, T *output, int element_cnt, cudaStream_t stream) {
78 LessOrEqualKernal<<<GET_BLOCKS(element_cnt), GET_THREADS, 0, stream>>>(input1, input2, output, element_cnt);
79 }
80
81 template void GreaterOrEqual(const float *input1, const float *input2, float *output, int element_cnt,
82 cudaStream_t stream);
83
84 template void GreaterOrEqual(const int *input1, const int *input2, int *output, int element_cnt, cudaStream_t stream);
85
86 template void LessOrEqual(const float *input1, const float *input2, float *output, int element_cnt,
87 cudaStream_t stream);
88
89 template void LessOrEqual(const int *input1, const int *input2, int *output, int element_cnt, cudaStream_t stream);
90
91 template void LogicalNot(const int32_t *input1, int32_t *output, int element_cnt, cudaStream_t stream);
92
93 template void LogicalAnd(const int32_t *input1, const int32_t *input2, int32_t *output, int element_cnt,
94 cudaStream_t stream);
95
96 template void LogicalOr(const int32_t *input1, const int32_t *input2, int32_t *output, int element_cnt,
97 cudaStream_t stream);
98