tesseract 4.1.1
Loading...
Searching...
No Matches
dotproductfma.cpp
Go to the documentation of this file.
1
2// File: dotproductfma.cpp
3// Description: Architecture-specific dot-product function.
4// Author: Stefan Weil
5//
6// (C) Copyright 2015, Google Inc.
7// Licensed under the Apache License, Version 2.0 (the "License");
8// you may not use this file except in compliance with the License.
9// You may obtain a copy of the License at
10// http://www.apache.org/licenses/LICENSE-2.0
11// Unless required by applicable law or agreed to in writing, software
12// distributed under the License is distributed on an "AS IS" BASIS,
13// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
14// See the License for the specific language governing permissions and
15// limitations under the License.
17
18#if !defined(__FMA__)
19#error Implementation only for FMA capable architectures
20#endif
21
22#include <immintrin.h>
23#include <cstdint>
24#include "dotproduct.h"
25
26namespace tesseract {
27
28// Computes and returns the dot product of the n-vectors u and v.
29// Uses Intel FMA intrinsics to access the SIMD instruction set.
30double DotProductFMA(const double* u, const double* v, int n) {
31 const unsigned quot = n / 8;
32 const unsigned rem = n % 8;
33 __m256d t0 = _mm256_setzero_pd();
34 __m256d t1 = _mm256_setzero_pd();
35 for (unsigned k = 0; k < quot; k++) {
36 __m256d f0 = _mm256_loadu_pd(u);
37 __m256d f1 = _mm256_loadu_pd(v);
38 t0 = _mm256_fmadd_pd(f0, f1, t0);
39 u += 4;
40 v += 4;
41 __m256d f2 = _mm256_loadu_pd(u);
42 __m256d f3 = _mm256_loadu_pd(v);
43 t1 = _mm256_fmadd_pd(f2, f3, t1);
44 u += 4;
45 v += 4;
46 }
47 t0 = _mm256_hadd_pd(t0, t1);
48 alignas(32) double tmp[4];
49 _mm256_store_pd(tmp, t0);
50 double result = tmp[0] + tmp[1] + tmp[2] + tmp[3];
51 for (unsigned k = 0; k < rem; k++) {
52 result += *u++ * *v++;
53 }
54 return result;
55}
56
57} // namespace tesseract.
double DotProductFMA(const double *u, const double *v, int n)