Mercurial > hgrepos > Python2 > PyMuPDF

comparison mupdf-source/thirdparty/tesseract/src/arch/dotproductfma.cpp @ 2:b50eed0cc0ef upstream

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
ADD: MuPDF v1.26.7: the MuPDF source as downloaded by a default build of PyMuPDF 1.26.4. The directory name has changed: no version number in the expanded directory now.
author Franz Glasner <fzglas.hg@dom66.de>
date Mon, 15 Sep 2025 11:43:07 +0200
parents
children
comparison
equal deleted inserted replaced
1:1d09e1dec1d9 2:b50eed0cc0ef
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.
16 ///////////////////////////////////////////////////////////////////////
17
18 #if !defined(__FMA__)
19 # if defined(__i686__) || defined(__x86_64__)
20 # error Implementation only for FMA capable architectures
21 # endif
22 #else
23
24 # include <immintrin.h>
25 # include <cstdint>
26 # include "dotproduct.h"
27
28 namespace tesseract {
29
30 // Computes and returns the dot product of the n-vectors u and v.
31 // Uses Intel FMA intrinsics to access the SIMD instruction set.
32 #if defined(FAST_FLOAT)
33 float DotProductFMA(const float *u, const float *v, int n) {
34 const unsigned quot = n / 16;
35 const unsigned rem = n % 16;
36 __m256 t0 = _mm256_setzero_ps();
37 __m256 t1 = _mm256_setzero_ps();
38 for (unsigned k = 0; k < quot; k++) {
39 __m256 f0 = _mm256_loadu_ps(u);
40 __m256 f1 = _mm256_loadu_ps(v);
41 t0 = _mm256_fmadd_ps(f0, f1, t0);
42 u += 8;
43 v += 8;
44 __m256 f2 = _mm256_loadu_ps(u);
45 __m256 f3 = _mm256_loadu_ps(v);
46 t1 = _mm256_fmadd_ps(f2, f3, t1);
47 u += 8;
48 v += 8;
49 }
50 t0 = _mm256_hadd_ps(t0, t1);
51 alignas(32) float tmp[8];
52 _mm256_store_ps(tmp, t0);
53 float result = tmp[0] + tmp[1] + tmp[2] + tmp[3] + tmp[4] + tmp[5] + tmp[6] + tmp[7];
54 for (unsigned k = 0; k < rem; k++) {
55 result += *u++ * *v++;
56 }
57 return result;
58 }
59 #else
60 double DotProductFMA(const double *u, const double *v, int n) {
61 const unsigned quot = n / 8;
62 const unsigned rem = n % 8;
63 __m256d t0 = _mm256_setzero_pd();
64 __m256d t1 = _mm256_setzero_pd();
65 for (unsigned k = 0; k < quot; k++) {
66 __m256d f0 = _mm256_loadu_pd(u);
67 __m256d f1 = _mm256_loadu_pd(v);
68 t0 = _mm256_fmadd_pd(f0, f1, t0);
69 u += 4;
70 v += 4;
71 __m256d f2 = _mm256_loadu_pd(u);
72 __m256d f3 = _mm256_loadu_pd(v);
73 t1 = _mm256_fmadd_pd(f2, f3, t1);
74 u += 4;
75 v += 4;
76 }
77 t0 = _mm256_hadd_pd(t0, t1);
78 alignas(32) double tmp[4];
79 _mm256_store_pd(tmp, t0);
80 double result = tmp[0] + tmp[1] + tmp[2] + tmp[3];
81 for (unsigned k = 0; k < rem; k++) {
82 result += *u++ * *v++;
83 }
84 return result;
85 }
86 #endif
87
88 } // namespace tesseract.
89
90 #endif