Mercurial > hgrepos > Python2 > PyMuPDF

comparison mupdf-source/thirdparty/tesseract/src/arch/intsimdmatrixrvv.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: intsimdmatrixrvv.cpp
3 // Description: matrix-vector product for 8-bit data on rvv.
4 // Author: sunyuechi
5 //
6 // Copyright (c) 2024 Institute of Software Chinese Academy of Sciences (ISCAS).
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 #ifdef HAVE_CONFIG_H
19 # include "config_auto.h" // for HAVE_RVV, ...
20 #endif
21
22 #if HAVE_RVV
23 # include "intsimdmatrix.h"
24 # include "tesstypes.h"
25
26 namespace tesseract {
27
28 static int DotProduct(const int8_t *u, const int8_t *v, int num) {
29 int total = 0;
30
31 asm __volatile__ (
32 " .option arch, +v \n\t"
33 " vsetvli t0,zero,e32,m8,ta,ma \n\t"
34 " vmv.v.i v0,0 \n\t"
35 "1: \n\t"
36 " vsetvli t0,%[num],e8,m2,ta,ma \n\t"
37 " vle8.v v16,0(%[u]) \n\t"
38 " vle8.v v24,0(%[v]) \n\t"
39 " sub %[num],%[num],t0 \n\t"
40 " vwmul.vv v8,v24,v16 \n\t"
41 " add %[u],%[u],t0 \n\t"
42 " add %[v],%[v],t0 \n\t"
43 " vsetvli zero,zero,e16,m4,tu,ma \n\t"
44 " vwadd.wv v0,v0,v8 \n\t"
45 " bnez %[num],1b \n\t"
46 " vsetvli t0,zero,e32,m8,ta,ma \n\t"
47 " vmv.s.x v8,zero \n\t"
48 " vredsum.vs v0,v0,v8 \n\t"
49 " vmv.x.s %[total],v0 \n\t"
50 : [u] "+r" (u),
51 [v] "+r" (v),
52 [num] "+r" (num),
53 [total] "+r" (total)
54 :
55 : "cc", "memory"
56 );
57
58 return total;
59 }
60
61 static void matrixDotVector(int dim1, int dim2, const int8_t *wi, const TFloat *scales,
62 const int8_t *u, TFloat *v) {
63 int num_out = dim1;
64 int num_in = dim2 - 1;
65 for (int i = 0; i < num_out; ++i) {
66 const int8_t *wi_start = wi + i * dim2;
67 int total = DotProduct(wi_start, u, num_in);
68 // Add in the bias and apply scaling.
69 v[i] = (total + wi_start[num_in] * INT8_MAX) * scales[i];
70 }
71 }
72
73 const IntSimdMatrix IntSimdMatrix::intSimdMatrixRVV = {
74 // Function.
75 matrixDotVector,
76 // Number of 32 bit outputs held in each register.
77 1,
78 // Maximum number of registers that we will use to hold outputs.
79 1,
80 // Number of 8 bit inputs in the inputs register.
81 1,
82 // Number of inputs in each weight group.
83 1
84 };
85
86 } // namespace tesseract.
87
88 #endif /* HAVE_RVV */