From f0fb6809e30986af8b127f9b68841f72c993c022 Mon Sep 17 00:00:00 2001
From: Stefan Weil <sw@weilnetz.de>
Date: Wed, 14 Jul 2021 19:07:05 +0200
Subject: [PATCH] Use SIMD instructions for DotProductNative

Signed-off-by: Stefan Weil <sw@weilnetz.de>
---
 Makefile.am             | 3 +++
 configure.ac            | 4 +++-
 src/arch/dotproduct.cpp | 3 +++
 3 files changed, 9 insertions(+), 1 deletion(-)

diff --git a/Makefile.am b/Makefile.am
index e95f18ea1..d56f6e48b 100644
--- a/Makefile.am
+++ b/Makefile.am
@@ -144,6 +144,9 @@ noinst_HEADERS += src/arch/simddetect.h
 noinst_LTLIBRARIES += libtesseract_native.la
 
 libtesseract_native_la_CXXFLAGS = -O3 -ffast-math
+if OPENMP_SIMD
+libtesseract_native_la_CXXFLAGS += -fopenmp-simd -DOPENMP_SIMD
+endif
 if MARCH_NATIVE_OPT
 libtesseract_native_la_CXXFLAGS += -march=native -mtune=native
 endif
diff --git a/configure.ac b/configure.ac
index 37a6bf452..bc5bb745e 100644
--- a/configure.ac
+++ b/configure.ac
@@ -186,6 +186,8 @@ esac
 
 AX_CHECK_COMPILE_FLAG([-march=native], [arch_native=true], [arch_native=false], [$WERROR])
 AM_CONDITIONAL([MARCH_NATIVE_OPT], $arch_native)
+AX_CHECK_COMPILE_FLAG([-fopenmp-simd], [openmp_simd=true], [openmp_simd=false], [$WERROR])
+AM_CONDITIONAL([OPENMP_SIMD], $openmp_simd)
 
 AC_ARG_WITH([extra-includes],
             [AS_HELP_STRING([--with-extra-includes=DIR],
@@ -284,7 +286,7 @@ m4_define([MY_CHECK_FRAMEWORK],
     ])
    if test "$my_cv_framework_$1"="yes"; then
      AC_DEFINE(AS_TR_CPP([HAVE_FRAMEWORK_$1]), 1,
-            [Define if you have the  $1 framework])
+            [Define if you have the $1 framework])
      AS_TR_CPP([FRAMEWORK_$1])="-framework $1"
      AC_SUBST(AS_TR_CPP([FRAMEWORK_$1]))
    fi]
diff --git a/src/arch/dotproduct.cpp b/src/arch/dotproduct.cpp
index 62bcc00ce..07bf10d68 100644
--- a/src/arch/dotproduct.cpp
+++ b/src/arch/dotproduct.cpp
@@ -21,6 +21,9 @@ namespace tesseract {
 // Computes and returns the dot product of the two n-vectors u and v.
 double DotProductNative(const double *u, const double *v, int n) {
   double total = 0.0;
+#if defined(OPENMP_SIMD)
+#pragma omp simd reduction(+:total)
+#endif
   for (int k = 0; k < n; ++k) {
     total += u[k] * v[k];
   }