From ea58420b637af02c376e158334421b79bd6c9b1c Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Jo=C3=A3o=20Silva?= <jgc3silva@gmail.com>
Date: Sun, 19 May 2024 22:47:10 +0100
Subject: [PATCH] Use lddqu instead of loadu since it may perform better in
 certain situations
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Signed-off-by: João Silva <jgc3silva@gmail.com>
---
 src/random/src/scramble.avx.c | 4 ++--
 src/random/src/scramble.sse.c | 6 +++---
 2 files changed, 5 insertions(+), 5 deletions(-)

diff --git a/src/random/src/scramble.avx.c b/src/random/src/scramble.avx.c
index fe7d4f45a..eb7d1191a 100644
--- a/src/random/src/scramble.avx.c
+++ b/src/random/src/scramble.avx.c
@@ -47,7 +47,7 @@ void scramble_data(unsigned char * _x,
     // apply static masks
     unsigned int i;
     for (i=0; i<t; i+=32) {
-        x = _mm256_loadu_si256((const __m256i_u *)&_x[i]);
+        x = _mm256_lddqu_si256((const __m256i_u *)&_x[i]);
 
 #if HAVE_AVX2
         x = _mm256_xor_si256(x, mask);
@@ -97,7 +97,7 @@ void unscramble_data_soft(unsigned char * _x,
     // apply static masks
     unsigned int i;
     for (i=0; i<t; i+=4) {
-        x = _mm256_loadu_si256((const __m256i_u *)&_x[8 * i]);
+        x = _mm256_lddqu_si256((const __m256i_u *)&_x[8 * i]);
         y = _mm256_sub_epi8(mask, x);
         x = _mm256_blendv_epi8(x, y, mask);
         _mm256_storeu_si256((__m256i_u *)&_x[8 * i], x);
diff --git a/src/random/src/scramble.sse.c b/src/random/src/scramble.sse.c
index 0ee97189c..75dcc9817 100644
--- a/src/random/src/scramble.sse.c
+++ b/src/random/src/scramble.sse.c
@@ -49,7 +49,7 @@ void scramble_data(unsigned char * _x,
     // apply static masks
     unsigned int i;
     for (i=0; i<t; i+=16) {
-        x = _mm_loadu_si128((const __m128i_u *)&_x[i]);
+        x = _mm_lddqu_si128((const __m128i_u *)&_x[i]);
 
         x = _mm_xor_si128(x, mask);
 
@@ -96,12 +96,12 @@ void unscramble_data_soft(unsigned char * _x,
     // apply static masks
     unsigned int i;
     for (i=0; i<t; i+=4) {
-        x = _mm_loadu_si128((const __m128i_u *)&_x[8 * i]);
+        x = _mm_lddqu_si128((const __m128i_u *)&_x[8 * i]);
         y = _mm_sub_epi8(mask01, x);
         x = _mm_blendv_epi8(x, y, mask01);
         _mm_storeu_si128((__m128i_u *)&_x[8 * i], x);
 
-        x = _mm_loadu_si128((const __m128i_u *)&_x[8 * i + 16]);
+        x = _mm_lddqu_si128((const __m128i_u *)&_x[8 * i + 16]);
         y = _mm_sub_epi8(mask23, x);
         x = _mm_blendv_epi8(x, y, mask23);
         _mm_storeu_si128((__m128i_u *)&_x[8 * i + 16], x);