From: Konstantinos Margaritis <konstantinos@vectorcamp.gr>
Date: Wed, 7 Sep 2022 10:20:01 +0000 (+0000)
Subject: [NEON] optimize mask1bit128, get rid of simd_onebit_masks
X-Git-Tag: vectorscan/5.4.8~1^2~2^2~3
X-Git-Url: http://git.ipfire.org/cgi-bin/gitweb.cgi?a=commitdiff_plain;h=0af2ba86165c469361fbfd9f34fd70aa2a53213d;p=thirdparty%2Fvectorscan.git

[NEON] optimize mask1bit128, get rid of simd_onebit_masks
---

diff --git a/src/util/arch/arm/simd_utils.h b/src/util/arch/arm/simd_utils.h
index 6447996c..45bcd23c 100644
--- a/src/util/arch/arm/simd_utils.h
+++ b/src/util/arch/arm/simd_utils.h
@@ -577,9 +577,9 @@ m128 variable_byte_shift_m128(m128 in, s32 amount) {
 static really_inline
 m128 mask1bit128(unsigned int n) {
     assert(n < sizeof(m128) * 8);
-    u32 mask_idx = ((n % 8) * 64) + 95;
-    mask_idx -= n / 8;
-    return loadu128(&simd_onebit_masks[mask_idx]);
+    static m128 onebit = { 1, 0 };
+    m128 mask = lshiftbyte_m128( onebit, n / 8 );
+    return lshift64_m128( mask, n % 8 );
 }
 
 // switches on bit N in the given vector.