diff --git a/textproc/krep/Makefile b/textproc/krep/Makefile
index 000f9c2c2e70..ea8965e9c32e 100644
--- a/textproc/krep/Makefile
+++ b/textproc/krep/Makefile
@@ -1,21 +1,21 @@
 PORTNAME=	krep
 DISTVERSIONPREFIX=	v
-DISTVERSION=	2.3.0
+DISTVERSION=	2.4.0
 CATEGORIES=	textproc
 
 MAINTAINER=	alven@FreeBSD.org
 COMMENT=	High-performance string search utility
 WWW=		https://github.com/davidesantangelo/krep/
 
 LICENSE=	BSD2CLAUSE
 LICENSE_FILE=	${WRKSRC}/LICENSE
 
 USES=		gmake
 USE_GITHUB=	yes
 GH_ACCOUNT=	davidesantangelo
 
 TEST_TARGET=	test
 
 PLIST_FILES=	bin/krep
 
 .include <bsd.port.mk>
diff --git a/textproc/krep/distinfo b/textproc/krep/distinfo
index 46beadf789ff..8443b1e45453 100644
--- a/textproc/krep/distinfo
+++ b/textproc/krep/distinfo
@@ -1,3 +1,3 @@
-TIMESTAMP = 1776331755
-SHA256 (davidesantangelo-krep-v2.3.0_GH0.tar.gz) = 058d02e9a34861b92695073ccc3243ca79bdc9b0dcc5b9e798720a22786d5ae6
-SIZE (davidesantangelo-krep-v2.3.0_GH0.tar.gz) = 80293
+TIMESTAMP = 1779654012
+SHA256 (davidesantangelo-krep-v2.4.0_GH0.tar.gz) = 1b0648938dae88b17c80108a6df83768beb5f061c5bbb65edeafd169c799ea8c
+SIZE (davidesantangelo-krep-v2.4.0_GH0.tar.gz) = 83147
diff --git a/textproc/krep/files/patch-krep.c b/textproc/krep/files/patch-krep.c
index f0215d13d627..1d6dbad2b9aa 100644
--- a/textproc/krep/files/patch-krep.c
+++ b/textproc/krep/files/patch-krep.c
@@ -1,39 +1,30 @@
---- krep.c.orig	2026-03-18 19:36:12 UTC
+--- krep.c.orig	2026-05-19 07:06:16 UTC
 +++ krep.c
-@@ -4503,6 +4503,27 @@ uint64_t memchr_short_search(const search_params_t *pa
- }
- 
- #ifdef __ARM_NEON
+@@ -4925,6 +4925,27 @@ uint64_t memchr_short_search(const search_params_t *pa
+ // Handles case-sensitive patterns of any length.
+ // Uses 2x 16-byte loads per iteration with fast mask extraction via
+ // vshrn/vmovn instead of store-to-memory.
 +
 +#ifdef __aarch64__
 +static inline bool
 +is_nonzero(uint8x16_t a)
 +{
 +	return (vmaxvq_u8(a) != 0);
 +}
 +#else
 +/* no vmaxvq_u8() on AArch32 */
 +static inline bool
 +is_nonzero(uint8x16_t a)
 +{
 +	uint8x8_t a8, a4;
 +
 +	a8 = vmax_u8(vget_low_u8(a), vget_high_u8(a));
 +	a4 = vpmax_u8(a8, a8);
 +
 +	return (vget_lane_u32(a4, 0) != 0);
 +}
 +#endif /* defined(__aarch64__) */
 +
  uint64_t neon_search(const search_params_t *params,
                       const char *text_start,
                       size_t text_len,
-@@ -4541,7 +4562,7 @@ uint64_t neon_search(const search_params_t *params,
- 
-         // Check if any match found
-         // vmaxvq_u8 returns the maximum value across the vector. If any byte matched (0xFF), result is 0xFF.
--        if (vmaxvq_u8(cmp) != 0)
-+        if (is_nonzero(cmp))
-         {
-             // Extract mask to find exact positions
-             // Since NEON doesn't have a direct movemask, we simulate it or iterate.