From e6edcfe20dc944dd4c90f81fa514a70a89eb0823 Mon Sep 17 00:00:00 2001
From: John Cupitt <jcupitt@gmail.com>
Date: Wed, 13 Jun 2018 16:23:21 +0100
Subject: [PATCH] improve accuracy of vector path

---
 ChangeLog                   | 1 +
 libvips/convolution/convi.c | 2 +-
 2 files changed, 2 insertions(+), 1 deletion(-)

diff --git a/ChangeLog b/ChangeLog
index 410f6bba..47fc40e5 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -5,6 +5,7 @@
 - configure test for g++ 7.2 and composite.cpp
 - don't Ping in magickload, too unreliable
 - ensure WebP can add metadata when compiled with libwebpmux [lovell]
+- improve accuracy of vector path convolution [felixbuenemann]
 
 12/2/18 started 8.6.3
 - use pkg-config to find libjpeg, if we can
diff --git a/libvips/convolution/convi.c b/libvips/convolution/convi.c
index 18b812cf..527c15e3 100644
--- a/libvips/convolution/convi.c
+++ b/libvips/convolution/convi.c
@@ -909,7 +909,7 @@ vips_convi_intize( VipsConvi *convi, VipsImage *M )
 	 * later, so 1.0 (for example) would become 128, which is outside
 	 * signed 8 bit. 
 	 */
-	shift = ceil( log2( mx + 1 ) );
+	shift = ceil( log2( mx ) + 1 );
 
 	/* We need to sum n_points, so we have to shift right before adding a
 	 * new value to make sure we have enough range.