From c20eb9dca0aedb9f7a5a148a3c953bcba05279e0 Mon Sep 17 00:00:00 2001
From: John Cupitt <jcupitt@gmail.com>
Date: Tue, 15 Mar 2016 17:34:31 +0000
Subject: [PATCH] remove old unrolled case for reduce

not used any more
---
 TODO                         |  6 ++++-
 libvips/resample/reduceh.cpp | 52 +++---------------------------------
 libvips/resample/reducev.cpp | 48 +++------------------------------
 3 files changed, 13 insertions(+), 93 deletions(-)

diff --git a/TODO b/TODO
index f55b57ba..992bc548 100644
--- a/TODO
+++ b/TODO
@@ -1,7 +1,11 @@
-
 - try orc version of reducev? and shrinkv? maybe shrinkh?
 
+	valgrind --tool=callgrind vipsthumbnail wtc.jpg -s 2000 --vips-info
 
+	self	called
+	54%	256	reducev_gen
+	25%	256	reduceh_gen
+	18%	loads	various jpeg decode
 
 - try SEQ_UNBUFFERED on jpg source, get out of order error?
 
diff --git a/libvips/resample/reduceh.cpp b/libvips/resample/reduceh.cpp
index a32412c5..5a2cd54d 100644
--- a/libvips/resample/reduceh.cpp
+++ b/libvips/resample/reduceh.cpp
@@ -179,46 +179,6 @@ reduceh_unsigned_int_tab( VipsReduceh *reduceh,
 	}
 }
 
-/* A 6-point interpolation on uint8 is the most common case ... unroll that.
- *
- * The inner loop here won't vectorise, but our inner loop doesn't run for
- * long enough for vectorisation to be useful :-( gcc says it needs about an
- * 11-point kernel for the vector version to be worthwhile.
- */
-static void inline
-reduceh_unsigned_uint8_6tab( VipsPel *out, const VipsPel *in,
-	const int bands, const int *cx )
-{
-	const int b1 = bands;
-	const int b2 = b1 + b1;
-	const int b3 = b1 + b2;
-	const int b4 = b2 + b2;
-	const int b5 = b1 + b4;
-
-	const int c0 = cx[0];
-	const int c1 = cx[1];
-	const int c2 = cx[2];
-	const int c3 = cx[3];
-	const int c4 = cx[4];
-	const int c5 = cx[5];
-
-	for( int z = 0; z < bands; z++ ) {
-		int cubich = unsigned_fixed_round( 
-			c0 * in[0] +
-			c1 * in[b1] +
-			c2 * in[b2] +
-			c3 * in[b3] +
-			c4 * in[b4] +
-			c5 * in[b5] ); 
-
-		cubich = VIPS_CLIP( 0, cubich, 255 ); 
-
-		out[z] = cubich;
-
-		in += 1;
-	}
-}
-
 template <typename T, int min_value, int max_value>
 static void inline
 reduceh_signed_int_tab( VipsReduceh *reduceh,
@@ -376,14 +336,10 @@ vips_reduceh_gen( VipsRegion *out_region, void *seq,
 
 			switch( in->BandFmt ) {
 			case VIPS_FORMAT_UCHAR:
-				if( reduceh->n_points == 6 )
-					reduceh_unsigned_uint8_6tab( 
-						q, p, bands, cxi );
-				else
-					reduceh_unsigned_int_tab
-						<unsigned char, UCHAR_MAX>(
-						reduceh,
-						q, p, bands, cxi );
+				reduceh_unsigned_int_tab
+					<unsigned char, UCHAR_MAX>(
+					reduceh,
+					q, p, bands, cxi );
 				break;
 
 			case VIPS_FORMAT_CHAR:
diff --git a/libvips/resample/reducev.cpp b/libvips/resample/reducev.cpp
index 9fd2f263..cdfd8a92 100644
--- a/libvips/resample/reducev.cpp
+++ b/libvips/resample/reducev.cpp
@@ -108,42 +108,6 @@ reducev_unsigned_int_tab( VipsReducev *reducev,
 	}
 }
 
-/* An unrolled version of ^^ for the most common case. 
- */
-static void inline
-reducev_unsigned_uint8_6tab( VipsPel *out, const VipsPel *in,
-	const int ne, const int lskip, const int *cy )
-{
-	const int l1 = lskip;
-	const int l2 = l1 + l1;
-	const int l3 = l1 + l2;
-	const int l4 = l2 + l2;
-	const int l5 = l4 + l1;
-
-	const int c0 = cy[0];
-	const int c1 = cy[1];
-	const int c2 = cy[2];
-	const int c3 = cy[3];
-	const int c4 = cy[4];
-	const int c5 = cy[5];
-
-	for( int z = 0; z < ne; z++ ) {
-		int sum = unsigned_fixed_round( 
-			c0 * in[0] +
-			c1 * in[l1] +
-			c2 * in[l2] +
-			c3 * in[l3] +
-			c4 * in[l4] +
-			c5 * in[l5] ); 
-
-		sum = VIPS_CLIP( 0, sum, 255 ); 
-
-		out[z] = sum;
-
-		in += 1;
-	}
-}
-
 template <typename T, int min_value, int max_value>
 static void inline
 reducev_signed_int_tab( VipsReducev *reducev,
@@ -290,14 +254,10 @@ vips_reducev_gen( VipsRegion *out_region, void *seq,
 
 		switch( in->BandFmt ) {
 		case VIPS_FORMAT_UCHAR:
-			if( reducev->n_points == 6 )
-				reducev_unsigned_uint8_6tab( 
-					q, p, ne, lskip, cyi );
-			else
-				reducev_unsigned_int_tab
-					<unsigned char, UCHAR_MAX>(
-					reducev,
-					q, p, ne, lskip, cyi );
+			reducev_unsigned_int_tab
+				<unsigned char, UCHAR_MAX>(
+				reducev,
+				q, p, ne, lskip, cyi );
 			break;
 
 		case VIPS_FORMAT_CHAR: