From: Matthew Fernandez <matthew.fernandez@gmail.com>
Date: Thu, 15 Oct 2020 02:32:58 +0000 (-0700)
Subject: inline and fuse a number of loops in majorization code
X-Git-Tag: 2.46.0~20^2^2~26^2~1
X-Git-Url: https://granicus.if.org/sourcecode?a=commitdiff_plain;h=d336cdacee0d51e0974c67010cb4d4a86c526e9c;p=graphviz

inline and fuse a number of loops in majorization code

This accelerates the Neato example from #1652 by ~9.5%.
---

diff --git a/lib/neatogen/stress.c b/lib/neatogen/stress.c
index f43ecb39b..8d840e895 100644
--- a/lib/neatogen/stress.c
+++ b/lib/neatogen/stress.c
@@ -1166,12 +1166,11 @@ int stress_majorization_kD_mkernel(vtx_data * graph,	/* Input graph in sparse re
 
 	    /* put into 'dist_accumulator' all squared distances between 'i' and 'i'+1,...,'n'-1 */
 	    for (k = 0; k < dim; k++) {
-		set_vector_valf(len, coords[k][i], tmp_coords);
-		vectors_mult_additionf(len, tmp_coords, -1,
-				       coords[k] + i + 1);
-		square_vec(len, tmp_coords);
-		vectors_additionf(len, tmp_coords, dist_accumulator,
-				  dist_accumulator);
+		size_t x;
+		for (x = 0; x < (size_t)len; ++x) {
+		    float tmp = coords[k][i] + -1.0f * (coords[k] + i + 1)[x];
+		    dist_accumulator[x] += tmp * tmp;
+		}
 	    }
 
 	    /* convert to 1/d_{ij} */