From: Rob Pike <r@golang.org>
Date: Tue, 11 Aug 2009 00:04:59 +0000 (-0700)
Subject: new timing data after compiler updates (fp regs, integer div and mod)
X-Git-Tag: weekly.2009-11-06~918
X-Git-Url: http://www.git.cypherpunks.su/?a=commitdiff_plain;h=ea8197cb4594ce5abd0b2bdb711d38b525b9edc5;p=gostls13.git

new timing data after compiler updates (fp regs, integer div and mod)

R=rsc
DELTA=90  (90 added, 0 deleted, 0 changed)
OCL=32993
CL=33001
---

diff --git a/test/bench/timing.log b/test/bench/timing.log
index 3df6714dfd..e73d061b19 100644
--- a/test/bench/timing.log
+++ b/test/bench/timing.log
@@ -141,3 +141,93 @@ threadring 50000000
 chameneos 6000000
 	gcc -O2 chameneosredux.c -lpthread	18.14u 276.52s 76.93r
 	gc chameneosredux	20.19u 0.01s 20.23r
+
+Aug 10 2009
+
+# new 6g with better fp registers, fast div and mod of integers
+# complete set of timings listed. significant changes marked ***
+
+fasta -n 25000000
+	# probably I/O library inefficiencies
+	gcc -O2 fasta.c	5.96u 0.00s 5.97r
+	gc fasta	10.59u 0.01s 10.61r
+	gc_B fasta	9.92u 0.02s 9.95r
+
+reverse-complement < output-of-fasta-25000000
+	# we don't know - memory cache behavior?
+	gcc -O2 reverse-complement.c	1.96u 1.56s 16.23r
+	gccgo -O2 reverse-complement.go	6.41u 0.62s 7.05r
+	gc reverse-complement	6.46u 0.70s 7.17r
+	gc_B reverse-complement	6.22u 0.72s 6.95r
+
+nbody 50000000
+	# math.Sqrt needs to be in assembly; inlining is probably the other 50%
+	gcc -O2 nbody.c	21.26u 0.01s 21.28r
+	gccgo -O2 nbody.go	116.68u 0.07s 116.80r
+	gc nbody	86.64u 0.01s 86.68r	# -14%
+	gc_B nbody	85.72u 0.02s 85.77r	# *** -17%
+
+binary-tree 15 # too slow to use 20
+	# memory allocation and garbage collection
+	gcc -O2 binary-tree.c -lm	0.87u 0.00s 0.87r
+	gccgo -O2 binary-tree.go	1.61u 0.47s 2.09r
+	gccgo -O2 binary-tree-freelist.go	0.00u 0.00s 0.01r
+	gc binary-tree	9.11u 0.01s 9.13r	# *** -5%
+	gc binary-tree-freelist	0.47u 0.01s 0.48r
+
+fannkuch 12
+	# bounds checking is half the difference
+	# rest might be registerization
+	gcc -O2 fannkuch.c	59.92u 0.00s 59.94r
+	gccgo -O2 fannkuch.go	65.54u 0.00s 65.58r
+	gc fannkuch	123.98u 0.01s 124.04r
+	gc_B fannkuch	90.75u 0.00s 90.78r
+
+regex-dna 100000
+	# regexp code is slow on trivial regexp
+	gcc -O2 regex-dna.c -lpcre	0.91u 0.00s 0.92r
+	gc regex-dna	27.25u 0.02s 27.28r
+	gc_B regex-dna	29.51u 0.03s 29.55r
+
+spectral-norm 5500
+	# possibly inline evalA
+	gcc -O2 spectral-norm.c -lm	11.57u 0.00s 11.57r
+	gccgo -O2 spectral-norm.go	12.07u 0.01s 12.08r
+	gc spectral-norm	23.99u 0.00s 24.00r
+	gc_B spectral-norm	23.73u 0.00s 23.75r
+
+k-nucleotide 1000000
+	# string maps are slower than glib string maps
+	gcc -O2 -I/usr/include/glib-2.0 -I/usr/lib/glib-2.0/include k-nucleotide.c -lglib-2.0	10.63u 0.02s 10.69r
+	gccgo -O2 k-nucleotide.go	23.19u 0.91s 24.12r
+	gc k-nucleotide	16.73u 0.04s 16.78r	# *** +5% (but this one seems to vary by more than that)
+	gc_B k-nucleotide	16.46u 0.04s 16.51r	# *** +5%
+
+mandelbrot 16000
+	gcc -O2 mandelbrot.c	56.16u 0.00s 56.16r
+	gccgo -O2 mandelbrot.go	57.41u 0.01s 57.42r
+	gc mandelbrot	64.05u 0.02s 64.08r	# *** -14%
+	gc_B mandelbrot	64.10u 0.02s 64.14r	# *** -14%
+
+meteor 16000
+	# we don't know
+	gcc -O2 meteor-contest.c	0.10u 0.00s 0.10r
+	gccgo -O2 meteor-contest.go	0.12u 0.00s 0.12r
+	gc meteor-contest	0.18u 0.00s 0.20r	# *** -25%
+	gc_B meteor-contest	0.17u 0.00s 0.18r	# *** -24%
+
+pidigits 10000
+	# bignum is slower than gmp
+	gcc -O2 pidigits.c -lgmp	2.57u 0.00s 2.57r
+	gc pidigits	71.82u 0.04s 71.89r
+	gc_B pidigits	71.84u 0.08s 71.98r
+
+threadring 50000000
+	gcc -O2 threadring.c -lpthread	30.91u 164.33s 204.57r
+	gccgo -O2 threadring.go	87.12u 460.04s 447.61r
+	gc threadring	38.55u 0.00s 38.56r	# *** +16%
+
+chameneos 6000000
+	gcc -O2 chameneosredux.c -lpthread	17.93u 323.65s 88.47r
+	gc chameneosredux	21.72u 0.00s 21.73r
+