test/bench/shootout: more speedups

Most significant in mandelbrot, from avoiding MOVSD between registers,
but there are others.

R=golang-dev, rsc
CC=golang-dev
https://golang.org/cl/6258063

diff --git a/test/bench/shootout/timing.log b/test/bench/shootout/timing.log
index 941d8ff19c94bdddfd048b8710d23f92e78902d8..c89f562e8efe74274dea31a1d46fa5486cd744d6 100644 (file)
--- a/test/bench/shootout/timing.log
+++ b/test/bench/shootout/timing.log
@@ -1023,6 +1023,7 @@ meteor 2098
        gc meteor-contest       0.13u 0.00s 0.13r # -13%
        gc_B meteor-contest     0.13u 0.00s 0.13r # -7%
 
+# May 30, 2012.
 # After http://codereview.appspot.com/6261051, restoring old code generated
 # for floating-point constants. Mandelbrot is back to its previous numbers.
 
@@ -1031,3 +1032,34 @@ mandelbrot 16000
        gccgo -O2 mandelbrot.go 41.72u 0.01s 41.90r
        gc mandelbrot   60.62u 0.00s 60.76r
        gc_B mandelbrot 60.68u 0.00s 60.82r
+
+# May 30, 2012.
+# After http://codereview.appspot.com/6248068, better FP code
+# by avoiding MOVSD between registers.
+# Plus some other timing changes that have crept in from other speedups,
+# from garbage collection to Printf.
+
+fasta -n 25000000
+       gc fasta        1.76u 0.00s 1.76r # -12%
+       gc_B fasta      1.71u 0.00s 1.72r # -12%
+
+nbody -n 50000000
+       gc nbody        17.56u 0.00s 17.60r # -8%
+       gc_B nbody      17.30u 0.00s 17.34r # -10%
+
+fannkuch 12
+       gc fannkuch-parallel    155.92u 0.01s 44.05r # -15%
+
+k-nucleotide 1000000
+       gc k-nucleotide 9.22u 0.01s 9.26r # -5%
+       gc k-nucleotide-parallel        9.23u 0.03s 3.26r # -9%
+       gc_B k-nucleotide       9.22u 0.03s 9.28r # -2%
+
+mandelbrot 16000
+       gc mandelbrot   44.80u 0.00s 44.90r # -27%
+       gc_B mandelbrot 44.81u 0.00s 44.92r # -26%
+
+pidigits 10000
+       gc pidigits     3.51u 0.00s 3.52r # -6%
+       gc_B pidigits   3.51u 0.00s 3.52r # -6%
+