Use fixed depth bench to make PGO builds more reproducible
authorJoost Vandevondele <Joost.VandeVondele@gmail.com>
Sat, 17 Sep 2016 11:37:11 +0000 (13:37 +0200)
committerMarco Costalba <mcostalba@gmail.com>
Sun, 18 Sep 2016 06:13:34 +0000 (08:13 +0200)
Discussed on fishcooking

proposal and objdump verification:
https://groups.google.com/d/msg/fishcooking/4_ausUwMXP0/EGPsMYqOFAAJ

verified no significant speed difference between depth and time:
https://groups.google.com/d/msg/fishcooking/4_ausUwMXP0/KazW5QZmFgAJ

stockfish_time - stats:
mean = 2207232.56        std = 7079.51        std/mean = 0.003207

stockfish_depth - stats:
mean = 2201783.57        std = 6356.69        std/mean = 0.002887

No functional change

src/Makefile

index cb55c8b..e5564ef 100644 (file)
@@ -32,7 +32,7 @@ PREFIX = /usr/local
 BINDIR = $(PREFIX)/bin
 
 ### Built-in benchmark for pgo-builds
-PGOBENCH = ./$(EXE) bench 16 1 1000 default time
+PGOBENCH = ./$(EXE) bench
 
 ### Object files
 OBJS = benchmark.o bitbase.o bitboard.o endgame.o evaluate.o main.o \