Optimization 2

tools/optimization/block_split_params.txt

@@ -0,0 +1,4 @@
+VARIABLE		MIN		INITIAL		MAX
+BLOCK_SIZE_SHIFT	0		12		20
+BLOCK_CUTOFF		0		200		2000
+OBSERVATIONS_PER_CHECK	10		512		10000

tools/optimization/optimize_block_split_params.r

@@ -0,0 +1,23 @@
+#!/usr/bin/Rscript
+
+PARAMS = read.table("block_split_params.txt", header=TRUE, row.names=1)
+
+fn = function(params) {
+
+    params = as.integer(params)
+    params = pmax(params, PARAMS$MIN)
+    params = pmin(params, PARAMS$MAX)
+
+    env = paste(rownames(PARAMS), "=", params, sep='', collapse=' ')
+
+    cat(env, "\n")
+
+    result = system(paste(env, "./try_block_params.sh"), intern=TRUE)
+    stopifnot(attr(result, "status") == 0)
+    result = as.numeric(result)
+    cat(result, "\n")
+    return(result)
+}
+
+result = optim(PARAMS$INITIAL, fn)
+print(result)

tools/optimization/results.txt

@@ -0,0 +1,15 @@
+testdata -6:
+	BLOCK_SIZE_SHIFT=20 BLOCK_CUTOFF=182 MIN_BLOCK_LENGTH=8071 MAX_BLOCK_LENGTH=299707 OBSERVATIONS_PER_CHECK=553
+	8727764
+
+moby.txt -12:
+	BLOCK_SIZE_SHIFT=20 BLOCK_CUTOFF=239 MIN_BLOCK_LENGTH=36336 MAX_BLOCK_LENGTH=303025 OBSERVATIONS_PER_CHECK=772 
+	476063 
+
+silesia -6:
+	BLOCK_SIZE_SHIFT=20 BLOCK_CUTOFF=189 MIN_BLOCK_LENGTH=10225 MAX_BLOCK_LENGTH=323665 OBSERVATIONS_PER_CHECK=556 
+	67927205 
+
+0b7d62c8-1ae5-11e6-818e-ca27821f62e9.png -12:
+    BLOCK_SIZE_SHIFT=20 BLOCK_CUTOFF=193 MIN_BLOCK_LENGTH=21726 MAX_BLOCK_LENGTH=300470 OBSERVATIONS_PER_CHECK=517
+    596524 

tools/optimization/try_block_params.sh

@@ -0,0 +1,23 @@
+#!/bin/bash
+
+set -e
+
+if [ -z "$LEVEL" -o -z "$INPUTFILE" ]; then
+	echo "Must specify LEVEL and INPUTFILE" 1>&2
+	exit 1
+fi
+
+topdir="$(dirname "$0")/../.."
+
+git checkout -f "$topdir/src" > /dev/null
+
+sed -i -e	\
+"
+    s/\<block_size >> 12\>/block_size >> $BLOCK_SIZE_SHIFT/
+    s/\<200 \* stats->num_observations/$BLOCK_CUTOFF * stats->num_observations/
+    s/num_new_observations < 512\>/num_new_observations < $OBSERVATIONS_PER_CHECK/
+"	\
+"$topdir/src/deflate_compress.c"
+
+make -C "$topdir" -j BUILD_BENCHMARK_PROGRAM=yes > /dev/null
+"$topdir/benchmark" -l $LEVEL -s 100000000 $INPUTFILE | grep Compressed | cut -f 4 -d ' '