dgryski

<dgryski@kamek[powbench] \ʕ◔ϖ◔ʔ/ > go test -test.bench=.
BenchmarkPolyPow-4    	 5000000	       286 ns/op
BenchmarkPolyFast-4   	100000000	        16.7 ns/op
PASS
ok  	github.com/dgryski/powbench	3.430s

dgryski

#!/usr/bin/perl

use warnings;
use strict;

my $bin=$ARGV[0];

my @line = split /\s+/, `nm $bin |grep runtime.buildVersion`;
my $addr = hex($line[0]);
my $end = $addr + 16;

dgryski

TEXT main.main(SB) /home/dgryski/Dropbox/GITS/gocode/src/github.com/dgryski/peachpy-examples/struct/main.go
	main.go:11	0x401000	64488b0c25f8ffffff	FS MOVQ FS:0xfffffff8, CX
	main.go:11	0x401009	483b6110		CMPQ 0x10(CX), SP
	main.go:11	0x40100d	0f86de000000		JBE 0x4010f1
	main.go:11	0x401013	4883ec60		SUBQ $0x60, SP
	main.go:11	0x401017	48896c2458		MOVQ BP, 0x58(SP)
	main.go:11	0x40101c	488d6c2458		LEAQ 0x58(SP), BP
	main.go:12	0x401021	488d05585d0900		LEAQ 0x95d58(IP), AX
	main.go:12	0x401028	48890424		MOVQ AX, 0(SP)
	main.go:12	0x40102c	e83fcb0000		CALL runtime.newobject(SB)

dgryski

package main

import (
	"bufio"
	"errors"
	"flag"
	"fmt"
	"log"
	"os"
	"runtime"

dgryski

from peachpy import *
from peachpy.x86_64 import *

r = Argument(ptr(const_uint64_t))
bits = Argument(ptr(const_uint64_t))
hashes = Argument(ptr(const_uint16_t))
length = Argument(size_t)

with Function("querySSE", (r, bits, hashes, length), float_, target=uarch.default + isa.fma3) as function:
    reg_r = GeneralPurposeRegister64()

dgryski

# github.com/dgryski/w
"".BenchmarkBar t=1 size=35 args=0x8 locals=0x0
	0x0000 00000 (/home/dgryski/Dropbox/GITS/gocode/src/github.com/dgryski/w/main_test.go:9)	TEXT	"".BenchmarkBar(SB), $0-8
	0x0000 00000 (/home/dgryski/Dropbox/GITS/gocode/src/github.com/dgryski/w/main_test.go:9)	NOP
	0x0000 00000 (/home/dgryski/Dropbox/GITS/gocode/src/github.com/dgryski/w/main_test.go:9)	NOP
	0x0000 00000 (/home/dgryski/Dropbox/GITS/gocode/src/github.com/dgryski/w/main_test.go:9)	FUNCDATA	$0, gclocals·87d20ce1b58390b294df80b886db78bf(SB)
	0x0000 00000 (/home/dgryski/Dropbox/GITS/gocode/src/github.com/dgryski/w/main_test.go:9)	FUNCDATA	$1, gclocals·33cdeccccebe80329f1fdbee7f5874cb(SB)
	0x0000 00000 (/home/dgryski/Dropbox/GITS/gocode/src/github.com/dgryski/w/main_test.go:10)	MOVQ	"".b+8(FP), AX
	0x0005 00005 (/home/dgryski/Dropbox/GITS/gocode/src/github.com/dgryski/w/main_test.go:10)	MOVQ	$0, CX
	0x0007 00007 (/home/dgryski/Dropbox/GITS/gocode/src/github.com/dgryski/w/main_test.go:10)	MOVQ	184(AX), DX

dgryski

// This is a port of the FAST Java version of
// https://jackmott.github.io/programming/2016/09/01/performance-in-the-large.html
// Original Go translation of naive version from https://gist.github.com/magiconair/68a524fc847ba2860893a799f637f532
//
// Code is not pretty!!! ;)
//
//

package main

dgryski

name           old time/op    new time/op     delta
Spooky8-4        43.4ns ± 2%     31.0ns ± 1%   -28.58%  (p=0.000 n=26+29)
Spooky16-4       69.3ns ± 1%     44.6ns ± 1%   -35.68%  (p=0.000 n=24+29)
Spooky40-4       85.6ns ± 0%     47.6ns ± 1%   -44.42%  (p=0.000 n=29+29)
Spooky64-4        126ns ± 0%       60ns ± 1%   -52.11%  (p=0.000 n=19+29)
Spooky1K-4       1.16µs ± 3%     0.39µs ± 1%   -66.45%  (p=0.000 n=26+27)
Spooky8K-4       8.05µs ± 2%     2.65µs ± 1%   -67.08%  (p=0.000 n=29+28)
Siphash8-4       23.6ns ± 0%     22.5ns ± 0%    -4.70%  (p=0.000 n=29+18)
Siphash16-4      27.6ns ± 2%     26.3ns ± 0%    -4.52%  (p=0.000 n=29+30)
Siphash40-4      40.1ns ± 1%     39.0ns ± 0%    -2.65%  (p=0.000 n=28+29)

dgryski

// +build !appengine

package mpchash

import (
	"reflect"
	"unsafe"
)

func stringToSlice(s string) []byte {

dgryski

<dgryski@kamek[w] \ʕ◔ϖ◔ʔ/ > cat main.go
package main

import (
	"fmt"
	"time"
)

func main() {