Merge pull request #527 from tkoziara/master

examples/sort added
This commit is contained in:
Dmitry Babokin
2013-06-25 10:11:39 -07:00
4 changed files with 451 additions and 0 deletions

8
examples/sort/Makefile Normal file
View File

@@ -0,0 +1,8 @@
EXAMPLE=sort
CPP_SRC=sort.cpp sort_serial.cpp
ISPC_SRC=sort.ispc
ISPC_TARGETS=sse2,sse4-x2,avx
#ISPC_FLAGS=-DDEBUG
include ../common.mk

134
examples/sort/sort.cpp Normal file
View File

@@ -0,0 +1,134 @@
/*
Copyright (c) 2013, Intel Corporation
All rights reserved.
Redistribution and use in source and binary forms, with or without
modification, are permitted provided that the following conditions are
met:
* Redistributions of source code must retain the above copyright
notice, this list of conditions and the following disclaimer.
* Redistributions in binary form must reproduce the above copyright
notice, this list of conditions and the following disclaimer in the
documentation and/or other materials provided with the distribution.
* Neither the name of Intel Corporation nor the names of its
contributors may be used to endorse or promote products derived from
this software without specific prior written permission.
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS
IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A
PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER
OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/
/* Author: Tomasz Koziara */
#include <stdio.h>
#include <stdlib.h>
#include <algorithm>
#include <iostream>
#include <iomanip>
#include "../timing.h"
#include "sort_ispc.h"
using namespace ispc;
extern void sort_serial (int n, unsigned int code[], int order[]);
/* progress bar by Ross Hemsley;
* http://www.rosshemsley.co.uk/2011/02/creating-a-progress-bar-in-c-or-any-other-console-app/ */
static inline void progressbar (unsigned int x, unsigned int n, unsigned int w = 50)
{
if (n < 100)
{
x *= 100/n;
n = 100;
}
if ((x != n) && (x % (n/100) != 0)) return;
using namespace std;
float ratio = x/(float)n;
int c = ratio * w;
cout << setw(3) << (int)(ratio*100) << "% [";
for (int x=0; x<c; x++) cout << "=";
for (int x=c; x<w; x++) cout << " ";
cout << "]\r" << flush;
}
int main (int argc, char *argv[])
{
int i, j, n = argc == 1 ? 1000000 : atoi(argv[1]), m = n < 100 ? 1 : 50, l = n < 100 ? n : RAND_MAX;
double tISPC1 = 0.0, tISPC2 = 0.0, tSerial = 0.0;
unsigned int *code = new unsigned int [n];
int *order = new int [n];
srand (0);
for (i = 0; i < m; i ++)
{
for (j = 0; j < n; j ++) code [j] = random() % l;
reset_and_start_timer();
sort_ispc (n, code, order, 1);
tISPC1 += get_elapsed_mcycles();
progressbar (i, m);
}
printf("[sort ispc]:\t[%.3f] million cycles\n", tISPC1);
srand (0);
for (i = 0; i < m; i ++)
{
for (j = 0; j < n; j ++) code [j] = random() % l;
reset_and_start_timer();
sort_ispc (n, code, order, 0);
tISPC2 += get_elapsed_mcycles();
progressbar (i, m);
}
printf("[sort ispc+tasks]:\t[%.3f] million cycles\n", tISPC2);
srand (0);
for (i = 0; i < m; i ++)
{
for (j = 0; j < n; j ++) code [j] = random() % l;
reset_and_start_timer();
sort_serial (n, code, order);
tSerial += get_elapsed_mcycles();
progressbar (i, m);
}
printf("[sort serial]:\t\t[%.3f] million cycles\n", tSerial);
printf("\t\t\t\t(%.2fx speedup from ISPC serial)\n", tSerial/tISPC1);
printf("\t\t\t\t(%.2fx speedup from ISPC with tasks)\n", tSerial/tISPC2);
delete code;
delete order;
return 0;
}

249
examples/sort/sort.ispc Normal file
View File

@@ -0,0 +1,249 @@
/*
Copyright (c) 2013, Intel Corporation
All rights reserved.
Redistribution and use in source and binary forms, with or without
modification, are permitted provided that the following conditions are
met:
* Redistributions of source code must retain the above copyright
notice, this list of conditions and the following disclaimer.
* Redistributions in binary form must reproduce the above copyright
notice, this list of conditions and the following disclaimer in the
documentation and/or other materials provided with the distribution.
* Neither the name of Intel Corporation nor the names of its
contributors may be used to endorse or promote products derived from
this software without specific prior written permission.
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS
IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A
PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER
OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/
/* Author: Tomasz Koziara */
task void histogram (uniform int span, uniform int n, uniform int64 code[], uniform int pass, uniform int hist[])
{
uniform int start = taskIndex*span;
uniform int end = taskIndex == taskCount-1 ? n : start+span;
uniform int strip = (end-start)/programCount;
uniform int tail = (end-start)%programCount;
int i = programCount*taskIndex + programIndex;
int g [256];
cfor (int j = 0; j < 256; j ++)
{
g[j] = 0;
}
cfor (int k = start+programIndex*strip; k < start+(programIndex+1)*strip; k ++)
{
unsigned int8 *c = (unsigned int8*) &code[k];
g[c[pass]] ++;
}
if (programIndex == programCount-1) /* remainder is processed by the last lane */
{
for (int k = start+programCount*strip; k < start+programCount*strip+tail; k ++)
{
unsigned int8 *c = (unsigned int8*) &code[k];
g[c[pass]] ++;
}
}
cfor (int j = 0; j < 256; j ++)
{
hist[j*programCount*taskCount+i] = g[j];
}
}
task void permutation (uniform int span, uniform int n, uniform int64 code[], uniform int pass, uniform int hist[], uniform int64 perm[])
{
uniform int start = taskIndex*span;
uniform int end = taskIndex == taskCount-1 ? n : start+span;
uniform int strip = (end-start)/programCount;
uniform int tail = (end-start)%programCount;
int i = programCount*taskIndex + programIndex;
int g [256];
cfor (int j = 0; j < 256; j ++)
{
g[j] = hist[j*programCount*taskCount+i];
}
cfor (int k = start+programIndex*strip; k < start+(programIndex+1)*strip; k ++)
{
unsigned int8 *c = (unsigned int8*) &code[k];
int l = g[c[pass]];
perm[l] = code[k];
g[c[pass]] = l+1;
}
if (programIndex == programCount-1) /* remainder is processed by the last lane */
{
for (int k = start+programCount*strip; k < start+programCount*strip+tail; k ++)
{
unsigned int8 *c = (unsigned int8*) &code[k];
int l = g[c[pass]];
perm[l] = code[k];
g[c[pass]] = l+1;
}
}
}
task void copy (uniform int span, uniform int n, uniform int64 from[], uniform int64 to[])
{
uniform int start = taskIndex*span;
uniform int end = taskIndex == taskCount-1 ? n : start+span;
foreach (i = start ... end)
{
to[i] = from[i];
}
}
task void pack (uniform int span, uniform int n, uniform unsigned int code[], uniform int64 pair[])
{
uniform int start = taskIndex*span;
uniform int end = taskIndex == taskCount-1 ? n : start+span;
foreach (i = start ... end)
{
pair[i] = ((int64)i<<32)+code[i];
}
}
task void unpack (uniform int span, uniform int n, uniform int64 pair[], uniform int unsigned code[], uniform int order[])
{
uniform int start = taskIndex*span;
uniform int end = taskIndex == taskCount-1 ? n : start+span;
foreach (i = start ... end)
{
code[i] = pair[i];
order[i] = pair[i]>>32;
}
}
task void addup (uniform int h[], uniform int g[])
{
uniform int * uniform u = &h[256*programCount*taskIndex];
uniform int i, x, y = 0;
for (i = 0; i < 256*programCount; i ++)
{
x = u[i];
u[i] = y;
y += x;
}
g[taskIndex] = y;
}
task void bumpup (uniform int h[], uniform int g[])
{
uniform int * uniform u = &h[256*programCount*taskIndex];
uniform int z = g[taskIndex];
foreach (i = 0 ... 256*programCount)
{
u[i] += z;
}
}
static void prefix_sum (uniform int num, uniform int h[])
{
uniform int * uniform g = uniform new int [num+1];
uniform int i;
launch[num] addup (h, g+1);
sync;
for (g[0] = 0, i = 1; i < num; i ++) g[i] += g[i-1];
launch[num] bumpup (h, g);
sync;
delete g;
}
export void sort_ispc (uniform int n, uniform unsigned int code[], uniform int order[], uniform int ntasks)
{
uniform int num = ntasks < 1 ? num_cores () : ntasks;
uniform int span = n / num;
uniform int hsize = 256*programCount*num;
uniform int * uniform hist = uniform new int [hsize];
uniform int64 * uniform pair = uniform new int64 [n];
uniform int64 * uniform temp = uniform new int64 [n];
uniform int pass, i;
#if DEBUG
if (n < 100)
{
print ("input: ");
for (i = 0; i < n; i ++) print ("%, ", code[i]);
print ("\n");
}
#endif
launch[num] pack (span, n, code, pair);
sync;
for (pass = 0; pass < 4; pass ++)
{
launch[num] histogram (span, n, pair, pass, hist);
sync;
prefix_sum (num, hist);
launch[num] permutation (span, n, pair, pass, hist, temp);
sync;
launch[num] copy (span, n, temp, pair);
sync;
}
launch[num] unpack (span, n, pair, code, order);
sync;
#if DEBUG
for (i = 0; i < n; i ++)
{
if (i > 0 && code[i-1] > code[i])
print ("ERR at % => % > %; ", i, code[i-1], code[i]);
}
if (n < 100)
{
print ("output: ");
for (i = 0; i < n; i ++) print ("%, ", code[i]);
print ("\n");
print ("order: ");
for (i = 0; i < n; i ++) print ("%, ", order[i]);
print ("\n");
}
#endif
delete hist;
delete pair;
delete temp;
}

View File

@@ -0,0 +1,60 @@
/*
Copyright (c) 2013, Intel Corporation
All rights reserved.
Redistribution and use in source and binary forms, with or without
modification, are permitted provided that the following conditions are
met:
* Redistributions of source code must retain the above copyright
notice, this list of conditions and the following disclaimer.
* Redistributions in binary form must reproduce the above copyright
notice, this list of conditions and the following disclaimer in the
documentation and/or other materials provided with the distribution.
* Neither the name of Intel Corporation nor the names of its
contributors may be used to endorse or promote products derived from
this software without specific prior written permission.
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS
IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A
PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER
OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/
/* Author: Tomasz Koziara */
#include <vector>
#include <algorithm>
#include <utility>
typedef std::pair<double,int> pair;
struct cmp
{
bool operator() (const pair& a, const pair& b) { return a.first < b.first; }
};
void sort_serial (int n, unsigned int code[], int order[])
{
std::vector<pair> pairs;
pairs.reserve (n);
for (int i = 0; i < n; i++) pairs.push_back (pair(code[i], i));
std::sort (pairs.begin(), pairs.end(), cmp());
int *o = order;
for (std::vector<pair>::const_iterator p = pairs.begin(); p != pairs.end(); ++p, ++o) *o = p->second;
}