working on sort
This commit is contained in:
Evghenii
275
examples_cuda/sort/sort1.ispc
Normal file
275
examples_cuda/sort/sort1.ispc
Normal file
@@ -0,0 +1,275 @@
|
||||
/*
|
||||
Copyright (c) 2013, Durham University
|
||||
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 Durham University 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 */
|
||||
|
||||
#ifdef __NVPTX__
|
||||
#warning "emitting DEVICE code"
|
||||
#define programCount warpSize()
|
||||
#define programIndex laneIndex()
|
||||
#define taskIndex blockIndex0()
|
||||
#define taskCount blockCount0()
|
||||
#define cfor for
|
||||
#define cif if
|
||||
#else
|
||||
#warning "emitting HOST code"
|
||||
#endif
|
||||
|
||||
task void histogram (uniform int span, uniform int n, uniform int64 code[], uniform int pass, uniform int hist[])
|
||||
{
|
||||
if (taskIndex >= taskCount) return;
|
||||
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[])
|
||||
{
|
||||
if (taskIndex >= taskCount) return;
|
||||
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[])
|
||||
{
|
||||
if (taskIndex >= taskCount) return;
|
||||
uniform int start = taskIndex*span;
|
||||
uniform int end = taskIndex == taskCount-1 ? n : start+span;
|
||||
|
||||
for (int i = programIndex + start; i < end; i += programCount)
|
||||
if (i < end)
|
||||
{
|
||||
to[i] = from[i];
|
||||
}
|
||||
}
|
||||
|
||||
task void pack (uniform int span, uniform int n, uniform unsigned int code[], uniform int64 pair[])
|
||||
{
|
||||
#if 0
|
||||
if (taskIndex >= taskCount) return;
|
||||
uniform int start = taskIndex*span;
|
||||
uniform int end = taskIndex == taskCount-1 ? n : start+span;
|
||||
|
||||
for (int i = programIndex + start; i < end; i += programCount)
|
||||
if (i < end)
|
||||
{
|
||||
pair[i] = ((int64)i<<32)+code[i];
|
||||
}
|
||||
#endif
|
||||
}
|
||||
|
||||
task void unpack (uniform int span, uniform int n, uniform int64 pair[], uniform int unsigned code[], uniform int order[])
|
||||
{
|
||||
if (taskIndex >= taskCount) return;
|
||||
uniform int start = taskIndex*span;
|
||||
uniform int end = taskIndex == taskCount-1 ? n : start+span;
|
||||
|
||||
for (int i = programIndex + start; i < end; i += programCount)
|
||||
if (i < end)
|
||||
{
|
||||
code[i] = pair[i];
|
||||
order[i] = pair[i]>>32;
|
||||
}
|
||||
}
|
||||
|
||||
task void addup (uniform int h[], uniform int g[])
|
||||
{
|
||||
if (taskIndex >= taskCount) return;
|
||||
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[])
|
||||
{
|
||||
if (taskIndex >= taskCount) return;
|
||||
uniform int * uniform u = &h[256*programCount*taskIndex];
|
||||
uniform int z = g[taskIndex];
|
||||
|
||||
for (int i = programIndex; i < 256*programCount; i += programCount)
|
||||
{
|
||||
u[i] += z;
|
||||
}
|
||||
}
|
||||
|
||||
static void prefix_sum (uniform int num, uniform int h[])
|
||||
{
|
||||
uniform int * uniform g = uniform new uniform 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 ? 13*4*8 : ntasks;
|
||||
uniform int span = n / num;
|
||||
uniform int hsize = 256*programCount*num;
|
||||
uniform int * uniform hist = uniform new uniform int [hsize];
|
||||
uniform int64 * uniform pair = uniform new uniform int64 [n];
|
||||
uniform int64 * uniform temp = uniform new uniform 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;
|
||||
#if 0
|
||||
|
||||
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
|
||||
#endif
|
||||
|
||||
delete hist;
|
||||
delete pair;
|
||||
delete temp;
|
||||
}
|
||||
Reference in New Issue
Block a user