aaron/ispc
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

+1

Browse Source
This commit is contained in:
Evghenii
2014-01-29 11:55:33 +01:00
parent 2634ae65fd
commit cb436d5b4a
1 changed files with 2 additions and 19 deletions
Download Patch File Download Diff File Expand all files Collapse all files

21
examples_ptx/radixSort/radixSort.ispc
View File

@@ -1,4 +1,4 @@
#define NUMBITS 4 #define NUMBITS 8
#define NUMDIGITS (1<<NUMBITS) #define NUMDIGITS (1<<NUMBITS)
typedef int64 Key; typedef int64 Key;
@@ -65,8 +65,6 @@ void sortPass(
const int unitScan = exclusive_scan_add(1); const int unitScan = exclusive_scan_add(1);
int lkeys[NUMDIGITS] = {0};
/* copy digit offset from Gmem to Lmem */ /* copy digit offset from Gmem to Lmem */
uniform int digitOffsets[NUMDIGITS]; uniform int digitOffsets[NUMDIGITS];
foreach (digit = 0 ... NUMDIGITS) foreach (digit = 0 ... NUMDIGITS)
@@ -77,6 +75,7 @@ void sortPass(
{ {
const int key = mask & ((unsigned int)keys[i] >> bit); const int key = mask & ((unsigned int)keys[i] >> bit);
int scatter; int scatter;
/* not a vector friendly loop */
foreach_active(iv) foreach_active(iv)
scatter = digitOffsets[key]++; scatter = digitOffsets[key]++;
sorted [scatter] = keys[i]; sorted [scatter] = keys[i];
@@ -100,7 +99,6 @@ void partialScanLocal(
uniform int (* uniform excScanBlock)[NUMDIGITS] = (uniform int (*)[NUMDIGITS])excScanAll; uniform int (* uniform excScanBlock)[NUMDIGITS] = (uniform int (*)[NUMDIGITS])excScanAll;
uniform int (* uniform partialSum)[NUMDIGITS] = (uniform int (*)[NUMDIGITS])partialSumAll; uniform int (* uniform partialSum)[NUMDIGITS] = (uniform int (*)[NUMDIGITS])partialSumAll;
#if 0
foreach (digit = 0 ... NUMDIGITS) foreach (digit = 0 ... NUMDIGITS)
{ {
int prev = bbeg == 0 ? excScanBlock[0][digit] : 0; int prev = bbeg == 0 ? excScanBlock[0][digit] : 0;
@@ -112,21 +110,6 @@ void partialScanLocal(
} }
partialSum[blockIdx][digit] = excScanBlock[bend-1][digit] + countsBlock[bend-1][digit]; partialSum[blockIdx][digit] = excScanBlock[bend-1][digit] + countsBlock[bend-1][digit];
} }
#else
int prev[NUMDIGITS];
for (int digit = 0; digit < NUMDIGITS; digit++)
prev[digit] = bbeg == 0 ? excScanBlock[0][digit] : 0;
foreach_tiled (block = bbeg ... bend, digit = 0 ... NUMDIGITS)
{
const int y = countsBlock[block][digit];
excScanBlock[block][digit] = prev[digit];
prev[digit] += y;
}
foreach (digit = 0 ... NUMDIGITS)
partialSum[blockIdx][digit] = excScanBlock[bend-1][digit] + countsBlock[bend-1][digit];
#endif
} }
task task