+runs.. next step is tunning
This commit is contained in:
@@ -84,7 +84,7 @@ void mergeSortGangKernel(
|
|||||||
s_key[programIndex + programCount] = srcKey[base + programIndex + programCount];
|
s_key[programIndex + programCount] = srcKey[base + programIndex + programCount];
|
||||||
s_val[programIndex + programCount] = srcVal[base + programIndex + programCount];
|
s_val[programIndex + programCount] = srcVal[base + programIndex + programCount];
|
||||||
|
|
||||||
for (uniform int stride = 1; stride < programCount; stride <<= 1)
|
for (uniform int stride = 1; stride < 2*programCount; stride <<= 1)
|
||||||
{
|
{
|
||||||
const int lPos = programIndex & (stride - 1);
|
const int lPos = programIndex & (stride - 1);
|
||||||
uniform int *baseKey = s_key + 2 * (programIndex - lPos);
|
uniform int *baseKey = s_key + 2 * (programIndex - lPos);
|
||||||
@@ -480,7 +480,6 @@ void mergeSort(
|
|||||||
|
|
||||||
for (uniform int stride = 2*programCount; stride < N; stride <<= 1)
|
for (uniform int stride = 2*programCount; stride < N; stride <<= 1)
|
||||||
{
|
{
|
||||||
// print ("stride= % N= % \n", stride, N);
|
|
||||||
uniform int lastSegmentElements = N % (2 * stride);
|
uniform int lastSegmentElements = N % (2 * stride);
|
||||||
|
|
||||||
//Find sample ranks and prepare for limiters merge
|
//Find sample ranks and prepare for limiters merge
|
||||||
|
|||||||
Reference in New Issue
Block a user