Skip to content

GitLab

"platforms/cuda/src/CudaBondedUtilities.cpp" did not exist on "cf112a25be3f27ae98bd58453d9dadfdb12cc7c0"
Commit 6c6d68d2 authored by peastman's avatar peastman
Browse files

Merge pull request #1499 from peastman/reallocneighbors 

Improved method for resizing neighbor list
parents c2be00a6 6cc2edd1
Show whitespace changes
Inline Side-by-side
Showing with 138 additions and 146 deletions
platforms/opencl/src/kernels/nonbonded.cl
View file @ 6c6d68d2
...@@ -200,6 +200,8 @@ __kernel void computeNonbonded( ...@@ -200,6 +200,8 @@ __kernel void computeNonbonded(
#ifdef USE_CUTOFF #ifdef USE_CUTOFF
unsigned int numTiles = interactionCount[0]; unsigned int numTiles = interactionCount[0];
if (numTiles > maxTiles)
return; // There wasn't enough memory for the neighbor list.
int pos = (int) (numTiles > maxTiles ? startTileIndex+warp*(long)numTileIndices/totalWarps : warp*(long)numTiles/totalWarps); int pos = (int) (numTiles > maxTiles ? startTileIndex+warp*(long)numTileIndices/totalWarps : warp*(long)numTiles/totalWarps);
int end = (int) (numTiles > maxTiles ? startTileIndex+(warp+1)*(long)numTileIndices/totalWarps : (warp+1)*(long)numTiles/totalWarps); int end = (int) (numTiles > maxTiles ? startTileIndex+(warp+1)*(long)numTileIndices/totalWarps : (warp+1)*(long)numTiles/totalWarps);
#else #else
...@@ -223,16 +225,12 @@ __kernel void computeNonbonded( ...@@ -223,16 +225,12 @@ __kernel void computeNonbonded(
int x, y; int x, y;
bool singlePeriodicCopy = false; bool singlePeriodicCopy = false;
#ifdef USE_CUTOFF #ifdef USE_CUTOFF
if (numTiles <= maxTiles) {
x = tiles[pos]; x = tiles[pos];
real4 blockSizeX = blockSize[x]; real4 blockSizeX = blockSize[x];
singlePeriodicCopy = (0.5f*periodicBoxSize.x-blockSizeX.x >= MAX_CUTOFF && singlePeriodicCopy = (0.5f*periodicBoxSize.x-blockSizeX.x >= MAX_CUTOFF &&
0.5f*periodicBoxSize.y-blockSizeX.y >= MAX_CUTOFF && 0.5f*periodicBoxSize.y-blockSizeX.y >= MAX_CUTOFF &&
0.5f*periodicBoxSize.z-blockSizeX.z >= MAX_CUTOFF); 0.5f*periodicBoxSize.z-blockSizeX.z >= MAX_CUTOFF);
} #else
else
#endif
{
y = (int) floor(NUM_BLOCKS+0.5f-SQRT((NUM_BLOCKS+0.5f)*(NUM_BLOCKS+0.5f)-2*pos)); y = (int) floor(NUM_BLOCKS+0.5f-SQRT((NUM_BLOCKS+0.5f)*(NUM_BLOCKS+0.5f)-2*pos));
x = (pos-y*NUM_BLOCKS+y*(y+1)/2); x = (pos-y*NUM_BLOCKS+y*(y+1)/2);
if (x < y || x >= NUM_BLOCKS) { // Occasionally happens due to roundoff error. if (x < y || x >= NUM_BLOCKS) { // Occasionally happens due to roundoff error.
...@@ -258,7 +256,7 @@ __kernel void computeNonbonded( ...@@ -258,7 +256,7 @@ __kernel void computeNonbonded(
while (skipTiles[currentSkipIndex] < pos) while (skipTiles[currentSkipIndex] < pos)
currentSkipIndex++; currentSkipIndex++;
includeTile = (skipTiles[currentSkipIndex] != pos); includeTile = (skipTiles[currentSkipIndex] != pos);
} #endif
if (includeTile) { if (includeTile) {
unsigned int atom1 = x*TILE_SIZE + tgx; unsigned int atom1 = x*TILE_SIZE + tgx;
......
platforms/opencl/src/kernels/nonbonded_cpu.cl
View file @ 6c6d68d2
...@@ -214,6 +214,8 @@ __kernel void computeNonbonded( ...@@ -214,6 +214,8 @@ __kernel void computeNonbonded(
#ifdef USE_CUTOFF #ifdef USE_CUTOFF
const unsigned int numTiles = interactionCount[0]; const unsigned int numTiles = interactionCount[0];
if (numTiles > maxTiles)
return; // There wasn't enough memory for the neighbor list.
int pos = (int) (numTiles > maxTiles ? (unsigned int) (startTileIndex+get_group_id(0)*(long)numTileIndices/get_num_groups(0)) : get_group_id(0)*(long)numTiles/get_num_groups(0)); int pos = (int) (numTiles > maxTiles ? (unsigned int) (startTileIndex+get_group_id(0)*(long)numTileIndices/get_num_groups(0)) : get_group_id(0)*(long)numTiles/get_num_groups(0));
int end = (int) (numTiles > maxTiles ? (unsigned int) (startTileIndex+(get_group_id(0)+1)*(long)numTileIndices/get_num_groups(0)) : (get_group_id(0)+1)*(long)numTiles/get_num_groups(0)); int end = (int) (numTiles > maxTiles ? (unsigned int) (startTileIndex+(get_group_id(0)+1)*(long)numTileIndices/get_num_groups(0)) : (get_group_id(0)+1)*(long)numTiles/get_num_groups(0));
#else #else
...@@ -234,16 +236,12 @@ __kernel void computeNonbonded( ...@@ -234,16 +236,12 @@ __kernel void computeNonbonded(
int x, y; int x, y;
bool singlePeriodicCopy = false; bool singlePeriodicCopy = false;
#ifdef USE_CUTOFF #ifdef USE_CUTOFF
if (numTiles <= maxTiles) {
x = tiles[pos]; x = tiles[pos];
real4 blockSizeX = blockSize[x]; real4 blockSizeX = blockSize[x];
singlePeriodicCopy = (0.5f*periodicBoxSize.x-blockSizeX.x >= MAX_CUTOFF && singlePeriodicCopy = (0.5f*periodicBoxSize.x-blockSizeX.x >= MAX_CUTOFF &&
0.5f*periodicBoxSize.y-blockSizeX.y >= MAX_CUTOFF && 0.5f*periodicBoxSize.y-blockSizeX.y >= MAX_CUTOFF &&
0.5f*periodicBoxSize.z-blockSizeX.z >= MAX_CUTOFF); 0.5f*periodicBoxSize.z-blockSizeX.z >= MAX_CUTOFF);
} #else
else
#endif
{
y = (int) floor(NUM_BLOCKS+0.5f-SQRT((NUM_BLOCKS+0.5f)*(NUM_BLOCKS+0.5f)-2*pos)); y = (int) floor(NUM_BLOCKS+0.5f-SQRT((NUM_BLOCKS+0.5f)*(NUM_BLOCKS+0.5f)-2*pos));
x = (pos-y*NUM_BLOCKS+y*(y+1)/2); x = (pos-y*NUM_BLOCKS+y*(y+1)/2);
if (x < y || x >= NUM_BLOCKS) { // Occasionally happens due to roundoff error. if (x < y || x >= NUM_BLOCKS) { // Occasionally happens due to roundoff error.
...@@ -262,7 +260,7 @@ __kernel void computeNonbonded( ...@@ -262,7 +260,7 @@ __kernel void computeNonbonded(
nextToSkip = end; nextToSkip = end;
} }
includeTile = (nextToSkip != pos); includeTile = (nextToSkip != pos);
} #endif
if (includeTile) { if (includeTile) {
// Load the data for this tile. // Load the data for this tile.
......
plugins/amoeba/platforms/cuda/src/kernels/multipoleElectrostatics.cu
View file @ 6c6d68d2
...@@ -488,6 +488,8 @@ extern "C" __global__ void computeElectrostatics( ...@@ -488,6 +488,8 @@ extern "C" __global__ void computeElectrostatics(
#ifdef USE_CUTOFF #ifdef USE_CUTOFF
const unsigned int numTiles = interactionCount[0]; const unsigned int numTiles = interactionCount[0];
if (numTiles > maxTiles)
return; // There wasn't enough memory for the neighbor list.
int pos = (int) (numTiles > maxTiles ? startTileIndex+warp*(long long)numTileIndices/totalWarps : warp*(long long)numTiles/totalWarps); int pos = (int) (numTiles > maxTiles ? startTileIndex+warp*(long long)numTileIndices/totalWarps : warp*(long long)numTiles/totalWarps);
int end = (int) (numTiles > maxTiles ? startTileIndex+(warp+1)*(long long)numTileIndices/totalWarps : (warp+1)*(long long)numTiles/totalWarps); int end = (int) (numTiles > maxTiles ? startTileIndex+(warp+1)*(long long)numTileIndices/totalWarps : (warp+1)*(long long)numTiles/totalWarps);
#else #else
...@@ -508,11 +510,8 @@ extern "C" __global__ void computeElectrostatics( ...@@ -508,11 +510,8 @@ extern "C" __global__ void computeElectrostatics(
int x, y; int x, y;
#ifdef USE_CUTOFF #ifdef USE_CUTOFF
if (numTiles <= maxTiles)
x = tiles[pos]; x = tiles[pos];
else #else
#endif
{
y = (int) floor(NUM_BLOCKS+0.5f-SQRT((NUM_BLOCKS+0.5f)*(NUM_BLOCKS+0.5f)-2*pos)); y = (int) floor(NUM_BLOCKS+0.5f-SQRT((NUM_BLOCKS+0.5f)*(NUM_BLOCKS+0.5f)-2*pos));
x = (pos-y*NUM_BLOCKS+y*(y+1)/2); x = (pos-y*NUM_BLOCKS+y*(y+1)/2);
if (x < y || x >= NUM_BLOCKS) { // Occasionally happens due to roundoff error. if (x < y || x >= NUM_BLOCKS) { // Occasionally happens due to roundoff error.
...@@ -535,7 +534,7 @@ extern "C" __global__ void computeElectrostatics( ...@@ -535,7 +534,7 @@ extern "C" __global__ void computeElectrostatics(
while (skipTiles[currentSkipIndex] < pos) while (skipTiles[currentSkipIndex] < pos)
currentSkipIndex++; currentSkipIndex++;
includeTile = (skipTiles[currentSkipIndex] != pos); includeTile = (skipTiles[currentSkipIndex] != pos);
} #endif
if (includeTile) { if (includeTile) {
unsigned int atom1 = x*TILE_SIZE + tgx; unsigned int atom1 = x*TILE_SIZE + tgx;
......
plugins/amoeba/platforms/cuda/src/kernels/multipoleFixedField.cu
View file @ 6c6d68d2
...@@ -592,6 +592,8 @@ extern "C" __global__ void computeFixedField( ...@@ -592,6 +592,8 @@ extern "C" __global__ void computeFixedField(
#ifdef USE_CUTOFF #ifdef USE_CUTOFF
const unsigned int numTiles = interactionCount[0]; const unsigned int numTiles = interactionCount[0];
if (numTiles > maxTiles)
return; // There wasn't enough memory for the neighbor list.
int pos = (int) (numTiles > maxTiles ? startTileIndex+warp*(long long)numTileIndices/totalWarps : warp*(long long)numTiles/totalWarps); int pos = (int) (numTiles > maxTiles ? startTileIndex+warp*(long long)numTileIndices/totalWarps : warp*(long long)numTiles/totalWarps);
int end = (int) (numTiles > maxTiles ? startTileIndex+(warp+1)*(long long)numTileIndices/totalWarps : (warp+1)*(long long)numTiles/totalWarps); int end = (int) (numTiles > maxTiles ? startTileIndex+(warp+1)*(long long)numTileIndices/totalWarps : (warp+1)*(long long)numTiles/totalWarps);
#else #else
...@@ -612,11 +614,8 @@ extern "C" __global__ void computeFixedField( ...@@ -612,11 +614,8 @@ extern "C" __global__ void computeFixedField(
int x, y; int x, y;
#ifdef USE_CUTOFF #ifdef USE_CUTOFF
if (numTiles <= maxTiles)
x = tiles[pos]; x = tiles[pos];
else #else
#endif
{
y = (int) floor(NUM_BLOCKS+0.5f-SQRT((NUM_BLOCKS+0.5f)*(NUM_BLOCKS+0.5f)-2*pos)); y = (int) floor(NUM_BLOCKS+0.5f-SQRT((NUM_BLOCKS+0.5f)*(NUM_BLOCKS+0.5f)-2*pos));
x = (pos-y*NUM_BLOCKS+y*(y+1)/2); x = (pos-y*NUM_BLOCKS+y*(y+1)/2);
if (x < y || x >= NUM_BLOCKS) { // Occasionally happens due to roundoff error. if (x < y || x >= NUM_BLOCKS) { // Occasionally happens due to roundoff error.
...@@ -639,7 +638,7 @@ extern "C" __global__ void computeFixedField( ...@@ -639,7 +638,7 @@ extern "C" __global__ void computeFixedField(
while (skipTiles[currentSkipIndex] < pos) while (skipTiles[currentSkipIndex] < pos)
currentSkipIndex++; currentSkipIndex++;
includeTile = (skipTiles[currentSkipIndex] != pos); includeTile = (skipTiles[currentSkipIndex] != pos);
} #endif
if (includeTile) { if (includeTile) {
unsigned int atom1 = x*TILE_SIZE + tgx; unsigned int atom1 = x*TILE_SIZE + tgx;
......
plugins/amoeba/platforms/cuda/src/kernels/multipoleInducedField.cu
View file @ 6c6d68d2
...@@ -454,6 +454,8 @@ extern "C" __global__ void computeInducedField( ...@@ -454,6 +454,8 @@ extern "C" __global__ void computeInducedField(
#ifdef USE_CUTOFF #ifdef USE_CUTOFF
const unsigned int numTiles = interactionCount[0]; const unsigned int numTiles = interactionCount[0];
if (numTiles > maxTiles)
return; // There wasn't enough memory for the neighbor list.
int pos = (int) (numTiles > maxTiles ? startTileIndex+warp*(long long)numTileIndices/totalWarps : warp*(long long)numTiles/totalWarps); int pos = (int) (numTiles > maxTiles ? startTileIndex+warp*(long long)numTileIndices/totalWarps : warp*(long long)numTiles/totalWarps);
int end = (int) (numTiles > maxTiles ? startTileIndex+(warp+1)*(long long)numTileIndices/totalWarps : (warp+1)*(long long)numTiles/totalWarps); int end = (int) (numTiles > maxTiles ? startTileIndex+(warp+1)*(long long)numTileIndices/totalWarps : (warp+1)*(long long)numTiles/totalWarps);
#else #else
...@@ -474,11 +476,8 @@ extern "C" __global__ void computeInducedField( ...@@ -474,11 +476,8 @@ extern "C" __global__ void computeInducedField(
int x, y; int x, y;
#ifdef USE_CUTOFF #ifdef USE_CUTOFF
if (numTiles <= maxTiles)
x = tiles[pos]; x = tiles[pos];
else #else
#endif
{
y = (int) floor(NUM_BLOCKS+0.5f-SQRT((NUM_BLOCKS+0.5f)*(NUM_BLOCKS+0.5f)-2*pos)); y = (int) floor(NUM_BLOCKS+0.5f-SQRT((NUM_BLOCKS+0.5f)*(NUM_BLOCKS+0.5f)-2*pos));
x = (pos-y*NUM_BLOCKS+y*(y+1)/2); x = (pos-y*NUM_BLOCKS+y*(y+1)/2);
if (x < y || x >= NUM_BLOCKS) { // Occasionally happens due to roundoff error. if (x < y || x >= NUM_BLOCKS) { // Occasionally happens due to roundoff error.
...@@ -501,7 +500,7 @@ extern "C" __global__ void computeInducedField( ...@@ -501,7 +500,7 @@ extern "C" __global__ void computeInducedField(
while (skipTiles[currentSkipIndex] < pos) while (skipTiles[currentSkipIndex] < pos)
currentSkipIndex++; currentSkipIndex++;
includeTile = (skipTiles[currentSkipIndex] != pos); includeTile = (skipTiles[currentSkipIndex] != pos);
} #endif
if (includeTile) { if (includeTile) {
unsigned int atom1 = x*TILE_SIZE + tgx; unsigned int atom1 = x*TILE_SIZE + tgx;
......
plugins/amoeba/platforms/cuda/src/kernels/pmeMultipoleElectrostatics.cu
View file @ 6c6d68d2
...@@ -555,6 +555,8 @@ extern "C" __global__ void computeElectrostatics( ...@@ -555,6 +555,8 @@ extern "C" __global__ void computeElectrostatics(
#ifdef USE_CUTOFF #ifdef USE_CUTOFF
const unsigned int numTiles = interactionCount[0]; const unsigned int numTiles = interactionCount[0];
if (numTiles > maxTiles)
return; // There wasn't enough memory for the neighbor list.
int pos = (int) (numTiles > maxTiles ? startTileIndex+warp*(long long)numTileIndices/totalWarps : warp*(long long)numTiles/totalWarps); int pos = (int) (numTiles > maxTiles ? startTileIndex+warp*(long long)numTileIndices/totalWarps : warp*(long long)numTiles/totalWarps);
int end = (int) (numTiles > maxTiles ? startTileIndex+(warp+1)*(long long)numTileIndices/totalWarps : (warp+1)*(long long)numTiles/totalWarps); int end = (int) (numTiles > maxTiles ? startTileIndex+(warp+1)*(long long)numTileIndices/totalWarps : (warp+1)*(long long)numTiles/totalWarps);
#else #else
...@@ -575,11 +577,8 @@ extern "C" __global__ void computeElectrostatics( ...@@ -575,11 +577,8 @@ extern "C" __global__ void computeElectrostatics(
int x, y; int x, y;
#ifdef USE_CUTOFF #ifdef USE_CUTOFF
if (numTiles <= maxTiles)
x = tiles[pos]; x = tiles[pos];
else #else
#endif
{
y = (int) floor(NUM_BLOCKS+0.5f-SQRT((NUM_BLOCKS+0.5f)*(NUM_BLOCKS+0.5f)-2*pos)); y = (int) floor(NUM_BLOCKS+0.5f-SQRT((NUM_BLOCKS+0.5f)*(NUM_BLOCKS+0.5f)-2*pos));
x = (pos-y*NUM_BLOCKS+y*(y+1)/2); x = (pos-y*NUM_BLOCKS+y*(y+1)/2);
if (x < y || x >= NUM_BLOCKS) { // Occasionally happens due to roundoff error. if (x < y || x >= NUM_BLOCKS) { // Occasionally happens due to roundoff error.
...@@ -602,7 +601,7 @@ extern "C" __global__ void computeElectrostatics( ...@@ -602,7 +601,7 @@ extern "C" __global__ void computeElectrostatics(
while (skipTiles[currentSkipIndex] < pos) while (skipTiles[currentSkipIndex] < pos)
currentSkipIndex++; currentSkipIndex++;
includeTile = (skipTiles[currentSkipIndex] != pos); includeTile = (skipTiles[currentSkipIndex] != pos);
} #endif
if (includeTile) { if (includeTile) {
unsigned int atom1 = x*TILE_SIZE + tgx; unsigned int atom1 = x*TILE_SIZE + tgx;
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment