Skip to content

GitLab

Unverified Commit fd66bc70 authored by David Clark's avatar David Clark Committed by GitHub
Browse files

cuFFT execution error checking (#2926) 



* Adds cuFFT error checking for execution

* Removes tabs
Co-authored-by: default avatarDavid Clark <daclark@nvidia.com>
parent b49b82ef
Hide whitespace changes
Inline Side-by-side
Showing with 37 additions and 17 deletions
platforms/cuda/src/CudaKernels.cpp
View file @ fd66bc70
...@@ -1293,10 +1293,15 @@ double CudaCalcNonbondedForceKernel::execute(ContextImpl& context, bool includeF ...@@ -1293,10 +1293,15 @@ double CudaCalcNonbondedForceKernel::execute(ContextImpl& context, bool includeF
cu.executeKernel(pmeFinishSpreadChargeKernel, finishSpreadArgs, gridSizeX*gridSizeY*gridSizeZ, 256); cu.executeKernel(pmeFinishSpreadChargeKernel, finishSpreadArgs, gridSizeX*gridSizeY*gridSizeZ, 256);
if (useCudaFFT) { if (useCudaFFT) {
if (cu.getUseDoublePrecision()) if (cu.getUseDoublePrecision()) {
cufftExecD2Z(fftForward, (double*) pmeGrid1.getDevicePointer(), (double2*) pmeGrid2.getDevicePointer()); cufftResult result = cufftExecD2Z(fftForward, (double*) pmeGrid1.getDevicePointer(), (double2*) pmeGrid2.getDevicePointer());
else if (result != CUFFT_SUCCESS)
cufftExecR2C(fftForward, (float*) pmeGrid1.getDevicePointer(), (float2*) pmeGrid2.getDevicePointer()); throw OpenMMException("Error executing FFT: "+cu.intToString(result));
} else {
cufftResult result = cufftExecR2C(fftForward, (float*) pmeGrid1.getDevicePointer(), (float2*) pmeGrid2.getDevicePointer());
if (result != CUFFT_SUCCESS)
throw OpenMMException("Error executing FFT: "+cu.intToString(result));
}
} }
else { else {
fft->execFFT(pmeGrid1, pmeGrid2, true); fft->execFFT(pmeGrid1, pmeGrid2, true);
...@@ -1315,10 +1320,15 @@ double CudaCalcNonbondedForceKernel::execute(ContextImpl& context, bool includeF ...@@ -1315,10 +1320,15 @@ double CudaCalcNonbondedForceKernel::execute(ContextImpl& context, bool includeF
cu.executeKernel(pmeConvolutionKernel, convolutionArgs, gridSizeX*gridSizeY*gridSizeZ, 256); cu.executeKernel(pmeConvolutionKernel, convolutionArgs, gridSizeX*gridSizeY*gridSizeZ, 256);
if (useCudaFFT) { if (useCudaFFT) {
if (cu.getUseDoublePrecision()) if (cu.getUseDoublePrecision()) {
cufftExecZ2D(fftBackward, (double2*) pmeGrid2.getDevicePointer(), (double*) pmeGrid1.getDevicePointer()); cufftResult result = cufftExecZ2D(fftBackward, (double2*) pmeGrid2.getDevicePointer(), (double*) pmeGrid1.getDevicePointer());
else if (result != CUFFT_SUCCESS)
cufftExecC2R(fftBackward, (float2*) pmeGrid2.getDevicePointer(), (float*) pmeGrid1.getDevicePointer()); throw OpenMMException("Error executing FFT: "+cu.intToString(result));
} else {
cufftResult result = cufftExecC2R(fftBackward, (float2*) pmeGrid2.getDevicePointer(), (float*) pmeGrid1.getDevicePointer());
if (result != CUFFT_SUCCESS)
throw OpenMMException("Error executing FFT: "+cu.intToString(result));
}
} }
else { else {
fft->execFFT(pmeGrid2, pmeGrid1, false); fft->execFFT(pmeGrid2, pmeGrid1, false);
...@@ -1353,10 +1363,15 @@ double CudaCalcNonbondedForceKernel::execute(ContextImpl& context, bool includeF ...@@ -1353,10 +1363,15 @@ double CudaCalcNonbondedForceKernel::execute(ContextImpl& context, bool includeF
cu.executeKernel(pmeDispersionFinishSpreadChargeKernel, finishSpreadArgs, dispersionGridSizeX*dispersionGridSizeY*dispersionGridSizeZ, 256); cu.executeKernel(pmeDispersionFinishSpreadChargeKernel, finishSpreadArgs, dispersionGridSizeX*dispersionGridSizeY*dispersionGridSizeZ, 256);
if (useCudaFFT) { if (useCudaFFT) {
if (cu.getUseDoublePrecision()) if (cu.getUseDoublePrecision()) {
cufftExecD2Z(dispersionFftForward, (double*) pmeGrid1.getDevicePointer(), (double2*) pmeGrid2.getDevicePointer()); cufftResult result = cufftExecD2Z(dispersionFftForward, (double*) pmeGrid1.getDevicePointer(), (double2*) pmeGrid2.getDevicePointer());
else if (result != CUFFT_SUCCESS)
cufftExecR2C(dispersionFftForward, (float*) pmeGrid1.getDevicePointer(), (float2*) pmeGrid2.getDevicePointer()); throw OpenMMException("Error executing FFT: "+cu.intToString(result));
} else {
cufftResult result = cufftExecR2C(dispersionFftForward, (float*) pmeGrid1.getDevicePointer(), (float2*) pmeGrid2.getDevicePointer());
if (result != CUFFT_SUCCESS)
throw OpenMMException("Error executing FFT: "+cu.intToString(result));
}
} }
else { else {
dispersionFft->execFFT(pmeGrid1, pmeGrid2, true); dispersionFft->execFFT(pmeGrid1, pmeGrid2, true);
...@@ -1375,10 +1390,15 @@ double CudaCalcNonbondedForceKernel::execute(ContextImpl& context, bool includeF ...@@ -1375,10 +1390,15 @@ double CudaCalcNonbondedForceKernel::execute(ContextImpl& context, bool includeF
cu.executeKernel(pmeDispersionConvolutionKernel, convolutionArgs, dispersionGridSizeX*dispersionGridSizeY*dispersionGridSizeZ, 256); cu.executeKernel(pmeDispersionConvolutionKernel, convolutionArgs, dispersionGridSizeX*dispersionGridSizeY*dispersionGridSizeZ, 256);
if (useCudaFFT) { if (useCudaFFT) {
if (cu.getUseDoublePrecision()) if (cu.getUseDoublePrecision()) {
cufftExecZ2D(dispersionFftBackward, (double2*) pmeGrid2.getDevicePointer(), (double*) pmeGrid1.getDevicePointer()); cufftResult result = cufftExecZ2D(dispersionFftBackward, (double2*) pmeGrid2.getDevicePointer(), (double*) pmeGrid1.getDevicePointer());
else if (result != CUFFT_SUCCESS)
cufftExecC2R(dispersionFftBackward, (float2*) pmeGrid2.getDevicePointer(), (float*) pmeGrid1.getDevicePointer()); throw OpenMMException("Error executing FFT: "+cu.intToString(result));
} else {
cufftResult result = cufftExecC2R(dispersionFftBackward, (float2*) pmeGrid2.getDevicePointer(), (float*) pmeGrid1.getDevicePointer());
if (result != CUFFT_SUCCESS)
throw OpenMMException("Error executing FFT: "+cu.intToString(result));
}
} }
else { else {
dispersionFft->execFFT(pmeGrid2, pmeGrid1, false); dispersionFft->execFFT(pmeGrid2, pmeGrid1, false);
...@@ -1764,7 +1784,7 @@ void CudaApplyMonteCarloBarostatKernel::scaleCoordinates(ContextImpl& context, d ...@@ -1764,7 +1784,7 @@ void CudaApplyMonteCarloBarostatKernel::scaleCoordinates(ContextImpl& context, d
float scalefZ = (float) scaleZ; float scalefZ = (float) scaleZ;
void* args[] = {&scalefX, &scalefY, &scalefZ, &numMolecules, cu.getPeriodicBoxSizePointer(), cu.getInvPeriodicBoxSizePointer(), void* args[] = {&scalefX, &scalefY, &scalefZ, &numMolecules, cu.getPeriodicBoxSizePointer(), cu.getInvPeriodicBoxSizePointer(),
cu.getPeriodicBoxVecXPointer(), cu.getPeriodicBoxVecYPointer(), cu.getPeriodicBoxVecZPointer(), cu.getPeriodicBoxVecXPointer(), cu.getPeriodicBoxVecYPointer(), cu.getPeriodicBoxVecZPointer(),
&cu.getPosq().getDevicePointer(), &moleculeAtoms.getDevicePointer(), &moleculeStartIndex.getDevicePointer()}; &cu.getPosq().getDevicePointer(), &moleculeAtoms.getDevicePointer(), &moleculeStartIndex.getDevicePointer()};
cu.executeKernel(kernel, args, cu.getNumAtoms()); cu.executeKernel(kernel, args, cu.getNumAtoms());
for (auto& offset : cu.getPosCellOffsets()) for (auto& offset : cu.getPosCellOffsets())
offset = mm_int4(0, 0, 0, 0); offset = mm_int4(0, 0, 0, 0);
......
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