void cudaColorSpinorField::saveCPUSpinorField(cpuColorSpinorField &dest) const {
if (nDim != dest.Ndim()) {
errorQuda("Number of dimensions %d %d don't match", nDim, dest.Ndim());
}
if (volume != dest.volume) {
errorQuda("Volumes %d %d don't match", volume, dest.volume);
}
if (SiteOrder() != dest.SiteOrder()) {
errorQuda("Subset orders don't match");
}
if (nColor != 3) {
errorQuda("Nc != 3 not yet supported");
}
if (siteSubset != dest.siteSubset) {
errorQuda("Subset types do not match %d %d", siteSubset, dest.siteSubset);
}
if (precision == QUDA_HALF_PRECISION) {
ColorSpinorParam param(*this); // acquire all attributes of this
param.precision = QUDA_SINGLE_PRECISION; // change precision
param.create = QUDA_COPY_FIELD_CREATE;
cudaColorSpinorField tmp(*this, param);
tmp.saveCPUSpinorField(dest);
return;
}
// (temporary?) bug fix for padding
memset(buffer, 0, bufferBytes);
cudaMemcpy(buffer, v, bytes, cudaMemcpyDeviceToHost);
#define SAVE_SPINOR_GPU_TO_CPU(myNs) \
if (precision == QUDA_DOUBLE_PRECISION) { \
if (dest.precision == QUDA_DOUBLE_PRECISION) { \
if (fieldOrder == QUDA_FLOAT_FIELD_ORDER) { \
unpackSpinor<3,myNs,1>((double*)dest.v, (double*)buffer, volume, pad, x, dest.total_length, total_length, \
dest.SiteSubset(), dest.SiteOrder(), dest.GammaBasis(), gammaBasis, dest.FieldOrder()); \
} else if (fieldOrder == QUDA_FLOAT2_FIELD_ORDER) { \
unpackSpinor<3,myNs,2>((double*)dest.v, (double*)buffer, volume, pad, x, dest.total_length, total_length, \
dest.SiteSubset(), dest.SiteOrder(), dest.GammaBasis(), gammaBasis, dest.FieldOrder()); \
} else if (fieldOrder == QUDA_FLOAT4_FIELD_ORDER) { \
errorQuda("double4 not supported"); \
} \
} else { \
if (fieldOrder == QUDA_FLOAT_FIELD_ORDER) { \
unpackSpinor<3,myNs,1>((float*)dest.v, (double*)buffer, volume, pad, x, dest.total_length, total_length, \
dest.SiteSubset(), dest.SiteOrder(), dest.GammaBasis(), gammaBasis, dest.FieldOrder()); \
} else if (fieldOrder == QUDA_FLOAT2_FIELD_ORDER) { \
unpackSpinor<3,myNs,2>((float*)dest.v, (double*)buffer, volume, pad, x, dest.total_length, total_length, \
dest.SiteSubset(), dest.SiteOrder(), dest.GammaBasis(), gammaBasis, dest.FieldOrder()); \
} else if (fieldOrder == QUDA_FLOAT4_FIELD_ORDER) { \
errorQuda("double4 not supported"); \
} \
} \
} else { \
if (dest.precision == QUDA_DOUBLE_PRECISION) { \
if (fieldOrder == QUDA_FLOAT_FIELD_ORDER) { \
unpackSpinor<3,myNs,1>((double*)dest.v, (float*)buffer, volume, pad, x, dest.total_length, total_length, \
dest.SiteSubset(), dest.SiteOrder(), dest.GammaBasis(), gammaBasis, dest.FieldOrder()); \
} else if (fieldOrder == QUDA_FLOAT2_FIELD_ORDER) { \
unpackSpinor<3,myNs,2>((double*)dest.v, (float*)buffer, volume, pad, x, dest.total_length, total_length, \
dest.SiteSubset(), dest.SiteOrder(), dest.GammaBasis(), gammaBasis, dest.FieldOrder()); \
} else if (fieldOrder == QUDA_FLOAT4_FIELD_ORDER) { \
unpackSpinor<3,myNs,4>((double*)dest.v, (float*)buffer, volume, pad, x, dest.total_length, total_length, \
dest.SiteSubset(), dest.SiteOrder(), dest.GammaBasis(), gammaBasis, dest.FieldOrder()); \
} \
} else { \
if (fieldOrder == QUDA_FLOAT_FIELD_ORDER) { \
unpackSpinor<3,myNs,1>((float*)dest.v, (float*)buffer, volume, pad, x, dest.total_length, total_length, \
dest.SiteSubset(), dest.SiteOrder(), dest.GammaBasis(), gammaBasis, dest.FieldOrder()); \
} else if (fieldOrder == QUDA_FLOAT2_FIELD_ORDER) { \
unpackSpinor<3,myNs,2>((float*)dest.v, (float*)buffer, volume, pad, x, dest.total_length, total_length, \
dest.SiteSubset(), dest.SiteOrder(), dest.GammaBasis(), gammaBasis, dest.FieldOrder()); \
} else if (fieldOrder == QUDA_FLOAT4_FIELD_ORDER) { \
unpackSpinor<3,myNs,4>((float*)dest.v, (float*)buffer, volume, pad, x, dest.total_length, total_length, \
dest.SiteSubset(), dest.SiteOrder(), dest.GammaBasis(), gammaBasis, dest.FieldOrder()); \
} \
} \
}
switch(nSpin){
case 1:
SAVE_SPINOR_GPU_TO_CPU(1);
break;
case 4:
SAVE_SPINOR_GPU_TO_CPU(4);
break;
default:
errorQuda("invalid number of spinors in function %s\n", __FUNCTION__);
}
#undef SAVE_SPINOR_GPU_TO_CPU
return;
}
void cudaColorSpinorField::loadCPUSpinorField(const cpuColorSpinorField &src) {
if (nDim != src.Ndim()) {
errorQuda("Number of dimensions %d %d don't match", nDim, src.Ndim());
}
if (volume != src.volume) {
errorQuda("Volumes %d %d don't match", volume, src.volume);
}
if (SiteOrder() != src.SiteOrder()) {
errorQuda("Subset orders don't match");
}
if (nColor != 3) {
errorQuda("Nc != 3 not yet supported");
}
if (siteSubset != src.siteSubset) {
errorQuda("Subset types do not match %d %d", siteSubset, src.siteSubset);
}
if (precision == QUDA_HALF_PRECISION) {
ColorSpinorParam param(*this); // acquire all attributes of this
param.precision = QUDA_SINGLE_PRECISION; // change precision
param.create = QUDA_COPY_FIELD_CREATE;
cudaColorSpinorField tmp(src, param);
copy(tmp);
return;
}
// (temporary?) bug fix for padding
memset(buffer, 0, bufferBytes);
#define LOAD_SPINOR_CPU_TO_GPU(myNs) \
if (precision == QUDA_DOUBLE_PRECISION) { \
if (src.precision == QUDA_DOUBLE_PRECISION) { \
if (fieldOrder == QUDA_FLOAT_FIELD_ORDER) { \
packSpinor<3,myNs,1>((double*)buffer, (double*)src.v, volume, pad, x, total_length, src.total_length, \
src.SiteSubset(), src.SiteOrder(), gammaBasis, src.GammaBasis(), src.FieldOrder()); \
} else if (fieldOrder == QUDA_FLOAT2_FIELD_ORDER) { \
packSpinor<3,myNs,2>((double*)buffer, (double*)src.v, volume, pad, x, total_length, src.total_length, \
src.SiteSubset(), src.SiteOrder(), gammaBasis, src.GammaBasis(), src.FieldOrder()); \
} else if (fieldOrder == QUDA_FLOAT4_FIELD_ORDER) { \
errorQuda("double4 not supported"); \
} \
} else { \
if (fieldOrder == QUDA_FLOAT_FIELD_ORDER) { \
packSpinor<3,myNs,1>((double*)buffer, (float*)src.v, volume, pad, x, total_length, src.total_length, \
src.SiteSubset(), src.SiteOrder(), gammaBasis, src.GammaBasis(), src.FieldOrder()); \
} else if (fieldOrder == QUDA_FLOAT2_FIELD_ORDER) { \
packSpinor<3,myNs,2>((double*)buffer, (float*)src.v, volume, pad, x, total_length, src.total_length, \
src.SiteSubset(), src.SiteOrder(), gammaBasis, src.GammaBasis(), src.FieldOrder()); \
} else if (fieldOrder == QUDA_FLOAT4_FIELD_ORDER) { \
errorQuda("double4 not supported"); \
} \
} \
} else { \
if (src.precision == QUDA_DOUBLE_PRECISION) { \
if (fieldOrder == QUDA_FLOAT_FIELD_ORDER) { \
packSpinor<3,myNs,1>((float*)buffer, (double*)src.v, volume, pad, x, total_length, src.total_length, \
src.SiteSubset(), src.SiteOrder(), gammaBasis, src.GammaBasis(), src.FieldOrder()); \
} else if (fieldOrder == QUDA_FLOAT2_FIELD_ORDER) { \
packSpinor<3,myNs,2>((float*)buffer, (double*)src.v, volume, pad, x, total_length, src.total_length, \
src.SiteSubset(), src.SiteOrder(), gammaBasis, src.GammaBasis(), src.FieldOrder()); \
} else if (fieldOrder == QUDA_FLOAT4_FIELD_ORDER) { \
packSpinor<3,myNs,4>((float*)buffer, (double*)src.v, volume, pad, x, total_length, src.total_length, \
src.SiteSubset(), src.SiteOrder(), gammaBasis, src.GammaBasis(), src.FieldOrder()); \
} \
} else { \
if (fieldOrder == QUDA_FLOAT_FIELD_ORDER) { \
packSpinor<3,myNs,1>((float*)buffer, (float*)src.v, volume, pad, x, total_length, src.total_length, \
src.SiteSubset(), src.SiteOrder(), gammaBasis, src.GammaBasis(), src.FieldOrder()); \
} else if (fieldOrder == QUDA_FLOAT2_FIELD_ORDER) { \
packSpinor<3,myNs,2>((float*)buffer, (float*)src.v, volume, pad, x, total_length, src.total_length, \
src.SiteSubset(), src.SiteOrder(), gammaBasis, src.GammaBasis(), src.FieldOrder()); \
} else if (fieldOrder == QUDA_FLOAT4_FIELD_ORDER) { \
packSpinor<3,myNs,4>((float*)buffer, (float*)src.v, volume, pad, x, total_length, src.total_length, \
src.SiteSubset(), src.SiteOrder(), gammaBasis, src.GammaBasis(), src.FieldOrder()); \
} \
} \
}
switch(nSpin){
case 1:
LOAD_SPINOR_CPU_TO_GPU(1);
break;
case 4:
LOAD_SPINOR_CPU_TO_GPU(4);
break;
default:
errorQuda("invalid number of spinors in function %s\n", __FUNCTION__);
}
#undef LOAD_SPINOR_CPU_TO_GPU
/* for (int i=0; i<length; i++) {
std::cout << i << " " << ((float*)src.v)[i] << " " << ((float*)buffer)[i] << std::endl;
}*/
cudaMemcpy(v, buffer, bytes, cudaMemcpyHostToDevice);
return;
}