__host__
inline void copy_to_host(T *host_data)
{
SHAKTI_SAFE_CUDA_CALL(cudaMemcpy2DFromArray(
host_data, _sizes[0] * sizeof(T), _array,
0, 0, _sizes[0]*sizeof(float), _sizes[1], cudaMemcpyDeviceToHost));
}
void RGBDOdometry::initICP(GPUTexture * filteredDepth, const float depthCutoff)
{
cudaArray * textPtr;
cudaGraphicsMapResources(1, &filteredDepth->cudaRes);
cudaGraphicsSubResourceGetMappedArray(&textPtr, filteredDepth->cudaRes, 0, 0);
cudaMemcpy2DFromArray(depth_tmp[0].ptr(0), depth_tmp[0].step(), textPtr, 0, 0, depth_tmp[0].colsBytes(), depth_tmp[0].rows(), cudaMemcpyDeviceToDevice);
cudaGraphicsUnmapResources(1, &filteredDepth->cudaRes);
for(int i = 1; i < NUM_PYRS; ++i)
{
pyrDown(depth_tmp[i - 1], depth_tmp[i]);
}
for(int i = 0; i < NUM_PYRS; ++i)
{
createVMap(intr(i), depth_tmp[i], vmaps_curr_[i], depthCutoff);
createNMap(vmaps_curr_[i], nmaps_curr_[i]);
}
cudaDeviceSynchronize();
}
cudaError_t cudaMemcpy3Dfix(const struct cudaMemcpy3DParms *param) {
const cudaMemcpy3DParms& p = *param;
// Use cudaMemcpy3D for 3D only
// But it does not handle 2D or 1D copies well
if (1<p.extent.depth) {
return cudaMemcpy3D( &p );
} else if (1<p.extent.height) {
// 2D copy
// Arraycopy
if (0 != p.srcArray && 0 == p.dstArray) {
return cudaMemcpy2DFromArray(p.dstPtr.ptr, p.dstPtr.pitch, p.srcArray, p.srcPos.x, p.srcPos.y, p.extent.width, p.extent.height, p.kind);
} else if(0 == p.srcArray && 0 != p.dstArray) {
return cudaMemcpy2DToArray(p.dstArray, p.dstPos.x, p.dstPos.y, p.srcPtr.ptr, p.srcPtr.pitch, p.extent.width, p.extent.height, p.kind);
} else if(0 != p.srcArray && 0 != p.dstArray) {
return cudaMemcpy2DArrayToArray( p.dstArray, p.dstPos.x, p.dstPos.y, p.srcArray, p.srcPos.x, p.srcPos.y, p.extent.width, p.extent.height, p.kind);
} else {
return cudaMemcpy2D( p.dstPtr.ptr, p.dstPtr.pitch, p.srcPtr.ptr, p.srcPtr.pitch, p.extent.width, p.extent.height, p.kind );
}
} else {
// 1D copy
// p.extent.width should not include pitch
EXCEPTION_ASSERT( p.extent.width == p.dstPtr.xsize );
EXCEPTION_ASSERT( p.extent.width == p.srcPtr.xsize );
// Arraycopy
if (0 != p.srcArray && 0 == p.dstArray) {
return cudaMemcpyFromArray(p.dstPtr.ptr, p.srcArray, p.srcPos.x, p.srcPos.y, p.extent.width, p.kind);
} else if(0 == p.srcArray && 0 != p.dstArray) {
return cudaMemcpyToArray(p.dstArray, p.dstPos.x, p.dstPos.y, p.srcPtr.ptr, p.extent.width, p.kind);
} else if(0 != p.srcArray && 0 != p.dstArray) {
return cudaMemcpyArrayToArray(p.dstArray, p.dstPos.x, p.dstPos.y, p.srcArray, p.srcPos.x, p.srcPos.y, p.extent.width, p.kind);
} else {
return cudaMemcpy( p.dstPtr.ptr, p.srcPtr.ptr, p.extent.width, p.kind );
}
}
}
SEXP R_auto_cudaMemcpy2DFromArray(SEXP r_dst, SEXP r_dpitch, SEXP r_src, SEXP r_wOffset, SEXP r_hOffset, SEXP r_width, SEXP r_height, SEXP r_kind)
{
SEXP r_ans = R_NilValue;
void * dst = GET_REF(r_dst, void );
size_t dpitch = REAL(r_dpitch)[0];
cudaArray_const_t src = (cudaArray_const_t) getRReference(r_src);
size_t wOffset = REAL(r_wOffset)[0];
size_t hOffset = REAL(r_hOffset)[0];
size_t width = REAL(r_width)[0];
size_t height = REAL(r_height)[0];
enum cudaMemcpyKind kind = (enum cudaMemcpyKind) INTEGER(r_kind)[0];
cudaError_t ans;
ans = cudaMemcpy2DFromArray(dst, dpitch, src, wOffset, hOffset, width, height, kind);
r_ans = Renum_convert_cudaError_t(ans) ;
return(r_ans);
}
cudaError_t WINAPI wine_cudaMemcpy2DFromArray( void *dst, size_t dpitch, const struct cudaArray *src, size_t wOffset, size_t hOffset, size_t width, size_t height, enum cudaMemcpyKind kind ) {
WINE_TRACE("\n");
return cudaMemcpy2DFromArray( dst, dpitch, src, wOffset, hOffset, width, height, kind );
}
