MCPcopy Create free account
hub / github.com/davisking/dlib / serialize

Function serialize

dlib/cuda/tensor.h:461–483  ·  view source on GitHub ↗

Source from the content-addressed store, hash-verified

459 };
460
461 inline void serialize(const tensor& item, std::ostream& out)
462 {
463 int version = 2;
464 serialize(version, out);
465 serialize(item.num_samples(), out);
466 serialize(item.k(), out);
467 serialize(item.nr(), out);
468 serialize(item.nc(), out);
469 byte_orderer bo;
470 auto sbuf = out.rdbuf();
471 for (auto d : item)
472 {
473 // Write out our data as 4byte little endian IEEE floats rather than using
474 // dlib's default float serialization. We do this because it will result in
475 // more compact outputs. It's slightly less portable but it seems doubtful
476 // that any CUDA enabled platform isn't going to use IEEE floats. But if one
477 // does we can just update the serialization code here to handle it if such a
478 // platform is encountered.
479 bo.host_to_little(d);
480 static_assert(sizeof(d)==4, "This serialization code assumes we are writing 4 byte floats");
481 sbuf->sputn((char*)&d, sizeof(d));
482 }
483 }
484
485 inline void deserialize(resizable_tensor& item, std::istream& in)
486 {

Callers

nothing calls this directly

Calls 5

host_to_littleMethod · 0.80
num_samplesMethod · 0.45
kMethod · 0.45
nrMethod · 0.45
ncMethod · 0.45

Tested by

no test coverage detected