Go to the source code of this file.
|
| module | hfx_compression_core_methods |
| | Contains routines for data compression: packing integers of Nbits bits each into a stream of 64-bit words, and unpacking them. Per number of bits, fypp generates a routine that assembles each packed word in a register from constant shifts and stores it once, for every group of 64 integers (which fills exactly Nbits words). Moreover, packing at a width per group of integers (the width the group needs), in 8 streams whose shifts are uniform over the lanes.
|
|
| subroutine, public | hfx_compression_core_methods::ints2bits_specific (nbits, ndata, packed_data, full_data) |
| | Packs integers into a stream of 64-bit words, Nbits bits per integer.
|
| subroutine, public | hfx_compression_core_methods::bits2ints_specific (nbits, ndata, packed_data, full_data) |
| | Unpacks integers from a stream of 64-bit words, Nbits bits per integer.
|
| integer function, public | hfx_compression_core_methods::ints2groups (nbits, ngroup, packed_data, full_data) |
| | Packs integers at a width per group, the width the group's largest magnitude needs: header bytes (the group size, then a width per group), then 8 interleaved streams (integer 8*c+l in stream l, all at the same bit position), so that every shift is uniform over the 8 lanes; a group is Ngroup/8 such columns at its width. The integers are signed, of Nbits bits (sign folded).
|
| integer function, public | hfx_compression_core_methods::groups2ints (packed_data, full_data) |
| | Unpacks integers that ints2groups packed (signed, the sign unfolded); the group size is read from the stream.
|