Skip to content

GitLab

wdata format concept · Changes

Page history
Update wdata format concept authored by Gabriel Wlazłowski's avatar Gabriel Wlazłowski
Hide whitespace changes
Inline Side-by-side
wdata-format-concept.md
View page @ 03d8e929
......@@ -56,7 +56,12 @@ link current_b current_a
const eF 0.5 MeV
const kF 1 1/fm
```
According to our experience, three types of variables (`real`, `complex`, `vector`) are sufficient and cover more than 90% of applications. The variables can be stored either in double or float precision.
According to our experience, three types of variables (`real`, `complex`, `vector`) are sufficient and cover more than 90% of applications.
Binary files store data as row arrays called `datablocks`:
![datablocks](uploads/4d9d6260250707f665bda6e6c4e921cf/datablocks.png)
The variables can be stored either in double or float precision.
| type | float | double |
|-----------|----------|-----------|
......@@ -64,26 +69,32 @@ According to our experience, three types of variables (`real`, `complex`, `vecto
| complex | `complex8` | `complex`, `complex16` |
| vector | `vector4` | `vector`, `vector8` |
However, we have implemented single-precision (`float` in C notation) types denoted as `real4`, `complex8`, and `vector4`. To be consistent with the notation, one can use `real8`, `complex16` or `vector8` exchangeable for `real`, `complex`, `vector`.
The size of the `datablock` depends on the variable type and the data dimensionality, and is computed according formula (result in bytes B):
`blocksize=blocklength*sizeB`
The `blocklength` depends on the data dimensionality (`datadim`)
|datadim | blocklength|
| -------| -----------|
|1 | `nx`|
|2 | `nx*ny`|
|3 | `nx*ny*nz`|
The `sizeB` depends on the data type
| type | sizeB | comment |
| -----| ------| -------|
|`real`, `real8` | 8 | `sizeof(double)` |
| `real4` | 4 | `sizeof(float)`
| `complex`, `complex16` | 16 | =8\*2 double\*(re,im) |
| `complex8` | 8 | =4\*2 float\*(re,im) |
| `vector(d)`, `vector8(d)` |8\*d| d-vector dimensionality, d=1,2,3(default) |
| `vector4(d)` |4\*d| d-vector dimensionality, d=1,2,3(default) |
Binary files store data as row arrays called `datablocks`:
![datablocks](uploads/4d9d6260250707f665bda6e6c4e921cf/datablocks.png)
Note that for vector variables, we use the following storage pattern:
![vecvar](uploads/39114e8349cccd7fb5f2f514dfcbfa77/vecvar.png)
The size of `datablock` depends on the variable type and the data dimensionality.
* *real*/*real8*: `blocksize=blocklength*8 Bytes`
* *real4*: `blocksize=blocklength*4 Bytes`
* *complex*/*complex16*: `blocksize=blocklength*16 Bytes`
* *complex8*: `blocksize=blocklength*8 Bytes`
* *vector*/*vector8*: `blocksize=blocklength*3*8 Bytes`
* *vector4*: `blocksize=blocklength*3*4 Bytes`
where `blocklength` is
* for datadim=3: `blocklength=nx*ny*nz`
* for datadim=2: `blocklength=nx*ny`
* for datadim=1: `blocklength=nx`
Note that for vector variables, we use the following storage pattern:
![vecvar](uploads/39114e8349cccd7fb5f2f514dfcbfa77/vecvar.png)
However, we have implemented single-precision (`float` in C notation) types denoted as `real4`, `complex8`, and `vector4`. To be consistent with the notation, one can use `real8`, `complex16` or `vector8` exchangeable for `real`, `complex`, `vector`.
To compute time associated with a given `icycle`, we use the formula
```
......