mirror of
https://github.com/paboyle/Grid.git
synced 2024-11-14 01:35:36 +00:00
Layout changes and typo fixes
This commit is contained in:
parent
56d1308477
commit
2223cf0f4b
@ -25,14 +25,14 @@ For the MPI interfaces the optional `-auto` suffix instructs the `configure` scr
|
||||
The following options can be use with the `--enable-shm=` option to target different shared memory behaviours (default `shmopen`):
|
||||
|
||||
| `<shm>` | Description |
|
||||
| ------------- | ----------------------------------------------------------------- |
|
||||
| ------------- | ------------------------------------------------------------------------------------------------------------------------------------------------ |
|
||||
| `shmnone` | uses anonymous spaces, use only for 1 MPI rank per node |
|
||||
| `shmopen` | uses `shm_open` to allocate a shared memory space for inter socket communications. It used a unique file name in `/dev/shm` associated to the user. |
|
||||
| `shmopen` | uses `shm_open` to allocate a shared memory space for inter socket communications. Uses a unique file name in `/dev/shm` associated to the user. |
|
||||
| `hugetlbfs` | optional [libhugetlbfs](https://github.com/libhugetlbfs/libhugetlbfs) support to map the shared memory allocation into huge 2M pages |
|
||||
|
||||
|
||||
### Other flags
|
||||
|
||||
`--enable-shmpath=<path>` to select the sharem memory map base path for [libhugetlbfs](https://github.com/libhugetlbfs/libhugetlbfs).
|
||||
`--enable-shmpath=<path>` to select the shared memory map base path for [libhugetlbfs](https://github.com/libhugetlbfs/libhugetlbfs).
|
||||
|
||||
{% include paginator.html %}
|
@ -66,7 +66,7 @@ Beside the generic implementation using stencils, GRID has optimised version of
|
||||
Flags at runtime can be used for the optimised paths
|
||||
|
||||
| Flag | Description |
|
||||
| ----------- | -------------------------------------- |
|
||||
| --------------------- | --------------------------------------------------------------------- |
|
||||
| `--dslash-generic` | This is the default option and used the implementation with stencils |
|
||||
| `--dslash-unroll` | This explicitly unroll the colour loops. It is tied to `Nc=3` |
|
||||
| `--dslash-asm` | This is specific for AVX512-F architectures and `Nc=3` |
|
||||
|
@ -14,12 +14,18 @@ Using HMC in Grid version 0.7.0
|
||||
These are the instructions to use the Generalised HMC on Grid version 0.7.0.
|
||||
Disclaimer: GRID is still under active development so any information provided here can be changed in future releases.
|
||||
|
||||
General
|
||||
Introduction
|
||||
=======
|
||||
|
||||
$$ \mathbf{X}\_{n,p} = \mathbf{A}\_{n,k} \mathbf{B}\_{k,p} $$
|
||||
The general problem is to generate a Markov Chain distributed according to the action $$S(\psi)$$ in order to compute observables expectaction values.
|
||||
|
||||
$$ \langle O \rangle = \frac{1}{Z} \int O e^{-S(\psi)} D\psi $$
|
||||
|
||||
The Hybrid Monte Carlo approach is to introduce ficticious random momenta to construct an Hamiltonian $$H(\psi)$$ and generate
|
||||
new configurations by integrating the corresponding Hamilton equations.
|
||||
|
||||
$$H(\psi) = \frac{1}{2} P^2 + S(\psi)$$
|
||||
|
||||
$$ \langle O \rangle = \int O e^{-S(\psi)} d\psi $$
|
||||
|
||||
Command line options
|
||||
====================
|
||||
|
@ -14,7 +14,10 @@
|
||||
&__inner-wrap {
|
||||
@include container;
|
||||
@include clearfix;
|
||||
padding: 1em 1em 1em;
|
||||
padding-top: 0.5em;
|
||||
padding-bottom: 0.5em;
|
||||
padding-left: 1em;
|
||||
padding-right: 1em;
|
||||
font-family: $sans-serif-narrow;
|
||||
|
||||
@include breakpoint($x-large) {
|
||||
|
@ -117,7 +117,7 @@
|
||||
&--overlay {
|
||||
position: relative;
|
||||
margin-bottom: 2em;
|
||||
padding: 3em 0;
|
||||
padding: 2em 0;
|
||||
@include clearfix;
|
||||
background-size: cover;
|
||||
background-repeat: no-repeat;
|
||||
|
Loading…
Reference in New Issue
Block a user