feat: I found my old blog
It only had 2 posts, but whatever
This commit is contained in:
110
content/posts/avr-memory-model.md
Normal file
110
content/posts/avr-memory-model.md
Normal file
@ -0,0 +1,110 @@
|
||||
+++
|
||||
date = '2019-03-24'
|
||||
draft = false
|
||||
title = 'AVR Memory Model: The Practical Explanation'
|
||||
tags = ['avr', 'programming', 'memory']
|
||||
categories = ['technical']
|
||||
+++
|
||||
|
||||
There is A LOT of people that have already explained this, but I personally don't feel like going through billions of
|
||||
forum posts, with most of them just dying out somewhere in 2008 with no better answer than "I figured it out, thanks".
|
||||
|
||||
Regardless. BIG shoutout to the amazing people at [AVR Freaks](https://www.avrfreaks.net/). They are really cool.
|
||||
Seriously. Make a user and ask them about anything, and they'll help you.
|
||||
|
||||
## Disclaimer
|
||||
|
||||
I have only been debugging the memory usage of a specific ATMega chip. I don't know if other AVR chip-types use the
|
||||
same, but this explanation should be valid for all MCUs that the
|
||||
[avr-libc](http://www.nongnu.org/avr-libc/user-manual/index.html) package supports.
|
||||
|
||||
I also assume that GNU/Linux is being used on the development computer.
|
||||
|
||||
### Open-Source development tools
|
||||
|
||||
The [avr-gcc](http://www.nongnu.org/avr-libc/user-manual/pages.html) compiler chain is an open source effort to have
|
||||
C/C++ for AVR Atmel chips. They do provide some rudimentary C++ support, but there's no STL and the `new` and `delete`
|
||||
keywords are not implemented by default. Even purely virtual functions doesn't work out of the box.
|
||||
|
||||
**But don't fret!** There are ways to implement those features manually. See my [other]() post about getting the build
|
||||
environment up and running.
|
||||
|
||||
## The Memory Model
|
||||
|
||||
As the avr-libc developers [explain](http://www.nongnu.org/avr-libc/user-manual/malloc.html), there's typically not a
|
||||
lot of RAM available on most many devices and therefore it's very important to keep track of how much memory you are
|
||||
using.
|
||||
|
||||
{{< centered image="/malloc-std.png" >}}
|
||||
|
||||
All of these symbols `SP`, `RAMEND`, `__data_start`, `__malloc_heap_start`, etc. Can be modified in the compiler, but
|
||||
the picture above gives the default layout (for an ATMega128 MCU). It goes without saying, that if you don't have an
|
||||
external RAM chip, you won't be able to utilize the extra RAM space for that. Otherwise, the memory addresses are pretty
|
||||
straight forward: `0x0100 => 256` bytes is the start of the memory, `0x10FF => 4351` bytes is the end. If you're
|
||||
wondering where the RAM ends on your specific MCU, you can usually simply open the spec-sheet of the chip and see the
|
||||
amount of available memory is in it.
|
||||
For the [ATMega128](https://www.microchip.com/wwwproducts/en/ATMEGA128) that number is 4096 (`4351 - 256 = 4095` (the
|
||||
spec-sheet also counts the 0th byte)).
|
||||
|
||||
## The avr-libc Memory Allocators
|
||||
|
||||
Now for the juicy part. whenever you `malloc` something in your program, the allocator first writes a 2-byte *free-list
|
||||
entry* that tells the system how big your object is.
|
||||
|
||||
Example:
|
||||
|
||||
```cpp
|
||||
/* ... */
|
||||
// Allocate an array of 5 integers
|
||||
int* my_heap_object = static_cast<int*>(malloc(sizeof(int) * 5));
|
||||
/* ... */
|
||||
```
|
||||
|
||||
Assuming that the memory has been cleared on chip-startup, the above example ends up with the memory setup looking like
|
||||
this: (Don't mind the specifc memory addresses. If you're curious, you can try doing this, by attaching `avr-gdb` to a
|
||||
simulator or On Chip Debugger (OCD)).
|
||||
|
||||
```
|
||||
gdb: > x/16xb my_heap_object
|
||||
0x800100: 0a 00 00 00 00 00 00 00
|
||||
0x800108: 00 00 00 00 00 00 00 00
|
||||
```
|
||||
|
||||
The first bytes at address `0x800100` are `0a` and `00`. These bytes are the *free-list* entry and explains how "big"
|
||||
the object is. When reading this, we have to remember that the model is littleengine-based (meaning that the bytes are
|
||||
switched),
|
||||
so we actually have the value of `0x000a`, meaning `10` in decimal. This makes a lot of sense, since we allocated 5
|
||||
`int`s, that is of size 2 (16bit integers).
|
||||
|
||||
The memory dump shows 16 bytes in total, so the last 4 bytes displayed in the gdb example are not part of the object.
|
||||
However, if you look at the Memory Model picture again, you can see that the `__brkval` value points to the biggest
|
||||
memory address that has not been allocated. In our example, if you check where the `__brkval` points to after our
|
||||
allocation, we get:
|
||||
|
||||
```
|
||||
gdb: > __brkval
|
||||
$ 268
|
||||
```
|
||||
|
||||
268 in hexadecimal is `0x10c`, and if interpreted as an address we get `0x80010c`, which fits very well with our
|
||||
example, since it is exactly 12 bytes away from where the free-list entry of `my_heap_object` is located at.
|
||||
|
||||
When `free`-ing the object again, the deallocator looks at the free-list entry at the given address, and wipes the
|
||||
free-list entry. **This is why you should not free a dangling pointer**. Freeing something that is not really free-list
|
||||
entry *will* result in undefined behaviour, and I think we all know how bad **that** is. (Even though the AVR
|
||||
environment is actually very good at handling it. In my experience, it usually just crashes and starts over.)
|
||||
However, as
|
||||
[explained](http://www.nongnu.org/avr-libc/user-manual/group__avr__stdlib.html#gafb8699abb1f51d920a176e695ff3be8a) in
|
||||
the avrlibc documentation, freeing the `NULL` value, doesn't do anything. So remember to assign your free'd pointers to
|
||||
`NULL` afterwards.
|
||||
|
||||
## Wrapping up
|
||||
|
||||
The memory allocators of AVR can be very confusing and if you don't keep your thoughts straight when programming, you
|
||||
can very easily get yourself into a lot of trouble. Since STL is not available to avr-gcc programmers, we dont have our
|
||||
glorious smart pointers, so we should implement them ourselves (or use arduino's implementations). That might become a
|
||||
future blogpost.
|
||||
|
||||
Regardless, I hope this helps the lost souls that are trying to actually use these tools.
|
||||
|
||||
{{< centered image="/6616144.png" >}}
|
||||
103
content/posts/bubble-graph.md
Normal file
103
content/posts/bubble-graph.md
Normal file
@ -0,0 +1,103 @@
|
||||
+++
|
||||
date = '2019-03-13'
|
||||
draft = false
|
||||
title = 'Plotting Circular Bubble Graph in LaTeX'
|
||||
tags = ['latex', 'programming']
|
||||
categories = ['technical']
|
||||
+++
|
||||
|
||||
When doing web-intelligence, you might need to visualize what's called inter-sentence similarity in a particular format.
|
||||
I personally haven't found an official name for these kinds of graphs, so I just simply call them Circular Bubble
|
||||
Graphs.
|
||||
|
||||
During a university project we needed such a graph in our report, and I got the idea of automatically plotting it
|
||||
through `gnuplot` and integrating it directly into our report with `gnuplottex`. You can see an example of the outcome
|
||||
of the script.
|
||||
|
||||
|
||||
|
||||
The script operates on a comma-seperated file (`.csv`). The data should be provided as a matrix of sentences assumed to
|
||||
be symmetric, with the cells containing a real number from 0 to 1, indicating the similarity between the column and the
|
||||
row. Because of this symmetric property, half of the matrix is ignored by the script. (It also ignores the diagonal,
|
||||
since sentence `23` will always have a maximum similarity to sentence `23`. It would also be hard to plot that line)
|
||||
|
||||
The whole script can be seen below, but you can also download it as a file
|
||||
[here](/sentence_similarity_graph.gnuplot). Make sure to set the `my_dataset` variable to your desired
|
||||
dataset. Example matrix can be downloaded [here](/example_similarities.csv).
|
||||
|
||||
```bash
|
||||
# Sentence similarity graph plotter
|
||||
# uncomment this for manual operation of the dataset plotted
|
||||
# my_dataset = "./sentence_similarities.csv" # ARG1
|
||||
set parametric
|
||||
set size square
|
||||
|
||||
# Styling
|
||||
set pointsize 7.5
|
||||
set style fill solid 1.0 border rgb 'grey30'
|
||||
set style line 1 lc rgb 'black' pt 6 lw 0.5
|
||||
|
||||
# Basically a one-dimensional circular coordinate system
|
||||
fx(t) = cos(t)
|
||||
fy(t) = sin(t)
|
||||
rownum = floor(system("wc -l ".my_dataset."")) +1
|
||||
coord(k) = (k/real(rownum))*(2*pi)
|
||||
fxx(t) = cos(coord(t))
|
||||
fyy(t) = sin(coord(t))
|
||||
|
||||
set trange [0:2*pi-(coord(1.0))]
|
||||
set sample rownum
|
||||
set noborder
|
||||
unset tics
|
||||
set xrange [-1.2:1.2]
|
||||
set yrange [-1.2:1.2]
|
||||
set title "Sentence inter-similarity graph"
|
||||
set multiplot
|
||||
refloptimization = 0
|
||||
do for [i = 0:rownum-1] {
|
||||
do for [j = refloptimization:rownum-1] {
|
||||
if (i != j) {
|
||||
# Get how many columns there are in the dataset.
|
||||
arrwidth = real(system("awk 'FNR == ".(i+1)." {print $".(j+1)."}' ".my_dataset.""))
|
||||
if (arrwidth > 0.0) {
|
||||
bubblerad = 0.125
|
||||
x1 = fxx(i)
|
||||
y1 = fyy(i)
|
||||
x2 = fxx(j)
|
||||
y2 = fyy(j)
|
||||
|
||||
dvx = x2-x1
|
||||
dvy = y2-y1
|
||||
dvl = sqrt((dvx ** 2) + (dvy ** 2))
|
||||
x1 = x1 + (dvx/dvl)*bubblerad
|
||||
y1 = y1 + (dvy/dvl)*bubblerad
|
||||
x2 = x2 - (dvx/dvl)*bubblerad
|
||||
y2 = y2 - (dvy/dvl)*bubblerad
|
||||
# Overleaf's arrow-width rendering is pretty terrible,
|
||||
# so we use a color-gradient to determine connection-strength.
|
||||
if (arrwidth > 0.2) {
|
||||
col = "#000000"
|
||||
} else {
|
||||
if (arrwidth < 0.1) {
|
||||
col = "#B8B8B8"
|
||||
} else {
|
||||
col = "#E4E4E4"
|
||||
}
|
||||
}
|
||||
|
||||
set arrow "".i.j."" from x1,y1 to x2,y2 nohead lw 0.5 lc rgb col
|
||||
#set label "H" at (fxx(j)-fxx(i)),(fyy(j)-fyy(i))
|
||||
show arrow "".i.j.""
|
||||
}
|
||||
}
|
||||
}
|
||||
refloptimization = refloptimization + 1
|
||||
}
|
||||
# Plot the circles
|
||||
plot '+' u (fx(t)):(fy(t)) w p ls 1 notitle
|
||||
|
||||
# Plot the sentence labels
|
||||
plot '+' u (fx(t)):(fy(t)):(sprintf("s.%d",$0+1)) with labels notitle
|
||||
```
|
||||
|
||||
{{< centered image="/6616144.png" >}}
|
||||
Reference in New Issue
Block a user