24 lines
989 B
Markdown
24 lines
989 B
Markdown
|
# What did work:
|
||
|
|
## Removing a double calculation when step
|
||
|
|
I changed `config[i] = 0.5*previous[i] + 0.25*previous[(i-1) % size]+ 0.25*previous[(i+1) % size];` to `config[i] = 0.5*previous[i] + 0.25*previous[(i-1) % size]*2;` in step
|
||
|
|
This made all the improvement as far as i can tell
|
||
|
|
|
||
|
|
|
||
|
|
## In shift_centre_to_origin compute the ratio before the for loop
|
||
|
|
```
|
||
|
|
const float ratio = sum/size;
|
||
|
|
for(long i = 0; i < size; i++) config[i] -= ratio;
|
||
|
|
```
|
||
|
|
This did not make a noticable difference
|
||
|
|
|
||
|
|
# What did not work and actually made it slower:
|
||
|
|
- Trying to find a fancy way to get a random bool in stochastic_unit_step
|
||
|
|
- Using the std::minmax_element function in elongation
|
||
|
|
- Changing the step function to be a void that changed the array directly made it about 8 secounds slower
|
||
|
|
|
||
|
|
# Compiler optimizations
|
||
|
|
I treid these flags: -Ofast -march=native -flto -fno-signed-zeros -fno-trapping-math
|
||
|
|
The improvements were in the margin of error
|
||
|
|
|
||
|
|
# Speedup
|
||
|
|
I got at best a 1.1 times speedup. See 12_speedup.png
|