Created
August 12, 2020 08:31
-
-
Save pb-uk/4ea0d7bd36a46199bc51cc87c8454d5b to your computer and use it in GitHub Desktop.
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This is part of a conversation resulting from a | |
[Post on PhysicsForums](https://www.physicsforums.com/threads/trying-to-decide-which-programming-language-i-want-to-learn.991177/post-6378518). | |
## Source code (simple Fibonnaci sequence)" | |
```cpp | |
#include <iostream> | |
using namespace std; | |
int main() { | |
int a[100] = {0, 1}; | |
int k = 2; | |
int f_k_minus_1 = 1; | |
int f_k_minus_2 = 1; | |
int f_k; | |
while (f_k < 1000) { | |
f_k = f_k_minus_1 + f_k_minus_2; | |
f_k_minus_2 = f_k_minus_1; | |
f_k_minus_1 = f_k; | |
// Either (bad): | |
a[k++] = f_k; | |
// Or (good): | |
// a[k] = f_k; | |
// k++; | |
} | |
cout << a[k - 1]; | |
return 0; | |
} | |
``` | |
## Compiled using `g++ -S` (i.e. no optimisation) g++ version 7.5.0 on x86_64 | |
Presumably because it is trying to make the object code easier to debug via a core dump the 'bad' code results in an additional save to memory. | |
### Using `a[k++]` | |
```asm | |
movl -432(%rbp), %eax ; Load k into ax. | |
leal 1(%rax), %edx ; Load k + 1 into dx. | |
movl %edx, -432(%rbp) ; Save dx into k. | |
cltq | |
movl -420(%rbp), %edx ; Load f_k into dx. | |
movl %edx, -416(%rbp,%rax,4) ; Save dx into a[ax] (the old k). | |
``` | |
### Using `a[k]; k++` | |
```asm | |
movl -432(%rbp), %eax ; Load k into ax. | |
cltq | |
movl -420(%rbp), %edx ; Load f_k into dx. | |
movl %edx, -416(%rbp,%rax,4) ; Save dx into a[ax] (we haven't incremented k yet). | |
addl $1, -432(%rbp) ; Add 1 to k; | |
``` | |
## Compiled using `g++ -S -O` g++ version 7.5.0 on x86_64 | |
Now the compiler is not worried about an informative core dump it uses registers for all intermediate values and comes up with almost the same object code for both 'good' and 'bad' sources - but separating the increment from the array indexing still saves 1 instruction, although only on the first iteration of the loop. | |
### Using `a[k++]` | |
```asm | |
.L3: | |
leal (%rsi,%rcx), %edx | |
movl %eax, %edi | |
movl %edx, -4(%rsp,%rax,4) | |
addq $1, %rax | |
movl %ecx, %esi | |
movl %edx, %ecx | |
cmpl $999, %edx | |
jle .L3 | |
``` | |
### Using `a[k]; k++;` | |
```asm | |
jmp .L3 | |
.L6: | |
movl %edx, %ecx | |
.L3: | |
leal (%rcx,%rsi), %edx | |
movl %edx, -4(%rsp,%rax,4) | |
movl %eax, %edi | |
addq $1, %rax | |
movl %ecx, %esi | |
cmpl $999, %edx | |
jle .L6 | |
``` | |
Conclusion: `a[i++];` may be quicker to type but is harder to maintain and may well run slower than `a[i]; i++;`. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment