| Issue |
56061
|
| Summary |
clang-x86 Missed optimizations with adding and multiplying by constants
|
| Labels |
|
| Assignees |
|
| Reporter |
AMS21
|
Watching Jason Tuners [video](https://youtu.be/NE_v4R1dPMA) sparked my interest and led me to compile a list of examples where clang generates more instructions than gcc for a simple example.
Give the following code:
```cpp
int get_value_2(int x) { return 42 + x * 2; }
int get_value_3(int x) { return 42 + x * 3; }
int get_value_5(int x) { return 42 + x * 5; }
int get_value_6(int x) { return 42 + x * 6; }
int get_value_9(int x) { return 42 + x * 9; }
int get_value_10(int x) { return 42 + x * 10; }
```
generates the following assembly:
```asm
get_value_2(int): # @get_value_2(int)
lea eax, [rdi + rdi]
add eax, 42
ret
get_value_3(int): # @get_value_3(int)
lea eax, [rdi + 2*rdi]
add eax, 42
ret
get_value_5(int): # @get_value_5(int)
lea eax, [rdi + 4*rdi]
add eax, 42
ret
get_value_6(int): # @get_value_6(int)
lea eax, [rdi + 2*rdi]
add eax, eax
add eax, 42
ret
get_value_9(int): # @get_value_9(int)
lea eax, [rdi + 8*rdi]
add eax, 42
ret
get_value_10(int): # @get_value_10(int)
lea eax, [rdi + 4*rdi]
add eax, eax
add eax, 42
ret
```
While gcc generates the following assembly:
```asm
get_value_2(int):
lea eax, [rdi+42+rdi]
ret
get_value_3(int):
lea eax, [rdi+42+rdi*2]
ret
get_value_5(int):
lea eax, [rdi+42+rdi*4]
ret
get_value_6(int):
lea eax, [rdi+rdi*2]
lea eax, [rax+42+rax]
ret
get_value_9(int):
lea eax, [rdi+42+rdi*8]
ret
get_value_10(int):
lea eax, [rdi+rdi*4]
lea eax, [rax+42+rax]
ret
```
[godbolt link](https://godbolt.org/z/erKE8GjWf)
![]()
_______________________________________________
llvm-bugs mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
