| Issue |
202366
|
| Summary |
[WebAssembly] Suboptimal lowering of int128 operations
|
| Labels |
new issue
|
| Assignees |
|
| Reporter |
anematode
|
https://godbolt.org/z/bhf67j17j
```cpp
#include <stdint.h>
uint32_t mul_hi(uint64_t a, uint32_t b) {
return (__int128)(a) * (__int128)(b) >> 64;
}
```
Seems like we always lower int128 multiplies to a library call even if the operands have a restricted range and could be expressed in a much simpler sequence.
```wat
mul_hi(unsigned long long, unsigned int):
global.get __stack_pointer
i32.const 16
i32.sub
local.tee 2
global.set __stack_pointer
local.get 2
local.get 1
i64.extend_i32_u
i64.const 0
local.get 0
i64.const 0
call __multi3
local.get 2
i32.load 8
local.set 1
local.get 2
i32.const 16
i32.add
global.set __stack_pointer
local.get 1
end_function
```
Found this while working on [this port](https://github.com/official-stockfish/Stockfish/pull/6875/changes#diff-e428c7a719856e301ac3978f5bb612bbafba0494838d9ec4c7fc5fbda8451410R368).
![]()
_______________________________________________
llvm-bugs mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
