| Issue |
178785
|
| Summary |
[X86] Missed Fold: `vgf2p8affineqb(x, m1) ^ vgf2p8affineqb(x, m2)` => `vgf2p8affineqb(x, m1 ^ m2)`
|
| Labels |
new issue
|
| Assignees |
|
| Reporter |
WalterKruger
|
The matrix in `vgf2p8affineqb` can be interpreted as determining which bits of the input are XORed together for each position. Following from that, XORing two affines of the same input can be folded by XORing both their matrices and immediates together into a single `vgf2p8affineqb` instruction.
```asm
src:
vgf2p8affineqb xmm1, xmm0, xmm1, 0
vgf2p8affineqb xmm0, xmm0, xmm2, 0
vpxor xmm0, xmm0, xmm1
ret
```
```asm
tgt:
vpxor xmm1, xmm2, xmm1
vgf2p8affineqb xmm0, xmm0, xmm1, 0
ret
```
https://godbolt.org/z/jK8eWPhb7
This can also be applied to `vgf2p8affineqb(x, m) ^ x`, as the standalone `x` can be interpreted as being affined by the neutral matrix.
![]()
_______________________________________________
llvm-bugs mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
