[InstCombine] Missing fold for (1 + 2 * a * b > 0) ? a : a * b --> a * b < 0 ? a * b : a

[k-arrows]

Test code:

int foo(int a, int b)
{
    return (1 + 2 * a * b > 0) ? a : a * b ;
}

clang

foo(int, int):                               // @foo(int, int)
        mul     w8, w0, w1
        mov     w9, #1                          // =0x1
        orr     w8, w9, w8, lsl #1
        cmp     w8, #1
        csinc   w8, w1, wzr, lt
        mul     w0, w8, w0
        ret

gcc

foo(int, int):
        mul     w1, w0, w1
        cmp     w1, 0
        csel    w0, w1, w0, lt
        ret

https://godbolt.org/z/b79ehqG77

[llvmbot]

@llvm/issue-subscribers-backend-aarch64

[junaire]

It's not applied to the AArch64 target but to other targets as well. I believe it's a missing fold in the middle end.
See https://alive2.llvm.org/ce/z/P-FXNQ

[junaire]

https://godbolt.org/z/dPh8P14q1

[k-arrows]

It's not applied to the AArch64 target but to other targets as well. I believe it's a missing fold in the middle end. See https://alive2.llvm.org/ce/z/P-FXNQ

Thank you for your comment. I'm sorry but the issue title was misleading.

[junaire]

Even these:

Positive * a * b + 1 > 0 ? a : a * b --> a * b < 0 : a * b ? a
Negative * a * b + 1 > 0 ? a : a * b --> a * b > 0 : a * b ? a

https://alive2.llvm.org/ce/z/9rD0BK

[junaire]

It's not applied to the AArch64 target but to other targets as well. I believe it's a missing fold in the middle end. See https://alive2.llvm.org/ce/z/P-FXNQ

Thank you for your comment. I'm sorry but the issue title was misleading.

Feel free to correct it!

[nikic]

Can you please confirm whether this report was based on a real-world motivation? I want to understand whether we should accept a non-principled fix for this or not.